CN118017564A - Energy storage method based on open source hong Meng system - Google Patents
Energy storage method based on open source hong Meng system Download PDFInfo
- Publication number
- CN118017564A CN118017564A CN202410419230.5A CN202410419230A CN118017564A CN 118017564 A CN118017564 A CN 118017564A CN 202410419230 A CN202410419230 A CN 202410419230A CN 118017564 A CN118017564 A CN 118017564A
- Authority
- CN
- China
- Prior art keywords
- data
- energy storage
- node
- edge node
- network
- 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
- 238000004146 energy storage Methods 0.000 title claims abstract description 182
- 238000000034 method Methods 0.000 title claims abstract description 80
- 238000010801 machine learning Methods 0.000 claims abstract description 20
- 238000004891 communication Methods 0.000 claims abstract description 18
- 238000012549 training Methods 0.000 claims abstract description 15
- 230000004044 response Effects 0.000 claims abstract description 12
- 238000013500 data storage Methods 0.000 claims abstract description 7
- 238000013467 fragmentation Methods 0.000 claims abstract description 6
- 238000006062 fragmentation reaction Methods 0.000 claims abstract description 6
- 238000004422 calculation algorithm Methods 0.000 claims description 49
- 230000005540 biological transmission Effects 0.000 claims description 45
- 238000009826 distribution Methods 0.000 claims description 43
- 230000036541 health Effects 0.000 claims description 30
- 239000013598 vector Substances 0.000 claims description 26
- 238000003860 storage Methods 0.000 claims description 23
- 238000004458 analytical method Methods 0.000 claims description 22
- 238000012545 processing Methods 0.000 claims description 21
- 238000013528 artificial neural network Methods 0.000 claims description 20
- 238000000605 extraction Methods 0.000 claims description 20
- 230000008859 change Effects 0.000 claims description 19
- 239000002245 particle Substances 0.000 claims description 18
- 238000013527 convolutional neural network Methods 0.000 claims description 17
- 238000004088 simulation Methods 0.000 claims description 17
- 238000007726 management method Methods 0.000 claims description 15
- 238000013507 mapping Methods 0.000 claims description 14
- 125000004122 cyclic group Chemical group 0.000 claims description 13
- 230000007246 mechanism Effects 0.000 claims description 13
- 238000005457 optimization Methods 0.000 claims description 13
- 235000018185 Betula X alpestris Nutrition 0.000 claims description 10
- 235000018212 Betula X uliginosa Nutrition 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 10
- 230000007774 longterm Effects 0.000 claims description 10
- 238000005192 partition Methods 0.000 claims description 9
- 238000007781 pre-processing Methods 0.000 claims description 9
- 230000009466 transformation Effects 0.000 claims description 8
- 230000010354 integration Effects 0.000 claims description 7
- 238000007619 statistical method Methods 0.000 claims description 7
- 239000000284 extract Substances 0.000 claims description 6
- 238000000638 solvent extraction Methods 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 2
- 238000011425 standardization method Methods 0.000 claims description 2
- 230000006870 function Effects 0.000 description 23
- 230000008569 process Effects 0.000 description 19
- 239000000306 component Substances 0.000 description 18
- 238000006243 chemical reaction Methods 0.000 description 9
- 230000000694 effects Effects 0.000 description 8
- 238000001514 detection method Methods 0.000 description 7
- 230000011218 segmentation Effects 0.000 description 7
- 230000002776 aggregation Effects 0.000 description 6
- 238000004220 aggregation Methods 0.000 description 6
- 238000012544 monitoring process Methods 0.000 description 6
- 239000008186 active pharmaceutical agent Substances 0.000 description 5
- 230000000306 recurrent effect Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 230000000007 visual effect Effects 0.000 description 4
- 230000015556 catabolic process Effects 0.000 description 3
- 238000012790 confirmation Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 3
- 238000006731 degradation reaction Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 230000018109 developmental process Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000011049 filling Methods 0.000 description 3
- 238000009434 installation Methods 0.000 description 3
- 230000015654 memory Effects 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 230000032683 aging Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000004140 cleaning Methods 0.000 description 2
- 238000007906 compression Methods 0.000 description 2
- 230000006835 compression Effects 0.000 description 2
- 238000013523 data management Methods 0.000 description 2
- 238000013499 data model Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000007599 discharging Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 230000008520 organization Effects 0.000 description 2
- 238000004806 packaging method and process Methods 0.000 description 2
- 230000002035 prolonged effect Effects 0.000 description 2
- 238000012163 sequencing technique Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 241001178520 Stomatepia mongo Species 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000033228 biological regulation Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 238000007621 cluster analysis Methods 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000013144 data compression Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000003111 delayed effect Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 230000001066 destructive effect Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 238000012407 engineering method Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000005562 fading Methods 0.000 description 1
- 238000005111 flow chemistry technique Methods 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000012806 monitoring device Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 230000002085 persistent effect Effects 0.000 description 1
- 238000006116 polymerization reaction Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000007639 printing Methods 0.000 description 1
- 238000010223 real-time analysis Methods 0.000 description 1
- 230000006798 recombination Effects 0.000 description 1
- 238000005215 recombination Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000008521 reorganization Effects 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 238000013515 script Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000000528 statistical test Methods 0.000 description 1
- 230000001502 supplementing effect Effects 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
- 230000001960 triggered effect Effects 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The application discloses an energy storage method based on an open source hong Meng system, which relates to the technical field of energy storage and comprises the following steps: obtaining geographic coordinates of energy storage equipment and a network connection topological structure of the energy storage equipment and a central node; constructing a distributed architecture comprising a central node and a plurality of edge nodes; deploying distributed data storage on the constructed distributed architecture by utilizing a distributed file system of the hong Mongolian system; collecting real-time data; extracting features through a parallelized machine learning model; transmitting the extracted characteristic data through fragmentation of a hong Monte system; the edge node acquires historical state data of the energy storage device from a distributed database of the central node; training a prediction model for predicting the capacity of the energy storage device using a machine learning model of the hong Mongolian system; generating a control instruction of the energy storage device; the control instructions are transmitted using an industrial timing communication protocol based on the OPCUA standard. Aiming at the problem of high response delay in an energy storage system in the prior art, the application reduces the response delay.
Description
Technical Field
The application relates to the technical field of energy storage, in particular to an energy storage method based on an open source hong Meng system.
Background
Along with the rapid increase of the specific gravity of renewable energy sources, the energy storage technology is widely applied as a key link of peak regulation and valley filling. In order to realize efficient collaborative management of large-scale energy storage equipment, the construction of a flexible and extensible energy storage management system becomes an important direction of technical development.
The existing energy storage management system generally adopts a centralized architecture, and core business logic in the system is concentrated in a central node for processing, so that the calculation and storage pressure of the central node is overlarge, and the expansibility of the system is limited. The mass equipment data is concentrated to the central node, the network transmission congestion is serious, and the response delay is obvious. A large amount of equipment data cannot be processed in real time, and the control effect is affected. The single point failure risk is larger and the reliability is not high. The interaction between the central node and the equipment is time-consuming and cannot be controlled in real time.
In the related art, for example, in CN117097766a, a data monitoring method and a data monitoring device for an energy storage system are provided, which are applied to a data monitoring system, where the data monitoring system includes a cloud platform and an energy storage system, and the energy storage system includes an energy storage device and an edge device, so that data transmission efficiency is improved. The method comprises the following steps: the cloud platform initiates a message queue telemetry transmission MQTT connection request to an edge device, and the edge device and an energy storage device keep in communication so as to establish MQTT connection with the energy storage system; the cloud platform confirms to establish the MQTT connection under the condition of receiving connection confirmation information from the edge equipment; the cloud platform sends a subscription request to the edge equipment, wherein the subscription request is used for acquiring energy storage data of the energy storage equipment; the cloud platform receives energy storage data sent by edge equipment, wherein the energy storage data are acquired by the edge equipment from the energy storage equipment. But in the scheme, the MQTT is a lightweight publish-subscribe mode protocol, so that the message transmission reliability is ensured to be weak. Network instability can lead to message loss, thereby increasing response delay.
Disclosure of Invention
1. Technical problem to be solved
Aiming at the problem of high response delay in an energy storage system in the prior art, the application provides an energy storage method based on an open source hong Meng system, and the response delay is reduced through the efficient management of mass energy storage devices by a cooperative center node and an edge node through the distributed architecture of the open source hong Meng system.
2. Technical proposal
The aim of the application is achieved by the following technical scheme.
The embodiment of the specification provides an energy storage method based on an open source hong Meng system, which comprises the following steps: obtaining geographic coordinates of energy storage equipment and a network connection topological structure of the energy storage equipment and a central node; the central node represents a data center deployed at the cloud, and the data center represents a server cluster; constructing a distributed architecture comprising a central node and a plurality of edge nodes according to the obtained geographic coordinates of the energy storage device and the network connection topological structure of the energy storage device and the central node; deploying distributed data storage on the constructed distributed architecture by utilizing a distributed file system of the hong Mongolian system; collecting real-time data of the energy storage equipment, and sending the real-time data to an edge node through a distributed network of the hong Monte system, wherein the real-time data comprises current and voltage; the edge node receives the real-time data, performs preprocessing, and performs feature extraction through a parallelized machine learning model to obtain feature data; the edge node sends the preprocessed real-time data to a distributed database on the central node for storage through a distributed soft bus network component of the hong-Meng system; the edge node transmits the extracted characteristic data to a distributed database on the central node for storage through the fragmentation transmission of the hong Monte system; the edge node acquires historical state data of the energy storage device from a distributed database of the central node; the edge node trains a prediction model for predicting the capacity of the energy storage device by utilizing a machine learning model of the hong Mongolian system based on the acquired historical state data; the central node acquires the preprocessed real-time data and the extracted characteristic data from the distributed database, and generates a control instruction of the energy storage device by combining the trained prediction model; and transmitting a control instruction generated by the central node to the energy storage equipment for management control by adopting an industrial time sequence communication protocol based on an OPCUA standard of an open platform communication unified architecture.
Wherein, open source hong Meng system: the distributed operating system developed by Hua-Cheng corporation supports various equipment terminals and realizes cross-platform interconnection and resource sharing. The Hongmon system adopts a distributed architecture design, provides core components such as a distributed file system, distributed data management, distributed soft bus and the like, and is convenient for a developer to construct distributed application. Wherein the edge node: the computing nodes are deployed near the data source or the equipment end and are responsible for data acquisition, preprocessing and real-time analysis. The edge nodes are generally configured with certain storage and computing resources, so that partial data processing tasks can be locally completed, and the load pressure of the center nodes is relieved. In the application, the edge node is used for collecting real-time data of the energy storage device and extracting characteristic data. Wherein, central node: and the high-performance computing nodes are deployed in the cloud or a data center and are responsible for storage, management and deep analysis of mass data. The central node generally adopts a large-scale distributed cluster architecture, and has strong computing and storage capabilities. In the application, the central node is used for converging the data uploaded by the edge node and generating the equipment control instruction by combining the machine learning model.
Wherein, machine learning model: a kind of artificial intelligence algorithm based on data driving automatically extracts characteristic rules and mapping relations contained in data through learning and training of historical data. Common machine learning models include neural networks, decision trees, support vector machines, and the like. In the application, a machine learning model is used for predicting the capacity change trend of the energy storage device. Wherein the distributed soft bus network component: the hong system provides a distributed communication mechanism that supports messaging and data exchange between different devices and applications. And loose coupling and flexible expansion among all modules of the distributed system are realized through a unified interface definition and registration discovery mechanism. In the present application, a distributed soft bus is used for real-time data transmission between an edge node and a center node. Wherein, fragmentation transmission: a data block transmission technology divides large data into a plurality of small data fragments for transmission, which can improve the efficiency and reliability of network transmission. In the present application, the fragmented transmission is used for the edge node to efficiently transmit the extracted feature data to the center node. Wherein OPCUA (Open Platform Communications Unified Architecture): industrial interoperability standards established by the OPC foundation support data communication and information exchange between industrial devices. OPCUA adopts a service-oriented architecture, and provides a set of general information model and interface specifications. In the application, the OPCUA is used for transmitting control instructions between the central node and the energy storage equipment, and the standardization and compatibility of communication are ensured.
