CN113465139A - Refrigeration optimization method, system, storage medium and equipment - Google Patents
Refrigeration optimization method, system, storage medium and equipment Download PDFInfo
- Publication number
- CN113465139A CN113465139A CN202110592444.9A CN202110592444A CN113465139A CN 113465139 A CN113465139 A CN 113465139A CN 202110592444 A CN202110592444 A CN 202110592444A CN 113465139 A CN113465139 A CN 113465139A
- Authority
- CN
- China
- Prior art keywords
- network
- parameters
- loss function
- output
- predicted
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F24—HEATING; RANGES; VENTILATING
- F24F—AIR-CONDITIONING; AIR-HUMIDIFICATION; VENTILATION; USE OF AIR CURRENTS FOR SCREENING
- F24F11/00—Control or safety arrangements
- F24F11/62—Control or safety arrangements characterised by the type of control or by internal processing, e.g. using fuzzy logic, adaptive control or estimation of values
- F24F11/63—Electronic processing
- F24F11/64—Electronic processing using pre-stored data
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F24—HEATING; RANGES; VENTILATING
- F24F—AIR-CONDITIONING; AIR-HUMIDIFICATION; VENTILATION; USE OF AIR CURRENTS FOR SCREENING
- F24F11/00—Control or safety arrangements
- F24F11/70—Control systems characterised by their outputs; Constructional details thereof
- F24F11/80—Control systems characterised by their outputs; Constructional details thereof for controlling the temperature of the supplied air
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F24—HEATING; RANGES; VENTILATING
- F24F—AIR-CONDITIONING; AIR-HUMIDIFICATION; VENTILATION; USE OF AIR CURRENTS FOR SCREENING
- F24F2110/00—Control inputs relating to air properties
- F24F2110/10—Temperature
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F24—HEATING; RANGES; VENTILATING
- F24F—AIR-CONDITIONING; AIR-HUMIDIFICATION; VENTILATION; USE OF AIR CURRENTS FOR SCREENING
- F24F2140/00—Control inputs relating to system states
- F24F2140/50—Load
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Combustion & Propulsion (AREA)
- Mechanical Engineering (AREA)
- Chemical & Material Sciences (AREA)
- Signal Processing (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Fuzzy Systems (AREA)
- Devices That Are Associated With Refrigeration Equipment (AREA)
- Feedback Control In General (AREA)
Abstract
The invention provides a refrigeration optimization method, a refrigeration optimization system, a storage medium and equipment, wherein the method comprises the following steps: taking the historical and current state parameters and action parameters of the refrigeration system as a first input value of a Q network and an input value of a U network, taking the predicted action parameters output by the U network as a second input value of the Q network, and obtaining the predicted state parameters and the power supply use efficiency by the Q network based on the first input value and the second input value; substituting the predicted state parameters and the power supply use efficiency into a loss function constructed in the Q network, and dynamically adjusting the output of the loss function to train the Q network; dynamically adjusting the output of the Q network to carry out U network training and obtain a U network training model; and inputting the historical and current state parameters and action parameters into a U network training model to obtain predicted action parameters for refrigeration. The invention realizes the temperature control of the refrigeration system and reduces the power supply use efficiency of the refrigeration system.
Description
Technical Field
The invention relates to the technical field of refrigeration, in particular to a refrigeration optimization method, a refrigeration optimization system, a storage medium and refrigeration optimization equipment.
Background
With the continuous acceleration of the digitization process, the promotion of the intelligent technology and the popularization of the application of services such as cloud computing and social networks, the data center plays an increasingly important role. A Modular Data Center (MDC) adopts a new concept to improve the operation efficiency of a data center in order to cope with a change in servers such as cloud computing and virtualization. The method has the advantages of rapid deployment, green energy conservation and flexible expansion. With the increase of application service demands, the size of the data center is larger and larger, and meanwhile, the energy consumption problem of the data center is concerned by people more and more. Therefore, the energy-saving operation of the data center has important economic value and social value. Refrigeration energy consumption is taken as a main source of MDC energy consumption, and the energy consumption level of the MDC determines the overall PUE (power supply use efficiency) of the MDC, so that the energy consumption problem of an MDC refrigeration system needs to be optimized.
