Power distribution network line fault positioning method, device and equipment

Abstract

The invention discloses a power distribution network line fault positioning method, device and equipment, and relates to the technical field of power distribution network line fault positioning, and the method comprises the steps: obtaining operation data and lightning ground lightning data; feature extraction is carried out on the operation data, spatial-temporal feature mapping is carried out on the lightning and ground lightning data, and multi-source feature parameters are generated; probabilistic reasoning is carried out on the input multi-source characteristic parameters through section agents deployed in a distributed mode and Bayesian network submodules of the section agents, and real-time fault probability values of all independent sections of the power distribution network are output; and taking the real-time fault probability value of each independent section as an input, dynamically coordinating an output result of each agent through a Nash equilibrium strategy, and generating a global consistency fault positioning decision. The method is used for solving the problem that a traditional line fault positioning method is difficult to effectively fuse heterogeneous data and is difficult to coordinate local judgment conflicts.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network line fault location, and more specifically, to a distribution network line fault location method, device and equipment. Background Art

[0002] The distribution network is a critical link in the power system's delivery of power to end users. Its operational reliability directly impacts power quality and the smooth functioning of socioeconomic activities. With the advancement of new power system construction, the increasingly complex distribution network structure, coupled with the large-scale integration of distributed power sources, the diversification of load types, and the frequent occurrence of extreme weather events, presents new technical challenges for line fault location. Traditional fault location methods, which primarily rely on electrical measurement data and protection device operation information, have significant limitations in practical applications: the difficulty in effectively integrating multi-source heterogeneous data and the lack of coordination mechanisms for conflicting distributed decisions.

[0003] For example, the invention patent application with publication number CN118797527A discloses a method for diagnosing DC distribution network line faults, which relates to the technical field of DC distribution networks and includes the following steps: before locating the line fault, obtaining slime mold model abnormal data, and when diagnosing the line fault type, obtaining network information fusion data; comparing the slime mold model abnormal data with the network information fusion data to obtain a first comparison result; combining the first comparison result with the constraint conditions of each diagnostic effect category to analyze and obtain the actual diagnostic effect category; optimizing the diagnostic scheme according to the actual diagnostic effect category; establishing an optimization effect evaluation model through machine learning based on the optimized slime mold model abnormal data and network information fusion data to generate an optimization effect evaluation coefficient; analyzing and obtaining the power grid line fault diagnosis effect; making a prediction for the effect of the DC distribution network fault diagnosis method, and improving the accuracy and stability of DC distribution network line fault diagnosis.

[0004] For example, the invention patent announcement with publication number: CN120354254A discloses a method and system for locating and identifying faults in an active distribution network based on a spatiotemporal graph network, which relates to the technical field of fault locating and identifying faults in an active distribution network. The method and system are as follows: acquiring data of each node in the active distribution network after a fault occurs, constructing structured graph data, marking the fault nodes and types, and constructing a sample data set; using the sample data set as source domain data, pre-training a fault diagnosis model based on a spatiotemporal graph network; in each iteration process, adopting a multi-scale adversarial perturbation addition method based on gradient optimization to generate adversarial sample data, screening data and adding it to the data set through a dynamic balance mechanism of data categories and a confidence evaluation mechanism, and using the updated data set to train the model; introducing target domain data, and adopting a transfer learning strategy based on dynamic kernelized maximum mean difference to fine-tune the pre-trained model; inputting the actually acquired node data into the model to achieve accurate positioning of the fault node and accurate identification of the fault type.

[0005] Existing solutions primarily rely on network information fusion data, but the fusion dimension of multi-source data is insufficient, making it difficult to cope with faults caused by natural disasters such as lightning strikes. Furthermore, they lack collaborative decision-making design for intelligent entities in distribution network segments, and the use of centralized diagnostic models cannot adapt to large-scale distributed power access scenarios. Although some solutions use spatiotemporal graph networks to process node data, their heterogeneous data collaboration capabilities are weak, and they lack a weighted fusion design for operational steady-state characteristics and lightning spatiotemporal characteristics. Furthermore, they rely on centralized transfer learning models for fault location, and are unable to resolve local judgment conflicts in ring networks or multi-power supply scenarios. In summary, existing technical solutions, either due to insufficient fusion of multi-source heterogeneous data or a lack of collaborative decision-making design for distribution network segments, have failed to effectively resolve the problems of integrating heterogeneous data and local judgment conflicts.

[0006] In view of the above problems, the present invention proposes a solution. Summary of the Invention

[0007] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a method, device and equipment for locating line faults in a distribution network. By extracting multi-source feature parameters, establishing a multi-agent Bayesian subnetwork and fusing the Bayesian subnetwork through a Nash equilibrium strategy, the problems that traditional line fault location methods are difficult to effectively integrate heterogeneous data and difficult to coordinate local judgment conflicts are solved.

