A method and system for operation and maintenance management of substation battery pack based on artificial intelligence
By adopting artificial intelligence-based methods in substations, the partition monitoring network and battery diagnosis model is solved, and the problem of lack of independent perception and automatic decision-making of battery pack management systems in the existing technology is solved, and refined monitoring and intelligence of battery packs are realized Integrated management improves operating efficiency and safety.
Patent Information
- Application Number
- CN202411181580.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-27
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-08-27
AI Technical Summary
The existing substation battery pack management system lacks independent perception and automatic decision-making capabilities, resulting in low efficiency in monitoring data processing and analysis, and the inability to detect potential problems or optimization opportunities in time. Traditional fault diagnosis methods cannot achieve real-time monitoring and accurate diagnosis, and lack intelligent algorithms. and big data technology support.
Using an artificial intelligence-based method, the battery pack is divided into battery strings, a partition monitoring network is built, battery information is collected in real time and monitoring performance is evaluated based on self-perception model, and a battery diagnosis model with learning evolution is used to analyze health status and output regulatory decisions. And optimize and adjust the monitoring network and parameters.
It realizes refined monitoring and management of the operating status of the battery pack, improves the accuracy and timeliness of fault diagnosis, provides intelligent management and optimized scheduling, improves operating efficiency, safety and reliability, and ensures the stable operation of the power grid. .
Smart Images

Figure CN118710257B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of substation operation and maintenance management, and in particular to an artificial intelligence-based substation battery pack operation and maintenance management method and system. Background Art
[0002] The battery pack is one of the important equipment in the substation, which is used to store electric energy and release it when needed to cope with load fluctuations in the power grid, power system failures and other situations, and to ensure the stable operation of the power grid. It is mainly composed of a series of batteries, usually using lead-acid, nickel-cadmium and lithium-ion battery technologies, with the characteristics of fast response, high energy density and long life. The battery pack plays an important role in the power system as a backup power source, peak load shifting and valley filling, and emergency backup, and plays an irreplaceable role in the safe and stable operation of the power grid.
[0003] With the continuous development of power systems and the growth of power demand, the operation and maintenance management technology of substation battery packs is also constantly improving. As an important part of the power system, the development of operation and maintenance management technology of substation battery packs is closely related to the safe and stable operation of substations. The operation and maintenance management of battery packs includes regular inspection, maintenance, monitoring and data analysis of battery packs to ensure their performance, safety and reliability.
[0004] Although substation battery technology has made significant progress in the past few years, there are still a series of shortcomings. First, the current battery management system often relies on manual monitoring and control, lacks autonomous perception and automatic decision-making capabilities, resulting in low efficiency in the processing and analysis of monitoring data, and cannot timely discover potential problems or optimization opportunities. Secondly, traditional fault diagnosis methods often rely on manual experience and regular inspections, and cannot achieve real-time monitoring and accurate diagnosis of the operating status of the battery pack, which can easily lead to missed faults or false alarms. In addition, the existing battery management system lacks intelligent algorithms and technical support in data analysis and decision optimization, and cannot make full use of big data and artificial intelligence technologies for data mining and pattern recognition, thus failing to achieve intelligent analysis and optimized scheduling of system operation. Summary of the invention
[0005] Based on this, it is necessary to provide an artificial intelligence-based substation battery group operation and maintenance management method and system to address the above technical problems.
[0006] In a first aspect, the present invention provides a method for operation and maintenance management of a substation battery pack based on artificial intelligence, the method comprising the following steps:
[0007] S1. Divide the battery pack into several battery strings, use a single battery string as a monitoring node, install independent monitoring equipment and sub-control switches, and build a partition monitoring network;
[0008] S2. Collect battery information during the operation of the battery pack in real time, store it locally and in the database, and evaluate the monitoring performance of the partition monitoring network based on the self-perception model;
[0009] S3. Analyze the health status of the battery pack based on the battery diagnosis model with learning evolution capability, and output the control decision of the battery pack according to the health status analysis result;
[0010] S4. Based on the monitoring performance evaluation results and control decisions, optimize and adjust the configuration parameters of the partition monitoring network and battery pack, record the parameter adjustment results, and automatically generate an operation and maintenance management report.
[0011] Furthermore, the battery information of the battery pack during operation is collected in real time, stored locally and in a database, and the monitoring performance of the partition monitoring network is evaluated based on the self-perception model, including the following steps:
[0012] S21, using monitoring equipment to collect battery information of corresponding monitoring nodes in real time, and store it in local storage in priority according to time sequence, and then transmit it to the supervision center through a communication link, wherein the battery information includes temperature, battery cell voltage, battery cell current, battery string voltage, battery string current, battery cell internal resistance and switch status;
[0013] S22, based on the pre-set battery information integration standard and classification principle, process the battery information uploaded by each monitoring node and store it in the database;
[0014] S23. Set a self-sensing period for the partition monitoring network. During the self-sensing period, calculate the information entropy gain of each monitoring node based on the self-sensing model to evaluate the monitoring performance of each monitoring node in the partition monitoring network.
