A hierarchical distributed power distribution protection and self-healing control method and system

Abstract

This application provides a hierarchical distributed power distribution protection and self-healing control method and system, relating to the technical field of regulating distributed power distribution systems. The method includes: establishing a knowledge base to store multiple training samples containing historical operating data and sample labels of electrical equipment, as well as a self-healing strategy corresponding to each fault label; then collecting real-time operating data of electrical equipment, constructing a BI-LSTM model and training it using the training samples, and inputting real-time data to output classification results after obtaining the trained model; finally, making decisions based on the classification results: if it is a fault label, the self-healing strategy in the knowledge base is invoked to regulate the electrical equipment; if it is not a fault label, no processing is performed, so as to realize intelligent fault prediction and processing, and improve the sensitivity and reliability of distributed power distribution system protection.

Description

Technical Field

[0001] This application relates to the technical field of regulating distributed power distribution systems, and in particular to a hierarchical distributed power distribution protection and self-healing control method and system. Background Technology

[0002] In the current power system field, the safe and stable operation of the distribution network is crucial to ensuring the normal operation of social production and life. With the widespread integration of distributed energy resources, the increasing demand for loads, and the increasingly complex network structure, the distribution network faces many severe challenges. From the perspective of distributed energy integration, distributed power sources such as wind power and photovoltaics are characterized by intermittency and volatility, which makes the power flow distribution of the distribution network more complex and difficult to predict.

[0003] Chinese invention patent application publication number CN108616114A discloses a distributed hierarchical protection system and method for a distribution network containing microgrids. This patent protects the distributed distribution network by setting different protection layers and using the upper protection layer as the lower protection layer.

[0004] However, the power distribution protection and control method provided by this patent only uses simple logical judgment when dealing with such complex and ever-changing power access situations. It cannot fully consider the operating status of electrical equipment and complex fault scenarios, resulting in low sensitivity and reliability of protection. Summary of the Invention

[0005] To improve the sensitivity and reliability of protection for distributed power distribution systems, this application provides a hierarchical distributed power distribution protection and self-healing control method and system.

[0006] In the first aspect, this application provides a hierarchical distributed power distribution protection and self-healing control method, which adopts the following technical solution:

[0007] A hierarchical distributed power distribution protection and self-healing control method includes the following steps:

[0008] Database construction: Establish a knowledge base that stores multiple training samples and self-healing strategies corresponding to each fault label. The training samples include: historical operating data and sample labels of various electrical devices. The sample labels include normal labels and multiple fault labels.

[0009] Prediction: Collect real-time operating data of various electrical devices, construct a BI-LSTM model, train the BI-LSTM model using training samples, obtain the trained BI-LSTM model, input the real-time operating data into the trained BI-LSTM model, and output the classification result;

[0010] Fault diagnosis: Determine whether the classification result is a fault label. If so, call the self-healing strategy in the knowledge base based on the classification result and use the self-healing strategy to regulate the electrical equipment; otherwise, do not take any action.

[0011] By adopting the above technical solution, this application achieves intelligent prediction and self-healing control of electrical equipment faults in distributed power distribution systems. By establishing a knowledge base, training a prediction model, and implementing corresponding self-healing strategies based on the prediction results, real-time monitoring of the operating status of electrical equipment and automatic fault handling are achieved, thereby improving the stability and reliability of the system. This application links fault tags with self-healing strategies, enabling the rapid identification of corresponding handling methods when faults are subsequently detected. Subsequently, this application constructs a BI-LSTM (Bidirectional Long Short-Term Memory) model for fault prediction. BI-LSTM can capture long-term dependencies in sequential data, making it suitable for processing time-series operating data of electrical equipment. This model can learn complex patterns and trends from historical data, thus more accurately predicting the current state of the equipment. Subsequently, this application trains the BI-LSTM model using training samples, enabling the model to output classification results based on the input real-time operating data. Through extensive training, the model can continuously optimize its parameters, improving the accuracy of prediction.

[0012] This application also inputs the real-time operating data of various electrical devices into the trained model to obtain classification results, enabling real-time monitoring of the equipment's operating status. Subsequently, this application determines whether the classification result is a fault label to quickly identify whether the equipment is in a faulty state. This application adopts a model-output-based judgment method, which is relatively objective and accurate, reducing the subjectivity and lag of manual judgment. When a fault is detected, this application can invoke self-healing strategies from the knowledge base based on the classification results to regulate the electrical equipment. This allows the application to automatically take measures to repair faults when they occur, reducing the impact of faults on the distributed power distribution system and improving the self-healing capability of the distributed power distribution system. This application combines a knowledge base and a BI-LSTM model to achieve automatic fault detection and self-healing regulation. When a fault occurs, this application can automatically take corresponding measures to repair it, reducing manual intervention and improving the operating efficiency and stability of the distributed power distribution system.