Further, obtaining the geographic coordinates of the energy storage device and the network connection topology of the energy storage device and the central node includes: acquiring geographic coordinate data of the energy storage equipment through a GPS, and transmitting the acquired geographic coordinate data to an edge node; after the edge node receives the geographic coordinate data, manhattan distance calculation is carried out through a Map Reduce frame, and the spatial distribution characteristics of the energy storage equipment are obtained; wherein the spatial distribution feature comprises a spatial coverage of the energy storage device; the edge node sends the calculated spatial distribution characteristics to the central node through a distributed soft bus network component of the hong Monte system; acquiring network connection information data of the energy storage equipment and the central node through a network topology discovery protocol LLDP, and sending the network connection information data to the edge node; the network connection information data comprises equipment identification, port information, time delay and link rate; after receiving the network connection information data, the edge node performs link statistical analysis through a network topology analysis tool Gephi to obtain network topology characteristics; wherein the network topology features include node degree distribution, average path length, and network diameter; the edge nodes send the obtained network topology characteristics to the central node through the distributed soft bus network components of the hong Monte system.
Wherein, map Reduce framework: a distributed computing model for parallel computing processing large-scale data sets. Map Reduce divides the computing task into two phases: the Map stage performs slicing and preprocessing on the data, and the Reduce stage performs summarizing and aggregation on the output result of the Map stage. By distributing tasks to multiple nodes for parallel execution, map Reduce can significantly improve the efficiency of data processing. According to the application, the edge node utilizes the Map Reduce frame to perform Manhattan distance calculation on the geographic coordinate data of the energy storage device, so that the spatial distribution characteristics are obtained rapidly.
Wherein, spatial distribution characteristics: an index describing the distribution and coverage of the energy storage device in the geographic space. Through cluster analysis on the equipment coordinates, hot spot areas and blank areas of equipment distribution can be identified, and the rationality and optimization potential of equipment deployment are evaluated. In the application, the spatial distribution characteristics are used for describing the geographic distribution mode of the energy storage equipment, and provide references for optimizing the equipment layout. Wherein, network topology discovery protocol LLDP (LINK LAYER Discovery Protocol): a data link layer protocol for exchanging identity, capability and adjacency status information between network devices. Through LLDP, the network management system can automatically discover the devices, ports and connection relations in the network, and a complete network topology diagram is constructed. In the application, LLDP is used for obtaining the network connection information between the energy storage device and the central node, including device identification, port, time delay, link rate, etc.
Wherein, network topology analysis tool Gephi: the open-source network visualization and analysis software provides rich layout algorithms and statistical indexes, and helps users to explore and understand the structural characteristics of the complex network. Through Gephi, a visual representation of the network topology may be generated, calculating various attribute metrics of the network, such as node degree distribution, average path length, network diameter, etc. In the application, the edge node uses Gephi to analyze the network connection information collected by LLDP, and generates visual network topology structure characteristics. Wherein, network topology characteristics: a series of indicators reflecting the connection relationships and organization patterns between network nodes. Common topological features include: node degree distribution (reflecting the distribution of the number of node connections), average path length (reflecting the average distance between nodes), network diameter (reflecting the longest shortest path in the network), etc. By analyzing the network topology characteristics, the connectivity, robustness and transmission efficiency of the network can be evaluated. In the application, the network topology features are used to characterize the quality and reliability of the communication link between the energy storage device and the central node.
Further, constructing a distributed architecture comprising a central node and a plurality of edge nodes, comprising: the central node receives the spatial distribution characteristic and the network topology characteristic data sent by the edge node; the central node adopts a greedy algorithm to set a plurality of groups of node position combination schemes according to the received spatial distribution characteristics and network topological structure characteristics; the central node sends the node position combination schemes of each group to the edge node; the edge node receives the node position combination schemes, and carries out network performance evaluation on each position combination scheme through a simulation method to obtain network time delay and load performance data of each scheme; the edge node returns the network delay and load performance data obtained by simulation to the center node; the center node selects the combination with the minimum network delay and the highest load performance as the optimal deployment position of the center node and the edge node; the central node sends the obtained optimal deployment position to the edge node through a distributed soft bus network component of the hong Monte system; and the edge node controls the energy storage equipment and the edge computing equipment to be deployed on the corresponding positions according to the received optimal deployment positions, and establishes connection with the central node to construct a distributed architecture.
Wherein, greedy algorithm: a common heuristic optimization algorithm attempts to achieve a globally optimal solution of the problem by local optimization of each step. The greedy algorithm takes the optimal decision in the current state in each step of selection in hope of achieving global optimum through a series of local optimum selections. In the application, a central node adopts a greedy algorithm to generate a plurality of groups of node position combination schemes, and each step selects the node position with the widest current coverage range and the best connectivity, thereby obtaining a globally better deployment scheme set. Wherein, network delay: the time required for data transmission from the source node to the target node reflects the data transmission efficiency of the network. The network delay is composed of multiple parts including transmission delay, propagation delay, processing delay, and queuing delay. Factors affecting network latency include network bandwidth, link distance, node processing power, network congestion, and the like. In the application, the edge node evaluates the network time delay under different node deployment schemes by a simulation method, and selects the scheme with the minimum time delay so as to ensure the instantaneity of data transmission.
Wherein, load performance: metrics of workload and resource utilization of each node in the distributed system are measured. Good load performance means that the processing tasks of the system are distributed uniformly among the nodes, and the situation that the load of individual nodes is excessive and the utilization of other nodes is insufficient is avoided. The load balancing system can fully exert the parallel processing capability of the distributed architecture and improve the overall performance of the system. According to the application, the edge node evaluates the load performance under different node deployment schemes by a simulation method, and selects the scheme with the most balanced load distribution and the highest resource utilization rate.
Specifically, the central node selects the combination with the minimum network delay and the highest load performance: the central node receives network delay and load performance data corresponding to each group of node position combination schemes from the edge nodes. The central node sorts the network delay data and selects a plurality of schemes with minimum delay as candidates. And in the candidate scheme with the minimum time delay, the central node further compares the load performance data, and selects the scheme with the most balanced load (such as the smallest load variance) and the highest resource utilization (such as the highest average load rate). If multiple schemes with similar load performances exist in the previous step, the central node can comprehensively evaluate according to other factors (such as node failure rate, deployment cost and the like) to select an optimal scheme. The central node takes the selected optimal node position combination scheme as an optimal deployment position and issues the optimal deployment position to the edge node through the distributed soft bus network component.
Further, deploying the distributed data store, comprising: respectively deploying No SQL databases on a central node and an edge node of the constructed distributed architecture to form a distributed database cluster; the central node configures redis cache middleware on the deployed No SQL database and sends configuration information to the edge node; and the edge node receives the configuration information of the redis cache middleware of the center node, and carries out corresponding configuration on the local No SQL database.
Wherein, noSQL database: compared with a traditional relational database (SQL database), the NoSQL database adopts a non-relational data model, and higher expandability, usability and flexibility are provided. NoSQL databases are typically based on key-value pairs, documents, column families, or graph-like data structures, and are adapted to handle large-scale unstructured or semi-structured data. Common NoSQL databases include MongoDB, cassandra, HBase, etc. In the application, the center node and the edge nodes are respectively deployed with NoSQL databases to form a distributed database cluster so as to support the efficient storage and inquiry of massive time sequence data.
Wherein, redis caches middleware: a memory-based key-value database provides high performance data caching and message queuing functions. Compared with disk storage, redis stores data in a memory, supports millisecond-level data read-write operation, and is commonly used for improving response speed and concurrency performance of an application system. Redis supports a variety of data structures, such as strings, hashes, lists, collections, etc., while providing advanced features of publish/subscribe, transactions, lua scripts, etc. In the application, the central node configures Redis cache middleware on the NoSQL database for caching frequently accessed hot spot data, thereby relieving the query pressure of the database and improving the response speed of the system. Wherein, configuration information: the information such as various parameters, options, environment variables and the like required by the running of the application system is used for controlling the behavior and performance of the system. Configuration information is typically stored in the form of configuration files, environment variables, databases, etc., which are read and applied at system start-up or run-time. Rational configuration management can improve flexibility, maintainability, and portability of the system. In the application, the central node sends the configuration information (such as host address, port number, database number, password, etc.) of the Redis cache middleware to the edge node, so that the cache service on the distributed node can be ensured to be normally connected and operated, and unified cache management is formed.
Further, according to the preprocessed real-time data, the edge node performs feature extraction through a parallelized machine learning model, including: the edge node divides the preprocessed real-time current and voltage data of the energy storage device into a plurality of data blocks according to time windows; inputting the data blocks into a plurality of convolutional neural networks; each convolutional neural network independently trains and updates network parameters, and outputs the mean value and variance of the residual life RUL of the energy storage equipment of the corresponding data block; the edge node aggregates the RUL mean value and the variance of each data block to obtain health characteristic data of the energy storage equipment; the edge node sends the obtained health characteristic data to the central node for storage through a distributed soft bus network component of the hong Mongolian system; the edge node organizes the real-time current and voltage data of the preprocessed energy storage equipment according to time sequence, and inputs the data into the circulating neural network; the circulating neural network obtains a long-term dependency relationship in the time series data through a gate control circulating unit GRU, and outputs a predicted value of the self-discharge rate and the internal resistance of the energy storage equipment; the edge node compares the predicted value of the self-discharge rate and the internal resistance output by the cyclic neural network with the historical value to obtain the increasing trend of the self-discharge rate and the internal resistance as the performance attenuation characteristic of the energy storage equipment; the edge nodes send the obtained performance attenuation characteristics to the central node for storage through the distributed soft bus network component of the hong Monte system.
Wherein the time window: the continuous time series data is divided into time segments of a fixed length, each time segment being referred to as a time window. The size of the time window is set according to the specific application scene and the data characteristics, and usually contains a certain number of sampling points. Through time window division, the long-time sequence can be converted into a plurality of short sequences, so that parallel processing and feature extraction are facilitated. In the application, the edge node divides current and voltage data into a plurality of data blocks according to time windows, and the data blocks are respectively input into a convolutional neural network for feature extraction.
Wherein, energy storage device remaining life RUL (Remaining Useful Life): representing the remaining time of the energy storage device from the current time to the failure time. The RUL of the energy storage equipment is accurately predicted, and the method has important significance in formulating maintenance strategies, optimizing operation modes and guaranteeing power supply reliability. By analyzing historical operating data and real-time monitoring data of the energy storage device, a machine learning model may be established to estimate the RUL. In the application, the convolution neural network outputs the RUL mean value and variance corresponding to each data block through extracting the characteristics of the real-time current and voltage data, and reflects the health state of the energy storage device in different time windows.