The optimization of the refrigeration energy consumption relates to a series of complex controls in the refrigeration system, such as the cooperative control of a cooling tower, a ventilation system and the like. A typical control strategy for a data center refrigeration system is implemented by setting a fixed value. For example, when the air conditioning system performs a cooling operation, a temperature value at an air outlet of the air conditioner needs to be set, and then the air conditioner starts the cooling system to deliver cold air to the data center and reduce the temperature of the data center to a specified value, so that most energy is consumed by the air conditioning system in the process. However, there is a challenge in how to select the optimal control parameters in the refrigeration system, and the traditional method mainly sets the optimal control parameters through manual experience, which has the disadvantages that the subjectivity of the setting process is strong and the quality of the parameter combination cannot be judged. In addition, some refrigeration system energy consumption optimization methods based on model estimation exist, and approximate model fitting is mainly performed based on thermal, electrical and mechanical principles, so that refrigeration system parameters can be given through models. However, the data center system is complex, the components are dynamically coupled, the mechanism modeling method is difficult, and the mechanism model under the dynamic system is difficult to provide. Therefore, the refrigeration system energy consumption optimization method based on the mechanism modeling method is difficult to give optimal refrigeration control parameters.
Disclosure of Invention
In view of the above, the present invention provides a refrigeration optimization method, system, storage medium and device, so as to solve the problems in the prior art that the manual setting of refrigeration parameters is highly subjective and the model method is difficult to provide optimal refrigeration control parameters, so as to achieve the purpose of reducing the energy consumption of the data center refrigeration system.
Based on the above purpose, the present invention provides a refrigeration optimization method, comprising the following steps:
taking the historical and current state parameters and action parameters of the refrigeration system as a first input value of a Q network and an input value of a U network, taking the predicted action parameters output by the U network as a second input value of the Q network, and obtaining the predicted state parameters and the power supply use efficiency by the Q network based on the first input value and the second input value;
substituting the predicted state parameters and the power supply use efficiency into a loss function constructed in the Q network, and dynamically adjusting the output of the loss function to train the Q network;
dynamically adjusting the output of the Q network to carry out U network training and obtain a U network training model;
and inputting the historical and current state parameters and action parameters into the U network training model to obtain predicted action parameters, and refrigerating based on the action parameters.
In some embodiments, substituting the predicted state parameters and power supply usage efficiency into a loss function constructed in the Q network, and dynamically adjusting the output of the loss function for Q network training comprises: the predicted state parameters and power usage efficiency are substituted into a loss function constructed in the Q network, and the Q network is trained by adjusting dynamic parameters of the Q network by minimizing an output value of the loss function by reducing the predicted power usage efficiency.
In some embodiments, dynamically adjusting the output of the Q network to perform U network training and obtain the U network training model comprises: and (3) minimizing the output value of the Q network to adjust the dynamic parameters of the U network so as to train the U network and obtain a U network training model.
In some embodiments, the status parameters include the IT load and ITs ambient temperature.
In some embodiments, substituting the predicted state parameter and power supply usage efficiency into a loss function constructed in the Q network further comprises: the predicted ambient temperature and power usage efficiency are substituted into a loss function constructed in the Q-network.
In some embodiments, the action parameter comprises an output temperature of a cooling device of the refrigeration system.
In some embodiments, the loss function further includes a superheat temperature setpoint.
In another aspect of the present invention, there is also provided a refrigeration optimization system, including:
the input module is configured to take the historical and current state parameters and action parameters of the refrigeration system as a first input value of the Q network and an input value of the U network, take the predicted action parameters output by the U network as a second input value of the Q network, and obtain the predicted state parameters and the power supply use efficiency by the Q network based on the first input value and the second input value;
the Q network training module is configured and used for substituting the predicted state parameters and the power supply use efficiency into a loss function constructed in the Q network and dynamically adjusting the output of the loss function so as to train the Q network;
the U network training module is configured for dynamically adjusting the output of the Q network so as to carry out U network training and obtain a U network training model; and
and the prediction module is configured for inputting the historical and current state parameters and action parameters into the U network training model to obtain predicted action parameters and refrigerating based on the action parameters.
In yet another aspect of the present invention, there is also provided a computer readable storage medium storing computer program instructions which, when executed, implement any one of the methods described above.
In yet another aspect of the present invention, a computer device is provided, which includes a memory and a processor, the memory storing a computer program, the computer program executing any one of the above methods when executed by the processor.
The invention has at least the following beneficial technical effects:
the method comprises the steps of adopting deep reinforcement learning and establishing an end-to-end refrigeration control strategy, acquiring action parameters and state parameters in a refrigeration system, limiting the requirement on the use efficiency of a power supply, establishing a loss function as an objective function by assisting environmental temperature constraint, adjusting based on the requirement on the objective function, training a Q network and a U network to obtain a trained U network training model, substituting historical and current state parameters and action parameters into the trained U network model when the predicted action parameters are required to be obtained, obtaining the predicted action parameters, realizing the temperature control of the refrigeration system, reducing the use efficiency of the power supply of the refrigeration system and reducing energy consumption.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other embodiments can be obtained by using the drawings without creative efforts.