[0013] Optionally, the method further includes:

[0014] Modeling: Obtain the topology of the distributed power distribution system, which includes the connection relationships and locations of electrical equipment in the distributed power distribution system; calculate the connection distance between electrical equipment based on the topology, classify electrical equipment with a connection distance less than a preset threshold into the same level, and add the same level label to electrical equipment belonging to the same level;

[0015] In the prediction step, real-time operating data of electrical equipment with the same hierarchical labels are collected in batches, and the classification results of each hierarchical label are obtained.

[0016] The distributed power distribution system topology obtained in this application encompasses the connection relationships and locations of electrical equipment. Connection relationships reflect the electrical connections between equipment, and location information helps assess the impact of the physical layout of electrical equipment on the operation of the distributed power distribution system. Subsequently, this application calculates the connection distances between electrical equipment based on the topology. These connection distances refer to electrical distances, including factors such as line impedance. Then, this application groups electrical equipment with connection distances less than a preset threshold into the same level and adds the same level label to equipment within the same level, thereby enabling the grouping of equipment in the distributed power distribution system according to their degree of interconnectivity.

[0017] This application also collected real-time operating data of electrical equipment with the same hierarchical labels in batches. By collecting data hierarchically, this application fully utilizes the hierarchical information and improves the correlation and systematic nature of the collected data. This application also obtained the classification results for each hierarchical label. The classification results may be a judgment of the operating status of the equipment at that level, such as normal or abnormal, or they may be a classification of the type of equipment fault. By adopting the above scheme, it is helpful to quickly locate and handle problems in distributed power distribution systems.

[0018] By dividing the data into hierarchical levels, this application simplifies complex distributed power distribution systems, facilitating management and analysis. Electrical equipment at different levels shares similar electrical characteristics; processing data hierarchically improves the accuracy and efficiency of the analysis.

[0019] Optionally, in the fault determination step, when the classification result is a fault label, the method further includes:

[0020] Isolation: Obtain the hierarchical label corresponding to the classification result, and record it as the first label. Record the electrical equipment corresponding to the first label as the target equipment. Isolate and control the target equipment until the fault is eliminated, and then release the target equipment from isolation.

[0021] This application identifies target devices by obtaining hierarchical labels corresponding to the classification results, enabling precise location of faulty electrical equipment. In distributed power distribution systems, numerous and interconnected devices exist. The hierarchical label-based location method employed in this application leverages the hierarchical division results from the previous modeling phase, resulting in more accurate and efficient fault location. After identifying the target device, this application isolates it, effectively preventing the fault from spreading to other devices or areas, minimizing the scope of the fault's impact, and ensuring the normal operation of other parts of the system. Furthermore, this application regulates the isolated target device until the fault is eliminated. Compared to large-scale investigation and processing of the entire distributed power distribution system, this application employs a target device-specific regulation method, enabling more efficient and precise control of the target device.

[0022] After the fault is cleared, this application also performs a de-isolation operation on the target equipment, enabling it to reconnect to the system and resume normal operation. By employing the above operations, this application can improve the recovery capability of the distributed power distribution system, reduce equipment downtime caused by faults, and ensure the stability and reliability of the system. Equipment at different levels has different electrical characteristics and functions. This application uses hierarchical labels for fault handling, making the fault handling process clearer and more orderly. Furthermore, this application can formulate different isolation and control strategies based on the characteristics of each level, further improving the efficiency and effectiveness of fault handling.

[0023] Optionally, the fault diagnosis step further includes:

[0024] A GNN model is constructed based on the topology. The graph nodes of the GNN model are electrical devices. The attributes of the graph nodes include: the device type of the electrical device, the historical operating data of the electrical device, the number of faults of the electrical device, and the protection action records. The graph nodes are processed into layers according to the hierarchical labels. The undirected edges of the GNN model are the connection relationships between electrical devices. Weights are assigned to the undirected edges in the GNN model. The weights are equal to the fault propagation probability calculated based on the historical operating data.

[0025] Construct the topology of the target device, input the topology of the target device into the GNN model, obtain the predicted fault propagation path, and obtain the target device under the fault propagation path, which is denoted as the fault device.

[0026] The isolation process involves isolating, regulating, or de-isolizing faulty equipment.