Wherein, self-discharge rate: refers to the rate of charge loss of the energy storage device in an open circuit state due to internal chemical reactions, leakage, and the like. The self-discharge rate is one of the key indicators affecting the performance and life of the energy storage device, and higher self-discharge rates can accelerate capacity decay and efficiency degradation of the energy storage device. By analyzing the charge and discharge data of the energy storage device, the magnitude and the change trend of the self-discharge rate can be estimated. In the application, the cyclic neural network outputs the predicted value of the self-discharge rate through extracting the characteristics of the real-time current and voltage time sequence, and is used for evaluating the performance attenuation condition of the energy storage equipment. Wherein, internal resistance: and the equivalent resistance inside the energy storage device is represented, and the energy loss of the energy storage device in the charging and discharging processes is reflected. The internal resistance directly influences the performance indexes such as charge and discharge efficiency, power output, thermal management and the like of the energy storage device. As the energy storage device is used and aged, the internal resistance may gradually increase, resulting in reduced performance. By analyzing the voltage and current responses of the energy storage device, the value and trend of the internal resistance can be estimated. According to the application, the circulating neural network outputs the predicted value of the internal resistance through extracting the characteristics of the real-time current and voltage time sequence, and is used for evaluating the performance attenuation condition of the energy storage equipment.
Wherein, health degree characteristic: and a group of indexes for comprehensively reflecting the current health state of the energy storage equipment, including capacity, self-discharge rate, internal resistance, RUL and the like. By fusion analysis of the health indexes, the health level and the residual life of the energy storage device can be comprehensively estimated. In the application, an edge node aggregates the average value and variance of each data block RUL output by a convolutional neural network to obtain the integral health characteristic of energy storage equipment, and sends the health characteristic to a central node for storage and decision analysis. Wherein, performance decay characteristics: describing the trend and rule that each performance index gradually decreases in the long-term use process of the energy storage equipment. Performance decay characteristics include capacity decay, increased self-discharge rate, increased internal resistance, etc., reflecting the aging process and life prediction of the energy storage device. By analyzing the change trend of the performance attenuation characteristics, the operation mode of the energy storage equipment can be optimized, the performance attenuation is delayed, and the service life is prolonged. In the application, the edge node compares the self-discharge rate and the internal resistance predicted value output by the cyclic neural network with the historical value, acquires the trend characteristic of performance attenuation, and sends the trend characteristic to the central node for storage and decision analysis.
Specifically, the extraction of health degree features: dividing data: the edge node divides the preprocessed real-time current and voltage data of the energy storage device into a plurality of data blocks according to a time window with a fixed length, and each data block comprises sampling points in a certain time range. Parallel convolutional neural networks: and respectively inputting the divided data blocks into a plurality of independent convolutional neural networks to perform feature extraction. Each convolutional neural network extracts local features in the data block through a convolutional layer and a pooling layer, and outputs the mean and variance of the residual life (RUL) of the energy storage device of the corresponding data block through a full connection layer. Model training: and each convolutional neural network is independently trained by using the data blocks, and network parameters are updated by optimizing the loss function, so that the accuracy of RUL prediction is improved. The loss function may be an evaluation index such as Mean Square Error (MSE) or Mean Absolute Error (MAE). Feature polymerization: and the edge node aggregates the RUL mean value and the variance of each data block to obtain the health degree characteristics of the energy storage device in the whole time range. The aggregation method can adopt strategies such as weighted average, voting and the like, and comprehensively considers the prediction results of all the data blocks. And (3) data transmission: the edge node sends the extracted health features to the central node for storage and subsequent analysis through the distributed soft bus network component of the hong Monte system.
Specifically, extraction of performance decay features: time series data organization: the edge nodes organize the preprocessed real-time current and voltage data of the energy storage device into a time sequence according to the sequence of time stamps, and each time step corresponds to one sampling point. Cyclic neural network: and inputting the time series data into a cyclic neural network for feature extraction. The recurrent neural network captures long-term dependencies in the time series through a recurrent connection and gating mechanism (such as GRU or LSTM) and outputs a state vector for each time step. Predicting performance indexes: the output state vector of the recurrent neural network is mapped to the performance indexes of the energy storage device, such as the self-discharge rate and the internal resistance, through the full connection layer. The network trains the prediction model by optimizing a loss function (such as MSE or MAE), and improves the accuracy of the performance index prediction. Trend analysis: and comparing the self-discharge rate and the internal resistance value predicted by the cyclic neural network with historical data by the edge node, and calculating the change rate and trend of the performance index. The change rate can be calculated by difference or slope, and the trend can be determined by threshold judgment or statistical test. Feature extraction: the change rate and trend of the self-discharge rate and the internal resistance are used as the performance attenuation characteristics of the energy storage device, and the aging rule and degradation speed of the device in the long-term use process are reflected. And (3) data transmission: the edge nodes send the extracted performance decay characteristics to the central node for storage and subsequent analysis through the distributed soft bus network components of the hong-Meng system.
Further, the distributed soft bus network component includes: the system comprises a message integration module, a service registration module and a request routing module; the message integration module receives the custom protocol data of the edge node or the center node and converts the custom protocol data into a standard protocol format according to a preset coding mapping rule; the converted standard protocol format is sent to a service registration module for service access; the service registration module receives a service access request and stores service instance information into a centralized etcd database; the request routing module acquires service instance information corresponding to the access request by inquiring the etcd database according to the service access request; according to the acquired service instance information, routing the access request to the node where the service instance is located by adopting a consistent hash algorithm, and carrying out load balancing; and converting the service response result data into a custom protocol format, and returning the custom protocol format to the edge node or the center node initiated by the request.
Wherein, the code mapping rule: refers to a set of conversion rules that convert custom protocol data into a standard protocol format. Because different data formats and communication protocols may be adopted between different nodes or systems, in order to implement interconnection and interworking and data sharing, it is necessary to map respective custom protocol data to a unified standard protocol format. The code mapping rule defines the corresponding relation between the custom protocol and the standard protocol, including data type, field name, data length, etc. The message integration module can automatically complete protocol conversion through a pre-configured coding mapping rule, so that seamless communication between heterogeneous systems is realized.
Wherein, service instance information: metadata of a service's deployment location, access mode, running state, etc. are described. In a micro-service architecture, multiple instances of a service are typically deployed on different nodes to improve service availability and load balancing capabilities. The service instance information includes service name, instance ID, IP address, port number, protocol type, etc. for uniquely identifying and locating a service instance. The service registration module stores the service instance information into the etcd database for query and use by other modules. When the service instance changes (such as capacity expansion, offline, etc.), the service instance information needs to be updated in time to ensure the accuracy of service call.
Wherein etcd database: an open-source distributed key-value storage system is commonly used for service registration and discovery in a micro-service architecture. The etcd adopts Raft consistency algorithm to realize distributed data synchronization, and ensures high availability and strong consistency of data. etcd provides a simple and easy-to-use RESTful API that supports the add-drop-check operation of key-value pairs. In the present application, the service registration module stores service instance information in the form of key-value pairs to the etcd database, where the keys are typically service names and the values are metadata of the service instances. And a plurality of etcd nodes form a cluster, and the automatic copying and fault recovery of data are realized through Raft protocols, so that the reliability of the system is improved.
Wherein, consistent hashing algorithm: an algorithm for load balancing in a distributed system, particularly for dynamically changing service instance clusters. The consistent hashing algorithm maps both service instances and requests to a hash ring of a fixed size, one location on the hash ring for each service instance. When a request arrives, the hash value of the request is calculated through a hash function, then the first service instance with the hash value larger than or equal to the hash value on the hash ring is searched clockwise, and the request is routed to the instance for processing. The consistent hash algorithm has good distribution uniformity and data stability, and only a small part of data needs to be moved when service examples are dynamically added and deleted, so that large-scale data migration is avoided. In the application, the request routing module adopts a consistent hash algorithm to carry out load balancing, calculates a hash value according to the characteristic information of the service access request, uniformly distributes the request to different service instances, and improves the concurrent processing capacity of the system.
Further, the method for transmitting the extracted characteristic data to the distributed database on the central node through the fragmentation transmission of the hong Monte system comprises the following steps: the edge node compresses the obtained health degree characteristic data and performance attenuation characteristic data of the energy storage equipment by adopting a Huffman coding algorithm; the edge node divides the compressed characteristic data into a plurality of data blocks by fibonacci dividing method, and adds CRC32 check code to each data block; the edge node generates an identification code for each data block by using an Object ID algorithm of MongoDB; the identification code comprises a time stamp and a sequence number; according to the identification code of the data block, the edge node adopts a APACHE KAFKA partitioning mechanism to send the data block with the same time stamp prefix to the same partition; the kafka consumer of the central node receives the data blocks according to the subareas and classifies the data blocks belonging to the same original characteristic data through the identification codes; the central node sorts and splices the classified data blocks according to the time stamp and the sequence number of the identification code, and original characteristic data are obtained through a Huffman decoding algorithm; the central node stores the obtained original characteristic data into a distributed database MongoDB, and constructs a data set indexed by device ID and time stamp in the MongoDB database.
Wherein, huffman coding algorithm: a lossless data compression algorithm constructs an optimal binary code tree based on the frequency of occurrence of characters. Huffman coding utilizes probability information of character occurrence to distribute shorter codes to characters with high occurrence frequency and longer codes to characters with low occurrence frequency, thereby achieving the purpose of compressing data. The Huffman coding process comprises the steps of counting character frequency, constructing a Huffman tree, generating a coding table, coding data and the like. Huffman coding is an entropy coding algorithm, can compress data to a theoretical limit close to information entropy, and has higher compression efficiency. In the application, the edge node compresses the health characteristic data and the performance attenuation characteristic data by adopting the Huffman coding, thereby reducing the bandwidth occupation of data transmission.
Wherein, the Fibonacci segmentation method comprises the following steps: a method for dividing data into a plurality of data blocks with unequal sizes, wherein the sizes of the data blocks follow the rule of a Fibonacci array. The Fibonacci number columns are defined in a recursive manner, each number being the sum of the first two numbers, shaped as: 1,1,2,3,5,8, 13, 21, 34. According to the size of data, the Fibonacci segmentation method selects the number in the Fibonacci sequence as the size of a data block, and sequentially divides the data into data blocks with the size of Fibonacci. Compared with fixed-size segmentation, the Fibonacci segmentation can better adapt to the uneven distribution of data, and reduce the number of data blocks and transmission delay. In the application, the edge node adopts a Fibonacci segmentation method to segment the compressed characteristic data into a plurality of data blocks, thereby improving the efficiency and reliability of data transmission.
Wherein, CRC32 check code: a commonly used data integrity checking algorithm is used to detect whether data is in error during transmission or storage. The CRC32 check code is based on the cyclic redundancy check (Cyclic Redundancy Check) principle, the data is regarded as a binary polynomial, and the remainder is obtained as the check code by carrying out modulo-2 division operation with a fixed generator polynomial. The sender attaches CRC32 check code after the data, the receiver calculates the check code of the received data through the same algorithm, compares the check code with the received check code, and if the check code is inconsistent, the receiver indicates that the data has errors. The CRC32 check code has higher error detection capability, and can detect most random errors and burst errors. In the application, the edge node adds CRC32 check code for each data block, thereby ensuring the integrity and reliability of data transmission.
Wherein Object ID algorithm: an algorithm in the mongo db database for generating a unique identifier is used to identify documents (documents). The Object ID is a 12-byte (24 hexadecimal characters) string consisting of a time stamp, a machine ID, a process ID, and a counter. Wherein the timestamp occupies 4 bytes, representing the time of Object ID generation; the machine ID occupies 3 bytes, representing the machine that generated the Object ID; the process ID occupies 2 bytes, representing the process that generates the Object ID; the counter occupies 3 bytes and is used to distinguish different Object IDs at the same time, in the same machine and in the same process. The Object ID can guarantee uniqueness in a distributed environment and also contains information about creation time. In the application, the edge node generates a unique identification code for each data block by using an Object ID algorithm, so that the subsequent data block classification and sequencing are facilitated.