FIG. 1 is a schematic diagram of a refrigeration optimization method provided in accordance with an embodiment of the present invention;
FIG. 2 is a schematic diagram of a data center refrigeration system according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of an improved Actor-critical network structure provided in accordance with an embodiment of the present invention;
FIG. 4 is a schematic diagram of a refrigeration optimization system provided in accordance with an embodiment of the present invention;
fig. 5 is a schematic diagram of a computer-readable storage medium for implementing a refrigeration optimization method according to an embodiment of the present invention;
fig. 6 is a schematic hardware structure diagram of a computer device for executing a refrigeration optimization method according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the following embodiments of the present invention are described in further detail with reference to the accompanying drawings.
It should be noted that all expressions using "first" and "second" in the embodiments of the present invention are used for distinguishing two non-identical entities with the same name or different parameters, and it is understood that "first" and "second" are only used for convenience of expression and should not be construed as limiting the embodiments of the present invention. Furthermore, the terms "comprises" and "comprising," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements does not include all of the other steps or elements inherent in the list.
In view of the above objects, a first aspect of an embodiment of the present invention proposes an embodiment of a refrigeration optimization method. Fig. 1 is a schematic diagram illustrating an embodiment of a refrigeration optimization method provided by the present invention. As shown in fig. 1, the embodiment of the present invention includes the following steps:
step S10, taking the history, current state parameters and action parameters of the refrigeration system as the first input value of the Q network and the input value of the U network, taking the predicted action parameters output by the U network as the second input value of the Q network, and obtaining the predicted state parameters and the power supply use efficiency by the Q network based on the first input value and the second input value;
step S20, substituting the predicted state parameters and the power supply use efficiency into a loss function constructed in the Q network, and dynamically adjusting the output of the loss function to train the Q network;
step S30, dynamically adjusting the output of the Q network to carry out U network training and obtain a U network training model;
and step S40, inputting the historical and current state parameters and action parameters into the U network training model to obtain the predicted action parameters, and cooling based on the action parameters.
In the prior art, the parameter setting of the traditional data center refrigeration system is set by depending on manual experience. For example, the air conditioning system sets temperature parameters in advance, then sets the parameters of the refrigeration system through manual experience, then starts the internal refrigerator to refrigerate, and the refrigeration system enables the ambient temperature to reach the set temperature through consuming certain electric energy. However, parameters of different subsystems of the data center are mutually influenced, the indoor environment temperature and the IT load are dynamically changed, and the manual experience setting has the defects of strong subjectivity and difficulty in judging the superiority and inferiority of parameter combination. The model-estimated refrigeration system energy consumption optimization method performs approximate modeling on refrigeration control parameters from the thermal, electrical and mechanical fields. However, the scale of the data center refrigerating system is large and complex, the coupling between systems is strong, the mechanism modeling method is difficult, and a mechanism model under a dynamic system is difficult to provide. Therefore, the refrigeration system energy consumption optimization method based on mechanism modeling has the defects that the mechanism modeling is complex and accurate optimization parameters are difficult to give. With the increase of the data center operation data volume and the continuous progress of machine learning and deep learning technologies, the optimization of the parameters of the refrigeration system based on data driving becomes a trend. Machine learning and deep learning energy consumption optimization aim at seeking optimal control parameters under different combinations to enable energy consumption to be the lowest, the method generally adopts a two-stage strategy, key characteristic parameters strongly related to the PUE (Power Usage efficiency) are selected in the first stage, characteristic reduction is carried out through data dimension reduction, and the association between the final characteristics and the PUE is established. In the second stage, the key characteristic parameters are selected dynamically by a parameter optimization method, so that the lowest overall energy consumption and the optimal PUE are realized. The parameter combinations generated by the method are tens of thousands, the optimization process is complicated, the efficiency is low, and the real-time performance of model reasoning is limited by the duration of the optimization process. Therefore, the method is difficult to be applied in a practical field.
In the prior art, an operator-critic-based reinforcement learning modeling method is also adopted, a depth certainty strategy gradient algorithm is used for modeling and designing an energy efficiency optimization automatic control scene of a data center, and the aim is to adjust the fan frequency and the cooling pump frequency of a cooling tower under control constraint so as to reduce the power of equipment on a cooling side. The method adopts an operator-critical modeling method to optimize the power of the cooling side, but in practical application, the following defects exist: firstly, the data center is large in scale and complex in structure, different refrigeration strategies adopted in different areas are different, and larger noise and abnormal values may exist in original data, so that the uncertainty of model prediction is increased; secondly, because the response period of the IT system is short, the IT system usually makes a response within milliseconds or microseconds, while the response period of the refrigeration system is long, the action issued by the current state needs a long time to reach the expected state, so that the response delay is caused, and the adjustment capability of the IT system on the dynamic system is poor.
