Power distribution network fault propagation path tracing method

Abstract

The invention provides a power distribution network fault propagation path tracing method. According to the method, the polymorphic measurement information and the system operation log in the power distribution network are collected, the polymorphic measurement information is preprocessed to generate the node feature data, the time sequence feature set is extracted in combination with the system operation log, and a high-quality data basis is provided for subsequent analysis. By constructing the information difference chart, the information interconnection relationship and change condition between the target components before and after the fault can be intuitively reflected, so that the dynamic characteristics of fault propagation are captured. Furthermore, sorting learning is performed on the time sequence feature set to generate sorting scores, and the fault propagation path is determined in combination with the information difference chart, so that a complete closed loop from data acquisition to path traceability is realized. According to the method, the accuracy of fault positioning is improved, the efficiency and reliability of fault tracing are remarkably improved through multi-dimensional data fusion and intelligent analysis, and powerful support is provided for stable operation of the power distribution network.

Description

Technical Field

[0001] The present application relates to the field of electrical fault detection, and in particular to a method for tracing the propagation path of a distribution network fault. Background Art

[0002] In recent years, as the scale of power dispatching and control systems has continued to expand, their internal information models have become increasingly complex and massive. When an anomaly or failure occurs in the system, a large number of alarm responses are triggered, placing significant pressure and time on operations and maintenance personnel. Traditional troubleshooting methods typically rely on analyzing the status data of each component one by one. However, in large-scale, low-latency, and fast-response power dispatching and control systems, this approach is no longer sufficient to meet the requirements for real-time and efficient determination of fault propagation paths. Power dispatching and control systems, characterized by low latency and rapid response, have numerous internal components and a complex network structure. Once a failure occurs, it often triggers a chain reaction, generating a series of redundant alarm messages.

[0003] Therefore, rapidly locating primary and secondary faults within massive amounts of alarm data has become a key technical challenge for ensuring power grid security and improving operation and maintenance efficiency. Because traditional time-series-based analysis methods struggle to directly trace specific fault propagation paths, a new approach that can effectively mine potential correlations within the system is urgently needed. While the massive amount of log information generated during daily operations may be trivial, it contains insights into the interactions and information transfer between system components, providing a crucial reference for inferring fault propagation paths. In recent years, learning-to-rank methods based on recommender systems have achieved significant results in information retrieval and personalized recommendation. By deeply mining log data, additional correlations between components can be established and ranked based on their degree of correlation, indirectly inferring fault propagation paths. Currently, existing technologies primarily focus on utilizing discretization, information entropy, and information difference graph models to diagnose and trace internal system faults. However, for large-scale power dispatching and control systems, single static information models struggle to capture the dynamic propagation characteristics of faults within complex systems.

[0004] In summary, a method for tracing the fault propagation path of distribution networks is needed to improve the efficiency and effectiveness of fault tracing. Summary of the Invention

[0005] The purpose of this application is to solve at least one of the above-mentioned technical deficiencies, especially the technical deficiencies of insufficient efficiency and effectiveness of fault tracing in the prior art.

[0006] In a first aspect, the present application provides a method for tracing the source of a distribution network fault propagation path, the method comprising:

[0007] Collect multi-state measurement information and system operation logs in the distribution network;

[0008] The polymorphic measurement information includes a plurality of electrical physical quantities during the operation of the distribution network, and the system operation log is used to indicate the operating status of the distribution network;

[0009] Preprocessing the polymorphic measurement information to generate node feature data;

[0010] constructing an information difference graph based on the node feature data;

[0011] The information difference graph is used to indicate the information interconnection relationship and information change between target components before and after the fault in the distribution network;

[0012] Obtaining a time series feature set according to the system operation log;

[0013] Performing ranking learning on the time series feature set to generate a ranking score;

[0014] A fault propagation path is determined according to the ranking score and the information difference graph.

[0015] As an optional implementation, the polymorphic measurement information includes voltage, current and power data;

[0016] The preprocessing of the polymorphic measurement information to generate node feature data includes:

[0017] Performing denoising and deduplication operations on the polymorphic measurement information to remove invalid or duplicate data, dividing each dimension in the processed polymorphic measurement information into a plurality of feature intervals, and calculating the centroid and sum of squared errors of each feature interval;

[0018] The boundary of the feature interval is dynamically adjusted based on the sum of squared errors to generate discretized node feature data.

[0019] As an optional implementation manner, constructing an information difference graph based on the node feature data includes:

[0020] Construct the first correlation matrix of each node before the fault and the second correlation matrix after the fault respectively;

[0021] The first correlation matrix and the second correlation matrix are used to reflect the information entropy and mutual information relationship between nodes;

[0022] Calculating a difference between the first correlation matrix and the second correlation matrix, and generating an information difference matrix based on the difference and the first correlation matrix;

[0023] Threshold processing and normalization processing are performed on the information difference matrix to construct an information difference map.

[0024] As an optional implementation manner, obtaining a time series feature set according to the system operation log includes:

[0025] Decompose the system operation log into time series to obtain the normal phase, abnormal phase, and recovery phase, and intercept the log fragments of the abnormal and recovery phases;

[0026] Extracting fault propagation features from the log segments, and generating a time series feature set based on the time sequence of the fault propagation features;

[0027] The fault propagation characteristics are used to indicate the fault development of the distribution network, including one or more characteristics of component word frequency, component operation frequency, reverse component frequency, log length and abnormal event density.

