WO2019154075A1 - 一种网络参数调优方法及装置 - Google Patents
一种网络参数调优方法及装置 Download PDFInfo
- Publication number
- WO2019154075A1 WO2019154075A1 PCT/CN2019/072708 CN2019072708W WO2019154075A1 WO 2019154075 A1 WO2019154075 A1 WO 2019154075A1 CN 2019072708 W CN2019072708 W CN 2019072708W WO 2019154075 A1 WO2019154075 A1 WO 2019154075A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- entity
- network
- parameter adjustment
- kpi
- network parameter
- Prior art date
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/08—Configuration management of networks or network elements
- H04L41/0803—Configuration setting
- H04L41/0823—Configuration setting characterised by the purposes of a change of settings, e.g. optimising configuration for enhancing reliability
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W16/00—Network planning, e.g. coverage or traffic planning tools; Network deployment, e.g. resource partitioning or cells structures
- H04W16/22—Traffic simulation tools or models
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/02—Arrangements for optimising operational condition
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/06—Testing, supervising or monitoring using simulated traffic
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W28/00—Network traffic management; Network resource management
- H04W28/16—Central resource management; Negotiation of resources or communication parameters, e.g. negotiating bandwidth or QoS [Quality of Service]
- H04W28/18—Negotiating wireless communication parameters
Definitions
- the embodiments of the present invention relate to the field of communications technologies, and in particular, to a network parameter tuning method and apparatus.
- Radio resource management is a reasonable allocation and effective management of limited radio resources in a mobile communication system, so that system performance and capacity reach a joint optimal state.
- the basic starting point is to flexibly allocate and dynamically adjust the available resources of the wireless transmission network in the case of uneven distribution of network traffic, channel characteristics due to channel fading and interference fluctuations, and the goal is to ensure network quality of service (Quality of Service). , QoS), to maximize spectrum utilization and system capacity.
- Radio resources mainly include time, frequency, power, and so on.
- the functions of RRM are based on the allocation and adjustment of wireless resources, including resource allocation, admission control, and load balancing.
- the mobile communication system uses a default configuration for many parameters in the RRM.
- the above parameters may be: pilot power, reference signal (RS) power, antenna downtilt, long term evolution (LTE) reusable
- RS reference signal
- LTE long term evolution
- MR measurement report
- the default configuration parameters cannot be adapted to the terminal's movement and traffic pattern changes, and the default configuration parameters are often not optimal for each cell, which does not maximize network efficiency.
- pilot power is part of the downlink power and shares the downlink power with other downlink channels.
- the default configured pilot power is relatively large, and the power ratio of other downlink channels will be smaller, and the traffic supported by other downlink channels will be reduced; the default configured pilot power accounts for Smaller, the cell range of the pilot signal service will be reduced. Therefore, the default configured pilot power does not maximize network efficiency.
- the embodiment of the present application provides a network parameter tuning method and device, which are used to implement online network parameter tuning.
- a network parameter tuning method is provided.
- the first functional entity implements a re-force learning (RL) method to implement network parameter tuning by interacting with other functional entities, and sends a network parameter adjustment action to the execution.
- the network element of the action and iteratively updates the evaluation model according to the change of the network environment state and the obtained return after the network element performs the network parameter adjustment action, and uses the updated evaluation model to determine the next network parameter adjustment action.
- RL re-force learning
- the first functional entity sends an evaluation model to the second functional entity, where the evaluation model is used by the second functional entity to determine a first network parameter adjustment action performed by the network element in the first network environment state
- the first function entity receives the information about the first network parameter adjustment action, and the first function entity adjusts the action information according to the information about the first network environment state, the first network parameter, and the network Updating the evaluation model, the information obtained by the element after performing the first network parameter adjustment action, and the information of the second network environment state after the network element performs the first network parameter adjustment action, Transmitting, by the first functional entity, the updated evaluation model to the second functional entity, the updated evaluation model being used by the second functional entity to determine a second performed by the network element in the second network environment state Network parameter adjustment action.
- the purpose of applying the method of enhancing learning to the network parameter tuning process is achieved.
- the rule before the first functional entity updates the evaluation model, receiving configuration information of the rule, the rule includes at least one of the following: a type and/or a calculation method of the network environment state, and a network parameter adjustment action Range and step size, type of reward, and/or calculation method; the first functional entity acquires from the third functional entity according to the type of the network environment state and is consistent with the type of the reward according to the rule And the first function entity determines, according to the network data, information that the network element obtains after performing the first network parameter adjustment action, and information about the second network environment state.
- the parameter definition or algorithm of the flexible programming RL is realized, so that the data of the training or updating evaluation model is richer and has diversity and applicability, and can be based on network requirements.
- Orchestration rules is the parameter definition or algorithm of the RL.
- the first functional entity is an analysis and modeling function AMF entity
- the second functional entity is a model execution function MEF entity
- the first functional entity sends an evaluation to the second functional entity
- the model can be understood as: the AMF entity sends an evaluation model to the MEF entity, and the MEF entity determines, according to the evaluation model, the first network parameter adjustment action before the adjustment is sent to the adaptive policy function APF entity, where the adjustment is performed.
- the first network parameter adjustment action is determined by the MEF entity according to the evaluation model, and the APF entity determines, according to the policy, whether the first network parameter adjustment action sent by the received MEF entity needs to be adjusted, if no adjustment is needed, And the APF entity sends the first network parameter adjustment action received from the MEF entity to the network element, and if the adjustment is required, the APF entity sends the adjusted first to the network element.
- the network parameter adjustment action the first function entity receiving the information about the first network parameter adjustment action, which may be understood as: the AMF entity receiving the APF entity Sending the adjusted first network parameter adjustment action information, the adjusted first network parameter adjustment action is actually a first network parameter adjustment action determined by the APF entity according to the policy, and the adjusted first network parameter adjustment
- the action includes the actual adjustment of the different actions sent by the original receiving MEF entity, and may not be adjusted, the same action as the original receiving MEF entity. In this way, through the interaction between the AMF entity and the MEF entity, and further through the interaction between the MEF entity and the APF entity, and the APF entity and the network element, the method of enhancing learning can be applied to the process of network parameter tuning.
- the first functional entity is an analysis and modeling function AMF entity
- the second functional entity is a model execution function MEF entity
- the first functional entity is according to the first network environment state.
- Information, information of the adjusted first network parameter adjustment action, information obtained by the network element after receiving the adjusted first network parameter adjustment action, and the network element performing the adjustment The subsequent first network parameter adjusts the information of the second network environment state after the action, and updates the evaluation model.
- the AMF entity also sends a category of key performance indicator KPIs and a KPI guarantee threshold for each KPI category to the MEF entity, the KPI category and a KPI guarantee threshold for each KPI category being used for: the MEF entity Determining whether the specific value of the KPI of each KPI category violates the KPI guarantee rule, and sending a judgment result to the APF entity, where the APF entity takes a rollback operation when the specific value of the KPI of any KPI category violates the KPI guarantee rule.
- the action after the back-off operation is used as the adjusted first network parameter adjustment action, and the KPI guarantee rule is not higher than or lower than the KPI guarantee threshold. In this way, it is ensured that the KPI does not deteriorate during the process of tuning the network parameters using the enhanced learning method.
- the first functional entity is an AMF entity
- the second functional entity is an APF entity
- the first functional entity sends an evaluation model to the second functional entity, where the first functional entity receives
- the information about the first network parameter adjustment action may be understood as: the AMF entity sends an evaluation model to the APF entity, where the APF entity determines the first network parameter adjustment action according to the evaluation model; the AMF entity receives Information about the first network parameter adjustment action sent by the APF entity.
- the AMF entity further sends a KPI type and a KPI guarantee threshold of each KPI category to the APF entity, the KPI type and the KPI guarantee threshold of each KPI category are used to:
- the APF entity determines whether the specific value of the KPI of each KPI category violates the KPI guarantee rule, and takes a rollback operation when the specific value of any KPI violates the KPI guarantee rule, and the action after the rollback operation is as described
- the first network parameter adjustment action In this way, it is ensured that the KPI does not deteriorate during the process of tuning the network parameters using the enhanced learning method.
- the AMF entity before the AMF entity sends the KPI type and the KPI guarantee threshold of each KPI category, the AMF entity receives information of a KPI guarantee rule, where the information of the KPI guarantee rule includes: a type of the KPI, The KPI guarantee threshold for each KPI category, and the content of the rollback operation.
- the method of enhancing learning is applied to each data analysis function entity of the access network, realizing online wireless network parameter tuning and optimizing network performance.
- the AMF entity feeds back the network parameter tuning action actually sent to the NE by the AMF entity, so that the AMF entity can update the evaluation model online, so that each data analysis function entity can implement online tuning network parameters.
- the method provided by the embodiment of the present application is introduced, so that machine learning can be applied to online tuning of wireless network parameters, and the application range of machine learning in a wireless network is expanded.
- the AMF can adjust the parameters of the learning method according to the parameters of the flexible learning method, and solve the problem that the existing network cannot be tuned due to insufficient data configuration.
- the defect has achieved a parameter tuning strategy that can be flexibly configured and deployed. This avoids the need to add and modify existing RLs for parameter tuning.
- a network parameter tuning method is provided.
- the model execution function MEF entity interacts with other functional entities, and the RL method is used to implement network parameter tuning, and the network parameter adjustment action is sent to the network element performing the action, and
- the evaluation model is iteratively updated according to the change of the network environment state and the reward obtained after the network element performs the network parameter adjustment action, and the updated network model parameter adjustment action is determined by using the updated evaluation model.
- the model execution function MEF entity receives an evaluation model sent by the analysis and modeling function AMF entity; the MEF entity determines, according to the evaluation model, the first network parameter executed by the network element in the first network environment state Adjusting the action, and transmitting the information about the first network parameter adjustment action to the adaptive policy function APF entity, where the adjusted first network parameter adjustment action is sent by the APF entity to the network element, after the adjusting
- the first network parameter adjustment action includes the first network parameter adjustment action sent by the MEF entity and the network parameter adjustment action performed by the APF entity to adjust the first network parameter adjustment action according to the policy
- the MEF The entity receives the updated evaluation model sent by the AMF entity, where the updated evaluation model is information that the AMF entity adjusts the action according to the information of the first network environment state, the adjusted first network parameter, and the Deriving information that the network element obtains after performing the adjusted first network parameter adjustment action, and the network element is executing
- the information about the second network environment state after the adjusted first network parameter adjustment action is updated; the MEF entity determine
- the MEF entity receives a category of a key performance indicator KPI sent by the AMF entity and a KPI guarantee threshold of each KPI category; the MEF entity determines whether a specific value of a KPI of each KPI category violates The KPI guarantees the rule, and sends a judgment result to the APF entity, where the APF entity takes a rollback operation when the specific value of any KPI violates the KPI guarantee rule, and the action after the rollback operation is
- the adjusted first network parameter adjustment action is that the KPI guarantee rule is not higher than or lower than the KPI guarantee threshold. In this way, it is ensured that the KPI does not deteriorate during the process of tuning the network parameters using the enhanced learning method.
- the method of enhancing learning is applied to each data analysis function entity of the access network, realizing online wireless network parameter tuning and optimizing network performance.
- the AMF entity feeds back the network parameter tuning action actually sent to the NE by the AMF entity, so that the AMF entity can update the evaluation model online, so that each data analysis function entity can implement online tuning network parameters.
- the method provided by the embodiment of the present application is introduced, so that machine learning can be applied to online tuning of wireless network parameters, and the application range of machine learning in a wireless network is expanded.
- the AMF can adjust the parameters of the learning method according to the parameters of the flexible learning method, and solve the problem that the existing network cannot be tuned due to insufficient data configuration.
- the defect has achieved a parameter tuning strategy that can be flexibly configured and deployed. This avoids the need to add and modify existing RLs for parameter tuning.
- the third aspect provides a network parameter tuning method.
- the adaptive policy function APF entity uses the RL method to implement network parameter tuning by interacting with other functional entities, and sends the network parameter adjustment action to the network element performing the action. And according to the change of the network environment state and the obtained return after the network element performs the network parameter adjustment action, iteratively updates the evaluation model, and uses the updated evaluation model to determine the next network parameter adjustment action.
- the APF entity receives a first network parameter adjustment action sent by the model execution function MEF entity, where the first network parameter adjustment action is determined by the MEF entity according to the evaluation model, and the APF entity will adjust the And the network parameter adjustment action is sent to the network element, where the network element performs the adjusted first network parameter adjustment action in the first network environment state, where the adjusted first network parameter adjustment action includes the MEF
- the APF entity sends the adjusted first network parameter adjustment action information to the An AMF entity for analyzing and modeling, according to the information of the first network environment state, the adjusted first network parameter adjustment action information, and the first time after the network element performs the adjustment Information obtained after the network parameter adjustment action is obtained, and after the network element performs the adjusted first network parameter adjustment action
- the APF entity further receives information that the specific value of the KPI sent by the MEF entity violates the KPI guarantee rule; the APF entity takes a rollback when the specific value of the KPI violates the KPI guarantee rule.
- the action after the back-off operation is used as the adjusted first network parameter adjustment operation. In this way, it is ensured that the KPI does not deteriorate during the process of tuning the network parameters using the enhanced learning method.
- the method of enhancing learning is applied to each data analysis function entity of the access network, realizing online wireless network parameter tuning and optimizing network performance.
- the AMF entity feeds back the network parameter tuning action actually sent to the NE by the AMF entity, so that the AMF entity can update the evaluation model online, so that each data analysis function entity can implement online tuning network parameters.
- the method provided by the embodiment of the present application is introduced, so that machine learning can be applied to online tuning of wireless network parameters, and the application range of machine learning in a wireless network is expanded.
- the AMF can adjust the parameters of the learning method according to the parameters of the flexible learning method, and solve the problem that the existing network cannot be tuned due to insufficient data configuration.
- the defect has achieved a parameter tuning strategy that can be flexibly configured and deployed. This avoids the need to add and modify existing RLs for parameter tuning.
- the fourth aspect provides a network parameter tuning method.
- the adaptive policy function APF entity uses the RL method to implement network parameter tuning by interacting with other functional entities, and sends the network parameter adjustment action to the network element performing the action. And according to the change of the network environment state and the obtained return after the network element performs the network parameter adjustment action, iteratively updates the evaluation model, and uses the updated evaluation model to determine the next network parameter adjustment action.
- the APF entity receives an evaluation model sent by the analysis and modeling function AMF entity, and the APF entity determines, according to the evaluation model, a first network parameter adjustment action performed by the network element in the first network environment state, And sending the first network parameter adjustment action to the network element, where the network element performs a first network parameter adjustment action in a first network environment state; the APF entity adjusts the first network parameter action
- the information is sent to the AMF entity, and is used by the AMF entity according to the information about the first network environment state, the information about the first network parameter adjustment action, and the network element obtains after performing the first network parameter adjustment action.
- the reported information and the information of the second network environment state of the network element after performing the first network parameter adjustment action are updated to update the evaluation model.
- the APF entity receives the type of the KPI sent by the AMF entity and the KPI guarantee threshold of each KPI type, and determines the KPI of each KPI type according to the type of the KPI and the KPI guarantee threshold of each KPI category. Whether the specific value violates the KPI guarantee rule, and when a specific value of any KPI violates the KPI guarantee rule, a rollback operation is taken, and the action after the rollback operation is used as the first network parameter adjustment action. In this way, it is ensured that the KPI does not deteriorate during the process of tuning the network parameters using the enhanced learning method.
- the method of enhancing learning is applied to each data analysis function entity of the access network, realizing online wireless network parameter tuning and optimizing network performance.
- the AMF entity feeds back the network parameter tuning action actually sent to the NE by the AMF entity, so that the AMF entity can update the evaluation model online, so that each data analysis function entity can implement online tuning network parameters.
- the method provided by the embodiment of the present application is introduced, so that machine learning can be applied to online tuning of wireless network parameters, and the application range of machine learning in a wireless network is expanded.
- the AMF can adjust the parameters of the learning method according to the parameters of the flexible learning method, and solve the problem that the existing network cannot be tuned due to insufficient data configuration.
- the defect has achieved a parameter tuning strategy that can be flexibly configured and deployed. This avoids the need to add and modify existing RLs for parameter tuning.
- a network parameter tuning apparatus having the functionality to implement any of the possible aspects of the first aspect and the first aspect described above.
- the functions may be implemented by hardware or by corresponding software implemented by hardware.
- the hardware or software includes one or more modules corresponding to the functions described above.
- the device can be a chip or an integrated circuit.
- the apparatus includes a transceiver for communicating with the other functional entity or network element, and a processor for executing a set of programs, when the program is executed,
- the apparatus may perform the method described in the first aspect and any of the possible designs of the first aspect above.
- the apparatus also includes a memory that stores the program executed by the processor.
- the device is an AMF entity for analysis and modeling functions.
- a network parameter tuning apparatus having the functionality to implement any of the possible aspects of the second aspect and the second aspect described above.
- the functions may be implemented by hardware or by corresponding software implemented by hardware.
- the hardware or software includes one or more modules corresponding to the functions described above.
- the device can be a chip or an integrated circuit.
- the apparatus includes a transceiver for communicating with the other functional entity or network element, and a processor for executing a set of programs, when the program is executed,
- the apparatus may perform the method described in any of the possible aspects of the second aspect and the second aspect described above.
- the apparatus also includes a memory that stores the program executed by the processor.
- the device performs a functional MEF entity for the model.
- a network parameter tuning apparatus having the functionality to implement any of the possible aspects of the third aspect and the third aspect described above.
- the functions may be implemented by hardware or by corresponding software implemented by hardware.
- the hardware or software includes one or more modules corresponding to the functions described above.
- the device can be a chip or an integrated circuit.
- the apparatus includes a transceiver for communicating with the other functional entity or network element, and a processor for executing a set of programs, when the program is executed,
- the apparatus may perform the methods described in any of the possible aspects of the third and third aspects above.
- the apparatus also includes a memory that stores the program executed by the processor.
- the device is an adaptive policy function APF entity.
- a network parameter tuning apparatus having the functionality to implement any of the possible aspects of the fourth aspect and the fourth aspect described above.
- the functions may be implemented by hardware or by corresponding software implemented by hardware.
- the hardware or software includes one or more modules corresponding to the functions described above.
- the device can be a chip or an integrated circuit.
- the apparatus includes a transceiver for communicating with the other functional entity or network element, and a processor for executing a set of programs, when the program is executed,
- the apparatus may perform the method described in any of the possible aspects of the fourth aspect and the fourth aspect above.
- the apparatus also includes a memory that stores the program executed by the processor.
- the device is an adaptive policy function APF entity.
- a chip is provided, the chip being connected to a memory or the chip comprising a memory for reading and executing a software program stored in the memory to implement any of the first aspect and the first aspect as described above Possible methods described in the design.
- a chip is provided, the chip being connected to a memory or the chip comprising a memory for reading and executing a software program stored in the memory to implement any of the second aspect and the second aspect as described above Possible methods described in the design.
- a chip is provided, the chip being connected to a memory or the chip comprising a memory for reading and executing a software program stored in the memory to implement any of the third aspect and the third aspect as described above A method described in a possible design.
- a chip is provided, the chip being connected to a memory or the chip comprising a memory for reading and executing a software program stored in the memory to implement any of the fourth aspect and the fourth aspect as described above A method described in a possible design.
- a thirteenth aspect a network parameter tuning method, comprising: a first functional entity sending an evaluation model to a second functional entity; the second functional entity receiving the evaluation model from the first functional entity, and according to Determining, by the evaluation model, a first network parameter adjustment action performed by the network element in a first network environment state; the first function entity receiving information of the first network parameter adjustment action; Information of the first network environment state, information of the first network parameter adjustment action, information that the network element obtains a reward after performing the first network parameter adjustment action, and the network element is performing the first Updating the evaluation model by information of a second network environment state after the network parameter adjustment action; the first function entity transmitting an updated evaluation model to the second function entity, the updated evaluation model being used for The second functional entity determines a second network parameter adjustment action performed by the network element in the second network environment state.
- the first functional entity is an analysis and modeling function AMF entity
- the second functional entity is a model execution function MEF entity
- the first functional entity sends to the second functional entity
- the evaluation model includes: the AMF entity sends an evaluation model to the MEF entity, and the MEF entity determines, according to the evaluation model, the first network parameter adjustment action before the adjustment is sent to the adaptive policy function APF entity, where the APF is The entity sends the adjusted first network parameter adjustment action to the network element; the first function entity receives the information about the first network parameter adjustment action, where the AMF entity receives the location sent by the APF entity The information of the adjusted first network parameter adjustment action is described.
- the first functional entity if the first functional entity is an AMF entity and the second functional entity is an APF entity, the first functional entity sends an evaluation model to the second functional entity, the first function Receiving, by the AMF entity, the information about the first network parameter adjustment action, the AMF entity sending an evaluation model to the APF entity, where the APF entity determines the first network parameter adjustment action according to the evaluation model; Information about the first network parameter adjustment action sent by the APF entity.
- a communication system comprising the apparatus of at least two aspects of the fifth aspect, the sixth aspect, the seventh aspect, and the eighth aspect.
- a computer storage medium is provided, stored with a computer program comprising instructions for performing any of the possible in-design methods of the various aspects and aspects described above.
- a computer program product comprising instructions, when executed on a computer, causes the computer to perform the methods described in any of the above aspects and aspects of any of the possible aspects.
- FIG. 1 is a schematic structural diagram of a system in an embodiment of the present application.
- FIG. 2 is a schematic diagram of a method for an RL according to an embodiment of the present application.
- FIG. 3 is a schematic flowchart of a method for tuning network parameters in an embodiment of the present application
- FIG. 4 is a second schematic flowchart of a method for tuning network parameters in an embodiment of the present application.
- FIG. 5 is a third schematic flowchart of a method for tuning network parameters in an embodiment of the present application.
- FIG. 7 is a fourth schematic flowchart of a method for tuning network parameters in an embodiment of the present application.
- FIG. 8 is a schematic structural diagram of a network parameter tuning apparatus according to an embodiment of the present application.
- FIG. 9 is a second schematic structural diagram of a network parameter tuning apparatus according to an embodiment of the present application.
- FIG. 10 is a third schematic structural diagram of a network parameter tuning apparatus according to an embodiment of the present application.
- FIG. 11 is a fourth schematic structural diagram of a network parameter tuning apparatus according to an embodiment of the present application.
- FIG. 12 is a fifth schematic structural diagram of a network parameter tuning apparatus according to an embodiment of the present application.
- FIG. 13 is a sixth schematic structural diagram of a network parameter tuning apparatus according to an embodiment of the present application.
- FIG. 14 is a schematic structural diagram of a network parameter tuning apparatus according to an embodiment of the present application.
- FIG. 15 is a schematic structural diagram of a network parameter tuning apparatus according to an embodiment of the present application.
- the present application provides a network parameter tuning method and apparatus for implementing online tuning of network parameters.
- the method and the device are based on the same inventive concept. Since the principles of the method and the device for solving the problem are similar, the implementation of the device and the method can be referred to each other, and the repeated description is not repeated.
- the network parameter tuning method provided by the embodiment of the present application is applicable to different wireless access technology communication systems, such as a third generation (3rd generation, 3G) communication system, a long term evolution (LTE) system, and a fifth. More likely communication systems, such as 5th generation (5G) communication systems.
- 3G third generation
- LTE long term evolution
- 5G 5th generation
- the access network is responsible for the wireless side access of the terminal, and the possible deployment modes of the access network (AN) device include: a centralized unit (CU) and a distributed unit (DU) separation scenario; Single site scenario.
- a single site includes a gNB/NR-NB, a transmission reception point (TRP), an evolved Node B (eNB), a radio network controller (RNC), and a Node B (Node B, NB), base station controller (BSC), base transceiver station (BTS), home base station (for example, home evolved NodeB, or home Node B, HNB), baseband unit (BBU) ), or wireless fidelity (Wifi) access point (AP), etc.
