Adaptive voltage control method for distributed power supply based on deep learning

Abstract

The present application provides a distributed power adaptive voltage control method based on deep learning, which relates to the field of power control technology. The method includes: adaptively partitioning the distributed power supply, establishing n cluster control chains, establishing historical compensation data of the distributed power supply, establishing a basic solution generation model, collecting the status data of the cluster area and sending it to the solution generation model, and monitoring the key nodes in the cluster area, generating node update status, and then generating rewards and performing action evaluation, executing iterative updates of the basic solution generation model, and making subsequent control decisions. This application can solve the problem that the existing technology cannot effectively manage multiple cluster areas at the same time, and cannot collect and monitor data in real time, which makes it difficult for the system to discover and solve potential problems in a timely manner, thereby affecting the stability of the power supply system, and realizing efficient, accurate and adaptive control of distributed power supplies, and improving the stability, reliability and continuous optimization capability of the system.

Description

Technical Field

[0001] The present application relates to the field of power supply control technology, and in particular to a distributed power supply adaptive voltage control method based on deep learning. Background Art

[0002] Distributed power sources, such as solar energy, wind energy and other renewable energy power generation equipment, have the characteristics of intermittency, randomness and unpredictability. These characteristics make the voltage output of distributed power sources unstable, and they need to be effectively controlled to ensure the stable operation of the power system. Voltage is an important parameter in the power system and has an important impact on the stable operation of the power system and the quality of power. If the voltage is too high or too low, it may cause damage to the power equipment and affect the normal operation of the power system. Therefore, it is very important to control the voltage of distributed power sources.

[0003] At present, traditional voltage control methods are mainly based on rules and experience, and perform static or simple dynamic control of voltage. However, due to the characteristics of distributed power sources and the complexity of power systems, traditional voltage control methods often have difficulty coping with complex and changeable voltage fluctuations and cannot achieve accurate voltage control.

[0004] In summary, existing technologies cannot effectively manage multiple cluster areas at the same time, and cannot collect and monitor data in real time, which makes it difficult for the system to detect and solve potential problems in a timely manner, thereby affecting the stability of the power supply system. Summary of the invention

[0005] The purpose of this application is to provide a distributed power supply adaptive voltage control method based on deep learning, so as to solve the problem that the existing technology cannot effectively manage multiple cluster areas at the same time and cannot collect and monitor data in real time, which makes it difficult for the system to detect and solve potential problems in time, thereby affecting the stability of the power supply system.

[0006] In view of the above problems, the present application provides a distributed power supply adaptive voltage control method based on deep learning.

[0007] In the first aspect, the present application provides a distributed power adaptive voltage control method based on deep learning, and the method is implemented by a distributed power adaptive voltage control system based on deep learning, wherein the method includes: adaptively partitioning the distributed power supply and establishing n cluster control chains, wherein each cluster area corresponds to an edge device and a sub-agent, and the n sub-agents are all connected to a shared update database, wherein the shared update database is a processing database that can be adaptively updated and iterated and is set in the cloud; establishing historical compensation data of the distributed power supply, wherein the historical compensation data is compensation data that has been verified and updated, and using the historical compensation data as training data to establish a basic scheme Generate a model and store the solution generation model in a shared update database; when any edge device collects status data of the corresponding cluster area, the status data is sent to the solution generation model to generate control parameters, control the edge device with the control parameters, monitor key nodes in the cluster area, and generate node update status; generate rewards according to the node update status and status data, evaluate the control parameters with the rewards, and share the action evaluation results with the shared update database; after the shared update database receives the action evaluation results, it executes iterative updates of the basic solution generation model, and makes subsequent control decisions of the distributed power source according to the iterative update results.