[0027] This application constructs a Generative Neural Network (GNN) model using electrical equipment as graph nodes and the connections between electrical equipment as undirected edges, fully considering the topology of distributed power distribution systems. The modeling method provided in this application reflects the electrical connections and mutual influences between electrical equipment, helping the GNN model understand the propagation mechanism of faults in distributed power distribution systems and providing a foundation for accurately predicting fault propagation paths. Subsequently, this application performs hierarchical processing on the graph nodes based on hierarchical labels, combining the results of previous topology hierarchical partitioning. Electrical equipment at different levels has similar electrical characteristics and functions; hierarchical processing allows the GNN model to better explore the connections between devices within the same level and the interactions between different levels, improving the analytical efficiency and accuracy of the GNN model. For example, when analyzing fault propagation, the spread of faults between different levels can be seen more clearly. Subsequently, this application assigns weights to the undirected edges in the GNN model, with the weight value equal to the fault propagation probability calculated based on historical operating data. By adopting the above scheme, this application enables the GNN model to consider the differences in the probability of fault propagation between different electrical equipment, making the prediction of fault propagation paths more consistent with reality. By learning the probability of fault propagation, GNN models can more accurately predict the scope and direction of the potential impact of a fault.

[0028] This application constructs the topology of the target device and inputs it into a GNN model to obtain the predicted fault propagation path. Based on this path, the faulty device can be accurately located, enabling a more comprehensive response to the fault situation. In the isolation step, this application performs isolation, control, or de-isolation operations on the faulty device, improving the targeting and effectiveness of fault handling and facilitating the rapid restoration of normal system operation.

[0029] Optionally, the method further includes:

[0030] Feature extraction is performed on the self-healing strategy corresponding to each fault tag stored in the knowledge base to obtain strategy features. A self-healing strategy template is formulated based on the strategy features, and the self-healing strategy template corresponding to the first tag is obtained.

[0031] Obtain the faulty devices and their number in the fault propagation path, and fuse the faulty devices and their number with the self-healing strategy template to obtain a dynamic self-healing strategy.

[0032] In the fault diagnosis process, a dynamic self-healing strategy is used to regulate the faulty equipment.

[0033] This application extracts features from the self-healing strategies corresponding to each fault label in the knowledge base, and then formulates self-healing strategy templates. This achieves the abstraction and standardization of self-healing strategies, transforming specific strategies into universal templates. The strategy templates can capture the common features of different self-healing strategies, enabling rapid reuse and adjustment of strategies when facing similar fault scenarios, thus improving the efficiency of fault handling.

[0034] Subsequently, this application obtains the corresponding self-healing strategy template based on the first tag, and then selects a suitable template based on the hierarchical information of the faulty device. Devices at different levels may have different fault characteristics and handling requirements. The above matching method can make the selected template more consistent with the actual situation of the faulty device, laying the foundation for formulating an effective dynamic self-healing strategy. Next, this application obtains the faulty devices and their number contained in the fault propagation path, and integrates this information with the self-healing strategy template to generate a dynamic self-healing strategy. By adopting the above scheme, this application considers the actual situation under specific fault scenarios, making the self-healing strategy no longer fixed, but dynamically adjustable according to the specific scope and severity of the fault. Using the dynamic self-healing strategy to regulate the faulty device in the fault judgment step allows for the adoption of the most suitable handling measures. Since the dynamic strategy is generated based on the specific fault situation, it can more accurately locate the fault problem, take effective repair and regulation measures, improve the success rate of fault handling, and reduce the impact of the fault on the system. Through precise fault regulation, this application can quickly restore the normal operation of the faulty device, reduce the propagation and spread of the fault in the system, thereby improving the stability and reliability of the entire distributed power distribution system.

[0035] Optionally, the method further includes:

[0036] Quantity judgment: Determine whether there are multiple fault tags. If so, use a genetic algorithm to sort the self-healing strategies corresponding to the multiple fault tags to obtain the sorting result of the self-healing strategies. According to the sorting result of the self-healing strategies, use the self-healing strategies to regulate the electrical equipment. If not, do not take any action.

[0037] When multiple fault labels appear, distributed power distribution systems face the choice of multiple self-healing strategies. Genetic algorithms, an optimization algorithm based on natural selection and genetic mechanisms, can find the optimal solution among numerous possible strategy combinations. By using a genetic algorithm to sort the self-healing strategies corresponding to multiple fault labels, this application can determine an optimal processing order, enabling the distributed power distribution system to sequentially regulate electrical equipment according to this order, thus improving fault handling efficiency. A reasonable sorting of self-healing strategies helps the system allocate resources more rationally. For example, for faults with significant impact on system operation and high severity, prioritizing the corresponding self-healing strategy can minimize the impact of the fault on the entire distributed power distribution system, ensuring stable system operation. Genetic algorithms are adaptive, dynamically adjusting the sorting of self-healing strategies based on the actual system conditions and fault characteristics. As the system's operating state changes and fault conditions differ, the genetic algorithm can continuously optimize the sorting results, making the fault handling process more intelligent and flexible. In complex situations where multiple faults occur simultaneously, manually judging and selecting the order of self-healing strategies can be very difficult and prone to errors. The automatic sorting function of genetic algorithms can simplify this decision-making process, provide operators with a clear fault handling sequence, reduce the impact of human factors on fault handling, and improve the accuracy and reliability of the process.