[0015] Furthermore, based on the pre-set battery information integration standard and classification principle, processing the battery information uploaded by each monitoring node and storing it in the database includes the following steps:
[0016] S221. Based on the integration standard and classification principle, the battery information is divided into corresponding attribute items, and the respective attribute labels are marked, and then format errors and duplicate data are eliminated;
[0017] S222. Formulate a unified coding standard and use a return-to-zero code method to encode the battery information so that the battery information meets the standardized input standard;
[0018] S223. Associate the battery information with the corresponding collection time and upload time, and store them in a database in the supervision center in the form of time series.
[0019] Furthermore, based on the self-perception model, the information entropy gain of each monitoring node is calculated, and the monitoring performance of each monitoring node in the partition monitoring network is evaluated, including the following steps:
[0020] S231, establishing a self-perception model based on a Bayesian network, and using the self-perception model to extract the dependency relationship and conditional probability between the monitoring node and its locally stored battery information;
[0021] S232, calculating the local condition entropy of each monitoring node using the local condition entropy formula;
[0022] S233, calculating the online condition entropy of each monitoring node using the online condition entropy formula;
[0023] S234. The difference between the online condition entropy and the local condition entropy is taken as the information entropy gain of the monitoring node, and compared with the preset information entropy gain threshold. If the information entropy gain of the monitoring node is less than the information entropy gain threshold, the monitoring performance of the monitoring node is determined to be qualified. If the information entropy gain of the monitoring node is greater than or equal to the information entropy gain threshold, the monitoring performance of the monitoring node is determined to be unqualified and there is a monitoring performance defect.
[0024] Furthermore, the local conditional entropy formula is:
[0025] ;
[0026] In the formula, A represents the local conditional entropy, H (·) represents information entropy, P (·) represents conditional probability, R i Indicates i Monitoring nodes, parent ( R i ) indicates the i Battery information stored locally in each monitoring node;
[0027] The online conditional entropy formula is:
[0028] ;
[0029] In the formula, B represents the online conditional entropy, D Represents the collection of battery information uploaded by monitoring nodes stored in the database.
[0030] Furthermore, based on the battery diagnosis model with learning evolution capability, the health status of the battery pack is analyzed, and according to the health status analysis result, the regulation decision of the battery pack is output, which includes the following steps:
[0031] S31, obtaining an online model library for battery pack health status assessment, using historical battery information and corresponding historical health status in the online model library as a source domain, and merging battery information of all monitoring nodes currently monitored by the battery pack into a battery data set as a target domain;
[0032] S32, setting the maximum mean difference index, evaluating the similarity between the sample data in the source domain and the target domain, and searching for the optimal historical model;
[0033] S33, extracting effective samples of the optimal historical model, constructing a mapping relationship from the source domain to the target domain, migrating historical knowledge information to the new battery diagnosis model, and performing model adaptive optimization;
[0034] S34, based on the iterative clustering strategy, retain the newly added sample data and update the online model library;
[0035] S35, inputting the battery information into the battery diagnosis model, analyzing the health status of each battery string in the current battery pack, and matching the analysis results with corresponding regulation decisions.
[0036] Furthermore, the maximum mean difference index is set to evaluate the similarity between the sample data in the source domain and the target domain. The search for the optimal historical model includes the following steps:
[0037] S321, sorting the sample data in the source domain according to the update time, and clustering the sample data in the source domain and the target domain using the K-means clustering algorithm in turn;
[0038] S322, using the maximum mean difference formula to calculate the maximum mean difference index between each source domain sample class and the target domain sample class;
[0039] S323, calculating the average value of the maximum mean difference index of all sample classes in the target domain, and selecting the reciprocal of the average value as the similarity degree between the source domain sample data and the target domain sample data;
[0040] S324, setting a similarity threshold, selecting source domain sample data with a similarity degree greater than the similarity threshold as the evaluation result, and matching the optimal historical model corresponding to the number of source domain samples.
[0041] Furthermore, extracting effective samples of the optimal historical model, constructing a mapping relationship from the source domain to the target domain, migrating historical knowledge information to the new battery diagnosis model, and performing model adaptive optimization include the following steps:
[0042] S331, combining all sample data of the target domain to form a target domain population, combining the matched source domain sample data to form a source domain population, and establishing a mapping relationship between the two populations;
[0043] S332, setting the upper limit and lower limit of the proportion of individuals migrating from the source domain population to the target domain population, calculating the migration ratio of the source domain population to the target domain population, and selecting a specific number of valid sample individuals from the source domain population according to the migration ratio. The migration ratio formula is:
[0044] ;
[0045] In the formula, G represents the migration ratio, S Indicates the similarity between the source domain sample data and the target domain sample data. p max represents the upper limit of the proportion of individuals migrating from the source domain population to the target domain population, p min It represents the lower limit of the proportion of individuals migrating from the source domain population to the target domain population;
[0046] S333, migrating valid sample individuals as historical knowledge information to the target domain population, forming a new target domain data population, and constructing a battery diagnosis model;
[0047] S334. Based on the particle swarm optimization algorithm, an adaptively optimized battery diagnosis model is obtained.
[0048] Furthermore, based on the iterative clustering strategy, the newly added sample data is retained and the online model library is updated, including the following steps:
[0049] S341, setting the update frequency and cycle duration of the online model library;
[0050] S342, clustering the newly generated target domain population using the K-means clustering algorithm, and selecting the individual with the smallest objective function value in the population, saving it to the representative individual set, and then calculating the Euclidean distance between two individuals in the representative individual set, and selecting the minimum distance value as the initial cluster radius;
[0051] S343. According to the update frequency and cycle duration, the representative individual set and initial class radius in each target domain population are regularly sampled as content updates of the online model library.