[0008] In the second aspect, the present application also provides a distributed power adaptive voltage control system based on deep learning, which is used to execute the distributed power adaptive voltage control method based on deep learning as described in the first aspect, wherein the system includes: a group control chain establishment module, which is used to adaptively partition the distributed power supply and establish n cluster control chains, wherein each cluster area corresponds to an edge device and a sub-agent, and the n sub-agents are all connected to a shared update database, which is a processing database that can be adaptively updated and iterated and is set in the cloud; a historical compensation data establishment module, which is used to establish historical compensation data of the distributed power supply, wherein the historical compensation data is verified and updated compensation data, and the historical compensation data is used as training data to establish a basic solution generation module. type, storing the solution generation model in the shared update database; a control parameter generation module, the control parameter generation module is used to send the status data of the corresponding cluster area to the solution generation model after any edge device collects the status data, generate control parameters, control the edge device with the control parameters, monitor the key nodes of the cluster area, and generate node update status; a database update module, the database update module is used to generate rewards according to the node update status and status data, evaluate the control parameters with the rewards, and share the action evaluation results with the shared update database; an iterative update module, the iterative update module is used to perform iterative updates of the basic solution generation model after the shared update database receives the action evaluation results, and make subsequent control decisions of distributed power sources according to the iterative update results.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] 1. By adaptively partitioning the distributed power supply, n cluster control chains are established, wherein each cluster area corresponds to an edge device and a sub-agent, and the n sub-agents are all connected to a shared update database, which is a processing database that can be adaptively updated and iterated and is set in the cloud; historical compensation data of the distributed power supply is established, wherein the historical compensation data is the compensation data that has been verified and updated, and the historical compensation data is used as training data to establish a basic solution generation model, and the solution generation model is stored in the shared update database; when any edge device collects the status data of the corresponding cluster area, the status data is sent to the solution generation model to generate control parameters, the edge device is controlled by the control parameters, and the key nodes of the cluster area are monitored to generate node update status; rewards are generated according to the node update status and status data, the control parameters are evaluated by the rewards, and the action evaluation results are shared to the shared update database; when the shared update database receives the action evaluation results, iterative updates of the basic solution generation model are performed, and subsequent control decisions of the distributed power supply are made according to the iterative update results. It realizes efficient, precise and adaptive control of distributed power sources, and improves the stability, reliability and continuous optimization capability of the system.

[0011] 2. Through adaptive partitioning, real-time data-driven control, key node monitoring, reward-based action evaluation, iterative update of the model and support of cloud database, efficient, precise and adaptive control of distributed power sources is achieved, improving the stability, reliability and continuous optimization capability of the system.

[0012] 3. The deep learning-based distributed power adaptive voltage control method is used instead of the existing method, which effectively solves the problem that the existing technology cannot effectively manage multiple cluster areas at the same time and cannot collect and monitor data in real time, which makes it difficult for the system to detect and solve potential problems in time, thus affecting the stability of the power supply system. It realizes efficient, precise and adaptive control of distributed power sources, and improves the stability, reliability and continuous optimization capability of the system.

[0013] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented according to the contents of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically cited below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0015] Figure 1 This is a flow chart of a distributed power adaptive voltage control method based on deep learning in this application;

[0016] Figure 2 This is a structural diagram of the distributed power supply adaptive voltage control system based on deep learning in this application.

[0017] Description of reference numerals:

[0018] A group control chain establishing module 11, a historical compensation data establishing module 12, a control parameter generating module 13, a database updating module 14, and an iterative updating module 15. DETAILED DESCRIPTION

[0019] This application provides a distributed power supply adaptive voltage control method based on deep learning, which solves the problem that the existing technology cannot effectively manage multiple cluster areas at the same time and cannot collect and monitor data in real time, which makes it difficult for the system to promptly discover and solve potential problems, thereby affecting the stability of the power supply system. It realizes efficient, precise and adaptive control of distributed power supplies, and improves the stability, reliability and continuous optimization capability of the system.

[0020] Below, the technical solutions in the present application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application. It should also be noted that, for the convenience of description, only the parts related to the present application are shown in the accompanying drawings, rather than all of them.

[0021] Embodiment 1

[0022] Please see attached Figure 1 The present application provides a distributed power adaptive voltage control method based on deep learning, wherein the method is applied to a distributed power adaptive voltage control system based on deep learning, and the method specifically comprises the following steps:

[0023] Adaptively partition the distributed power supply and establish n cluster control chains, where each cluster area corresponds to an edge device and a sub-agent, and the n sub-agents are all connected to a shared update database, which is a processing database set in the cloud that can be adaptively updated and iterated;

[0024] Specifically, each cluster area corresponds to an edge device, which is responsible for collecting real-time data of distributed power sources in the area, such as voltage, current, power, etc., and sending these data to the corresponding sub-agents. The sub-agents perform voltage control or other related controls on the distributed power sources in the area based on the received data and the preset control strategy or algorithm. At the same time, these n sub-agents are connected to a shared update database, which is set up in the cloud and can be updated and iterated adaptively. The sub-agents can obtain the latest control strategies, model parameters or other relevant information from the cloud database at any time to achieve more accurate and intelligent control. At the same time, each sub-agent can also feed back its own operating data, control experience, etc. to the cloud database to provide data support for the update and iteration of the database.