[0038] Optionally, the method further includes:

[0039] Strategy judgment: After the current self-healing strategy is completed, the proportion of the number of devices that successfully eliminated the fault to the total number of faulty devices is counted and recorded as the fault elimination rate; after the second self-healing strategy is completed, it is determined whether the current fault elimination rate is greater than the previous fault elimination rate. If so, no action is taken; otherwise, the decomposition strategy steps are executed.

[0040] Decomposition Strategy: Decompose the self-healing strategy corresponding to the current fault elimination rate to obtain multiple sub-strategies. Execute the sub-strategies one by one. Calculate the fault elimination rate of each sub-strategy and record it as the first data. Delete the sub-strategies whose first data is less than the previous fault elimination rate from the current self-healing strategy. Integrate the remaining sub-strategies into a new self-healing strategy. Replace the current self-healing strategy with the new self-healing strategy and execute the strategy judgment steps until all self-healing strategies are executed in sequence.

[0041] This application dynamically evaluates the effectiveness of each self-healing strategy by calculating the fault clearance rate and comparing it with the fault clearance rates after the execution of adjacent strategies. If the current fault clearance rate is not higher than the previous one, it indicates that the current strategy has a problem and needs to be optimized. This dynamic evaluation mechanism enables the distributed power distribution system to adjust its strategies based on the actual fault handling effect, improving the adaptability and flexibility of fault handling. Subsequently, this application decomposes the ineffective self-healing strategies into multiple sub-strategies. By executing each sub-strategy one by one and calculating its fault clearance rate, the sub-strategies with better performance are selected and then integrated into a new self-healing strategy. This approach allows for fine-tuning of the self-healing strategy, removing ineffective or inefficient parts, retaining and strengthening effective parts, thereby continuously optimizing the self-healing strategy and improving the efficiency of fault handling.

[0042] After decomposing and optimizing the self-healing strategy, the new strategy can handle faults more effectively. Because it is selected based on actual fault elimination rates, it better meets the needs of the current fault scenario and can better cope with faults of different types and degrees, thereby improving the success rate of fault handling. By continuously executing the steps of strategy judgment and strategy decomposition, this application can gradually eliminate faults until all self-healing strategies are executed sequentially. This step-by-step approach ensures comprehensive and in-depth fault handling, reducing situations where faults cannot be resolved due to the ineffectiveness of a single strategy.

[0043] Optionally, the decomposition strategy further includes the following steps:

[0044] A genetic algorithm is used to sort each sub-strategy, and the sorting results are obtained. The sub-strategies are then executed one by one according to the sorting results.

[0045] When faced with multiple sub-strategies, this application uses a genetic algorithm to sort them, determining the most advantageous execution order for fault elimination. For example, for some interrelated sub-strategies, a reasonable execution order can avoid redundant operations, reduce fault handling time and resource consumption, and thus improve overall fault handling efficiency. Different sub-strategies may have different effects on faulty equipment, and there may be synergistic effects between them. During the sorting process, the genetic algorithm can comprehensively consider the interrelationships between sub-strategies, ensuring that sub-strategies are executed in the order that maximizes synergy, thereby better utilizing the advantages of each sub-strategy and improving the effectiveness of fault elimination.

[0046] Genetic algorithms are adaptive, dynamically adjusting the sorting results based on the specific fault conditions and the actual effectiveness of sub-strategies. As the fault handling process progresses, the state and fault characteristics of the distributed power distribution system may change. Genetic algorithms can adapt to these changes in real time, re-evaluating the merits of sub-strategies and adjusting the execution order to handle the fault in the most efficient way.

[0047] Secondly, this application provides a hierarchical distributed power distribution protection and self-healing control system, which adopts the following technical solution:

[0048] A hierarchical distributed power distribution protection and self-healing control system includes: a memory and a processor.

[0049] The memory contains a computer-readable storage medium;

[0050] When the processor processes a computer program stored on the computer-readable storage medium, it implements the method.

[0051] In summary, this application includes at least one of the following beneficial technical effects:

[0052] 1. This application combines a knowledge base and a BI-LSTM model to achieve automatic fault detection and self-healing control. When a fault occurs, this application can automatically take corresponding measures to repair it, reducing manual intervention and improving the operating efficiency and stability of the distributed power distribution system.

[0053] 2. By dividing the data into hierarchical layers, this application simplifies complex distributed power distribution systems, facilitating management and analysis. Electrical equipment at different layers possess similar electrical characteristics; processing data hierarchically improves the accuracy and efficiency of the analysis. Attached Figure Description

[0054] Figure 1 This is a flowchart of Embodiment 1 of this application. Detailed Implementation

[0055] The present application will be further described in detail below with reference to the accompanying drawings.