- the single site is gNB/NR-NB.
- the CU supports protocols such as radio resource control (RRC), packet data convergence protocol (PDCP), and service data adaptation protocol (SDAP).
- RRC radio resource control
- PDCP packet data convergence protocol
- SDAP service data adaptation protocol
- the CU is generally deployed at the central office and has rich computing resources.
- the DU mainly supports radio link control (RLC), media access control (MAC), and physical layer (PHY) protocols.
- RLC radio link control
- MAC media access control
- PHY physical layer
- the DU generally adopts a distributed deployment mode. In a normal case, one CU needs to connect more than one DU.
- the gNB has the functions of CU and DU and is usually deployed as a single site. DU and gNB are limited by factors such as the size and power consumption of the device, and usually the computing resources are limited.
- the operation support system (OSS) of the access network is mainly used to configure parameters of the terminal device, collect alarms, performance statistics, running status, and logs of the terminal device.
- a terminal device also called a user equipment (UE), a mobile station (MS), a mobile terminal (MT), etc., is a device that provides voice and/or data connectivity to users.
- the terminal device includes a handheld device having a wireless connection function, an in-vehicle device, and the like.
- the terminal devices can be: mobile phones, tablets, laptops, PDAs, mobile internet devices (MIDs), wearable devices, virtual reality (VR) devices, augmented reality ( Augmented reality, AR) wireless terminal in equipment, industrial control, wireless terminal in self driving, wireless terminal in remote medical surgery, smart grid A wireless terminal, a wireless terminal in a transportation safety, a wireless terminal in a smart city, or a wireless terminal in a smart home.
- MIDs mobile internet devices
- VR virtual reality
- AR Augmented reality
- wireless terminal in equipment industrial control
- wireless terminal in self driving wireless terminal in remote medical surgery
- smart grid A wireless terminal, a wireless terminal in a transportation safety, a wireless terminal in a smart city, or a wireless terminal in a smart home.
- the network parameter tuning method provided by the embodiment of the present application is implemented by interaction between four functional entities, including: data service function (DSF) entity, analysis, and construction. Analyzing and modeling function (AMF) entity, model execution function (MEF) entity, and adaptive policy function (APF) entity.
- DSF data service function
- AMF Analyzing and modeling function
- MEF model execution function
- API adaptive policy function
- the signaling interaction between the four functional entities is performed, and the network parameter adjustment is implemented by using a reinforcement learning (RL) method, and the network parameter adjustment action is sent to the network element performing the action, and
- the evaluation model is iteratively updated according to the change of the network environment state and the reward obtained after the network element performs the network parameter adjustment action, and the updated network model parameter adjustment action is determined by using the updated evaluation model.
- the DSF is used to collect data and pre-process the collected data, provide the AMF with the data needed to train or update the evaluation model, and provide the MEF with the network data needed to perform the evaluation model.
- network data may also be simply referred to as data.
- the AMF is used to subscribe to the data required to train or update the evaluation model from the DSF, train or update the evaluation model based on the subscribed data, and send the evaluation model to the MEF. And, after receiving the network parameter adjustment action of the APF feedback, iteratively updates the evaluation model according to the network parameter adjustment action, and sends the updated evaluation model to the MEF or the APF.
- the MEF is used to obtain an evaluation model from the AMF, obtain network data from the DSF, use the evaluation model to perform online prediction on the network data, obtain network parameter adjustment actions, and send the network parameter adjustment action to the APF.
- APF is used to trigger a policy (such as a conflict handling policy) based on the result of analysis or prediction to change the network status, such as parameter adjustment, traffic engineering, resource scheduling, and the like.
- a policy such as a conflict handling policy
- the network parameter adjustment action is obtained from the MEF, and the network parameter adjustment action is sent to the network element that actually performs network parameter adjustment, thereby improving the capacity or performance of the network element. And, for feeding back the network parameter adjustment action to the AMF.
- the above four functional entities may be deployed on the network elements of the access network, for example, may be deployed on the CU, DU, gNB, and OSS of the access network. Specifically, the four functional entities may be deployed in the same network element or distributed in different network elements. Or, in some application scenarios, the four functional entities in the same network element perform signaling interaction to complete the network parameter tuning method in the embodiment of the present application; in other application scenarios, the functions deployed in different network elements are performed.
- the network parameter tuning method of the embodiment of the present application is implemented by performing signaling interaction between the interfaces of the network elements. For example, for some parameters with high real-time requirements, the computing resources of the DU are limited, and the evaluation model can be trained or updated by the AMF entity in the CU, and the training model or the updated evaluation model is executed by the MEF entity in the DU.
- the network parameters involved in the embodiments of the present application may refer to various parameters in the RRM, or various parameters in a radio transmission technology (RTT), or various parameters in the operation and maintenance system.
- the network parameters may be: pilot power, RS power, antenna downtilt, LTE reusable level difference threshold, MR interference decision threshold, and the like.
- Figure 2 shows a schematic diagram of the method of RL.
- RL is the process by which an agent takes a series of actions in the environment to achieve the maximum cumulative return. Agents must constantly interact with the environment to get the best strategy through trial and error. By enhancing learning, the agent can know what state of action should be taken.
- the process of RL can be expressed as a Markov decision process.
- the state is represented by s, a represents the action, Q(s, a) represents an estimate of the overall reward obtained for action a in state s, and r is an immediate return for this action.
- Q(s, a) can be a table, a function relationship, or a neural network model.
- the initial state of the environment is s t
- the agent selects an action a t in A in the initial state, and performs the action a t , wherein the action a t is an action that maximizes Q(s, a).
- the agent After the agent performs the action a t , it will get the return r t , and the state of the environment will change to become the next state s t+1 .
- the agent updates Q(s, a) according to [s t+1 , s t , a t , r t ].
- the agent selects the next action a t+1 in the next state s t+1 , and performs the action a t+1 , and performs the action a t+1 , the action a t+1 is such that the updated Q (s) , a) the biggest action.
- the agent After the agent performs action a t+1 , it will get a return r t+1 , and the state of the environment will change to become the next state s t+2 .
- the agent continues to iteratively update Q(s, a). And so on.
- the evaluation model in combination with the RL method, can be equivalent to Q(s, a), the network environment state is equivalent to s, and the network parameter adjustment action is equivalent to a.
- the flow of the method for tuning network parameters in the embodiment of the present application is specifically as follows.
- Step 301 The AMF entity receives configuration information of a rule of the RL.
- the rules of the RL may be simply referred to as rules.
- the rules include the type of network environment state and/or the calculation method of the network environment state, the range of network parameter adjustment actions, the step size of the network parameter adjustment action, the type of reward, and/or the calculation method of the reward.
- the types of the network environment state include: the load of the cell, the number of users, the current pilot power, the load difference between the local cell and the neighboring cell, and the reference signal received power (RSRP). Distribution / RSRP mean variance.
- the calculation method of the network environment state includes: when the type of the network environment state is the RSRP distribution/RSRP mean variance, the calculation method of the network environment state is the calculation method of the RSRP distribution/RSRP mean variance; when the type of the network environment state is the cell and When the load of the neighboring area is poor, the calculation method of the network environment state is the calculation method of the load difference between the cell and the neighboring cell.
- the range of the network parameter adjustment operation is the minimum value of the pilot power to the maximum value of the pilot power.
- the pilot power range is [MaxTxPower/10-13, MaxTxPower/10-7], where MaxTxPower is the maximum output power.
- the minimum pilot power is 5% of the maximum output power and the maximum is 20% of the maximum output power.
- the network parameter adjustment step size can be 1 dB. For example, an increase of 1 dB or a decrease of 1 dB based on the current pilot power.
- the types of rewards can be traffic, user perceived rate, and spectral efficiency.
- the calculation method of the reward may include: traffic/load; or traffic* key performance indicator (KPI), and the KPI may be a circuit switching (CS) call drop rate; or , the way in which traffic or load is calculated, such as traffic or load is the weighted sum of multiple cells.
- KPI traffic* key performance indicator
- CS circuit switching
- the AMF entity sends an evaluation model to the MEF entity, and the MEF entity receives the evaluation model sent by the AMF entity.
- the evaluation model may include a correspondence between a network environment state and a network parameter adjustment action, that is, according to the evaluation model, when the network environment state is known, an optimal network parameter adjustment action may be determined.
- the parameter definition or algorithm of the RL can be improved only when each new version is released, and the released version cannot be changed within the period, and needs to be Wait until the next release of the new version to improve the RL parameter definition or algorithm again.
- the parameter definition or algorithm of the flexible RL is realized by receiving the configuration information of the rule on the AMF entity setting interface, so that the data of the training or updating evaluation model is richer, diverse and applicable, and the rules can be arranged according to the network requirement. .
- the rule is the parameter definition or algorithm of the RL.
- Step 301 is an optional step.
- Step 302 The AMF entity acquires network data required for training the evaluation model from the DSF entity.
- the AMF entity subscribes the network data required by the training evaluation model to the DSF entity, and the DSF entity receives the subscription message of the AMF entity.
- the AMF entity determines the type of the network environment state and the type of the reward according to the configuration information of the rule received in step 301.
- the AMF entity subscribes to the DSF entity for network data of the network environment state of the kind, and the network data may be a specific value of the network environment state of the type, or may be source data of the network environment state of the type, and the AMF is according to the network environment state.
- the calculation method calculates a specific value of the network environment state of the kind by the source data.
- the type of the network environment state is a pilot parameter
- the AMF entity subscribes the data of the pilot parameter to the DSF entity.
- the type of the network environment state is the load difference between the cell and the neighboring cell
- the AMF entity subscribes to the load of the local cell and the load of the neighboring cell to the DSF entity, and then subscribes according to the calculation method of the load difference between the cell and the neighboring cell.
- the obtained load of the own cell and the load of the neighboring cell calculate the load difference.
- the AMF entity subscribes to the DSF entity for network data of the kind of reward, the network data may be a specific value of the reward of the category, or may be the source data of the reward of the category, and the AMF passes the calculation method according to the reward.
- the source data calculates the specific value of the return for that category.
- the type of reward is traffic
- the AMF entity subscribes to the data of the traffic to the DSF entity.
- the calculation method of the return is traffic/load
- the AMF entity subscribes the data of the traffic and the load to the DSF entity.
- Step 303 The DSF entity sends the subscribed network data to the AMF entity, and the AMF entity receives the network data sent by the DSF entity.
- Step 304 The AMF entity trains the evaluation model according to the subscribed network data.
- Step 305 The AMF entity sends an evaluation model to the MEF entity, and the MEF entity receives the evaluation model sent by the AMF entity.
- the evaluation model sent by the AMF entity may include: a type of the network environment state and/or a calculation method of the network environment state, a range of the network parameter adjustment action, a step size of the network parameter adjustment action, and a network environment state and the network parameter. Adjust the correspondence of actions.
- the AMF entity may also send the specific value of the network environment state of the type obtained in step 302 in this step.
- Step 306 The MEF entity subscribes to the DSF entity for network data required to perform the evaluation model, and the DSF entity receives the subscription message of the MEF entity.
- the MEF entity subscribes the network data of the network environment of the kind to the DSF entity according to the type of the network environment state and/or the calculation method of the network environment state sent by the AMF entity received in step 305.
- Step 307 The DSF entity sends the subscribed network data to the MEF entity, and the MEF entity receives the network data sent by the DSF entity.
- step 306 and step 307 may be omitted.
- Step 308 The MEF entity determines, according to the evaluation model sent by the AMF entity, the first network parameter adjustment action performed by the NE in the first network environment state.
- the MEF entity determines, according to the network data obtained in step 307, a specific value of the current network environment state, which is recorded as the first network environment state. Or the specific value of the current network environment state sent by the AMF entity is recorded as the first network environment state.
- the MEF entity determines, according to the evaluation model sent by the AMF entity, the network parameter adjustment action corresponding to the first network environment state, and records it as the first network parameter adjustment action.
- the first network parameter adjustment action is an action that maximizes the return, and is determined by the evaluation model.
- the step size of the network parameter adjustment action is adjusted based on the previous network parameter.
- Step 309 The MEF entity sends the first network parameter adjustment action before the adjustment to the APF entity, and the APF entity receives the adjustment first network parameter adjustment action delivered by the MEF entity.
- Step 310 The APF entity sends the adjusted first network parameter adjustment action to the NE, and the NE receives and performs the adjusted first network parameter adjustment action sent by the APF entity.
- the first network parameter adjustment action before the adjustment is determined by the MEF entity according to the evaluation model, and the APF entity determines, according to the policy, whether the first network parameter adjustment action sent by the received MEF entity needs to be adjusted, if no adjustment is needed.
- the APF entity sends the first network parameter adjustment action received from the MEF entity to the network element. If the adjustment is required, the APF entity sends the adjusted first network parameter adjustment action to the network element.
- the adjusted first network parameter adjustment action is actually a first network parameter adjustment action determined by the APF entity according to the policy, and the adjusted first network parameter adjustment action includes the actually adjusted different action sent by the original receiving MEF entity. It is also possible that there is no adjustment, the same action as the original MEF entity sent.
- the first network parameter adjustment action sent by the APF to other functional entities or network elements is actually a first network parameter adjustment action according to the policy adjustment, of course, according to The policy determines that the first network parameter is not adjusted, and the APF sends the first network parameter adjustment action received from the MEF.
- This description applies to the entire embodiment.
- the network environment state changes, and it is assumed to become the second network environment state.
- a reward is generated, which is assumed to be recorded as the first return.
- the DSF can obtain the network data of the second network environment state and the generated first reward data from the NE, and can also be obtained by other means.
- Step 311 The APF entity sends the adjusted first network parameter adjustment action to the AMF entity, and the AMF entity receives the adjusted first network parameter adjustment action sent by the APF entity.
- step 310 and step 311 has no strict sequence, and the execution order can be exchanged.
- the AMF entity acquires data of the second network environment state from the DSF entity, and acquires, after the adjusted first network parameter adjustment action sent by the NE performing the APF entity from the DSF entity. The first return of the data.
- the AMF entity updates the evaluation model according to the second network environment state, the first reward, the first network environment state, and the first network parameter adjustment action.
- the first return is an immediate return obtained by the NE performing the first network parameter adjustment action in the first network environment state, and the AMF entity can determine, according to the first return, whether the first network parameter adjustment action performed by the NE brings a positive to the network environment. Or negative impact, iteratively update the evaluation model according to the positive or negative impact, so that the updated evaluation model can obtain the maximum overall return, that is, can maximize the behavior of the NE, so that the NE performs Network parameter adjustment actions can achieve better network efficiency.
- the AMF entity sends an updated evaluation model to the MEF entity, and the MEF entity receives the updated evaluation model sent by the AMF entity.
- the MEF entity determines the second network parameter adjustment action performed in the second network environment state according to the updated evaluation model sent by the AMF entity.
- the MEF entity sends the second network parameter adjustment action to the APF entity, and the APF entity receives the second network parameter adjustment action delivered by the MEF entity.
- the APF entity sends the second network parameter adjustment action to the NE, and the NE receives and performs the second network parameter adjustment action sent by the APF entity.
- the network environment state changes, and it is assumed to become the third network environment state.
- a reward is generated, which is assumed to be recorded as a second return.
- the DSF can obtain the network data of the third network environment state and the generated second reward data from the NE, and can also be obtained by other means.
- the APF entity sends a second network parameter adjustment action to the AMF entity, and the AMF entity receives the second network parameter adjustment action sent by the APF entity.
- the operations of steps 302 to 311 are repeatedly performed.
- the AMF entity iteratively updates the evaluation model and sends it to the MEF entity for execution.
- the MEF determines the network parameter adjustment action that the NE needs to perform according to the updated evaluation model, and the MEF entity delivers the NE to the NE through the APF entity.
- the MEF entity feeds back the network parameter adjustment action performed this time to the AMF entity, so that the AMF entity iteratively updates the evaluation model.
- the method of enhancing learning is applied to each data analysis function entity of the access network, realizing online wireless network parameter tuning and optimizing network performance.
- the AMF entity feeds back the network parameter tuning action actually sent to the NE by the AMF entity, so that the AMF entity can update the evaluation model online, so that each data analysis function entity can implement online tuning network parameters.
- the method provided by the embodiment of the present application is introduced, so that machine learning can be applied to online tuning of wireless network parameters, and the application range of machine learning in a wireless network is expanded.
- the AMF can adjust the parameters of the learning method according to the parameters of the flexible learning method, and solve the problem that the existing network cannot be tuned due to insufficient data configuration.
- the defect has achieved a parameter tuning strategy that can be flexibly configured and deployed. This avoids the need to add and modify existing RLs for parameter tuning.
- the embodiment of the present application proposes several other possible network parameter tuning implementation manners through adjustment or modification of partial steps. Here are some other possible implementations.
- Steps 401 to 404 are the same as steps 301 to 304, and the details are not described again.
- Step 405 The AMF entity sends a type of network environment state and/or a calculation method of the network environment state to the MEF entity, and the MEF entity receives a type of the network environment state sent by the AMF entity and/or a calculation method of the network environment state.
- the AMF entity may also send the specific value of the network environment state of the type obtained in step 402 in this step.
- Step 406 The AMF entity sends an evaluation model to the APF entity, and the APF entity receives the evaluation model sent by the AMF entity.
- the evaluation model sent by the AMF entity may include: a range of network parameter adjustment actions, a step size of the network parameter adjustment action, and a correspondence between the network environment state and the network parameter adjustment action.
- Step 407 The MEF entity subscribes to the DSF entity for network data required to perform the evaluation model, and the DSF entity receives the subscription message of the MEF entity.
- the MEF entity subscribes the network data of the network environment of the kind to the DSF entity according to the type of the network environment state and/or the calculation method of the network environment state sent by the AMF entity received in step 405.
- Step 408 The DSF entity sends the subscribed network data to the MEF entity, and the MEF entity receives the network data sent by the DSF entity.
- Step 409 The MEF entity determines a specific value of the network environment state according to the network data sent by the DSF entity.
- step 407 to step 409 may be omitted.
- the specific value of the network environment state determined by the MEF entity may be recorded as the first network environment state.
- Step 410 The MEF entity sends a specific value of the network environment state to the APF entity. Or the specific value of the current network environment state sent by the AMF entity is recorded as the first network environment state.
- Step 411 The APF entity determines, according to the specific value of the network environment state and the evaluation model sent by the AMF entity, the first network parameter adjustment action performed by the NE in the first network environment state.
- the APF entity determines, according to the evaluation model sent by the AMF entity, the network parameter adjustment action corresponding to the first network environment state, and records it as the first network parameter adjustment action.
- the first network parameter adjustment action is an action that maximizes the return, and is determined by the evaluation model.
- the step size of the network parameter adjustment action is adjusted based on the previous network parameter.
- Step 412 The APF entity sends the first network parameter adjustment action to the NE, and the NE receives and performs the first network parameter adjustment action sent by the APF entity.
- the network environment state changes, and it is assumed to become the second network environment state.
- a reward is generated, which is assumed to be recorded as the first return.
- the DSF can obtain the network data of the second network environment state and the generated first reward data from the NE, and can also be obtained by other means.
- Step 413 The APF entity sends the first network parameter adjustment action to the AMF entity, and the AMF entity receives the first network parameter adjustment action sent by the APF entity.
- step 412 and step 413 has no strict sequence, and the execution order can be exchanged.
- the AMF entity subscribes the data of the second network environment state to the DSF entity, and subscribes to the data of the first return generated after the NE performs the first network parameter adjustment action sent by the APF entity. .
- the AMF entity updates the evaluation model according to the second network environment state, the first reward, the first network environment state, and the first network parameter adjustment action.
- the first return is an immediate return obtained by the NE performing the first network parameter adjustment action in the first network environment state, and the AMF entity can determine, according to the first return, whether the first network parameter adjustment action performed by the NE brings a positive to the network environment. Or negative impact, iteratively update the evaluation model according to the positive or negative impact, so that the updated evaluation model can obtain the maximum overall return, that is, can maximize the behavior of the NE, so that the NE performs Network parameter adjustment actions can achieve better network efficiency.
- the AMF entity sends information of the second network environment status to the MEF entity.
- the AMF entity sends an updated evaluation model to the APF entity, and the APF entity receives the updated evaluation model sent by the AMF entity.
- the APF entity determines the second network parameter adjustment action performed in the second network environment state according to the updated evaluation model sent by the AMF entity.
- the APF entity sends a second network parameter adjustment action to the NE, and the NE receives and performs the second network parameter adjustment action sent by the APF entity.
- the network environment state changes, and it is assumed to become the third network environment state.
- a reward is generated, which is assumed to be recorded as a second return.
- the DSF can obtain the network data of the third network environment state and the generated second reward data from the NE, and can also be obtained by other means.
- the APF entity sends a second network parameter adjustment action to the AMF entity, and the AMF entity receives the second network parameter adjustment action sent by the APF entity.
- the operations of steps 402 to 413 are repeatedly performed.
- the AMF entity iteratively updates the evaluation model and sends it to the APF entity for execution.
- the APF entity determines the network parameter adjustment action that the NE needs to perform according to the updated evaluation model, and the APF entity sends the network parameter adjustment action to the APF entity.
- the NE, and the APF entity feeds back the network parameter adjustment action performed this time to the AMF entity, so that the AMF entity iteratively updates the evaluation model.
- the above implementation separates the two processes of calculating the network environment state and determining the network parameter adjustment action.
- the method of enhancing learning is applied to each data analysis function entity of the access network, realizing online wireless network parameter tuning and optimizing network performance.
- the AMF entity feeds back the network parameter tuning action actually sent to the NE by the AMF entity, so that the AMF entity can update the evaluation model online, so that each data analysis function entity can implement online tuning network parameters.
- the method provided by the embodiment of the present application is introduced, so that machine learning can be applied to online tuning of wireless network parameters, and the application range of machine learning in a wireless network is expanded.
- the AMF can adjust the parameters of the learning method according to the parameters of the flexible learning method, and solve the problem that the existing network cannot be tuned due to insufficient data configuration.
- the defect has achieved a parameter tuning strategy that can be flexibly configured and deployed. This avoids the need to add and modify existing RLs for parameter tuning.
- Step 501 The AMF entity receives the configuration information of the rule and the information of the KPI guarantee rule.
- step 301 The specific steps and contents of the configuration information of the rule are as described in step 301, and details are not described herein again.
- the information of the KPI guarantee rule includes: the type of the KPI, the KPI guarantee threshold for each type, the KPI guarantee rule, the content of the rollback operation, and the conflict handling policy.
- the KPI guarantee rule is that the value of the KPI is not higher or lower than the KPI guarantee threshold.
- the rule for the rollback operation is the rollback operation taken when the specific value of any KPI violates the KPI guarantee rules.
- the content of the rollback operation is: if the specific value of the KPI violates the KPI guarantee rule, the network parameter is adjusted to a default value, or an action opposite to the network parameter adjustment action determined last time according to the evaluation model is performed.
- KPIs include, but are not limited to, radio resource control (RRC) establishment success rate, packet switched (PS) radio access bearer (RAB), CS RAB, CS call rate, PS. Call drop rate, adaptive multi rate (AMR) voice service traffic.
- RRC radio resource control
- PS packet switched
- RAB radio access bearer
- CS RAB CS call rate
- PS Call drop rate
- AMR adaptive multi rate
- the guaranteed threshold of the KPI can be 110% or 90% or 3 ⁇ of the historical statistics.
- the KPI guarantee rule is: the CS call rate is not higher than the 3 ⁇ of the past month distribution, or the CS call rate is not higher than 110% of the past week average, or the RRC establishment success rate is not lower than the average of the past two weeks. 90%.