[0052] In a second aspect, the present invention also provides an artificial intelligence-based substation battery group operation and maintenance management system, the system comprising:
[0053] The monitoring module is used to divide the battery pack into several battery strings, and use a single battery string as a monitoring node, install independent monitoring equipment and sub-control switches, and build a partition monitoring network;
[0054] The perception module is used to collect battery information during the operation of the battery pack in real time, store it locally and in the database, and evaluate the monitoring performance of the partition monitoring network based on the self-perception model;
[0055] A decision module is used to analyze the health status of the battery pack based on a battery diagnosis model with learning and evolution capabilities, and output a control decision for the battery pack based on the health status analysis results;
[0056] Configuration module, which is used to optimize and adjust the configuration parameters of the partition monitoring network and battery pack based on the monitoring performance evaluation results and control decisions, record the parameter adjustment results, and generate operation and maintenance management reports;
[0057] Among them, the monitoring module, the perception module, the decision module and the configuration module are connected in sequence.
[0058] The beneficial effects of the present invention are:
[0059] 1. By dividing the battery group into battery strings and building a partition monitoring network, refined monitoring and management of the operating status of the battery group can be achieved; real-time battery information is collected and the monitoring network performance is evaluated based on the self-perception model to ensure the accuracy and stability of the monitoring system; the battery diagnosis model based on learning evolutionary capabilities is used to analyze the health status of the battery group and output control decisions, which improves the accuracy and timeliness of system fault diagnosis. On this basis, the control decision is optimized to adjust the monitoring network and parameters, realizing intelligent management and optimized scheduling of the battery group, providing convenience for decision support and operation and maintenance management, and comprehensively improving the operating efficiency, safety and reliability of the substation battery group.
[0060] 2. The information entropy gain of each monitoring node is calculated through the self-perception model, and the monitoring performance of each monitoring node in the partition monitoring network is evaluated, which helps operation and maintenance personnel to promptly discover whether the monitoring capability of the monitoring node meets the standard, so as to timely adjust and optimize the monitoring network, improve the monitoring effect and management level of the battery group operating status, and ensure the safe and stable operation of the power grid.
[0061] 3. By acquiring historical battery information and health status from the online model library, migrating historical knowledge information to a new battery diagnostic model, and performing adaptive model optimization, the diagnosis and evaluation of the health status of the battery pack can be realized. At the same time, the online model library is updated through an iterative clustering strategy to keep the model updated and optimized, thereby improving the accuracy and reliability of the diagnostic model, providing a scientific basis and decision-making support for the operation and maintenance management of the battery pack, and improving the safety, stability and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS
[0062] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:
[0063] Figure 1 The present invention is a flowchart of an artificial intelligence-based substation battery group operation and maintenance management method according to an embodiment of the present invention.
[0064] Figure 2 The present invention is a system principle block diagram of an artificial intelligence-based substation battery group operation and maintenance management system according to an embodiment of the present invention.
[0065] Figure numbers: 1. Monitoring module; 2. Perception module; 3. Decision module; 4. Configuration module. DETAILED DESCRIPTION
[0066] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0067] See also Figure 1 , provides a substation battery group operation and maintenance management method based on artificial intelligence, the method comprising the following steps:
[0068] S1. Divide the battery pack into several battery strings, use a single battery string as a monitoring node, install independent monitoring equipment and sub-control switches, and build a partition monitoring network.
[0069] In the description of the present invention, the batteries in the battery pack are connected in series to form a plurality of battery strings, each of which contains a plurality of battery cells, and the number of battery strings is determined according to the system design and the capacity of the battery pack.
[0070] Each battery string is regarded as a monitoring node, that is, a separate monitoring unit. Each monitoring node is equipped with independent monitoring equipment to monitor the key parameters of the battery string such as voltage, current, temperature, etc. in real time. Corresponding monitoring equipment is installed at each monitoring node, including sensors, data collectors and other equipment, which can accurately collect and transmit the operating data of the battery string for subsequent analysis and management.
[0071] The sub-control switch of the battery string is a device used to control the charge and discharge status of each battery string and protect the battery string from overcharge, over-discharge and other problems. It is installed at the connection end of each battery string and serves as a switch control device between the battery string and the external circuit. Its main functions include:
[0072] 1. The sub-control switch can control the charging and discharging process of the battery string according to system requirements. When charging, it can monitor the voltage and current of the battery string and control the size of the charging current to avoid overcharging and damage to the battery. When discharging, it can ensure that the current output of the battery string is within a safe range to prevent damage caused by over-discharging of the battery.
[0073] 2. The sub-control switch has functions such as overvoltage protection, undervoltage protection, and overcurrent protection. It can disconnect the battery string from the external circuit in time, protect the battery from overcharging, overdischarging, overloading and other problems, and ensure the safe operation of the battery pack.
[0074] 3. For multiple battery cells connected in series, the sub-control switch can realize the balance control of the battery string. By adjusting the charge and discharge state of each battery cell, the voltage between each battery cell is kept in a relatively balanced state, thus improving the overall performance and service life of the battery pack.