[0028] As an optional implementation, performing ranking learning on the time series feature set to generate a ranking score includes:

[0029] Perform preliminary sorting based on the fault propagation characteristics through an integrated learning algorithm to generate preliminary sorting results;

[0030] Inputting the preliminary ranking results into the target ranking learning model, and generating a ranking score by fitting the minimization of the association loss function;

[0031] Among them, the ranking score is used to indicate the fault correlation of the nodes corresponding to each target component in the distribution network in fault propagation, and the correlation loss function is used to indicate the difference between the component ranking score output by the target ranking learning model and the actual fault propagation correlation order.

[0032] As an optional implementation manner, determining the fault propagation path according to the ranking score and the information difference map includes:

[0033] Determining a plurality of recommended nodes according to the ranking scores;

[0034] Determining a target component according to an element relationship of each of the recommended nodes in the information difference graph;

[0035] According to the target component, an associated component corresponding to the target component is determined, and according to the target component and the corresponding associated component, a fault propagation path is determined.

[0036] As an optional implementation, the method further includes:

[0037] A visualization process is used to display the fault propagation path in the form of an interactive graphical interface, annotating the ranking scores and edge weight change rates of key nodes in the fault propagation path;

[0038] A fault reporting process for generating a multi-level fault report including the location of the original fault, its propagation path and corresponding timing, and the impact range of the secondary fault;

[0039] The fault alarm process is used to push alarm information to the target operation and maintenance terminal and trigger the corresponding emergency response process according to the fault severity level.

[0040] In a second aspect, the present application provides a distribution network fault propagation path tracing device, the device comprising:

[0041] Acquisition module, used to collect polymorphic measurement information and system operation logs in the distribution network;

[0042] The polymorphic measurement information includes a plurality of electrical physical quantities during the operation of the distribution network, and the system operation log is used to indicate the operating status of the distribution network;

[0043] A processing module, configured to pre-process the polymorphic measurement information to generate node feature data;

[0044] The processing module is further configured to construct an information difference graph based on the node feature data;

[0045] The information difference graph is used to indicate the information interconnection relationship and information change between target components before and after the fault in the distribution network;

[0046] The processing module is further configured to obtain a time series feature set based on the system operation log;

[0047] The processing module is further configured to perform ranking learning on the time series feature set to generate a ranking score;

[0048] The processing module is further configured to determine a fault propagation path based on the ranking score and the information difference graph.

[0049] In a third aspect, the present application provides a computer device comprising one or more processors and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the one or more processors, the steps of the method described in the first aspect are performed.

[0050] In a fourth aspect, the present application provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the method described in the first aspect.

[0051] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0052] Based on any of the above embodiments, the distribution network fault propagation path tracing method proposed in this application, through the collaborative collection of polymorphic measurement information and system operation logs, comprehensively covers the electrical physical quantities and operating status data during the operation of the distribution network, providing a rich data foundation for fault tracing. In the data preprocessing stage, the structure of the node feature data is optimized through denoising, deduplication and discretization processing, reducing the computational complexity. By constructing an information difference graph, the information interconnection relationship and changes between the target components before and after the fault are intuitively reflected, providing a key basis for the dynamic analysis of fault propagation. Furthermore, the ranking learning based on the time series feature set generates a ranking score, and the fault propagation path is accurately determined in combination with the information difference graph, realizing a complete closed loop from data collection to path tracing. This method significantly improves the accuracy and efficiency of fault location through multi-dimensional data fusion and intelligent analysis. In the subsequent process, the introduction of visual display, multi-level fault reporting and intelligent alarm mechanism not only enhances the timeliness of fault response, but also provides scientific support for operation and maintenance decision-making. Compared with traditional methods, this solution achieves comprehensive optimization in data quality, analysis depth and response efficiency, providing reliable technical support for the stable operation and rapid fault handling of the distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0054] Figure 1 A flow chart of a method for tracing the source of a distribution network fault propagation path provided in one embodiment of the present application;

[0055] Figure 2 A flow chart of a method for tracing the source of a distribution network fault propagation path provided in one embodiment of the present application;

[0056] Figure 3 A flow chart of a method for tracing the source of a distribution network fault propagation path provided in one embodiment of the present application;

[0057] Figure 4 This is a diagram of the internal structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0059] As power dispatching and control systems continue to expand, their internal information models are becoming increasingly complex and massive. When a system anomaly or failure occurs, it triggers a large number of alarm responses, placing significant pressure and time on operations and maintenance personnel to troubleshoot. Traditional troubleshooting methods typically rely on analyzing the status data of each component one by one. However, in the large-scale, low-latency, and fast-response power dispatching and control systems, this approach is no longer sufficient to efficiently and effectively determine the fault propagation path in real time.

[0060] Electric power dispatching and control systems are characterized by low latency and rapid response. However, they possess numerous internal components and a complex network structure. Failures often trigger chain reactions, generating a series of redundant alarms. Rapidly locating primary and secondary faults within this vast amount of alarm data has become a key technical challenge for ensuring power grid security and improving operational efficiency. Because traditional time-series-based analysis methods struggle to directly trace specific fault propagation paths, a new approach that can effectively mine potential correlations within the system is urgently needed. While the massive amount of log information generated during daily operations is trivial, it captures the interactions and information transfer between system components, providing a valuable insight for inferring fault propagation paths. Learning-to-rank methods based on recommender systems have achieved remarkable results in information retrieval and personalized recommendations. By deeply mining log data, additional correlations between components can be established and ranked according to their degree of correlation, indirectly inferring fault propagation paths.