[0056] Example 1: This example discloses a hierarchical distributed power distribution protection and self-healing control method, referring to... Figure 1 The method includes: S11 database construction, S12 prediction, and S13 fault judgment. First, a knowledge base is established through the database construction step, storing multiple training samples containing historical operating data of electrical equipment and sample labels (including normal and multiple fault labels), as well as a self-healing strategy corresponding to each fault label. Next, in the prediction stage, real-time operating data of the electrical equipment is collected, a BI-LSTM model is constructed, and it is trained using the training samples. After obtaining the trained model, real-time data is input to output the classification result. Finally, in the fault judgment stage, a decision is made based on the classification result. If it is a fault label, the self-healing strategy in the knowledge base is invoked to regulate the electrical equipment; if it is not a fault label, no action is taken, thus achieving intelligent prediction and handling of system faults. This embodiment includes the following steps:

[0057] The S11 database is used to build a knowledge base. This knowledge base primarily stores two core types of content: training samples and self-healing strategies. Training samples are collected from the historical operation of electrical equipment, containing historical operating data for various devices at different time periods. This data includes electrical parameters such as voltage, current, and power, comprehensively reflecting the operating status of the equipment. Sample labels assign clear attributes to this historical operating data. The "normal" label identifies data where the equipment is in normal operating condition, while multiple "fault" labels correspond to different types of faults, such as short-circuit faults and overload faults, giving the historical operating data clear classification attributes. Self-healing strategies are historical processing plans developed for each fault label. These plans are based on professional knowledge and practical experience, including specific operational steps such as fault isolation and equipment repair, providing direct guidance for subsequent fault handling.

[0058] S12 predicts that real-time operating data of various electrical devices is collected through sensors, reflecting the current actual operating status of the electrical devices. Subsequently, this application constructs a BI-LSTM (Bidirectional Long Short-Term Memory) model, an advanced deep learning model particularly suitable for processing data with time-series characteristics. The BI-LSTM model is trained using training samples stored in a knowledge base. During training, the BI-LSTM model continuously learns the features and patterns in historical data and adjusts its parameters. After repeated training, a trained BI-LSTM model is obtained, enabling it to analyze and judge input data. Finally, the real-time collected operating data is input into the trained BI-LSTM model. After processing, the BI-LSTM model outputs the corresponding classification result, determining whether the equipment is currently in normal operation or a fault state.

[0059] S13 Fault Judgment: This step first judges the classification result output by the BI-LSTM model. If the classification result shows a fault label, it indicates that the electrical equipment has malfunctioned. In this case, based on the classification result, this application will call the self-healing strategy corresponding to the fault label from the knowledge base. Then, according to the specific operation steps set in the self-healing strategy, the faulty electrical equipment will be controlled, such as disconnecting the faulty circuit and adjusting equipment parameters, to achieve fault repair and handling. If the classification result is a normal label, it indicates that the electrical equipment is in normal operation, and no action is taken; real-time monitoring of the equipment continues.

[0060] This application first constructs a knowledge base that stores training samples (including historical operating data of electrical equipment and normal and multiple types of fault labels) and self-healing strategies corresponding to each fault label; then it collects real-time operating data of electrical equipment, constructs a BI-LSTM model and trains it with training samples, inputs real-time data into the trained model and outputs classification results; if the classification result is a fault label, the self-healing strategy in the knowledge base is called to regulate the electrical equipment; if it is not a fault label, no processing is performed. This application realizes intelligent processing of electrical equipment faults.

[0061] Example 2: This example differs from Example 1 in that the method further includes:

[0062] S21 Modeling: This embodiment obtains the topology of the distributed power distribution system. This topology records in detail the connections between electrical devices, such as which devices are connected in series, which are connected in parallel, and their specific locations in the actual physical space. The connections determine the power transmission path and the mutual influence between devices, while location information helps to consider the potential impact of geographical factors on the operation of the distributed power distribution system. Subsequently, this application calculates the connection distances between electrical devices based on the topology. These connection distances are not simply physical distances, but rather electrical distances that comprehensively consider factors such as electrical parameters and line characteristics. By setting a preset connection distance threshold, electrical devices with connection distances less than this threshold are grouped into the same level, and devices in the same level are given the same level label. Through this hierarchical division method, this application can group electrical devices with close electrical connections in the distributed power distribution system together, facilitating subsequent targeted data processing and analysis by level, and simplifying the management and research of complex systems.