[0075] S2. Collect battery information during the operation of the battery pack in real time, store it locally and in the database respectively, and evaluate the monitoring performance of the partition monitoring network based on the self-perception model.
[0076] In the description of the present invention, real-time collection of battery information during the operation of the battery pack, storage in local and database, and evaluation of the monitoring performance of the partition monitoring network based on the self-perception model include the following steps:
[0077] S21. Use monitoring equipment to collect battery information of corresponding monitoring nodes in real time, and store it in local storage in time sequence, and then transmit it to the supervision center through the communication link. The battery information includes temperature, battery cell voltage, battery cell current, battery string voltage, battery string current, battery cell internal resistance and switch status.
[0078] S22. Based on the pre-set battery information integration standard and classification principle, process the battery information uploaded by each monitoring node and store it in the database.
[0079] In the description of the present invention, based on the pre-set integration standard and classification principle of battery information, processing the battery information uploaded by each monitoring node and storing it in the database includes the following steps:
[0080] S221. Based on the integration standard and classification principle, the battery information is divided into corresponding attribute items, and the respective attribute labels are marked, and then format errors and duplicate data are eliminated.
[0081] Specifically, the integration standards and classification principles include determining the battery information items that need to be collected, such as temperature, voltage, current, internal resistance, etc., as well as the classification and attribute definition of each item. The battery information uploaded from the monitoring node is divided according to the integration standards and classification principles, and each information is assigned to the corresponding attribute item. For example, the collected temperature information is classified into the temperature attribute item, the voltage information is classified into the voltage attribute item, and so on. For each attribute item, the corresponding attribute label needs to be calibrated for subsequent identification and processing. The attribute label uses a phrase or a simple identifier that can clearly express the meaning and function of the attribute item.
[0082] After the attribute items and attribute labels are divided and calibrated, the battery information is checked and screened. Data with incorrect formats are eliminated, including data types that do not meet requirements and data range abnormalities. At the same time, duplicate data is excluded to ensure the uniqueness of each piece of data.
[0083] S222. Develop a unified coding standard and use the return-to-zero code method to encode the battery information so that the battery information meets the standardized entry standard.
[0084] Specifically, we first need to determine the coding principle, that is, determine the coding rules for each attribute, including considerations such as the data type and value range of the attribute. For example, for the temperature attribute, it is stipulated that it must be encoded with a certain accuracy within a certain range. Then, for each attribute of the battery information, formulate corresponding coding rules. According to the formulated coding rules, encode each attribute in the battery information, map the attribute value to the corresponding coding value, and ensure the uniqueness and identifiability of the coding. In the return-to-zero code method, a reference value is selected, and the coding values of all attributes are encoded relative to this reference value.
[0085] S223. Associate the battery information with the corresponding collection time and upload time, and store them in a database in the supervision center in the form of time series.
[0086] S23. Set a self-sensing period for the partition monitoring network. During the self-sensing period, calculate the information entropy gain of each monitoring node based on the self-sensing model to evaluate the monitoring performance of each monitoring node in the partition monitoring network.
[0087] In the description of the present invention, based on the self-perception model, calculating the information entropy gain of each monitoring node and evaluating the monitoring performance of each monitoring node in the partition monitoring network include the following steps:
[0088] S231. Establish a self-perception model based on the Bayesian network, and use the self-perception model to extract the dependency and conditional probability between the monitoring node and its locally stored battery information.
[0089] Among them, the Bayesian Network is a probabilistic graphical model used to represent the dependency relationship between random variables. The Bayesian network dependency model refers to a model that uses the Bayesian network to describe the dependency relationship between variables and uses this dependency relationship for reasoning and decision-making.
[0090] In a Bayesian network, variables are represented as nodes, and dependencies between nodes are represented as directed edges. Each node represents a random variable, and the state (value) of the node represents the possible values of the random variable. Directed edges represent dependencies between variables, that is, the state of a node depends on the state of its parent node. Each node is also associated with a probability distribution, which describes the conditional probability distribution of the node state given the state of its parent node.
[0091] In the present invention, variables used to describe monitoring nodes and battery information are determined, including battery information such as temperature, voltage, current, and the relationship between monitoring nodes, and each variable corresponds to a node in the Bayesian network. Next, the structure of the network is designed according to the characteristics of the problem and prior knowledge. The connection between nodes in the Bayesian network represents the dependency between variables. According to the actual situation, the dependency between monitoring nodes and battery information is determined, and the network structure is constructed.
[0092] For each node, determine the conditional probability distribution given its parent node. The conditional probability needs to be determined by historical data analysis. For example, estimate the probability distribution of battery current at a specific temperature by analyzing historical data. Finally, use the collected data to learn the network parameters, that is, determine the conditional probability distribution of each node.
[0093] S232. Calculate the local condition entropy of each monitoring node using the local condition entropy formula.
[0094] In the description of the present invention, the local conditional entropy formula is:
[0095] .
[0096] In the formula, A represents the local conditional entropy, H (·) represents information entropy, P (·) represents conditional probability, R i Indicates i Monitoring nodes, parent ( R i ) indicates the i The battery information is stored locally on each monitoring node.