- the content of the rollback operation may be a rollback operation when the CS drop rate is higher than the 3 ⁇ distribution of the past month, that is, the network parameter is adjusted to a default value, or the network parameter determined according to the last evaluation model is executed. Adjust an action in the opposite direction of the action.
- the conflict handling policy refers to which result is preferentially executed when the network parameter adjustment action determined according to the evaluation model violates the KPI guarantee rule.
- the conflict processing policy is defined as the network parameter adjustment action determined according to the evaluation model is inconsistent with the network parameter adjustment action determined according to the KPI guarantee rule
- the network parameter adjustment action determined according to the KPI guarantee rule takes precedence, that is, according to the evaluation model.
- the network parameter adjustment action determined according to the evaluation model is ignored or not performed.
- Step 502 The AMF entity subscribes the training evaluation model and the network data required to calculate the KPI threshold to the DSF entity, and the DSF entity receives the subscription message of the AMF entity.
- Step 503 The DSF entity sends the subscribed network data to the AMF entity, and the AMF entity receives the network data sent by the DSF entity.
- Step 504 The AMF entity trains the evaluation model according to the subscribed network data, and calculates a KPI threshold.
- the KPI guarantee threshold is 3 ⁇ of the CS drop rate distribution of the past month
- the AMF counts the mean and variance of the CS call rate, and calculates the value of 3 ⁇ , which is assumed to be 0.12%. That is, the KPI threshold for the CS call drop rate is 0.12%.
- Step 505 The AMF entity sends an evaluation model to the MEF entity, and the MEF entity receives the evaluation model sent by the AMF entity.
- the evaluation model sent by the AMF entity may include: a type of network environment state and/or a calculation method of the network environment state, a range of network parameter adjustment actions, a step size of the network parameter adjustment action, a network environment state, and network parameter adjustment.
- the correspondence of actions, and the type of KPI and the KPI guarantee threshold for each KPI category may also send the specific value of the network environment state of each KPI type obtained in step 502 and the specific value of the KPI of each KPI category in this step.
- Step 506 The AMF entity sends a conflict handling policy to the APF entity, and the APF entity receives the conflict handling policy sent by the AMF entity.
- Step 507 The MEF entity subscribes to the DSF entity for network data required to perform the evaluation model, and the DSF entity receives the subscription message of the MEF entity.
- the MEF entity subscribes the network data of the network environment of the kind to the DSF entity according to the type of the network environment state and/or the calculation method of the network environment state sent by the AMF entity received in step 505; and the MEF entity according to the step The 505 receives the KPI type sent by the AMF entity and the KPI guarantee threshold of each KPI type, and subscribes the KPI data of each KPI category to the DSF entity.
- the subscribed KPI data is the current network CS call rate and the AMR voice service. Traffic volume.
- Step 508 The DSF entity sends the subscribed network data to the MEF entity, and the MEF entity receives the network data sent by the DSF entity.
- step 507 and steps may be omitted. 508.
- Step 509 The MEF entity determines the network parameter adjustment action according to the evaluation model sent by the AMF entity, and determines whether the specific value of the current KPI violates the KPI guarantee rule.
- the MEF entity determines a specific value of the current network environment state according to the network data sent by the DSF entity, and records the state of the first network environment. Or the specific value of the current network environment state sent by the AMF entity in step 505 is recorded as the first network environment state.
- the MEF entity determines a specific value of the KPI of each KPI category according to the network data sent by the DSF entity in step 508, or the MEF entity receives the specific value of the KPI of each KPI category sent by the AMF entity.
- the MEF entity determines, according to the evaluation model sent by the AMF entity, a network parameter adjustment action corresponding to the first network environment state.
- the MEF entity also determines whether the specific value of the KPI violates the KPI guarantee rule for any kind of KPI, that is, whether the specific value of the KPI is higher or lower than the KPI threshold.
- the MEF entity may indicate, according to the violation identifier, whether the specific value of the KPI violates the KPI guarantee rule. For example, if the specific value of any type of KPI violates the KPI guarantee rule, the record KPI violation flag is 1. If the specific value of the KPI does not violate the KPI guarantee rule, the record KPI violation flag is 0.
- Step 510 The MEF entity sends the network parameter adjustment action, which is determined by the evaluation model, the network parameter adjustment action, and the specific value of the KPI, to the APF entity, and the APF entity receives the network parameter adjustment action delivered by the MEF entity. critical result.
- the judgment result may be a KPI violation identifier.
- Step 511 The APF entity determines, according to the received conflict processing policy sent by the AMF, the first network parameter adjustment action performed by the NE in the first network environment state.
- the APF entity takes a fallback operation, for example, adjusts the network parameter to a default value, or performs an action opposite to the last network parameter adjustment action.
- the action after the rollback operation is used as the first network parameter adjustment action.
- the network parameter adjustment action determined according to the evaluation model is used as the first network parameter adjustment action.
- Step 512 The APF entity sends the first network parameter adjustment action to the NE, and the NE receives and performs the first network parameter adjustment action sent by the APF entity.
- the network environment state changes, and it is assumed to become the second network environment state.
- a reward is generated, which is assumed to be recorded as the first return.
- the DSF can obtain the network data of the second network environment state and the generated first reward data from the NE, and can also be obtained by other means.
- Step 513 The APF entity sends the first network parameter adjustment action to the AMF entity, and the AMF entity receives the first network parameter adjustment action sent by the APF entity.
- step 512 and step 513 are not strictly sequential, and the execution order can be exchanged.
- the AMF entity subscribes the data of the second network environment state to the DSF entity, and subscribes to the data of the first return generated after the NE performs the first network parameter adjustment action sent by the APF entity. And subscribing to the KPI data after the NE performs the first network parameter adjustment action.
- the AMF entity updates the evaluation model according to the second network environment state, the first reward, the first network environment state, and the first network parameter adjustment action. And the AMF entity calculates the KPI threshold and the specific value of the KPI based on the subscribed KPI data for any kind of KPI.
- the first return is an immediate return obtained by the NE performing the first network parameter adjustment action in the first network environment state, and the AMF entity can determine, according to the first return, whether the first network parameter adjustment action performed by the NE brings a positive to the network environment. Or negative impact, iteratively update the evaluation model according to the positive or negative impact, so that the updated evaluation model can obtain the maximum overall return, that is, can maximize the behavior of the NE, so that the NE performs Network parameter adjustment actions can achieve better network efficiency.
- the AMF entity sends the updated evaluation model and the recalculated KPI threshold to the MEF entity, and the MEF entity receives the updated evaluation model sent by the AMF entity and the recalculated KPI threshold.
- the MEF entity acquires the KPI data after the first network parameter adjustment action from the DSF entity.
- the MEF entity determines the network parameter adjustment action performed in the second network environment state according to the updated evaluation model sent by the AMF entity. For any type of KPI, it is determined whether the specific value of the KPI after the first network parameter adjustment action violates the KPI guarantee rule, wherein the KPI guarantee rule herein includes the recalculated KPI threshold.
- the specific value of the KPI after the first network parameter adjustment action obtained by the MEF entity may be returned from the information sent by the AMF in the process of returning to step 505, or the MEF entity may return the new KPI data obtained by performing step 507 and step 508. to make sure.
- the MEF entity sends the network parameter adjustment action determined according to the updated evaluation model and the judgment result of the KPI after the first network parameter adjustment action violates the KPI guarantee rule to the APF.
- the entity, the APF entity receives the network parameter adjustment action and the judgment result delivered by the MEF entity. This judgment result is a judgment result for any kind of KPI.
- the APF entity takes a rollback operation when the specific value of the KPI after the first network parameter adjustment action violates the KPI guarantee rule for any type of KPI, and the APF entity will fall back.
- the action after the operation is used as the second network parameter adjustment action; if for any type of KPI, the specific value of the KPI after the first network parameter adjustment action does not violate the KPI guarantee rule, the network determined according to the updated evaluation model The parameter adjustment action is used as the second network parameter adjustment action.
- the APF entity sends a second network parameter adjustment action to the NE, and the NE receives and performs the second network parameter adjustment action sent by the APF entity.
- the network environment state changes, and it is assumed to become the third network environment state.
- a reward is generated, which is assumed to be recorded as a second return.
- the DSF can obtain the network data of the third network environment state and the generated second reward data from the NE, and can also be obtained by other means.
- the APF entity sends a second network parameter adjustment action to the AMF entity, and the AMF entity receives the second network parameter adjustment action sent by the APF entity.
- the operations of steps 502 to 513 are repeatedly performed.
- the AMF entity iteratively updates the evaluation model and sends it to the MEF entity for execution.
- the MEF determines the network parameter adjustment action that the NE needs to perform according to the updated evaluation model, and the MEF entity delivers the NE to the NE through the APF entity.
- the MEF entity feeds back the network parameter adjustment action performed this time to the AMF entity, so that the AMF entity iteratively updates the evaluation model.
- the method of enhancing learning is applied to each data analysis function entity of the access network, realizing online wireless network parameter tuning and optimizing network performance.
- the AMF entity feeds back the network parameter tuning action actually sent to the NE by the AMF entity, so that the AMF entity can update the evaluation model online, so that each data analysis function entity can implement online tuning network parameters.
- the method provided by the embodiment of the present application is introduced, so that machine learning can be applied to online tuning of wireless network parameters, and the application range of machine learning in a wireless network is expanded.
- the AMF can adjust the parameters of the learning method according to the parameters of the flexible learning method, and solve the problem that the existing network cannot be tuned due to insufficient data configuration.
- the defect has achieved a parameter tuning strategy that can be flexibly configured and deployed. This avoids the need to add and modify existing RLs for parameter tuning.
- the KPI does not deteriorate during the wireless network parameter tuning process, and the requirements of the wireless network environment for the KPI are ensured.
- step 401 in the implementation manner shown in FIG. 4, it may be further optimized to receive information of the KPI guarantee rule by the AMF entity; in step 404, the AMF entity also calculates the KPI. Threshold.
- the AMF entity also sends a conflict handling policy to the APF entity.
- step 411 the APF entity determines, according to the received conflict processing policy sent by the AMF, the first network parameter that is executed by the NE in the first network environment state. Adjust the action.
- Steps 701 to 704 are the same as steps 301 to 304, and are not described here.
- Step 705 The AMF entity determines a first network parameter adjustment action performed by the NE in the first network environment state.
- Step 706 The AMF entity sends the first network parameter adjustment action to the NE, and the NE receives and performs the first network parameter adjustment action sent by the AMF entity.
- the AMF entity subscribes the data of the second network environment state to the DSF entity, and subscribes to the data of the first return generated after the NE performs the first network parameter adjustment action sent by the APF entity.
- the AMF entity updates the evaluation model according to the second network environment state, the first reward, the first network environment state, and the first network parameter adjustment action.
- the first return is an immediate return obtained by the NE performing the first network parameter adjustment action in the first network environment state, and the AMF entity can determine, according to the first return, whether the first network parameter adjustment action performed by the NE brings a positive to the network environment. Or negative impact, iteratively update the evaluation model according to the positive or negative impact, so that the updated evaluation model can obtain the maximum overall return, that is, can maximize the behavior of the NE, so that the NE performs Network parameter adjustment actions can achieve better network efficiency.
- the AMF entity determines a second network parameter adjustment action performed in the second network environment state according to the updated evaluation model.
- the AMF entity sends a second network parameter adjustment action to the NE, and the NE receives and performs the second network parameter adjustment action sent by the AMF entity.
- the network environment state changes, and it is assumed to become the third network environment state.
- a reward is generated, which is assumed to be recorded as a second return.
- the DSF can obtain the network data of the third network environment state and the generated second reward data from the NE, and can also be obtained by other means.
- the AMF entity iteratively updates the evaluation model during each iteration, and determines the network parameter adjustment actions that the NE needs to perform according to the updated evaluation model.
- the AMF gets the last updated evaluation model. Next, proceed to step 707 and the following steps.
- the AMF can determine whether the evaluation model is mature according to the feedback of the environment, that is, whether the optimal network parameter can be obtained according to the evaluation model. Thereby determining the end of the above cycle step.
- Step 707 The AMF entity sends the last updated evaluation model to the MEF entity, and the MEF entity receives the last updated evaluation model sent by the AMF entity.
- Step 708 The MEF entity subscribes to the DSF entity for network data required to perform the last updated evaluation model, and the DSF entity receives the subscription message of the MEF entity.
- Step 709 The DSF entity sends the subscribed network data to the MEF entity, and the MEF entity receives the network data sent by the DSF entity.
- Step 710 The MEF entity determines a specific value of the network environment state according to the network data sent by the DSF entity, and the MEF entity determines the network parameter adjustment action performed by the NE in the current network environment state according to the last updated evaluation model sent by the AMF entity.
- This network parameter adjustment action is determined by the evaluation model that was last updated.
- Step 711 The MEF entity sends the determined network parameter adjustment action to the APF entity, and the APF entity receives the network parameter adjustment action delivered by the MEF entity.
- Step 712 The APF entity sends the network parameter adjustment action to the NE, and the NE receives and performs the network parameter adjustment action sent by the APF entity.
- the method of enhancing learning is applied to each data analysis function entity of the access network, realizing online wireless network parameter tuning and optimizing network performance.
- the method provided by the embodiment of the present application is introduced, so that machine learning can be applied to online tuning of wireless network parameters, and the application range of machine learning in a wireless network is expanded.
- the AMF can adjust the parameters of the learning method according to the parameters of the flexible learning method, and solve the problem that the existing network cannot be tuned due to insufficient data configuration.
- the defect has achieved a parameter tuning strategy that can be flexibly configured and deployed. This avoids the need to add and modify existing RLs for parameter tuning.
- the loop process of steps 702 to 706 can be regarded as the process of evaluating the model training, and the optimal evaluation model is obtained.
- the process of steps 707 to 711 can be regarded as the process of performing the evaluation model, or is predicted. process.
- the implementation method realizes the separation of the evaluation model training and the prediction, and the interface between the AMF entity and the NE is added in the implementation manner, so that when the model training and the prediction are separated, the network parameter adjustment action can still be issued, and the online implementation can be realized. Network parameter tuning.
- the MEF entity that is involved in the process of delivering the network parameter adjustment action to the NE by the APF entity may also be changed to the MEF entity without passing through the APF entity in a simple scenario. Directly send the network parameter adjustment action to the NE.
- the above simple scenario may be a scenario that does not require APF decision, such as a modulation and coding scheme (MCS) order selection.
- MCS modulation and coding scheme
- the DSF entity stores data of various types of network environment states, data of various types of returns, data of various types of KPIs, and AMF entities and MEF entities pass.
- the DSF entity is subscribed to the training or the data needed to evaluate the model to complete the function.
- different identifiers or index values may be used to distinguish the types of the different parameters, and the AMF entity and the MEF entity only need to send the data to the DSF entity.
- the identifier of the required category data can subscribe to the required data, and the DSF entity can determine the data subscribed by the AMF entity and the MEF entity according to the received identifier and feed back the data.
- the state index 1 is used to identify the type of network environment such as ⁇ pilot power, number of cell-dedicated channels, number of handovers ⁇ .
- the embodiment of the present application further provides a network parameter tuning device 800, which is used to perform the embodiments provided by the present application.
- the above network parameter tuning method, the network parameter tuning device 800 includes:
- the sending unit 801 is configured to send, to the second functional entity, an evaluation model, where the evaluation model is used by the second functional entity to determine a first network parameter adjustment action performed by the network element in the first network environment state;
- the receiving unit 802 is configured to receive information about the first network parameter adjustment action.
- the model training unit 803 is configured to: according to the information about the first network environment state, the information of the first network parameter adjustment action, the information obtained by the network element after performing the first network parameter adjustment action, and the network element executing the first network parameter Updating the evaluation model by adjusting information of the second network environment state after the action;
- the sending unit 801 is further configured to send the updated evaluation model to the second function entity, where the updated evaluation model is used by the second function entity to determine a second network parameter adjustment action performed by the network element in the second network environment state.
- the receiving unit 802 is further configured to:
- the rule includes at least one of the following: the type of the network environment state and/or the computing device, the range and step size of the network parameter adjustment action, the type of the reward, and/or Computing device
- the model training unit 803 is further configured to: obtain, according to a rule received by the receiving unit 802, network data that is consistent with the type of the network environment state and is consistent with the type of the report, and is configured to determine the network element according to the network data. Information that is rewarded after performing the first network parameter adjustment action, and information of the second network environment state.
- the device 800 is an analysis and modeling function AMF entity and the second functional entity is a model execution function MEF entity, then:
- the sending unit 801 is specifically configured to send an evaluation model to the MEF entity, where the MEF entity determines, according to the evaluation model, the first network parameter adjustment action before the adjustment is sent to the adaptive policy function APF entity, and the APF entity sends the information to the network element. Adjusted first network parameter adjustment action;
- the receiving unit 802 is specifically configured to: receive information about the adjusted first network parameter adjustment action sent by the APF entity;
- the model training unit 803 is specifically configured to: according to the information of the first network environment state, the adjusted first network parameter adjustment action information, the network element obtains the reward information after performing the adjusted first network parameter adjustment action, and the network The element updates the evaluation model by performing information of the second network environment state after the adjusted first network parameter adjustment action.
- the sending unit 801 is further configured to:
- the KPI type and the KPI guarantee threshold of each KPI type are used by the MEF entity to determine whether the specific value of the KPI of each KPI category violates the KPI guarantee.
- the rule sends a judgment result to the APF entity.
- the APF entity takes a rollback operation when the specific value of the KPI of any KPI category violates the KPI guarantee rule.
- the action after the rollback operation is used as the adjusted first network parameter adjustment action, KPI.
- the guarantee rule is no higher or lower than the KPI guarantee threshold.
- the device 800 is an AMF entity and the second functional entity is an APF entity, then:
- the sending unit 801 is specifically configured to: send an evaluation model to the APF entity, where the APF entity determines, according to the evaluation model, the first network parameter adjustment action;
- the receiving unit 802 is specifically configured to receive information about a first network parameter adjustment action sent by the APF entity.
- the sending unit 801 is further configured to:
- the type of the KPI and the KPI guarantee threshold of each KPI type are sent to the APF entity.
- the type of the KPI and the KPI guarantee threshold of each KPI type are used by the APF entity to determine whether the specific value of the KPI of each KPI category violates the KPI guarantee rule.
- the rollback operation is taken when the specific value of any KPI violates the KPI guarantee rule, and the action after the rollback operation is used as the first network parameter adjustment action.
- the receiving unit 802 is further configured to:
- the KPI guarantee rule information includes: the type of the KPI, the KPI guarantee threshold of each KPI type, and the fallback. The content of the operation.
- the embodiment of the present application further provides a network parameter tuning apparatus 900, where the network parameter tuning apparatus 900 is configured to perform the embodiments provided by the present application.
- the above network parameter tuning method, the network parameter tuning device 900 includes:
- the receiving unit 901 is configured to receive an evaluation model sent by the analysis and modeling function AMF entity;
- a determining unit 902 configured to determine, according to the evaluation model received by the receiving unit 901, a first network parameter adjustment action performed by the network element in a first network environment state;
- the sending unit 903 is configured to send the information about the first network parameter adjustment action to the adaptive policy function APF entity, and the APF entity sends the adjusted first network parameter adjustment action to the network element, and the adjusted first network parameter adjustment
- the action includes a first network parameter adjustment action sent by the MEF entity and a network parameter adjustment action performed by the APF entity to adjust the first network parameter adjustment action according to the policy;
- the receiving unit 901 is further configured to receive the updated evaluation model sent by the AMF entity, where the updated evaluation model is information that the AMF entity adjusts the action according to the information of the first network environment state, the adjusted first network parameter, and the network element is The information obtained after the adjusted first network parameter adjustment action is obtained, and the information of the second network environment state after the network element performs the adjusted first network parameter adjustment action is updated;
- the determining unit 902 is further configured to determine, according to the updated evaluation model received by the receiving unit 901, a second network parameter adjustment action performed by the network element in the second network environment state.
- the receiving unit 901 is further configured to: receive a category of a key performance indicator KPI sent by the AMF entity, and a KPI guarantee threshold of each KPI category;
- the determining unit 902 is further configured to: determine whether the specific value of the KPI of each KPI category violates the KPI guarantee rule, and the sending unit 903 is further configured to send the judgment result to the APF entity, where the specific value of the APF entity violates the KPI in any KPI
- the rollback operation is performed when the rule is guaranteed.
- the action after the rollback operation is used as the adjusted first network parameter adjustment action.
- the KPI guarantee rule is not higher than or lower than the KPI guarantee threshold.
- the embodiment of the present application further provides a network parameter tuning device 1000, which is used to perform the embodiments provided by the present application.
- the above network parameter tuning method, the network parameter tuning device 1000 includes:
- the receiving unit 1001 is configured to receive a first network parameter adjustment action sent by the model execution function MEF entity, where the first network parameter adjustment action is determined by the MEF entity according to the evaluation model;
- the sending unit 1002 is configured to send the adjusted first network parameter adjustment action to the network element, where the network element performs the adjusted first network parameter adjustment action in the first network environment state, and the adjusted first network is configured.
- the parameter adjustment action includes a first network parameter adjustment action sent by the MEF entity and a network parameter adjustment action adjusted according to the policy to adjust the first network parameter adjustment action;
- the sending unit 1002 is further configured to send the adjusted information about the first network parameter adjustment action to the analysis and modeling function AMF entity, where the AMF entity adjusts according to the information of the first network environment state and the adjusted first network parameter.
- the receiving unit 1001 is further configured to: receive whether the specific value of the KPI sent by the MEF entity violates the information of the KPI guarantee rule;
- the device 1000 further includes a processing unit 1003.
- the processing unit 1003 is configured to take a rollback operation when the specific value of the KPI violates the KPI guarantee rule, and adjust the action after the rollback operation as the adjusted first network parameter.
- the embodiment of the present application further provides a network parameter tuning apparatus 1100, which is used to perform the embodiments of the present application.
- the above network parameter tuning method, the network parameter tuning device 1100 includes:
- the receiving unit 1101 is configured to receive an evaluation model sent by the analysis and modeling function AMF entity;
- the determining unit 1102 is configured to determine, according to the evaluation model received by the receiving unit 1101, a first network parameter adjustment action performed by the network element in the first network environment state, and send the first network parameter adjustment action to the network element, where the network element is used for the network element Performing a first network parameter adjustment action in a first network environment state;
- the sending unit 1103 is configured to send information about the first network parameter adjustment action determined by the determining unit 1102 to the AMF entity, where the information about the first network environment state, the first network parameter adjustment action, and the network element are used by the AMF entity.
- the information obtained after the execution of the first network parameter adjustment action and the information of the second network environment state after the network element performs the first network parameter adjustment action are updated to update the evaluation model.
- the receiving unit 1101 is further configured to: receive a category of the KPI sent by the AMF entity, and a KPI guarantee threshold of each KPI category;
- the determining unit 1102 is further configured to determine, according to the type of the KPI received by the receiving unit 1101 and the KPI guarantee threshold of each KPI category, whether the specific value of the KPI of each KPI category violates the KPI guarantee rule, and the specific value of any KPI When the KPI guarantee rule is violated, the rollback operation is taken, and the action after the rollback operation is used as the first network parameter adjustment action.
- the embodiment of the present application further provides a network parameter tuning apparatus 1200, where the network parameter tuning apparatus 1200 is configured to perform the foregoing network parameter tuning method.
- the network parameter device 1200 includes a transceiver 1201, a processor 1202, and a memory 1203. Memory 1203 is optional.
- the processor 1202 is configured to invoke a set of programs that, when executed, cause the processor 1202 to perform the network parameter tuning method described above.
- the memory 1203 is for storing a program executed by the processor 1202.