[0063] In S12 prediction, real-time operating data of electrical equipment with the same hierarchical label is collected in batches according to hierarchical labels. Equipment at the same level has similar electrical characteristics, making its operating data highly correlated and valuable for reference, resulting in more systematic and targeted data. This real-time operating data is then input into a pre-trained BI-LSTM model. Through the analysis and processing of the BI-LSTM model, the classification results corresponding to each hierarchical label are obtained, thereby determining whether the electrical equipment at each level is currently in a normal operating state or a fault state. This hierarchical data collection and prediction method can more accurately capture changes in equipment operating status, improving the accuracy and efficiency of prediction.

[0064] When the classification result predicted in S12 is a fault label, it means that there is a faulty electrical device in the distributed power distribution system. At this time, the sub-step of fault judgment in S13, S131 isolation, needs to be executed.

[0065] S131 Isolation: This application, by obtaining the hierarchical label (i.e., the first label) corresponding to the classification result, can quickly locate the level where the fault occurs, and then determine the electrical equipment corresponding to that hierarchical label as the target equipment. Isolation is performed on the target equipment, cutting off its connection with other parts of the distributed power distribution system to prevent further fault propagation. Simultaneously, the target equipment is regulated, such as adjusting equipment parameters and attempting to repair the fault. After the fault is successfully eliminated, the target equipment is de-isolated, allowing it to reconnect to the distributed power distribution system and resume normal operation. This scheme can quickly and accurately locate and handle faulty equipment when a fault occurs, minimizing the impact of the fault on the entire distributed power distribution system.

[0066] Example 3: The difference between this example and Example 2 is that the fault judgment in S13 further includes:

[0067] A Generative Neural Network (GNN) model is constructed based on the topology, where graph nodes represent electrical devices. Each node has predefined attributes, including the device type (e.g., transformer, circuit breaker), historical operating data (e.g., voltage, current, power, etc., changing over time), fault count (the frequency of past faults), and protection action records (information on device-triggered protection mechanisms). These attributes comprehensively and meticulously describe the characteristics and historical states of the electrical devices, providing rich information for the GNN model and helping it to more accurately learn and analyze the relationships between devices and the patterns of fault occurrence.

[0068] Based on the hierarchical labels added to electrical equipment in the previous S21 modeling, the graph nodes are processed into layers, thereby grouping electrical equipment with similar electrical characteristics and connections into the same layer. This allows the GNN model to better capture the differences and connections between different layers, improving the analysis efficiency and accuracy of the GNN model.

[0069] In the GNN model, undirected edges represent the connections between electrical devices, reflecting the electrical associations between them. Weights are assigned to these undirected edges, with the weight value equal to the fault propagation probability calculated based on historical operating data.

[0070] By conducting in-depth analysis and statistics of historical operational data, the likelihood of a fault propagating from one device to another can be estimated; this likelihood is known as the fault propagation probability. The weight allocation in this step allows the GNN model to consider the probabilistic differences in fault propagation during the learning process, thereby more accurately predicting the fault propagation path.

[0071] After identifying the faulty target device, its topology is constructed. The topology of the faulty target device describes the connection relationships and hierarchical information between the target device and other related devices.

[0072] The topology of the target device is input into a pre-built GNN model. Based on its learned device features, connectivity relationships, and fault propagation probabilities, the GNN model predicts possible fault propagation paths. By analyzing edge weights and device attributes between graph nodes, the GNN model can infer other devices most likely to be affected by the fault, thus obtaining the predicted fault propagation path.

[0073] Based on the predicted fault propagation path, the target devices along the path are identified and marked as faulty devices. These faulty devices are those that may be affected during the fault propagation process and require further processing and monitoring.

[0074] Obtain the hierarchical label (i.e., the first label) corresponding to the classification result, and identify the electrical equipment corresponding to the first label as the target equipment. Combining the hierarchical division and the analysis results of the GNN model, the specific equipment or group of equipment where the fault occurred can be located more accurately.

[0075] Isolate the identified faulty equipment by disconnecting it from other parts of the distributed power distribution system to prevent further spread of the fault. Simultaneously, based on the specific fault details and self-healing strategies in the knowledge base, adjust the faulty equipment, such as by modifying equipment parameters or restarting it, to attempt to eliminate the fault. Once the fault is successfully eliminated, remove the equipment from isolation, allowing it to reconnect to the system and resume normal operation.

[0076] In other embodiments, the method further includes:

[0077] For each fault tag stored in the knowledge base, feature extraction is performed on the self-healing strategy corresponding to it. Key features and operational steps within the strategy are identified to obtain strategy characteristics. Based on these strategy characteristics, a self-healing strategy template is developed. The strategy template is an abstraction and summary of self-healing strategies for similar faults.