[0097] S233. Calculate the online condition entropy of each monitoring node using the online condition entropy formula.
[0098] Among them, the online conditional entropy formula is:
[0099] .
[0100] In the formula, B represents the online conditional entropy, D Represents the collection of battery information uploaded by monitoring nodes stored in the database.
[0101] S234. The difference between the online condition entropy and the local condition entropy is taken as the information entropy gain of the monitoring node, and compared with the preset information entropy gain threshold. If the information entropy gain of the monitoring node is less than the information entropy gain threshold, the monitoring performance of the monitoring node is determined to be qualified. If the information entropy gain of the monitoring node is greater than or equal to the information entropy gain threshold, the monitoring performance of the monitoring node is determined to be unqualified and there is a monitoring performance defect.
[0102] Specifically, for each monitoring node, online conditional entropy refers to the conditional entropy calculated using online collected data, which reflects the monitoring performance of the current monitoring node; local conditional entropy refers to the conditional entropy calculated using locally stored historical data, which reflects the benchmark performance of the monitoring node. The difference between the online conditional entropy and the local conditional entropy is taken as the information entropy gain of the monitoring node. The information entropy gain represents the information gain brought by using online data relative to using local data, and is one of the indicators for evaluating monitoring performance.
[0103] The calculated information entropy gain is compared with the preset information entropy gain threshold. The information entropy gain threshold is a threshold set according to system requirements and performance indicators to determine whether the monitoring performance of the monitoring node is qualified. If the information entropy gain of the monitoring node is less than the information entropy gain threshold, the monitoring performance of the monitoring node is judged to be qualified, indicating that its online data collection capability is relatively good; if the information entropy gain is greater than or equal to the information entropy gain threshold, the monitoring performance of the monitoring node is judged to be unqualified, there are monitoring performance defects, and further investigation and improvement are needed.
[0104] S3. Based on the battery diagnosis model with learning and evolution capabilities, the health status of the battery pack is analyzed, and the control decision of the battery pack is output according to the health status analysis results.
[0105] In the description of the present invention, based on the battery diagnosis model with learning evolution capability, analyzing the health status of the battery pack, and outputting the regulation decision of the battery pack according to the health status analysis result includes the following steps:
[0106] S31. Obtain an online model library for battery pack health status assessment, use historical battery information and corresponding historical health status in the online model library as the source domain, and merge battery information of all monitoring nodes currently monitored by the battery pack into a battery data set as the target domain.
[0107] The online model library is a database or data warehouse that stores historical data and corresponding health status, which is used for battery pack health status assessment and prediction. In this library, the battery information collected in the past and the corresponding health status labels are saved.
[0108] S32. Set the maximum mean difference index, evaluate the similarity between the sample data in the source domain and the target domain, and search for the optimal historical model.
[0109] In the description of the present invention, the maximum mean difference index is set to evaluate the similarity between the sample data in the source domain and the target domain, and the search for the optimal historical model includes the following steps:
[0110] S321 , sorting the sample data in the source domain according to the update time, and clustering the sample data in the source domain and the target domain using the K-means clustering algorithm in turn.
[0111] Specifically, the data in the source domain is divided into several different categories, and the K-means clustering algorithm is used. The algorithm divides the data into K categories, where K is a pre-specified parameter. Using the K-means clustering algorithm to cluster the sample data in the source domain can group similar data points into the same category and determine the center point of each category.
[0112] S322. Use the maximum mean difference formula to calculate the maximum mean difference index between each source domain sample class and the target domain sample class.
[0113] Among them, Maximum Mean Discrepancy (MMD) is a method used to measure the similarity between two sample sets, and is often used in kernel methods. In machine learning, MMD is widely used in tasks such as domain adaptation, kernel selection, and generative adversarial networks (GANs). The core idea of MMD is to judge the similarity between two sample sets by comparing their sample distributions.
[0114] S323. Calculate the average value of the maximum mean difference index of all sample classes in the target domain, and select the reciprocal of the average value as the similarity degree between the source domain sample data and the target domain sample data.
[0115] S324, setting a similarity threshold, selecting source domain sample data with a similarity degree greater than the similarity threshold as the evaluation result, and matching the optimal historical model corresponding to the number of source domain samples.
[0116] S33. Extract valid samples of the optimal historical model, construct a mapping relationship from the source domain to the target domain, migrate historical knowledge information to the new battery diagnosis model, and perform model adaptive optimization.
[0117] In the description of the present invention, extracting effective samples of the optimal historical model, constructing a mapping relationship from the source domain to the target domain, migrating historical knowledge information to a new battery diagnosis model, and performing model adaptive optimization include the following steps:
[0118] S331. All sample data of the target domain are combined to form a target domain population, and the matched source domain sample data are combined to form a source domain population, and a mapping relationship between the two populations is established.
[0119] S332, setting the upper limit and lower limit of the proportion of individuals migrating from the source domain population to the target domain population, calculating the migration ratio of the source domain population to the target domain population, and selecting a specific number of valid sample individuals from the source domain population according to the migration ratio. The migration ratio formula is:
[0120] .
[0121] In the formula, G represents the migration ratio, S Indicates the similarity between the source domain sample data and the target domain sample data. p max represents the upper limit of the proportion of individuals migrating from the source domain population to the target domain population, p min Indicates the lower limit of the proportion of individuals migrating from the source domain population to the target domain population.