- the function module sending unit 801 and the receiving unit 802 in FIG. 8 can be implemented by the transceiver 1201, and the model training unit 803 can be implemented by the processor 1202.
- the processor 1202 may be a central processing unit (CPU), a network processor (NP), or a combination of a CPU and an NP.
- CPU central processing unit
- NP network processor
- the processor 1202 may further include a hardware chip.
- the hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof.
- the PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a general array logic (GAL), or any combination thereof.
- the memory 1203 may include a volatile memory such as a random-access memory (RAM); the memory 1203 may also include a non-volatile memory such as a flash memory (flash) Memory), hard disk drive (HDD) or solid-state drive (SSD); the memory 1203 may also include a combination of the above types of memory.
- RAM random-access memory
- non-volatile memory such as a flash memory (flash) Memory), hard disk drive (HDD) or solid-state drive (SSD); the memory 1203 may also include a combination of the above types of memory.
- the embodiment of the present application further provides a network parameter tuning apparatus 1300, where the network parameter tuning apparatus 1300 is configured to perform the foregoing network parameter tuning method.
- the network parameter device 1300 includes a transceiver 1301, a processor 1302, and a memory 1303. Memory 1303 is optional.
- the processor 1302 is configured to invoke a set of programs that, when executed, cause the processor 1302 to perform the network parameter tuning method described above.
- the memory 1303 is for storing a program executed by the processor 1302.
- the function module receiving unit 901 and the transmitting unit 903 in FIG. 9 can be implemented by the transceiver 1301, and the determining unit 902 can be implemented by the processor 1302.
- the processor 1302 may be a central processing unit (CPU), a network processor (NP), or a combination of a CPU and an NP.
- CPU central processing unit
- NP network processor
- the processor 1302 may further include a hardware chip.
- the hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof.
- the PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a general array logic (GAL), or any combination thereof.
- the memory 1303 may include a volatile memory such as a random-access memory (RAM); the memory 1303 may also include a non-volatile memory such as a flash memory (flash) Memory), hard disk drive (HDD) or solid-state drive (SSD); the memory 1303 may also include a combination of the above types of memories.
- RAM random-access memory
- non-volatile memory such as a flash memory (flash) Memory), hard disk drive (HDD) or solid-state drive (SSD); the memory 1303 may also include a combination of the above types of memories.
- the embodiment of the present application further provides a network parameter tuning apparatus 1400, where the network parameter tuning apparatus 1400 is configured to perform the foregoing network parameter tuning method.
- the network parameter device 1400 includes a transceiver 1401, a processor 1402, and a memory 1403. Memory 1403 is optional.
- the processor 1402 is configured to invoke a set of programs that, when executed, cause the processor 1402 to perform the network parameter tuning method described above.
- the memory 1403 is for storing a program executed by the processor 1402.
- the function module receiving unit 1001 and the transmitting unit 1002 in FIG. 10 may be implemented by a transceiver 1401, and the processing unit 1003 may be implemented by the processor 1402.
- the processor 1402 may be a central processing unit (CPU), a network processor (NP), or a combination of a CPU and an NP.
- CPU central processing unit
- NP network processor
- the processor 1402 may further include a hardware chip.
- the hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof.
- the PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a general array logic (GAL), or any combination thereof.
- the memory 1403 may include a volatile memory such as a random-access memory (RAM); the memory 1403 may also include a non-volatile memory such as a flash memory (flash) Memory), hard disk drive (HDD) or solid-state drive (SSD); the memory 1403 may also include a combination of the above types of memories.
- RAM random-access memory
- non-volatile memory such as a flash memory (flash) Memory), hard disk drive (HDD) or solid-state drive (SSD); the memory 1403 may also include a combination of the above types of memories.
- the embodiment of the present application further provides a network parameter tuning apparatus 1500, where the network parameter tuning apparatus 1500 is configured to perform the foregoing network parameter tuning method.
- the network parameter device 1500 includes a transceiver 1501, a processor 1502, and a memory 1503.
- Memory 1503 is optional.
- the processor 1502 is configured to invoke a set of programs that, when executed, cause the processor 1502 to perform the network parameter tuning method described above.
- the memory 1503 is for storing a program executed by the processor 1502.
- the function module receiving unit 1101 and the transmitting unit 1103 in FIG. 11 can be implemented by the transceiver 1501, and the determining unit 1102 can be implemented by the processor 1502.
- the processor 1502 may be a central processing unit (CPU), a network processor (NP), or a combination of a CPU and an NP.
- CPU central processing unit
- NP network processor
- the processor 1502 may further include a hardware chip.
- the hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof.
- the PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a general array logic (GAL), or any combination thereof.
- the memory 1503 may include a volatile memory such as a random-access memory (RAM); the memory 1503 may also include a non-volatile memory such as a flash memory (flash) Memory), hard disk drive (HDD) or solid-state drive (SSD); the memory 1503 may also include a combination of the above types of memory.
- RAM random-access memory
- non-volatile memory such as a flash memory (flash) Memory), hard disk drive (HDD) or solid-state drive (SSD)
- the memory 1503 may also include a combination of the above types of memory.
- the embodiment of the present application further provides a chip, including a processor, for supporting the apparatus to implement the functions involved in the first functional entity in the network parameter tuning method.
- the chip is coupled to a memory or the chip includes a memory for storing program instructions and data necessary for the device.
- the embodiment of the present application further provides a chip, including a processor, for supporting the apparatus to implement the functions involved in the MEF in the network parameter tuning method.
- the chip is coupled to a memory or the chip includes a memory for storing program instructions and data necessary for the device.
- the embodiment of the present application further provides a chip, including a processor, for supporting the apparatus to implement the functions involved in the APF entity in the network parameter tuning method.
- the chip is coupled to a memory or the chip includes a memory for storing program instructions and data necessary for the device.
- the embodiment of the present application further provides a chip, including a processor, for supporting the apparatus to implement the functions involved in the APF entity in the network parameter tuning method.
- the chip is coupled to a memory or the chip includes a memory for storing program instructions and data necessary for the device.
- the embodiment of the present application provides a computer storage medium, which stores a computer program, and the computer program includes a method for performing the above network parameter tuning.
- the embodiment of the present application provides a computer program product comprising instructions, which when executed on a computer, causes the computer to perform the above network parameter tuning method.
- embodiments of the present application can be provided as a method, system, or computer program product.
- the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment in combination of software and hardware.
- the application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) including computer usable program code.
- the computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture comprising the instruction device.
- the apparatus implements the functions specified in one or more blocks of a flow or a flow and/or block diagram of the flowchart.
- These computer program instructions can also be loaded onto a computer or other programmable data processing device such that a series of operational steps are performed on a computer or other programmable device to produce computer-implemented processing for execution on a computer or other programmable device.
- the instructions provide steps for implementing the functions specified in one or more of the flow or in a block or blocks of a flow diagram.
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Quality & Reliability (AREA)
- Mobile Radio Communication Systems (AREA)
- Telephonic Communication Services (AREA)
Abstract
一种网络参数调优方法及装置,用以实现在线的网络参数调优。该方法为:第一功能实体向第二功能实体发送评估模型,该评估模型用于第二功能实体确定网元在第一网络环境状态下执行的第一网络参数调整动作;第一功能实体接收第一网络参数调整动作的信息;第一功能实体根据第一网络环境状态的信息、第一网络参数调整动作的信息、网元在执行第一网络参数调整动作之后获得回报的信息、以及网元在执行第一网络参数调整动作之后的第二网络环境状态的信息,更新评估模型;第一功能实体向第二功能实体发送更新后的评估模型,所述更新后的评估模型用于第二功能实体确定网元在第二网络环境状态执行的第二网络参数调整动作。
Description
本申请要求在2018年02月09日提交中国专利局、申请号为201810135977.2、发明名称为“一种网络参数调优方法及装置”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
本申请实施例涉及通信技术领域,尤其涉及一种网络参数调优方法及装置。
无线资源管理(radio resource management,RRM)是对移动通信***中有限的无线资源进行合理分配和有效管理,使***性能和容量达到联合最佳状态。其基本出发点是在网络话务量分布不均匀、信道特性因信道衰落和干扰而起伏变化等情况下,灵活分配和动态调整无线传输网络的可用资源,目标是在保证网络服务质量(Quality of Service,QoS)的前提下,最大限度地提高频谱利用率和***容量。无线资源主要包括时间、频率、功率等。RRM所具有的功能都是以无线资源的分配和调整为基础来展开的,主要包括资源分配、接纳控制、负荷均衡等几个方面。
移动通信***在RRM中很多参数使用默认的配置,例如,上述参数可以是:导频功率、参考信号(reference signal,RS)功率、天线下倾角、长期演进(long term evolution,LTE)可复用电平差门限、测量报告(measurement report,MR)干扰判决门限。但是默认配置的参数不能适应终端的移动和流量模式的变化,并且默认配置的参数往往不是对每个小区来说都是最优的,这样不能最大化网络效率。以导频功率为例,导频功率是下行功率的一部分,与其他下行信道共享下行功率。在一定的发射机功率下,默认配置的导频功率的占比较大,其他下行信道的功率占比就会较小,其他下行信道所支持的业务量就会减少;默认配置的导频功率占比较小,导频信号服务的小区范围就会减小。因此默认配置的导频功率不能最大化网络效率。
发明内容
本申请实施例提供一种网络参数调优的方法及装置,用以实现在线的网络参数调优。
本申请实施例提供的具体技术方案如下:
第一方面,提供一种网络参数调优方法,第一功能实体通过与其他功能实体的交互,采用增强学习(reinforcement learning,RL)的方法实现网络参数调优,将网络参数调整动作发送给执行动作的网元,并根据网元执行网络参数调整动作后网络环境状态的改变和获得的回报,来迭代更新评估模型,利用更新后的评估模型确定下一次的网络参数调整动作。
在一个可能的设计中,第一功能实体向第二功能实体发送评估模型,所述评估模型用于所述第二功能实体确定网元在第一网络环境状态下执行的第一网络参数调整动作,所述第一功能实体接收所述第一网络参数调整动作的信息,所述第一功能实体根据所述第一网络环境状态的信息、所述第一网络参数调整动作的信息、所述网元在执行所述第一网络参 数调整动作之后获得回报的信息、以及所述网元在执行所述第一网络参数调整动作之后的第二网络环境状态的信息,更新所述评估模型,所述第一功能实体向所述第二功能实体发送更新后的评估模型,所述更新后的评估模型用于所述第二功能实体确定所述网元在所述第二网络环境状态执行的第二网络参数调整动作。从而达到将增强学习的方法应用到网络参数调优过程中的目的。
在一个可能的设计中,所述第一功能实体更新所述评估模型之前,接收规则的配置信息,所述规则包括以下至少一种:网络环境状态的种类和/或计算方法、网络参数调整动作的范围和步长、回报的种类和/或计算方法;所述第一功能实体根据所述规则,从第三功能实体获取与所述网络环境状态的种类一致并和所述回报的种类一致的网络数据;所述第一功能实体根据所述网络数据,确定所述网元在执行所述第一网络参数调整动作之后获得回报的信息、以及所述第二网络环境状态的信息。这样,通过在第一功能实体设置接口接收规则的配置信息,实现了灵活编排RL的参数定义或算法,使得训练或更新评估模型的数据更加丰富和具有多样性和适用性,能够根据网络需求来编排规则。其中,规则即RL的参数定义或算法。
在一个可能的设计中,所述第一功能实体为分析和建模功能AMF实体,所述第二功能实体为模型执行功能MEF实体,则:所述第一功能实体向第二功能实体发送评估模型,可以理解为:AMF实体向MEF实体发送评估模型,用于所述MEF实体根据所述评估模型确定并向自适应策略功能APF实体下发调整前的第一网络参数调整动作,所述调整前的第一网络参数调整动作为所述MEF实体根据所述评估模型确定的,APF实体根据策略确定是否需要对接收到的MEF实体发送的第一网络参数调整动作进行调整,若不需要调整,则由所述APF实体向所述网元下发从所述MEF实体接收到的第一网络参数调整动作,若需要调整,则由所述APF实体向所述网元下发调整后的第一网络参数调整动作;所述第一功能实体接收所述第一网络参数调整动作的信息,可以理解为:所述AMF实体接收所述APF实体发送的所述调整后的第一网络参数调整动作的信息,该调整后的第一网络参数调整动作实际上为APF实体根据策略确定的第一网络参数调整动作,调整后的第一网络参数调整动作包括实际调整过的与原来接收MEF实体发送的不同的动作,也有可能并没有调整,与原来接收MEF实体发送的相同的动作。这样,通过AMF实体与MEF实体的交互,进一步通过MEF实体与APF实体、以及APF实体与网元的交互,能够将增强学习的方法应用到网络参数调优的过程中去。
在一个可能的设计中,所述第一功能实体为分析和建模功能AMF实体,所述第二功能实体为模型执行功能MEF实体,则所述第一功能实体根据所述第一网络环境状态的信息、所述调整后的第一网络参数调整动作的信息、所述网元在执行所述调整后的第一网络参数调整动作之后获得回报的信息、以及所述网元在执行所述调整后的第一网络参数调整动作之后的第二网络环境状态的信息,更新所述评估模型。