The parameter configuration subjectivity and randomness of the artificial experience refrigeration system are high, the adaptability to the working condition parameters with indoor temperature and IT load dynamic changes is poor, and the PUE optimization is difficult to realize. The method for configuring the parameters of the refrigerating system based on the model has the characteristics of difficult mechanism model modeling and poor dynamic working condition adaptability. The parameter optimization method adopting the two-stage method has numerous parameter combinations, the optimal parameter searching process is complicated, the efficiency is low, and the real-time performance of model reasoning is limited by the duration of the optimizing process. The modeling method based on operator-critic reinforcement learning has the problems of poor anti-interference performance, hysteresis in the prediction process and incapability of realizing optimal PUE. In conclusion, the existing artificial experience configuration method, model parameter configuration method, two-stage parameter optimization method and reinforcement learning optimization method have the problem that the optimal PUE cannot be realized.
In this embodiment, both the Q network and the U network belong to a neural network, and the Q network and the U network are combined into a reinforcement learning framework of an Actor-Critic (representing an algorithm), so as to establish an end-to-end energy consumption optimization model.
The embodiment of the invention adopts deep reinforcement learning and establishes an end-to-end refrigeration control strategy, acquires action parameters and state parameters in a refrigeration system, establishes a loss function as an objective function by limiting the requirement of the use efficiency of a power supply and assisting environmental temperature constraint, adjusts the loss function based on the requirement of the objective function, trains a Q network and a U network to obtain a trained U network training model, substitutes historical and current state parameters and action parameters into the trained U network model when the predicted action parameters are required to be obtained, obtains the predicted action parameters, realizes the temperature control of the refrigeration system and reduces the use efficiency of the power supply of the refrigeration system.
In some embodiments, substituting the predicted state parameters and power supply usage efficiency into a loss function constructed in the Q network, and dynamically adjusting the output of the loss function for Q network training comprises: the predicted state parameters and power usage efficiency are substituted into a loss function constructed in the Q network, and the Q network is trained by adjusting dynamic parameters of the Q network by minimizing an output value of the loss function by reducing the predicted power usage efficiency.
In some embodiments, dynamically adjusting the output of the Q network to perform U network training and obtain the U network training model comprises: and (3) minimizing the output value of the Q network to adjust the dynamic parameters of the U network so as to train the U network and obtain a U network training model.
In some embodiments, the status parameters include the IT load and ITs ambient temperature. In this embodiment, IT represents an internet technology, and refers to an information technology developed and established on the basis of a computer technology.
In some embodiments, substituting the predicted state parameter and power supply usage efficiency into a loss function constructed in the Q network further comprises: the predicted ambient temperature and power usage efficiency are substituted into a loss function constructed in the Q-network.
In some embodiments, the action parameter comprises an output temperature of a cooling device of the refrigeration system. Specifically, the cooling device comprises a direct evaporative cooler, an indirect evaporative cooler, a direct evaporative cooling pipe, a water chiller and a water chiller cooling pipe.
In some embodiments, the loss function further includes a superheat temperature setpoint.
Fig. 2 shows a schematic diagram of a data center refrigeration system according to an embodiment of the invention. As shown in fig. 2, the number of servers in different areas of the data center is different from the system structure, so the adopted refrigeration strategies are different, and this is the caseIn the embodiment of the invention, a two-region refrigeration system which is composed of a direct evaporative cooling system and a cooling system of a cooling device as refrigeration modes is used as a monitoring object for PUE optimization. In the direct evaporative cooling system, the direct evaporative cooler and the indirect evaporative cooler perform temperature control on the data center area 1 through the direct evaporative cooling pipe. In the cooling system of the cooling device, the water chiller controls the temperature of the data center area 2 through the cooling pipe of the water chiller. Wherein the output temperature of the direct evaporative cooler is TdecThe output temperature of the indirect evaporative cooler is TiecThe output temperature of the direct evaporative cooling tube is TdcThe output temperature of the water chiller is TcwThe output temperature of the cooling pipe of the water chilling unit is Tcc。
In the PUE optimization of the multi-zone refrigeration system, the temperature variable of the refrigeration system is used as an action parameter, and the adopted five action parameters are respectively as follows: t isiec、Tdec、Tdc、Tcw、Tcc. At ambient temperature TenvAnd IT load LitlAs a status parameter. Five action parameters in the refrigeration system energy efficiency optimization model are control variables, and two state parameters are observation variables, so that system modeling is carried out. The constructed energy efficiency optimization model is shown as a formula (1).
In the formula (1), T1,T2The temperature of the IT load air outlets in the area 1 and the area 2, namely the ambient temperature; t isSIs a superheat temperature setpoint; xiPUEThe PUE value is obtained. Temperature T through IT load air outlet in data center area 11Temperature T of air outlet of IT load in area 22And building a loss function L by the PUE of the system, wherein the goal of system optimization is to control five action parameters to enable different areas of the data center to reach a specified temperature, and simultaneously enable the PUE of the system to be the lowest, namely, the loss function L is minimized by carrying out parameter optimization within the range of the five action parameters.