[0008] To achieve the above-mentioned objectives, the present invention provides the following technical solutions: a distribution network line fault location method, characterized in that it includes the following steps: obtaining operating data and lightning-to-ground flash data; performing feature extraction on the operating data, and performing spatiotemporal feature mapping on the lightning-to-ground flash data to generate multi-source feature parameters; performing probabilistic reasoning on the input multi-source feature parameters through distributedly deployed segment intelligent agents and their Bayesian network submodules, and outputting real-time fault probability values ​​for each independent segment of the distribution network; using the real-time fault probability values ​​of each independent segment as input, dynamically coordinating the output results of each intelligent agent through a Nash equilibrium strategy to generate a globally consistent fault location decision.

[0009] In a preferred embodiment, the operating data includes the binary code value of the isolation section status, the action timing characteristic value and the feeder effective value, the ground lightning data includes the spatial correlation coefficient of the line section and the ground lightning fault time difference coupling degree, and the multi-source characteristic parameters include the operating steady-state characteristic parameters and the lightning spatiotemporal characteristic parameters.

[0010] In a preferred embodiment, the feature extraction of the operating data and the spatiotemporal feature mapping of the lightning-to-ground flash data are performed to generate multi-source feature parameters, specifically: the operating data is standardized and the operating steady-state feature parameters are generated by splicing feature vectors; the distribution network topology is constructed, and the lightning-to-ground flash data and the topology are spatially integrated to generate lightning spatiotemporal feature parameters; the operating steady-state feature parameters and the lightning spatiotemporal feature parameters are weightedly integrated to output the multi-source feature parameters.

[0011] In a preferred embodiment, the distributed deployed segment agent and its Bayesian network submodule perform probabilistic reasoning on the input multi-source characteristic parameters and output the real-time fault probability value of each independent segment of the distribution network, specifically: according to the topology of the distribution network, the line is divided into several physical segments, each physical segment is associated with an agent, and a Bayesian network submodule is constructed for each agent; the multi-source characteristic parameters are input into the corresponding agent as observation data of the Bayesian network submodule; the conditional probability table of the Bayesian network is set based on historical data and expert experience; according to the observation data and the conditional probability table, the posterior probability of the fault of each segment is calculated by the Bayesian probability propagation algorithm; based on the posterior probability, the real-time fault probability value of each segment is determined by the posterior probability estimation algorithm.

[0012] In a preferred embodiment, the construction of the distribution network topology structure is specifically as follows: obtaining line model data and grid spatial coordinate information; extracting node distribution and device connection relationships based on the line model data and grid spatial coordinate information; using a topology verification algorithm to perform connectivity verification on the node distribution and device connection relationships to generate verified topology information; integrating the verified topology information to construct a distribution network topology structure including physical connection relationships.

[0013] In a preferred embodiment, the real-time fault probability value of each independent segment is used as input, and the output results of each intelligent agent are dynamically coordinated through the Nash equilibrium strategy to generate a globally consistent fault location decision. Specifically, the Bayesian network submodule of each intelligent agent is defined as a game participant, and the strategy space parameter of each intelligent agent is the real-time fault probability value of the corresponding segment; a profit function is constructed for each intelligent agent, and its function value is a weighted combination of the fault location accuracy index and the adjacent segment fault correlation index; the strategy space parameters of each intelligent agent are dynamically adjusted through an iterative optimization algorithm until the profit function of all intelligent agents reaches a preset threshold and the strategy space parameter combination conforms to the Nash equilibrium strategy; the strategy space parameters that satisfy the Nash equilibrium strategy are output as the globally consistent fault location decision.

[0014] In a preferred embodiment, the policy space parameters of each agent are dynamically adjusted through an iterative optimization algorithm, specifically: the policy space parameters of each agent are used as optimization variables, and the updated value of the policy space parameters of each agent in the current iteration round is calculated based on a preset benefit function expression; the updated value of the policy space parameters is used as the initial value of a new round of iteration, and the policy space parameter optimization calculation is repeated for all agents; the iteration is continued until the convergence conditions are met, and the convergence conditions include that the changes in the policy space parameters of all agents in two adjacent iterations are less than a preset convergence threshold, and the benefit function values ​​of all agents reach a preset benefit function threshold state.

[0015] In a preferred embodiment, the line is divided into several physical sections according to the distribution network topology, and each physical section is associated with an intelligent agent. Specifically, according to the distribution network topology, the line is divided into several physical sections with the protection device as the boundary; an intelligent agent is created for each physical section, and the intelligent agent is constructed based on the collected connection relationship between adjacent sections, the type parameters of the section power equipment, the initial fault probability distribution generated by historical fault statistics, and the fault propagation weight coefficient generated based on the topological distance.

[0016] A distribution network line fault location device is characterized by including a data acquisition module, a feature extraction and mapping module, a probabilistic reasoning module and a global decision module, and there are connections between the modules: the data acquisition module acquires operating data and lightning-to-ground flash data; the feature extraction and mapping module extracts features from the operating data and performs spatiotemporal feature mapping on the lightning-to-ground flash data to generate multi-source feature parameters; the probabilistic reasoning module performs probabilistic reasoning on the input multi-source feature parameters through distributed deployed segment intelligent agents and their Bayesian network submodules, and outputs real-time fault probability values ​​for each independent segment of the distribution network; the global decision module takes the real-time fault probability values ​​of each independent segment as input, dynamically coordinates the output results of each intelligent agent through a Nash equilibrium strategy, and generates a globally consistent fault location decision.