[0122] S333. Migrate valid sample individuals as historical knowledge information to the target domain population to form a new target domain data population, and build a battery diagnosis model.
[0123] Specifically, the source domain sample individuals that best match the target domain are selected as valid samples. These samples have high similarity and can be used as historical knowledge information for migration. The selected valid sample individuals are migrated from the source domain population to the target domain population as new target domain data. The historical knowledge information can be introduced into the target domain to enrich the target domain data and improve the accuracy and robustness of the battery diagnosis model. Then, the historical knowledge information migrated to the target domain and the data of the target domain itself are used to build a new battery diagnosis model. The model will evaluate and diagnose the health status of the battery based on the data characteristics and historical knowledge information of the target domain, including the use of various machine learning algorithms or statistical models to build this diagnosis model, such as support vector machines, neural networks, decision trees, etc.
[0124] S334. Based on the particle swarm optimization algorithm, an adaptively optimized battery diagnosis model is obtained.
[0125] Specifically, the Particle Swarm Optimization (PSO) algorithm is an evolutionary computing algorithm that simulates the foraging behavior of a flock of birds and seeks the optimal solution through cooperation and competition between individuals. The steps of the adaptively optimized battery diagnosis model are as follows:
[0126] Initialize particle swarm: randomly generate a set of particles, each particle represents a possible battery diagnosis model, and these particles move in a multi-dimensional space, where each dimension represents a model parameter.
[0127] Evaluate fitness: For each particle, calculate its fitness on the target domain data based on the battery diagnosis model it represents. Fitness is determined by evaluating the prediction accuracy of the model or other indicators.
[0128] Update speed and position: According to the principle of particle swarm algorithm, the speed and position of each particle are updated according to its current position and speed, its individual optimal position and the group optimal position.
[0129] Iterative update: Repeat the update of velocity and position until the stopping condition is reached, such as reaching the maximum number of iterations or the particle's position converges to a stable state.
[0130] Select the optimal model: After the iteration, according to the fitness of each particle, the battery diagnosis model represented by the particle with the highest fitness is selected as the optimal model.
[0131] S34. Based on the iterative clustering strategy, the newly added sample data is retained and the online model library is updated.
[0132] In the description of the present invention, based on the iterative clustering strategy, retaining the newly added sample data and updating the online model library include the following steps:
[0133] S341. Set the update frequency and cycle duration of the online model library.
[0134] S342. Use the K-means clustering algorithm to cluster the newly generated target domain population, select the individual with the smallest objective function value in the population, save it to the representative individual set, calculate the Euclidean distance between each two individuals in the representative individual set, and select the minimum distance value as the initial cluster radius.
[0135] S343. According to the update frequency and cycle duration, the representative individual set and initial class radius in each target domain population are regularly sampled as content updates of the online model library.
[0136] S35, inputting the battery information into the battery diagnosis model, analyzing the health status of each battery string in the current battery pack, and matching the analysis results with corresponding regulation decisions.
[0137] Specifically, the built battery diagnosis model with learning and evolution capabilities is used to input preprocessed battery information data to analyze and diagnose the health status. Based on the input data, the model predicts the health status of each battery string, such as healthy, sub-healthy or faulty status.
[0138] According to the analysis results of the battery diagnosis model, the corresponding control decisions are matched.
[0139] For example, if the battery pack is in a healthy state, it means that the battery pack cell string is in good working condition, the performance is normal, and there is no obvious abnormality. Its control decision is to continue normal operation, perform routine inspections and maintenance regularly, and keep the charge and discharge states balanced.
[0140] If the battery pack is in a sub-healthy state, it means that the battery pack cell string has a slight performance degradation or pre-failure sign, but has not seriously affected the normal operation of the entire system. Its control decision is to adjust the charging and discharging strategy, increase the maintenance cycle, and strengthen the monitoring frequency to delay the decline of battery performance and try to avoid failures.
[0141] If the battery pack is in a faulty state, it means that the battery string has a serious fault and the performance has significantly decreased, which may cause abnormal or interrupted system operation. The control decision is to replace the faulty battery string in time, repair the battery pack or related equipment, and ensure the stable operation of the system. At the same time, the root cause analysis is carried out to prevent similar faults from happening again.
[0142] If the battery pack is undercharged or under-discharged, it means that the battery pack string has insufficient performance during the charging or discharging process and cannot meet the system requirements. Its control decision is to adjust the charging and discharging parameters and optimize the charging and discharging strategy to ensure that the battery pack can meet the energy needs of the system in a timely manner.
[0143] S4. Based on the monitoring performance evaluation results and control decisions, optimize and adjust the configuration parameters of the partition monitoring network and battery pack, record the parameter adjustment results, and automatically generate an operation and maintenance management report.
[0144] See also Figure 2 The present invention also provides a substation battery group operation and maintenance management system based on artificial intelligence, the system comprising:
[0145] The monitoring module 1 is used to divide the battery group into several battery strings, and use a single battery string as a monitoring node, install independent monitoring equipment and sub-control switches, and build a partition monitoring network.
[0146] The perception module 2 is used to collect battery information during the operation of the battery pack in real time, store it locally and in the database respectively, and evaluate the monitoring performance of the partition monitoring network based on the self-perception model.