[0025] Establishing historical compensation data of distributed power sources, wherein the historical compensation data is verified and updated compensation data, using the historical compensation data as training data, establishing a basic solution generation model, and storing the solution generation model in a shared update database;

[0026] Specifically, first, collect relevant historical data from the distributed power system. These data can come from the system's monitoring equipment, control equipment or other related equipment. The collected data should include key parameters such as voltage, current, power, and other information related to environmental factors, equipment status, etc. The collected raw data may contain errors or outliers, and they are verified and cleaned. The purpose of data verification is to ensure the accuracy and reliability of the data for subsequent training and model building. Over time, the operating status and environmental factors of the distributed power system may change. Therefore, historical compensation data also needs to be continuously updated and adjusted to reflect these changes. Data updates can be achieved by regularly collecting new data, updating old data, or introducing new data sources. It can be used as training data to establish a basic solution generation model. This model can learn how to generate effective control solutions based on the voltage fluctuations in historical data and the adjustment of control strategies. Through technologies such as deep learning, the model can automatically extract features and laws in the data and generate control solutions that adapt to different situations. Once the basic solution generation model is established, it can be stored in a shared update database. In this way, each sub-agent can obtain the model from the database when needed and generate corresponding control solutions based on real-time data. At the same time, with the continuous addition of new historical data and the updating and iteration of the model, the basic solution generation model can also be continuously optimized and improved to adapt to the changes and development of distributed power systems.

[0027] When any edge device collects status data of the corresponding cluster area, the status data is sent to the solution generation model to generate control parameters, the edge device is controlled by the control parameters, and key nodes in the cluster area are monitored to generate node update status;

[0028] Specifically, the edge device will continuously monitor the status data of the cluster area where it is located, including but not limited to key parameters such as voltage, current, power, temperature, and equipment operating status. The collected status data will be transmitted to the solution generation model in real time or regularly through the communication network. After receiving the status data, the solution generation model analyzes the data and generates corresponding control parameters. These control parameters are designed to optimize the operating status of the distributed power supply to ensure that the goals of voltage stability, equipment safety, and high energy efficiency are achieved. The generated control parameters will be sent back to the corresponding edge device, and the edge device will adjust and control the distributed power supply in real time according to the received control parameters. This may include adjusting voltage, current output, changing working mode, starting or shutting down equipment, etc. While executing control, the key nodes in the cluster area are monitored in real time. Key nodes usually refer to devices or nodes that play an important role in the distributed power system, such as converters, inverters, etc. Monitoring these nodes helps to detect potential problems or faults in a timely manner. Once the status of the key node is detected to change, the node update status is generated and fed back to the solution generation model. In this way, the model can be adjusted and optimized according to the latest node status data to improve the accuracy and adaptability of the control strategy.

[0029] Generate rewards according to the node update status and status data, use the rewards to evaluate the control parameters, and share the action evaluation results to the shared update database;

[0030] Specifically, the calculation of rewards can be based on multiple factors, such as voltage stability, energy efficiency, and equipment operating status. For example, if the control parameters successfully maintain the voltage within the target range and the energy efficiency is also improved, the reward value will increase accordingly. Use the reward value to evaluate the control parameters. It can be regarded as a quantitative evaluation of the effect of the control parameters. If the reward value is high, it means that the control parameters have achieved good results in practical applications, otherwise it means that the control strategy needs to be adjusted or optimized. The results of the action evaluation can provide feedback for subsequent model training. The performance of the model can be optimized by adjusting the weights or parameters of the model to generate more accurate and effective control parameters. The results of the action evaluation are shared to the shared update database. In this way, each sub-agent can access the latest evaluation results to understand the actual effect of the control parameters. Sharing evaluation results helps promote collaborative learning and optimization between different cluster areas. For example, the control parameters of one cluster area may perform well in specific situations, and other areas can learn from these experiences to improve their own control strategies.

[0031] After the shared update database receives the action evaluation result, it executes iterative update of the basic solution generation model, and makes subsequent control decisions of the distributed power source according to the iterative update result.