[0017] A distribution network line fault locating device, characterized by comprising at least one processor; and an input / output interface communicatively connected to the at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to perform the distribution network line fault locating method.

[0018] The technical effects and advantages of the distribution network line fault location method, device and equipment of the present invention are as follows: The distribution network line fault location method, device, and equipment proposed in this invention successfully break down the barriers between distribution network operating data and lightning monitoring data through multi-source data fusion technology. Distribution network operating data includes key information such as isolation section status and switch action information, while lightning monitoring data involves important parameters such as ground lightning location and lightning strike intensity. These heterogeneous data are originally difficult to work together. However, this invention extracts steady-state features from operating data and performs spatiotemporal feature mapping on ground lightning data, effectively integrating the two into unified feature parameters. This provides a comprehensive and reliable data foundation for subsequent fault location, solving the pain point of traditional line fault location methods that make heterogeneous data difficult to work together.

[0019] For local fault diagnosis, the multi-agent Bayesian network submodule enables precise calculation of fault probabilities in each section. The distribution network is divided into multiple sections, each corresponding to an agent and equipped with a Bayesian network submodule. Combining historical data with a conditional probability table established by expert experience, each agent can accurately calculate the fault probability of its section based on input characteristic parameters, significantly improving the reliability of local fault diagnosis.

[0020] In terms of global decision-making, the Nash equilibrium strategy is used to coordinate the various agents, effectively overcoming the challenge of conflicting local judgments. Each agent's Bayesian network submodule, acting as a participant in the game, ultimately achieves globally consistent fault location by defining a reasonable payoff function and iteratively calculating strategy space parameters. This process ensures that the judgments of each segment are coordinated and unified, avoiding confusion caused by local discrepancies.

[0021] It effectively solves the problem that traditional line fault location methods are difficult to effectively integrate heterogeneous data and difficult to coordinate local judgment conflicts. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic flow chart of a method for locating a distribution network line fault provided by an embodiment of the present invention.

[0023] Figure 2 A schematic diagram of the structure of a distribution network line fault locating device provided by an embodiment of the present invention.

[0024] Figure 3 A schematic diagram of the structure of a distribution network line fault locating device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0026] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, the elements defined by the phrase "comprising..." do not exclude the presence of other identical elements in the process, method, article, or device comprising the elements.

[0027] Example 1, Figure 1 The present invention provides a method for locating a fault in a distribution network, comprising the following steps: S1, obtain operation data and lightning ground flash data; S2, extract features from the operating data and perform spatiotemporal feature mapping on the lightning-to-ground flash data to generate multi-source feature parameters; S3, through the distributed deployment of the segment agent and its Bayesian network submodule, performs probabilistic reasoning on the input multi-source feature parameters and outputs the real-time fault probability value of each independent segment of the distribution network; S4, taking the real-time fault probability value of each independent segment as input, dynamically coordinates the output results of each intelligent agent through the Nash equilibrium strategy to generate a globally consistent fault location decision.

[0028] The distribution network line fault location method, device, and equipment proposed in this invention successfully break down the barriers between distribution network operating data and lightning monitoring data through multi-source data fusion technology. Distribution network operating data includes key information such as isolation section status and switch action information, while lightning monitoring data involves important parameters such as ground lightning location and lightning strike intensity. These heterogeneous data are originally difficult to work together. However, this invention extracts steady-state features from operating data and performs spatiotemporal feature mapping on ground lightning data, effectively integrating the two into unified feature parameters. This provides a comprehensive and reliable data foundation for subsequent fault location, solving the pain point of traditional line fault location methods that make heterogeneous data difficult to work together.

[0029] For local fault diagnosis, the multi-agent Bayesian network submodule enables precise calculation of fault probabilities in each section. The distribution network is divided into multiple sections, each corresponding to an agent and equipped with a Bayesian network submodule. Combining historical data with a conditional probability table established by expert experience, each agent can accurately calculate the fault probability of its section based on input characteristic parameters, significantly improving the reliability of local fault diagnosis.

[0030] In terms of global decision-making, the Nash equilibrium strategy is used to coordinate the various agents, effectively overcoming the challenge of conflicting local judgments. Each agent's Bayesian network submodule, acting as a participant in the game, ultimately achieves globally consistent fault location by defining a reasonable payoff function and iteratively calculating strategy space parameters. This process ensures that the judgments of each segment are coordinated and unified, avoiding confusion caused by local discrepancies.

[0031] It effectively solves the problem that traditional line fault location methods are difficult to effectively integrate heterogeneous data and difficult to coordinate local judgment conflicts.

[0032] S1, obtain operating data and lightning ground flash data.