在一个可能的设计中,AMF实体还向MEF实体发送关键性能指标KPI的种类和每个KPI种类的KPI保障阈值,所述KPI种类和每个KPI种类的KPI保障阈值用于:所述MEF实体判断每个KPI种类的KPI的具体值是否违反KPI保障规则,并向APF实体发送判断结果,由所述APF实体在任一KPI种类的KPI的具体值违反所述KPI保障规则时采取回退操作,所述回退操作后的动作作为所述调整后的第一网络参数调整动作,所述KPI保障规则为不高于或者不低于所述KPI保障阈值。这样,能够保证了在使用增强学习方法实现 网络参数调优的过程中KPI不恶化。
在一个可能的设计中,所述第一功能实体为AMF实体,所述第二功能实体为APF实体,则所述第一功能实体向第二功能实体发送评估模型,所述第一功能实体接收所述第一网络参数调整动作的信息,可以理解为:AMF实体向APF实体发送评估模型,用于所述APF实体根据所述评估模型确定所述第一网络参数调整动作;所述AMF实体接收所述APF实体发送的所述第一网络参数调整动作的信息。这样,通过AMF实体与与APF实体、以及APF实体与网元的交互,能够将增强学习的方法应用到网络参数调优的过程中去。
在一个可能的设计中,所述AMF实体还向所述APF实体发送KPI的种类和每个KPI种类的KPI保障阈值,所述KPI的种类和每个KPI种类的KPI保障阈值用于:所述APF实体判断每个KPI种类的KPI的具体值是否违反KPI保障规则,并在任意一种KPI的具体值违反所述KPI保障规则时采取回退操作,所述回退操作后的动作作为所述第一网络参数调整动作。这样,能够保证了在使用增强学习方法实现网络参数调优的过程中KPI不恶化。
在一个可能的设计中,所述AMF实体发送KPI的种类和每个KPI种类的KPI保障阈值之前,所述AMF实体接收KPI保障规则的信息,所述KPI保障规则的信息包括:KPI的种类、每个KPI种类的KPI保障阈值、和回退操作的内容。
综上所述,将增强学习的方法应用于接入网的各个数据分析功能实体,实现了实时在线的无线网络参数调优,优化了网络性能。通过APF实体向AMF实体反馈实际下发给NE的网络参数调优动作,使得AMF实体能够在线更新评估模型,使得各个数据分析功能实体能够实现在线调优网络参数。引入本申请实施例提供的方法,使得机器学习能够应用于无线网络参数的在线调优,扩大了机器学习在无线网络中的应用范围。通过向AMF实体编排规则的配置,使得AMF可以根据灵活配置的用于增强学习方法中的参数来实现网络参数调优,解决了现网因为网络参数配置固定带来的数据不足导致不能进行调优的缺陷,实现了可灵活配置和部署的参数调优策略,避免了现有的RL应用于参数调优时新增和修改case需要通过软硬件版本升级实现。
第二方面,提供一种网络参数调优方法,通过模型执行功能MEF实体与其他功能实体的交互,采用RL的方法实现网络参数调优,将网络参数调整动作发送给执行动作的网元,并根据网元执行网络参数调整动作后网络环境状态的改变和获得的回报,来迭代更新评估模型,利用更新后的评估模型确定下一次的网络参数调整动作。
在一个可能的设计中,模型执行功能MEF实体接收分析和建模功能AMF实体发送的评估模型;所述MEF实体根据所述评估模型确定网元在第一网络环境状态下执行的第一网络参数调整动作,并将所述第一网络参数调整动作的信息发送给自适应策略功能APF实体,由所述APF实体将调整后的第一网络参数调整动作发送给所述网元,所述调整后的第一网络参数调整动作包括所述MEF实体发送的所述第一网络参数调整动作以及所述APF实体根据策略对所述第一网络参数调整动作进行调整后的网络参数调整动作,所述MEF实体接收所述AMF实体发送的更新后的评估模型,所述更新后的评估模型是AMF实体根据所述第一网络环境状态的信息、所述调整后的第一网络参数调整动作的信息、所述网元在执行所述调整后的第一网络参数调整动作之后获得回报的信息、以及所述网元在执行所述调整后的第一网络参数调整动作之后的第二网络环境状态的信息进行更新的;所述MEF实体根据更新后的评估模型确定所述网元在所述第二网络环境状态执行的第二网络参数调整动作。这样,通过MEF实体与AMF实体的交互,进一步通过MEF实体与APF实体、 以及APF实体与网元的交互,能够将增强学习的方法应用到网络参数调优的过程中去。
在一个可能的设计中,所述MEF实体接收所述AMF实体发送的关键性能指标KPI的种类和每个KPI种类的KPI保障阈值;所述MEF实体判断每个KPI种类的KPI的具体值是否违反KPI保障规则,并向所述APF实体发送判断结果,由所述APF实体在任意一种KPI的具体值违反所述KPI保障规则时采取回退操作,所述回退操作后的动作作为所述调整后的第一网络参数调整动作,所述KPI保障规则为不高于或者不低于所述KPI保障阈值。这样,能够保证了在使用增强学习方法实现网络参数调优的过程中KPI不恶化。
综上所述,将增强学习的方法应用于接入网的各个数据分析功能实体,实现了实时在线的无线网络参数调优,优化了网络性能。通过APF实体向AMF实体反馈实际下发给NE的网络参数调优动作,使得AMF实体能够在线更新评估模型,使得各个数据分析功能实体能够实现在线调优网络参数。引入本申请实施例提供的方法,使得机器学习能够应用于无线网络参数的在线调优,扩大了机器学习在无线网络中的应用范围。通过向AMF实体编排规则的配置,使得AMF可以根据灵活配置的用于增强学习方法中的参数来实现网络参数调优,解决了现网因为网络参数配置固定带来的数据不足导致不能进行调优的缺陷,实现了可灵活配置和部署的参数调优策略,避免了现有的RL应用于参数调优时新增和修改case需要通过软硬件版本升级实现。
第三方面,提供一种网络参数调优方法,自适应策略功能APF实体通过与其他功能实体的交互,采用RL的方法实现网络参数调优,将网络参数调整动作发送给执行动作的网元,并根据网元执行网络参数调整动作后网络环境状态的改变和获得的回报,来迭代更新评估模型,利用更新后的评估模型确定下一次的网络参数调整动作。
在一个可能的设计中,APF实体接收模型执行功能MEF实体发送的第一网络参数调整动作,所述第一网络参数调整动作是MEF实体根据评估模型确定的,所述APF实体将调整后的第一网络参数调整动作下发给网元,用于所述网元在第一网络环境状态下执行调整后的第一网络参数调整动作,所述调整后的第一网络参数调整动作包括所述MEF实体发送的所述第一网络参数调整动作以及根据策略对所述第一网络参数调整动作进行调整后的网络参数调整动作;所述APF实体将调整后的第一网络参数调整动作的信息发送给分析和建模功能AMF实体,用于AMF实体根据所述第一网络环境状态的信息、所述调整后的第一网络参数调整动作的信息、所述网元在执行所述调整后的第一网络参数调整动作之后获得回报的信息、以及所述网元在执行所述调整后的第一网络参数调整动作之后的第二网络环境状态的信息,来更新所述评估模型。这样,通过APF实体与AMF实体的交互,进一步通过APF实体与MEF实体、以及APF实体与网元的交互,能够将增强学习的方法应用到网络参数调优的过程中去。
在一个可能的设计中,所述APF实体还接收所述MEF实体发送的KPI的具体值是否违反KPI保障规则的信息;所述APF实体在KPI的具体值违反所述KPI保障规则时采取回退操作,将所述回退操作后的动作作为所述调整后的第一网络参数调整动作。这样,能够保证了在使用增强学习方法实现网络参数调优的过程中KPI不恶化。
综上所述,将增强学习的方法应用于接入网的各个数据分析功能实体,实现了实时在线的无线网络参数调优,优化了网络性能。通过APF实体向AMF实体反馈实际下发给NE的网络参数调优动作,使得AMF实体能够在线更新评估模型,使得各个数据分析功能实体能够实现在线调优网络参数。引入本申请实施例提供的方法,使得机器学习能够应用于 无线网络参数的在线调优,扩大了机器学习在无线网络中的应用范围。通过向AMF实体编排规则的配置,使得AMF可以根据灵活配置的用于增强学习方法中的参数来实现网络参数调优,解决了现网因为网络参数配置固定带来的数据不足导致不能进行调优的缺陷,实现了可灵活配置和部署的参数调优策略,避免了现有的RL应用于参数调优时新增和修改case需要通过软硬件版本升级实现。
第四方面,提供一种网络参数调优方法,自适应策略功能APF实体通过与其他功能实体的交互,采用RL的方法实现网络参数调优,将网络参数调整动作发送给执行动作的网元,并根据网元执行网络参数调整动作后网络环境状态的改变和获得的回报,来迭代更新评估模型,利用更新后的评估模型确定下一次的网络参数调整动作。
在一个可能的设计中,APF实体接收分析和建模功能AMF实体发送的评估模型,所述APF实体根据所述评估模型确定网元在第一网络环境状态下执行的第一网络参数调整动作,并将所述第一网络参数调整动作发送给所述网元,用于所述网元在第一网络环境状态下执行第一网络参数调整动作;所述APF实体将第一网络参数调整动作的信息发送给所述AMF实体,用于AMF实体根据所述第一网络环境状态的信息、所述第一网络参数调整动作的信息、所述网元在执行所述第一网络参数调整动作之后获得回报的信息、以及所述网元在执行所述第一网络参数调整动作之后的第二网络环境状态的信息,来更新所述评估模型。从而达到将增强学习的方法应用到网络参数调优过程中的目的。
在一个可能的设计中,APF实体接收AMF实体发送的KPI的种类和每个KPI种类的KPI保障阈值,根据所述KPI的种类和每个KPI种类的KPI保障阈值,判断每个KPI种类的KPI的具体值是否违反KPI保障规则,在任意一种KPI的具体值违反所述KPI保障规则时采取回退操作,将所述回退操作后的动作作为所述第一网络参数调整动作。这样,能够保证了在使用增强学习方法实现网络参数调优的过程中KPI不恶化。
综上所述,将增强学习的方法应用于接入网的各个数据分析功能实体,实现了实时在线的无线网络参数调优,优化了网络性能。通过APF实体向AMF实体反馈实际下发给NE的网络参数调优动作,使得AMF实体能够在线更新评估模型,使得各个数据分析功能实体能够实现在线调优网络参数。引入本申请实施例提供的方法,使得机器学习能够应用于无线网络参数的在线调优,扩大了机器学习在无线网络中的应用范围。通过向AMF实体编排规则的配置,使得AMF可以根据灵活配置的用于增强学习方法中的参数来实现网络参数调优,解决了现网因为网络参数配置固定带来的数据不足导致不能进行调优的缺陷,实现了可灵活配置和部署的参数调优策略,避免了现有的RL应用于参数调优时新增和修改case需要通过软硬件版本升级实现。
第五方面,提供一种网络参数调优装置,该装置具有实现上述第一方面和第一方面的任一种可能的设计的功能。所述功能可以通过硬件实现,也可以通过硬件执行相应的软件实现。所述硬件或软件包括一个或多个与上述功能相对应的模块。
在一个可能的设计中,该装置可以是芯片或者集成电路。
在一个可能的设计中,该装置包括收发器和处理器,该收发器用于该装置与其他功能实体或网元之间进行通信,处理器用于执行一组程序,当程序被执行时,所述装置可以执行上述第一方面和第一方面的任一种可能的设计中所述的方法。
在一个可能的设计中,该装置还包括存储器,该存储器存储有所述处理器执行的所述程序。
在一个可能的设计中,该装置为分析和建模功能AMF实体。
第六方面,提供一种网络参数调优装置,该装置具有实现上述第二方面和第二方面的任一种可能的设计的功能。所述功能可以通过硬件实现,也可以通过硬件执行相应的软件实现。所述硬件或软件包括一个或多个与上述功能相对应的模块。
在一个可能的设计中,该装置可以是芯片或者集成电路。
在一个可能的设计中,该装置包括收发器和处理器,该收发器用于该装置与其他功能实体或网元之间进行通信,处理器用于执行一组程序,当程序被执行时,所述装置可以执行上述第二方面和第二方面的任一种可能的设计中所述的方法。
在一个可能的设计中,该装置还包括存储器,该存储器存储有所述处理器执行的所述程序。
在一个可能的设计中,该装置为模型执行功能MEF实体。
第七方面,提供一种网络参数调优装置,该装置具有实现上述第三方面和第三方面的任一种可能的设计的功能。所述功能可以通过硬件实现,也可以通过硬件执行相应的软件实现。所述硬件或软件包括一个或多个与上述功能相对应的模块。
在一个可能的设计中,该装置可以是芯片或者集成电路。
在一个可能的设计中,该装置包括收发器和处理器,该收发器用于该装置与其他功能实体或网元之间进行通信,处理器用于执行一组程序,当程序被执行时,所述装置可以执行上述第三方面和第三方面的任一种可能的设计中所述的方法。
在一个可能的设计中,该装置还包括存储器,该存储器存储有所述处理器执行的所述程序。
在一个可能的设计中,该装置为自适应策略功能APF实体。
第八方面,提供一种网络参数调优装置,该装置具有实现上述第四方面和第四方面的任一种可能的设计的功能。所述功能可以通过硬件实现,也可以通过硬件执行相应的软件实现。所述硬件或软件包括一个或多个与上述功能相对应的模块。
在一个可能的设计中,该装置可以是芯片或者集成电路。
在一个可能的设计中,该装置包括收发器和处理器,该收发器用于该装置与其他功能实体或网元之间进行通信,处理器用于执行一组程序,当程序被执行时,所述装置可以执行上述第四方面和第四方面的任一种可能的设计中所述的方法。
在一个可能的设计中,该装置还包括存储器,该存储器存储有所述处理器执行的所述程序。
在一个可能的设计中,该装置为自适应策略功能APF实体。
第九方面,提供一种芯片,该芯片与存储器相连或者该芯片包括存储器,用于读取并执行所述存储器中存储的软件程序,以实现如上述第一方面和第一方面的任一种可能的设计中所述的方法。
第十方面,提供一种芯片,该芯片与存储器相连或者该芯片包括存储器,用于读取并执行所述存储器中存储的软件程序,以实现如上述第二方面和第二方面的任一种可能的设计中所述的方法。
第十一方面,提供一种芯片,该芯片与存储器相连或者该芯片包括存储器,用于读取并执行所述存储器中存储的软件程序,以实现如上述第三方面和第三方面的任一种可能的设计中所述的方法。
第十二方面,提供一种芯片,该芯片与存储器相连或者该芯片包括存储器,用于读取并执行所述存储器中存储的软件程序,以实现如上述第四方面和第四方面的任一种可能的设计中所述的方法。
第十三方面,提供一种网络参数调优方法,包括:第一功能实体向第二功能实体发送评估模型;所述第二功能实体从所述第一功能实体接收所述评估模型,并根据所述评估模型确定网元在第一网络环境状态下执行的第一网络参数调整动作;所述第一功能实体接收所述第一网络参数调整动作的信息;所述第一功能实体根据所述第一网络环境状态的信息、所述第一网络参数调整动作的信息、所述网元在执行所述第一网络参数调整动作之后获得回报的信息、以及所述网元在执行所述第一网络参数调整动作之后的第二网络环境状态的信息,更新所述评估模型;所述第一功能实体向所述第二功能实体发送更新后的评估模型,所述更新后的评估模型用于所述第二功能实体确定所述网元在所述第二网络环境状态执行的第二网络参数调整动作。
在一个可能的设计中,若所述第一功能实体为分析和建模功能AMF实体,所述第二功能实体为模型执行功能MEF实体,则:所述第一功能实体向第二功能实体发送评估模型,包括:AMF实体向MEF实体发送评估模型,用于所述MEF实体根据所述评估模型确定并向自适应策略功能APF实体下发调整前的第一网络参数调整动作,由所述APF实体向所述网元下发调整后的第一网络参数调整动作;所述第一功能实体接收所述第一网络参数调整动作的信息,包括:所述AMF实体接收所述APF实体发送的所述调整后的第一网络参数调整动作的信息。
在一个可能的设计中,若所述第一功能实体为AMF实体,所述第二功能实体为APF实体,则:所述第一功能实体向第二功能实体发送评估模型,所述第一功能实体接收所述第一网络参数调整动作的信息,包括:AMF实体向APF实体发送评估模型,用于所述APF实体根据所述评估模型确定所述第一网络参数调整动作;所述AMF实体接收所述APF实体发送的所述第一网络参数调整动作的信息。
第十三方面,提供了一种通信***,该通信***包括第五方面、第六方面、第七方面和第八方面中至少两方面所述的装置。
第十四方面,提供一种计算机存储介质,存储有计算机程序,该计算机程序包括用于执行上述各方面和各方面的任一可能的设计中方法的指令。
第十五方面,提供了一种包含指令的计算机程序产品,当其在计算机上运行时,使得计算机执行上述各方面和各方面的任一可能的设计中所述的方法。
图1为本申请实施例中***架构示意图;
图2为本申请实施例中RL的方法示意图;
图3为本申请实施例中网络参数调优的方法流程示意图之一;
图4为本申请实施例中网络参数调优的方法流程示意图之二;
图5为本申请实施例中网络参数调优的方法流程示意图之三;
图6为本申请实施例中历史统计的CS掉话率分布;
图7为本申请实施例中网络参数调优的方法流程示意图之四;
图8为本申请实施例中网络参数调优装置的结构示意图之一;
图9为本申请实施例中网络参数调优装置的结构示意图之二;
图10为本申请实施例中网络参数调优装置的结构示意图之三;
图11为本申请实施例中网络参数调优装置的结构示意图之四;
图12为本申请实施例中网络参数调优装置的结构示意图之五;
图13为本申请实施例中网络参数调优装置的结构示意图之六;
图14为本申请实施例中网络参数调优装置的结构示意图之七;
图15为本申请实施例中网络参数调优装置的结构示意图之八。
本申请提供一种网络参数调优方法及装置,用以实现对网络参数进行在线的调优。其中,方法和装置是基于同一发明构思的,由于方法及装置解决问题的原理相似,因此装置与方法的实施可以相互参见,重复之处不再赘述。
下面将结合附图,对本申请实施例进行详细描述。
本申请实施例提供的网络参数调优方法适用于不同的无线接入技术的通信***中,例如第三代(3rd Generation,3G)通信***、长期演进(long term evolution,LTE)***、第五代(5th generation,5G)通信***等更多可能的通信***中。
接入网负责终端的无线侧接入,接入网(access network,AN)设备可能的部署形态包括:集中式单元(centralized unit,CU)和分布式单元(distributed unit,DU)分离场景;以及单站点的场景。单站点包括gNB/NR-NB、传输接收点(transmission reception point,TRP)、演进型节点B(evolved Node B,eNB)、无线网络控制器(radio network controller,RNC)、节点B(Node B,NB)、基站控制器(base station controller,BSC)、基站收发台(base transceiver station,BTS)、家庭基站(例如,home evolved NodeB,或home Node B,HNB)、基带单元(base band unit,BBU),或无线保真(wireless fidelity,Wifi)接入点(access point,AP)等。在5G通信***中,单站点为gNB/NR-NB。其中,CU支持无线资源控制(radio resource control,RRC)、分组数据汇聚协议(packet data convergence protocol,PDCP)、业务数据适配协议(service data adaptation protocol,SDAP)等协议。CU一般会部署在中心局点,具有较为丰富的计算资源。DU主要支持无线链路控制层(radio link control,RLC)、媒体接入控制层(media access control,MAC)和物理层(PHY)协议。DU一般采用分布式部署方式,在通常情况下一个CU要连接一个以上的DU。gNB具有CU和DU的功能,并且通常作为单站点的形态部署。DU和gNB受限于设备的体积、功耗等因素,通常计算资源较为有限。
接入网的运营支持***(operation support system,OSS)主要用于配置终端设备的参数、收集终端设备的告警、性能统计、运行状态和日志等信息数据。终端设备,又称之为用户设备(user equipment,UE)、移动台(mobile station,MS)、移动终端(mobile terminal,MT)等,是一种向用户提供语音和/或数据连通性的设备。例如,终端设备包括具有无线连接功能的手持式设备、车载设备等。目前,终端设备可以是:手机(mobile phone)、平板电脑、笔记本电脑、掌上电脑、移动互联网设备(mobile internet device,MID)、可穿戴设备,虚拟现实(virtual reality,VR)设备、增强现实(augmented reality,AR)设备、工业控制(industrial control)中的无线终端、无人驾驶(self driving)中的无线终端、远程手 术(remote medical surgery)中的无线终端、智能电网(smart grid)中的无线终端、运输安全(transportation safety)中的无线终端、智慧城市(smart city)中的无线终端,或智慧家庭(smart home)中的无线终端等。
如图1所示,本申请实施例提供的网络参数调优方法通过四个功能实体之间的交互来实现,四个功能实体包括:数据服务功能(data service function,DSF)实体、分析和建模功能(analyzing and modeling function,AMF)实体、模型执行功能(model execution function,MEF)实体、和自适应策略功能(adaptive policy function,APF)实体。本申请实施例中,通过上述四个功能实体之间进行信令交互,采用增强学习(reinforcement learning,RL)的方法实现网络参数调优,将网络参数调整动作发送给执行动作的网元,并根据网元执行网络参数调整动作后网络环境状态的改变和获得的回报,来迭代更新评估模型,利用更新后的评估模型确定下一次的网络参数调整动作。
以下介绍在实现网络参数调优时上述四个功能实体所执行的功能。
DSF,用于采集数据,并对采集的数据进行预处理,向AMF提供训练或更新评估模型所需的数据,并向MEF提供执行评估模型时所需的网络数据。以下描述中,网络数据也可以简称为数据。
AMF,用于从DSF订阅训练或更新评估模型所需的数据,根据订阅的数据训练或更新评估模型,将评估模型发送给MEF。以及,用于在接收到APF反馈的网络参数调整动作后,根据网络参数调整动作进行迭代更新评估模型,并将更新的评估模型发送给MEF或APF。
MEF,用于从AMF获取评估模型,并从DSF获取网络数据,采用评估模型对网络数据进行在线预测,获得网络参数调整动作,将网络参数调整动作发送给APF。
APF,用于根据分析或预测的结果来触发策略(如冲突处理策略),以改变网络状态,如调参、流量工程、资源调度等。具体用于从MEF获取网络参数调整动作,将网络参数调整动作发送给实际执行网络参数调整的网元,从而改善网元的容量或者性能。以及,用于将网络参数调整动作反馈给AMF。
上述四个功能实体可以部署在接入网的网元上,例如,可以部署在接入网的CU、DU、gNB和OSS上。具体的,四个功能实体可以部署在同一个网元中,也可以分散的部署在不同的网元中。或者说,在一些应用场景下,同一个网元中的四个功能实体进行信令交互来完成本申请实施例的网络参数调优方法;在另一些应用场景下,部署在不同网元的功能实体通过网元之间的接口进行信令交互来完成本申请实施例的网络参数调优方法。例如,对于一些实时性要求高的参数,DU的计算资源受限,可以通过CU中的AMF实体来训练或更新评估模型,通过DU中的MEF实体来执行训练或更新后的评估模型。
需要说明的是,上述四个功能实体的名称本申请中不作限定,本领域技术人员将上述功能实体的名称更换为其它名称而执行相同的功能,均属于本申请保护的范围。
本申请实施例涉及的网络参数可以是指RRM中的各种参数,或无线传输技术(radio transmission technology,RTT)中的各种参数,或运维***中的各种参数。例如网络参数可以是:导频功率、RS功率、天线下倾角、LTE可复用电平差门限、MR干扰判决门限等。
以下将结合附图对本申请实施例提供的网络参数调优方法做详细说明。
为方便对本申请实施例的理解,首先介绍一下RL的方法。
图2示出了RL的方法示意图。RL是指智能体(agent)如何在环境中采取一系列行为获 得最大的累积回报的过程。智能体要不断与环境进行交互,通过试错的方式来获得最佳策略。通过增强学习,智能体能够知道在什么状态下应该采取什么行为。
RL的过程可以表示成一个马尔科夫决策过程。用s表示状态,a表示动作,Q(s,a)表示对状态s下动作a得到的总体回报的一个估计,r为此动作的立即回报。Q(s,a)可以是一个表格,也可以是一种函数关系,也可以是一个神经网络模型。
环境的初始状态为s
t,智能体在初始状态下选择A中的一个动作a
t,并执行该动作a
t,其中该动作a
t为使Q(s,a)最大的动作。智能体在执行动作a
t后,会获得回报r
t,且环境的状态会发生改变,变为下一个状态s
t+1。智能体根据[s
t+1,s
t,a
t,r
t]来更新Q(s,a)。智能体在下一个状态s
t+1下选择下一个动作a
t+1,并执行动作a
t+1,并执行该动作a
t+1,该动作a
t+1为使得更新后的Q(s,a)最大的动作。智能体在执行动作a
t+1后,会获得回报r
t+1,且环境的状态会发生改变,变为下一个状态s
t+2。智能体继续迭代更新Q(s,a)。以此类推。
以下叙述中,结合RL的方法来说,评估模型可以相当于Q(s,a),网络环境状态相当于s,网络参数调整动作相当于a。
如图3所示,本申请实施例的网络参数调优的方法流程具体如下所述。
步骤301、AMF实体接收RL的规则的配置信息。
RL的规则可以简称为规则。该规则包括网络环境状态的种类和/或网络环境状态的计算方法、网络参数调整动作的范围和网络参数调整动作的步长、回报的种类和/或回报的计算方法。
以网络参数为导频功率为例,网络环境状态的种类包括:本小区负载、用户数、当前导频功率、本小区和邻区的负载差、参考信号接收功率(reference signal received power,RSRP)分布/RSRP均值方差。网络环境状态的计算方法包括:当网络环境状态的种类为RSRP分布/RSRP均值方差时,网络环境状态的计算方法为RSRP分布/RSRP均值方差的计算方式;当网络环境状态的种类为本小区和邻区的负载差时,网络环境状态的计算方法为本小区和邻区的负载差的计算方式。网络参数调整动作的范围为导频功率的最小值~导频功率的最大值。例如,导频功率的范围为[MaxTxPower/10-13,MaxTxPower/10-7],其中,MaxTxPower为最大输出功率。或者,导频功率的最小值为最大输出功率的5%,最大值为最大输出功率的20%。网络参数调整步长可以为1dB。例如,在当前导频功率的基础上增加1dB或减少1dB。假设增加1dB后的值超过导频功率的最大值,则将本次网络参数调整动作设定为调整到导频功率的最大值;假设减少1dB后的值低于导频功率的最小值,则将本次网络参数调整动作设定为调整到导频功率的最小值。回报的种类可以为流量(traffic)、用户感知速率、频谱效率。回报的计算方法可包括:流量/负载(traffic/load);或者,流量(traffic)*关键性能指标(key performance indicator,KPI),KPI可以是电路交换(circuit switching,CS)掉话率;或者,流量或负载的计算方式,如流量或负载为多个小区的加权和。
AMF实体向MEF实体发送评估模型,MEF实体接收AMF实体发送的评估模型。
其中,评估模型可以包括网络环境状态与网络参数调整动作的对应关系,即根据评估模型,在已知网络环境状态时,可以确定最优的网络参数调整动作。
由于现有技术中将RL应用于参数调优的方法只能是按照软硬件版本,只有在每次发行新版本时才可以改进RL的参数定义或算法,发行的版本在周期内不能更改,需要等到 下次发行新版本才能再次改进RL的参数定义或算法。本步骤通过在AMF实体设置接口接收规则的配置信息,实现了灵活编排RL的参数定义或算法,使得训练或更新评估模型的数据更加丰富和具有多样性和适用性,能够根据网络需求来编排规则。其中,规则即RL的参数定义或算法。
步骤301为可选的步骤。
步骤302、AMF实体从DSF实体获取训练评估模型所需要的网络数据。
可选的,AMF实体向DSF实体订阅训练评估模型所需要的网络数据,DSF实体接收AMF实体的订阅消息。