FIG. 3 showsIn the schematic diagram of the improved Actor-critical Network structure of the embodiment of the invention, the output in the penultimate layer (y) in the Q Network (Q-Network) is the environmental temperature parameter TenvAnd PUE data ξPUEBy adopting the network, not only can the complexity of network training be reduced, but also the output value of the second last layer can visually display the predicted temperature and the PUE value. The ambient temperature and PUE values of the network output of the penultimate layer are substituted into equation (1) to calculate the loss function. In a U-Network (mu-Network), the last layer (y) of the Q-Network is passedr) The output values are minimized for U-network training. The last layer of the U-network represents the output predicted action parameter a.
In other embodiments, based on the above training method, in order to improve the anti-interference capability of the algorithm model, a cyclic decision method based on variable step length of historical data is adopted. Specifically, the step length of the selected historical data is recorded as delta, and the historical state parameter is recorded as st-δ,st-δ+2,...st-1(ii) a The historical motion parameter is at-δ,at-δ+2,...at-1. Over time, historical data is input in a circulating mode, and noise suppression of the network can be enhanced. In order to optimize the matching between the state parameters and the motion parameters and alleviate the problem of motion parameter hysteresis, the duration of the observation state is set to be tau, that is, the prediction step is reduced, for example, tau is 1min, that is, the motion a after 1 minute is predictedt+1。
In still other embodiments, in order to ensure the model training precision and generalization performance in practical applications, the present invention uses different levels of training verification data sets to perform cross verification during the offline training phase, for example, data partitioning is performed by a training verification ratio of 8:2, and model parameters are saved by verifying the optimal precision in the data sets in an Actor network (μ -network) and a critical network (Q-network). Initializing a weight W at the beginning of the Actor-Critic network training, generating a random weight by a normal distribution function of a sigma standard deviation, and limiting sigma in a smaller range so as to improve the stability of model training; the learning rates of Q-network and mu-network are updated by dynamically adjusting the gradient by a staged step type learning rate.
The embodiment of the invention can effectively adapt to the influence of parameter changes such as dynamic environment, IT load and the like on control parameters by adopting a deep learning method, has strong adaptability to online adjustment, and can dynamically adjust the running state of a refrigeration system; the model is optimized through a series of optimization algorithms such as historical data cyclic decision and the like, so that the predicted action parameters are further optimized, and the robustness of the model (the capability of representing the survival of the system under abnormal and dangerous conditions) is improved; the energy consumption of the refrigeration system is effectively reduced, the PUE of the system and the operation cost of the data center are reduced, and the economic benefit of enterprises is improved.
In a second aspect of the embodiments of the present invention, a refrigeration optimization system is also provided. Fig. 4 is a schematic diagram of an embodiment of the refrigeration optimization system provided by the present invention. As shown in fig. 4, a refrigeration optimization system includes: the input module 10 is configured to use historical and current state parameters and action parameters of the refrigeration system as first input values of a Q network and input values of a U network, use predicted action parameters output by the U network as second input values of the Q network, and obtain predicted state parameters and power supply use efficiency by the Q network based on the first input values and the second input values; a Q network training module 20 configured to substitute the predicted state parameters and power supply use efficiency into a loss function constructed in the Q network, and dynamically adjust an output of the loss function to perform Q network training; a U network training module 30 configured to dynamically adjust an output of the Q network to perform U network training and obtain a U network training model; and a prediction module 40 configured to input the historical and current state parameters and the motion parameters into the U network training model to obtain predicted motion parameters and perform cooling based on the motion parameters.
The refrigeration optimization system provided by the embodiment of the invention adopts deep reinforcement learning and establishes an end-to-end refrigeration control strategy, trains a Q network and a U network by acquiring action parameters and state parameters in the refrigeration system, limiting the requirements on the use efficiency of a power supply and establishing a loss function as an objective function by assisting environmental temperature constraint, and adjusting based on the requirements on the objective function to obtain a trained U network training model, and substitutes historical and current state parameters and action parameters into the trained U network model when the predicted action parameters are required to be obtained, so as to obtain the predicted action parameters, realize the temperature control of the refrigeration system, and reduce the use efficiency of the power supply of the refrigeration system.
In some embodiments, Q-network training module 20 is further configured to substitute the predicted state parameters and power usage efficiency into a loss function constructed in the Q-network and train the Q-network by adjusting dynamic parameters of the Q-network by minimizing an output value of the loss function by reducing the predicted power usage efficiency.
In some embodiments, the U-network training module 30 is further configured to adjust the dynamic parameters of the U-network by minimizing the output values of the Q-network so that the U-network is trained and the U-network training model is obtained.