[0032] Specifically, the shared update database first receives the action evaluation results from the edge devices and sub-agents. These results contain the performance evaluation of the control parameters in actual applications and are an important basis for the iterative update of the model. The basic solution generation model is iteratively updated using the received action evaluation results. This includes weight adjustment or structural optimization of the deep learning model. By comparing the model prediction results with the actual action effects, the system can identify deficiencies in the model and make corresponding improvements. The iterative update process may include retraining the model, adjusting model parameters, introducing new features or algorithms, etc. The goal is to improve the prediction accuracy and adaptability of the model so that it can generate control parameters more accurately. After the model is iteratively updated, its performance needs to be verified. This can be done by evaluating the prediction accuracy, stability and other indicators of the model on the test data set. The verification process helps ensure that the updated model can achieve better results in actual applications. Once the validity of the iterative update results is verified, the updated basic solution generation model is used to generate subsequent control parameters. These parameters will be used to guide real-time control decisions of distributed power sources, including voltage adjustment, energy distribution, fault prediction, etc.

[0033] Furthermore, the present application also includes:

[0034] Establish the voltage control reward function as follows:

[0035]

[0036] Among them, r n is the reward of sub-agent n, Ω n，O is the set of observation points in the sub-agent n region, V i is the voltage amplitude of node i, and V0 is the rated voltage of the network.

[0037] Specifically, when establishing the voltage control reward function, the sub-agent is encouraged to maintain the voltage amplitude in the cluster area close to the rated voltage of the network, while avoiding the voltage amplitude being too high or too low to prevent equipment damage or system instability. The reward function will encourage the sub-agent to adjust the control parameters to minimize the deviation between the voltage amplitude and the rated voltage at all observation points in the cluster area, thereby achieving stable voltage control.

[0038] Furthermore, the present application also includes:

[0039] The historical compensation data is subjected to data smoothing processing, and the specific processing is as follows:

[0040] Step 1: extracting device features of corresponding devices from the historical compensation data, and dividing the historical compensation data into independent data and common data based on the device feature extraction results;

[0041] Step 2: Using the initial state data in the general data as clustering features, perform free clustering of data in the general data to generate clustering results;

[0042] Step 3: Sort the solutions under the same clustering result in sequence using the final state data in the common data, and optimize the parameters of the initial control parameters based on the sequence sorting results;

[0043] Step 4: Complete data smoothing based on clustering results, parameter optimization results and independent data;

[0044] The smoothed data is used as training data to build a basic solution generation model.

[0045] Specifically, first, extract the features related to each device from the historical compensation data. These features may include the model, capacity, operating status, historical fault records, etc. of the device. Based on the results of the device feature extraction, the historical compensation data is divided into two parts: independent data and general data. Independent data refers to data that is closely related to only a specific device or scenario, while general data is data applicable to multiple devices or scenarios. Extract the initial state data, such as the initial state data of a certain time period, as clustering features from the general data. Use clustering algorithms such as K-means, hierarchical clustering, etc. to freely cluster the general data. The goal of clustering is to find naturally formed groups or patterns in the data. For each clustering result, use the final state data, such as the end state data of a certain time period, to sequentially sort the solutions within the cluster. Based on the results of the sequential sorting, use optimization algorithms such as gradient descent, genetic algorithms, etc. to optimize the initial control parameters. The goal of optimization is to find parameter settings that can minimize errors or maximize performance. Combining the clustering results, parameter optimization results, and independent data, the historical compensation data is smoothed. Interpolation, filtering, or time series analysis can be applied. The goal of smoothing is to eliminate noise or outliers in the data and make the data smoother and more continuous. The smoothed data is used as training data to build a basic solution generation model. Use the training data to build a basic solution generation model.

[0046] Furthermore, the present application also includes:

[0047] According to the clustering result, a preset search step length is configured with the initial state data as the initial control parameter of the matching data, wherein the preset search step length includes a step length direction and a step length;

[0048] After the state data is sent to the solution generation model, the solution generation model matches and obtains basic control parameters, and modifies the basic control parameters by corresponding preset search steps, and outputs control parameters according to the modified results.