[0061] Currently, existing technologies focus on using discretization processing, information entropy, and information difference graph models to diagnose and trace internal system faults. However, for large-scale power dispatching and control systems, a single static information model is difficult to capture the dynamic propagation characteristics of faults in complex systems. Therefore, by combining the ranking learning method with the information difference graph model, the dynamic process decomposition and feature extraction of the system log are used to generate a fault association recommendation list. Combined with the difference in information before and after the fault, it can more accurately reflect the fault propagation path. This method not only breaks through the limitations of traditional time series backtracking, but also improves the accuracy and efficiency of fault source location and fault propagation path reasoning, providing a new technical approach for the intelligent maintenance of large-scale power dispatching and control systems.

[0062] In summary, this application aims to utilize the relevance of log information, obtain fault correlation ranking through a recommendation system, and combine it with an information difference graph model to quickly locate the source of the fault and infer the fault propagation path, thereby significantly improving the efficiency of system fault investigation and the level of safe operation of the power grid. The technical concept of this application is that, through the collaborative collection of polymorphic measurement information and system operation logs, the electrical physical quantities and operating status data during the operation of the distribution network are fully covered, providing a rich data foundation for fault tracing. In the data preprocessing stage, the structure of the node feature data is optimized and the computational complexity is reduced through denoising, deduplication and discretization. By constructing an information difference graph, the information interconnection relationship and changes between the target components before and after the fault are intuitively reflected, providing a key basis for the dynamic analysis of fault propagation. Furthermore, ranking learning based on the time series feature set generates a ranking score, and the fault propagation path is accurately determined in combination with the information difference graph, realizing a complete closed loop from data collection to path tracing. This method significantly improves the accuracy and efficiency of fault location through multi-dimensional data fusion and intelligent analysis. In subsequent processes, the introduction of visual display, multi-level fault reporting, and intelligent alerting mechanisms not only enhances the timeliness of fault response but also provides scientific support for operation and maintenance decision-making. Compared with traditional methods, this solution achieves comprehensive optimization in data quality, analysis depth, and response efficiency, providing reliable technical support for the stable operation of the distribution network and rapid fault resolution.

[0063] The method provided in this application is described in detail below based on corresponding implementation methods in some actual application scenarios.

[0064] See also Figure 1 , Figure 1 A flow chart of a method for tracing the source of a distribution network fault propagation path provided by an embodiment of the present application is shown as follows: Figure 1 As shown, the method includes:

[0065] S101, collecting multi-state measurement information and system operation logs in the distribution network;

[0066] The polymorphic measurement information includes multiple electrical physical quantities during the operation of the distribution network. The system operation log is used to indicate the operating status of the distribution network. The system operation log records the operating status and event information of each component. These data serve as the basis for subsequent fault propagation path analysis.

[0067] S102, pre-processing the polymorphic measurement information to generate node feature data;

[0068] The preprocessing process includes data standard formatting, denoising, deduplication, and the use of discretization technology to construct feature intervals for key node data in polymorphic measurement data, and calculate the sum of squared errors between the centroid of each interval and the collected samples to provide data support for the quantification of node information. For details, please refer to the relevant implementation methods.

[0069] S103: constructing an information difference graph based on the node feature data;

[0070] The information difference graph is used to indicate the information interconnection relationship and information change between target components before and after the fault in the distribution network;

[0071] Specifically, the information entropy model can be used to measure the node information in a single directed information flow. The information entropy is used to indicate the change in the degree of information confusion before and after a failure occurs between nodes. In specific applications, the mathematical model of the information entropy can be obtained by calculating the matrix indicating the information between nodes, or by generating an associated matrix after calculating the information entropy. The basic formulas for calculating the information entropy of each node include:

[0072]

[0073] m represents the information of each node in the directed information flow L; g(m) represents the output probability of the directed information flow L taking the value m; M represents all possible output probabilities of m in the directed information flow L.

[0074] Furthermore, correlation matrices were constructed for both cases with and without fault information. Matrix difference operations were performed to obtain information difference matrices, which were then thresholded and normalized to generate a difference map reflecting the information changes of each key component under fault conditions. By comparing the data differences in the information correlation matrices before and after the fault, the changes in the information entropy of each node and the mutual information between nodes can be intuitively reflected. An information difference map was constructed, in which the nodes represent key components in the distribution network, and the edge weights represent the rate of change of information transfer between nodes.

[0075] S104: Obtain a time series feature set based on the system operation log;

[0076] Dynamic process decomposition and feature extraction of log data can transform the originally scattered log information into a feature set with time-series correlation, providing richer input information for the ranking learning model.

[0077] System operation logs are divided into three phases, normal, abnormal, and recovery, based on the temporal dynamics of the process. Log sequences from the abnormal and recovery phases are specifically captured. Feature selection techniques are used to extract characteristic information from the log data that indicates the degree of association between components, such as component word frequency, inverse document ratio, usage frequency, historical query click-through rate, historical query display rate, and query relevance level. The higher the query relevance level, the greater the fault relevance. By marking the query relevance of normal, abnormal, and recovery segments as 0, 2, and 1, respectively, a time-advanced sequence set queue is obtained. For details, please refer to the description of the relevant implementation method.

[0078] S105: performing ranking learning on the time series feature set to generate a ranking score;

[0079] An initial sorting algorithm is used to preliminarily sort the time series feature set obtained from the system operation logs, generating a preliminary list of fault association recommendations. This step primarily explores potential component associations in the logs, laying the foundation for further optimization of the subsequent sorting model.

[0080] During the ranking learning process, the initial ranking results are used as input for training using a LambdaMART-based ranking learning model. This model uses iterative training to minimize the relevance loss function using various component log features (such as component log word frequency, inverse component frequency, and log length). The model then outputs a final ranking score for each system component. This ranking score reflects the criticality and relevance of each component in the fault propagation process, providing a quantitative basis for reasoning about fault propagation paths.

