Intelligent control method and system for fault location and rapid reliability recovery of power distribution network

Abstract

The invention provides an intelligent control method and system for fault positioning and rapid reliability recovery of a power distribution network, and relates to the technical field of power grids, and the method comprises the steps: collecting real-time data of each node of the power distribution network, inputting the real-time data into a pre-trained deep learning model, rapidly recognizing the fault type, and accurately positioning a fault point. Generating a preliminary fault assessment report; based on the generated preliminary fault assessment report, starting a multi-objective optimization algorithm, simulating various possible fault isolation and power grid reconstruction schemes by using a reinforcement learning technology, and selecting an optimal fault isolation and power grid reconstruction scheme; after fault isolation and power grid reconstruction are completed, the system recovery state is continuously monitored, detailed event reports and optimization suggestions are generated, and data support is provided for future fault prevention and system improvement.

Description

Technical Field

[0001] The present invention relates to power grid technology, and in particular to an intelligent control method and system for fault location and rapid reliability recovery of a distribution network. Background Art

[0002] With the rapid development of smart grids and the continuous growth of electricity demand, the reliability and power supply quality of distribution networks are facing higher and higher requirements. As an important part of the power system, the distribution network is directly facing the end users, and its operating status directly affects the users' electricity experience and the normal progress of social and economic activities. However, due to the complex structure and wide distribution of the distribution network, and its susceptibility to external factors, faults are inevitable. Therefore, quickly and accurately locating faults and achieving rapid recovery of power supply reliability have become key issues in the current operation and management of distribution networks.

[0003] The traditional distribution network fault handling method mainly relies on manual inspection and experience judgment. This method has the following disadvantages: Long fault location time: Manual inspection takes a lot of time, especially in bad weather or at night, which seriously affects the efficiency of fault handling. Low positioning accuracy: Relying on experience can easily lead to misjudgment, resulting in the expansion of the fault scope or improper handling.

[0004] Recovery strategies are not optimized enough: Manually formulating recovery plans is difficult to consider all factors in a short period of time, which may lead to suboptimal choices. Difficulty in dealing with complex failures: When faced with multiple failures or cascading failures, manual decision-making is prone to errors. Information transmission lags: On-site conditions are difficult to transmit to the control center in real time, affecting the timeliness and accuracy of decision-making.

[0005] In addition, although the existing distribution network automation system has improved the management efficiency of the distribution network to a certain extent, it still has problems such as slow response speed and poor adaptability when facing complex and changeable fault conditions. Especially in the context of large-scale distributed energy access and dynamic changes in power load, traditional centralized control strategies are difficult to meet the needs of fast and flexible control.

[0006] Therefore, there is an urgent need to develop an intelligent control system for distribution network fault location and rapid reliability recovery with higher intelligence, faster response and more optimized decision-making. Summary of the invention

[0007] The embodiments of the present invention provide an intelligent control method and system for fault location and rapid reliability recovery of a distribution network, which can solve the problems in the prior art.

[0008] According to a first aspect of the embodiments of the present invention, Provides intelligent control methods for fault location and rapid reliability recovery of distribution networks, including: Collect real-time data from each node of the distribution network, including voltage, current and power data, and input the real-time data into a pre-trained deep learning model; the deep learning model is based on a hybrid architecture of a convolutional neural network and a long short-term memory network, and quickly identifies the fault type and accurately locates the fault point by analyzing the time series characteristics and spatial distribution characteristics of the real-time data; based on the identified fault type and fault point location, combined with the preset distribution network topology information, evaluate the fault impact scope and severity, and generate a preliminary fault assessment report; Based on the generated preliminary fault assessment report, a multi-objective optimization algorithm is initiated, which comprehensively considers factors such as power supply reliability, economy and user importance; multiple possible fault isolation and grid reconstruction schemes are simulated using reinforcement learning technology, and each scheme is simulated and evaluated; the optimal fault isolation and grid reconstruction scheme is selected through iterative optimization; based on the selected optimal scheme, a detailed intelligent control instruction sequence is formulated, including specific measures such as circuit breaker operation, load transfer and distributed power generation scheduling; The formulated intelligent control instruction sequence is sent to the relevant intelligent terminal devices through a secure encrypted channel; after receiving the instructions, the intelligent terminal devices gradually execute the control operations according to the preset security protocol; during the execution process, the status information of each node is collected in real time, and this information is fed back to the central control system; the central control system uses edge computing technology to quickly process feedback information locally and evaluate the execution effect of the plan in real time; if an abnormal situation is detected or the execution effect is poor, the emergency response mechanism is immediately triggered and the control strategy is dynamically adjusted; after completing fault isolation and grid reconstruction, the system recovery status is continuously monitored, and detailed event reports and optimization suggestions are generated to provide data support for future fault prevention and system improvement.

[0009] In an optional embodiment, The deep learning model is based on a hybrid architecture of a convolutional neural network and a long short-term memory network. By analyzing the time series characteristics and spatial distribution characteristics of the real-time data, the fault type can be quickly identified and the fault point can be accurately located. By deploying smart sensors at key nodes of the distribution network, voltage, current and power data are collected in real time at a sampling frequency of 100 times per second, and the data is pre-processed, including denoising, standardization and time window segmentation, to form a data format suitable for deep learning model input; The preprocessed data is input into a deep learning model with a multi-channel structure, which corresponds to voltage, current and power data respectively. The data of each channel first passes through a series of convolutional layers and pooling layers. The convolutional layers use three sizes of convolution kernels: 3×3, 5×5 and 7×7, with 64 filters for each size. Each convolution operation is followed by a maximum pooling layer. The feature maps output by the convolution layer and the pooling layer are reshaped and input into the long short-term memory network layer, wherein the long short-term memory network layer contains 128 hidden units; Adding a self-attention layer after the LSTM layer enables the model to adaptively focus on the most relevant time steps and features; The output of the self-attention layer passes through a fully connected layer, and finally outputs the probability distribution of the fault type through a softmax classifier; The deep learning model is trained by combining large-scale simulated data with actual fault data. The cross entropy loss function and Adam optimizer are used. The learning rate is initially set to 0.001, and a learning rate decay strategy is adopted. At the same time, the dropout technology is applied to prevent overfitting, and the dropout rate is set to 0.5. The trained deep learning model is used to analyze the input real-time data, complete fault identification and location within milliseconds, and output fault type and fault location information. The fault type identification accuracy rate is not less than 98.5%, and the average error of fault location is less than 50 meters.

[0010] In an optional embodiment, According to the identified fault type and fault point location, combined with the preset distribution network topology information, the fault impact scope and severity are evaluated, and a preliminary fault assessment report is generated, including: Constructing a digital twin model of a distribution network, wherein the digital twin model of a distribution network is based on graph database technology, and represents transformers, switches, lines, loads, and power generation equipment in the distribution network as nodes, and represents the connection relationship between the nodes as directed edges, wherein the directed edges represent the direction of power flow; Setting attribute information for the node, the attribute information including device identification, rated capacity, current load, voltage level, location coordinates, device type, rated current, device status, line length, impedance parameter, load type and importance level; Through the data interface with the distribution network management system, the distribution network topology and equipment information are imported into the graph database, and a real-time data update mechanism is established with the monitoring and data acquisition system to dynamically update the attribute information of the nodes; Receiving a fault node identifier and fault type information, and locating a fault starting point in the distribution network digital twin model based on the fault node identifier; Using a depth-first search algorithm, starting from the fault starting point and following the power flow direction represented by the directed edge, the area affected by the fault is gradually explored; During the search process of the depth-first search algorithm, based on the attribute information of the currently accessed node and the adjacent nodes and in combination with the fault type, it is determined whether the fault will propagate to the adjacent nodes; The determination includes: when the fault type is a short circuit fault, checking whether there is an untriggered protection device between the current node and the adjacent node; when the fault type is an open circuit fault, determining whether the adjacent node relies on the current node for power supply; Add nodes that are determined to be affected by the fault to the affected area list, and record the parent node of each affected node for reconstructing the fault propagation path; Based on the affected area list and the fault propagation path, generate a fault impact assessment result including an affected node list and a fault propagation sequence; By using graphic visualization technology, the fault impact assessment result is converted into a visualization graphic to intuitively display the fault impact range and propagation path; Verify the method based on historical fault data, compare the predicted impact range with the actual situation, and optimize the propagation judgment logic and parameters in the depth-first search algorithm based on the verification results; An adaptive learning mechanism is implemented to automatically adjust and optimize the judgment parameters and logic of the method according to the feedback information of each actual fault.