[0147] The decision module 3 is used to analyze the health status of the battery pack based on the battery diagnosis model with learning evolution capability, and output the control decision of the battery pack according to the health status analysis result.
[0148] Configuration module 4 is used to optimize and adjust the configuration parameters of the partition monitoring network and battery pack based on the monitoring performance evaluation results and control decisions, record the parameter adjustment results, and generate an operation and maintenance management report.
[0149] Among them, the monitoring module 1, the perception module 2, the decision module 3 and the configuration module 4 are connected in sequence.
[0150] In summary, with the help of the above technical solution of the present invention, by dividing the battery group into battery strings and constructing a partition monitoring network, the refined monitoring and management of the operating status of the battery group is realized; the battery information is collected in real time and the monitoring network performance is evaluated based on the self-perception model to ensure the accuracy and stability of the monitoring system; the battery diagnosis model based on learning evolution is used to analyze the health status of the battery group and output the control decision, which improves the accuracy and timeliness of the system fault diagnosis, and on this basis, the control decision is optimized to adjust the monitoring network and parameters, realizing the intelligent management and optimized scheduling of the battery group, providing the convenience of decision support and operation and maintenance management, and comprehensively improving the operation efficiency, safety and reliability of the substation battery group. The information entropy gain of each monitoring node is calculated through the self-perception model, and the monitoring performance of each monitoring node in the partition monitoring network is evaluated, which helps the operation and maintenance personnel to timely discover whether the monitoring capability of the monitoring node meets the standard, so as to timely adjust and optimize the monitoring network, improve the monitoring effect and management level of the operating status of the battery group, and ensure the safe and stable operation of the power grid. By obtaining historical battery information and health status from the online model library, migrating historical knowledge information to a new battery diagnostic model, and performing model adaptive optimization, the diagnosis and evaluation of the health status of the battery pack is realized. At the same time, the online model library is updated through an iterative clustering strategy to keep the model updated and optimized, thereby improving the accuracy and reliability of the diagnostic model, providing a scientific basis and decision-making support for the operation and maintenance management of the battery pack, and improving the safety, stability and reliability of the system.
[0151] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.
Claims
1. A method for operation and maintenance management of substation battery packs based on artificial intelligence, It is characterized in that The method comprises the following steps: S1. Divide the battery pack into several battery strings, use a single battery string as a monitoring node, install independent monitoring equipment and sub-control switches, and build a partition monitoring network; S2. Real-time collection of battery information during the operation of the battery pack, storage in local and database locations, and evaluation of the monitoring performance of the partition monitoring network based on a self-perception model; S3. Analyze the health status of the battery pack based on the battery diagnosis model with learning evolution capability, and output the control decision of the battery pack according to the health status analysis result; S4. Based on the monitoring performance evaluation results and control decisions, optimize and adjust the configuration parameters of the partition monitoring network and battery pack, record the parameter adjustment results, and automatically generate an operation and maintenance management report; The real-time collection of battery information during the operation of the battery pack, storing it locally and in a database respectively, and evaluating the monitoring performance of the partition monitoring network based on a self-perception model includes the following steps: S21, using the monitoring device to collect battery information of the corresponding monitoring node in real time, and storing it in a local memory in priority according to the time sequence, and then transmitting it to the supervision center through a communication link, wherein the battery information includes temperature, battery cell voltage, battery cell current, battery string voltage, battery string current, battery cell internal resistance and switch status; S22, based on the pre-set battery information integration standard and classification principle, processing the battery information uploaded by each monitoring node and storing it in the database; S23, setting a self-sensing period of the partition monitoring network, during which, based on the self-sensing model, the information entropy gain of each monitoring node is calculated to evaluate the monitoring performance of each monitoring node in the partition monitoring network; The step of calculating the information entropy gain of each monitoring node based on the self-perception model and evaluating the monitoring performance of each monitoring node in the partitioned monitoring network comprises the following steps: S231, establishing a self-perception model based on a Bayesian network, and using the self-perception model to extract the dependency relationship and conditional probability between the monitoring node and its locally stored battery information; S232, calculating the local condition entropy of each monitoring node using a local condition entropy formula; S233, calculating the online condition entropy of each monitoring node using an online condition entropy formula; S234, taking the difference between the online condition entropy and the local condition entropy as the information entropy gain of the monitoring node, and comparing it with the preset information entropy gain threshold; if the information entropy gain of the monitoring node is less than the information entropy gain threshold, the monitoring performance of the monitoring node is determined to be qualified; if the information entropy gain of the monitoring node is greater than or equal to the information entropy gain threshold, the monitoring performance of the monitoring node is determined to be unqualified and there is a monitoring performance defect; The method of analyzing the health status of the battery pack based on the battery diagnosis model with learning evolution capability and outputting the regulation decision of the battery pack according to the health status analysis result includes the following steps: S31, obtaining an online model library for battery pack health status assessment, using historical battery information and corresponding historical health status in the online model library as a source domain, and merging the battery information of all monitoring nodes currently monitored by the battery pack into a battery data set as a target domain; S32, setting a maximum mean difference index, evaluating the similarity between the sample data in the source domain and the target domain, and searching for an optimal historical model; S33, extracting valid samples of the optimal historical model, constructing a mapping relationship from the source domain to the target domain, migrating historical knowledge information to the new battery diagnosis model, and performing model adaptive optimization; S34, based on the iterative clustering strategy, retain the newly added sample data and update the online model library; S35, inputting the battery information into the battery diagnosis model, analyzing the health status of each battery string in the current battery pack, and matching the control decision corresponding to the analysis result; The local condition entropy formula is: ; In the formula, A represents the local conditional entropy; H (·) represents information entropy; P (·) represents conditional probability; R i Indicates i Monitoring nodes; parent ( R i ) indicates the i Battery information stored locally in each monitoring node; The online conditional entropy formula is: ; In the formula, B represents the online conditional entropy; D Represents the collection of battery information uploaded by monitoring nodes stored in the database; H (·) represents information entropy; P (·) represents conditional probability; R i Indicates i Monitoring nodes.