[0033] In this embodiment, the operating data includes the binary code value of the isolation section status, the action timing characteristic value and the feeder effective value; the ground lightning data includes the spatial correlation coefficient of the line section and the ground lightning fault time difference coupling degree; the multi-source characteristic parameters include the operating steady-state characteristic parameters and the lightning spatiotemporal characteristic parameters.

[0034] S2, extract features from the operating data, and perform spatiotemporal feature mapping on the lightning ground flash data to generate multi-source feature parameters.

[0035] In this embodiment, the feature extraction of the operating data and the spatiotemporal feature mapping of the lightning ground flash data are performed to generate multi-source feature parameters, specifically: The operation data is standardized and the operation steady-state characteristic parameters are generated by splicing the characteristic vectors; Construct a distribution network topology, and perform spatial fusion of ground-to-ground lightning data with the topology to generate lightning spatiotemporal characteristic parameters. The operational steady-state characteristic parameters and lightning spatiotemporal characteristic parameters are weightedly fused to output multi-source characteristic parameters.

[0036] Obtaining steady-state characteristic parameters of the operating data based on the isolation section state, switch action information, and feeder short-circuit current, wherein the isolation section state is binary-coded to obtain a binary code value, the switch action information is time-series-coded to obtain an action timing characteristic value, and the feeder short-circuit current is effectively extracted to obtain a feeder effective value. The binary code value, the action timing characteristic value, and the feeder effective value are normalized and concatenated with a characteristic vector to obtain the steady-state characteristic parameters of the operating data; The specific steps for obtaining the distribution network topology and spatiotemporal lightning characteristic parameters are as follows: First, digital modeling of the distribution network topology is performed. Basic line data from the distribution network geographic information system is collected, including the specific coordinates of towers, line models, the locations of branch nodes, and the connectivity between nodes. Based on this data, a line topology matrix is ​​constructed, with matrix elements identifying the connectivity between different line segments. A line spatial distribution vector, consisting of the spatial coordinates of each line segment, is generated. Second, the spatial correlation coefficient is calculated. Data such as the location coordinates, occurrence time, and lightning current intensity of ground-to-ground lightning are obtained from the ground-to-ground lightning monitoring system. For each line segment, the vertical distance from the ground-to-ground lightning location to the segment is calculated. A spatial attenuation coefficient is set. Based on the vertical distance and attenuation coefficient, the spatial correlation coefficient is calculated using an exponential function. A larger value indicates a closer spatial correlation between the ground-to-ground lightning and the line segment. Third, the lightning strike intensity withstand value is calculated. The lightning withstand level parameter for each line segment is extracted, which takes into account factors such as tower grounding resistance and insulator withstand voltage. Calculate the ratio of lightning current intensity to the lightning withstand capability of the line segment. Then, use the Sigmoid function to map this ratio to a lightning strike intensity tolerance value. When the ratio is greater than or equal to 1, the tolerance value approaches 1, indicating that the lightning strike intensity may exceed the line's tolerance capacity. Step 4: Calculate the temporal coupling. Determine the specific time when the line fault occurs and calculate the time difference between the lightning strike and the line fault. Set a time window threshold, which is determined based on the line protection operation time. When the time difference is within the threshold, the temporal coupling decreases linearly with increasing time difference. When the time difference exceeds the threshold, the temporal coupling is zero. Step 5: Determine the spatiotemporal lightning characteristic parameters through weighted fusion. Use the analytic hierarchy process to determine the weight coefficients for the spatial correlation coefficient, lightning strike intensity tolerance value, and temporal coupling. These three parameters are weighted and summed according to the weights to obtain preliminary characteristic values ​​for each line segment. These preliminary characteristic values ​​for all line segments are normalized to form the lightning spatiotemporal characteristic vector.

[0037] The operational steady-state characteristic parameters and lightning spatiotemporal characteristic parameters are weightedly fused to obtain multi-source characteristic parameters.

[0038] The calculation formula for the steady-state characteristic parameters is:

[0039] in, Indicates the The steady-state characteristic parameters of each line section, , , is the steady-state weight coefficient, is the binary coded value of the isolated segment status, is the action timing characteristic value after normalization of the switch action information, For the The effective value of the feeder for each line section.

[0040] The calculation formula for lightning spatiotemporal characteristic parameters is:

[0041] in, Indicates the Lightning spatiotemporal characteristic parameters of each line section, , , is the lightning spatiotemporal weight coefficient, For the The spatial correlation coefficient of the line section, is the lightning intensity tolerance value in space and time. is the time difference coupling degree of ground-to-ground lightning fault.

[0042] The calculation formula for multi-source characteristic parameters is:

[0043] in, Indicates the A set of multi-source characteristic parameters for each line section, is the feature parameter fusion weight coefficient.

[0044] S3, through the distributed deployment of segment intelligent agents and their Bayesian network submodules, performs probabilistic reasoning on the input multi-source feature parameters and outputs the real-time fault probability value of each independent segment of the distribution network.