[0011] In an optional embodiment, Based on the generated preliminary fault assessment report, a multi-objective optimization algorithm is initiated, which comprehensively considers the factors of power supply reliability, economy and user importance, including: receiving a preliminary fault assessment report, the preliminary fault assessment report including an affected area, a fault type, a fault location, and an estimated degree of impact; Based on the preliminary fault assessment report, a multi-objective optimization algorithm is started, wherein the multi-objective optimization algorithm adopts an improved non-dominated sorting genetic algorithm II to balance the three objectives of power supply reliability, economy and user importance; For the power supply reliability target, the system average interruption frequency index and system average interruption duration index are calculated and minimized. For the economic target, the equipment replacement cost, labor cost and power loss cost are considered and the total cost is minimized. For the user importance target, a five-level importance classification is adopted and different weights are assigned to each level of users. Integer coding is used for chromosome coding, and each gene represents the state of a controllable switch; the population size is set to 100, the evolutionary generations are set to 500, the single-point crossover operation is used, the crossover probability is 0.8, and the uniform mutation operation is used, with a mutation probability of 0.1; Introduce an adaptive crossover and mutation probability adjustment mechanism, calculate the average fitness value of the population every 10 generations, and when the average fitness value calculated three times changes less than 1%, increase the crossover probability by 0.05 and the mutation probability by 0.01 until the preset upper limit is reached; Network topology constraints, voltage constraints and current constraints are set. The voltage constraint requires that the voltage deviation of each node does not exceed ±7% of the rated voltage, and the current constraint requires that the line current does not exceed 80% of the rated capacity. The penalty function method is used to punish individuals that violate the constraints. In each generation of evolution, the fast non-dominated sorting method is used to stratify individuals, and the crowding distance calculation is used to maintain population diversity. A set of non-dominated solutions is output as the Pareto optimal solution set for decision makers to choose the most suitable fault recovery plan.

[0012] In an optional embodiment, Using reinforcement learning technology, we simulated a variety of possible fault isolation and grid reconstruction solutions, and conducted simulation evaluation on each solution. Through iterative optimization, we selected the optimal fault isolation and grid reconstruction solution, including: Construct a distribution network environment model, which includes network topology, load distribution, generator location and switch status information; define the environment state, which is represented by a high-dimensional vector, including the voltage, current, power factor parameters of each node, and the opening and closing state of the switch; set an action space, which is defined as the combination of operations of all controllable switches, and each action corresponds to a specific fault isolation and network reconstruction scheme; design a reward function, which takes into account the degree of power supply recovery, network loss, voltage quality and number of operations; Constructing a deep Q network, the deep Q network includes a multi-layer convolutional neural network and a fully connected layer, the input layer receives the environment state, and the output layer corresponds to the Q value of each possible action; An experience replay mechanism is set up, and a fixed-capacity experience pool is used to store transfer samples; a target network update mechanism is set up, and the target network parameters are updated every fixed training step; an ε-greedy strategy is used for action selection, and the initial ε value of the ε-greedy strategy is set to 1, and a fixed ratio is decayed each round until a minimum value is reached; and a training process is executed, including the following sub-steps: Reset the distribution network environment model and obtain the initial state; select actions according to the ε-greedy strategy; execute the selected action to obtain the reward and the next state; store the transfer samples in the experience pool; randomly extract batch samples from the experience pool for training; use the mean square error loss function and Adam optimizer to update the deep Q network parameters; regularly update the target network; The trained deep Q network is tested and evaluated, and the test evaluation includes single fault, multiple fault and cascading fault scenarios; based on the test evaluation results, the optimal fault isolation and power grid reconstruction scheme is selected.

[0013] In an optional embodiment, The formulated intelligent control instruction sequence is sent to the relevant intelligent terminal devices through a secure encrypted channel; after receiving the instruction, the intelligent terminal device gradually executes the control operation according to the preset security protocol; during the execution process, the status information of each node is collected in real time and fed back to the central control system; the central control system uses edge computing technology to quickly process the feedback information locally and evaluate the execution effect of the scheme in real time; if an abnormal situation is detected or the execution effect is not good, the emergency response mechanism is immediately triggered, and the control strategy is dynamically adjusted, including: Performing test evaluation on the trained deep Q network, wherein the test evaluation includes single fault, multiple fault and cascading fault scenarios; Select the best fault isolation and grid reconstruction solution based on the test evaluation results; The selected optimal solution is converted into an intelligent control instruction sequence and sent to the relevant intelligent terminal device through a secure encryption channel based on elliptic curve cryptography; After receiving the instruction, the intelligent terminal device gradually executes the control operation according to the preset safety protocol, which includes the environment check before the operation, the operation interval time control and the operation sequence enforcement; During the execution process, the status information of each node is collected in real time through a high-precision sensor network, and this information is fed back to the central control system through a multi-level data aggregation structure; The central control system uses a layered edge computing architecture to quickly process feedback information locally and evaluate the execution effect of the solution in real time; If an abnormal situation is detected or the execution effect is poor, the emergency response mechanism will be triggered immediately, and the control strategy will be dynamically adjusted using a multi-objective optimization algorithm.

[0014] According to a second aspect of the embodiments of the present invention, Provides intelligent control system for distribution network fault location and rapid reliability recovery, including: The first unit is used to collect real-time data of each node of the distribution network, including voltage, current and power data, and input the real-time data into a pre-trained deep learning model; the deep learning model is based on a hybrid architecture of a convolutional neural network and a long short-term memory network, and quickly identifies the fault type and accurately locates the fault point by analyzing the time series characteristics and spatial distribution characteristics of the real-time data; based on the identified fault type and fault point location, combined with the preset distribution network topology information, the fault impact scope and severity are evaluated, and a preliminary fault assessment report is generated; The second unit is used to start a multi-objective optimization algorithm based on the generated preliminary fault assessment report, which comprehensively considers factors such as power supply reliability, economy and user importance; uses reinforcement learning technology to simulate multiple possible fault isolation and power grid reconstruction schemes, and conducts simulation evaluation on each scheme; selects the optimal fault isolation and power grid reconstruction scheme through iterative optimization; and formulates a detailed intelligent control instruction sequence based on the selected optimal scheme, including specific measures such as circuit breaker operation, load transfer and distributed power generation scheduling; The third unit is used to send the formulated intelligent control instruction sequence to the relevant intelligent terminal devices through a secure encrypted channel; after receiving the instructions, the intelligent terminal devices gradually execute the control operations according to the preset security protocol; during the execution process, the status information of each node is collected in real time and this information is fed back to the central control system; the central control system uses edge computing technology to quickly process feedback information locally and evaluate the execution effect of the plan in real time; if an abnormal situation is detected or the execution effect is poor, the emergency response mechanism is immediately triggered and the control strategy is dynamically adjusted; after completing fault isolation and power grid reconstruction, the system recovery status is continuously monitored, and detailed event reports and optimization suggestions are generated to provide data support for future fault prevention and system improvement.