[0081] The preliminary ranking results were further optimized through the LambdaMART model, making the correlation scores of each component more accurate, thereby providing a basis for accurate reasoning of the fault propagation path.

[0082] S106: Determine a fault propagation path according to the ranking score and the information difference graph.

[0083] By screening and marking components with higher ranking scores, backtracking and pruning methods are used to infer the specific path of fault propagation from the original fault to the outside, and finally the fault tracing results are determined, and accurate reasoning of the fault propagation path is achieved, thereby quickly locating the original fault and secondary faults.

[0084] Figure 2 A flow chart of a method for tracing the source of a distribution network fault propagation path provided by an embodiment of the present application, combined with Figure 1 The relevant steps in Figure 2 Shows a Figure 1The specific implementation of the corresponding steps in the actual application scenario can be used as an example to illustrate the functions of the following system. Correspondingly, the present application provides a system for implementing a method for tracing the propagation path of a distribution network fault, including a data acquisition unit for collecting polymorphic measurement data (such as voltage, current, power, etc.) in the distribution network and operation log information of the power dispatching and control system. A data preprocessing unit is used to format, denoise, deduplicate and discretize the collected data to generate measurement feature intervals of key nodes in the power grid and log data preprocessing results.

[0085] The node information measurement and information difference graph construction unit is used to measure node information using the information entropy model, construct a correlation matrix containing the states before and after the fault, and generate an information difference matrix through matrix difference; the difference matrix is ​​thresholded and normalized to generate an information difference graph model to reflect the correlation changes between nodes under the fault state.

[0086] The ranking learning unit is used to perform process decomposition and feature extraction on preprocessed log data to generate a set of sequences with a time-advancing relationship. It uses an initial ranking algorithm to obtain a preliminary fault association recommendation list and inputs it into the LambdaMART-based ranking learning model. Through iterative training, it outputs the final ranking score, which reflects the correlation between each component in fault propagation.

[0087] The fault propagation path inference unit is used to filter and mark components with high correlation based on the fault association recommendation list output by the ranking learning unit and the status information of nodes and edges in the information difference graph. It uses backtracking and pruning algorithms to infer the path of fault propagation from the source fault to the outside, determine the fault tracing results, and present them to the operation and maintenance personnel in graphical or text form.

[0088] The data acquisition unit can realize real-time data acquisition. The initial sorting algorithm in the sorting learning unit adopts a preset sorting algorithm, and then uses the LambdaMART algorithm to optimize the component correlation; the fault propagation path reasoning unit combines the comprehensive data of the recommendation system and the information difference graph to trace back and determine the fault path.

[0089] The distribution network fault propagation path tracing method provided by the present application comprehensively covers the electrical physical quantities and operating status data during the operation of the distribution network by collecting polymorphic measurement information and system operation logs in the distribution network. On this basis, the polymorphic measurement information is preprocessed to generate node feature data, and the time series feature set is extracted in combination with the system operation log to provide a high-quality data basis for subsequent analysis. By constructing an information difference graph, the information interconnection relationship and changes between the target components before and after the fault can be intuitively reflected, thereby capturing the dynamic characteristics of fault propagation. Furthermore, by sorting and learning the time series feature set to generate a ranking score, and combining the information difference graph to determine the fault propagation path, a complete closed loop from data collection to path tracing is achieved. This method not only improves the accuracy of fault location, but also significantly improves the efficiency and reliability of fault tracing through multi-dimensional data fusion and intelligent analysis, providing strong support for the stable operation of the distribution network.

[0090] As an optional implementation, the polymorphic measurement information includes voltage, current and power data;

[0091] The preprocessing of the polymorphic measurement information to generate node feature data includes:

[0092] Performing denoising and deduplication operations on the polymorphic measurement information to remove invalid or duplicate data, dividing each dimension in the processed polymorphic measurement information into a plurality of feature intervals, and calculating the centroid and sum of squared errors of each feature interval;

[0093] The boundary of the feature interval is dynamically adjusted based on the sum of squared errors to generate discretized node feature data.

[0094] This implementation effectively removes invalid or duplicate data and improves data purity by denoising and deduplicating the voltage, current, and power data in polymorphic measurement information. Furthermore, the processed data is divided into multiple feature intervals, and the centroid and sum of squared errors are calculated. Discrete node feature data is generated by dynamically adjusting the feature interval boundaries, making the data more consistent with subsequent modeling requirements. This process not only optimizes the data structure but also reduces computational complexity through discretization, enhancing the interpretability and practicality of the feature data and laying a solid foundation for constructing information difference maps.

[0095] As an optional implementation manner, constructing an information difference graph based on the node feature data includes:

[0096] Construct the first correlation matrix of each node before the fault and the second correlation matrix after the fault respectively;

[0097] The first correlation matrix and the second correlation matrix are used to reflect the information entropy and mutual information relationship between nodes;

[0098] Calculating a difference between the first correlation matrix and the second correlation matrix, and generating an information difference matrix based on the difference and the first correlation matrix;

[0099] Threshold processing and normalization processing are performed on the information difference matrix to construct an information difference map.

[0100] The specific process of constructing the fault information correlation matrix includes:

[0101] Construct the first correlation matrix D that does not contain fault information:

[0102]

[0103] Obtain n characteristic time series. Assume that the information entropy of each single directed information flow is {S1; S2...Sn}. Then the mutual information entropy of any two directed information flows can be expressed as {H1→2, H1→3, …, H1→n, H2→1, H2→3, …, H2→n, …, Hn→n}.