[0045] In this embodiment, the distributed deployed segment agent and its Bayesian network submodule perform probabilistic reasoning on the input multi-source feature parameters and output the real-time fault probability value of each independent segment of the distribution network, specifically: According to the topology of the distribution network, the line is divided into several physical sections. Each physical section is associated with an intelligent agent, and a Bayesian network submodule is constructed for each intelligent agent. Input the multi-source feature parameters into the corresponding agent as the observation data of the Bayesian network submodule; Set the conditional probability table of the Bayesian network based on historical data and expert experience; Based on the observed data and conditional probability table, the posterior probability of each section fault is calculated using the Bayesian probability propagation algorithm; Based on the posterior probability, the real-time fault probability value of each section is determined by a posterior probability estimation algorithm.

[0046] Based on the distribution network topology, the lines are divided into several physical segments, each of which serves as an intelligent agent. The distribution network topology includes key information such as line connections, node distribution, and equipment layout. Using this as a basis for segmentation ensures that each segment has relatively independent electrical characteristics and fault manifestations. Protection devices are used as clear boundaries during segmentation, as their placement is often related to factors such as line criticality and load distribution. This allows for more targeted fault detection and isolation within each physical segment. Once each physical segment is assigned the attributes of an intelligent agent, it possesses the capabilities of autonomous perception, analysis, and decision-making. To enable the intelligent agent to perform its work more accurately, detailed attribute information of each physical section is collected, including the connection relationship between adjacent sections (to clarify the relationship with other sections and facilitate information exchange and collaborative judgment), section equipment type parameters (such as line material, length, tower type, etc., which affect the failure probability and characteristics of the section), initial fault probability distribution (based on historical fault data statistics, serving as the initial reference for fault probability calculation) and fault propagation weight coefficient (reflecting the possibility of fault propagation from the current section to adjacent sections). Corresponding intelligent agents are created based on these attributes, so that each intelligent agent can accurately represent the characteristics of its section. The conditional probability table is key to the functioning of the Bayesian network submodule, reflecting the probabilistic dependencies between different variables. Statistical analysis based on historical data (including numerous past failure cases, corresponding characteristic parameters, and actual failure outcomes) reveals objective probabilistic patterns between variables. Combined with expert experience (with a deep understanding of distribution network failure mechanisms and the impact of various factors), statistical results are refined and improved to create a reasonable and accurate conditional probability table, ensuring the reliability and accuracy of the Bayesian network submodule's reasoning.

[0047] The formula for calculating input evidence is:

[0048] in For the The Bayesian network submodule of each line segment inputs evidence, A collection of input evidence for each Bayesian network submodule.

[0049] The formula for calculating the posterior probability of failure in each section is:

[0050] in, is the posterior probability of failure in each section, In the section fault state, is the prior probability of section failure (initial distribution), is the conditional probability of evidence appearing under a given fault condition (from the conditional probability table), is the marginal probability of the evidence.

[0051] The calculation formula for the probability of failure in each section is:

[0052] in, is the probability of failure in each section, is the posterior probability weight, is the number of devices in the segment, For the section The historical failure probability of the equipment, For the Type weight of class device, is the historical failure impact coefficient.

[0053] In this embodiment, the construction of the distribution network topology is specifically as follows: Obtain line model data and grid spatial coordinate information; Extract node distribution and equipment connection relationships based on line model data and grid spatial coordinate information; A topology verification algorithm is used to perform connectivity verification on the node distribution and device connection relationship to generate verified topology information; Integrate the verified topology information and construct the distribution network topology structure including the physical connection relationship.

[0054] First, basic data collection and integration are carried out. By connecting to the distribution network dispatching system, geographic information system (GIS), and equipment management platform, line model data and grid spatial coordinate information are simultaneously acquired. Line model data includes line physical parameters (such as conductor type, cross-sectional area, and impedance), equipment attributes (such as transformer capacity, switch type, and transformer ratio), and hierarchical relationships (such as the subordinate relationship between trunk lines and branches). Grid spatial coordinate information includes the longitude and latitude coordinates, altitude, and relative position relationships of each device (such as the coordinates of the connection points between towers and lines). Based on this data, spatial topology analysis algorithms are used to extract node distribution information of the distribution network (nodes include substation busbar nodes, switch station nodes, branch box nodes, and user access nodes). Furthermore, the connection relationships between devices are organized based on their physical connection characteristics (such as the connection between cable heads and switches, and the fixed relationship between conductors and towers), forming an initial node-connection relationship map. Secondly, a topology verification algorithm is used to verify the initial information. The core verification algorithm includes a connectivity check based on an adjacency matrix: using distribution network nodes as matrix elements, an adjacency matrix is ​​constructed (an element value of 1 in the matrix indicates that the two nodes are directly connected, and a value of 0 indicates that they are not connected). The matrix is ​​traversed through a depth-first search (DFS) or breadth-first search (BFS) to verify whether each node is connected, and to identify isolated nodes (such as branch boxes not connected to the main network) or false connections (such as lines that are recorded as connected in the database but are actually physically disconnected). In addition, device attribute consistency verification (such as verifying whether the rated current of a switch matches the current carrying capacity of the connected line to avoid topological logic errors caused by parameter conflicts) and spatial location rationality verification (such as calculating the line length between two nodes using GIS coordinates and comparing it with the actual line length to verify the authenticity of the connection relationship). Through multi-dimensional verification, erroneous information is eliminated and missing associations are supplemented to form a verified node distribution and device connection relationship dataset. Finally, the verified information is integrated and processed. The spatial coordinates of the nodes, the physical parameters of the equipment, and the verified connection relationships are mapped to construct a topological model containing the three-dimensional information of "node location - equipment attributes - connection method." The model must clearly mark the physical location of each node (such as the specific address of the substation and the coordinates of the tower), the technical parameters of the equipment (such as the voltage level of the transformer and the operation method of the switch), and the physical form of the connection (such as the specific location of the overhead line or cable connection and the connection point). Through data standardization (such as a unified coordinate system and standardized equipment naming rules), structured distribution network topology data is generated, providing basic support for subsequent applications such as power flow calculation, fault location, and network reconstruction. S4, taking the real-time fault probability value of each independent segment as input, dynamically coordinates the output results of each intelligent agent through the Nash equilibrium strategy to generate a globally consistent fault location decision.