[0078] Subsequently, this step retrieves the corresponding self-healing strategy template based on the first tag, obtains information such as the faulty devices and their number within the fault propagation path, and integrates these actual fault conditions with the self-healing strategy template. Considering different fault scenarios and the number of devices, the strategy template is adjusted and optimized to generate a dynamic self-healing strategy. The dynamic self-healing strategy can better adapt to specific fault situations, improving the targeting and effectiveness of fault handling.

[0079] In S13 fault diagnosis, a generated dynamic self-healing strategy is used to regulate the faulty equipment. Following the operating procedures and parameter settings specified in the dynamic strategy, the faulty equipment is precisely handled to eliminate the fault as quickly as possible and restore the system to normal operation.

[0080] This embodiment achieves accurate prediction of fault propagation paths by constructing a GNN model. Combined with hierarchical equipment management and the generation and application of dynamic self-healing strategies, it forms a complete and efficient fault handling solution for distributed power distribution systems, thereby effectively improving the reliability and stability of distributed power distribution systems.

[0081] Example 4: This example differs from Example 1 in that the method further includes:

[0082] S31 Quantity Judgment: This step statistically analyzes the fault labels in the classification results to determine whether there is a single fault label or multiple fault labels. If the judgment result shows that there are multiple fault labels, it means that the distributed power distribution system is facing a relatively complex fault situation. At this time, a genetic algorithm is used to rank the self-healing strategies corresponding to multiple fault labels. The genetic algorithm is an optimization algorithm based on natural selection and genetic mechanisms. It simulates the biological evolution process and finds the optimal solution through continuous iteration and optimization. In this step, the genetic algorithm comprehensively considers various factors, such as the severity of the fault, the implementation cost of the self-healing strategy, and the impact of the strategy on system stability, to evaluate and rank the self-healing strategies. After ranking, the control system will regulate the electrical equipment in sequence according to the ranking results, prioritizing the self-healing strategies with better effects and more urgent needs, so as to improve the efficiency and overall effectiveness of fault handling.

[0083] If the assessment result shows only a single fault tag, it indicates that the fault is relatively simple. In this case, the distributed power distribution system does not require any additional processing.

[0084] The S32 strategy determines that after the current self-healing strategy is completed, the number of devices that successfully eliminated faults is counted, and the proportion of these devices to the total number of faulty devices is calculated. This proportion is recorded as the fault elimination rate. The fault elimination rate is an important indicator for measuring the effectiveness of the self-healing strategy, as it directly reflects the effectiveness of the self-healing strategy in eliminating faults.

[0085] Starting from the completion of the second self-healing strategy, the distributed power distribution system compares the current fault clearance rate with the previous fault clearance rate. If the current fault clearance rate is greater than the previous fault clearance rate, it indicates that the current strategy is more effective than the previous one, and the distributed power distribution system does not take any action, continuing to execute the next self-healing strategy according to the ranking results. If the current fault clearance rate is less than or equal to the previous fault clearance rate, it indicates that the current strategy is ineffective, requiring optimization of the current self-healing strategy and the execution of the S33 decomposition strategy.

[0086] The S33 decomposition strategy decomposes the self-healing strategy corresponding to the current fault elimination rate into multiple sub-strategies. Each sub-strategy represents a specific operation or combination of operations in the original strategy. Then, a genetic algorithm is used again to rank each sub-strategy. The genetic algorithm evaluates and ranks the sub-strategies based on factors such as their expected effect, implementation difficulty, and impact on the system, determining the execution order of the sub-strategies.

[0087] According to the order of the sub-strategies, the distributed power distribution system executes the sub-strategies one by one. For each sub-strategy executed, the distributed power distribution system calculates the fault clearance rate for that sub-strategy and records it as the first data point. After execution, the distributed power distribution system reverts to the state before executing the sub-strategy. Subsequently, this embodiment compares the first data point with the previous fault clearance rate to determine the effectiveness of the sub-strategy's execution.

[0088] In the current self-healing strategy, the distributed power distribution system deletes sub-strategies whose first data is less than the previous fault clearance rate, as these sub-strategies are considered ineffective. Then, the remaining sub-strategies are integrated into a new self-healing strategy. This new strategy is an optimized strategy that improves the fault clearance rate. After replacing the current self-healing strategy with the new one, the S32 strategy judgment is executed again to continue evaluating the effectiveness of the new strategy. This process iterates continuously until all self-healing strategies are executed sequentially.

[0089] This embodiment can dynamically adjust and optimize the self-healing strategy when faced with complex fault conditions, continuously improve the efficiency and effectiveness of fault handling, and ensure the stable operation of the distributed power distribution system.

[0090] Example 5: This example discloses a hierarchical distributed power distribution protection and self-healing control system, the system including: a memory and a processor.

[0091] The memory contains a computer-readable storage medium;

[0092] When the processor processes a computer program stored on the computer-readable storage medium, it implements the method.