[0049] Specifically, according to the clustering results, the initial state data of each cluster center is selected as representative data. These initial state data reflect the initial state of the system under different operating conditions. A preset search step size for the initial control parameters is configured for each cluster. The search step size includes the step size direction and the step size length. The direction indicates the positive and negative trend of the parameter adjustment, and the length indicates the amplitude of the parameter adjustment. Based on historical data and expert knowledge, the possible direction of each parameter in the optimization process, such as increase or decrease, is determined. It can be determined based on the importance of the parameter, historical optimization experience, or experiments. A longer step size may speed up the search process, but may also cause the optimal solution to be missed; while a shorter step size may be more refined, but the search speed is slower. Real-time or recent state data is sent to the solution generation model. The model matches or generates a set of basic control parameters from the preset control parameter set based on the input state data. The basic control parameters are corrected by the preset search step size. The basic control parameters are corrected using the preset search step size previously configured for each cluster. The correction process may include increasing, decreasing, or keeping the parameters unchanged, depending on the step size direction and length. The corrected control parameters are applied to the system and their effects are observed. Based on the feedback effect, the search step size and direction can be further adjusted for iterative optimization. The effect of the control parameters can be evaluated by comparing the system performance before and after correction, such as voltage stability and energy efficiency. The best control parameters are selected as the final output. These parameters will be used to guide the real-time control of the distributed power system.

[0050] Furthermore, the present application also includes:

[0051] Feeding back the action evaluation result to the feedback optimization layer of the solution generation model;

[0052] After the feedback optimization layer receives the action evaluation result, the step length is updated according to the preset search step length according to the action evaluation result. The updating steps are as follows:

[0053] A: Determine whether the action evaluation result is an optimization result relative to the final state parameter corresponding to the initial control parameter;

[0054] B1: If it is an optimization result, a unit step optimization analysis is performed according to the optimization degree and the length of the preset search step, and an adjustment step is generated according to the analysis result. The adjustment step is a bidirectional test adjustment step, and the search end point of the preset search step is used as the new starting point, and the step is updated with the bidirectional test adjustment step;

[0055] B2: If it is a non-optimized result, return to the basic control parameters, use the median step of the preset search step as the update step, and perform a bidirectional step update starting from the basic control parameters;

[0056] The iterative update of the basic solution generation model is completed according to the step size update result.

[0057] Specifically, the actual results after all agents perform control actions, i.e., the action evaluation results, are collected. These action evaluation results are transmitted to the feedback optimization layer of the solution generation model. The feedback optimization layer receives the action evaluation results. It is determined whether the action evaluation results are optimized results relative to the final state parameters of the corresponding initial control parameters. If the action evaluation results are optimized results, the degree of optimization is analyzed. Unit step optimization analysis is performed based on the degree of optimization and the length of the preset search step. An adjustment step is generated based on the analysis results, which is a two-way test adjustment step. The search end point of the preset search step is used as the new starting point, and the step size is updated using the two-way test adjustment step. If the action evaluation result is a non-optimized result, it returns to the basic control parameters. The median step size of the preset search step size is used as the update step size. Starting from the basic control parameters, the median step size is used for two-way step size updates. According to the step size update results, the basic solution generation model is iteratively updated. This includes adjusting the internal parameters, logic, or search strategy of the model. With more data feedback and iterative updates, the model will gradually optimize its ability to generate control parameters and improve the stability and efficiency of the system.

[0058] Furthermore, the present application also includes:

[0059] Determine whether the fallback value meets the preset deviation ratio;

[0060] If the preset deviation ratio is met, a penalty factor is generated;

[0061] The update step size is penalized for searching in the same direction as the preset search step size according to the penalty factor.

[0062] Specifically, the current system performance is evaluated and compared with the expected performance. The difference between the actual performance and the expected performance, i.e., the backoff value, is calculated. An acceptable performance deviation ratio is preset, which is used to determine whether the backoff value is within the tolerable range. The calculated backoff value is compared with the preset deviation ratio. If the backoff value meets or exceeds the preset deviation ratio, it indicates that the current control strategy may have problems. In this case, a penalty factor is generated. The penalty factor is usually a value greater than 1 and is used to increase the size of the subsequent search step so that the system adjusts the control parameters more cautiously in the next iteration. According to the generated penalty factor, the update step size is penalized in the same direction. In the next iteration, the search step size will be multiplied by the penalty factor, thereby reducing the search range or increasing the caution of the search. Although the step size is penalized or increased, the search direction remains unchanged, so that the system can still continue to find a better solution on the current search path. After applying the penalty factor, the search step information in the model is updated. According to the adjusted step size, the control parameters are recalculated and applied to the system.

[0063] Furthermore, the present application also includes:

[0064] Record the control results of n cluster control chains and generate steady-state identification of the cluster control chains;

[0065] generating an active control compensation according to the steady-state indicator;

[0066] The active control compensation is used to perform pre-compensation control of n cluster control chains.