[0015] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0016] According to a fourth aspect of the embodiments of the present invention, A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0017] The method of this application can complete the impact assessment within a few seconds after the fault occurs, providing accurate decision-making basis for fault isolation and system recovery. For a typical urban distribution network, containing 5,000 nodes and 7,500 edges, this method can complete a complete fault impact assessment within 3 seconds with an accuracy rate of more than 95%. This efficient and accurate assessment capability is of great significance for quickly responding to distribution network faults, minimizing the scope of fault impact, and improving the reliability and resilience of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A schematic diagram of a flow chart of an intelligent control method for fault location and rapid reliability recovery of a distribution network according to an embodiment of the present invention; Figure 2It is a structural schematic diagram of an intelligent control system for fault location and rapid reliability recovery of a distribution network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0021] Figure 1 FIG. 1 is a flow chart of an intelligent control method for fault location and rapid reliability recovery of a distribution network according to an embodiment of the present invention. Figure 1 As shown, the method includes: S101. Collect real-time data of each node in the distribution network, including voltage, current and power data, and input the real-time data into a pre-trained deep learning model; the deep learning model is based on a hybrid architecture of a convolutional neural network and a long short-term memory network, and quickly identifies the fault type and accurately locates the fault point by analyzing the time series characteristics and spatial distribution characteristics of the real-time data; based on the identified fault type and fault point location, combined with the preset distribution network topology information, evaluate the fault impact scope and severity, and generate a preliminary fault assessment report; S102. Based on the generated preliminary fault assessment report, a multi-objective optimization algorithm is started, which comprehensively considers factors such as power supply reliability, economy and user importance; multiple possible fault isolation and power grid reconstruction schemes are simulated using reinforcement learning technology, and each scheme is simulated and evaluated; the optimal fault isolation and power grid reconstruction scheme is selected through iterative optimization; according to the selected optimal scheme, a detailed intelligent control instruction sequence is formulated, including specific measures such as circuit breaker operation, load transfer and distributed power generation scheduling; S103. Send the formulated intelligent control instruction sequence to the relevant intelligent terminal devices through a secure encrypted channel; after receiving the instructions, the intelligent terminal devices gradually execute the control operations according to the preset security protocol; during the execution process, the status information of each node is collected in real time, and this information is fed back to the central control system; the central control system uses edge computing technology to quickly process the feedback information locally and evaluate the execution effect of the plan in real time; if an abnormal situation is detected or the execution effect is poor, the emergency response mechanism is immediately triggered and the control strategy is dynamically adjusted; after completing fault isolation and grid reconstruction, the system recovery status is continuously monitored, and detailed event reports and optimization suggestions are generated to provide data support for future fault prevention and system improvement.

[0022] In an optional embodiment, The deep learning model is based on a hybrid architecture of a convolutional neural network and a long short-term memory network. By analyzing the time series characteristics and spatial distribution characteristics of the real-time data, the fault type can be quickly identified and the fault point can be accurately located. By deploying smart sensors at key nodes of the distribution network, voltage, current and power data are collected in real time at a sampling frequency of 100 times per second, and the data is pre-processed, including denoising, standardization and time window segmentation, to form a data format suitable for deep learning model input; The preprocessed data is input into a deep learning model with a multi-channel structure, which corresponds to voltage, current and power data respectively. The data of each channel first passes through a series of convolutional layers and pooling layers. The convolutional layers use three sizes of convolution kernels: 3×3, 5×5 and 7×7, with 64 filters for each size. Each convolution operation is followed by a maximum pooling layer. The feature maps output by the convolution layer and the pooling layer are reshaped and input into the long short-term memory network layer, wherein the long short-term memory network layer contains 128 hidden units; Adding a self-attention layer after the LSTM layer enables the model to adaptively focus on the most relevant time steps and features; The output of the self-attention layer passes through a fully connected layer, and finally outputs the probability distribution of the fault type through a softmax classifier; The deep learning model is trained by combining large-scale simulated data with actual fault data. The cross entropy loss function and Adam optimizer are used. The learning rate is initially set to 0.001, and a learning rate decay strategy is adopted. At the same time, the dropout technology is applied to prevent overfitting, and the dropout rate is set to 0.5. The trained deep learning model is used to analyze the input real-time data, complete fault identification and location within milliseconds, and output fault type and fault location information. The fault type identification accuracy rate is not less than 98.5%, and the average error of fault location is less than 50 meters.

[0023] For example, in order to achieve rapid identification and accurate location of distribution network faults, this study proposes a hybrid deep learning architecture based on convolutional neural network (CNN) and long short-term memory network (LSTM). This model makes full use of the advantages of CNN in spatial feature extraction and the ability of LSTM in time series analysis to effectively handle complex fault modes in distribution networks.

[0024] First, intelligent sensors deployed at key nodes of the distribution network collect voltage, current, and power data in real time. These data are sampled at a high frequency (e.g., 100 times per second) to capture transient fault characteristics. The collected raw data is preprocessed, including denoising, standardization, and time window segmentation, to form a data format suitable for deep learning model input.

[0025] At the input layer of the model, a multi-channel structure is adopted, corresponding to voltage, current and power data respectively. The data of each channel first passes through a series of convolutional layers and pooling layers to extract spatial distribution features. The size and number of convolution kernels are optimized to capture fault features to the greatest extent. Specifically, 3×3, 5×5 and 7×7 convolution kernels are used, with 64 filters set for each size. Each convolution operation is followed by a maximum pooling layer to reduce the feature dimension and improve the translation invariance of the model.

[0026] The feature map output by the convolutional layer is then reshaped and input to the LSTM layer. The LSTM layer contains 128 hidden units for analyzing time series features. Through the gating mechanism, LSTM is able to effectively capture long-term dependencies, which is crucial for identifying certain slowly evolving fault types. The output of the LSTM layer passes through a fully connected layer and finally outputs the probability distribution of the fault type through a softmax classifier.

[0027] In order to improve the localization accuracy of the model, an attention mechanism is introduced. After the LSTM layer, a self-attention layer is added to enable the model to adaptively focus on the most relevant time steps and features. This mechanism significantly improves the localization ability of the model in complex fault scenarios.

[0028] The model training adopts a method that combines large-scale simulated data with limited actual fault data. A large number of samples containing various fault types and locations are generated through power system simulation software, and historical fault records are collected as a supplement. The cross entropy loss function and Adam optimizer are used in the training process. The learning rate is initially set to 0.001, and the learning rate decay strategy is adopted. To prevent overfitting, the dropout technology is applied, and the dropout rate is set to 0.5.

[0029] Finally, the trained model was able to complete fault identification and location within milliseconds. On the test set, the model's fault type recognition accuracy reached 98.5%, and the average error of fault location was less than 50 meters. This efficient and accurate fault diagnosis provides a reliable foundation for subsequent fault isolation and system recovery, thereby significantly improving the reliability and recovery speed of the distribution network.

[0030] In an optional embodiment, According to the identified fault type and fault point location, combined with the preset distribution network topology information, the fault impact scope and severity are evaluated, and a preliminary fault assessment report is generated, including: Constructing a digital twin model of a distribution network, wherein the digital twin model of a distribution network is based on graph database technology, and represents transformers, switches, lines, loads, and power generation equipment in the distribution network as nodes, and represents the connection relationship between the nodes as directed edges, wherein the directed edges represent the direction of power flow; Setting attribute information for the node, the attribute information including device identification, rated capacity, current load, voltage level, location coordinates, device type, rated current, device status, line length, impedance parameter, load type and importance level; Through the data interface with the distribution network management system, the distribution network topology and equipment information are imported into the graph database, and a real-time data update mechanism is established with the monitoring and data acquisition system to dynamically update the attribute information of the nodes; Receiving a fault node identifier and fault type information, and locating a fault starting point in the distribution network digital twin model based on the fault node identifier; Using a depth-first search algorithm, starting from the fault starting point and following the power flow direction represented by the directed edge, the area affected by the fault is gradually explored; During the search process of the depth-first search algorithm, based on the attribute information of the currently accessed node and the adjacent nodes and in combination with the fault type, it is determined whether the fault will propagate to the adjacent nodes; The determination includes: when the fault type is a short circuit fault, checking whether there is an untriggered protection device between the current node and the adjacent node; when the fault type is an open circuit fault, determining whether the adjacent node relies on the current node for power supply; Add nodes that are determined to be affected by the fault to the affected area list, and record the parent node of each affected node for reconstructing the fault propagation path; Based on the affected area list and the fault propagation path, generate a fault impact assessment result including an affected node list and a fault propagation sequence; By using graphic visualization technology, the fault impact assessment result is converted into a visualization graphic to intuitively display the fault impact range and propagation path; Verify the method based on historical fault data, compare the predicted impact range with the actual situation, and optimize the propagation judgment logic and parameters in the depth-first search algorithm based on the verification results; An adaptive learning mechanism is implemented to automatically adjust and optimize the judgment parameters and logic of the method according to the feedback information of each actual fault.