Wherein APACHE KAFKA: a distributed stream processing platform for building real-time data pipes and stream applications. Kafka operates in a publish-subscribe mode, with a Producer (Producer) publishing data to a specified Topic (Topic), and a Consumer (Consumer) subscribing to and consuming data from the Topic. Kafka introduced the concept of partitioning (Partition) to divide a topic into multiple partitions, each of which can independently store and consume data, improving the ability to process in parallel. Kafka adopts log-based persistent storage, and ensures the reliability and the sequency of data. At the same time, kafka also provides rich APIs and integration functions, supporting multi-language clients and stream processing frameworks. In the application, the edge node sends the data blocks with the same time stamp prefix to the same partition by using the partition mechanism of Kafka, thereby realizing the ordered transmission and load balancing of the data.
Further, the edge node trains a predictive model for predicting the capacity of the energy storage device based on the historical state data using a machine learning model of the hong Monte system, comprising: the edge node acquires historical state data of the energy storage device from a distributed database MongoDB, wherein the historical state data comprises current and voltage time sequence data in the past year; the edge node adopts a statistical analysis method based on differential entropy to calculate the differential entropy value in each time window of the historical state data; the edge node is provided with sliding windows, and the differential entropy value of each window forms a differential entropy characteristic vector; by sliding a time window on a time axis, a differential entropy feature vector sequence is obtained; the edge node takes the differential entropy feature vector sequence as input, a plurality of clusters are divided by a BIRCH incremental clustering algorithm, and the center point of the cluster reflects the mode of the state of the energy storage device under different working conditions; the edge node adopts a z-score standardization method to normalize the data in each cluster; the edge node takes the cluster data after normalization processing as a training set and is used for training a prediction model; and constructing a sequence prediction model comprising a multi-layer LSTM network and residual connection, and performing model training by using the obtained training set to obtain a prediction model for predicting the capacity of the energy storage equipment.
Wherein the distributed database MongoDB: a NoSQL database based on documents (documents) stores data in a JSON-like (BSON) format. MongoDB supports flexible data models, and fields can be dynamically added and modified without the need for pre-defining table structures. MongoDB provides high availability, extensible and distributed data storage and query functions, realizes redundant backup and automatic fault switching of data through a copy set (REPLICA SET), and realizes horizontal splitting and load balancing of data through a slicing (sharding). MongoDB also supports rich index types, aggregation operation, geographic position query and other functions, and is suitable for large-scale, high-concurrency and real-time data management scenes. In the application, the historical state data is stored in the MongoDB database, so that the edge node can acquire and process the data efficiently.
The statistical analysis method based on the differential entropy comprises the following steps: a method for analyzing complexity and change rules of time sequence data. The differential entropy (DIFFERENTIAL ENTROPY) is the information entropy of the continuous random variable, reflecting the degree of uncertainty of the random variable. For time series data, the differential entropy can be obtained by calculating the entropy value of probability distribution of first order difference (difference between two adjacent data points). The larger the differential entropy is, the more severe the change of time sequence data is, and the larger the amount of information is contained; the smaller the differential entropy, the more gradual the change in the time series data, and the smaller the amount of information contained. By calculating the differential entropy values within different time windows, the complexity characteristics of the time series data on different time scales can be characterized. In the application, the edge node adopts a statistical analysis method based on differential entropy to extract the complexity characteristic of the historical state data, and provides effective characteristic representation for subsequent clustering and prediction.
Wherein, differential entropy feature vector: the feature vector, which consists of differential entropy values within a time window, represents the complexity characteristics of the time series data within the time window. And (3) dividing the time sequence data into a plurality of time slices by setting a time window with a fixed size, and calculating a differential entropy value for the data in each time slice to obtain a differential entropy characteristic vector. The differential entropy feature vector can capture the change mode of time sequence data in a local time range, and reflects the dynamic characteristics of the data. In the application, the edge node calculates the differential entropy value of each time window to form a differential entropy characteristic vector which is used for representing the state characteristics of the energy storage equipment in different time periods.
Wherein, the differential entropy feature vector sequence: a sequence formed by arranging a plurality of differential entropy feature vectors in time sequence represents the complexity variation trend of time series data in a continuous time period. And moving the window on a time axis by sliding the time window with a fixed step length, and calculating the differential entropy characteristic vector in each window to obtain a differential entropy characteristic vector sequence. The differential entropy feature vector sequence can describe the long-term change rule of time sequence data, and embody the evolution process of the data. According to the method, the edge node generates a differential entropy feature vector sequence in a sliding window mode, and the differential entropy feature vector sequence is used as input of a clustering and prediction model to capture dynamic changes of states of the energy storage equipment.
Wherein, BIRCH incremental clustering algorithm: balancedIterativeReducingandClusteringusingHierarchies an incremental algorithm for large-scale data clustering. BIRCH represents a clustering structure of data by building a hierarchical cluster feature tree (Clustering Feature Tree, CFTree), each node containing a set of statistical information (e.g., number of data points, linear sum, square sum, etc.) of the data. When a new data point arrives, BIRCH inserts it into the nearest leaf node based on the distance metric and updates the cluster features along the path. When the data amount of one node exceeds a threshold value, node splitting is performed. BIRCH can process data in an incremental mode, does not need to scan a data set for multiple times, and has higher clustering efficiency. In the application, the edge node adopts BIRCH algorithm to cluster the differential entropy feature vector sequence to obtain a cluster which reflects the state modes of the energy storage equipment under different working conditions.
Wherein, LSTM network: long Short-Term Memory, a recurrent neural network for processing time series data. LSTM overcomes the gradient elimination and gradient explosion problems of the traditional circulating neural network by introducing a gating mechanism (input gate, forgetting gate and output gate) and a memory unit, and can effectively learn the long-term dependency relationship. The hidden state of the LSTM is transferred in the time dimension, and information is selectively reserved and updated through a gating mechanism, so that dynamic modeling of time sequence data is realized. LSTM has been widely used in speech recognition, natural language processing, time series prediction, and other fields. In the application, the prediction model adopts a multilayer LSTM network, and the long-term trend of the capacity change of the energy storage equipment is captured by learning the clustered time sequence data, so that the prediction of the future capacity is realized.
Further, according to the preprocessed real-time data, the extracted feature data and the trained prediction model, the central node generates a control instruction of the energy storage device, including: the center node receives the energy storage equipment real-time state data preprocessed by the edge node; the central node acquires health characteristics and performance attenuation characteristics of the energy storage equipment from a distributed database MongoDB; the central node aligns the acquired real-time state data, health degree characteristics and performance attenuation characteristics according to time and then takes the aligned real-time state data, health degree characteristics and performance attenuation characteristics as input vectors; predicting through the trained prediction model to obtain a prediction curve of the capacity change of the energy storage equipment in a period of time in the future; the central node extracts peak-valley characteristics of the obtained prediction curve, and a multi-scale analysis method based on wavelet transformation is adopted to obtain local maximum value points and minimum value points of the prediction curve; the central node calculates control parameters for adjusting the charge and discharge current and voltage of the energy storage device by adopting an incremental PID algorithm according to the acquired local maximum value point and the acquired local minimum value point; the central node encapsulates the calculated control parameters into control instructions, the control instructions are issued to the edge nodes, the edge nodes convert the control instructions into CAN bus frame formats, and the energy storage equipment is managed.
Wherein, time alignment: the process of synchronizing and aligning data from different data sources or different time stamps results in data that remains consistent in the time dimension. When time series data are processed, due to factors such as frequency, delay, interruption and the like of data acquisition, the problem of inconsistent time stamps may exist for data from different sources. The time alignment maps the data to the uniform time stamp by interpolation, deletion, complementation and other methods, so that the continuity and the synchronism of the data on a time axis are ensured. Common time alignment methods include nearest neighbor interpolation, linear interpolation, spline interpolation, and the like. In the application, the central node aligns the real-time state data, the health degree characteristic and the performance attenuation characteristic according to time to form a unified input vector, thereby facilitating the processing and analysis of the prediction model.
Wherein, the multi-scale analysis method based on wavelet transformation comprises the following steps: a method for analyzing features of time series data at different frequency scales. Wavelet transformation achieves multi-scale decomposition of signals by converting time domain signals into time-frequency domain signals. Wavelet transformation uses a scalable and translatable wavelet basis function to capture local features of a signal at different scales by changing the scale and position of the basis function. The wavelet transformation decomposes the signal into a low-frequency approximation coefficient and a high-frequency detail coefficient, and the functions of denoising, compressing, extracting features and the like of the signal are realized through analysis and reconstruction of the coefficients. Common wavelet basis functions include Haar wavelets, daubechies wavelets, morlet wavelets, and the like. In the application, a central node processes a prediction curve by adopting a multi-scale analysis method based on wavelet transformation, and extracts local extremum characteristics of the curve on different frequency scales for subsequent control parameter calculation.
Wherein, peak valley characteristic extraction: a process of identifying and extracting local maximum points (peaks) and minimum points (valleys) from the time series data. The peak-valley characteristics reflect the local variation trend and key turning points of the time sequence data, and have important significance for analyzing the fluctuation rule of the data and predicting future trend. The peak-valley feature extraction generally adopts a sliding window and threshold comparison method, and local extremum points meeting the conditions are searched in time sequence data by setting window size and threshold conditions. Common peak-valley feature extraction algorithms include local maximum minimum method, derivative method, template matching method, and the like. In the application, a central node performs peak-valley characteristic extraction on a prediction curve, and local maximum value points and minimum value points of the curve are obtained and used as key control points for adjusting charge and discharge of energy storage equipment.
Wherein, CAN bus frame format: and Controller Area Network the data frame format of the bus is used for data communication on the CAN bus. The CAN bus is a serial communication protocol oriented to messages and is widely applied to the fields of automobiles, industrial automation and the like. The CAN bus frame consists of a frame start bit, an arbitration section, a control section, a data section, a CRC section, an ACK section and a frame end bit. Wherein the arbitration segment contains an identifier and a remote transmission request bit for determining the priority and type of the frame; the control section comprises a data length code and reserved bits; the data segment contains the data content actually transmitted; the CRC segment is used for error detection; the ACK segment is used for acknowledgement by the recipient. The CAN bus works in a multi-master mode, and the bus conflict is solved through a non-destructive arbitration mechanism. In the application, the control instruction needs to be converted into a CAN bus frame format so as to be transmitted to the energy storage equipment on the CAN bus, thereby realizing the control and management of the equipment.
Further, the incremental PID algorithm adopts a particle swarm optimization-based adaptive incremental PID control algorithm. Specifically, the controller constructs a target optimization function of the PID controller according to the wave crest and wave trough parameters, the historical control parameters and the state feedback of the energy storage device at the current moment, and the optimization target is to minimize the deviation between the actual capacity curve and the prediction curve; initializing a particle group, wherein each particle represents a group of possible PID parameter combinations, and the position coordinates of the particles represent values of a proportional coefficient, an integral coefficient and a differential coefficient; the particle swarm continuously updates the position coordinates of the particles through iterative search, and the optimization strategy comprises the following steps: evaluating the fitness function of each particle, wherein the fitness function is the same as the objective function of the PID controller; updating the historical optimal position and the global optimal position of each particle according to the fitness function value; adjusting the speed and position of each particle according to the individual optimal position and the global optimal position; when the iteration termination condition is met, taking the PID parameter corresponding to the global optimal position as an optimal control parameter at the current moment, and generating a current voltage instruction for controlling the charge and discharge of the energy storage device; and the controller evaluates the control effect of the current PID parameters according to the state feedback of the energy storage equipment, and when the deviation of the actual capacity curve and the predicted curve exceeds a preset threshold value, the controller triggers the next round of particle swarm optimization, dynamically adjusts the PID parameters and realizes the self-adaptive incremental control.