[0055] In this embodiment, the real-time fault probability value of each independent segment is input, and the output results of each intelligent agent are dynamically coordinated through the Nash equilibrium strategy to generate a globally consistent fault location decision, specifically: The Bayesian network submodule of each agent is defined as a game participant, and the strategy space parameter of each agent is the real-time failure probability value of the corresponding segment; A profit function is constructed for each intelligent agent, whose function value is a weighted combination of the fault location accuracy index and the adjacent section fault correlation index; Dynamically adjust the policy space parameters of each agent through an iterative optimization algorithm until the payoff function of all agents reaches the preset threshold and the policy space parameter combination conforms to the Nash equilibrium strategy; The policy space parameters that satisfy the Nash equilibrium strategy are output as the globally consistent fault location decision.

[0056] The preset threshold of the profit function is a benchmark value that comprehensively considers the fault location accuracy and the correlation between adjacent section faults.

[0057] The Bayesian network submodule of each agent is used as a game participant, and the strategy space parameter of each agent is the probability of failure of each segment output by each agent, that is, the agent The strategy is , , is the total number of segments, Representing an agent Judgment section The probability of failure.

[0058] The calculation formula for fault location accuracy is:

[0059] in, For fault location accuracy, For segment The actual fault status, Representing an agent Judgment section The probability of failure.

[0060] The calculation formula for the fault correlation degree of adjacent sections is:

[0061] in, is the fault correlation degree of adjacent sections, For intelligent agents The number of neighboring agents, is the set of neighboring agents, Representing an agent Judgment section The probability of failure.

[0062] The profit function expression is:

[0063] in, For intelligent agents The function return, is the weight parameter of the profit function.

[0064] The calculation formula for global consistency fault location results is:

[0065] in, is the global consistency fault location result, For intelligent agents The weight of Representing an agent Judgment section The probability of failure.

[0066] In this embodiment, the strategy space parameters of each agent are dynamically adjusted by the iterative optimization algorithm, specifically: Taking the policy space parameters of each agent as the optimization variable, the updated value of the policy space parameters of each agent in the current iteration round is calculated based on the preset profit function expression; Use the updated policy space parameter values ​​as the initial values ​​for the next iteration, and repeat the policy space parameter optimization calculation for all agents. The iteration continues until the convergence conditions are met. The convergence conditions include that the changes in the strategy space parameters of all agents in two adjacent iterations are less than the preset convergence threshold, and the profit function values ​​of all agents reach the preset profit function threshold state.

[0067] During each iteration, each agent is selected as the iterative agent for policy optimization. For the currently selected iterative agent, the system fixes the policy space parameters of all other agents in the current iteration. These fixed parameters serve as a reference for the iterative agents to optimize their own policies. At this point, the iterative agents treat their own policy space parameters as optimization variables that require adjustment. Their core goal is to improve their own reward function value toward a preset threshold by changing these parameters. Based on a payoff function expression that takes into account factors such as fault localization accuracy and correlation with neighboring agent policies, the iterative agent uses gradient ascent to calculate new policy space parameters. This method adjusts parameters along the direction of fastest payoff growth. By continuously calculating the partial derivatives of the payoff function with respect to each policy parameter, it determines the magnitude and direction of parameter adjustments, thereby quickly finding new policy space parameters that improve the payoff function. When an iterative agent completes the optimization and update of its policy space parameters, the system temporarily stores the updated parameters and selects the next agent as the new iterative agent. The process of fixing the parameters of other agents and optimizing its own parameters repeats. This iteration ends when all agents have completed the update of their policy space parameters. The system then uses the updated policy space parameters of all agents in this iteration as the initial values ​​for the next iteration, and a new iteration begins. This cycle repeats, with each agent optimizing its own parameters while the other agents' parameters are fixed. At the end of each iteration, the system calculates the change in the policy space parameters of all agents after the current iteration compared to the policy space parameters at the end of the previous iteration. When the change in the policy space parameters of all agents between two consecutive iterations is less than the preset convergence threshold, it indicates that the policy space parameters of each agent have stabilized, and further iterations will have little effect on improving the value of the reward function. At this point, the policy space of each agent has reached the preset threshold of the reward function, and the iteration process terminates.