[0031] For example, the fault impact assessment method based on the distribution network digital twin model and the depth-first search algorithm is an innovative distribution network fault analysis technology. This method achieves fast and accurate fault impact range assessment and propagation path tracking by building an accurate digital twin model and combining it with an efficient search algorithm. The following is the detailed technical content of this method: The construction of the digital twin model of the distribution network is the basis of this method. Neo4j is selected as the graph database platform to use its efficient graph traversal performance to represent the complex distribution network structure. In Neo4j, multiple node types are created, including transformer nodes, switch nodes, line nodes, load nodes, and power generation nodes. Each node type defines a specific set of attributes. For example, the transformer node contains attributes such as ID, rated capacity, current load, voltage level, and location coordinates; the switch node contains attributes such as ID, type (circuit breaker / disconnector), rated current, and state (open / closed); the line node contains attributes such as ID, length, impedance parameters, and rated current; the load node contains attributes such as ID, type (residential / commercial / industrial), importance level, and current power.

[0032] The connection relationship between nodes is represented by directed edges. The direction of the edge represents the direction of power flow. The attributes of the edge include connection type, current direction, etc. For example, the edge from the transformer node to the line node represents the flow of power from the transformer to the line. This structure enables the system to quickly track the power flow and fault propagation path.

[0033] Data import is a key step in model building. Through the data interface with the existing distribution network management system, the complete distribution network topology and equipment information are extracted. Using Neo4j's batch import tool, this data is efficiently imported into the graph database. For example, for a medium-sized distribution network with 1,000 nodes and 1,500 edges, batch import can be completed in a few minutes.

[0034] In order to maintain the real-time nature of the model, a real-time data update mechanism with the SCADA system is designed. Every 5 seconds, the system automatically obtains the latest equipment status data from SCADA, such as the current load of the transformer, the opening and closing status of the switch, and updates the properties of the corresponding nodes. This ensures that the digital twin model can accurately reflect the dynamic state of the actual distribution network.

[0035] The depth-first search (DFS) algorithm is the core of fault impact assessment. When receiving the fault node ID and fault type information, the algorithm first locates the fault starting point in the digital twin model. Then, the DFS is implemented using a stack data structure, starting from the fault point and gradually exploring the affected area along the power flow.

[0036] In the DFS process, the key is to determine whether the fault will propagate to the adjacent nodes. This judgment logic varies according to the fault type and node characteristics. For example, for a short-circuit fault, the algorithm checks whether there is an untriggered protection device between the current node and the adjacent node. If there is an untriggered protection device, the fault will propagate to the adjacent node. For an open-circuit fault, the algorithm determines whether the adjacent node depends on the current node for power supply. If the dependency exists, the adjacent node will be affected.

[0037] The algorithm uses a set data structure to record visited nodes to avoid repeated visits. At the same time, a dictionary structure is used to store the parent node information of each affected node for subsequent reconstruction of the fault propagation path. This method ensures that no affected nodes are missed and can efficiently track the exact path of fault propagation.

[0038] In order to improve the processing efficiency of large-scale networks, a parallel DFS algorithm is implemented. The distribution network is divided into multiple subgraphs, and each subgraph can be independently processed by DFS. For example, for a large distribution network containing 10,000 nodes, it can be divided into 10 subgraphs and processed simultaneously using 10 parallel threads, which significantly reduces the calculation time.

[0039] The output and visualization of the results are designed to make it easier for decision makers to understand and use the evaluation results. The system generates output in JSON format, including a list of affected nodes and fault propagation paths. For example, a typical output may contain 50 affected nodes and 35 propagation relationships between them. Use the NetworkX library to convert this data into a visualization graph to intuitively display the fault impact range and propagation path. Nodes can use different colors and sizes to indicate their type and importance, and the thickness of the edge can indicate the current size.

[0040] Model validation and optimization are key to ensuring the accuracy and adaptability of the method. Use historical failure data to validate the model and compare the impact range predicted by the model with the actual situation. For example, backtest 100 failure events that occurred in the past year and calculate the accuracy of the model prediction. If the accuracy is less than 90%, the propagation judgment logic and weight parameters in the DFS algorithm need to be adjusted.

[0041] Implement an adaptive learning mechanism to automatically optimize model parameters based on feedback from each actual fault. For example, if the model underestimates the impact of a certain type of fault multiple times, the system will automatically increase the weight of the corresponding propagation judgment logic. This mechanism enables the model to continuously learn and improve and adapt to the dynamic changes of the distribution network.

[0042] Through the above detailed technical steps, this method can complete the impact assessment within a few seconds after the fault occurs, providing accurate decision-making basis for fault isolation and system recovery. For a typical urban distribution network, which contains 5,000 nodes and 7,500 edges, this method can complete a complete fault impact assessment within 3 seconds with an accuracy rate of more than 95%. This efficient and accurate assessment capability is of great significance for quickly responding to distribution network faults, minimizing the scope of fault impact, and improving the reliability and resilience of the power grid.

[0043] In an optional embodiment, Based on the generated preliminary fault assessment report, a multi-objective optimization algorithm is initiated, which comprehensively considers the factors of power supply reliability, economy and user importance, including: receiving a preliminary fault assessment report, the preliminary fault assessment report including an affected area, a fault type, a fault location, and an estimated degree of impact; Based on the preliminary fault assessment report, a multi-objective optimization algorithm is started, wherein the multi-objective optimization algorithm adopts an improved non-dominated sorting genetic algorithm II to balance the three objectives of power supply reliability, economy and user importance; For the power supply reliability target, the system average interruption frequency index and system average interruption duration index are calculated and minimized. For the economic target, the equipment replacement cost, labor cost and power loss cost are considered and the total cost is minimized. For the user importance target, a five-level importance classification is adopted and different weights are assigned to each level of users. Integer coding is used for chromosome coding, and each gene represents the state of a controllable switch; the population size is set to 100, the evolutionary generations are set to 500, the single-point crossover operation is used, the crossover probability is 0.8, and the uniform mutation operation is used, with a mutation probability of 0.1; Introduce an adaptive crossover and mutation probability adjustment mechanism, calculate the average fitness value of the population every 10 generations, and when the average fitness value calculated three times changes less than 1%, increase the crossover probability by 0.05 and the mutation probability by 0.01 until the preset upper limit is reached; Network topology constraints, voltage constraints and current constraints are set. The voltage constraint requires that the voltage deviation of each node does not exceed ±7% of the rated voltage, and the current constraint requires that the line current does not exceed 80% of the rated capacity. The penalty function method is used to punish individuals that violate the constraints. In each generation of evolution, the fast non-dominated sorting method is used to stratify individuals, and the crowding distance calculation is used to maintain population diversity. A set of non-dominated solutions is output as the Pareto optimal solution set for decision makers to choose the most suitable fault recovery plan.

[0044] Exemplarily, based on the generated preliminary fault assessment report, a multi-objective optimization algorithm is started, which comprehensively considers the factors of power supply reliability, economy and user importance. The preliminary fault assessment report provides key inputs for the multi-objective optimization algorithm, including the affected area, fault type, fault location and estimated impact. Based on this information, a multi-objective optimization algorithm is designed to generate the optimal fault recovery plan. The algorithm adopts an improved non-dominated sorting genetic algorithm II (NSGA-II) to balance the three conflicting goals of power supply reliability, economy and user importance.

[0045] First, for the power supply reliability goal, the algorithm considers the system average interruption frequency index (SAIFI) and the system average interruption duration index (SAIDI). SAIFI is calculated by dividing the number of power outages per user in a year by the total number of users, while SAIDI is the sum of the annual power outage time per user divided by the total number of users. For example, if there are 10 power outages in a region in a year, affecting a total of 1,000 users and a cumulative power outage time of 500 hours, then the SAIFI is 0.01 times / user·year and the SAIDI is 0.5 hours / user·year. The algorithm aims to minimize these two indicators and improve the overall power supply reliability.