3. Advantageous effects
Compared with the prior art, the application has the advantages that:
By introducing the distributed architecture design and the distributed capability of the hong Meng system, reasonable configuration and scheduling of computing storage resources are realized. And the node deployment is optimized by using a greedy algorithm and a simulation method, so that the operation efficiency and the expandability of the system are improved. The distributed NoSQL database cluster and the redis cache are used, so that the fault tolerance and the access speed of data storage are enhanced;
The feature extraction is performed by combining the convolutional neural network and the cyclic neural network, so that the limitation of a single model is overcome. Through parallelization and time sequence modeling, the health degree and performance attenuation characteristics of the energy storage equipment are accurately obtained, and a reliable basis is provided for subsequent capacity prediction and life management;
Aiming at historical state data, a characteristic engineering method based on differential entropy and incremental clustering is provided. The information quantity change of the differential entropy description data is utilized, the equipment state modes under different working conditions are adaptively identified through a clustering algorithm, the internal association of the data is mined from two dimensions of time and space, and a high-quality model training set is constructed;
And a fragmented data transmission mechanism is adopted to divide large data into small data blocks for coding, checking and parallel transmission, so that the network load is reduced, and the efficiency and reliability of data transmission are improved. The kafka partitioning mechanism and the data block identification algorithm ensure the order and the integrity of the data, and meet the requirement of real-time data return;
the advantage of long-term dependency of LSTM extraction time sequence data is fully utilized, the problem of network degradation is relieved by residual connection, the long-term and short-term accuracy of capacity prediction of the energy storage equipment is remarkably improved, and powerful support is provided for optimizing a scheduling strategy;
The multi-scale analysis method using wavelet transformation extracts local characteristics of the prediction curve, and overcomes the defect that the traditional method is difficult to describe non-stationary signals. The self-adaptive incremental PID algorithm automatically adjusts the control gain according to the system state change, so that smooth and stable control of the charging and discharging processes of the energy storage equipment is realized, and the service life of the equipment is prolonged;
based on the OPCUA standard and the CAN bus, the interconnection and intercommunication between heterogeneous equipment and the network are realized. The consistent hash algorithm and the kafka partitioning mechanism provide high-efficiency load balancing and concurrent processing capacity, and the expandability and the robustness of the system are enhanced;
By adopting the techniques of Huffman coding, CRC checking, data blocking and the like, the compression rate, error detection and correction capability and concurrency of data transmission are improved, the data volume and bandwidth occupation are reduced, and the transmission requirement under the environment of the Internet of things is met. The MongoDB-based data set index accelerates the query and the aggregation analysis of time series data, and provides a high-efficiency and convenient means for data mining.
Drawings
FIG. 1 is an exemplary flow chart of a method of storing energy based on an open source Hongmon system according to some embodiments of the present description;
FIG. 2 is an exemplary flow chart for building a distributed architecture according to some embodiments of the present description;
Fig. 3 is an exemplary flow chart of characteristic data packet transmissions shown in accordance with some embodiments of the present description;
FIG. 4 is an exemplary flow diagram for constructing a predictive model according to some embodiments of the present disclosure;
FIG. 5 is an exemplary flow chart for generating control instructions according to some embodiments of the present description.
Detailed Description
The method and system provided in the embodiments of the present specification are described in detail below with reference to the accompanying drawings.
FIG. 1 is an exemplary flow chart of an energy storage method based on an open source hong-mo system according to some embodiments of the present description, obtaining geographic coordinates of an energy storage device and a network connection topology of the energy storage device and a central node; constructing a distributed architecture comprising a central node and a plurality of edge nodes according to the obtained geographic coordinates of the energy storage device and the network connection topological structure of the energy storage device and the central node; deploying distributed data storage on the constructed distributed architecture by utilizing a distributed file system of the hong Mongolian system; collecting real-time data of the energy storage equipment, and sending the real-time data to an edge node for preprocessing through a distributed network of the hong Monte system, wherein the real-time data comprises current and voltage; according to the preprocessed real-time data, the edge node performs feature extraction through a parallelized machine learning model; the preprocessed real-time data is sent to a distributed database on a central node through a distributed soft bus network component of the hong Monte system; transmitting the extracted characteristic data to a distributed database on a central node through the fragmentation transmission of the hong Monte system; the edge node acquires historical state data of the energy storage device from a distributed database of the central node; the edge node trains a prediction model for predicting the capacity of the energy storage device by utilizing a machine learning model of the hong Mongolian system based on the historical state data; according to the preprocessed real-time data, the extracted characteristic data and the trained prediction model, the central node generates a control instruction of the energy storage device; and an industrial time sequence communication protocol based on the OPCUA standard is adopted to transmit control instructions to manage the energy storage equipment.
Acquiring energy storage equipment information: and obtaining the geographic coordinates of the energy storage equipment and the network connection topological structure of the energy storage equipment and the central node. And installing a GPS module for each energy storage device, acquiring longitude and latitude coordinates of an installation position, acquiring map coordinates of the position of the device by inquiring a geographic information database, manually recording the physical address of the device, and converting the physical address into the map coordinates. Installing a network detection tool on the equipment, actively detecting the link connection performance with the central node, deploying a network monitoring tool on the central node, analyzing the equipment access log, deducing the network topology, consulting a network deployment drawing, acquiring the connection relation of a switch and a router between the equipment and the central node, manually investigating the network access mode of the equipment, and recording the physical network link between the equipment and the central node. The development data acquisition module is used for transmitting the acquired coordinates and topology data to a central data server, and the central server is used for intensively storing the coordinates and network topology information of the equipment, providing the spatial distribution and network topology structure of the visual page display equipment and providing data support for subsequent distributed deployment, network optimization and the like.
FIG. 2 is an exemplary flow chart for building a distributed architecture, analyzing energy storage device information, according to some embodiments of the present description: and carrying out Manhattan distance calculation on the geographic coordinates of the energy storage equipment to obtain the spatial distribution characteristics. Inquiring the obtained geographic coordinate data of the equipment, selecting target equipment participating in calculation, and recording longitude and latitude coordinates of each equipment. Setting the center reference point coordinate as/>Calculating the absolute value of the difference between the horizontal and vertical coordinates of each device to the reference point: /(I)And the sum of the transverse distance and the longitudinal distance is the Manhattan distance from the equipment to the reference point, and the Manhattan distance from all the equipment to the reference point is calculated by repeating the steps. And calculating Manhattan distances from all the devices to the reference point, and arranging the distance values in ascending order to obtain an ordered list of the devices. Distance range intervals such as [0, 100], [101, 500] and the like are set, the number of devices in each interval is counted, and the proportion of the number of devices in each interval to the total number is calculated. Marking the position of each device on the map, representing different distance intervals by different colors, and observing the spatial distribution condition of the devices on the map. Comparing the duty ratio of the number of the devices in each interval, analyzing the aggregation degree of the device distribution, evaluating the uniformity of the spatial distribution, and extracting the device dispersion and aggregation characteristics.
And carrying out link statistical analysis on the network connection topological structure of the energy storage equipment and the central node to obtain the network topological structure characteristics. Collecting network deployment drawings, extracting network connection relations between equipment and a central node, actively detecting link components between the equipment and the central node by using a network detection tool, and summarizing topology connection data acquired by various channels to a statistical system. The number and the length of direct links from the equipment to the central node are counted, the type, the number and the flow processing capacity of network equipment on the links are analyzed, the response time and the packet loss rate of the single-hop links are calculated, and the reliability and the network delay performance of the links are evaluated. And analyzing key link components affecting network indexes, finding out the correlation and main influencing factors of link statistics indexes, and summarizing key characteristics of a network structure. According to network indexes and key characteristics, the capacity of the existing network for supporting distributed deployment is evaluated, and possible network problems of the distributed deployment system are analyzed, so that basis is provided for network optimization.
Constructing a distributed architecture: and setting node position combinations by adopting a greedy algorithm according to the spatial distribution characteristics and the network topological structure characteristics. And referring to the installation list and the debugging report of the equipment from the project management platform, acquiring the specific installation address of the equipment in the report, for example, xx building xx layer xx room, if the report is not noted in detail, re-working and checking the recorded address. Using the Geo-coding API, the address is converted to latitude and longitude coordinates, e.g., input "XXXX", and the API returns the latitude 116.3x latitude 39.9x corresponding to the address. And recording the device names and the converted coordinates together in a table, forming a complete device coordinate data set by one device, adding attribute information such as device types, areas and the like, and enriching the data set. And importing the Gode/hundred degree/Google map data by using the Arc GIS geographic information system software, and uploading the tidied equipment coordinate data set. A grid division is set for the map, for example, 1 km x1 km, and the number of grids is determined according to the map size and the device density. Traversing each grid area, counting the number of devices in the grid by using SQL, and storing the corresponding relation between each grid and the number of devices. For each grid, calculating density = number of devices/grid area, obtaining a distribution density value of devices in the grid area, the density value being used to determine the priority area and node location.
And obtaining a network topology graph file provided by a network department, marking backbone links such as backbone optical fibers, routers and the like on the topology graph by using a marking pen, and distinguishing the backbone links at different levels in color. And marking the positions of the core router and the key switch on the topological graph, recording parameters such as the model numbers, the port numbers, the capacities and the like of the devices, and summarizing a core network device list. In a test environment, bandwidth utilization rate data of a main link are collected, a monitoring tool is used for counting delay jitter data of the main link, and a main link performance analysis report is generated. And displaying the main link distribution and performance data through a chart in a visual way, generating a PDF report, and providing the PDF report for an addressing team to refer to as a basis for determining a network core area. And marking the trend and distribution condition of the main links on the map, determining the region with concentrated network speed and capacity by combining the link indexes, and drawing the boundary range of the region to form a network core region. And opening the map by using CAD or GIS software, drawing the range of the network core area on the map, filling the area, and displaying the boundary of the labeling range. Alternative physical locations around the perimeter of the core area are checked, and candidate points adjacent to or partially covering the area are selected, enabling the node deployment location to access the core link to reduce overhead.
And collecting the geographic coordinate information of all the devices, accurately marking the position of each device on the digital map, and distinguishing different icons by different types of devices. For map drawing grids, for example, 1 km x1 km, the number of grids is reasonably set according to the size of the map, so that a sufficient sample size in each grid is ensured. The number of devices in each grid can be counted by traversing each grid area, and the number can be counted automatically or manually by using the counting function of the map software. For each grid, calculating density = number of devices/grid area, obtaining a device distribution density value of each grid, and determining a target area by using the density data. And (3) carrying out sequencing comparison on the equipment distribution density of each grid, calculating the percentage of each grid density value to the total density, and identifying the grid area with the highest equipment density. The grid with the device density ranking of 2-3 is selected as a target area, the high-density area represents more concentrated and aggregated devices, and the number of devices in the target area accounts for more than 80% of the total number. The range of a target area is marked on the map by using a rectangle, and the target area can be a combination of a plurality of high-density grids to form an area with more densely distributed equipment.