[0068] In this embodiment, the line is divided into several physical segments according to the distribution network topology, and each physical segment is associated with an intelligent agent, specifically: According to the topology of the distribution network, the line is divided into several physical sections with the protection device as the boundary; An intelligent agent is created for each physical segment, which is constructed based on the collected connection relationships between adjacent segments, type parameters of segment power equipment, initial fault probability distribution generated by historical fault statistics, and fault propagation weight coefficient generated based on topological distance.

[0069] Based on the distribution network topology, the lines are divided into several physical sections with protection devices as boundaries. The distribution network topology clearly presents key information such as the line's direction, node distribution, and equipment connection relationships, and is an important basis for segmentation. The main function of protection devices such as circuit breakers and fuses is to quickly operate when a line fault occurs, cut off the fault current, and prevent the fault from spreading. Dividing physical sections based on protection devices as boundaries ensures that the protection devices in each section can function independently when a fault occurs, achieving rapid fault isolation. It also makes the scope of each section relatively clear and electrically independent, laying the foundation for subsequent fault location and analysis. For example, in a feeder, the line section between two adjacent circuit breakers can be regarded as an independent physical section. For each physical segment, the adjacent segment connectivity, segment equipment type parameters, initial fault probability distribution, and fault propagation weight coefficient are collected. A corresponding intelligent agent is created based on these attributes. Adjacent segment connectivity refers to the connection method and location between the current segment and other segments. Understanding this relationship helps the intelligent agent consider the potential impact of adjacent segments when making fault assessments, enabling information exchange and coordination. Segment equipment type parameters include the model, specifications, and material of equipment such as lines, transformers, and switches within the segment. Different types of equipment have different fault characteristics, and these parameters are crucial for the intelligent agent to analyze segment fault probabilities. The initial fault probability distribution is derived from historical fault data for that segment type. It reflects the probability of a segment failing under normal operation and provides a reference for the intelligent agent's initial assessment. The fault propagation weight coefficient reflects the likelihood of a fault propagating from the current segment to adjacent segments. A larger coefficient indicates a higher probability of fault propagation. After collecting this attribute information, it is set as the core attributes of the intelligent agent, thus creating an intelligent agent that corresponds to each physical segment. With these attributes, each intelligent agent can accurately represent the characteristics of its physical segment and independently carry out fault monitoring, analysis, and judgment. It also provides data support for the collaborative operation of intelligent agents.

[0070] Example 2, Figure 2 The present invention provides a device for locating a distribution network line fault method, which includes a data acquisition module, a feature extraction and mapping module, a probability reasoning module, and a global decision module. There are connections between the modules: Data acquisition module, to obtain operation data and lightning ground flash data; The feature extraction and mapping module extracts features from operational data and performs spatiotemporal feature mapping on lightning-to-ground flash data to generate multi-source feature parameters. The probabilistic reasoning module uses distributed segment agents and their Bayesian network submodules to perform probabilistic reasoning on the input multi-source feature parameters and output the real-time fault probability value of each independent segment of the distribution network; The global decision-making module takes the real-time fault probability value of each independent segment as input, dynamically coordinates the output results of each intelligent agent through the Nash equilibrium strategy, and generates a globally consistent fault location decision.

[0071] A distribution network line fault location device, characterized in that it includes an electronic device, characterized in that the electronic device includes: Example 3, Figure 3The present invention provides a device for a distribution network line fault locating method, comprising: at least one processor; and an input / output interface communicatively connected to the at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to perform the distribution network line fault locating method.

[0072] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0073] The above embodiments may be implemented in whole or in part through software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0074] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0075] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0076] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0077] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for locating a fault in a distribution network, characterized in that: The following steps are involved: Obtain operation data and lightning ground flash data; Extract features from operational data and perform spatiotemporal feature mapping on lightning-to-ground flash data to generate multi-source feature parameters; Through the distributed deployment of the segment agent and its Bayesian network submodule, probabilistic reasoning is performed on the input multi-source characteristic parameters to output the real-time fault probability value of each independent segment of the distribution network; Taking the real-time fault probability value of each independent segment as input, the output results of each intelligent agent are dynamically coordinated through the Nash equilibrium strategy to generate a globally consistent fault location decision.

2. The method for locating a distribution network line fault according to claim 1, wherein: The operating data includes the binary code value of the isolation section status, the action timing characteristic value and the feeder effective value; the ground lightning data includes the spatial correlation coefficient of the line section and the ground lightning fault time difference coupling degree; the multi-source characteristic parameters include the operating steady-state characteristic parameters and the lightning spatiotemporal characteristic parameters.