[0093] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A hierarchical distributed power distribution protection and self-healing control method, characterized in that, include: Database construction: Establish a knowledge base that stores multiple training samples and self-healing strategies corresponding to each fault label. The training samples include: historical operating data and sample labels of various electrical devices. The sample labels include normal labels and multiple fault labels. Prediction: Collect real-time operating data of various electrical devices, construct a BI-LSTM model, train the BI-LSTM model using training samples, obtain the trained BI-LSTM model, input the real-time operating data into the trained BI-LSTM model, and output the classification result; Fault diagnosis: Determine whether the classification result is a fault label. If so, call the self-healing strategy in the knowledge base based on the classification result and use the self-healing strategy to regulate the electrical equipment; otherwise, do not take any action. Modeling: Obtain the topology of the distributed power distribution system, which includes the connection relationships and locations of electrical equipment in the distributed power distribution system; calculate the connection distance between electrical equipment based on the topology, classify electrical equipment with a connection distance less than a preset threshold into the same level, and add the same level label to electrical equipment belonging to the same level; In the prediction step, real-time operating data of electrical equipment with the same hierarchical labels are collected in batches, and the classification results of each hierarchical label are obtained. In the fault determination step, when the classification result is a fault label, the method further includes: Isolation: Obtain the hierarchical label corresponding to the classification result, and record it as the first label. Record the electrical equipment corresponding to the first label as the target equipment. Isolate and control the target equipment until the fault is eliminated, and then release the target equipment from isolation. A GNN model is constructed based on the topology. The graph nodes of the GNN model are electrical devices. The attributes of the graph nodes include: the device type of the electrical device, the historical operating data of the electrical device, the number of faults of the electrical device, and the protection action records. The graph nodes are processed into layers according to the hierarchical labels. The undirected edges of the GNN model are the connection relationships between electrical devices. Weights are assigned to the undirected edges in the GNN model. The weights are equal to the fault propagation probability calculated based on the historical operating data. Construct the topology of the target device, input the topology of the target device into the GNN model, obtain the predicted fault propagation path, and obtain the target device under the fault propagation path, which is denoted as the fault device. The isolation process involves isolating, regulating, or de-isolizing faulty equipment.

2. The hierarchical distributed power distribution protection and self-healing control method according to claim 1, characterized in that, The method further includes: Feature extraction is performed on the self-healing strategy corresponding to each fault tag stored in the knowledge base to obtain strategy features. A self-healing strategy template is formulated based on the strategy features, and the self-healing strategy template corresponding to the first tag is obtained. Obtain the faulty devices and their number in the fault propagation path, and fuse the faulty devices and their number with the self-healing strategy template to obtain a dynamic self-healing strategy. In the fault diagnosis process, a dynamic self-healing strategy is used to regulate the faulty equipment.

3. The hierarchical distributed power distribution protection and self-healing control method according to claim 1 or 2, characterized in that, The method further includes: Quantity judgment: Determine whether there are multiple fault tags. If so, use a genetic algorithm to sort the self-healing strategies corresponding to the multiple fault tags, obtain the sorting result of the self-healing strategies, and regulate the electrical equipment according to the sorting result of the self-healing strategies; otherwise, do not process.

4. The hierarchical distributed power distribution protection and self-healing control method according to claim 3, characterized in that, The method further includes: Strategy judgment: After the current self-healing strategy is completed, the proportion of the number of devices that successfully eliminated the fault to the total number of faulty devices is counted and recorded as the fault elimination rate; after the second self-healing strategy is completed, it is determined whether the current fault elimination rate is greater than the previous fault elimination rate. If so, no action is taken; otherwise, the decomposition strategy steps are executed. Decomposition Strategy: Decompose the self-healing strategy corresponding to the current fault elimination rate to obtain multiple sub-strategies. Execute the sub-strategies one by one. Calculate the fault elimination rate of each sub-strategy and record it as the first data. Delete the sub-strategies whose first data is less than the previous fault elimination rate from the current self-healing strategy. Integrate the remaining sub-strategies into a new self-healing strategy. Replace the current self-healing strategy with the new self-healing strategy and execute the strategy judgment steps until all self-healing strategies are executed in sequence.

5. The hierarchical distributed power distribution protection and self-healing control method according to claim 4, characterized in that, The steps of the decomposition strategy also include: A genetic algorithm is used to sort each sub-strategy, and the sorting results are obtained. The sub-strategies are then executed one by one according to the sorting results.

6. A hierarchical distributed power distribution protection and self-healing control system, characterized in that, include: Memory and processor The memory contains a computer-readable storage medium; When the processor processes a computer program stored on the computer-readable storage medium, it implements the method as described in any one of claims 1-5.

Images