And evaluating the network performance of each node position combination by a simulation method, and selecting the optimal combination as the positions of the center node and the edge node. And deploying energy storage equipment at an optimal position to construct a distributed architecture. And evaluating mainstream network simulation software such as OPNET, NS2 and the like, and selecting the OPNET to perform network modeling simulation. And importing a physical topological connection diagram of an actual network, constructing the same network structure in the OPNET, and configuring a router and switch model consistent with the actual environment. Parameters such as bandwidth and delay of the simulation link are set, parameter values are set by referring to performance indexes of the actual link, and the simulation environment is ensured to be consistent with the actual network. And according to actual service statistics, configuring a similar flow model, and generating corresponding message flow according to different service types, so as to ensure that the simulated network load is close to the actual condition. And the method runs for multiple times, adjusts parameters to achieve the effect of being matched with an actual network, and can be used for evaluating a node position scheme after reaching a certain precision. Different combinations of center node and edge node locations are proposed, such as scheme a: [ center node 1, edge nodes a, b, c ]. And setting the node position of each scheme in the simulation platform, and keeping other parameters unchanged. And running multiple rounds of simulation on each scheme, and collecting data such as network delay, packet loss rate and the like. And analyzing the network index data result of each scheme, and comparing the differences of network delay, packet loss rate and the like of different schemes. And selecting a node position combination scheme with the optimal network index as a scheme for actual deployment of equipment. And comparing indexes such as time delay, packet loss rate and the like of each scheme, taking cost factors into consideration, performing cost performance analysis, and determining a node deployment scheme with optimal network performance. And selecting the node position combination with the best network index as the final deployment blueprint of the center node and the edge node. And actually building a data center at the selected optimal center node position, and deploying a core router, a server group and the like. And erecting a micro-module machine room at the selected optimal edge node position, and deploying an edge computing server, a cache and the like. The central node and each edge node are connected to form a location-based distributed computing network architecture. And testing the network, checking indexes such as time delay and the like, and confirming that the network index reaches the expected value, or else, tuning.
Deploying distributed data storage: and deploying the No SQL database on the central node and the edge node to form a distributed database cluster. And the Redis cache middleware is configured to improve the data access efficiency. Evaluating MongoDB, cassandra the No SQL database, selecting Cassandra for distributed deployment. And constructing a Cassandra database cluster in a central data center, configuring Seed nodes, and managing other nodes. And deploying a Cassandra database on the selected edge node, configuring the Cassandra database as a non-Seed node, and adding the Cassandra database into a cluster of the central node. The test center node is connected with the database on the edge node, and the data can be normally accessed and synchronized through confirmation. And installing a Redis cache middleware on each database node, wherein database operation is performed through the Redis, so that the access speed is improved.
Collecting real-time data: real-time data of the energy storage device, including current and voltage, is collected. Collecting real-time data of the energy storage equipment, and sending the real-time data to an edge node for preprocessing through a distributed network of the hong Monte system, wherein the real-time data comprises current and voltage; according to the preprocessed real-time data, the edge node performs feature extraction through a parallelized machine learning model, and the method comprises the following steps: the method comprises the steps of adopting a convolutional neural network, taking current and voltage of energy storage equipment as input, and outputting the mean value and variance of the residual life RUL of the energy storage equipment as health characteristics of the energy storage equipment; and cleaning and formatting the collected real-time current and voltage data to generate structured time sequence data for model input. The convolutional neural network model CNN is designed, an input layer is a current/voltage time sequence, and an output layer outputs the mean value and variance of the residual life RUL. The CNN model is deployed on each edge node, and parallel training of the model is realized by using a distributed deep learning framework. Inputting the preprocessed data into a CNN model, and outputting RUL mean and variance as health degree characteristics. And the CNN model is continuously trained through the new data, so that the accuracy of the extracted features of the model is improved. And summarizing the characteristic results extracted by each edge node to a central node, and uniformly carrying out health evaluation and fault prediction.
And a cyclic neural network is adopted, current and voltage of the energy storage device are used as time series input, and the increasing trend of the self-discharge rate and the internal resistance of the energy storage device is output as the performance attenuation characteristic of the energy storage device. And (3) carrying out standardization processing on the collected current and voltage time sequence data, unifying time intervals, and converting the time sequence data into a model input format. And designing a cyclic neural network model RNN, wherein an input layer is a current/voltage time sequence, and an output layer outputs the increasing trend of the self-discharge rate and the internal resistance. The cyclic neural network model is deployed at each edge node, and parallel computation is realized by using a distributed deep learning framework. The preprocessed current/voltage time sequence is input to the RNN model, and the self-discharge rate and the internal resistance trend are output as performance attenuation characteristics. And the RNN model is continuously trained through new data, so that the accuracy of the output characteristics is improved. And sending the feature extraction result of the edge node to a central node, and uniformly carrying out equipment health assessment and fault prediction.
The preprocessed real-time data is sent to a distributed database on a central node through a distributed soft bus network component of the hong Monte system; the message integration module adopts a configuration coding mapping mode to perform data format conversion between a custom protocol and a standard protocol; and (3) preprocessing such as checking, filtering, formatting and the like is performed on the collected data at the edge node, a soft bus component provided by the hong Mongolian OS is integrated at the edge node and the central node, the bus component is configured, and communication parameters and security authentication are set. And the edge node encapsulates the preprocessed real-time data into a message, calls a release interface of the soft bus assembly and releases the message. The central node registers subscription specified topics with the soft bus, and can receive messages when the edge node publishes the messages. The soft bus component distributes messages in the network to realize data transmission between the center node and the edge node. And signature verification, encryption and other operations are performed in the data transmission process, so that the safety and reliability of message transmission are ensured. And adjusting the data sending rate according to the network bandwidth to avoid congestion. And analyzing a data custom protocol sent by the edge node, and defining information such as fields, lengths, types, codes and the like.
A general standard protocol, such as JSON or XML, is selected and the syntax structure and requirements of the standard protocol are studied. And comparing the two protocols, establishing a mapping relation table between the fields, and definitely mapping the data in the custom format to the fields in the standard format. And using the language used by the central node to realize a format conversion function, and converting input data into standard output according to the mapping relation. And calling a conversion function at the central node message processing module, receiving the custom input, and mapping and outputting a standard format. And simulating and sending custom protocol data, checking a conversion output result, and verifying conversion integrity by reverse conversion.
The service registration module stores service instance information on the management network node by adopting a centralized etcd database; and the request routing module dynamically routes and forwards the received service access request to the node where the corresponding service instance is located according to the stored service instance information so as to balance the load. And installing an etcd database on a central node server, deploying 3 to 5 etcd nodes by adopting a cluster mode, and configuring an etcd cluster to form a high-availability key-value storage. The registration format of the service instance information, such as the service name, the IP address, the port, the version number and the like, is formulated, and the service instance is described by using the JSON format. Registration logic is realized in the micro-service application code, and when the service is started, the registration logic is connected with the etcd client and the instance information is written. Service instance information is registered in a defined format using a write API provided by the etcd client. The etcd database generates instance information of the service for subsequent service discovery and invocation. The request routing module realizes etcd client terminal and subscribes to the instance information change of the appointed service from etcd. The latest instance information list of the service is obtained from etcd and is converted into a local routing table, wherein the local routing table comprises instance addresses, loads and the like. When the instance list is changed, the local routing table is dynamically updated, and the routing information of the available instances is always maintained. And when the request is received, searching a routing table according to the service name of the request, and selecting an instance by using a load balancing algorithm. According to the instance address, a remote service is invoked, forwarding the request to the selected instance process. If the call fails, the instance is searched again and forwarded, and the high availability of the service is realized.
FIG. 3 is an exemplary flow chart of feature data packet transmission at an edge node for extracting device state features using a machine learning model to generate digital feature data describing device operational states, according to some embodiments of the present description. And analyzing the value domain and probability distribution condition of the characteristic data, and constructing an optimal Huffman coding tree. And generating variable length codes for each characteristic value according to the Huffman tree to form a mapping table from the characteristic value to the Huffman code. According to the mapping table, huffman coding is used to replace the original characteristic value, and the characteristic data is losslessly compressed into a coded bit stream. A reasonable block size is set, such as 256KB for one block, according to the transmission requirements. The bit stream is recursively split using a hierarchical partitioning method, first to cut out primary blocks, then to cut out secondary blocks, and so on. Each time a block is cut out, a block object is generated, which contains information such as block data, block number, start position, etc. Each generated block is assigned a unique ID number, which is convenient for the receiving end to reorganize. A block information table is maintained storing each block ID and location in the bitstream. And merging all the blocks, checking the correctness of the segmentation, and ensuring that no error occurs in the hierarchical segmentation. And transmitting or carrying out subsequent processing on each generated data block.
A class library or module is introduced in the code that generates the Object Id, such as the Object Id in MongoDB or a custom Object Id generator. The Object Id class construction method is called to create an Object, and the required construction parameters such as machine Id, process Id, time stamp, etc. are imported. A generation method, such as generate (x), of the Object Id Object is invoked, which returns a new Object Id value based on the initialized parameters. The generated Object Id is converted into a character String, so that the character String is convenient to transmit and store, and the to String (x) method of the Object Id can be called. A unique Object Id is generated for each data block Object and stored in the Id field of the data block Object. Maintaining a mapping table of Object ids and data blocks facilitates finding the corresponding data block from the Object Id. And carrying Object Id information corresponding to each block during data transmission, and using the Object Id information for reorganization of a receiving end.
Topic is created in the Kafka cluster and used for transmitting data blocks, and the number of Topic partitions is specified and used for realizing parallel transmission. The edge node, acting as a Kafka producer, obtains the data block and encapsulates the data block content and the block ID into a message. The producer issues a message to Topic with block ID as Key, and the message will be evenly distributed to different partitions. The central node subscribes to the Topic as a consumer and pulls the message from the partition to obtain the original data block. The data block content and corresponding block ID are parsed from the message. The block ID is used as a key, the content of the data block is used as a value, and the data block is written into a database, such as a CASSANDRA, HBASE distributed database. And returning a confirmation message to the producer.
The database has stored the data blocks transmitted from the edge nodes, uses the block ID as a key, uses the database interface to sort the query data blocks by block ID field, and returns an ordered set of data block results. And connecting the data block contents one by one according to the block ID sequence, and recovering the data block contents into the original compressed bit stream data. And carrying out Huffman decoding on the recombined bit stream to restore the original characteristic data. Comparing the recombined and decompressed data with the original characteristic data, and verifying that the recombined and decompressed data are completely consistent and have no difference. If the verification is passed, the transmission and recombination of the data block are confirmed to be completed correctly, otherwise, the problem causing inconsistency is found and corrected. And submitting the correctly recombined characteristic data to a subsequent analysis module for use.
Collecting historical current and voltage time sequence data of energy storage equipment, dividing the time sequence data into equally spaced cells, and calculating probability distribution in each cell. For each cell, calculate probability distribution/>, of the next cellFor each x, calculate/>And/>Log ratio/>. And calculating the expectation of the logarithmic ratio r to obtain a final differential entropy value, wherein the final differential entropy value represents the data distribution difference of two adjacent intervals. According to the data time range, a proper sliding window size is set, for example, 1 hour, each time the sliding is performed for one time step, the differential entropy of the data in the current window is calculated, a differential entropy value sequence is obtained, the change in the whole time range is represented, a line drawing is used for drawing the differential entropy time sequence, the X axis is time, and the Y axis is the differential entropy value. Analyzing the fluctuation condition of the line graph, finding peak-valley characteristic points of the differential entropy, and judging the change condition of data distribution. The differential entropy time sequence is used as a numerical feature vector of the state of the equipment, and the evolution of the state along with time is reflected. The differential entropy feature vector is input into a subsequent machine learning model for analysis tasks such as state prediction or anomaly detection. Meanwhile, the differential entropy characteristics obtained through calculation according to the original data are reserved, and the backup is used for model optimization or debugging.
Collecting differential entropy time series data of equipment as a sample, setting the number k of clusters, establishing a BIRCH model, determining the number k of desired cluster categories according to requirements, establishing the BIRCH model, designating the number k of clusters, and initializing CF tree root nodes. Sample data that needs to be clustered, such as a differential entropy time series of the device, is read. And sequentially taking out each sample, inserting the samples into a CF tree of BIRCH, finding out leaf nodes closest to the leaf nodes, and updating node statistical information. If the number of samples in the node is excessive, a splitting mechanism is triggered to split into two leaf nodes, and the CF tree structure is adjusted. After splitting, the statistical information of the father node is updated upwards, and the CF tree is adjusted layer by layer. Repeating until all samples are inserted, and obtaining the CF tree constructed by the whole samples. The clustering result can be directly read from the CF tree, and the leaf nodes are the sample clusters. The mean μ is calculated for each clustered data sample, the standard deviation σ is calculated for each sample, and the Z-score is calculated for each sample data x: . The Z-score of each sample was mapped to a standard normal distribution, the mean of which was 0 and the standard deviation was 1. The histogram of the normalized sample is drawn to approximate a bell-shaped curve of a standard normal distribution. And storing the standardized sample data as the input of model training. And meanwhile, the original mean value and standard deviation of the sample are saved and are used for carrying out inverse standardization on the model output.