In some embodiments, the status parameters include the IT load and ITs ambient temperature.
In some embodiments, Q-network training module 20 further includes a loss function module configured to substitute the predicted ambient temperature and power usage efficiency into a loss function constructed in the Q-network.
In some embodiments, the action parameter comprises an output temperature of a cooling device of the refrigeration system.
In some embodiments, the loss function further includes a superheat temperature setpoint.
In a third aspect of the embodiment of the present invention, a computer-readable storage medium is further provided, and fig. 5 is a schematic diagram of a computer-readable storage medium implementing a refrigeration optimization method according to an embodiment of the present invention. As shown in fig. 5, the computer-readable storage medium 3 stores computer program instructions 31, the computer program instructions 31 implementing the method of any one of the above embodiments when executed.
It is to be understood that all embodiments, features and advantages set forth above with respect to the refrigeration optimization method according to the invention apply equally, without conflict therewith, to the refrigeration optimization system and to the storage medium according to the invention.
In a fourth aspect of the embodiments of the present invention, there is further provided a computer device, including a memory 402 and a processor 401, where the memory stores a computer program, and the computer program, when executed by the processor, implements the method of any one of the above embodiments.
Fig. 6 is a schematic hardware structure diagram of an embodiment of a computer device for executing a refrigeration optimization method according to the present invention. Taking the computer device shown in fig. 6 as an example, the computer device includes a processor 401 and a memory 402, and may further include: an input device 403 and an output device 404. The processor 401, the memory 402, the input device 403 and the output device 404 may be connected by a bus or other means, and fig. 6 illustrates an example of a connection by a bus. The input device 403 may receive entered numeric or character information and generate key signal inputs relating to user settings and function controls of the refrigeration optimization system. The output device 404 may include a display device such as a display screen.
The memory 402, which is a non-volatile computer-readable storage medium, may be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as program instructions/modules corresponding to the refrigeration optimization method in the embodiments of the present application. The memory 402 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created by use of the refrigeration optimization method, and the like. Further, the memory 402 may include high speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid state storage device. In some embodiments, memory 402 may optionally include memory located remotely from processor 401, which may be connected to local modules via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The processor 401 executes various functional applications of the server and data processing by running nonvolatile software programs, instructions and modules stored in the memory 402, that is, implements the refrigeration optimization method of the above method embodiment.
Finally, it should be noted that the computer-readable storage medium (e.g., memory) herein can be either volatile memory or nonvolatile memory, or can include both volatile and nonvolatile memory. By way of example, and not limitation, nonvolatile memory can include Read Only Memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM), which can act as external cache memory. By way of example and not limitation, RAM is available in a variety of forms such as synchronous RAM (DRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), and Direct Rambus RAM (DRRAM). The storage devices of the disclosed aspects are intended to comprise, without being limited to, these and other suitable types of memory.
Those of skill would further appreciate that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the disclosure herein may be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as software or hardware depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the disclosed embodiments of the present invention.
The various illustrative logical blocks, modules, and circuits described in connection with the disclosure herein may be implemented or performed with the following components designed to perform the functions herein: a general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination of these components. A general purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. A processor may also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP, and/or any other such configuration.
The foregoing is an exemplary embodiment of the present disclosure, but it should be noted that various changes and modifications could be made herein without departing from the scope of the present disclosure as defined by the appended claims. The functions, steps and/or actions of the method claims in accordance with the disclosed embodiments described herein need not be performed in any particular order. Furthermore, although elements of the disclosed embodiments of the invention may be described or claimed in the singular, the plural is contemplated unless limitation to the singular is explicitly stated.
It should be understood that, as used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly supports the exception. It should also be understood that "and/or" as used herein is meant to include any and all possible combinations of one or more of the associated listed items. The numbers of the embodiments disclosed in the embodiments of the present invention are merely for description, and do not represent the merits of the embodiments.
Those of ordinary skill in the art will understand that: the discussion of any embodiment above is meant to be exemplary only, and is not intended to intimate that the scope of the disclosure, including the claims, of embodiments of the invention is limited to these examples; within the idea of an embodiment of the invention, also technical features in the above embodiment or in different embodiments may be combined and there are many other variations of the different aspects of the embodiments of the invention as described above, which are not provided in detail for the sake of brevity. Therefore, any omissions, modifications, substitutions, improvements, and the like that may be made without departing from the spirit and principles of the embodiments of the present invention are intended to be included within the scope of the embodiments of the present invention.