[0046] Secondly, the economic goal mainly considers the cost factors in the fault recovery process. This includes equipment replacement costs, labor costs, power loss costs, etc. For example, the cost of replacing a 10kV / 400V transformer is about 100,000 yuan, and the labor cost of an emergency repair is about 5,000 yuan / hour. The algorithm achieves the economic goal by minimizing the total cost.

[0047] Third, the user importance target reflects the different needs of different users for power supply continuity. The algorithm uses a five-level importance classification: particularly important users (such as hospitals and data centers), important users (such as large industrial enterprises), relatively important users (such as small and medium-sized enterprises), general users (such as residential areas) and non-important users (such as temporary electricity). Each type of user is assigned a different weight, for example, a particularly important user has a weight of 1.0, while a non-important user has a weight of 0.2. The algorithm prioritizes the restoration of power supply to high-weight users.

[0048] In the implementation process, the chromosome encoding of the NSGA-II algorithm uses integer encoding, and each gene represents the state of a controllable switch (0 means open, 1 means closed). The population size is set to 100, and the evolutionary generation is 500. The crossover operation uses single-point crossover with a crossover probability of 0.8; the mutation operation uses uniform mutation with a mutation probability of 0.1.

[0049] In order to accelerate convergence and improve the quality of the solution, an adaptive crossover and mutation probability adjustment mechanism is introduced. When the population diversity decreases, the crossover and mutation probabilities are increased; otherwise, they are reduced. Specifically, the average fitness value of the population is calculated once every 10 generations. If the average fitness value calculated three times in a row changes by less than 1%, the crossover probability is increased by 0.05 and the mutation probability is increased by 0.01 until the preset upper limit is reached.

[0050] The constraints of the algorithm include network topology constraints, voltage constraints, and current constraints. The network topology constraints ensure that the restored network structure is radial; the voltage constraints require that the voltage deviation of each node does not exceed ±7% of the rated voltage; the current constraints ensure that the line current does not exceed 80% of the rated capacity. Individuals that violate the constraints are punished through the penalty function method to reduce their probability of being selected.

[0051] In each generation of evolution, the fast non-dominated sorting method is used to stratify individuals, and the crowding distance calculation is used to maintain population diversity. Finally, the algorithm outputs a set of non-dominated solutions, namely the Pareto optimal solution set. Decision makers can choose the most suitable fault recovery solution from this set of solutions according to actual conditions.

[0052] In order to verify the effectiveness of the algorithm, a simulation test was conducted on a city distribution network. The distribution network consists of 200 distribution transformers, 300 lines and 500 user nodes. A medium-scale fault involving 30 user nodes was simulated. The algorithm was run on a computer with an Intel Core i7 processor and 16GB of memory, and it took about 2 minutes to obtain the optimal solution set. The final selected recovery scheme reduced SAIFI by 15% and SAIDI by 20%, while keeping the recovery cost within 90% of the budget and ensuring that all particularly important users and 95% of important users were restored to power within 30 minutes.

[0053] This multi-objective optimization method not only allows a fault recovery plan that balances various factors to be quickly developed, but also provides decision makers with multiple options, enhancing the flexibility of decision making. With the continuous optimization of the algorithm and the application of more practical cases, this method is expected to become a powerful tool for distribution network fault recovery decision-making and improve the overall reliability and resilience of the power grid.

[0054] In an optional embodiment, Using reinforcement learning technology, we simulated a variety of possible fault isolation and grid reconstruction solutions, and conducted simulation evaluation on each solution. Through iterative optimization, we selected the optimal fault isolation and grid reconstruction solution, including: Construct a distribution network environment model, which includes network topology, load distribution, generator location and switch status information; define the environment state, which is represented by a high-dimensional vector, including the voltage, current, power factor parameters of each node, and the opening and closing state of the switch; set an action space, which is defined as the combination of operations of all controllable switches, and each action corresponds to a specific fault isolation and network reconstruction scheme; design a reward function, which takes into account the degree of power supply recovery, network loss, voltage quality and number of operations; Constructing a deep Q network, the deep Q network includes a multi-layer convolutional neural network and a fully connected layer, the input layer receives the environment state, and the output layer corresponds to the Q value of each possible action; An experience replay mechanism is set up, and a fixed-capacity experience pool is used to store transfer samples; a target network update mechanism is set up, and the target network parameters are updated every fixed training step; an ε-greedy strategy is used for action selection, and the initial ε value of the ε-greedy strategy is set to 1, and a fixed ratio is decayed each round until a minimum value is reached; and a training process is executed, including the following sub-steps: Reset the distribution network environment model and obtain the initial state; select actions according to the ε-greedy strategy; execute the selected action to obtain the reward and the next state; store the transfer samples in the experience pool; randomly extract batch samples from the experience pool for training; use the mean square error loss function and Adam optimizer to update the deep Q network parameters; regularly update the target network; The trained deep Q network is tested and evaluated, and the test evaluation includes single fault, multiple fault and cascading fault scenarios; based on the test evaluation results, the optimal fault isolation and power grid reconstruction scheme is selected.

[0055] Exemplarily, a distribution network environment model is first constructed. The model contains key information such as network topology, load distribution, generator location and switch status. Take a medium-sized distribution network as an example. The network contains 200 nodes, 300 lines, 50 controllable switches, and 10 distributed generators. The network topology is represented by a graph structure, and each node stores its connection relationship, type (load, generator or transformer) and parameters. The load distribution is generated based on historical data and load forecasting models, taking into account daily load curves and seasonal changes. The generator location and capacity are input according to the actual configuration, including renewable energy such as photovoltaic and wind power. The switch state is initialized to normal operation.

[0056] Next, define the environmental state. The environmental state is represented by a high-dimensional vector, which contains parameters such as voltage, current, power factor, etc. of each node, as well as the opening and closing states of all switches. For the above 200-node distribution network, the state vector dimension is 800 (3 parameters for each node, plus 50 switch states). The voltage is expressed in per-unit values, which usually range from 0.95 to 1.05. The current is also expressed in per-unit values ​​and is normalized according to the rated capacity. The power factor range is usually between 0.8 and 1. The switch state is represented by 0 and 1 for closed and open.

[0057] Setting the action space is the next step. The action space is defined as the combination of operations of all controllable switches. For 50 controllable switches, there are theoretically 2^50 combinations. However, considering the actual operation restrictions and network security constraints, invalid actions can be reduced by preprocessing, such as prohibiting the simultaneous operation of adjacent switches. Finally, a discrete action space containing about 10,000 valid actions is obtained. Each action corresponds to a specific fault isolation and network reconstruction scheme.

[0058] Designing a reward function is the key to reinforcement learning. The reward function considers four aspects: the degree of power supply recovery, network loss, voltage quality, and number of operations. The degree of power supply recovery is expressed by the proportion of loads that have restored power, ranging from 0 to 1. Network loss is expressed by the proportion of loss increase relative to the normal operating state, which should ideally be close to 1. Voltage quality is expressed by the maximum value of the voltage deviation of all nodes, and the smaller the deviation, the better. The number of operations is expressed by the ratio of the number of switch actions to the maximum allowed number of operations, and reconstruction is encouraged to be completed with the least number of operations. These four indicators are weighted and summed to obtain the final reward value, and the weights are adjusted according to actual needs, for example, they can be set to 0.4, 0.3, 0.2, and 0.1.

[0059] Building a deep Q network is the core of decision-making. The network structure includes three layers of convolutional neural network and two layers of fully connected layers. The input layer receives an 800-dimensional state vector, and the output layer corresponds to the Q value of 10,000 possible actions. The specific structure is as follows: the first convolutional layer uses 32 3x3 convolution kernels with a step size of 1; the second convolutional layer uses 64 3x3 convolution kernels with a step size of 1; the third convolutional layer uses 64 3x3 convolution kernels with a step size of 1; the first fully connected layer contains 256 neurons; the second fully connected layer contains 128 neurons; the output layer contains 10,000 neurons. All hidden layers use the ReLU activation function, and the output layer uses the linear activation function.