FIG. 4 is an exemplary flow chart for constructing a predictive model according to some embodiments of the present description, setting the number of LSTM layers, e.g., 3 layers, of a network according to problem complexity. And adding residual connection between adjacent LSTM layers, enabling the LSTM output of each layer to enter the next layer and the residual connection at the same time, enabling a residual path to be directly connected, avoiding the fading disappearance of gradients in multiple layers, selecting a mean square error as a loss function of a regression problem, carrying out difference between network output and a true value, carrying out averaging after squaring, enabling a loss function value to be minimum by adjusting network parameters, enabling a residual connection acceleration model to converge, obtaining a better training effect, and obtaining a trained model for predicting equipment states. And dividing partial data from the data set to serve as a test set, inputting test set samples into the trained model, and obtaining prediction output of the model to the samples. And comparing the predicted output with a real target, and calculating an accuracy index such as MSE, MAE, R < 2 >.
FIG. 5 is an exemplary flow chart for generating control instructions according to some embodiments of the present description, the control instructions being generated: and (3) preprocessing the collected real-time data, such as cleaning, denoising, supplementing and missing, extracting health degree characteristics and performance attenuation characteristics of the equipment, and reading a pre-trained LSTM prediction model. A future period of time, such as 1 day, 1 week, etc., is specified that requires prediction. The preprocessing data and the characteristics are input into a prediction model, and the model outputs a prediction result of a corresponding period. The prediction result is a sequence of capacity values for the future time period, constituting a prediction curve of capacity over time. And visually drawing a predicted capacity curve, and visually checking the curve form. And analyzing the peak-valley characteristics of the curve, and providing basis for subsequent control parameter calculation. And reading capacity curve data of a future period predicted by the model, performing smooth filtering on the curve, removing high-frequency noise, and calculating first-order differential of the curve to obtain slopes sequences. Finding the positive and negative variation points of slopes sequences, which are the peak-valley points of the curve. And judging peaks and valleys according to the positive and negative directions of slopes before and after the zero point, and recording the position and the numerical value of each peak and valley. Outputting the specific parameters of each detected peak and valley: location, time of day, numerical value, etc. And marking the found peak and valley points on the original graph, checking whether the marked peak and valley points are accurate, and transmitting the peak Gu Canshu to a subsequent control module for use.
Setting the number of particles, randomly initializing the position and speed of the particles, inputting the wave crest and wave trough parameters of the prediction curve obtained by the previous detection, setting the number of the particles, initializing the position and speed, and combining PID parameters by the position. And establishing a simulation environment by using the wave crest and wave trough parameters, inputting PID parameters of each particle position into the simulation environment, and running simulation to obtain a control effect. And analyzing the control result of the simulation, designing an effect evaluation function, and calculating the proper value of each particle. And evaluating all particles to find the proper highest, namely the current global optimal PID parameter. And (3) searching a better solution by iteratively updating the particle position, and finally outputting the optimal PID parameters found in the optimizing process. And setting the maximum iteration number, stopping when the maximum iteration number is reached, and outputting a final global optimal solution after the iteration is finished, wherein the PID parameter combination effect is optimal. Three PID parameters are assigned to the current and voltage controllers.
The method comprises the steps of obtaining PID control parameters of current and voltage obtained through calculation and optimization, creating an empty data packet according to a protocol supported by equipment, such as Modbus TCP, according to a protocol format, packaging the PID control parameters of the current and the voltage into a specified field of the data packet, adding target control values of the current and the voltage into the data packet, filling a communication address of the equipment into the data packet, calculating and adding a check code according to protocol requirements, converting packaged instruction data into a byte stream format, printing and checking the content of the instruction data packet before transmission, and transmitting a packaged control instruction through a network interface. And writing a client application program by using the OPCUA development kit, and establishing connection with an OPCUA server of the industrial control equipment through a network. And creating nodes of the control parameters in the address space, packaging the control instructions to be sent into data of the nodes, issuing a request for writing the control nodes by the client, subscribing the request by the server, and writing the control instructions. And carrying out real-time reliable transmission by using the industrial time sequence mode of OPCUA, and analyzing a control instruction after receiving the request by the server and transmitting the control instruction to corresponding industrial control equipment. And the industrial control equipment returns the execution state to the OPCUA server. The client subscribes and receives the execution state of the equipment, and closed-loop control of the industrial control equipment is realized through the OPCUA.
Claims (10)
1. An energy storage method based on an open source hong Meng system, comprising:
obtaining geographic coordinates of energy storage equipment and a network connection topological structure of the energy storage equipment and a central node; the central node represents a data center deployed at the cloud, and the data center represents a server cluster;
Constructing a distributed architecture comprising a central node and a plurality of edge nodes according to the obtained geographic coordinates of the energy storage device and the network connection topological structure of the energy storage device and the central node;
deploying distributed data storage on the constructed distributed architecture by utilizing a distributed file system of the hong Mongolian system;
Collecting real-time data of the energy storage equipment, and sending the real-time data to an edge node through a distributed network of the hong Monte system, wherein the real-time data comprises current and voltage;
the edge node receives the real-time data, performs preprocessing, and performs feature extraction through a parallelized machine learning model to obtain feature data;
The edge node sends the preprocessed real-time data to a distributed database on the central node for storage through a distributed soft bus network component of the hong-Meng system;
The edge node transmits the extracted characteristic data to a distributed database on the central node for storage through the fragmentation transmission of the hong Monte system;
The edge node acquires historical state data of the energy storage device from a distributed database of the central node;
the edge node trains a prediction model for predicting the capacity of the energy storage device by utilizing a machine learning model of the hong Mongolian system based on the acquired historical state data;
the central node acquires the preprocessed real-time data and the extracted characteristic data from the distributed database, and generates a control instruction of the energy storage device by combining the trained prediction model;
And transmitting a control instruction generated by the central node to the energy storage equipment for management control by adopting an industrial time sequence communication protocol based on an OPCUA standard of an open platform communication unified architecture.
2. An open source Hongmon system based energy storage method as defined in claim 1, wherein:
The method for obtaining the geographic coordinates of the energy storage device and the network connection topological structure of the energy storage device and the central node comprises the following steps:
Acquiring geographic coordinate data of the energy storage equipment through a GPS, and transmitting the acquired geographic coordinate data to an edge node;
after the edge node receives the geographic coordinate data, manhattan distance calculation is carried out through a Map Reduce frame, and the spatial distribution characteristics of the energy storage equipment are obtained; wherein the spatial distribution feature comprises a spatial coverage of the energy storage device;
the edge node sends the calculated spatial distribution characteristics to the central node through a distributed soft bus network component of the hong Monte system;
acquiring network connection information data of the energy storage equipment and the central node through a network topology discovery protocol LLDP, and sending the network connection information data to the edge node; the network connection information data comprises equipment identification, port information, time delay and link rate;
After receiving the network connection information data, the edge node performs link statistical analysis through a network topology analysis tool Gephi to obtain network topology characteristics; wherein the network topology features include node degree distribution, average path length, and network diameter;
the edge nodes send the obtained network topology characteristics to the central node through the distributed soft bus network components of the hong Monte system.
3. An open source Hongmon system based energy storage method as defined in claim 2, wherein:
constructing a distributed architecture comprising a central node and a plurality of edge nodes, comprising:
the central node receives the spatial distribution characteristic and the network topology characteristic data sent by the edge node;
The central node adopts a greedy algorithm to set a plurality of groups of node position combination schemes according to the received spatial distribution characteristics and network topological structure characteristics;
The central node sends the node position combination schemes of each group to the edge node;
The edge node receives the node position combination schemes, and carries out network performance evaluation on each position combination scheme through a simulation method to obtain network time delay and load performance data of each scheme;
The edge node returns the network delay and load performance data obtained by simulation to the center node;
the center node selects the combination with the minimum network delay and the highest load performance as the optimal deployment position of the center node and the edge node;
the central node sends the obtained optimal deployment position to the edge node through a distributed soft bus network component of the hong Monte system;
and the edge node controls the energy storage equipment and the edge computing equipment to be deployed on the corresponding positions according to the received optimal deployment positions, and establishes connection with the central node to construct a distributed architecture.
4. An open source Hongmon system based energy storage method according to claim 3, wherein:
Deploying a distributed data store, comprising:
Respectively deploying No SQL databases on a central node and an edge node of the constructed distributed architecture to form a distributed database cluster;
the central node configures redis cache middleware on the deployed No SQL database and sends configuration information to the edge node;
And the edge node receives the configuration information of the redis cache middleware of the center node, and carries out corresponding configuration on the local No SQL database.
5. An open source Hongmon system based energy storage method as defined in claim 4, wherein:
according to the preprocessed real-time data, the edge node performs feature extraction through a parallelized machine learning model, and the method comprises the following steps:
The edge node divides the preprocessed real-time current and voltage data of the energy storage device into a plurality of data blocks according to time windows; inputting the data blocks into a plurality of convolutional neural networks;
Each convolutional neural network independently trains and updates network parameters, and outputs the mean value and variance of the residual life RUL of the energy storage equipment of the corresponding data block;
The edge node aggregates the RUL mean value and the variance of each data block to obtain health characteristic data of the energy storage equipment;
The edge node sends the obtained health characteristic data to the central node for storage through a distributed soft bus network component of the hong Mongolian system;
The edge node organizes the real-time current and voltage data of the preprocessed energy storage equipment according to time sequence, and inputs the data into the circulating neural network;
The circulating neural network obtains a long-term dependency relationship in the time series data through a gate control circulating unit GRU, and outputs a predicted value of the self-discharge rate and the internal resistance of the energy storage equipment;
The edge node compares the predicted value of the self-discharge rate and the internal resistance output by the cyclic neural network with the historical value to obtain the increasing trend of the self-discharge rate and the internal resistance as the performance attenuation characteristic of the energy storage equipment;
The edge nodes send the obtained performance attenuation characteristics to the central node for storage through the distributed soft bus network component of the hong Monte system.
6. An open source hong Meng system based energy storage method according to any one of claims 1 to 5, wherein:
A distributed soft bus network component, comprising: the system comprises a message integration module, a service registration module and a request routing module;
the message integration module receives the custom protocol data of the edge node or the center node and converts the custom protocol data into a standard protocol format according to a preset coding mapping rule; the converted standard protocol format is sent to a service registration module for service access;
The service registration module receives a service access request and stores service instance information into a centralized etcd database;
The request routing module acquires service instance information corresponding to the access request by inquiring the etcd database according to the service access request; according to the acquired service instance information, routing the access request to the node where the service instance is located by adopting a consistent hash algorithm, and carrying out load balancing; and converting the service response result data into a custom protocol format, and returning the custom protocol format to the edge node or the center node initiated by the request.