[0104] Constructing a second correlation matrix E containing fault information;

[0105] Matrix difference is used to calculate F = (E – D) / D, and threshold processing is performed on F to generate a new node information difference matrix F′, and then the information difference graph is generated.

[0106] This implementation comprehensively reflects the information entropy and mutual information relationships between nodes by constructing a first correlation matrix for each node before the fault and a second correlation matrix after the fault. An information difference matrix is ​​generated by calculating the difference between the first and second correlation matrices, and this matrix is ​​thresholded and normalized to ultimately construct an information difference map. This method clearly depicts the impact of a fault on the information interconnection relationships in the distribution network, intuitively demonstrating the dynamic characteristics of information changes during fault propagation, and providing a mathematical basis for subsequent fault path determination.

[0107] As an optional implementation manner, obtaining a time series feature set according to the system operation log includes:

[0108] Decompose the system operation log into time series to obtain the normal phase, abnormal phase, and recovery phase, and intercept the log fragments of the abnormal and recovery phases;

[0109] Extracting fault propagation features from the log segments, and generating a time series feature set based on the time sequence of the fault propagation features;

[0110] The fault propagation characteristics are used to indicate the fault development of the distribution network, including one or more characteristics of component word frequency, component operation frequency, reverse component frequency, log length and abnormal event density.

[0111] This implementation decomposes system operation logs into time series, precisely dividing them into normal, abnormal, and recovery phases. It then extracts log snippets from the abnormal and recovery phases. Based on this, it extracts fault propagation characteristics and generates a time series feature set based on chronological order. This method preprocesses the underlying log data, improving its validity. It not only captures the temporal characteristics of fault development but also comprehensively reflects the dynamic process of fault propagation through multidimensional feature extraction, providing rich and structured input data for learning to rank.

[0112] As an optional implementation, performing ranking learning on the time series feature set to generate a ranking score includes:

[0113] Perform preliminary sorting based on the fault propagation characteristics through an integrated learning algorithm to generate preliminary sorting results;

[0114] Inputting the preliminary ranking results into the target ranking learning model, and generating a ranking score by fitting the minimization of the association loss function;

[0115] Among them, the ranking score is used to indicate the fault correlation of the nodes corresponding to each target component in the distribution network in fault propagation, and the correlation loss function is used to indicate the difference between the component ranking score output by the target ranking learning model and the actual fault propagation correlation order.

[0116] The algorithm corresponding to the preliminary sorting may include an ensemble learning forest algorithm, and specific implementation methods may include:

[0117] Input training set , minority class label Lmin, forest size S, model ratio q{ratio of RF models in key regions}, number of leaves per tree L;

[0118] A balanced dataset is obtained by using a data partitioning mixed sampling method , the boundary area samples are sampled, and after sampling, we get ; The majority of safe zone samples Cluster into multiple clusters, perform random undersampling on each (the number of samples is half of the number of samples in each cluster), and get ;Retain minority class safe zone samples ; The final comprehensive ;

[0119] Based on the balanced dataset Get the filtered set , a balanced dataset is obtained by sampling with replacement ; Train a decision tree on each subset ; Sampling with replacement filters the dataset to obtain a subset ; Train the decision tree by using the subset corresponding to each filtered decision tree ; Finally, in practical applications, all decision tree outputs can be integrated to obtain preliminary ranking results:

[0120]

[0121] Figure 3 A flow chart of a method for tracing the source of a distribution network fault propagation path provided by an embodiment of the present application is provided in Figure 3 , the target ranking learning model includes iterative training of the LambdaMART algorithm based on the ensemble learning forest algorithm as the initialization:

[0122] Input: the number of trees in the set N, the number of training samples (the number of leaves per tree L), and the learning rate η;

[0123] Based on the ensemble learning forest algorithm, that is, the ELFC algorithm in the figure, the basic regression tree model set F(x) is initialized:

[0124] Traverse all training data (different document pairs), calculate the index change caused by swapping the positions of each document pair, calculate the gradient λ of each document pair, and then calculate the weight ω of each sample.

[0125]

[0126] Create a regression tree that fits λ and generates a regression tree with L (number of leaves) as its leaf nodes

[0127] Find the leaf value based on the approximate Newton step size and the contraction size :

[0128]

[0129] Update the model and incorporate the regression tree learned in each iteration into the F(x) model. Regularize each iteration and perform a penalty at the learning rate η.

[0130]

[0131] The final output is the linear combination of regression trees .

[0132] The ranking scores of system components are fitted by minimizing the association loss function, which is used to measure the difference between the component ranking scores output by the ranking learning model and the true fault propagation association order.

[0133] The following parameters were set for LambdaMART: base learner 'boosting_type': gbrt', number of iterations 'num_iterations': 200, learning rate 'learning_rate': 0.01, minimum number of samples in a leaf node 'min data in leaf': 30, number of leaves 'num_leaves': 31. The default parameters were used for random forest. The number of decision trees for both random forest and ensemble learning forest was 200. The number of neighbors in the ensemble learning forest was k=13, the synthesis factor b=0.5, the forest size S=200, and the model ratio g=0.5.

[0134] This implementation uses an ensemble learning algorithm to perform a preliminary ranking of fault propagation features, generating a preliminary ranking result. This result is then fed into a target ranking learning model and fitted to generate a ranking score by minimizing a correlation loss function. The ranking score accurately indicates the correlation between target components in fault propagation, while the correlation loss function ensures that the model output is consistent with the actual fault propagation sequence. This approach not only improves the accuracy of the ranking results but also enhances the interpretability of fault correlation through model optimization, providing a scientific basis for the final determination of the fault propagation path.