[0067] Specifically, the control result data of n cluster control chains are collected. This includes the execution of control actions, changes in system status, and any relevant performance indicators. The control result data of each cluster control chain is analyzed to detect whether it has reached a steady state. Steady state usually refers to a state in which the system state remains unchanged or changes very little over a certain period of time. For cluster control chains that have reached a steady state, a steady state identifier is generated. This identifier can be a Boolean value, for example, true indicates a steady state and false indicates a non-steady state. Based on the steady state identifier, an active control compensation strategy is formulated. This strategy can include different compensation methods, such as parameter adjustment, control logic optimization, etc. According to the compensation strategy, the required active control compensation amount is calculated, including a comprehensive analysis of the system state, historical control data, and steady state identifiers. The calculated active control compensation is applied to the n cluster control chains. This can be achieved by modifying control parameters, adjusting control logic, or introducing new control actions. After the pre-compensation control is implemented, the system state and control results are continuously monitored to ensure the effectiveness of the compensation measures. If it is found that the compensation is insufficient or excessive, the compensation strategy and compensation amount can be adjusted in time.

[0068] In summary, the distributed power supply adaptive voltage control method based on deep learning provided in this application has the following technical effects:

[0069] 1. By adaptively partitioning the distributed power supply, n cluster control chains are established, wherein each cluster area corresponds to an edge device and a sub-agent, and the n sub-agents are all connected to a shared update database, which is a processing database that can be adaptively updated and iterated and is set in the cloud; historical compensation data of the distributed power supply is established, wherein the historical compensation data is the compensation data that has been verified and updated, and the historical compensation data is used as training data to establish a basic solution generation model, and the solution generation model is stored in the shared update database; when any edge device collects the status data of the corresponding cluster area, the status data is sent to the solution generation model to generate control parameters, the edge device is controlled by the control parameters, and the key nodes of the cluster area are monitored to generate node update status; rewards are generated according to the node update status and status data, the control parameters are evaluated by the rewards, and the action evaluation results are shared to the shared update database; when the shared update database receives the action evaluation results, iterative updates of the basic solution generation model are performed, and subsequent control decisions of the distributed power supply are made according to the iterative update results. It realizes efficient, precise and adaptive control of distributed power sources, and improves the stability, reliability and continuous optimization capability of the system.

[0070] 2. Through adaptive partitioning, real-time data-driven control, key node monitoring, reward-based action evaluation, iterative update of the model and support of cloud database, efficient, precise and adaptive control of distributed power sources is achieved, improving the stability, reliability and continuous optimization capability of the system.

[0071] 3. The deep learning-based distributed power adaptive voltage control method is used instead of the existing method, which effectively solves the problem that the existing technology cannot effectively manage multiple cluster areas at the same time and cannot collect and monitor data in real time, which makes it difficult for the system to detect and solve potential problems in time, thus affecting the stability of the power supply system. It realizes efficient, precise and adaptive control of distributed power sources, and improves the stability, reliability and continuous optimization capability of the system.

[0072] Embodiment 2

[0073] Based on the distributed power adaptive voltage control method based on deep learning in the aforementioned embodiment, the present application also provides a distributed power adaptive voltage control system based on deep learning, please refer to the attached Figure 2 , the system comprising:

[0074] A group control chain establishment module 11, which is used to adaptively partition the distributed power supply and establish n cluster control chains, wherein each cluster area corresponds to an edge device and a sub-agent, and the n sub-agents are all connected to a shared update database, which is a processing database set in the cloud that can be adaptively updated and iterated;

[0075] A historical compensation data establishment module 12, wherein the historical compensation data establishment module 12 is used to establish historical compensation data of distributed power sources, wherein the historical compensation data is verified and updated compensation data, and the historical compensation data is used as training data to establish a basic solution generation model, and the solution generation model is stored in a shared update database;

[0076] A control parameter generation module 13, wherein the control parameter generation module 13 is used to send the status data of the corresponding cluster area to the solution generation model after any edge device collects the status data, generate control parameters, control the edge device with the control parameters, monitor key nodes in the cluster area, and generate node update status;

[0077] A database update module 14, the database update module 14 is used to generate rewards according to the node update status and status data, perform action evaluation on the control parameters with the rewards, and share the action evaluation results to the shared update database;

[0078] The iterative update module 15 is used to execute iterative update of the basic solution generation model after the shared update database receives the action evaluation result, and make subsequent control decisions of the distributed power source according to the iterative update result.