具体的,AMF实体根据步骤301接收到的规则的配置信息,确定网络环境状态的种类以及回报的种类。AMF实体向DSF实体订阅该种类的网络环境状态的网络数据,该网络数据可以是该种类的网络环境状态的具体值,也可以是该种类的网络环境状态的源数据,AMF根据网络环境状态的计算方法,通过该源数据计算出该种类的网络环境状态的具体值。例如网络环境状态的种类为导频参数,AMF实体向DSF实体订阅导频参数的数据。又例如,网络环境状态的种类为本小区和邻区的负载差,AMF实体向DSF实体订阅本小区的负载以及邻区的负载,再根据本小区和邻区的负载差的计算方法,通过订阅得到的本小区的负载和邻区的负载计算出负载差。同样的,AMF实体向DSF实体订阅该种类的回报的网络数据,该网络数据可以是该种类的回报的具体值,也可以是该种类的回报的源数据,AMF根据回报的计算方法,通过该源数据计算出该种类的回报的具体值。例如,回报的种类为流量,AMF实体向DSF实体订阅流量的数据。又例如回报的计算方式为流量/负载,AMF实体向DSF实体订阅流量和负载的数据。
步骤303、DSF实体向AMF实体发送订阅的网络数据,AMF实体接收DSF实体发送的网络数据。
步骤304、AMF实体根据订阅的网络数据训练评估模型。
步骤305、AMF实体向MEF实体发送评估模型,MEF实体接收AMF实体发送的评估模型。
具体的,AMF实体发送的评估模型中可以包括:网络环境状态的种类和/或网络环境状态的计算方法、网络参数调整动作的范围和网络参数调整动作的步长、以及网络环境状态与网络参数调整动作的对应关系。可选的,AMF实体在本步骤中还可以一并下发步骤302中获得的该种类的网络环境状态的具体值。
步骤306、MEF实体向DSF实体订阅执行评估模型所需要的网络数据,DSF实体接收MEF实体的订阅消息。
具体地,MEF实体根据步骤305中接收到的AMF实体发送的网络环境状态的种类和/或网络环境状态的计算方法,向DSF实体订阅该种类的网络环境的网络数据。
步骤307、DSF实体向MEF实体发送订阅的网络数据,MEF实体接收DSF实体发送的网络数据。
需要说明的是,若在步骤305中AMF实体向MEF实体已经发送了步骤302中获得的该种类的网络环境状态的具体值,则可省略步骤306和步骤307。
步骤308、MEF实体根据AMF实体发送的评估模型,确定NE在第一网络环境状态下执行的第一网络参数调整动作。
具体的,MEF实体根据步骤307获得的网络数据,确定当前的网络环境状态的具体值, 记为第一网络环境状态。或者将AMF实体发送的当前的网络环境状态的具体值记为第一网络环境状态。MEF实体根据AMF实体发送的评估模型,确定第一网络环境状态对应的网络参数调整动作,记为第一网络参数调整动作。第一网络参数调整动作为使得回报最大的一个动作,由评估模型来确定,可选的,在上一个网络参数的基础上按照网络参数调整动作的步长来调整。
步骤309、MEF实体将调整前的第一网络参数调整动作下发给APF实体,APF实体接收MEF实体下发的调整前第一网络参数调整动作。
步骤310、APF实体将调整后的第一网络参数调整动作下发给NE,NE接收并执行APF实体发送的调整后的第一网络参数调整动作。
具体的,上述调整前的第一网络参数调整动作为MEF实体根据评估模型确定的,APF实体根据策略确定是否需要对接收到的MEF实体发送的第一网络参数调整动作进行调整,若不需要调整,则APF实体向网元下发从MEF实体接收到的第一网络参数调整动作,若需要调整,则APF实体向网元下发调整后的第一网络参数调整动作。该调整后的第一网络参数调整动作实际上为APF实体根据策略确定的第一网络参数调整动作,调整后的第一网络参数调整动作包括实际调整过的与原来接收MEF实体发送的不同的动作,也有可能并没有调整,与原来接收MEF实体发送的相同的动作。
于此,需要说明的是,本申请实施例中,APF向其他功能实体或网元发送的第一网络参数调整动作,实际上是根据策略调整后的第一网络参数调整动作,当然,若根据策略确定第一网络参数不用调整,则APF发送的是从MEF接收到的第一网络参数调整动作。该说明适用于整个实施例。
在NE执行APF实体发送的调整后的第一网络参数调整动作之后,网络环境状态会发生改变,假设变为第二网络环境状态。并且在NE执行APF实体发送的调整后的第一网络参数调整动作之后,会产生回报,假设记为第一回报。DSF可以从NE处获取第二网络环境状态的网络数据以及产生的第一回报的数据,也可以通过其他方式来获取。
步骤311、APF实体将调整后的第一网络参数调整动作发送给AMF实体,AMF实体接收APF实体发送的调整后的第一网络参数调整动作。
需要说明的是,步骤310和步骤311之间的执行没有严格的先后顺序,可以交换执行顺序。
返回执行步骤302~步骤311。
具体的,在返回执行步骤302的过程中,AMF实体从DSF实体获取第二网络环境状态的数据,以及从DSF实体获取在NE执行APF实体发送的调整后的第一网络参数调整动作之后所产生的第一回报的数据。
在返回执行步骤304的过程中,AMF实体根据第二网络环境状态、第一回报、第一网络环境状态、和第一网络参数调整动作,来更新评估模型。第一回报是NE在第一网络环境状态下执行第一网络参数调整动作获得的立即回报,AMF实体能够根据第一回报确定NE所执行的第一网络参数调整动作是否给网络环境带来正向或者负向的影响,根据带来的正向或者负向的影响来迭代更新评估模型,使得更新后的评估模型能够获得最大的总体回报,也就是能够使得NE的行为性能达到最大,使得NE执行的网络参数调整动作能够获得更好的网络效率。
在返回执行步骤305的过程中,AMF实体向MEF实体发送更新后的评估模型,MEF 实体接收AMF实体发送的更新后的评估模型。
在返回执行步骤308的过程中,MEF实体根据AMF实体发送的更新后的评估模型,确定在第二网络环境状态下执行的第二网络参数调整动作。
在返回执行步骤309的过程中,MEF实体将第二网络参数调整动作下发给APF实体,APF实体接收MEF实体下发的第二网络参数调整动作。
在返回执行步骤310的过程中,APF实体将第二网络参数调整动作下发给NE,NE接收并执行APF实体发送的第二网络参数调整动作。
在NE执行APF实体发送的第二网络参数调整动作之后,网络环境状态会发生改变,假设变为第三网络环境状态。并且在NE执行APF实体发送的第二网络参数调整动作之后,会产生回报,假设记为第二回报。DSF可以从NE处获取第三网络环境状态的网络数据以及产生的第二回报的数据,也可以通过其他方式来获取。
在返回执行步骤311的过程中,APF实体将第二网络参数调整动作发送给AMF实体,AMF实体接收APF实体发送的第二网络参数调整动作。
接下来继续重复执行步骤302~步骤311的操作。AMF实体在每一次重复执行的过程中,迭代更新评估模型,并发送给MEF实体执行,MEF根据更新的评估模型确定NE需要执行的网络参数调整动作,MEF实体通过APF实体下发给NE,并且MEF实体将本次执行的网络参数调整动作反馈给AMF实体,以使得AMF实体迭代更新评估模型。
综上所述,将增强学习的方法应用于接入网的各个数据分析功能实体,实现了实时在线的无线网络参数调优,优化了网络性能。通过APF实体向AMF实体反馈实际下发给NE的网络参数调优动作,使得AMF实体能够在线更新评估模型,使得各个数据分析功能实体能够实现在线调优网络参数。引入本申请实施例提供的方法,使得机器学习能够应用于无线网络参数的在线调优,扩大了机器学习在无线网络中的应用范围。通过向AMF实体编排规则的配置,使得AMF可以根据灵活配置的用于增强学习方法中的参数来实现网络参数调优,解决了现网因为网络参数配置固定带来的数据不足导致不能进行调优的缺陷,实现了可灵活配置和部署的参数调优策略,避免了现有的RL应用于参数调优时新增和修改case需要通过软硬件版本升级实现。
基于图3所示的网络参数调优的方法同一发明构思,本申请实施例通过部分步骤的调整或改动,提出了几种其它可能的网络参数调优的实现方式。以下具体介绍一下其它几种可能的实现方式。
如图4所示,本申请实施例的另一种网络参数调优的方法流程具体如下所述。
步骤401~步骤404与步骤301~步骤304相同,重复之处不再赘述。
步骤405、AMF实体向MEF实体发送网络环境状态的种类和/或网络环境状态的计算方法,MEF实体接收AMF实体发送的网络环境状态的种类和/或网络环境状态的计算方法。可选的,AMF实体在本步骤中还可以一并下发步骤402中获得的该种类的网络环境状态的具体值。
步骤406、AMF实体向APF实体发送评估模型,APF实体接收AMF实体发送的评估模型。
具体的,AMF实体发送的评估模型中可以包括:网络参数调整动作的范围和网络参数调整动作的步长、以及网络环境状态与网络参数调整动作的对应关系。
步骤407、MEF实体向DSF实体订阅执行评估模型所需要的网络数据,DSF实体接收 MEF实体的订阅消息。
具体地,MEF实体根据步骤405中接收到的AMF实体发送的网络环境状态的种类和/或网络环境状态的计算方法,向DSF实体订阅该种类的网络环境的网络数据。
步骤408、DSF实体向MEF实体发送订阅的网络数据,MEF实体接收DSF实体发送的网络数据。
步骤409、MEF实体根据DSF实体发送的网络数据,确定网络环境状态的具体值。
需要说明的是,若在步骤405中AMF实体向MEF实体已经发送了网络环境状态的具体值,则可省略步骤407~步骤409。
这里MEF实体确定的网络环境状态的具体值,可以记为第一网络环境状态。
步骤410、MEF实体向APF实体发送网络环境状态的具体值。或者将AMF实体发送的当前的网络环境状态的具体值记为第一网络环境状态。
步骤411、APF实体根据网络环境状态的具体值、和AMF实体发送的评估模型,确定NE在第一网络环境状态下执行的第一网络参数调整动作。
APF实体根据AMF实体发送的评估模型,确定第一网络环境状态对应的网络参数调整动作,记为第一网络参数调整动作。第一网络参数调整动作为使得回报最大的一个动作,由评估模型来确定,可选的,在上一个网络参数的基础上按照网络参数调整动作的步长来调整。
步骤412、APF实体将第一网络参数调整动作下发给NE,NE接收并执行APF实体发送的第一网络参数调整动作。
在NE执行APF实体发送的第一网络参数调整动作之后,网络环境状态会发生改变,假设变为第二网络环境状态。并且在NE执行APF实体发送的第一网络参数调整动作之后,会产生回报,假设记为第一回报。DSF可以从NE处获取第二网络环境状态的网络数据以及产生的第一回报的数据,也可以通过其他方式来获取。
步骤413、APF实体将第一网络参数调整动作发送给AMF实体,AMF实体接收APF实体发送的第一网络参数调整动作。
需要说明的是,步骤412和步骤413之间的执行没有严格的先后顺序,可以交换执行顺序。
返回执行步骤402~步骤413。
具体的,在返回执行步骤402的过程中,AMF实体向DSF实体订阅第二网络环境状态的数据,以及订阅在NE执行APF实体发送的第一网络参数调整动作之后所产生的第一回报的数据。
在返回执行步骤404的过程中,AMF实体根据第二网络环境状态、第一回报、第一网络环境状态、和第一网络参数调整动作,来更新评估模型。第一回报是NE在第一网络环境状态下执行第一网络参数调整动作获得的立即回报,AMF实体能够根据第一回报确定NE所执行的第一网络参数调整动作是否给网络环境带来正向或者负向的影响,根据带来的正向或者负向的影响来迭代更新评估模型,使得更新后的评估模型能够获得最大的总体回报,也就是能够使得NE的行为性能达到最大,使得NE执行的网络参数调整动作能够获得更好的网络效率。
在返回执行步骤405的过程中,AMF实体向MEF实体发送第二网络环境状态的信息。
在返回执行步骤406的过程中,AMF实体向APF实体发送更新后的评估模型,APF 实体接收AMF实体发送的更新后的评估模型。
在返回执行步骤411的过程中,APF实体根据AMF实体发送的更新后的评估模型,确定在第二网络环境状态下执行的第二网络参数调整动作。
在返回执行步骤412的过程中,APF实体将第二网络参数调整动作下发给NE,NE接收并执行APF实体发送的第二网络参数调整动作。
在NE执行APF实体发送的第二网络参数调整动作之后,网络环境状态会发生改变,假设变为第三网络环境状态。并且在NE执行APF实体发送的第二网络参数调整动作之后,会产生回报,假设记为第二回报。DSF可以从NE处获取第三网络环境状态的网络数据以及产生的第二回报的数据,也可以通过其他方式来获取。
在返回执行步骤413的过程中,APF实体将第二网络参数调整动作发送给AMF实体,AMF实体接收APF实体发送的第二网络参数调整动作。
接下来继续重复执行步骤402~步骤413的操作。AMF实体在每一次重复执行的过程中,迭代更新评估模型,并发送给APF实体执行,APF实体根据更新的评估模型确定NE需要执行的网络参数调整动作,APF实体将网络参数调整动作下发给NE,并且APF实体将本次执行的网络参数调整动作反馈给AMF实体,以使得AMF实体迭代更新评估模型。
综上所述,上述实现方式将网络环境状态的计算和确定网络参数调整动作这两个过程分离开来。将增强学习的方法应用于接入网的各个数据分析功能实体,实现了实时在线的无线网络参数调优,优化了网络性能。通过APF实体向AMF实体反馈实际下发给NE的网络参数调优动作,使得AMF实体能够在线更新评估模型,使得各个数据分析功能实体能够实现在线调优网络参数。引入本申请实施例提供的方法,使得机器学习能够应用于无线网络参数的在线调优,扩大了机器学习在无线网络中的应用范围。通过向AMF实体编排规则的配置,使得AMF可以根据灵活配置的用于增强学习方法中的参数来实现网络参数调优,解决了现网因为网络参数配置固定带来的数据不足导致不能进行调优的缺陷,实现了可灵活配置和部署的参数调优策略,避免了现有的RL应用于参数调优时新增和修改case需要通过软硬件版本升级实现。
如图5所示,本申请实施例的另一种网络参数调优的方法流程具体如下所述。
步骤501、AMF实体接收规则的配置信息以及KPI保障规则的信息。
其中,规则的配置信息的具体步骤和内容如步骤301所述,在此不再赘述。
KPI保障规则的信息包括:KPI的种类、每一种类的KPI保障阈值、KPI保障规则、回退操作的内容、冲突处理策略。KPI保障规则为KPI的值不高于或者不低于KPI保障阈值。回退操作的规则为当任意一种KPI的具体值违反KPI保障规则时,采取的回退操作。回退操作的内容为:若KPI的具体值违反KPI保障规则,则将网络参数调整为默认值,或者执行与上次根据评估模型确定的网络参数调整动作相反方向的一个动作。KPI的种类包括但不限于无线资源控制(radio resource control,RRC)建立成功率、分组交换(packet switched,PS)无线接入承载(radio access bearer,RAB)、CS RAB、CS掉话率、PS掉话率、自适应多速率(adaptive multi rate,AMR)语音业务话务量。
以KPI的种类为CS掉话率为例,该类KPI的保障阈值可以为历史统计值的110%或90%或者3σ。例如KPI保障规则为:CS掉话率不高于过去一个月分布的3σ,或者CS掉话率不高于过去一周平均值的110%,或者RRC建立成功率不低于过去两周的平均值的90%。回退操作的内容可以为当CS掉话率高于过去一个月分布的3σ时,采取回退操作, 即,将网络参数调整为默认值,或者,执行与上次根据评估模型确定的网络参数调整动作相反方向的一个动作。冲突处理策略是指,当根据评估模型确定的网络参数调整动作违反KPI保障规则时,优先执行哪个结果。例如,将冲突处理策略定义为根据评估模型确定的网络参数调整动作与根据KPI保障规则确定的网络参数调整动作不一致时,以根据KPI保障规则确定的网络参数调整动作优先,即,根据评估模型确定的网络参数调整动作违反KPI保障规则时,忽略或不执行根据评估模型确定的网络参数调整动作。
步骤502、AMF实体向DSF实体订阅训练评估模型和计算KPI阈值所需要的网络数据,DSF实体接收AMF实体的订阅消息。
步骤503、DSF实体向AMF实体发送订阅的网络数据,AMF实体接收DSF实体发送的网络数据。
步骤504、AMF实体根据订阅的网络数据训练评估模型,并计算KPI阈值。
例如,以图6为例,是过去一个月的CS掉话率分布。若KPI的保障阈值为过去一个月CS掉话率分布的3σ,则AMF统计CS掉话率的均值和方差,计算出3σ的值,假设为0.12%。即得到CS掉话率的KPI阈值为0.12%。
步骤505、AMF实体向MEF实体发送评估模型,MEF实体接收AMF实体发送的评估模型。
具体的,AMF实体发送的评估模型中可以包括:网络环境状态的种类和/或网络环境状态的计算方法、网络参数调整动作的范围和网络参数调整动作的步长、网络环境状态与网络参数调整动作的对应关系、以及KPI的种类和每个KPI种类的KPI保障阈值。可选的,AMF实体在本步骤中还可以一并下发步骤502中获得的每个KPI种类的网络环境状态的具体值和每个KPI种类的KPI的具体值。
步骤506、AMF实体向APF实体下发冲突处理策略,APF实体接收AMF实体发送的冲突处理策略。
步骤507、MEF实体向DSF实体订阅执行评估模型所需要的网络数据,DSF实体接收MEF实体的订阅消息。
具体地,MEF实体根据步骤505中接收到的AMF实体发送的网络环境状态的种类和/或网络环境状态的计算方法,向DSF实体订阅该种类的网络环境的网络数据;以及,MEF实体根据步骤505中接收到AMF实体发送的KPI的种类和每个KPI种类的KPI保障阈值,向DSF实体订阅每个KPI种类的KPI数据,例如,订阅的KPI数据为当前网络CS掉话率、AMR语音业务话务量。
步骤508、DSF实体向MEF实体发送订阅的网络数据,MEF实体接收DSF实体发送的网络数据。
需要说明的是,若在步骤505中AMF实体向MEF实体已经发送了每个KPI种类的网络环境状态的具体值、以及发送了每个KPI种类的KPI的具体值,则可省略步骤507和步骤508。
步骤509、MEF实体根据AMF实体发送的评估模型确定网络参数调整动作,并判断当前KPI的具体值是否违反了KPI保障规则。
具体的,MEF实体根据DSF实体发送的网络数据,确定当前的网络环境状态的具体值,记为第一网络环境状态。或者将AMF实体在步骤505中发送的当前的网络环境状态的具体值记为第一网络环境状态。MEF实体根据DSF实体在步骤508中发送的网络数据, 确定每个KPI种类的KPI的具体值,或者MEF实体接收AMF实体发送的每个KPI种类的KPI的具体值。MEF实体根据AMF实体发送的评估模型,确定第一网络环境状态对应的网络参数调整动作。MEF实体还针对任一种类的KPI,判断KPI的具体值是否违反了KPI保障规则,即KPI的具体值是否高于或低于KPI阈值。可选的,MEF实体可以根据违反标识来指示KPI的具体值是否违反KPI保障规则。例如,若任一种类的KPI的具体值违反了KPI保障规则,则记录KPI违反标识为1,若KPI的具体值不违反KPI保障规则,则记录KPI违反标识为0。
步骤510、MEF实体将根据评估模型确定的网络参数调整动作、和KPI的具体值是否违反KPI保障规则的判断结果,下发给APF实体,APF实体接收MEF实体下发的网络参数调整动作和该判断结果。
具体的,该判断结果可以是KPI违反标识。
步骤511、APF实体根据接收到的AMF发送的冲突处理策略确定NE在第一网络环境状态下执行的第一网络参数调整动作。
具体的,若任一种类的KPI的具体值违反了KPI保障规则,则APF实体采取回退操作,例如,将网络参数调整为默认值,或者执行与上一次网络参数调整动作相反的动作。将回退操作后的动作作为第一网络参数调整动作。若任一种类的KPI的具体值均不违反KPI保障规则,则将根据评估模型确定的网络参数调整动作作为第一网络参数调整动作。
步骤512、APF实体将第一网络参数调整动作下发给NE,NE接收并执行APF实体发送的第一网络参数调整动作。
在NE执行APF实体发送的第一网络参数调整动作之后,网络环境状态会发生改变,假设变为第二网络环境状态。并且在NE执行APF实体发送的第一网络参数调整动作之后,会产生回报,假设记为第一回报。DSF可以从NE处获取第二网络环境状态的网络数据以及产生的第一回报的数据,也可以通过其他方式来获取。
步骤513、APF实体将第一网络参数调整动作发送给AMF实体,AMF实体接收APF实体发送的第一网络参数调整动作。
需要说明的是,步骤512和步骤513之间的执行没有严格的先后顺序,可以交换执行顺序。
返回执行步骤502~步骤513。
具体的,在返回执行步骤502的过程中,AMF实体向DSF实体订阅第二网络环境状态的数据,以及订阅在NE执行APF实体发送的第一网络参数调整动作之后所产生的第一回报的数据,以及订阅NE执行第一网络参数调整动作之后的KPI数据。
在返回执行步骤504的过程中,AMF实体根据第二网络环境状态、第一回报、第一网络环境状态、和第一网络参数调整动作,来更新评估模型。并且AMF实体针对任一种类的KPI,根据订阅的KPI数据计算KPI阈值和KPI的具体值。第一回报是NE在第一网络环境状态下执行第一网络参数调整动作获得的立即回报,AMF实体能够根据第一回报确定NE所执行的第一网络参数调整动作是否给网络环境带来正向或者负向的影响,根据带来的正向或者负向的影响来迭代更新评估模型,使得更新后的评估模型能够获得最大的总体回报,也就是能够使得NE的行为性能达到最大,使得NE执行的网络参数调整动作能够获得更好的网络效率。
在返回执行步骤505的过程中,AMF实体向MEF实体发送更新后的评估模型和重新 计算的KPI阈值,MEF实体接收AMF实体发送的更新后的评估模型和重新计算的KPI阈值。
在返回执行步骤507和步骤508的过程中,针对任一种类的KPI,MEF实体从DSF实体获取第一网络参数调整动作之后的KPI数据。
在返回执行步骤509的过程中,MEF实体根据AMF实体发送的更新后的评估模型,确定在第二网络环境状态下执行的网络参数调整动作。并针对任一种类的KPI,判断第一网络参数调整动作之后的KPI的具体值是否违反了KPI保障规则,其中,这里的KPI保障规则中包括重新计算的KPI阈值。MEF实体获得的第一网络参数调整动作之后的KPI的具体值可以来自于返回执行步骤505的过程中接收AMF发送的信息,也可以MEF实体通过返回执行步骤507和步骤508获取的新的KPI数据来确定。
在返回执行步骤510的过程中,MEF实体将根据更新后的评估模型确定的网络参数调整动作、以及第一网络参数调整动作之后的KPI的具体值是否违反KPI保障规则的判断结果下发给APF实体,APF实体接收MEF实体下发的该网络参数调整动作以及判断结果。该判断结果是针对任一种类的KPI的判断结果。
在返回执行步骤511的过程中,APF实体根据冲突处理策略,针对任一种类的KPI,在第一网络参数调整动作之后的KPI的具体值违反KPI保障规则时,采取回退操作,将回退操作后的动作作为第二网络参数调整动作;若针对任一种类的KPI,第一网络参数调整动作之后的KPI的具体值均不违反KPI保障规则,则将根据更新后的评估模型确定的网络参数调整动作作为第二网络参数调整动作。
在返回执行步骤512的过程中,APF实体将第二网络参数调整动作下发给NE,NE接收并执行APF实体发送的第二网络参数调整动作。
在NE执行APF实体发送的第二网络参数调整动作之后,网络环境状态会发生改变,假设变为第三网络环境状态。并且在NE执行APF实体发送的第二网络参数调整动作之后,会产生回报,假设记为第二回报。DSF可以从NE处获取第三网络环境状态的网络数据以及产生的第二回报的数据,也可以通过其他方式来获取。
在返回执行步骤513的过程中,APF实体将第二网络参数调整动作发送给AMF实体,AMF实体接收APF实体发送的第二网络参数调整动作。
接下来继续重复执行步骤502~步骤513的操作。AMF实体在每一次重复执行的过程中,迭代更新评估模型,并发送给MEF实体执行,MEF根据更新的评估模型确定NE需要执行的网络参数调整动作,MEF实体通过APF实体下发给NE,并且MEF实体将本次执行的网络参数调整动作反馈给AMF实体,以使得AMF实体迭代更新评估模型。综上所述,将增强学习的方法应用于接入网的各个数据分析功能实体,实现了实时在线的无线网络参数调优,优化了网络性能。通过APF实体向AMF实体反馈实际下发给NE的网络参数调优动作,使得AMF实体能够在线更新评估模型,使得各个数据分析功能实体能够实现在线调优网络参数。引入本申请实施例提供的方法,使得机器学习能够应用于无线网络参数的在线调优,扩大了机器学习在无线网络中的应用范围。通过向AMF实体编排规则的配置,使得AMF可以根据灵活配置的用于增强学习方法中的参数来实现网络参数调优,解决了现网因为网络参数配置固定带来的数据不足导致不能进行调优的缺陷,实现了可灵活配置和部署的参数调优策略,避免了现有的RL应用于参数调优时新增和修改case需要通过软硬件版本升级实现。进一步的,通过KPI保障策略,使得无线网络参数调优过程中 KPI不会恶化,保证了无线网络环境对KPI的要求。
基于图5所示的实现方式的构思,在图4所述的实现方式中的步骤401中,也可以进一步的优化为AMF实体接收KPI保障规则的信息;在步骤404中,AMF实体还计算KPI阈值。在步骤405中,AMF实体还会向APF实体下发冲突处理策略;在步骤411中,APF实体根据接收到的AMF发送的冲突处理策略确定NE在第一网络环境状态下执行的第一网络参数调整动作。这几个步骤的实现过程可参见图5中步骤501、步骤504、步骤506、步骤511,重复之处在此不再赘述。
如图7所示,本申请实施例的另一种网络参数调优的方法流程具体如下所述。
步骤701~步骤704与步骤301~步骤304相同,在此不再赘述。
步骤705、AMF实体确定NE在第一网络环境状态下执行的第一网络参数调整动作。
步骤706、AMF实体将第一网络参数调整动作下发给NE,NE接收并执行AMF实体发送的第一网络参数调整动作。
返回执行步骤701~步骤706。
在返回执行步骤702的过程中,AMF实体向DSF实体订阅第二网络环境状态的数据,以及订阅在NE执行APF实体发送的第一网络参数调整动作之后所产生的第一回报的数据。
在返回执行步骤704的过程中,AMF实体根据第二网络环境状态、第一回报、第一网络环境状态、和第一网络参数调整动作,来更新评估模型。第一回报是NE在第一网络环境状态下执行第一网络参数调整动作获得的立即回报,AMF实体能够根据第一回报确定NE所执行的第一网络参数调整动作是否给网络环境带来正向或者负向的影响,根据带来的正向或者负向的影响来迭代更新评估模型,使得更新后的评估模型能够获得最大的总体回报,也就是能够使得NE的行为性能达到最大,使得NE执行的网络参数调整动作能够获得更好的网络效率。
在返回执行步骤705的过程中,AMF实体根据更新后的评估模型,确定在第二网络环境状态下执行的第二网络参数调整动作。
在返回执行步骤706的过程中,AMF实体向NE发送第二网络参数调整动作,NE接收并执行AMF实体发送的第二网络参数调整动作。
在NE执行AMF实体发送的第二网络参数调整动作之后,网络环境状态会发生改变,假设变为第三网络环境状态。并且在NE执行AMF实体发送的第二网络参数调整动作之后,会产生回报,假设记为第二回报。DSF可以从NE处获取第三网络环境状态的网络数据以及产生的第二回报的数据,也可以通过其他方式来获取。
接下来继续重复执行步骤702~步骤706的操作。AMF实体在每一次重复执行的过程中,迭代更新评估模型,并根据更新的评估模型确定NE需要执行的网络参数调整动作。
在循环结束后,AMF获得最后一次更新的评估模型。接下来继续执行步骤707以及后面的步骤。
具体的,AMF可以根据环境的反馈来确定评估模型是否成熟,即是否能够根据该评估模型得到最优的网络参数。从而决定上述循环步骤的结束。
步骤707、AMF实体向MEF实体发送最后一次更新的评估模型,MEF实体接收AMF实体发送的最后一次更新的评估模型。
步骤708、MEF实体向DSF实体订阅执行最后一次更新的评估模型所需要的网络数据, DSF实体接收MEF实体的订阅消息。
步骤709、DSF实体向MEF实体发送订阅的网络数据,MEF实体接收DSF实体发送的网络数据。
步骤710、MEF实体根据DSF实体发送的网络数据,确定网络环境状态的具体值,MEF实体根据AMF实体发送的最后一次更新的评估模型,确定NE在当前网络环境状态下执行的网络参数调整动作。
该网络参数调整动作由最后一次更新的评估模型来确定。
步骤711、MEF实体将确定的网络参数调整动作下发给APF实体,APF实体接收MEF实体下发的网络参数调整动作。
步骤712、APF实体将网络参数调整动作下发给NE,NE接收并执行APF实体发送的网络参数调整动作。
重复执行上述步骤707~步骤712。