[0060] To improve learning efficiency and stability, an experience replay mechanism is set up. An experience pool with a capacity of 100,000 is used to store transfer samples. Each sample contains the current state, the action performed, the reward obtained, and the next state. When the experience pool is full, the first-in-first-out strategy is used to replace the old samples.

[0061] The target network update mechanism is used to stabilize the training process. The target network parameters are updated every 1000 training steps, and the parameters of the main network are directly copied to the target network. The ε-greedy strategy is used for action selection. The initial ε value is set to 1, decayed by 0.995 in each training round, and the minimum value is set to 0.01. This allows more exploration in the early stages of training, and gradually increases the use of learned knowledge as learning progresses.

[0062] The training process includes multiple sub-steps. First, the distribution network environment model is reset and an initial fault state is randomly generated. Then, the action is selected according to the ε-greedy strategy, with a probability of ε for random selection and a probability of 1-ε for selecting the action with the largest Q value. The selected action is executed, the new network state is obtained through power flow calculation, and the reward value is calculated. This transfer sample is stored in the experience pool. 256 samples are randomly selected from the experience pool to form a batch for training. The mean square error is used as the loss function, and the Adam optimizer is used to update the deep Q network parameters, and the learning rate is set to 0.001. The target network is updated every 1000 steps. Repeat the above steps until the preset termination condition is reached, such as the number of training rounds reaches 10,000 or the average reward value tends to be stable.

[0063] The trained deep Q network was tested and evaluated. Various fault scenarios were set, including single fault (such as a single line tripping), multiple faults (such as multiple lines tripping at the same time), and cascading faults (a fault triggering a chain reaction). Each scenario was tested 100 times, and indicators such as average recovery time, recovery rate, network loss, and voltage quality were recorded.

[0064] Finally, based on the test evaluation results, the optimal fault isolation and grid reconstruction solution is selected. Usually, a solution with balanced performance in various indicators is selected, such as being able to complete fault isolation and reconstruction within 10 seconds, restore more than 95% of the load, control network losses within 105% of the original level, and all node voltage deviations not exceeding ±5% of the rated voltage. Such a solution can not only quickly restore power supply, but also ensure the economy and safety of the reconstructed network.

[0065] In an optional embodiment, The formulated intelligent control instruction sequence is sent to the relevant intelligent terminal devices through a secure encrypted channel; after receiving the instruction, the intelligent terminal device gradually executes the control operation according to the preset security protocol; during the execution process, the status information of each node is collected in real time and fed back to the central control system; the central control system uses edge computing technology to quickly process the feedback information locally and evaluate the execution effect of the scheme in real time; if an abnormal situation is detected or the execution effect is not good, the emergency response mechanism is immediately triggered, and the control strategy is dynamically adjusted, including: Performing test evaluation on the trained deep Q network, wherein the test evaluation includes single fault, multiple fault and cascading fault scenarios; Select the best fault isolation and grid reconstruction solution based on the test evaluation results; The selected optimal solution is converted into an intelligent control instruction sequence and sent to the relevant intelligent terminal device through a secure encryption channel based on elliptic curve cryptography; After receiving the instruction, the intelligent terminal device gradually executes the control operation according to the preset safety protocol, which includes the environment check before the operation, the operation interval time control and the operation sequence enforcement; During the execution process, the status information of each node is collected in real time through a high-precision sensor network, and this information is fed back to the central control system through a multi-level data aggregation structure; The central control system uses a layered edge computing architecture to quickly process feedback information locally and evaluate the execution effect of the solution in real time; If an abnormal situation is detected or the execution effect is poor, the emergency response mechanism will be triggered immediately, and the control strategy will be dynamically adjusted using a multi-objective optimization algorithm.

[0066] For example, to ensure the secure transmission of control instructions, an encryption scheme based on elliptic curve cryptography is adopted. First, the central control system exchanges public keys with each smart terminal device in advance. When sending instructions, the central system uses the AES-256 algorithm to encrypt the instruction content, and the key is generated by the ECDH key exchange protocol. The encrypted content is then digitally signed using the ECDSA algorithm to ensure the integrity and non-repudiation of the instruction.

[0067] The transmission channel uses the TLS 1.3 protocol to provide end-to-end encryption protection. To prevent man-in-the-middle attacks, the certificate pinning technology is implemented to pre-place the public key certificate of the legitimate server in the terminal device. In addition, a two-way authentication mechanism is introduced, requiring the terminal device to also provide a certificate to further improve security.

[0068] Considering the complexity of the distribution network, a distributed instruction delivery strategy is adopted. The central system first decomposes the global instruction into multiple local instructions, and then sends them to the relevant terminals through multi-channel parallel transmission. Each instruction contains a unique serial number and timestamp for subsequent synchronous execution and audit tracking.

[0069] After receiving the command, the intelligent terminal device will gradually execute the control operation according to the preset safety protocol; After receiving the encrypted instruction, the smart terminal device first performs decryption and signature verification. After verification, the device will check the timeliness and sequence integrity of the instruction to prevent replay attacks and instruction tampering. For distributed instructions, the device will wait for all related instructions to arrive before processing to ensure the consistency of the operation.

[0070] When executing control operations, the terminal equipment strictly follows the preset safety protocol. The protocol includes mechanisms such as environmental inspection before operation, operation interval time control, and operation sequence enforcement. For example, before executing a switch operation, the device will check whether the line current is within the safe range; the operation interval between adjacent switches is not less than 100 milliseconds to prevent grid oscillation; for multi-step operations, they must be executed in a predetermined order, and skipping or repetition is not allowed.

[0071] To cope with possible communication interruptions, the terminal device also implements limited local decision-making capabilities. When it detects that the communication interruption with the central system exceeds a preset threshold (such as 5 seconds), the device can autonomously perform some key operations according to the locally stored emergency plan to maintain basic system stability.

[0072] During the execution process, the status information of each node is collected in real time and fed back to the central control system; In order to achieve real-time monitoring of the distribution network status, a high-precision sensor network is deployed at key nodes. These sensors include voltage transformers, current transformers, power factor meters, etc. The sampling frequency can reach 60 times per second to ensure that transient changes are captured. At the same time, the intelligent switch device itself also has a status monitoring function, which can provide information such as switch position, operation times, and operation time.

[0073] After local preprocessing, the collected data is transmitted to the central control system through a multi-level data aggregation structure. Using edge computing technology, edge servers are deployed near the data source to perform preliminary analysis and compression on the original data, reducing the transmission bandwidth requirements. Data transmission adopts a publish-subscribe model and uses the MQTT protocol to ensure the real-time and reliability of data.

[0074] To cope with large-scale data flows, a distributed stream processing framework, such as Apache Flink, is used on the central control system to achieve millisecond-level data processing latency. The system also integrates a data quality control module to ensure the accuracy of feedback information through methods such as data consistency checks and outlier detection.

[0075] The central control system uses edge computing technology to quickly process feedback information locally and evaluate the execution effect of the solution in real time; The central control system adopts a layered architecture to distribute data processing tasks to computing nodes at different levels. At the edge layer, lightweight neural network models, such as compressed convolutional neural networks, are deployed to quickly identify abnormal patterns in power grid parameters. These models are optimized with transfer learning technology and can adapt to the characteristics of power grids in different regions.

[0076] The middle layer uses fog computing nodes, which are responsible for more complex analysis tasks, such as short-term load forecasting and power flow calculation. Ensemble learning methods are used here to combine the outputs of multiple machine learning models (such as random forests, gradient boosting trees, etc.) to improve prediction accuracy. The prediction results are compared with the actual feedback data to evaluate the execution effect of the solution.

[0077] The core layer implements global optimization and decision-making. Reinforcement learning techniques, such as the Deep Deterministic Policy Gradient (DDPG) algorithm, are used to dynamically adjust the parameters of the control strategy based on historical data and current status. The system also integrates a digital twin model for rapid simulation and prediction of the potential impact of different control strategies.