[0079] Furthermore, the system further comprises a function establishment module, wherein the function establishment module is used to:

[0080] Establish the voltage control reward function as follows:

[0081]

[0082] Among them, r n is the reward of sub-agent n, Ω n , O is the set of observation points in the sub-agent n region, V i is the voltage amplitude of node i, and V0 is the rated voltage of the network.

[0083] Furthermore, the system further comprises a model building generation module, wherein the model building generation module is used to:

[0084] The historical compensation data is subjected to data smoothing processing, and the specific processing is as follows:

[0085] Step 1: extracting device features of corresponding devices from the historical compensation data, and dividing the historical compensation data into independent data and common data based on the device feature extraction results;

[0086] Step 2: Using the initial state data in the general data as clustering features, perform free clustering of data in the general data to generate clustering results;

[0087] Step 3: Sort the solutions under the same clustering result in sequence using the final state data in the common data, and optimize the parameters of the initial control parameters based on the sequence sorting results;

[0088] Step 4: Complete data smoothing based on clustering results, parameter optimization results and independent data;

[0089] The smoothed data is used as training data to build a basic solution generation model.

[0090] Furthermore, the system further comprises a control parameter output module, wherein the control parameter output module is used for:

[0091] According to the clustering result, a preset search step length is configured with the initial state data as the initial control parameter of the matching data, wherein the preset search step length includes a step length direction and a step length;

[0092] After the state data is sent to the solution generation model, the solution generation model matches and obtains basic control parameters, and modifies the basic control parameters by corresponding preset search steps, and outputs control parameters according to the modified results.

[0093] Furthermore, the system further comprises an iterative update module, wherein the iterative update module is used for:

[0094] Feeding back the action evaluation result to the feedback optimization layer of the solution generation model;

[0095] After the feedback optimization layer receives the action evaluation result, the step length is updated according to the preset search step length according to the action evaluation result. The updating steps are as follows:

[0096] A: Determine whether the action evaluation result is an optimization result relative to the final state parameter corresponding to the initial control parameter;

[0097] B1: If it is an optimization result, a unit step optimization analysis is performed according to the optimization degree and the length of the preset search step, and an adjustment step is generated according to the analysis result. The adjustment step is a bidirectional test adjustment step, and the search end point of the preset search step is used as the new starting point, and the step is updated with the bidirectional test adjustment step;

[0098] B2: If it is a non-optimized result, return to the basic control parameters, use the median step of the preset search step as the update step, and perform a bidirectional step update starting from the basic control parameters;

[0099] The iterative update of the basic solution generation model is completed according to the step size update result.

[0100] Furthermore, the system further comprises a penalty factor generation module, wherein the penalty factor generation module is used to:

[0101] Determine whether the fallback value meets the preset deviation ratio;

[0102] If the preset deviation ratio is met, a penalty factor is generated;

[0103] The update step size is penalized for searching in the same direction as the preset search step size according to the penalty factor.

[0104] Furthermore, the system further comprises a compensation control module, wherein the compensation control module is used for:

[0105] Record the control results of n cluster control chains and generate steady-state identification of the cluster control chains;

[0106] generating an active control compensation according to the steady-state indicator;

[0107] The active control compensation is used to perform pre-compensation control of n cluster control chains.

[0108] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. Figure 1 The distributed power adaptive voltage control method based on deep learning and the specific examples in the first embodiment are also applicable to the distributed power adaptive voltage control system based on deep learning in this embodiment. Through the above detailed description of the distributed power adaptive voltage control method based on deep learning, those skilled in the art can clearly know the distributed power adaptive voltage control system based on deep learning in this embodiment, so for the sake of brevity of the specification, it will not be described in detail here. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0109] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0110] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalent technology, the present application is also intended to include these modifications and variations.