7. An open source Hongmon system based energy storage method as defined in claim 6, wherein:
Transmitting the extracted feature data to a distributed database on a central node via fragmented transmission of the hong system, comprising:
The edge node compresses the obtained health degree characteristic data and performance attenuation characteristic data of the energy storage equipment by adopting a Huffman coding algorithm;
The edge node divides the compressed characteristic data into a plurality of data blocks by fibonacci dividing method, and adds CRC32 check code to each data block;
the edge node generates an identification code for each data block by using an Object ID algorithm of MongoDB; the identification code comprises a time stamp and a sequence number;
According to the identification code of the data block, the edge node adopts a APACHE KAFKA partitioning mechanism to send the data block with the same time stamp prefix to the same partition;
The kafka consumer of the central node receives the data blocks according to the subareas and classifies the data blocks belonging to the same original characteristic data through the identification codes;
The central node sorts and splices the classified data blocks according to the time stamp and the sequence number of the identification code, and original characteristic data are obtained through a Huffman decoding algorithm;
The central node stores the obtained original characteristic data into a distributed database MongoDB, and constructs a data set indexed by device ID and time stamp in the MongoDB database.
8. An open source Hongmon system based energy storage method as defined in claim 7, wherein:
the edge node trains a predictive model for predicting energy storage device capacity based on historical state data using a machine learning model of the hong Monte system, comprising:
The edge node acquires historical state data of the energy storage device from a distributed database MongoDB, wherein the historical state data comprises current and voltage time sequence data in the past year;
The edge node adopts a statistical analysis method based on differential entropy to calculate the differential entropy value in each time window of the historical state data;
the edge node is provided with sliding windows, and the differential entropy value of each window forms a differential entropy characteristic vector; by sliding a time window on a time axis, a differential entropy feature vector sequence is obtained;
the edge node takes the differential entropy feature vector sequence as input, a plurality of clusters are divided by a BIRCH incremental clustering algorithm, and the center point of the cluster reflects the mode of the state of the energy storage device under different working conditions;
the edge node adopts a z-score standardization method to normalize the data in each cluster;
the edge node takes the cluster data after normalization processing as a training set and is used for training a prediction model;
And constructing a sequence prediction model comprising a multi-layer LSTM network and residual connection, and performing model training by using the obtained training set to obtain a prediction model for predicting the capacity of the energy storage equipment.
9. An open source Hongmon system based energy storage method as defined in claim 8, wherein:
According to the preprocessed real-time data, the extracted characteristic data and the trained prediction model, the central node generates a control instruction of the energy storage device, and the method comprises the following steps:
The center node receives the energy storage equipment real-time state data preprocessed by the edge node;
the central node acquires health characteristics and performance attenuation characteristics of the energy storage equipment from a distributed database MongoDB;
the central node aligns the acquired real-time state data, health degree characteristics and performance attenuation characteristics according to time and then takes the aligned real-time state data, health degree characteristics and performance attenuation characteristics as input vectors; predicting through the trained prediction model to obtain a prediction curve of the capacity change of the energy storage equipment in a period of time in the future;
The central node extracts peak-valley characteristics of the obtained prediction curve, and a multi-scale analysis method based on wavelet transformation is adopted to obtain local maximum value points and minimum value points of the prediction curve;
the central node calculates control parameters for adjusting the charge and discharge current and voltage of the energy storage device by adopting an incremental PID algorithm according to the acquired local maximum value point and the acquired local minimum value point;
the central node encapsulates the calculated control parameters into control instructions, the control instructions are issued to the edge nodes, the edge nodes convert the control instructions into CAN bus frame formats, and the energy storage equipment is managed.
10. An open source hong Meng system based energy storage method as defined in claim 9, wherein:
the incremental PID algorithm adopts a self-adaptive incremental PID control algorithm based on particle swarm optimization.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410419230.5A CN118017564B (en) | 2024-04-09 | 2024-04-09 | Energy storage method based on open source hong Meng system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410419230.5A CN118017564B (en) | 2024-04-09 | 2024-04-09 | Energy storage method based on open source hong Meng system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN118017564A true CN118017564A (en) | 2024-05-10 |
| CN118017564B CN118017564B (en) | 2024-06-21 |
Family
ID=90943282
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410419230.5A Active CN118017564B (en) | 2024-04-09 | 2024-04-09 | Energy storage method based on open source hong Meng system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118017564B (en) |
Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN211790786U (en) * | 2019-12-19 | 2020-10-27 | 中国电力科学研究院有限公司 | Client side energy storage energy management system |
| CN112543429A (en) * | 2020-11-25 | 2021-03-23 | 北京科技大学 | Configurable edge computing gateway with autonomous decision making function |
| CN113094174A (en) * | 2021-04-09 | 2021-07-09 | 贵州电网有限责任公司 | Method for coordination control of power distribution network terminal autonomous networking based on edge computing technology |
| CN114422371A (en) * | 2022-01-20 | 2022-04-29 | 重庆邮电大学 | Elastic micro-service system based on distributed and container virtualization and implementation method |
| CN114416415A (en) * | 2022-01-17 | 2022-04-29 | 南京新联电子股份有限公司 | Remote online fault detection method and system for Hongmon operating system and storage medium |
| CN114979142A (en) * | 2022-05-16 | 2022-08-30 | 深圳市金瑞铭科技有限公司 | Data processing method and system for edge equipment of cloud platform |
| US11444338B1 (en) * | 2021-11-30 | 2022-09-13 | Knoetik Solutions, Inc. | Smart battery system |
| CN115392373A (en) * | 2022-08-25 | 2022-11-25 | 浙江英集动力科技有限公司 | Deep reinforcement learning-based energy management method for multi-region comprehensive energy system |
| CN115733246A (en) * | 2022-11-11 | 2023-03-03 | 阳光电源股份有限公司 | Energy storage power station data monitoring and analyzing system based on edge cloud |
| CN115827579A (en) * | 2023-01-28 | 2023-03-21 | 华北电力大学 | Distributed energy storage informatization control and operation platform architecture |
| CN115993549A (en) * | 2021-11-17 | 2023-04-21 | 江苏天合储能有限公司 | Correction method, device, system and medium of energy storage battery management system |
| CN117335557A (en) * | 2023-09-04 | 2024-01-02 | 华为数字能源技术有限公司 | Electric power measurement and control device and power generation system |
| CN117408470A (en) * | 2023-10-27 | 2024-01-16 | 国网江苏综合能源服务有限公司 | Collaborative optimization method for comprehensive energy monitoring operation based on edge autonomy |
| CN117498400A (en) * | 2024-01-03 | 2024-02-02 | 长峡数字能源科技(湖北)有限公司 | Distributed photovoltaic and energy storage data processing method and system |
| CN117596247A (en) * | 2023-11-28 | 2024-02-23 | 中移(苏州)软件技术有限公司 | Resource monitoring and performance evaluation method based on heterogeneous edge computing system |
| CN117674139A (en) * | 2024-01-30 | 2024-03-08 | 国网辽宁省电力有限公司丹东供电公司 | Internet of things-based distributed energy management method and system |
-
2024
- 2024-04-09 CN CN202410419230.5A patent/CN118017564B/en active Active
Patent Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN211790786U (en) * | 2019-12-19 | 2020-10-27 | 中国电力科学研究院有限公司 | Client side energy storage energy management system |
| CN112543429A (en) * | 2020-11-25 | 2021-03-23 | 北京科技大学 | Configurable edge computing gateway with autonomous decision making function |
| CN113094174A (en) * | 2021-04-09 | 2021-07-09 | 贵州电网有限责任公司 | Method for coordination control of power distribution network terminal autonomous networking based on edge computing technology |
| CN115993549A (en) * | 2021-11-17 | 2023-04-21 | 江苏天合储能有限公司 | Correction method, device, system and medium of energy storage battery management system |
| US11444338B1 (en) * | 2021-11-30 | 2022-09-13 | Knoetik Solutions, Inc. | Smart battery system |
| CN114416415A (en) * | 2022-01-17 | 2022-04-29 | 南京新联电子股份有限公司 | Remote online fault detection method and system for Hongmon operating system and storage medium |
| CN114422371A (en) * | 2022-01-20 | 2022-04-29 | 重庆邮电大学 | Elastic micro-service system based on distributed and container virtualization and implementation method |
| CN114979142A (en) * | 2022-05-16 | 2022-08-30 | 深圳市金瑞铭科技有限公司 | Data processing method and system for edge equipment of cloud platform |
| CN115392373A (en) * | 2022-08-25 | 2022-11-25 | 浙江英集动力科技有限公司 | Deep reinforcement learning-based energy management method for multi-region comprehensive energy system |
| CN115733246A (en) * | 2022-11-11 | 2023-03-03 | 阳光电源股份有限公司 | Energy storage power station data monitoring and analyzing system based on edge cloud |
| CN115827579A (en) * | 2023-01-28 | 2023-03-21 | 华北电力大学 | Distributed energy storage informatization control and operation platform architecture |
| CN117335557A (en) * | 2023-09-04 | 2024-01-02 | 华为数字能源技术有限公司 | Electric power measurement and control device and power generation system |
| CN117408470A (en) * | 2023-10-27 | 2024-01-16 | 国网江苏综合能源服务有限公司 | Collaborative optimization method for comprehensive energy monitoring operation based on edge autonomy |
| CN117596247A (en) * | 2023-11-28 | 2024-02-23 | 中移(苏州)软件技术有限公司 | Resource monitoring and performance evaluation method based on heterogeneous edge computing system |
| CN117498400A (en) * | 2024-01-03 | 2024-02-02 | 长峡数字能源科技(湖北)有限公司 | Distributed photovoltaic and energy storage data processing method and system |
| CN117674139A (en) * | 2024-01-30 | 2024-03-08 | 国网辽宁省电力有限公司丹东供电公司 | Internet of things-based distributed energy management method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN118017564B (en) | 2024-06-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111177095B (en) | Log analysis method, device, computer equipment and storage medium | |
| CN109564568B (en) | Apparatus, method and machine-readable storage medium for distributed dataset indexing | |
| US10169720B2 (en) | Systems and methods for machine learning using classifying, clustering, and grouping time series data | |
| CN105069703B (en) | A kind of electrical network mass data management method | |
| CN115412947B (en) | Fault simulation method and system based on digital twin and AI algorithm | |
| CN110798517B (en) | Decentralized cluster load balancing method and system, mobile terminal and storage medium | |
| US11429566B2 (en) | Approach for a controllable trade-off between cost and availability of indexed data in a cloud log aggregation solution such as splunk or sumo | |
| CN113900810A (en) | Distributed graph processing method, system and storage medium | |
| CN115358487A (en) | Federal learning aggregation optimization system and method for power data sharing | |
| CN112148578A (en) | IT fault defect prediction method based on machine learning | |
| CN113407600B (en) | Enhanced real-time calculation method for dynamically synchronizing multi-source large table data in real time | |
| CN116502162A (en) | Abnormal computing power federal detection method, system and medium in edge computing power network | |
| CN112667747B (en) | Dynamic configuration multi-database distributed persistence method supporting user-defined plug-in | |
| CN113541986B (en) | Fault prediction method and device for 5G slice and computing equipment | |
| CN118017564B (en) | Energy storage method based on open source hong Meng system | |
| CN113360353A (en) | Test server and cloud platform | |
| CN112925964A (en) | Big data acquisition method based on cloud computing service and big data acquisition service system | |
| CN116226139B (en) | Distributed storage and processing method and system suitable for large-scale ocean data | |
| CN111736774A (en) | Redundant data processing method and device, server and storage medium | |
| CN107463540B (en) | Electric energy quality data processing method and electric energy quality monitoring device | |
| CN114296785A (en) | Log data modeling method and system | |
| CN112925831A (en) | Big data mining method and big data mining service system based on cloud computing service | |
| CN115374101A (en) | Rail transit station level data management system | |
| Wang et al. | A Storage Optimization Model for Cloud Servers in Integrated Communication, Sensing, and Computation System | |
| CN116600090B (en) | Operation and maintenance control method and system for IPC network camera |
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 |