综上所述,将增强学习的方法应用于接入网的各个数据分析功能实体,实现了实时在线的无线网络参数调优,优化了网络性能。引入本申请实施例提供的方法,使得机器学习能够应用于无线网络参数的在线调优,扩大了机器学习在无线网络中的应用范围。通过向AMF实体编排规则的配置,使得AMF可以根据灵活配置的用于增强学习方法中的参数来实现网络参数调优,解决了现网因为网络参数配置固定带来的数据不足导致不能进行调优的缺陷,实现了可灵活配置和部署的参数调优策略,避免了现有的RL应用于参数调优时新增和修改case需要通过软硬件版本升级实现。进一步的,步骤702~步骤706的循环过程可以看做是评估模型训练的过程,得到最优的评估模型,步骤707~步骤711的过程可以看做是执行评估模型的过程,或者说是预测的过程。该实现方式实现了评估模型训练和预测的分离,该实现方式中增加了AMF实体与NE之间的接口,使得在评估模型训练和预测的分离时,仍然可以下发网络参数调整动作,实现在线网络参数调优。
另外,在上述图3~图7所示的各方法中,所涉及的MEF实体通过APF实体下发网络参数调整动作给NE的过程,在简单场景下,也可以更改为MEF实体不经过APF实体直接下发网络参数调整动作给NE。上述简单场景可以是不需要APF决策的场景,如调制编码方案(modulation and coding scheme,MCS)阶数选择。在上述图3~图7所示的各方法中,DSF实体存有各种类的网络环境状态的数据、各种类的回报的数据、各种类的KPI的数据,AMF实体和MEF实体通过向DSF实体订阅训练或者执行评估模型所需要的数据来完成功能。为了节省各功能实体间数据通信的消息长度以节约信令消耗,本申请实施例中,可以采用不同的标识或者索引值来区别上述不同参数的种类,AMF实体和MEF实体只需向DSF实体发送所需种类数据的标识,就可以订阅所需的数据,DSF实体可以根据接收到的标识来确定AMF实体和MEF实体订阅的数据并反馈该数据。例如,对于网络环境状态来说,用状态索引1来标识{导频功率,小区专用信道UE数,切换数}等种类的网络环境状态。
基于与上述网络参数调优方法的同一发明构思,如图8所示,本申请实施例还提供了一种网络参数调优装置800,该网络参数调优装置800用于执行本申请实施例提供的上述网络参数调优方法,该网络参数调优装置800包括:
发送单元801,用于向第二功能实体发送评估模型,评估模型用于第二功能实体确定网元在第一网络环境状态下执行的第一网络参数调整动作;
接收单元802,用于接收第一网络参数调整动作的信息;
模型训练单元803,用于根据第一网络环境状态的信息、第一网络参数调整动作的信息、网元在执行第一网络参数调整动作之后获得回报的信息、以及网元在执行第一网络参数调整动作之后的第二网络环境状态的信息,更新评估模型;
发送单元801,还用于向第二功能实体发送更新后的评估模型,更新后的评估模型用于第二功能实体确定网元在第二网络环境状态执行的第二网络参数调整动作。
可选的,接收单元802还用于:
在模型训练单元803更新评估模型之前,接收规则的配置信息,规则包括以下至少一种:网络环境状态的种类和/或计算装置、网络参数调整动作的范围和步长、回报的种类和/或计算装置;
模型训练单元803还用于,根据接收单元802接收的规则,从第三功能实体获取与网络环境状态的种类一致并和回报的种类一致的网络数据;以及,用于根据网络数据,确定网元在执行第一网络参数调整动作之后获得回报的信息、以及第二网络环境状态的信息。
可选的,若装置800为分析和建模功能AMF实体,第二功能实体为模型执行功能MEF实体,则:
发送单元801具体用于,向MEF实体发送评估模型,用于MEF实体根据评估模型确定并向自适应策略功能APF实体下发调整前的第一网络参数调整动作,由APF实体向网元下发调整后的第一网络参数调整动作;
接收单元802具体用于,接收APF实体发送的调整后的第一网络参数调整动作的信息;
模型训练单元803具体用于,根据第一网络环境状态的信息、调整后的第一网络参数调整动作的信息、网元在执行调整后的第一网络参数调整动作之后获得回报的信息、以及网元在执行调整后的第一网络参数调整动作之后的第二网络环境状态的信息,更新评估模型。
可选的,发送单元801还用于:
向MEF实体发送关键性能指标KPI的种类和每个KPI种类的KPI保障阈值,KPI种类和每个KPI种类的KPI保障阈值用于:MEF实体判断每个KPI种类的KPI的具体值是否违反KPI保障规则,并向APF实体发送判断结果,由APF实体在任一KPI种类的KPI的具体值违反KPI保障规则时采取回退操作,回退操作后的动作作为调整后的第一网络参数调整动作,KPI保障规则为不高于或者不低于KPI保障阈值。
可选的,若装置800为AMF实体,第二功能实体为APF实体,则:
发送单元801具体用于,向APF实体发送评估模型,用于APF实体根据评估模型确定第一网络参数调整动作;
接收单元802具体用于,接收APF实体发送的第一网络参数调整动作的信息。
可选的,发送单元801还用于:
向APF实体发送KPI的种类和每个KPI种类的KPI保障阈值,KPI的种类和每个KPI种类的KPI保障阈值用于:APF实体判断每个KPI种类的KPI的具体值是否违反KPI保障规则,并在任意一种KPI的具体值违反KPI保障规则时采取回退操作,回退操作后的动作作为第一网络参数调整动作。
可选的,接收单元802还用于:
在发送单元801发送KPI的种类和每个KPI种类的KPI保障阈值之前,还接收KPI 保障规则的信息,KPI保障规则的信息包括:KPI的种类、每个KPI种类的KPI保障阈值、和回退操作的内容。
基于与上述网络参数调优方法的同一发明构思,如图9所示,本申请实施例还提供了一种网络参数调优装置900,该网络参数调优装置900用于执行本申请实施例提供的上述网络参数调优方法,该网络参数调优装置900包括:
接收单元901,用于接收分析和建模功能AMF实体发送的评估模型;
确定单元902,用于根据接收单元901接收的评估模型确定网元在第一网络环境状态下执行的第一网络参数调整动作;
发送单元903,用于将第一网络参数调整动作的信息发送给自适应策略功能APF实体,由APF实体将调整后的第一网络参数调整动作发送给网元,调整后的第一网络参数调整动作包括MEF实体发送的第一网络参数调整动作以及APF实体根据策略对第一网络参数调整动作进行调整后的网络参数调整动作;
接收单元901,还用于接收AMF实体发送的更新后的评估模型,更新后的评估模型是AMF实体根据第一网络环境状态的信息、调整后的第一网络参数调整动作的信息、网元在执行调整后的第一网络参数调整动作之后获得回报的信息、以及网元在执行调整后的第一网络参数调整动作之后的第二网络环境状态的信息进行更新的;
确定单元902,还用于根据接收单元901接收到的更新后的评估模型确定网元在第二网络环境状态执行的第二网络参数调整动作。
可选的,接收单元901还用于,接收AMF实体发送的关键性能指标KPI的种类和每个KPI种类的KPI保障阈值;
确定单元902还用于,判断每个KPI种类的KPI的具体值是否违反KPI保障规则,发送单元903,还用于向APF实体发送判断结果,由APF实体在任意一种KPI的具体值违反KPI保障规则时采取回退操作,回退操作后的动作作为调整后的第一网络参数调整动作,KPI保障规则为不高于或者不低于KPI保障阈值。
基于与上述网络参数调优方法的同一发明构思,如图10所示,本申请实施例还提供了一种网络参数调优装置1000,该网络参数调优装置1000用于执行本申请实施例提供的上述网络参数调优方法,该网络参数调优装置1000包括:
接收单元1001,用于接收模型执行功能MEF实体发送的第一网络参数调整动作,第一网络参数调整动作是MEF实体根据评估模型确定的;
发送单元1002,用于将调整后的第一网络参数调整动作下发给网元,用于网元在第一网络环境状态下执行调整后的第一网络参数调整动作,调整后的第一网络参数调整动作包括MEF实体发送的第一网络参数调整动作以及根据策略对第一网络参数调整动作进行调整后的网络参数调整动作;
发送单元1002,还用于将调整后的第一网络参数调整动作的信息发送给分析和建模功能AMF实体,用于AMF实体根据第一网络环境状态的信息、调整后的第一网络参数调整动作的信息、网元在执行调整后的第一网络参数调整动作之后获得回报的信息、以及网元在执行调整后的第一网络参数调整动作之后的第二网络环境状态的信息,来更新评估模型。
可选的,接收单元1001还用于,接收MEF实体发送的KPI的具体值是否违反KPI保障规则的信息;
装置1000还包括处理单元1003,处理单元1003用于在KPI的具体值违反KPI保障规则时采取回退操作,将回退操作后的动作作为调整后的第一网络参数调整动作。
基于与上述网络参数调优方法的同一发明构思,如图11所示,本申请实施例还提供了一种网络参数调优装置1100,该网络参数调优装置1100用于执行本申请实施例提供的上述网络参数调优方法,该网络参数调优装置1100包括:
接收单元1101,用于接收分析和建模功能AMF实体发送的评估模型;
确定单元1102,用于根据接收单元1101接收的评估模型确定网元在第一网络环境状态下执行的第一网络参数调整动作,并将第一网络参数调整动作发送给网元,用于网元在第一网络环境状态下执行第一网络参数调整动作;
发送单元1103,用于将确定单元1102确定的第一网络参数调整动作的信息发送给AMF实体,用于AMF实体根据第一网络环境状态的信息、第一网络参数调整动作的信息、网元在执行第一网络参数调整动作之后获得回报的信息、以及网元在执行第一网络参数调整动作之后的第二网络环境状态的信息,来更新评估模型。
可选的,接收单元1101还用于:接收AMF实体发送的KPI的种类和每个KPI种类的KPI保障阈值;
确定单元1102还用于,根据接收单元1101接收的KPI的种类和每个KPI种类的KPI保障阈值,判断每个KPI种类的KPI的具体值是否违反KPI保障规则,在任意一种KPI的具体值违反KPI保障规则时采取回退操作,将回退操作后的动作作为第一网络参数调整动作。
基于上述方法实施例的同一发明构思,如图12所示,本申请实施例还提供了一种网络参数调优装置1200,该网络参数调优装置1200用于执行上述网络参数调优方法,该网络参数装置1200包括:收发器1201、处理器1202和存储器1203。存储器1203为可选的。处理器1202用于调用一组程序,当程序被执行时,使得处理器1202执行上述网络参数调优方法。存储器1203用于存储处理器1202执行的程序。图8中的功能模块发送单元801、接收单元802可以通过收发器1201来实现,模型训练单元803可以通过处理器1202来实现。
处理器1202可以是中央处理器(central processing unit,CPU),网络处理器(network processor,NP)或者CPU和NP的组合。
处理器1202还可以进一步包括硬件芯片。上述硬件芯片可以是专用集成电路(application-specific integrated circuit,ASIC),可编程逻辑器件(programmable logic device,PLD)或其组合。上述PLD可以是复杂可编程逻辑器件(complex programmable logic device,CPLD),现场可编程逻辑门阵列(field-programmable gate array,FPGA),通用阵列逻辑(generic array logic,GAL)或其任意组合。
存储器1203可以包括易失性存储器(volatile memory),例如随机存取存储器(random-access memory,RAM);存储器1203也可以包括非易失性存储器(non-volatile memory),例如快闪存储器(flash memory),硬盘(hard disk drive,HDD)或固态硬盘(solid-state drive,SSD);存储器1203还可以包括上述种类的存储器的组合。
基于上述方法实施例的同一发明构思,如图13所示,本申请实施例还提供了一种网络参数调优装置1300,该网络参数调优装置1300用于执行上述网络参数调优方法,该网络参数装置1300包括:收发器1301、处理器1302和存储器1303。存储器1303为可选的。 处理器1302用于调用一组程序,当程序被执行时,使得处理器1302执行上述网络参数调优方法。存储器1303用于存储处理器1302执行的程序。图9中的功能模块接收单元901、发送单元903可以通过收发器1301来实现,确定单元902可以通过处理器1302来实现。
处理器1302可以是中央处理器(central processing unit,CPU),网络处理器(network processor,NP)或者CPU和NP的组合。
处理器1302还可以进一步包括硬件芯片。上述硬件芯片可以是专用集成电路(application-specific integrated circuit,ASIC),可编程逻辑器件(programmable logic device,PLD)或其组合。上述PLD可以是复杂可编程逻辑器件(complex programmable logic device,CPLD),现场可编程逻辑门阵列(field-programmable gate array,FPGA),通用阵列逻辑(generic array logic,GAL)或其任意组合。
存储器1303可以包括易失性存储器(volatile memory),例如随机存取存储器(random-access memory,RAM);存储器1303也可以包括非易失性存储器(non-volatile memory),例如快闪存储器(flash memory),硬盘(hard disk drive,HDD)或固态硬盘(solid-state drive,SSD);存储器1303还可以包括上述种类的存储器的组合。
基于上述方法实施例的同一发明构思,如图14所示,本申请实施例还提供了一种网络参数调优装置1400,该网络参数调优装置1400用于执行上述网络参数调优方法,该网络参数装置1400包括:收发器1401、处理器1402和存储器1403。存储器1403为可选的。处理器1402用于调用一组程序,当程序被执行时,使得处理器1402执行上述网络参数调优方法。存储器1403用于存储处理器1402执行的程序。图10中的功能模块接收单元1001、发送单元1002可以通过收发器1401来实现,处理单元1003可以通过处理器1402来实现。
处理器1402可以是中央处理器(central processing unit,CPU),网络处理器(network processor,NP)或者CPU和NP的组合。
处理器1402还可以进一步包括硬件芯片。上述硬件芯片可以是专用集成电路(application-specific integrated circuit,ASIC),可编程逻辑器件(programmable logic device,PLD)或其组合。上述PLD可以是复杂可编程逻辑器件(complex programmable logic device,CPLD),现场可编程逻辑门阵列(field-programmable gate array,FPGA),通用阵列逻辑(generic array logic,GAL)或其任意组合。
存储器1403可以包括易失性存储器(volatile memory),例如随机存取存储器(random-access memory,RAM);存储器1403也可以包括非易失性存储器(non-volatile memory),例如快闪存储器(flash memory),硬盘(hard disk drive,HDD)或固态硬盘(solid-state drive,SSD);存储器1403还可以包括上述种类的存储器的组合。
基于上述方法实施例的同一发明构思,如图15所示,本申请实施例还提供了一种网络参数调优装置1500,该网络参数调优装置1500用于执行上述网络参数调优方法,该网络参数装置1500包括:收发器1501、处理器1502和存储器1503。存储器1503为可选的。处理器1502用于调用一组程序,当程序被执行时,使得处理器1502执行上述网络参数调优方法。存储器1503用于存储处理器1502执行的程序。图11中的功能模块接收单元1101、发送单元1103可以通过收发器1501来实现,确定单元1102可以通过处理器1502来实现。
处理器1502可以是中央处理器(central processing unit,CPU),网络处理器(network processor,NP)或者CPU和NP的组合。
处理器1502还可以进一步包括硬件芯片。上述硬件芯片可以是专用集成电路 (application-specific integrated circuit,ASIC),可编程逻辑器件(programmable logic device,PLD)或其组合。上述PLD可以是复杂可编程逻辑器件(complex programmable logic device,CPLD),现场可编程逻辑门阵列(field-programmable gate array,FPGA),通用阵列逻辑(generic array logic,GAL)或其任意组合。
存储器1503可以包括易失性存储器(volatile memory),例如随机存取存储器(random-access memory,RAM);存储器1503也可以包括非易失性存储器(non-volatile memory),例如快闪存储器(flash memory),硬盘(hard disk drive,HDD)或固态硬盘(solid-state drive,SSD);存储器1503还可以包括上述种类的存储器的组合。
为了实现上述图8或图12所述的装置的功能,本申请实施例还提供一种芯片,包括处理器,用于支持该装置实现上述网络参数调优方法中第一功能实体所涉及的功能。在一种可能的设计中,该芯片与存储器连接或者该芯片包括存储器,该存储器用于保存该装置必要的程序指令和数据。
为了实现上述图9或图13所述的装置的功能,本申请实施例还提供一种芯片,包括处理器,用于支持该装置实现上述网络参数调优方法中MEF所涉及的功能。在一种可能的设计中,该芯片与存储器连接或者该芯片包括存储器,该存储器用于保存该装置必要的程序指令和数据。
为了实现上述图10或图14所述的装置的功能,本申请实施例还提供一种芯片,包括处理器,用于支持该装置实现上述网络参数调优方法中APF实体所涉及的功能。在一种可能的设计中,该芯片与存储器连接或者该芯片包括存储器,该存储器用于保存该装置必要的程序指令和数据。
为了实现上述图11或图15所述的装置的功能,本申请实施例还提供一种芯片,包括处理器,用于支持该装置实现上述网络参数调优方法中APF实体所涉及的功能。在一种可能的设计中,该芯片与存储器连接或者该芯片包括存储器,该存储器用于保存该装置必要的程序指令和数据。
本申请实施例提供了一种计算机存储介质,存储有计算机程序,该计算机程序包括用于执行上述网络参数调优方法。
本申请实施例提供了一种包含指令的计算机程序产品,当其在计算机上运行时,使得计算机执行上述网络参数调优方法。
本领域内的技术人员应明白,本申请的实施例可提供为方法、***、或计算机程序产品。因此,本申请可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且,本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本申请是参照根据本申请实施例的方法、设备(***)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器,使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中,使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品,该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上,使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理,从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
尽管已描述了本申请的优选实施例,但本领域内的技术人员一旦得知了基本创造性概念,则可对这些实施例作出另外的变更和修改。所以,所附权利要求意欲解释为包括优选实施例以及落入本申请范围的所有变更和修改。
显然,本领域的技术人员可以对本申请实施例进行各种改动和变型而不脱离本申请实施例的精神和范围。这样,倘若本申请实施例的这些修改和变型属于本申请权利要求及其等同技术的范围之内,则本申请也意图包含这些改动和变型在内。
Claims (35)
- 一种网络参数调优方法,其特征在于,包括:第一功能实体向第二功能实体发送评估模型,所述评估模型用于所述第二功能实体确定网元在第一网络环境状态下执行的第一网络参数调整动作;所述第一功能实体接收所述第一网络参数调整动作的信息;所述第一功能实体根据所述第一网络环境状态的信息、所述第一网络参数调整动作的信息、所述网元在执行所述第一网络参数调整动作之后获得回报的信息、以及所述网元在执行所述第一网络参数调整动作之后的第二网络环境状态的信息,更新所述评估模型;所述第一功能实体向所述第二功能实体发送更新后的评估模型,所述更新后的评估模型用于所述第二功能实体确定所述网元在所述第二网络环境状态执行的第二网络参数调整动作。
- 如权利要求1所述的方法,其特征在于,所述第一功能实体更新所述评估模型之前,还包括:所述第一功能实体接收规则的配置信息,所述规则包括以下至少一种:网络环境状态的种类和/或计算方法、网络参数调整动作的范围和步长、回报的种类和/或计算方法;所述第一功能实体根据所述规则,从第三功能实体获取与所述网络环境状态的种类一致并和所述回报的种类一致的网络数据;所述第一功能实体根据所述网络数据,确定所述网元在执行所述第一网络参数调整动作之后获得回报的信息、以及所述第二网络环境状态的信息。
- 如权利要求1或2所述的方法,其特征在于,若所述第一功能实体为分析和建模功能AMF实体,所述第二功能实体为模型执行功能MEF实体,则:所述第一功能实体向第二功能实体发送评估模型,包括:AMF实体向MEF实体发送评估模型,用于所述MEF实体根据所述评估模型确定并向自适应策略功能APF实体下发调整前的第一网络参数调整动作,由所述APF实体向所述网元下发调整后的第一网络参数调整动作;所述第一功能实体接收所述第一网络参数调整动作的信息,包括:所述AMF实体接收所述APF实体发送的所述调整后的第一网络参数调整动作的信息;所述第一功能实体根据所述第一网络环境状态的信息、所述第一网络参数调整动作的信息、所述网元在执行所述第一网络参数调整动作之后获得回报的信息、以及所述网元在执行所述第一网络参数调整动作之后的第二网络环境状态的信息,更新所述评估模型,包括:所述第一功能实体根据所述第一网络环境状态的信息、所述调整后的第一网络参数调整动作的信息、所述网元在执行所述调整后的第一网络参数调整动作之后获得回报的信息、以及所述网元在执行所述调整后的第一网络参数调整动作之后的第二网络环境状态的信息,更新所述评估模型。
- 如权利要求3所述的方法,其特征在于,所述方法还包括:AMF实体还向MEF实体发送关键性能指标KPI的种类和每个KPI种类的KPI保障阈值,所述KPI种类和每个KPI种类的KPI保障阈值用于:所述MEF实体判断每个KPI种 类的KPI的具体值是否违反KPI保障规则,并向APF实体发送判断结果,由所述APF实体在任一KPI种类的KPI的具体值违反所述KPI保障规则时采取回退操作,所述回退操作后的动作作为所述调整后的第一网络参数调整动作,所述KPI保障规则为不高于或者不低于所述KPI保障阈值。
- 如权利要求1或2所述的方法,其特征在于,若所述第一功能实体为AMF实体,所述第二功能实体为APF实体,则:所述第一功能实体向第二功能实体发送评估模型,所述第一功能实体接收所述第一网络参数调整动作的信息,包括:AMF实体向APF实体发送评估模型,用于所述APF实体根据所述评估模型确定所述第一网络参数调整动作;所述AMF实体接收所述APF实体发送的所述第一网络参数调整动作的信息。
- 如权利要求5所述的方法,其特征在于,所述方法还包括:所述AMF实体还向所述APF实体发送KPI的种类和每个KPI种类的KPI保障阈值,所述KPI的种类和每个KPI种类的KPI保障阈值用于:所述APF实体判断每个KPI种类的KPI的具体值是否违反KPI保障规则,并在任意一种KPI的具体值违反所述KPI保障规则时采取回退操作,所述回退操作后的动作作为所述第一网络参数调整动作。
- 如权利要求4或6所述的方法,其特征在于,所述AMF实体发送KPI的种类和每个KPI种类的KPI保障阈值之前,还包括:所述AMF实体接收KPI保障规则的信息,所述KPI保障规则的信息包括:KPI的种类、每个KPI种类的KPI保障阈值、和回退操作的内容。
- 一种网络参数调优方法,其特征在于,包括:模型执行功能MEF实体接收分析和建模功能AMF实体发送的评估模型;所述MEF实体根据所述评估模型确定网元在第一网络环境状态下执行的第一网络参数调整动作,并将所述第一网络参数调整动作的信息发送给自适应策略功能APF实体,由所述APF实体将调整后的第一网络参数调整动作发送给所述网元,所述调整后的第一网络参数调整动作包括所述MEF实体发送的所述第一网络参数调整动作以及所述APF实体根据策略对所述第一网络参数调整动作进行调整后的网络参数调整动作;所述MEF实体接收所述AMF实体发送的更新后的评估模型,所述更新后的评估模型是AMF实体根据所述第一网络环境状态的信息、所述调整后的第一网络参数调整动作的信息、所述网元在执行所述调整后的第一网络参数调整动作之后获得回报的信息、以及所述网元在执行所述调整后的第一网络参数调整动作之后的第二网络环境状态的信息进行更新的;所述MEF实体根据更新后的评估模型确定所述网元在所述第二网络环境状态执行的第二网络参数调整动作。
- 如权利要求8所述的方法,其特征在于,所述方法还包括:所述MEF实体接收所述AMF实体发送的关键性能指标KPI的种类和每个KPI种类的KPI保障阈值;所述MEF实体判断每个KPI种类的KPI的具体值是否违反KPI保障规则,并向所述APF实体发送判断结果,由所述APF实体在任意一种KPI的具体值违反所述KPI保障规 则时采取回退操作,所述回退操作后的动作作为所述调整后的第一网络参数调整动作,所述KPI保障规则为不高于或者不低于所述KPI保障阈值。
- 一种网络参数调优方法,其特征在于,包括:自适应策略功能APF实体接收模型执行功能MEF实体发送的第一网络参数调整动作,所述第一网络参数调整动作是MEF实体根据评估模型确定的;所述APF实体将调整后的第一网络参数调整动作下发给网元,用于所述网元在第一网络环境状态下执行调整后的第一网络参数调整动作,所述调整后的第一网络参数调整动作包括所述MEF实体发送的所述第一网络参数调整动作以及根据策略对所述第一网络参数调整动作进行调整后的网络参数调整动作;所述APF实体将调整后的第一网络参数调整动作的信息发送给分析和建模功能AMF实体,用于AMF实体根据所述第一网络环境状态的信息、所述调整后的第一网络参数调整动作的信息、所述网元在执行所述调整后的第一网络参数调整动作之后获得回报的信息、以及所述网元在执行所述调整后的第一网络参数调整动作之后的第二网络环境状态的信息,来更新所述评估模型。
- 如权利要求10所述的方法,其特征在于,所述方法还包括:所述APF实体还接收所述MEF实体发送的KPI的具体值是否违反KPI保障规则的信息;所述APF实体在KPI的具体值违反所述KPI保障规则时采取回退操作,将所述回退操作后的动作作为所述调整后的第一网络参数调整动作。
- 一种网络参数调优方法,其特征在于,包括:自适应策略功能APF实体接收分析和建模功能AMF实体发送的评估模型;所述APF实体根据所述评估模型确定网元在第一网络环境状态下执行的第一网络参数调整动作,并将所述第一网络参数调整动作发送给所述网元,用于所述网元在第一网络环境状态下执行第一网络参数调整动作;所述APF实体将第一网络参数调整动作的信息发送给所述AMF实体,用于AMF实体根据所述第一网络环境状态的信息、所述第一网络参数调整动作的信息、所述网元在执行所述第一网络参数调整动作之后获得回报的信息、以及所述网元在执行所述第一网络参数调整动作之后的第二网络环境状态的信息,来更新所述评估模型。