3. The method for locating a distribution network line fault according to claim 2, wherein: The feature extraction of the operating data and the spatiotemporal feature mapping of the lightning ground flash data are performed to generate multi-source feature parameters, specifically: The operation data is standardized and the operation steady-state characteristic parameters are generated by splicing the characteristic vectors; Construct a distribution network topology, and perform spatial fusion of ground-to-ground lightning data with the topology to generate lightning spatiotemporal characteristic parameters. The operational steady-state characteristic parameters and lightning spatiotemporal characteristic parameters are weightedly fused to output multi-source characteristic parameters.

4. The method for locating a fault in a distribution network according to claim 3, wherein: The distributed deployed segment agent and its Bayesian network submodule perform probabilistic reasoning on the input multi-source feature parameters and output the real-time fault probability value of each independent segment of the distribution network, specifically: According to the topology of the distribution network, the line is divided into several physical sections. Each physical section is associated with an intelligent agent, and a Bayesian network submodule is constructed for each intelligent agent. Input the multi-source feature parameters into the corresponding agent as the observation data of the Bayesian network submodule; Set the conditional probability table of the Bayesian network based on historical data and expert experience; Based on the observed data and conditional probability table, the posterior probability of each section fault is calculated using the Bayesian probability propagation algorithm; Based on the posterior probability, the real-time fault probability value of each section is determined by a posterior probability estimation algorithm.

5. The method for locating a distribution network line fault according to claim 4, characterized in that: The construction of the distribution network topology structure is specifically as follows: Obtain line model data and grid spatial coordinate information; Extract node distribution and equipment connection relationships based on line model data and grid spatial coordinate information; A topology verification algorithm is used to perform connectivity verification on the node distribution and device connection relationship to generate verified topology information; Integrate the verified topology information and construct the distribution network topology structure including the physical connection relationship.

6. The method for locating a fault in a distribution network according to claim 5, wherein: The real-time fault probability value of each independent segment is used as input, and the output results of each intelligent agent are dynamically coordinated through the Nash equilibrium strategy to generate a globally consistent fault location decision. Specifically: The Bayesian network submodule of each agent is defined as a game participant, and the strategy space parameter of each agent is the real-time failure probability value of the corresponding segment; A profit function is constructed for each intelligent agent, whose function value is a weighted combination of the fault location accuracy index and the adjacent section fault correlation index; Dynamically adjust the policy space parameters of each agent through an iterative optimization algorithm until the payoff function of all agents reaches the preset threshold and the policy space parameter combination conforms to the Nash equilibrium strategy; The policy space parameters that satisfy the Nash equilibrium strategy are output as the globally consistent fault location decision.

7. The method for locating a fault in a distribution network according to claim 4, wherein: The dynamic adjustment of the strategy space parameters of each agent through the iterative optimization algorithm is specifically as follows: Taking the policy space parameters of each agent as the optimization variable, the updated value of the policy space parameters of each agent in the current iteration round is calculated based on the preset profit function expression; Use the updated policy space parameter values ​​as the initial values ​​for the next iteration, and repeat the policy space parameter optimization calculation for all agents. The iteration continues until the convergence conditions are met. The convergence conditions include that the changes in the strategy space parameters of all agents in two adjacent iterations are less than the preset convergence threshold, and the profit function values ​​of all agents reach the preset profit function threshold state.

8. The method for locating a fault in a distribution network according to claim 4, wherein: The line is divided into several physical sections according to the distribution network topology, and each physical section is associated with an intelligent agent, specifically: According to the topology of the distribution network, the line is divided into several physical sections with the protection device as the boundary; An intelligent agent is created for each physical segment, which is constructed based on the collected connection relationships between adjacent segments, type parameters of segment power equipment, initial fault probability distribution generated by historical fault statistics, and fault propagation weight coefficient generated based on topological distance.

9. A device using the distribution network line fault location method according to any one of claims 1 to 8, characterized in that: It includes data acquisition module, feature extraction and mapping module, probability reasoning module and global decision module. There are connections between modules: Data acquisition module, to obtain operation data and lightning ground flash data; The feature extraction and mapping module extracts features from operational data and performs spatiotemporal feature mapping on lightning-to-ground flash data to generate multi-source feature parameters. The probabilistic reasoning module uses distributed segment agents and their Bayesian network submodules to perform probabilistic reasoning on the input multi-source feature parameters and output the real-time fault probability value of each independent segment of the distribution network; The global decision-making module takes the real-time fault probability value of each independent segment as input, dynamically coordinates the output results of each intelligent agent through the Nash equilibrium strategy, and generates a globally consistent fault location decision.

10. An electronic device, characterized in that: The electronic device comprises: At least one processor; and an input / output interface communicatively connected to the at least one processor; a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the distribution network line fault location method according to any one of claims 1 to 8.

Images