2. According to claim 1, a method for operation and maintenance management of a substation battery pack based on artificial intelligence, It is characterized in that The processing of the battery information uploaded by each monitoring node based on the pre-set integration standard and classification principle of the battery information and storing it in the database includes the following steps: S221, based on the integration standard and classification principle, the battery information is divided into corresponding attribute items, and the respective attribute labels are marked, and format errors and duplicate data are removed; S222, formulating a unified coding standard, and using a return-to-zero code method to encode the battery information so that the battery information meets the standardized input standard; S223. Associate the battery information with the corresponding collection time and upload time, and store them in a database in the supervision center in the form of time series.
3. According to claim 2, a method for operation and maintenance management of a substation battery pack based on artificial intelligence, It is characterized in that The setting of the maximum mean difference index, evaluating the similarity between the sample data in the source domain and the target domain, and searching for the optimal historical model include the following steps: S321, sorting the sample data in the source domain according to the update time, and clustering the sample data in the source domain and the target domain using the K-means clustering algorithm in turn; S322, using the maximum mean difference formula to calculate the maximum mean difference index between each source domain sample class and the target domain sample class; S323, calculating the average value of the maximum mean difference index of all sample classes in the target domain, and selecting the reciprocal of the average value as the similarity degree between the source domain sample data and the target domain sample data; S324, setting a similarity threshold, selecting source domain sample data with a similarity degree greater than the similarity threshold as the evaluation result, and matching the optimal historical model corresponding to the number of source domain samples.
4. According to claim 3, a method for operation and maintenance management of a substation battery pack based on artificial intelligence, It is characterized in that The extracting of valid samples of the optimal historical model, constructing a mapping relationship from a source domain to a target domain, migrating historical knowledge information to a new battery diagnosis model, and performing model adaptive optimization includes the following steps: S331, combining all sample data of the target domain to form a target domain population, combining the matched source domain sample data to form a source domain population, and establishing a mapping relationship between the two populations; S332, setting an upper limit and a lower limit of the proportion of individuals migrating from the source domain population to the target domain population, calculating the migration ratio of the source domain population to the target domain population, and selecting a specific number of valid sample individuals from the source domain population according to the migration ratio. The migration ratio formula is: ; In the formula, G represents the migration ratio; S Indicates the degree of similarity between the source domain sample data and the target domain sample data; p max It represents the upper limit of the proportion of individuals that migrate from the source domain population to the target domain population; p min It represents the lower limit of the proportion of individuals migrating from the source domain population to the target domain population; S333, migrating valid sample individuals as historical knowledge information to the target domain population to form a new target domain data population, and constructing a battery diagnosis model; S334. Based on the particle swarm optimization algorithm, an adaptively optimized battery diagnosis model is obtained.
5. According to claim 4, a method for operation and maintenance management of substation battery packs based on artificial intelligence, It is characterized in that The method of retaining newly added sample data and updating the online model library based on the iterative clustering strategy includes the following steps: S341, setting the update frequency and cycle duration of the online model library; S342, clustering the newly generated target domain population using the K-means clustering algorithm, and selecting the individual with the smallest objective function value in the population, saving it to a representative individual set, and then calculating the Euclidean distance between two individuals in the representative individual set, and selecting the minimum distance value as the initial cluster radius; S343. According to the update frequency and cycle duration, regularly sample the representative individual set and initial class radius in each target domain population as content update of the online model library.
6. An artificial intelligence-based substation battery group operation and maintenance management system, used to implement the artificial intelligence-based substation battery group operation and maintenance management method according to any one of claims 1 to 5, It is characterized in that The system includes: The monitoring module is used to divide the battery pack into several battery strings, and use a single battery string as a monitoring node, install independent monitoring equipment and sub-control switches, and build a partition monitoring network; A perception module, used to collect battery information during the operation of the battery pack in real time, store it locally and in a database, and evaluate the monitoring performance of the partition monitoring network based on a self-perception model; A decision module is used to analyze the health status of the battery pack based on a battery diagnosis model with learning and evolution capabilities, and output a control decision for the battery pack based on the health status analysis results; Configuration module, which is used to optimize and adjust the configuration parameters of the partition monitoring network and battery pack based on the monitoring performance evaluation results and control decisions, record the parameter adjustment results, and generate operation and maintenance management reports; Among them, the monitoring module, the perception module, the decision module and the configuration module are connected in sequence.