- 如权利要求12所述的方法,其特征在于,所述方法还包括:APF实体接收AMF实体发送的KPI的种类和每个KPI种类的KPI保障阈值,根据所述KPI的种类和每个KPI种类的KPI保障阈值,判断每个KPI种类的KPI的具体值是否违反KPI保障规则,在任意一种KPI的具体值违反所述KPI保障规则时采取回退操作,将所述回退操作后的动作作为所述第一网络参数调整动作。
- 一种网络参数调优装置,其特征在于,包括:发送单元,用于向第二功能实体发送评估模型,所述评估模型用于所述第二功能实体确定网元在第一网络环境状态下执行的第一网络参数调整动作;接收单元,用于接收所述第一网络参数调整动作的信息;模型训练单元,用于根据所述第一网络环境状态的信息、所述第一网络参数调整动作的信息、所述网元在执行所述第一网络参数调整动作之后获得回报的信息、以及所述网元在执行所述第一网络参数调整动作之后的第二网络环境状态的信息,更新所述评估模型;所述发送单元,还用于向所述第二功能实体发送更新后的评估模型,所述更新后的评估模型用于所述第二功能实体确定所述网元在所述第二网络环境状态执行的第二网络参数调整动作。
- 如权利要求14所述的装置,其特征在于,所述接收单元还用于:在所述模型训练单元更新所述评估模型之前,接收规则的配置信息,所述规则包括以下至少一种:网络环境状态的种类和/或计算装置、网络参数调整动作的范围和步长、回报的种类和/或计算装置;所述模型训练单元还用于,根据所述接收单元接收的规则,从第三功能实体获取与所述网络环境状态的种类一致并和所述回报的种类一致的网络数据;以及,用于根据所述网络数据,确定所述网元在执行所述第一网络参数调整动作之后获得回报的信息、以及所述第二网络环境状态的信息。
- 如权利要求14或15所述的装置,其特征在于,若所述装置为分析和建模功能AMF实体,所述第二功能实体为模型执行功能MEF实体,则:所述发送单元具体用于,向MEF实体发送评估模型,用于所述MEF实体根据所述评估模型确定并向自适应策略功能APF实体下发调整前的第一网络参数调整动作,由所述APF实体向所述网元下发调整后的第一网络参数调整动作;所述接收单元具体用于,接收所述APF实体发送的所述调整后的第一网络参数调整动作的信息;所述模型训练单元具体用于,根据所述第一网络环境状态的信息、所述调整后的第一网络参数调整动作的信息、所述网元在执行所述调整后的第一网络参数调整动作之后获得回报的信息、以及所述网元在执行所述调整后的第一网络参数调整动作之后的第二网络环境状态的信息,更新所述评估模型。
- 如权利要求16所述的装置,其特征在于,所述发送单元还用于:向MEF实体发送关键性能指标KPI的种类和每个KPI种类的KPI保障阈值,所述KPI种类和每个KPI种类的KPI保障阈值用于:所述MEF实体判断每个KPI种类的KPI的具体值是否违反KPI保障规则,并向APF实体发送判断结果,由所述APF实体在任一KPI种类的KPI的具体值违反所述KPI保障规则时采取回退操作,所述回退操作后的动作作为所述调整后的第一网络参数调整动作,所述KPI保障规则为不高于或者不低于所述KPI保障阈值。
- 如权利要求14或15所述的装置,其特征在于,若所述装置为AMF实体,所述第二功能实体为APF实体,则:所述发送单元具体用于,向APF实体发送评估模型,用于所述APF实体根据所述评估模型确定所述第一网络参数调整动作;所述接收单元具体用于,接收所述APF实体发送的所述第一网络参数调整动作的信息。
- 如权利要求18所述的装置,其特征在于,所述发送单元还用于:向所述APF实体发送KPI的种类和每个KPI种类的KPI保障阈值,所述KPI的种类 和每个KPI种类的KPI保障阈值用于:所述APF实体判断每个KPI种类的KPI的具体值是否违反KPI保障规则,并在任意一种KPI的具体值违反所述KPI保障规则时采取回退操作,所述回退操作后的动作作为所述第一网络参数调整动作。
- 如权利要求17或19所述的装置,其特征在于,所述接收单元还用于:在所述发送单元发送KPI的种类和每个KPI种类的KPI保障阈值之前,还接收KPI保障规则的信息,所述KPI保障规则的信息包括:KPI的种类、每个KPI种类的KPI保障阈值、和回退操作的内容。
- 一种网络参数调优装置,其特征在于,包括:接收单元,用于接收分析和建模功能AMF实体发送的评估模型;确定单元,用于根据所述接收单元接收的评估模型确定网元在第一网络环境状态下执行的第一网络参数调整动作,并将所述第一网络参数调整动作的信息发送给自适应策略功能APF实体,由所述APF实体将调整后的第一网络参数调整动作发送给所述网元,所述调整后的第一网络参数调整动作包括所述MEF实体发送的所述第一网络参数调整动作以及所述APF实体根据策略对所述第一网络参数调整动作进行调整后的网络参数调整动作;所述接收单元,还用于接收所述AMF实体发送的更新后的评估模型,所述更新后的评估模型是AMF实体根据所述第一网络环境状态的信息、所述调整后的第一网络参数调整动作的信息、所述网元在执行所述调整后的第一网络参数调整动作之后获得回报的信息、以及所述网元在执行所述调整后的第一网络参数调整动作之后的第二网络环境状态的信息进行更新的;所述确定单元,还用于根据所述接收单元接收到的更新后的评估模型确定所述网元在所述第二网络环境状态执行的第二网络参数调整动作。
- 如权利要求21所述的装置,其特征在于,所述接收单元还用于,接收所述AMF实体发送的关键性能指标KPI的种类和每个KPI种类的KPI保障阈值;所述确定单元还用于,判断每个KPI种类的KPI的具体值是否违反KPI保障规则,并向所述APF实体发送判断结果,由所述APF实体在任意一种KPI的具体值违反所述KPI保障规则时采取回退操作,所述回退操作后的动作作为所述调整后的第一网络参数调整动作,所述KPI保障规则为不高于或者不低于所述KPI保障阈值。
- 一种网络参数调优装置,其特征在于,包括:接收单元,用于接收模型执行功能MEF实体发送的第一网络参数调整动作,所述第一网络参数调整动作是MEF实体根据评估模型确定的;发送单元,用于将调整后的第一网络参数调整动作下发给网元,用于所述网元在第一网络环境状态下执行调整后的第一网络参数调整动作,所述调整后的第一网络参数调整动作包括所述MEF实体发送的所述第一网络参数调整动作以及根据策略对所述第一网络参数调整动作进行调整后的网络参数调整动作;所述发送单元,还用于将调整后的第一网络参数调整动作的信息发送给分析和建模功能AMF实体,用于AMF实体根据所述第一网络环境状态的信息、所述调整后的第一网络参数调整动作的信息、所述网元在执行所述调整后的第一网络参数调整动作之后获得回报的信息、以及所述网元在执行所述调整后的第一网络参数调整动作之后的第二网络环境状 态的信息,来更新所述评估模型。
- 如权利要求23所述的装置,其特征在于,所述接收单元还用于,接收所述MEF实体发送的KPI的具体值是否违反KPI保障规则的信息;所述装置还包括处理单元,所述处理单元用于在KPI的具体值违反所述KPI保障规则时采取回退操作,将所述回退操作后的动作作为所述调整后的第一网络参数调整动作。
- 一种网络参数调优装置,其特征在于,包括:接收单元,用于接收分析和建模功能AMF实体发送的评估模型;确定单元,用于根据所述接收单元接收的评估模型确定网元在第一网络环境状态下执行的第一网络参数调整动作,并将所述第一网络参数调整动作发送给所述网元,用于所述网元在第一网络环境状态下执行第一网络参数调整动作;发送单元,用于将所述确定单元确定的第一网络参数调整动作的信息发送给所述AMF实体,用于AMF实体根据所述第一网络环境状态的信息、所述第一网络参数调整动作的信息、所述网元在执行所述第一网络参数调整动作之后获得回报的信息、以及所述网元在执行所述第一网络参数调整动作之后的第二网络环境状态的信息,来更新所述评估模型。
- 如权利要求25所述的装置,其特征在于,所述接收单元还用于:接收AMF实体发送的KPI的种类和每个KPI种类的KPI保障阈值;所述确定单元还用于,根据所述接收单元接收的所述KPI的种类和每个KPI种类的KPI保障阈值,判断每个KPI种类的KPI的具体值是否违反KPI保障规则,在任意一种KPI的具体值违反所述KPI保障规则时采取回退操作,将所述回退操作后的动作作为所述第一网络参数调整动作。
- 一种计算机可读存储介质,其特征在于,所述计算机存储介质中存储有计算机可读指令,当计算机读取并执行所述计算机可读指令时,使得计算机执行如权利要求1-13任意一项所述的方法。
- 一种计算机程序产品,其特征在于,当计算机读取并执行所述计算机程序产品时,使得计算机执行如权利要求1-13任意一项所述的方法。
- 一种芯片,其特征在于,所述芯片与存储器相连或者所述芯片包括所述存储器,用于读取并执行所述存储器中存储的软件程序,以实现如权利要求1-13任意一项所述的方法。
- 一种网络参数调优方法,其特征在于,包括:第一功能实体向第二功能实体发送评估模型;所述第二功能实体从所述第一功能实体接收所述评估模型,并根据所述评估模型确定网元在第一网络环境状态下执行的第一网络参数调整动作;所述第一功能实体接收所述第一网络参数调整动作的信息;所述第一功能实体根据所述第一网络环境状态的信息、所述第一网络参数调整动作的信息、所述网元在执行所述第一网络参数调整动作之后获得回报的信息、以及所述网元在执行所述第一网络参数调整动作之后的第二网络环境状态的信息,更新所述评估模型;所述第一功能实体向所述第二功能实体发送更新后的评估模型,所述更新后的评估模型用于所述第二功能实体确定所述网元在所述第二网络环境状态执行的第二网络参数调整动作。
- 如权利要求30所述的方法,其特征在于,若所述第一功能实体为分析和建模功能AMF实体,所述第二功能实体为模型执行功能MEF实体,则:所述第一功能实体向第二功能实体发送评估模型,包括:AMF实体向MEF实体发送评估模型,用于所述MEF实体根据所述评估模型确定并向自适应策略功能APF实体下发调整前的第一网络参数调整动作,由所述APF实体向所述网元下发调整后的第一网络参数调整动作;所述第一功能实体接收所述第一网络参数调整动作的信息,包括:所述AMF实体接收所述APF实体发送的所述调整后的第一网络参数调整动作的信息。
- 如权利要求30所述的方法,其特征在于,若所述第一功能实体为AMF实体,所述第二功能实体为APF实体,则:所述第一功能实体向第二功能实体发送评估模型,所述第一功能实体接收所述第一网络参数调整动作的信息,包括:AMF实体向APF实体发送评估模型,用于所述APF实体根据所述评估模型确定所述第一网络参数调整动作;所述AMF实体接收所述APF实体发送的所述第一网络参数调整动作的信息。
- 一种通信***,其特征在于,包括第一功能实体和第二功能实体;其中:所述第一功能实体,用于向第二功能实体发送评估模型;所述第二功能实体,用于从所述第一功能实体接收所述评估模型,并根据所述评估模型确定网元在第一网络环境状态下执行的第一网络参数调整动作;所述第一功能实体,还用于接收所述第一网络参数调整动作的信息;以及根据所述第一网络环境状态的信息、所述第一网络参数调整动作的信息、所述网元在执行所述第一网络参数调整动作之后获得回报的信息、以及所述网元在执行所述第一网络参数调整动作之后的第二网络环境状态的信息,更新所述评估模型;以及向所述第二功能实体发送更新后的评估模型,所述更新后的评估模型用于所述第二功能实体确定所述网元在所述第二网络环境状态执行的第二网络参数调整动作。
- 如权利要求33所述的通信***,其特征在于,若所述第一功能实体为分析和建模功能AMF实体,所述第二功能实体为模型执行功能MEF实体,则:所述第一功能实体用于:向MEF实体发送评估模型,用于所述MEF实体根据所述评估模型确定并向自适应策略功能APF实体下发调整前的第一网络参数调整动作,由所述APF实体向所述网元下发调整后的第一网络参数调整动作;所述第一功能实体用于:接收所述APF实体发送的所述调整后的第一网络参数调整动作的信息。
- 如权利要求33所述的通信***,其特征在于,若所述第一功能实体为AMF实体,所述第二功能实体为APF实体,则:所述第一功能实体用于:向APF实体发送评估模型,所述评估模型用于所述APF实体根据所述评估模型确定所述第一网络参数调整动作;所述AMF实体接收所述APF实体发送的所述第一网络参数调整动作的信息。
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
EP19750464.0A EP3739929A4 (en) | 2018-02-09 | 2019-01-22 | NETWORK PARAMETER OPTIMIZATION ADJUSTMENT METHOD AND DEVICE |
US16/985,405 US11240103B2 (en) | 2018-02-09 | 2020-08-05 | Network parameter optimization method and apparatus |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810135977.2A CN110139325B (zh) | 2018-02-09 | 2018-02-09 | 一种网络参数调优方法及装置 |
CN201810135977.2 | 2018-02-09 |
Related Child Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US16/985,405 Continuation US11240103B2 (en) | 2018-02-09 | 2020-08-05 | Network parameter optimization method and apparatus |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2019154075A1 true WO2019154075A1 (zh) | 2019-08-15 |
Family
ID=67549166
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2019/072708 WO2019154075A1 (zh) | 2018-02-09 | 2019-01-22 | 一种网络参数调优方法及装置 |
Country Status (4)
Country | Link |
---|---|
US (1) | US11240103B2 (zh) |
EP (1) | EP3739929A4 (zh) |
CN (1) | CN110139325B (zh) |
WO (1) | WO2019154075A1 (zh) |
Families Citing this family (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110139325B (zh) * | 2018-02-09 | 2021-08-13 | 华为技术有限公司 | 一种网络参数调优方法及装置 |
EP3979692A4 (en) * | 2019-07-22 | 2022-09-14 | Guangdong Oppo Mobile Telecommunications Corp., Ltd. | WIRELESS COMMUNICATIONS METHOD AND APPARATUS |
CN112752327B (zh) * | 2019-10-29 | 2023-10-20 | 上海华为技术有限公司 | 功率调节方法和接入网设备 |
CN112884159B (zh) * | 2019-11-30 | 2024-06-18 | 华为技术有限公司 | 模型更新***、模型更新方法及相关设备 |
CN114982268A (zh) * | 2020-01-21 | 2022-08-30 | 华为技术有限公司 | 一种无线参数调整方法和装置 |
CN112512077B (zh) * | 2020-12-15 | 2023-08-11 | 中国联合网络通信集团有限公司 | 一种上行速率的评估方法及装置 |
US20220190990A1 (en) * | 2020-12-16 | 2022-06-16 | Qualcomm Incorporated | Network-configured training procedure |
CN113904931A (zh) * | 2020-12-24 | 2022-01-07 | 常熟市祺迹网络信息科技有限公司 | 一种基于获客信息处理的网络参数调整方法、装置及服务器 |
US11792664B2 (en) * | 2021-06-16 | 2023-10-17 | Verizon Patent And Licensing Inc. | Virtualized architecture for system parameter identification and network component configuration with reinforcement learning |
CN116126657A (zh) * | 2021-11-12 | 2023-05-16 | 华为云计算技术有限公司 | 参数寻优方法和装置 |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101237697A (zh) * | 2008-02-26 | 2008-08-06 | 中兴通讯股份有限公司 | 无线本地环路***中实现空中数据管理功能的装置及方法 |
CN102045734A (zh) * | 2010-12-10 | 2011-05-04 | 上海百林通信软件有限公司 | 一种基于自动场景分析的td-scdma***参数方法 |
CN104158700A (zh) * | 2014-08-13 | 2014-11-19 | 福建富士通信息软件有限公司 | 基于网络环境识别的智能网络配置调整***及调整方法 |
WO2016029483A1 (zh) * | 2014-08-30 | 2016-03-03 | 华为技术有限公司 | 一种网络参数调整方法和基站设备 |
EP3136650A1 (en) * | 2015-08-31 | 2017-03-01 | Accenture Global Services Limited | Method and system for optimizing network parameters to improve customer satisfaction of network content |
Family Cites Families (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6058260A (en) * | 1995-06-12 | 2000-05-02 | The United States Of America As Represented By The Secretary Of The Army | Methods and apparatus for planning and managing a communications network |
JP2012511292A (ja) * | 2008-12-10 | 2012-05-17 | エヌイーシー ヨーロッパ リミテッド | 基盤ネットワーク上で少なくとも1つの仮想ネットワークを動作させる方法および仮想ネットワーク環境 |
WO2012154751A1 (en) * | 2011-05-08 | 2012-11-15 | Infinetics Technologies, Inc. | Flexible radix switching network |
US10327159B2 (en) * | 2014-12-09 | 2019-06-18 | Futurewei Technologies, Inc. | Autonomous, closed-loop and adaptive simulated annealing based machine learning approach for intelligent analytics-assisted self-organizing-networks (SONs) |
US10200884B2 (en) * | 2015-01-14 | 2019-02-05 | Futurewei Technologies, Inc. | Analytics-assisted, multi-agents, self-learning, self-managing, flexible and adaptive framework for intelligent SON |
US9456362B2 (en) * | 2015-01-19 | 2016-09-27 | Viavi Solutions Uk Limited | Techniques for dynamic network optimization using geolocation and network modeling |
US10445641B2 (en) * | 2015-02-06 | 2019-10-15 | Deepmind Technologies Limited | Distributed training of reinforcement learning systems |
CN110139325B (zh) * | 2018-02-09 | 2021-08-13 | 华为技术有限公司 | 一种网络参数调优方法及装置 |
EP3742669B1 (en) * | 2019-05-20 | 2023-11-08 | Nokia Technologies Oy | Machine learning in radio access networks |
-
2018
- 2018-02-09 CN CN201810135977.2A patent/CN110139325B/zh active Active
-
2019
- 2019-01-22 WO PCT/CN2019/072708 patent/WO2019154075A1/zh unknown
- 2019-01-22 EP EP19750464.0A patent/EP3739929A4/en active Pending
-
2020
- 2020-08-05 US US16/985,405 patent/US11240103B2/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101237697A (zh) * | 2008-02-26 | 2008-08-06 | 中兴通讯股份有限公司 | 无线本地环路***中实现空中数据管理功能的装置及方法 |
CN102045734A (zh) * | 2010-12-10 | 2011-05-04 | 上海百林通信软件有限公司 | 一种基于自动场景分析的td-scdma***参数方法 |
CN104158700A (zh) * | 2014-08-13 | 2014-11-19 | 福建富士通信息软件有限公司 | 基于网络环境识别的智能网络配置调整***及调整方法 |
WO2016029483A1 (zh) * | 2014-08-30 | 2016-03-03 | 华为技术有限公司 | 一种网络参数调整方法和基站设备 |
EP3136650A1 (en) * | 2015-08-31 | 2017-03-01 | Accenture Global Services Limited | Method and system for optimizing network parameters to improve customer satisfaction of network content |
Non-Patent Citations (1)
Title |
---|
See also references of EP3739929A4 |
Also Published As
Publication number | Publication date |
---|---|
CN110139325A (zh) | 2019-08-16 |
EP3739929A1 (en) | 2020-11-18 |
US20200366557A1 (en) | 2020-11-19 |
CN110139325B (zh) | 2021-08-13 |
EP3739929A4 (en) | 2021-03-10 |
US11240103B2 (en) | 2022-02-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2019154075A1 (zh) | 一种网络参数调优方法及装置 | |
US11451452B2 (en) | Model update method and apparatus, and system | |
CN109526029B (zh) | 一种业务优化方法、介质、相关装置和设备 | |
CN101502058B (zh) | 用于异步无线通信的条件请求的方法、装置、接入点和接入终端 | |
CN104137575B (zh) | 蜂窝式网络中收集以及提供多样业务信息的方法及装置 | |
EP2754271B1 (en) | Methods and apparatus for implementing a self optimizing-organizing network manager | |
US20150257147A1 (en) | Method, collaboration node, and network side device for inter-network collaboration | |
KR101104506B1 (ko) | Comp 운영을 위한 기지국 | |
JP2021503251A (ja) | PcellまたはPScellを管理するための方法および装置 | |
CN106793047B (zh) | 一种上行功率控制方法及基站 | |
WO2013044804A1 (zh) | 一种同频干扰协调方法和装置 | |
CN103856312A (zh) | 载波激活和去激活的方法、设备 | |
CN104022859A (zh) | 一种基于own的自适应流量分载技术 | |
JP7208360B2 (ja) | Mdt測定ログ送信方法、端末および可読能記憶媒体 | |
CN116325890A (zh) | 网络切片动态拥塞控制 | |
CN103828425B (zh) | 一种自适应多速率的编码速率调整方法和设备 | |
TW202337243A (zh) | 調整無線接取網路資源的方法、裝置及相關無線接取網路 | |
CN112243296B (zh) | 辅小区激活方法及装置 | |
JP6560753B2 (ja) | モバイルセルラーネットワークにおけるデータ送信電力を管理するための方法 | |
CN116017489A (zh) | 一种传输信息的方法及装置 | |
KR102301818B1 (ko) | 무선 통신 시스템에서 상향링크 커버리지 제어 방법 및 장치 | |
JP2015154184A (ja) | 無線通信システム、基地局装置及び制御装置 | |
WO2024125292A1 (zh) | 功率调整的方法与通信装置 | |
CN114205059B (zh) | 信息交互方法、装置及存储介质 | |
US20240196471A1 (en) | Apparatus, Methods, and Computer Programs |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 19750464 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
ENP | Entry into the national phase |
Ref document number: 2019750464 Country of ref document: EP Effective date: 20200814 |