Claims

1. A distributed power adaptive voltage control method based on deep learning, characterized in that: The method comprises: Adaptively partition the distributed power supply and establish n cluster control chains, where each cluster area corresponds to an edge device and a sub-agent, and the n sub-agents are all connected to a shared update database, which is a processing database set in the cloud that can be adaptively updated and iterated; Establishing historical compensation data of distributed power sources, wherein the historical compensation data is verified and updated compensation data, using the historical compensation data as training data, establishing a basic solution generation model, and storing the solution generation model in a shared update database; When any edge device collects status data of the corresponding cluster area, the status data is sent to the solution generation model to generate control parameters, the edge device is controlled by the control parameters, and key nodes in the cluster area are monitored to generate node update status; Generate rewards according to the node update status and status data, use the rewards to evaluate the control parameters, and share the action evaluation results to the shared update database; After the shared update database receives the action evaluation result, it executes an iterative update of the basic solution generation model, and makes subsequent control decisions of the distributed power source according to the iterative update result; The method of using the historical compensation data as training data to establish a basic solution generation model includes: The historical compensation data is subjected to data smoothing processing, and the specific processing is as follows: Step 1: extracting device features of corresponding devices from the historical compensation data, and dividing the historical compensation data into independent data and common data based on the device feature extraction results; Step 2: using the initial state data in the general data as clustering features, performing free clustering of the data in the general data to generate a clustering result; Step 3: sorting the solutions under the same clustering result according to the final state data in the general data, and optimizing the parameters of the initial control parameters according to the sorting results; Step 4: completing data smoothing processing according to the clustering results, parameter optimization results and the independent data; The smoothed data is used as training data to build a basic solution generation model; According to the clustering result, a preset search step length is configured with the initial state data as the initial control parameter of the matching data, wherein the preset search step length includes a step length direction and a step length; After the state data is sent to the solution generation model, the solution generation model matches and obtains basic control parameters, and modifies the basic control parameters by corresponding preset search steps, and outputs control parameters according to the modified results; Wherein, after the shared update database receives the action evaluation result, iterative update of the basic solution generation model is performed, including: Feeding back the action evaluation result to the feedback optimization layer of the solution generation model; After the feedback optimization layer receives the action evaluation result, the step length is updated according to the preset search step length according to the action evaluation result. The updating steps are as follows: A: Determine whether the action evaluation result is an optimization result relative to the final state parameter corresponding to the initial control parameter; B1: If it is an optimization result, a unit step optimization analysis is performed according to the optimization degree and the length of the preset search step, and an adjustment step is generated according to the analysis result. The adjustment step is a bidirectional test adjustment step, and the search end point of the preset search step is used as the new starting point, and the step is updated with the bidirectional test adjustment step; B2: If it is a non-optimized result, return to the basic control parameters, use the median step of the preset search step as the update step, and perform a bidirectional step update starting from the basic control parameters; Complete the iterative update of the basic solution generation model according to the step size update result; Wherein, if the result is not optimized, it also includes: Determine whether the fallback value meets the preset deviation ratio; If the preset deviation ratio is met, a penalty factor is generated; The update step size is penalized for searching in the same direction as the preset search step size according to the penalty factor.

2. The method according to claim 1, characterized in that The method comprises: Establish the voltage control reward function as follows: Among them, r n is the reward of sub-agent n, Ω n，O is the set of observation points in the sub-agent n region, V i is the voltage amplitude of node i, and V0 is the rated voltage of the network.

3. The method according to claim 1, characterized in that The method comprises: Record the control results of n cluster control chains and generate steady-state identification of the cluster control chains; generating an active control compensation according to the steady-state indicator; The active control compensation is used to perform pre-compensation control of n cluster control chains.

4. Distributed power adaptive voltage control system based on deep learning, characterized in that: For implementing the steps of the method according to any one of claims 1 to 3, the system comprises: A group control chain establishment module, which is used to adaptively partition the distributed power supply and establish n cluster control chains, wherein each cluster area corresponds to an edge device and a sub-agent, and the n sub-agents are all connected to a shared update database, which is a processing database set in the cloud that can be adaptively updated and iterated; A historical compensation data establishment module, which is used to establish historical compensation data of distributed power sources, wherein the historical compensation data is verified and updated compensation data, and the historical compensation data is used as training data to establish a basic solution generation model, and the solution generation model is stored in a shared update database; A control parameter generation module, wherein when any edge device collects status data of a corresponding cluster area, the control parameter generation module is used to send the status data to the solution generation model, generate control parameters, control the edge device with the control parameters, monitor key nodes in the cluster area, and generate node update status; A database update module, the database update module is used to generate rewards according to the node update status and status data, perform action evaluation on the control parameters with the rewards, and share the action evaluation results to the shared update database; An iterative update module, wherein the iterative update module is used to execute iterative update of the basic solution generation model after the shared update database receives the action evaluation result, and make subsequent control decisions of the distributed power source according to the iterative update result.

Images