[0078] If an abnormal situation is detected or the execution effect is poor, the emergency response mechanism is immediately triggered and the control strategy is adjusted dynamically; anomaly detection adopts a multi-dimensional and multi-scale method. In the time dimension, the long short-term memory network (LSTM) is used to detect anomalies in time series data; in the spatial dimension, the graph neural network is applied to analyze the abnormal association between nodes. The system also integrates a rule-based expert system to encode the experience and knowledge of power experts for identifying complex fault modes.

[0079] When an abnormality is detected, the system triggers different levels of emergency response according to the severity and scope of the abnormality. For local minor abnormalities, the system will make small adjustments within the current control strategy framework, such as fine-tuning the operation sequence of certain switches. For large-scale or serious abnormalities, the system will initiate pre-defined emergency plans, such as quickly switching to backup power or implementing load shedding.

[0080] When dynamically adjusting the control strategy, the system will comprehensively consider multiple objectives, including power supply reliability, power quality, economy, etc. A multi-objective optimization algorithm, such as NSGA-III, is used to quickly generate a set of Pareto optimal adjustment plans. Then, through the fuzzy comprehensive evaluation method, combined with the weight of each objective in the current situation, the most suitable plan is selected for implementation.

[0081] The whole process forms a closed-loop control system. After executing the adjusted control strategy, the system will continuously monitor its effect and further optimize it based on new feedback information to ensure that the distribution network can maintain a stable and efficient operating state in the face of various abnormal situations.

[0082] Figure 2 FIG. 1 is a schematic diagram of the structure of an intelligent control system for fault location and rapid reliability recovery of a distribution network according to an embodiment of the present invention. Figure 2 As shown, the system comprises: The first unit is used to collect real-time data of each node of the distribution network, including voltage, current and power data, and input the real-time data into a pre-trained deep learning model; the deep learning model is based on a hybrid architecture of a convolutional neural network and a long short-term memory network, and quickly identifies the fault type and accurately locates the fault point by analyzing the time series characteristics and spatial distribution characteristics of the real-time data; based on the identified fault type and fault point location, combined with the preset distribution network topology information, the fault impact scope and severity are evaluated, and a preliminary fault assessment report is generated; The second unit is used to start a multi-objective optimization algorithm based on the generated preliminary fault assessment report, which comprehensively considers factors such as power supply reliability, economy and user importance; uses reinforcement learning technology to simulate multiple possible fault isolation and power grid reconstruction schemes, and conducts simulation evaluation on each scheme; selects the optimal fault isolation and power grid reconstruction scheme through iterative optimization; and formulates a detailed intelligent control instruction sequence based on the selected optimal scheme, including specific measures such as circuit breaker operation, load transfer and distributed power generation scheduling; The third unit is used to send the formulated intelligent control instruction sequence to the relevant intelligent terminal devices through a secure encrypted channel; after receiving the instructions, the intelligent terminal devices gradually execute the control operations according to the preset security protocol; during the execution process, the status information of each node is collected in real time and this information is fed back to the central control system; the central control system uses edge computing technology to quickly process feedback information locally and evaluate the execution effect of the plan in real time; if an abnormal situation is detected or the execution effect is poor, the emergency response mechanism is immediately triggered and the control strategy is dynamically adjusted; after completing fault isolation and power grid reconstruction, the system recovery status is continuously monitored, and detailed event reports and optimization suggestions are generated to provide data support for future fault prevention and system improvement.

[0083] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0084] A fourth aspect of the embodiments of the present invention is: A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0085] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent control method for fault location and rapid reliability recovery of distribution network, characterized in that: include: Collect real-time data from each node of the distribution network, including voltage, current and power data, and input the real-time data into a pre-trained deep learning model; The deep learning model is based on a hybrid architecture of a convolutional neural network and a long short-term memory network. By analyzing the time series characteristics and spatial distribution characteristics of the real-time data, the fault type can be quickly identified and the fault point can be accurately located. According to the identified fault type and fault point location, combined with the preset distribution network topology information, the fault impact scope and severity are evaluated, and a preliminary fault assessment report is generated. Based on the generated preliminary fault assessment report, a multi-objective optimization algorithm is initiated, which comprehensively considers the factors of power supply reliability, economy and user importance; Using reinforcement learning technology, simulate multiple possible fault isolation and grid reconstruction schemes, and conduct simulation evaluation on each scheme; Through iterative optimization, the optimal fault isolation and grid reconstruction scheme is selected; based on the selected optimal scheme, a detailed sequence of intelligent control instructions is formulated, including specific measures for circuit breaker operation, load transfer, and distributed generation scheduling; The formulated intelligent control instruction sequence is sent to the relevant intelligent terminal device through a secure encrypted channel; after the intelligent terminal device receives the instruction, it gradually executes the control operation according to the preset security protocol; During the execution process, the status information of each node is collected in real time and fed back to the central control system. The central control system uses edge computing technology to quickly process feedback information locally and evaluate the execution effect of the plan in real time. If an abnormal situation is detected or the execution effect is poor, the emergency response mechanism is immediately triggered and the control strategy is dynamically adjusted. After completing fault isolation and grid reconstruction, the system recovery status is continuously monitored, and detailed event reports and optimization suggestions are generated to provide data support for future fault prevention and system improvement.

2. The method according to claim 1, characterized in that The deep learning model is based on a hybrid architecture of a convolutional neural network and a long short-term memory network. By analyzing the time series characteristics and spatial distribution characteristics of the real-time data, the fault type can be quickly identified and the fault point can be accurately located. By deploying smart sensors at key nodes of the distribution network, voltage, current and power data are collected in real time at a sampling frequency of 100 times per second, and the data is pre-processed, including denoising, standardization and time window segmentation, to form a data format suitable for deep learning model input; The preprocessed data is input into a deep learning model with a multi-channel structure, which corresponds to voltage, current and power data respectively. The data of each channel first passes through a series of convolutional layers and pooling layers. The convolutional layers use three sizes of convolution kernels: 3×3, 5×5 and 7×7, with 64 filters for each size. Each convolution operation is followed by a maximum pooling layer. The feature maps output by the convolution layer and the pooling layer are reshaped and input into the long short-term memory network layer, wherein the long short-term memory network layer contains 128 hidden units; Adding a self-attention layer after the LSTM layer enables the model to adaptively focus on the most relevant time steps and features; The output of the self-attention layer passes through a fully connected layer, and finally outputs the probability distribution of the fault type through a softmax classifier; The deep learning model is trained by combining large-scale simulated data with actual fault data. The cross entropy loss function and Adam optimizer are used. The learning rate is initially set to 0.001, and a learning rate decay strategy is adopted. At the same time, the dropout technology is applied to prevent overfitting, and the dropout rate is set to 0.

5. The trained deep learning model is used to analyze the input real-time data, complete fault identification and location within milliseconds, and output fault type and fault location information. The fault type identification accuracy rate is not less than 98.5%, and the average error of fault location is less than 50 meters.