Claims (10)
1. A refrigeration optimization method, comprising the steps of:
taking the historical and current state parameters and action parameters of the refrigeration system as a first input value of a Q network and an input value of a U network, taking the predicted action parameters output by the U network as a second input value of the Q network, and obtaining the predicted state parameters and the power supply use efficiency by the Q network based on the first input value and the second input value;
substituting the predicted state parameters and the power supply use efficiency into a loss function constructed in the Q network, and dynamically adjusting the output of the loss function to train the Q network;
dynamically adjusting the output of the Q network to carry out U network training and obtain a U network training model;
and inputting historical and current state parameters and action parameters into the U network training model to obtain predicted action parameters, and refrigerating based on the action parameters.
2. The method of claim 1, wherein substituting the predicted state parameters and power supply usage efficiency into a loss function constructed in a Q network, and dynamically adjusting an output of the loss function for Q network training comprises:
substituting the predicted state parameters and the power supply use efficiency into a loss function constructed in the Q network, and minimizing the output value of the loss function by reducing the predicted power supply use efficiency to adjust the dynamic parameters of the Q network so as to train the Q network.
3. The method of claim 1, wherein dynamically adjusting the output of the Q network to perform U network training and obtain a U network training model comprises:
and (3) minimizing the output value of the Q network to adjust the dynamic parameters of the U network so as to train the U network and obtain a U network training model.
4. The method of claim 1, wherein the status parameters include IT load and ITs ambient temperature.
5. The method of claim 4, wherein substituting the predicted state parameters and power supply usage efficiency into a loss function constructed in a Q network further comprises:
substituting the predicted ambient temperature and power supply usage efficiency into a loss function constructed in a Q-network.
6. The method of claim 1, wherein the action parameter comprises an output temperature of a cooling device of the refrigeration system.
7. The method of claim 1, wherein the loss function further comprises a superheat temperature setpoint.
8. A refrigeration optimization system, comprising:
the input module is configured to use historical and current state parameters and action parameters of the refrigeration system as first input values of a Q network and input values of a U network, use predicted action parameters output by the U network as second input values of the Q network, and obtain predicted state parameters and power supply use efficiency through the Q network based on the first input values and the second input values;
the Q network training module is configured to substitute the predicted state parameters and the power supply use efficiency into a loss function constructed in a Q network, and dynamically adjust the output of the loss function to perform Q network training;
the U network training module is configured for dynamically adjusting the output of the Q network so as to carry out U network training and obtain a U network training model; and
and the prediction module is configured to input historical and current state parameters and action parameters into the U network training model to obtain predicted action parameters and perform refrigeration based on the action parameters.
9. A computer-readable storage medium, characterized in that computer program instructions are stored which, when executed, implement the method according to any one of claims 1-7.
10. A computer device comprising a memory and a processor, characterized in that the memory has stored therein a computer program which, when executed by the processor, performs the method according to any one of claims 1-7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110592444.9A CN113465139B (en) | 2021-05-28 | 2021-05-28 | Refrigeration optimization method, system, storage medium and equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110592444.9A CN113465139B (en) | 2021-05-28 | 2021-05-28 | Refrigeration optimization method, system, storage medium and equipment |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113465139A true CN113465139A (en) | 2021-10-01 |
CN113465139B CN113465139B (en) | 2022-11-08 |
Family
ID=77871631
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110592444.9A Active CN113465139B (en) | 2021-05-28 | 2021-05-28 | Refrigeration optimization method, system, storage medium and equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113465139B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114838496A (en) * | 2022-04-22 | 2022-08-02 | 江苏风神空调集团股份有限公司 | Air conditioner silencer performance detection method based on artificial intelligence |
CN115388586A (en) * | 2022-08-24 | 2022-11-25 | 南栖仙策(南京)高新技术有限公司 | Refrigeration control model, device, electronic equipment and storage medium |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20140003113A (en) * | 2012-06-29 | 2014-01-09 | 한국전자통신연구원 | System for controlling internal temperature of internet data center and method thereof |
KR20150067852A (en) * | 2013-12-10 | 2015-06-19 | 한국전자통신연구원 | Airflow management system for data center and management method of the same |
KR20180082764A (en) * | 2017-01-11 | 2018-07-19 | 인하대학교 산학협력단 | Apparatus for thermal environment prediction based on machine learning and method thereof |