[0135] As an optional implementation manner, determining the fault propagation path according to the ranking score and the information difference map includes:

[0136] Determining a plurality of recommended nodes according to the ranking scores;

[0137] Determining a target component according to an element relationship of each of the recommended nodes in the information difference graph;

[0138] According to the target component, an associated component corresponding to the target component is determined, and according to the target component and the corresponding associated component, a fault propagation path is determined.

[0139] In specific application scenarios, a fault-related recommendation list can be generated based on the component ranking scores output by the ranking learning model; the recommendation list can be matched with the status information of the nodes and edges in the information difference graph, and components with high correlation and significant information change rate can be marked; the interconnection relationship between the marked components can be traced back to infer the path of the fault propagation from the source to the periphery step by step, and finally the fault tracing result can be determined.

[0140] This implementation determines multiple recommended nodes based on ranking scores, combines the element relationships between each recommended node in the information difference graph to determine the target component, and further determines its associated components based on the target component, ultimately constructing the fault propagation path. This method, through the collaborative analysis of ranking scores and information difference graphs, accurately locates the fault propagation path, avoiding the limitations of traditional methods that rely on a single data source and significantly improving the comprehensiveness and accuracy of fault tracing.

[0141] As an optional implementation, the method further includes:

[0142] A visualization process is used to display the fault propagation path in the form of an interactive graphical interface, annotating the ranking scores and edge weight change rates of key nodes in the fault propagation path;

[0143] A fault reporting process for generating a multi-level fault report including the location of the original fault, its propagation path and corresponding timing, and the impact range of the secondary fault;

[0144] The fault alarm process is used to push alarm information to the target operation and maintenance terminal and trigger the corresponding emergency response process according to the fault severity level.

[0145] This implementation uses a visual process to display the fault propagation path in an interactive graphical interface, annotating the ranking scores of key nodes and the rate of change of edge weights, making the fault propagation process more intuitive. Furthermore, the fault reporting process generates multi-level fault reports (including the location of the original fault, the propagation path, and the timing), and the fault alarm process pushes alarm information to the target operation and maintenance terminal, triggering the emergency response process. This achieves closed-loop management from fault tracing to resolution. This approach not only improves the efficiency of fault response but also enhances the scientific nature and timeliness of operation and maintenance decision-making through the integration of multi-dimensional information.

[0146] The present application also provides a device for tracing a distribution network fault propagation path, the device comprising:

[0147] Acquisition module, used to collect polymorphic measurement information and system operation logs in the distribution network;

[0148] The polymorphic measurement information includes a plurality of electrical physical quantities during the operation of the distribution network, and the system operation log is used to indicate the operating status of the distribution network;

[0149] A processing module, configured to pre-process the polymorphic measurement information to generate node feature data;

[0150] The processing module is further configured to construct an information difference graph based on the node feature data;

[0151] The information difference graph is used to indicate the information interconnection relationship and information change between target components before and after the fault in the distribution network;

[0152] The processing module is further configured to obtain a time series feature set based on the system operation log;

[0153] The processing module is further configured to perform ranking learning on the time series feature set to generate a ranking score;

[0154] The processing module is further configured to determine a fault propagation path based on the ranking score and the information difference graph.

[0155] This application comprehensively covers the electrical physical quantities and operating status data during the operation of the distribution network by collecting polymorphic measurement information and system operation logs in the distribution network. On this basis, the polymorphic measurement information is preprocessed to generate node feature data, and the time series feature set is extracted in combination with the system operation log to provide a high-quality data basis for subsequent analysis. By constructing an information difference graph, the information interconnection relationship and changes between the target components before and after the fault can be intuitively reflected, thereby capturing the dynamic characteristics of fault propagation. Furthermore, by sorting and learning the time series feature set to generate a sorting score, and combining the information difference graph to determine the fault propagation path, a complete closed loop from data collection to path tracing is achieved. This method not only improves the accuracy of fault location, but also significantly improves the efficiency and reliability of fault tracing through multi-dimensional data fusion and intelligent analysis, providing strong support for the stable operation of the distribution network.

[0156] As an optional implementation, the polymorphic measurement information includes voltage, current and power data;

[0157] The specific manner in which the processing module pre-processes the polymorphic measurement information to generate node feature data includes:

[0158] Performing denoising and deduplication operations on the polymorphic measurement information to remove invalid or duplicate data, dividing each dimension in the processed polymorphic measurement information into a plurality of feature intervals, and calculating the centroid and sum of squared errors of each feature interval;

[0159] The boundary of the feature interval is dynamically adjusted based on the sum of squared errors to generate discretized node feature data.

[0160] This implementation effectively removes invalid or duplicate data and improves data purity by denoising and deduplicating the voltage, current, and power data in polymorphic measurement information. Furthermore, the processed data is divided into multiple feature intervals, and the centroid and sum of squared errors are calculated. Discrete node feature data is generated by dynamically adjusting the feature interval boundaries, making the data more consistent with subsequent modeling requirements. This process not only optimizes the data structure but also reduces computational complexity through discretization, enhancing the interpretability and practicality of the feature data and laying a solid foundation for constructing information difference maps.

[0161] As an optional implementation manner, the specific manner in which the processing module constructs the information difference graph based on the node feature data includes:

[0162] Construct the first correlation matrix of each node before the fault and the second correlation matrix after the fault respectively;

[0163] The first correlation matrix and the second correlation matrix are used to reflect the information entropy and mutual information relationship between nodes;

[0164] Calculating a difference between the first correlation matrix and the second correlation matrix, and generating an information difference matrix based on the difference and the first correlation matrix;

[0165] Threshold processing and normalization processing are performed on the information difference matrix to construct an information difference map.

[0166] This implementation comprehensively reflects the information entropy and mutual information relationships between nodes by constructing a first correlation matrix for each node before the fault and a second correlation matrix after the fault. An information difference matrix is ​​generated by calculating the difference between the first and second correlation matrices, and this matrix is ​​thresholded and normalized to ultimately construct an information difference map. This method clearly depicts the impact of a fault on the information interconnection relationships in the distribution network, intuitively demonstrating the dynamic characteristics of information changes during fault propagation, and providing a mathematical basis for subsequent fault path determination.

[0167] As an optional implementation manner, the specific manner in which the processing module obtains the time series feature set according to the system operation log includes:

[0168] Decompose the system operation log into time series to obtain the normal phase, abnormal phase, and recovery phase, and intercept the log fragments of the abnormal and recovery phases;

[0169] Extracting fault propagation features from the log segments, and generating a time series feature set based on the time sequence of the fault propagation features;

[0170] The fault propagation characteristics are used to indicate the fault development of the distribution network, including one or more characteristics of component word frequency, component operation frequency, reverse component frequency, log length and abnormal event density.

[0171] This implementation decomposes system operation logs into time series, precisely dividing them into normal, abnormal, and recovery phases. It then extracts log snippets from the abnormal and recovery phases. Based on this, it extracts fault propagation characteristics and generates a time series feature set based on chronological order. This method preprocesses the underlying log data, improving its validity. It not only captures the temporal characteristics of fault development but also comprehensively reflects the dynamic process of fault propagation through multidimensional feature extraction, providing rich and structured input data for learning to rank.

[0172] As an optional implementation manner, the processing module performs ranking learning on the time series feature set to generate a ranking score in a specific manner including:

[0173] Perform preliminary sorting based on the fault propagation characteristics through an integrated learning algorithm to generate preliminary sorting results;

[0174] Inputting the preliminary ranking results into the target ranking learning model, and generating a ranking score by fitting the minimization of the association loss function;

[0175] Among them, the ranking score is used to indicate the fault correlation of the nodes corresponding to each target component in the distribution network in fault propagation, and the correlation loss function is used to indicate the difference between the component ranking score output by the target ranking learning model and the actual fault propagation correlation order.

[0176] This implementation uses an ensemble learning algorithm to perform a preliminary ranking of fault propagation features, generating a preliminary ranking result. This result is then fed into a target ranking learning model and fitted to generate a ranking score by minimizing a correlation loss function. The ranking score accurately indicates the correlation between target components in fault propagation, while the correlation loss function ensures that the model output is consistent with the actual fault propagation sequence. This approach not only improves the accuracy of the ranking results but also enhances the interpretability of fault correlation through model optimization, providing a scientific basis for the final determination of the fault propagation path.

[0177] As an optional implementation manner, the processing module determines a specific manner of the fault propagation path according to the ranking score and the information difference map, including:

[0178] Determining a plurality of recommended nodes according to the ranking scores;

[0179] Determining a target component according to an element relationship of each of the recommended nodes in the information difference graph;

[0180] According to the target component, an associated component corresponding to the target component is determined, and according to the target component and the corresponding associated component, a fault propagation path is determined.

[0181] This implementation determines multiple recommended nodes based on ranking scores, combines the element relationships between each recommended node in the information difference graph to determine the target component, and further determines its associated components based on the target component, ultimately constructing the fault propagation path. This method, through the collaborative analysis of ranking scores and information difference graphs, accurately locates the fault propagation path, avoiding the limitations of traditional methods that rely on a single data source and significantly improving the comprehensiveness and accuracy of fault tracing.

[0182] As an optional implementation manner, the processing module is further configured to execute:

[0183] A visualization process is used to display the fault propagation path in the form of an interactive graphical interface, annotating the ranking scores and edge weight change rates of key nodes in the fault propagation path;

[0184] A fault reporting process for generating a multi-level fault report including the location of the original fault, its propagation path and corresponding timing, and the impact range of the secondary fault;

[0185] The fault alarm process is used to push alarm information to the target operation and maintenance terminal and trigger the corresponding emergency response process according to the fault severity level.

[0186] This implementation uses a visual process to display the fault propagation path in an interactive graphical interface, annotating the ranking scores of key nodes and the rate of change of edge weights, making the fault propagation process more intuitive. Furthermore, the fault reporting process generates multi-level fault reports (including the location of the original fault, the propagation path, and the timing), and the fault alarm process pushes alarm information to the target operation and maintenance terminal, triggering the emergency response process. This achieves closed-loop management from fault tracing to resolution. This approach not only improves the efficiency of fault response but also enhances the scientific nature and timeliness of operation and maintenance decision-making through the integration of multi-dimensional information.

[0187] It should be noted that it should be understood that the division of the various modules of the above device is merely a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. Moreover, these modules can all be implemented in the form of software called by a processing element; or they can all be implemented in the form of hardware; or some modules can be implemented in the form of software called by a processing element, and some modules can be implemented in the form of hardware. For example, the processing module can be a separately established processing element, or it can be integrated into a chip of the above device. In addition, it can also be stored in the memory of the above device in the form of program code, and called by a processing element of the above device to perform the functions of the above-mentioned module. The implementation of other modules is similar. In addition, these modules can all or partly be integrated together, or they can be implemented independently. The processing element here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each of the above modules can be completed by the hardware integrated logic circuit in the processor element or by instructions in the form of software.

[0188] Schematically, as Figure 4 As shown, Figure 4 This is a schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. The computer device 300 can be provided as a server. Figure 4 Computer device 300 includes a processing component 302, which further includes one or more processors, and a memory resource represented by memory 301 for storing instructions executable by processing component 302, such as an application. The application stored in memory 301 may include one or more modules, each corresponding to a set of instructions. In addition, processing component 302 is configured to execute the instructions to perform the method of any of the above embodiments.

[0189] The computer device 300 may further include a power supply component 303 configured to perform power management of the computer device 300, a wired or wireless network interface 304 configured to connect the computer device 300 to a network, and an input / output (I / O) interface 305. The computer device 300 may operate based on an operating system stored in the memory 301, such as Windows Server™, Mac OS X™, Unix™, Linux™, Free BSD™, or the like.

[0190] Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0191] An embodiment of the present application provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute a method as provided in any embodiment.

[0192] Finally, 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 variations thereof are intended to cover non-exclusive inclusion, such 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, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0193] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referenced to each other.

[0194] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for tracing the propagation path of a distribution network fault, characterized in that: The method comprises: Collect multi-state measurement information and system operation logs in the distribution network; The polymorphic measurement information includes a plurality of electrical physical quantities during the operation of the distribution network, and the system operation log is used to indicate the operating status of the distribution network; Preprocessing the polymorphic measurement information to generate node feature data; constructing an information difference graph based on the node feature data; The information difference graph is used to indicate the information interconnection relationship and information change between target components before and after the fault in the distribution network; Obtaining a time series feature set according to the system operation log; Performing ranking learning on the time series feature set to generate a ranking score; A fault propagation path is determined according to the ranking score and the information difference graph.

2. The method according to claim 1, characterized in that The multi-state measurement information includes voltage, current and power data; The preprocessing of the polymorphic measurement information to generate node feature data includes: Performing denoising and deduplication operations on the polymorphic measurement information to remove invalid or duplicate data, dividing each dimension in the processed polymorphic measurement information into a plurality of feature intervals, and calculating the centroid and sum of squared errors of each feature interval; The boundary of the feature interval is dynamically adjusted based on the sum of squared errors to generate discretized node feature data.

3. The method according to claim 1, characterized in that The constructing of an information difference graph based on the node feature data includes: Construct the first correlation matrix of each node before the fault and the second correlation matrix after the fault respectively; The first correlation matrix and the second correlation matrix are used to reflect the information entropy and mutual information relationship between nodes; Calculating a difference between the first correlation matrix and the second correlation matrix, and generating an information difference matrix based on the difference and the first correlation matrix; Threshold processing and normalization processing are performed on the information difference matrix to construct an information difference map.

4. The method according to claim 1, wherein The step of obtaining a time series feature set according to the system operation log includes: Decompose the system operation log into time series to obtain the normal phase, abnormal phase, and recovery phase, and intercept the log fragments of the abnormal and recovery phases; Extracting fault propagation features from the log segments, and generating a time series feature set based on the time sequence of the fault propagation features; The fault propagation characteristics are used to indicate the fault development of the distribution network, including one or more characteristics of component word frequency, component operation frequency, reverse component frequency, log length and abnormal event density.

5. The method according to claim 4, characterized in that The performing ranking learning on the time series feature set to generate a ranking score includes: Perform preliminary sorting based on the fault propagation characteristics through an ensemble learning algorithm to generate preliminary sorting results; Inputting the preliminary ranking results into the target ranking learning model, and generating a ranking score by fitting the minimization of the association loss function; Among them, the ranking score is used to indicate the fault correlation of the nodes corresponding to each target component in the distribution network in fault propagation, and the correlation loss function is used to indicate the difference between the component ranking score output by the target ranking learning model and the actual fault propagation correlation order.

6. The method according to claim 5, characterized in that The determining of the fault propagation path according to the ranking score and the information difference graph includes: Determining a plurality of recommended nodes according to the ranking scores; Determining a target component according to an element relationship of each of the recommended nodes in the information difference graph; According to the target component, an associated component corresponding to the target component is determined, and according to the target component and the corresponding associated component, a fault propagation path is determined.

7. The method according to any one of claims 1 to 6, characterized in that The method further comprises: A visualization process is used to display the fault propagation path in the form of an interactive graphical interface, annotating the ranking scores and edge weight change rates of key nodes in the fault propagation path; A fault reporting process for generating a multi-level fault report including the location of the original fault, its propagation path and corresponding timing, and the impact range of the secondary fault; The fault alarm process is used to push alarm information to the target operation and maintenance terminal and trigger the corresponding emergency response process according to the fault severity level.

8. A distribution network fault propagation path tracing device, characterized in that: The device comprises: Acquisition module, used to collect polymorphic measurement information and system operation logs in the distribution network; The polymorphic measurement information includes a plurality of electrical physical quantities during the operation of the distribution network, and the system operation log is used to indicate the operating status of the distribution network; A processing module, configured to pre-process the polymorphic measurement information to generate node feature data; The processing module is further configured to construct an information difference graph based on the node feature data; The information difference graph is used to indicate the information interconnection relationship and information change between target components before and after the fault in the distribution network; The processing module is further configured to obtain a time series feature set based on the system operation log; The processing module is further configured to perform ranking learning on the time series feature set to generate a ranking score; The processing module is further configured to determine a fault propagation path based on the ranking score and the information difference graph.

9. A computer device, characterized in that: The method comprises one or more processors and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the one or more processors, the steps of the method according to any one of claims 1 to 7 are performed.

10. A storage medium, characterized in that: The storage medium stores computer-readable instructions, which, when executed by one or more processors, enable the one or more processors to perform the steps of the method according to any one of claims 1 to 7.

Images