3. The method according to claim 1, characterized in that According to the identified fault type and fault point location, combined with the preset distribution network topology information, the fault impact scope and severity are evaluated, and a preliminary fault assessment report is generated, including: Constructing a digital twin model of a distribution network, wherein the digital twin model of a distribution network is based on graph database technology, and represents transformers, switches, lines, loads, and power generation equipment in the distribution network as nodes, and represents the connection relationship between the nodes as directed edges, wherein the directed edges represent the direction of power flow; Setting attribute information for the node, the attribute information including device identification, rated capacity, current load, voltage level, location coordinates, device type, rated current, device status, line length, impedance parameter, load type and importance level; Through the data interface with the distribution network management system, the distribution network topology and equipment information are imported into the graph database, and a real-time data update mechanism is established with the monitoring and data acquisition system to dynamically update the attribute information of the nodes; Receiving a fault node identifier and fault type information, and locating a fault starting point in the distribution network digital twin model based on the fault node identifier; Using a depth-first search algorithm, starting from the fault starting point and following the power flow direction represented by the directed edge, the area affected by the fault is gradually explored; During the search process of the depth-first search algorithm, based on the attribute information of the currently accessed node and the adjacent nodes and in combination with the fault type, it is determined whether the fault will propagate to the adjacent nodes; The determination includes: when the fault type is a short circuit fault, checking whether there is an untriggered protection device between the current node and the adjacent node; when the fault type is an open circuit fault, determining whether the adjacent node relies on the current node for power supply; Add nodes that are determined to be affected by the fault to the affected area list, and record the parent node of each affected node for reconstructing the fault propagation path; Based on the affected area list and the fault propagation path, generate a fault impact assessment result including an affected node list and a fault propagation sequence; By using graphic visualization technology, the fault impact assessment result is converted into a visualization graphic to intuitively display the fault impact range and propagation path; Verify the method based on historical fault data, compare the predicted impact range with the actual situation, and optimize the propagation judgment logic and parameters in the depth-first search algorithm based on the verification results; An adaptive learning mechanism is implemented to automatically adjust and optimize the judgment parameters and logic of the method according to the feedback information of each actual fault.

4. The method according to claim 1, characterized in that: Based on the generated preliminary fault assessment report, a multi-objective optimization algorithm is initiated, which comprehensively considers the factors of power supply reliability, economy and user importance, including: receiving a preliminary fault assessment report, the preliminary fault assessment report including an affected area, a fault type, a fault location, and an estimated degree of impact; Based on the preliminary fault assessment report, a multi-objective optimization algorithm is started, wherein the multi-objective optimization algorithm adopts an improved non-dominated sorting genetic algorithm II to balance the three objectives of power supply reliability, economy and user importance; For the power supply reliability target, the system average interruption frequency index and system average interruption duration index are calculated and minimized. For the economic target, the equipment replacement cost, labor cost and power loss cost are considered and the total cost is minimized. For the user importance target, a five-level importance classification is adopted and different weights are assigned to each level of users. Integer coding is used for chromosome coding, and each gene represents the state of a controllable switch; the population size is set to 100, the evolutionary generations are set to 500, the single-point crossover operation is used, the crossover probability is 0.8, and the uniform mutation operation is used, with a mutation probability of 0.1; Introduce an adaptive crossover and mutation probability adjustment mechanism, calculate the average fitness value of the population every 10 generations, and when the average fitness value calculated three times changes less than 1%, increase the crossover probability by 0.05 and the mutation probability by 0.01 until the preset upper limit is reached; Network topology constraints, voltage constraints and current constraints are set. The voltage constraint requires that the voltage deviation of each node does not exceed ±7% of the rated voltage, and the current constraint requires that the line current does not exceed 80% of the rated capacity. The penalty function method is used to punish individuals that violate the constraints. In each generation of evolution, the fast non-dominated sorting method is used to stratify individuals, and the crowding distance calculation is used to maintain population diversity. A set of non-dominated solutions is output as the Pareto optimal solution set for decision makers to choose the most suitable fault recovery plan.

5. The method according to claim 1, characterized in that Using reinforcement learning technology, simulate multiple possible fault isolation and grid reconstruction schemes, and conduct simulation evaluation on each scheme; Through iterative optimization, the optimal fault isolation and grid reconstruction solutions are selected, including: Construct a distribution network environment model, which includes network topology, load distribution, generator location and switch status information; define the environment state, which is represented by a high-dimensional vector, including the voltage, current, power factor parameters of each node, and the opening and closing state of the switch; set an action space, which is defined as the combination of operations of all controllable switches, and each action corresponds to a specific fault isolation and network reconstruction scheme; design a reward function, which takes into account the degree of power supply recovery, network loss, voltage quality and number of operations; Constructing a deep Q network, the deep Q network includes a multi-layer convolutional neural network and a fully connected layer, the input layer receives the environment state, and the output layer corresponds to the Q value of each possible action; An experience replay mechanism is set up, and a fixed-capacity experience pool is used to store transfer samples; a target network update mechanism is set up, and the target network parameters are updated every fixed training step; an ε-greedy strategy is used for action selection, and the initial ε value of the ε-greedy strategy is set to 1, and a fixed ratio is decayed each round until a minimum value is reached; and a training process is executed, including the following sub-steps: Reset the distribution network environment model and obtain the initial state; select actions according to the ε-greedy strategy; execute the selected action to obtain the reward and the next state; store the transfer samples in the experience pool; randomly extract batch samples from the experience pool for training; use the mean square error loss function and Adam optimizer to update the deep Q network parameters; regularly update the target network; The trained deep Q network is tested and evaluated, and the test evaluation includes single fault, multiple fault and cascading fault scenarios; based on the test evaluation results, the optimal fault isolation and power grid reconstruction scheme is selected.

6. The method according to claim 1, characterized in that The formulated intelligent control instruction sequence is sent to the relevant intelligent terminal device through a secure encrypted channel; after the intelligent terminal device receives the instruction, it gradually executes the control operation according to the preset security protocol; During the execution process, the status information of each node is collected in real time and fed back to the central control system. The central control system uses edge computing technology to quickly process the feedback information locally and evaluate the execution effect of the solution in real time. If an abnormal situation or poor execution effect is detected, the emergency response mechanism is immediately triggered, and the control strategy is dynamically adjusted, including: Performing test evaluation on the trained deep Q network, wherein the test evaluation includes single fault, multiple fault and cascading fault scenarios; Select the best fault isolation and grid reconstruction solution based on the test evaluation results; The selected optimal solution is converted into an intelligent control instruction sequence and sent to the relevant intelligent terminal device through a secure encryption channel based on elliptic curve cryptography; After receiving the instruction, the intelligent terminal device gradually executes the control operation according to the preset safety protocol, which includes the environment check before the operation, the operation interval time control and the operation sequence enforcement; During the execution process, the status information of each node is collected in real time through a high-precision sensor network, and this information is fed back to the central control system through a multi-level data aggregation structure; The central control system uses a layered edge computing architecture to quickly process feedback information locally and evaluate the execution effect of the solution in real time; If an abnormal situation is detected or the execution effect is poor, the emergency response mechanism will be triggered immediately, and the control strategy will be dynamically adjusted using a multi-objective optimization algorithm.

7. An intelligent control system for fault location and rapid reliability recovery of distribution network, used to implement the method according to any one of claims 1 to 6, characterized in that: include: The first unit is used to collect real-time data of each node of the distribution network, including voltage, current and power data, and input the real-time data into a pre-trained deep learning model; The deep learning model is based on a hybrid architecture of a convolutional neural network and a long short-term memory network. By analyzing the time series characteristics and spatial distribution characteristics of the real-time data, the fault type can be quickly identified and the fault point can be accurately located. According to the identified fault type and fault point location, combined with the preset distribution network topology information, the fault impact scope and severity are evaluated, and a preliminary fault assessment report is generated. The second unit is used to start a multi-objective optimization algorithm based on the generated preliminary fault assessment report, which comprehensively considers factors such as power supply reliability, economy and user importance; Using reinforcement learning technology, simulate multiple possible fault isolation and grid reconstruction schemes, and conduct simulation evaluation on each scheme; select the optimal fault isolation and grid reconstruction scheme through iterative optimization; formulate detailed intelligent control instruction sequences based on the selected optimal scheme, including specific measures such as circuit breaker operation, load transfer and distributed power generation scheduling; The third unit is used to send the formulated intelligent control instruction sequence to the relevant intelligent terminal device through a secure encrypted channel; after the intelligent terminal device receives the instruction, it gradually executes the control operation according to the preset security protocol; During the execution process, the status information of each node is collected in real time and fed back to the central control system. The central control system uses edge computing technology to quickly process feedback information locally and evaluate the execution effect of the plan in real time. If an abnormal situation is detected or the execution effect is poor, the emergency response mechanism is immediately triggered and the control strategy is dynamically adjusted. After completing fault isolation and grid reconstruction, the system recovery status is continuously monitored, and detailed event reports and optimization suggestions are generated to provide data support for future fault prevention and system improvement.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images