CN109478254A (en) * | 2016-05-20 | 2019-03-15 | 渊慧科技有限公司 | Neural network is trained using composition gradient |
WO2019227273A1 (en) * | 2018-05-28 | 2019-12-05 | Alibaba Group Holding Limited | Hierarchical concept based neural network model for data center power usage effectiveness prediction |
CN111351180A (en) * | 2020-03-06 | 2020-06-30 | 上海外高桥万国数据科技发展有限公司 | System and method for realizing energy conservation and temperature control of data center by applying artificial intelligence |
CN111582604A (en) * | 2020-05-20 | 2020-08-25 | 中国工商银行股份有限公司 | Data processing method and device, electronic device, and medium |
CN111811110A (en) * | 2020-08-28 | 2020-10-23 | 创新奇智(南京)科技有限公司 | Control method and device of refrigerating unit, electronic equipment and storage medium |
WO2020227983A1 (en) * | 2019-05-15 | 2020-11-19 | Alibaba Group Holding Limited | Hybrid-learning neural network architecture |
-
2021
- 2021-05-28 CN CN202110592444.9A patent/CN113465139B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20140003113A (en) * | 2012-06-29 | 2014-01-09 | 한국전자통신연구원 | System for controlling internal temperature of internet data center and method thereof |
KR20150067852A (en) * | 2013-12-10 | 2015-06-19 | 한국전자통신연구원 | Airflow management system for data center and management method of the same |
CN109478254A (en) * | 2016-05-20 | 2019-03-15 | 渊慧科技有限公司 | Neural network is trained using composition gradient |
KR20180082764A (en) * | 2017-01-11 | 2018-07-19 | 인하대학교 산학협력단 | Apparatus for thermal environment prediction based on machine learning and method thereof |
WO2019227273A1 (en) * | 2018-05-28 | 2019-12-05 | Alibaba Group Holding Limited | Hierarchical concept based neural network model for data center power usage effectiveness prediction |
WO2020227983A1 (en) * | 2019-05-15 | 2020-11-19 | Alibaba Group Holding Limited | Hybrid-learning neural network architecture |
CN111351180A (en) * | 2020-03-06 | 2020-06-30 | 上海外高桥万国数据科技发展有限公司 | System and method for realizing energy conservation and temperature control of data center by applying artificial intelligence |
CN111582604A (en) * | 2020-05-20 | 2020-08-25 | 中国工商银行股份有限公司 | Data processing method and device, electronic device, and medium |
CN111811110A (en) * | 2020-08-28 | 2020-10-23 | 创新奇智(南京)科技有限公司 | Control method and device of refrigerating unit, electronic equipment and storage medium |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114838496A (en) * | 2022-04-22 | 2022-08-02 | 江苏风神空调集团股份有限公司 | Air conditioner silencer performance detection method based on artificial intelligence |
CN114838496B (en) * | 2022-04-22 | 2024-02-23 | 北京百车宝科技有限公司 | Air conditioner muffler performance detection method based on artificial intelligence |
CN115388586A (en) * | 2022-08-24 | 2022-11-25 | 南栖仙策(南京)高新技术有限公司 | Refrigeration control model, device, electronic equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN113465139B (en) | 2022-11-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113465139B (en) | Refrigeration optimization method, system, storage medium and equipment | |
JP7016407B2 (en) | Energy optimization of the cooling unit through smart supply air temperature setpoint control | |
CN112577161B (en) | Air conditioner energy consumption model training method and air conditioner system control method | |
CN102213475B (en) | Adaptive management method for power consumption of data center | |
CN112050397A (en) | Method and system for regulating and controlling temperature of machine room | |
CN110410942B (en) | Energy-saving optimization control method and system for cold and heat source machine room | |
US8825184B2 (en) | Multivariable optimization of operation of vapor compression systems | |
WO2015074573A1 (en) | Method, and device for controlling the output of the air volume and memory medium | |
Koehler et al. | Building temperature distributed control via explicit MPC and “Trim and Respond” methods | |
CN110609474B (en) | Data center energy efficiency optimization method based on reinforcement learning | |
CN104833154B (en) | Chilled water loop control method based on fuzzy PID and neural internal model | |
CN101655272A (en) | Energy-saving control management system of network central air conditioner and method thereof | |
CN112283890A (en) | Cold and heat quantity control method and device suitable for building heating and ventilation equipment monitoring system | |
CN112906966B (en) | Method, system, medium and equipment for optimizing load of central air conditioner water chilling unit | |
CN112413762B (en) | Parameter optimization method and system for cooling water system of refrigerating room | |
CN115682324A (en) | Energy-saving optimization control method, device, equipment and storage medium for central air-conditioning system | |
Petrie et al. | Energy efficient control methods of HVAC systems for smart campus | |
CN109612055A (en) | A kind of feedforward fuzzy control method of air-conditioner water system | |
CN116907036A (en) | Deep reinforcement learning water chilling unit control method based on cold load prediction | |
CN115049141A (en) | Control method and control device for refrigeration system and refrigeration system | |
CN107560065A (en) | A kind of intelligent air-conditioner control method and server | |
KR102183484B1 (en) | Air-conditioner of predicting air conditioning load using artificial intelligence, cloud server, and method of operating thereof | |
CN111415036A (en) | Load optimization distribution method for parallel connection cold machines of central air-conditioning system | |
CN112963950B (en) | Air conditioner control method and device, storage medium and air conditioner | |
KR20180138372A (en) | Method for controlling of chillers optimally using an online/offline hybrid machine learning models |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |