High-voltage cable pipe gallery sheath ground loop fault source ai positioning method and system

Abstract

This invention relates to the field of cable fault location, and particularly to an AI-based method and system for locating grounding circulation fault sources in high-voltage cable tunnel sheaths. The method includes: collecting circulating current signals from nodes in the tunnel sheath grounding system to construct an initial dataset; calculating mutual information entropy values ​​for node pairs and extracting frequency components to construct a time-varying coupling matrix; statistically analyzing the number of nodes with enhanced coupling; identifying causal emerging nodes through clustering to construct an emerging event graph; selecting the earliest causal emerging node as the starting node; calculating the propagation path entropy value based on the time-varying coupling matrix to construct an entropy-reducing propagation network; calculating the weighted sum of the out-degree centrality and betweenness centrality of network nodes as a topological centrality index; selecting the node with the maximum value as a candidate fault source node; and calculating the matching distance between the frequency components of the candidate fault source node and a preset fault characteristic frequency database to output the location identifier and fault mode. This method achieves accurate location and type identification of grounding circulation faults in high-voltage cable tunnel sheaths.

Description

Technical Field

[0001] This invention relates to the field of cable fault location technology, and in particular to an AI method and system for locating fault sources of grounding loop current in high-voltage cable tunnel sheaths. Background Technology

[0002] High-voltage cable tunnels are vital channels for urban power transmission, and their sheath grounding systems are used to limit induced voltage and guide fault current. When abnormal circulating current occurs in the sheath grounding loop, it usually indicates a grounding fault, sheath damage, or poor contact in the grounding box. In existing technologies, maintenance personnel often rely on periodic measurements of the circulating current amplitude, triggering alarms by setting fixed thresholds, or using offline methods such as infrared thermography and partial discharge detection for assistance. This conventional approach, based on a single threshold and offline detection, struggles to capture the dynamic changes of the circulating current over time and fails to effectively distinguish the coupling relationships between different nodes.

[0003] With the development of smart grids, some solutions have begun to introduce online monitoring and trend analysis of circulating current data, identifying abnormal intervals by statistically analyzing the standard deviation or rate of change of circulating current amplitude in historical data. However, these methods are still limited to the independent processing of signals from a single node, ignoring the temporal and frequency domain correlations of circulating currents at different locations in the grounding system. Since the sheath grounding loop is a multi-node interconnected network structure, the electromagnetic coupling effect caused by a fault will gradually propagate along the cable path. Judging solely by the amplitude abrupt change of a single node can easily lead to misjudging interference from non-faulty nodes as a fault source, or missing weak characteristics in the initial stage of a fault.

[0004] Current technologies lack systematic analysis of circulating current propagation paths. When circulating current fluctuations occur simultaneously at multiple nodes, maintenance personnel often cannot quickly locate the source of the fault and can only troubleshoot one by one or rely on experience to infer the cause. This approach, which relies on manual experience, is not only inefficient but also limited by the professional level of personnel, making it unable to meet the fault location needs in complex coupled scenarios. Furthermore, the identification of fault characteristic frequencies usually only reaches the level of spectrum analysis and does not establish a correspondence with typical fault modes (such as single-point grounding, multi-point grounding, insulation failure, etc.), resulting in a lack of clear fault type information in the location results, making it difficult to guide the formulation of subsequent maintenance strategies. Summary of the Invention

[0005] This invention provides an AI-based method and system for locating grounding loop current fault sources in high-voltage cable tunnel sheaths, which can solve the problems in the prior art.

[0006] A first aspect of the present invention provides an AI method for locating the source of a grounding loop current fault in the sheath of a high-voltage cable gallery, comprising:

[0007] Collect circulating current signals from each node of the pipe gallery sheath grounding system to construct the original dataset;

[0008] The mutual information entropy of the node pairs in the original dataset is calculated within the sliding time window. The time series gradient of the mutual information entropy is used to identify the time corresponding to the gradient peak as the coupling enhancement time. The frequency component corresponding to the coupling enhancement time is extracted, and a time-varying coupling matrix is ​​constructed.

[0009] The number of coupling enhancements generated by each node in the time-varying coupling matrix is ​​counted. The nodes whose number of coupling enhancements is located at the cluster center of the maximum value are identified by the clustering algorithm as causal emergent nodes, and an emergent event graph is constructed.

[0010] The earliest causal emergence node in the emergence event graph is selected as the starting node. The propagation path entropy value is calculated based on the time-varying coupling matrix. The propagation path with the entropy value of subsequent nodes being less than that of the preceding node is selected to construct an entropy reduction propagation network.

[0011] The weighted sum of the out-degree centrality and betweenness centrality of each node in the entropy reduction propagation network is calculated as the topological centrality index, and the node corresponding to the maximum value of the topological centrality index is selected as the candidate fault source node.

[0012] Calculate the matching distance between the frequency components of the candidate fault source node and the preset fault characteristic frequency library, select the fault mode corresponding to the minimum matching distance, and output the location identifier and fault mode type of the candidate fault source node.

[0013] The mutual information entropy of node pairs in the original dataset is calculated within a sliding time window. The time series gradient of the calculated mutual information entropy is used to identify the moment corresponding to the gradient peak as the coupling enhancement moment. The frequency components corresponding to the coupling enhancement moment are extracted, and the time-varying coupling matrix is ​​constructed, including:

[0014] For any two nodes in the original dataset, extract the corresponding circulation signal segments of each node within the sliding time window, calculate the mutual information entropy value between the circulation signal segments, and arrange the mutual information entropy values ​​in chronological order to construct a time series of mutual information entropy values.

[0015] The time series of mutual information entropy values ​​is subjected to first-order difference operation to obtain the time series gradient. The time series gradient is traversed to identify continuous positive gradient intervals. The gradient cumulative sum within the continuous positive gradient interval is calculated. The termination time of the interval corresponding to the maximum value of the gradient cumulative sum is selected as the time corresponding to the gradient peak. The time corresponding to the gradient peak is marked as the coupling enhancement time.

[0016] Extract the circulating signal segment within the sliding time window where the coupling enhancement moment is located, perform Fourier transform on the circulating signal segment to obtain the spectral distribution, extract the dominant frequency corresponding to the maximum amplitude from the spectral distribution as the frequency component corresponding to the coupling enhancement moment, and calculate the amplitude ratio of the dominant frequency as the weighting coefficient of the frequency component.

[0017] Using node pairs as matrix row and column indices, and the coupling enhancement time, frequency components, and weight coefficients as matrix element contents, a time-varying coupling matrix is ​​constructed, where each matrix element records the frequency components and weight coefficients of the corresponding node pair at the coupling enhancement time.

[0018] The number of coupling enhancements generated by each node in the time-varying coupling matrix is ​​statistically analyzed. Nodes whose coupling enhancement numbers are located at the cluster centers of the maximum values ​​are identified as causal emerging nodes using a clustering algorithm. An emerging event graph is then constructed, including:

[0019] Traverse the time-varying coupling matrix to extract all coupling enhancement moments of each node, calculate the time interval between adjacent coupling enhancement moments of each node, construct a probability distribution for the time interval and calculate the information entropy, and use the information entropy as the number of coupling enhancements generated by each node.

[0020] Cluster the number of coupling enhancements generated by each node, extract the frequency components and weight coefficients corresponding to the nodes whose number of coupling enhancements is located at the cluster center of the maximum value, calculate the autocorrelation coefficient of the frequency component weighted sequence, use the ratio of the autocorrelation coefficient to the information entropy as the instability index, and select the nodes whose instability index is greater than the median as causal emergence nodes.

[0021] Extract all coupling enhancement moments and frequency components corresponding to causal emergent nodes in the time-varying coupling matrix. Calculate the cross-correlation function for the coupling enhancement moments of any two causal emergent nodes. Identify the time delay corresponding to the peak value of the cross-correlation function to determine the propagation direction. Calculate the frequency component difference between the two causal emergent nodes as the propagation intensity.

[0022] An emergent event graph is constructed by using causal emergent nodes as graph nodes, propagation direction as directed edges, propagation intensity as directed edge weights, and the earliest coupling enhancement moment as the temporal attribute of the graph node.

[0023] For any two causal emergent nodes, calculate the cross-correlation function at the moment of enhanced coupling, identify the time delay corresponding to the peak of the cross-correlation function to determine the propagation direction, and calculate the frequency component difference between the two causal emergent nodes as the propagation strength, including:

[0024] Fourier spectrum decomposition is performed on the frequency components of each causal emergence node to obtain the amplitude at the fundamental frequency and integer multiples of the power frequency harmonics. The harmonic order corresponding to the harmonic frequency with the largest amplitude at each coupling enhancement moment is identified as the dominant harmonic order.

[0025] For any two causal emergence nodes, select coupling enhancement times with the same dominant harmonic order to construct a set of time pairs with the same frequency, calculate the time difference of each time pair in the set of time pairs with the same frequency, and construct a probability density distribution function as a cross-correlation function by statistically analyzing the occurrence frequency of the time difference.

[0026] The time difference corresponding to the maximum probability density is extracted as the time delay corresponding to the peak of the cross-correlation function, and the propagation direction is determined based on the sign of the time delay.

[0027] The dominant harmonic order of node pairs with propagation direction is extracted to construct a time series. The moment when the dominant harmonic order changes is identified as the harmonic transition moment. The standard deviation of the difference between the harmonic transition moments of the two nodes is calculated. The reciprocal of the standard deviation is normalized and used as the frequency component difference degree. The frequency component difference degree is used as the propagation intensity.

[0028] The earliest causal emergence node in the emergent event graph is selected as the starting node. The propagation path entropy is calculated based on the time-varying coupling matrix. An entropy-reducing propagation network is constructed by selecting propagation paths where the entropy value of subsequent nodes is less than that of their predecessors.

[0029] Extract the emergence time of all causal emergence nodes from the emergence event graph, and mark the causal emergence node with the earliest emergence time as the starting node;

[0030] Extract all outgoing edges of the starting node and the subsequent nodes pointed to by the outgoing edges from the emergent event graph. Obtain the frequency components and weight coefficients of the starting node and each subsequent node in the time-varying coupling matrix. Calculate the weighted variance of the frequency components as the frequency domain dispersion.

[0031] For all coupling enhancement moments in the time-varying coupling matrix, calculate the Shannon entropy of the time interval as the time-domain random entropy. Use the weighted sum of the frequency domain dispersion and the time-domain random entropy as the node entropy value. Calculate the node entropy value of the starting node and the node entropy value of each subsequent node.

[0032] Filter subsequent nodes whose entropy value is less than that of the starting node, construct a propagation path by connecting the directed edges between the starting node and the filtered subsequent nodes and their weights, and record the entropy values ​​of the subsequent nodes in the propagation path as the propagation path entropy value.

[0033] Using the selected subsequent nodes as new starting nodes, the process is repeated iteratively until there are no subsequent nodes whose entropy value is less than that of the preceding node. All propagation paths are then merged to construct an entropy reduction propagation network.

[0034] The weighted sum of the out-degree centrality and betweenness centrality of each node in the entropy reduction propagation network is calculated as the topological centrality index. The node corresponding to the maximum value of the topological centrality index is selected as the candidate fault source node, including:

[0035] Extract all outgoing edges of each node from the entropy reduction propagation network, identify the node entropy difference between the subsequent node pointed to by the outgoing edge and the current node, and count the number of outgoing edges whose node entropy difference exceeds the entropy reduction threshold as the out-degree centrality of each node.

[0036] Extract the propagation paths between all node pairs from the entropy-reducing propagation network, select the paths in which the entropy values ​​between adjacent nodes continuously decrease as monotonic entropy-reducing paths, and count the number of times each node appears as an intermediate node in the monotonic entropy-reducing path as the betweenness centrality of each node.

[0037] Extract the dominant harmonic order corresponding to each node in the entropy reduction propagation network, identify nodes whose dominant harmonic order is the same as the dominant harmonic order of the starting node as co-frequency nodes, set the first weight coefficient group for the out-degree centrality and betweenness centrality of co-frequency nodes, and set the second weight coefficient group for the out-degree centrality and betweenness centrality of non-co-frequency nodes.

[0038] The out-degree centrality and betweenness centrality of each node are weighted and summed according to the corresponding weight coefficient group. The weighted sum is used as the topological centrality index of each node. The maximum value is extracted from the topological centrality index of all nodes, and the node corresponding to the maximum value of the topological centrality index is marked as a candidate fault source node.

[0039] Calculate the matching distance between the frequency components of the candidate fault source node and a preset fault feature frequency database, select the fault mode corresponding to the minimum matching distance, and output the location identifier and fault mode type of the candidate fault source node, including:

[0040] Frequency components are extracted from the time-varying coupling matrix corresponding to the candidate fault source node. The relationship between the occurrence time of each frequency component and the emergence time of the candidate fault source node is identified. Frequency components whose occurrence time is earlier than the emergence time are marked as leading frequency components, and frequency components whose occurrence time is later than the emergence time are marked as response frequency components. A set of leading frequency components and a set of response frequency components are constructed.

[0041] Extract the standard leader frequency component set and standard response frequency component set for each fault mode from the preset fault characteristic frequency library. Calculate the set similarity between the leader frequency component set and the standard leader frequency component set as the leader matching degree. Calculate the set similarity between the response frequency component set and the standard response frequency component set as the response matching degree. Use the absolute value of the difference between the leader matching degree and the response matching degree as the matching distance.

[0042] Select the fault mode corresponding to the minimum matching distance from all fault modes, extract the node identifier information of the candidate fault source node as the location identifier, and output the location identifier and fault mode of the candidate fault source node.

[0043] A second aspect of the present invention provides an AI-based system for locating grounding loop current fault sources in high-voltage cable gallery sheaths, comprising:

[0044] The data acquisition unit is used to collect the circulating current signals of each node in the grounding system of the pipe gallery sheath and construct the original dataset;

[0045] The time-varying coupling unit is used to calculate the mutual information entropy of node pairs in the original dataset within a sliding time window. The time series gradient of the mutual information entropy value is used to identify the moment corresponding to the gradient peak as the coupling enhancement moment. The frequency component corresponding to the coupling enhancement moment is extracted to construct the time-varying coupling matrix.

[0046] Emergent graph units are used to count the number of coupling enhancements generated by each node in the time-varying coupling matrix. Clustering algorithms are used to identify nodes whose number of coupling enhancements is located at the cluster center of the maximum value as causal emergent nodes, and an emergent event graph is constructed.

[0047] Entropy reduction propagation unit is used to select the earliest causal emergence node from the emergence event graph as the starting node, calculate the propagation path entropy value based on the time-varying coupling matrix, and select the propagation path where the entropy value of subsequent nodes is less than that of the previous node to construct the entropy reduction propagation network.

[0048] The topological centrality unit is used to calculate the weighted sum of the out-degree centrality and betweenness centrality of each node in the entropy reduction propagation network as the topological centrality index, and the node corresponding to the maximum value of the topological centrality index is selected as the candidate fault source node.

[0049] The fault matching unit is used to calculate the matching distance between the frequency components of the candidate fault source node and the preset fault characteristic frequency library, select the fault mode corresponding to the minimum matching distance, and output the location identifier and fault mode type of the candidate fault source node.

[0050] A third aspect of the present invention provides an electronic device, comprising:

[0051] processor;

[0052] Memory used to store processor-executable instructions;

[0053] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0054] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0055] This invention utilizes mutual information entropy calculation and time-series gradient analysis to accurately identify moments of enhanced coupling within a sliding window, extracting key frequency components to construct a time-varying coupling matrix, effectively distinguishing between abnormal signal coupling caused by faults and normal fluctuation interference. A clustering algorithm automatically counts the number of enhanced couplings at each node, identifies causal emerging nodes, and constructs an emerging event graph, capturing the earliest emerging starting node in time sequence to avoid misjudgment and omission of fault propagation paths. Based on the time-varying coupling matrix, the propagation path entropy is calculated, and propagation links with decreasing entropy along the path are selected to construct an entropy-reducing propagation network, strictly adhering to the physical propagation laws of fault signals and eliminating interference from reverse or irrelevant paths. The weighted sum of the out-degree centrality and betweenness centrality of nodes in the network is calculated as a topological centrality index, comprehensively reflecting the connectivity breadth and path control of nodes in the propagation topology, eliminating the limitations of a single index, and enhancing the robustness of the positioning results in complex pipe gallery electromagnetic environments. The system calculates the matching distance between the frequency components of candidate fault source nodes and the preset fault characteristic frequency library, automatically matches fault modes using the minimum distance criterion, and outputs the location identifier and fault type of candidate fault sources. This achieves full automation from signal acquisition to fault diagnosis, eliminating the need for manual experience threshold setting and reducing the technical threshold for operation and maintenance personnel. Attached Figure Description

[0056] Figure 1 This is a flowchart illustrating the AI ​​method for locating the grounding loop current fault source in the high-voltage cable gallery sheath according to an embodiment of the present invention.

[0057] Figure 2 This is a flowchart illustrating the candidate fault source node identification process in an embodiment of the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0060] Figure 1 This is a flowchart illustrating the AI ​​method for locating the grounding circulation fault source in the sheath of a high-voltage cable gallery according to an embodiment of the present invention.

[0061] The AI ​​method for locating the fault source of grounding circulation current in the sheath of high-voltage cable corridors includes:

[0062] Collect circulating current signals from each node of the pipe gallery sheath grounding system to construct the original dataset;

[0063] The mutual information entropy of the node pairs in the original dataset is calculated within the sliding time window. The time series gradient of the mutual information entropy is used to identify the time corresponding to the gradient peak as the coupling enhancement time. The frequency component corresponding to the coupling enhancement time is extracted, and a time-varying coupling matrix is ​​constructed.

[0064] The number of coupling enhancements generated by each node in the time-varying coupling matrix is ​​counted. The nodes whose number of coupling enhancements is located at the cluster center of the maximum value are identified by the clustering algorithm as causal emergent nodes, and an emergent event graph is constructed.

[0065] The earliest causal emergence node in the emergence event graph is selected as the starting node. The propagation path entropy value is calculated based on the time-varying coupling matrix. The propagation path with the entropy value of subsequent nodes being less than that of the preceding node is selected to construct an entropy reduction propagation network.

[0066] The weighted sum of the out-degree centrality and betweenness centrality of each node in the entropy reduction propagation network is calculated as the topological centrality index, and the node corresponding to the maximum value of the topological centrality index is selected as the candidate fault source node.

[0067] Calculate the matching distance between the frequency components of the candidate fault source node and the preset fault characteristic frequency library, select the fault mode corresponding to the minimum matching distance, and output the location identifier and fault mode type of the candidate fault source node.

[0068] The mutual information entropy of node pairs in the original dataset is calculated within a sliding time window. The time series gradient of the calculated mutual information entropy is used to identify the moment corresponding to the gradient peak as the coupling enhancement moment. The frequency components corresponding to the coupling enhancement moment are extracted, and the time-varying coupling matrix is ​​constructed, including:

[0069] For any two nodes in the original dataset, extract the corresponding circulation signal segments of each node within the sliding time window, calculate the mutual information entropy value between the circulation signal segments, and arrange the mutual information entropy values ​​in chronological order to construct a time series of mutual information entropy values.

[0070] The time series of mutual information entropy values ​​is subjected to first-order difference operation to obtain the time series gradient. The time series gradient is traversed to identify continuous positive gradient intervals. The gradient cumulative sum within the continuous positive gradient interval is calculated. The termination time of the interval corresponding to the maximum value of the gradient cumulative sum is selected as the time corresponding to the gradient peak. The time corresponding to the gradient peak is marked as the coupling enhancement time.

[0071] Extract the circulating signal segment within the sliding time window where the coupling enhancement moment is located, perform Fourier transform on the circulating signal segment to obtain the spectral distribution, extract the dominant frequency corresponding to the maximum amplitude from the spectral distribution as the frequency component corresponding to the coupling enhancement moment, and calculate the amplitude ratio of the dominant frequency as the weighting coefficient of the frequency component.

[0072] Using node pairs as matrix row and column indices, and the coupling enhancement time, frequency components, and weight coefficients as matrix element contents, a time-varying coupling matrix is ​​constructed, where each matrix element records the frequency components and weight coefficients of the corresponding node pair at the coupling enhancement time.

[0073] In high-voltage cable tunnel sheath grounding systems, the circulating current signals collected by each monitoring node often exhibit nonlinear and non-stationary characteristics, making it difficult for traditional linear correlation analysis methods to effectively capture the dynamic coupling relationship between nodes. Therefore, for any pair of nodes in the original dataset, circulating current signal segments corresponding to each node are extracted within a sliding time window. The length of the sliding time window is determined based on a combination of the sampling frequency and the typical fluctuation period of the signal, with the window step size set as a certain proportion of the window length (usually 10% to 30%) to ensure sufficient overlap between adjacent windows, thereby continuously tracking the evolution of the coupling relationship between node pairs over time. For tunnels containing... The grounding system of each monitoring node has a total of [number] node pairs. Each pair of nodes must independently complete the signal extraction operation within the aforementioned sliding window.

[0074] For each sliding time window, the mutual information entropy value between the two extracted circulating current signal segments is calculated. Mutual information entropy measures the statistical dependence between two random variables, is not limited by linear assumptions, and is suitable for describing the nonlinear coupling enhancement phenomenon caused by faults such as insulation degradation and abnormal contact resistance in grounding circulating current signals. Let the nodes... With nodes In the The circulating signal segments within each time window are respectively and Then the mutual information entropy value The calculation is based on the joint probability distribution and individual marginal probability distributions of the two signals. After discretizing the continuous signal using kernel density estimation or equal-frequency binning, probability estimates are obtained, thus completing the numerical calculation of the mutual information entropy value. The mutual information entropy values ​​corresponding to all time windows are then calculated. Arranged sequentially according to window time, forming node pairs. Mutual information entropy value time series ,in This represents the total number of windows. This time series fully records the trajectory of the coupling strength between node pairs over time, providing fundamental data for subsequent gradient analysis.

[0075] A first-order differencing operation is performed on the mutual information entropy time series to obtain the time series gradient sequence. The first-order differencing operation calculates the difference in mutual information entropy values ​​between adjacent time windows, reflecting the rate of change of coupling strength within a unit time step. The gradient sequence is iterated to identify continuous positive gradient intervals, i.e., consecutive time periods where the gradient value is continuously greater than zero. Within each continuous positive gradient interval, the gradient values ​​at each time step are accumulated to obtain the gradient sum, which reflects the overall increase in coupling strength within that interval. After iterating through all continuous positive gradient intervals, the termination time of the interval corresponding to the maximum gradient sum is selected as the time corresponding to the gradient peak, and this time is marked as the coupling enhancement moment. This approach effectively eliminates transient random fluctuations, focusing on critical moments when coupling strength significantly and continuously increases, accurately reflecting the initial characteristics of abnormal coupling between nodes caused by faults. For each node pair, the moment of coupling enhancement is determined independently using the above method.

[0076] Determine the coupling enhancement time Next, a segment of the circulating signal within the sliding time window at that moment is extracted, and a Fourier transform is performed on this signal segment to obtain its spectral distribution. The Fourier transform decomposes the time-domain signal into a superposition of different frequency components. The amplitude corresponding to each frequency point in the spectral distribution reflects the energy proportion of that frequency component in the signal. The frequency corresponding to the maximum amplitude is identified from the spectral distribution and defined as the dominant frequency. The dominant frequency is the frequency component corresponding to the moment of enhanced coupling. The dominant frequency can represent the main frequency channel of energy transfer between the two nodes at the moment of enhanced coupling. For different types of grounding system faults (such as single-point grounding failure, multi-point grounding short circuit, sheath insulation damage, etc.), the dominant frequency often exhibits different characteristic value ranges. Therefore, the dominant frequency is an important basis for subsequent fault mode matching.

[0077] In acquiring dominant frequency Simultaneously, the amplitude proportion of the dominant frequency is calculated as the weighting coefficient of the frequency component. Specifically, weighting coefficients Defined as the ratio of the amplitude corresponding to the dominant frequency to the sum of the amplitudes of all frequency points in the spectral distribution, i.e. ,in The amplitude at the dominant frequency, For the first The amplitude at each frequency point. The weighting coefficient reflects the degree of energy concentration of the dominant frequency in the entire spectrum: when the weighting coefficient is large, it indicates that the energy at the moment of enhanced coupling is highly concentrated at a single frequency, which has a strong indication of fault characteristics; when the weighting coefficient is small, it indicates that the spectrum is more dispersed, and the enhanced coupling may be due to the superposition effect of multiple frequencies, which needs to be distinguished in subsequent analysis.

[0078] Construct a time-varying coupling matrix using node pairs as row and column indices. The first matrix Line number Column element corresponding node With nodes The constructed node pairs contain three pieces of information in their element content: the coupling enhancement time. Dominant frequency and weighting coefficients For node pairs that do not exhibit significant coupling enhancement (i.e., the cumulative sum of gradients across all consecutive positive gradient intervals is below a preset threshold), the corresponding matrix elements are recorded as null values ​​or filled with zero values, indicating that no statistically significant coupling enhancement event occurred for that node pair during the observation period. Time-varying coupling matrix The coupling enhancement characteristics of all node pairs in the pipe gallery grounding system were fully recorded in a structured manner, providing a unified data foundation for the subsequent construction of emergent event maps and fault propagation path analysis.

[0079] The construction of time-varying coupling matrices requires attention to the handling of matrix symmetry. Because the mutual information entropy value itself possesses symmetry, node pairs... With node pair The calculated mutual information entropy values ​​are the same across time series, therefore the matrix Numerically symmetrical, the structure allows for storage of only the upper or lower triangular portion to conserve computational resources. Furthermore, regarding the selection of the sliding time window parameter, a window length that is too short can lead to inaccurate probability estimation, affecting the reliability of the mutual information entropy value; a window length that is too long will smooth out transient coupling enhancement events, reducing the accuracy of moment identification. In practice, it is recommended that the window length cover at least five complete cycles of the signal's fundamental frequency and be adjusted based on the actual sampling rate to balance computational accuracy and time resolution requirements. Through the above complete process, the time-varying coupling matrix can accurately characterize the time-frequency features of the dynamic coupling relationship between nodes in the high-voltage cable tunnel sheath grounding system during fault evolution, providing reliable input data for subsequent fault source localization based on causal emergence theory.

[0080] The number of coupling enhancements generated by each node in the time-varying coupling matrix is ​​statistically analyzed. Nodes whose coupling enhancement numbers are located at the cluster centers of the maximum values ​​are identified as causal emerging nodes using a clustering algorithm. An emerging event graph is then constructed, including:

[0081] Traverse the time-varying coupling matrix to extract all coupling enhancement moments of each node, calculate the time interval between adjacent coupling enhancement moments of each node, construct a probability distribution for the time interval and calculate the information entropy, and use the information entropy as the number of coupling enhancements generated by each node.

[0082] Cluster the number of coupling enhancements generated by each node, extract the frequency components and weight coefficients corresponding to the nodes whose number of coupling enhancements is located at the cluster center of the maximum value, calculate the autocorrelation coefficient of the frequency component weighted sequence, use the ratio of the autocorrelation coefficient to the information entropy as the instability index, and select the nodes whose instability index is greater than the median as causal emergence nodes.

[0083] Extract all coupling enhancement moments and frequency components corresponding to causal emergent nodes in the time-varying coupling matrix. Calculate the cross-correlation function for the coupling enhancement moments of any two causal emergent nodes. Identify the time delay corresponding to the peak value of the cross-correlation function to determine the propagation direction. Calculate the frequency component difference between the two causal emergent nodes as the propagation intensity.

[0084] An emergent event graph is constructed by using causal emergent nodes as graph nodes, propagation direction as directed edges, propagation intensity as directed edge weights, and the earliest coupling enhancement moment as the temporal attribute of the graph node.

[0085] Traversing the time-varying coupling matrix For each node, extract the set of moments in which the coupling of that node is enhanced across all time windows. For a given node... The set of times when coupling is enhanced is denoted as ,in For nodes The total number of times coupling enhancement occurred, sorted in ascending chronological order. The time interval sequence between adjacent coupling enhancement moments was calculated. ,in , This serves as the interval index. Histogram statistics are performed on the time interval sequence to obtain the probability distribution of each interval. Information entropy is calculated based on this probability distribution. ,Right now ,in For histogram interval index, For nodes The time interval falls into the first The probability of each interval. Information entropy. This reflects the degree of randomness in the temporal distribution of node coupling enhancement behavior: if nodes frequently and regularly generate coupling enhancement, the interval distribution will be concentrated. Smaller; if the coupling enhancement time is scattered, then Relatively large. With As a node This generates a quantitative representation of the coupling enhancement quantity, thereby transforming the original counting problem into an entropy measure of distribution characteristics, avoiding the limitation that simple counting cannot distinguish between regularity and randomness.

[0086] Collect the information entropy values ​​of all nodes and perform K-Means clustering to divide all nodes into several clusters. Extract the set of nodes corresponding to the cluster centers with the largest mean information entropy, i.e., the set of nodes whose coupling enhancement number is located at the maximum value cluster center, and denote it as the candidate node set. .right Each node Extract its time-varying coupling matrix The frequency components corresponding to all coupling enhancement moments. and weighting coefficients ( (Construct a frequency component weighted sequence based on the index of the number of times the coupling enhancement is applied to this node). , of which The elements are For weighted sequences Calculate the autocorrelation coefficient The normalized value of the correlation function at zero delay is taken as... The representative quantity, A larger value indicates a stronger periodicity in the frequency component weighted sequence, suggesting that the coupling behavior of the nodes exhibits stable frequency repetition characteristics. The autocorrelation coefficient... With information entropy The ratio is defined as the node Instability index ,Right now . A larger value indicates that the frequency coupling of the nodes is highly repetitive, and the timing of its enhanced coupling is relatively concentrated. This is consistent with the continuous and regular circulating current anomalies caused by insulation degradation or grounding faults in sheath grounding systems. Calculate the candidate node set. Median of all node instability indices Select the one that satisfies The nodes are used as causal emergence nodes, forming a causal emergence node set.

[0087] Extract all coupling enhancement times and frequency components corresponding to each node in the time-varying coupling matrix of the causal emergent node set. For any two causal emergent nodes... and Extract their coupling enhancement time sequences respectively and ( For nodes The coupling enhancement order index is used to transform the two time series into binary time series with equal time steps. Then, the cross-correlation function is calculated for the two binary series, where... For time-delayed variables. Identify the cross-correlation function in Time delay corresponding to the peak value within the range .like , indicating nodes The coupling enhancement time generally precedes the node. The direction of transmission is from point to ;like The direction of propagation is from point to This directional judgment logic is based on the causal sequence, that is, the node where the coupling enhancement occurs first is more likely to be the source of the abnormal signal or an intermediate propagation node.

[0088] Based on the determined propagation direction, the two causal emergence nodes are further calculated. and Frequency component difference between This serves as a quantitative indicator of propagation strength. Extracting nodes. frequency component set With nodes frequency component set Calculate the weighted mean of the two sets respectively. and The weight is the corresponding weight coefficient. and Frequency component difference Defined as the normalized difference between the weighted means of the two, i.e. . The smaller the value, the more similar the two nodes are in frequency characteristics, the higher the propagation strength, and the closer the coupling relationship between the two nodes. The larger the value, the weaker the propagation intensity, indicating that the frequency coupling correlation between the two nodes is relatively independent. In high-voltage cable tunnels, circulating current anomalies caused by the same fault source will propagate along the grounding line to adjacent nodes. During the propagation process, the similarity of frequency characteristics will decrease as the propagation distance increases. Therefore, using the frequency difference as a reverse measure of propagation intensity has physical rationality.

[0089] Based on the above calculation results, each node in the causal emergent node set is used as a graph node, the directed relationship determined by the propagation direction is used as a directed edge, and the reciprocal of the propagation intensity is used as the weight of the directed edge (the smaller the difference, the larger the weight, indicating stronger propagation). The earliest time of coupling enhancement of each node is used as the temporal attribute of the graph node to construct an emergent event graph. The graph fully records the temporal relationship, frequency coupling propagation path, and propagation intensity of each causal emergent node, providing a structured input basis for subsequent construction of entropy reduction propagation network based on time-varying coupling matrix and location of candidate fault source nodes. In actual utility tunnel scenarios, the emergent event graph can intuitively reflect the dynamic process of grounding circulation anomalies spreading from the initial excitation node to surrounding nodes, effectively supporting the source analysis of fault propagation links.

[0090] For any two causal emergent nodes, calculate the cross-correlation function at the moment of enhanced coupling, identify the time delay corresponding to the peak of the cross-correlation function to determine the propagation direction, and calculate the frequency component difference between the two causal emergent nodes as the propagation strength, including:

[0091] Fourier spectrum decomposition is performed on the frequency components of each causal emergence node to obtain the amplitude at the fundamental frequency and integer multiples of the power frequency harmonics. The harmonic order corresponding to the harmonic frequency with the largest amplitude at each coupling enhancement moment is identified as the dominant harmonic order.

[0092] For any two causal emergence nodes, select coupling enhancement times with the same dominant harmonic order to construct a set of time pairs with the same frequency, calculate the time difference of each time pair in the set of time pairs with the same frequency, and construct a probability density distribution function as a cross-correlation function by statistically analyzing the occurrence frequency of the time difference.

[0093] The time difference corresponding to the maximum probability density is extracted as the time delay corresponding to the peak of the cross-correlation function, and the propagation direction is determined based on the sign of the time delay.

[0094] The dominant harmonic order of node pairs with propagation direction is extracted to construct a time series. The moment when the dominant harmonic order changes is identified as the harmonic transition moment. The standard deviation of the difference between the harmonic transition moments of the two nodes is calculated. The reciprocal of the standard deviation is normalized and used as the frequency component difference degree. The frequency component difference degree is used as the propagation intensity.

[0095] To determine the propagation relationship between causal emerging nodes, Fourier spectral decomposition of the frequency components of each node is required. For each causal emerging node, a Discrete Fourier Transform is performed on the circulating signal segment near the coupling enhancement moment to extract the amplitudes at the fundamental frequency (50Hz) and integer multiples of the fundamental frequency, i.e., the spectral amplitudes corresponding to each harmonic, such as 100Hz, 150Hz, and 200Hz. For each coupling enhancement moment, all harmonic frequency points are traversed, and the harmonic order corresponding to the harmonic frequency with the largest amplitude is identified and recorded as the dominant harmonic order at that moment. The dominant harmonic order reflects the frequency component with the most concentrated energy in the current coupled enhancement event, and serves as the fundamental information carrier for subsequent propagation direction determination and propagation intensity calculation.

[0096] For any two causal emergence nodes and From their respective coupling enhancement time sequences, times with the same dominant harmonic order are selected, and the nodes are... The dominant harmonic order is The coupling enhancement moment is denoted as ,node The dominant harmonic order is the same The coupling enhancement moment is denoted as ,in and These are the event indices in their respective sequences. All time pairs that satisfy the condition of having the same dominant harmonic order are considered. Forming a set of time pairs with the same frequency For sets For each time pair in the dataset, calculate the time difference. The frequency of all time differences is counted, and the frequency is normalized to construct a probability density distribution function. This function is then used as the node. With nodes The cross-correlation function between them. This statistical method based on time pairs with the same frequency avoids the dependence of traditional cross-correlation calculations on signal continuity and equal-interval sampling, and is suitable for practical scenarios in utility tunnel environments where circulating signals are missing or non-stationary.

[0097] Extract the time difference corresponding to the maximum probability density value from the above probability density distribution function, and use it as the time delay corresponding to the peak of the cross-correlation function. Time delay The numerical sign directly determines the direction of propagation: when When, it indicates the node The coupling enhancement events generally occur earlier than the node. The direction of transmission is from point to ;when At that time, the direction of transmission was from point to ;when When the probability density distribution function is close to zero and exhibits a broad peak or bimodal characteristic, it is considered that there is no significant unidirectional propagation relationship between the two nodes, and directed edges are not established in the emergent event graph. By repeating the above process for all causal emergent node pairs, the direction labeling of directed edges in the emergent event graph is completed, forming a directed graph structure with propagation direction information.

[0098] After determining the propagation direction, the propagation strength between node pairs with valid propagation directions is further calculated. This is for node pairs that have established directed propagation relationships. Extract nodes respectively With nodes The dominant harmonic orders recorded at their respective coupling enhancement time series are arranged in chronological order to construct a dominant harmonic order time series. and ,in and They are nodes With nodes The total number of coupling enhancements. Traversing the time series of their respective dominant harmonic orders, identify changes in the dominant harmonic order between two adjacent coupling enhancement moments (i.e., At the moment of [the change in harmonic frequency], this moment is recorded as the harmonic transition moment. For nodes... Extracting the harmonic transition time sequence For nodes Extracting the harmonic transition time sequence .

[0099] The harmonic transition moment reflects the time point at which the dominant energy frequency component of a node switches. During the propagation of fault disturbances along the pipe gallery liner, the harmonic transition at the source node often occurs before that at the downstream node, and the difference in the timing of these two harmonic transitions shows a relatively concentrated distribution across multiple events. If the node... With nodes Harmonic transition time difference sequence ( The standard deviation of the paired index for harmonic transition events A smaller value indicates that the two nodes are highly synchronized in the harmonic transition timing, their frequency component evolution paths are similar, and their propagation coupling is tight; conversely, if... A larger value indicates that the frequency components of the two nodes evolve relatively independently, and the propagation coupling is weak.

[0100] Based on the above analysis, the propagation intensity is defined as the standard deviation. The normalized result of the reciprocal. Let the set of standard deviations of all node pairs with directed propagation relationships be . For each node pair Perform max-min normalization to obtain the normalized frequency component difference. The specific calculation is as follows: ,in and These are the minimum and maximum values ​​of the inverse values ​​of all nodes, respectively. The range of values ​​is The larger the value, the more synchronized the two nodes are in the harmonic transition timing, and the stronger the propagation coupling. As an emergent event map The propagation intensity weights of the corresponding directed edges are assigned to complete the weight assignment of the graph.

[0101] In practical engineering applications, high-voltage cable tunnel sheath grounding systems are affected by various interference factors, and different fault modes (such as single-phase grounding faults, insulation aging, and sheath damage) exhibit significant differences in harmonic distribution characteristics. By capturing the frequency component switching patterns of fault disturbances during propagation through the jump behavior of the dominant harmonic order, it is possible to effectively distinguish between transient coupling enhancement caused by external load fluctuations and systemic coupling diffusion caused by continuous excitation from the fault source. The method of constructing the set of events with the same frequency ensures that the propagation direction is determined only based on events with the same physical excitation frequency, eliminating spurious correlations between different harmonic components and improving the reliability of propagation path identification. Finally, the propagation direction and propagation intensity together constitute an emergent event map. The complete properties of the directed weighted edges provide accurate graph structure input for the subsequent construction of the entropy reduction propagation network and the topological centrality analysis of candidate fault source nodes.

[0102] The earliest causal emergence node in the emergent event graph is selected as the starting node. The propagation path entropy is calculated based on the time-varying coupling matrix. An entropy-reducing propagation network is constructed by selecting propagation paths where the entropy value of subsequent nodes is less than that of their predecessors.

[0103] Extract the emergence time of all causal emergence nodes from the emergence event graph, and mark the causal emergence node with the earliest emergence time as the starting node;

[0104] Extract all outgoing edges of the starting node and the subsequent nodes pointed to by the outgoing edges from the emergent event graph. Obtain the frequency components and weight coefficients of the starting node and each subsequent node in the time-varying coupling matrix. Calculate the weighted variance of the frequency components as the frequency domain dispersion.

[0105] For all coupling enhancement moments in the time-varying coupling matrix, calculate the Shannon entropy of the time interval as the time-domain random entropy. Use the weighted sum of the frequency domain dispersion and the time-domain random entropy as the node entropy value. Calculate the node entropy value of the starting node and the node entropy value of each subsequent node.

[0106] Filter subsequent nodes whose entropy value is less than that of the starting node, construct a propagation path by connecting the directed edges between the starting node and the filtered subsequent nodes and their weights, and record the entropy values ​​of the subsequent nodes in the propagation path as the propagation path entropy value.

[0107] Using the selected subsequent nodes as new starting nodes, the process is repeated iteratively until there are no subsequent nodes whose entropy value is less than that of the preceding node. All propagation paths are then merged to construct an entropy reduction propagation network.

[0108] After constructing the emergent event graph, it is necessary to extract the emergence time corresponding to each causal emergent node from the graph. The emergence time is defined as the timestamp of the first occurrence of coupling enhancement at a node, that is, the earliest time when the node is recorded as the coupling enhancement in the time-varying coupling matrix. By traversing all causal emergent nodes in the emergent event graph and comparing the emergence times of each node, the causal emergent node with the smallest emergence time value (i.e., the earliest occurrence) is marked as the starting node. If multiple nodes emerge at the same time and are all the earliest, they are sorted according to their out-degree in the time-varying coupling matrix, and the node with the largest degree is taken as the starting node to ensure that the extension of the subsequent propagation path has sufficient information.

[0109] Extracting the starting node from the emergent event graph All outgoing edges, and the set of nodes pointed to by the outgoing edges are denoted as . This refers to the set of nodes that are the direct successors of the starting node. For the starting node... With sets Each subsequent node in ( (For the enumeration index of subsequent nodes), read node pairs from the time-varying coupling matrix. All corresponding dominant frequencies and its weighting coefficients The weighted average of these frequency components is denoted as... The weighted variance of each frequency component is obtained by calculating the square of the difference between each frequency component and the weighted mean, and then summing the results according to the weighting coefficients. This variance is defined as the frequency domain dispersion. The formula is ,in For node pairs in the th The dominant frequency at each frequency point These are the corresponding weighting coefficients. The greater the frequency domain dispersion, the more dispersed the frequency coupling between the node pairs, and the higher the frequency uncertainty of signal propagation.

[0110] After completing the frequency domain dispersion calculation, the node pairs in the time-varying coupling matrix are then analyzed. From all the coupling enhancement time sequences, the time interval sequences between adjacent coupling enhancement times are extracted. Histogram statistics are performed on these time interval sequences, and the frequencies of each interval are normalized to a probability distribution. Then, the Shannon entropy of this probability distribution is calculated to obtain the temporal random entropy. The formula is ,in For node pairs Coupling enhancement time interval falls into the first The probability of each histogram interval. The larger the temporal random entropy, the more random and less regular the coupling enhancement events are in their temporal distribution, corresponding to higher signal propagation uncertainty.

[0111] The frequency domain dispersion and the time domain random entropy are weighted and summed to obtain the node pairs. propagation path entropy The formula is ,in and These are the weighting coefficients for frequency domain dispersion and time domain random entropy, respectively, with a sum of 1. The specific values ​​can be set according to the relative importance of frequency domain and time domain information in the actual engineering scenario; the default value is [value missing]. , For the starting node Its own node entropy Based on In the time-varying coupling matrix, the frequency components corresponding to all outgoing edges and the coupling enhancement time are summarized and calculated in the same way, and the average of the entropy values ​​of all outgoing edge propagation paths is taken as the mean. The node entropy value.

[0112] Calculate the starting node separately node entropy and each subsequent node node entropy Next, filter those that meet the criteria. The subsequent nodes are then used to form a filtered set of subsequent nodes. For each node in the set of subsequent nodes... , start node To subsequent nodes Directed edges between nodes are preserved, and the weight of these directed edges is set to the weight of the node pair in the time-varying coupling matrix. The average of the frequency component weighting coefficients, and also record the subsequent nodes. node entropy This serves as the propagation path entropy value for that propagation path. If subsequent nodes... The node entropy value does not satisfy the entropy reduction condition, that is... If the corresponding directed edge is not included in the entropy reduction propagation network, the propagation path of that branch will terminate.

[0113] Complete the starting node After one round of expansion, each node in the subsequent node set is used as a new starting node, and the above steps are repeated: extract the outgoing edges and subsequent nodes of the current starting node from the emergent event graph, calculate the frequency domain dispersion and time domain random entropy, obtain the node entropy value of each subsequent node, filter the subsequent nodes that satisfy the condition that the node entropy value is less than the node entropy value of the current starting node, retain the corresponding directed edges and weights, and record the propagation path entropy value. The iterative process continues until all nodes in the current traversal layer have no subsequent nodes that satisfy the entropy reduction condition, that is, for each node in the current layer... All its subsequent nodes All meet The iteration terminates.

[0114] To prevent loops in the graph from causing infinite iterations, the set of visited nodes is recorded each time a node is added to the entropy reduction propagation network. If a subsequent node has already appeared in the current propagation path, that node is skipped and not expanded upon. This mechanism ensures that the entropy reduction propagation network is a directed acyclic graph, which can accurately reflect the physical propagation law of fault signals spreading unidirectionally from the source.

[0115] By summing up the directed edges and their weights retained from all iterations, a complete entropy reduction propagation network is formed. This network starts with the node... The root node is defined as the network node, and subsequent nodes selected from each layer are its child nodes. Each directed edge in the network satisfies the entropy reduction constraint that "the entropy value of the subsequent node is strictly less than the entropy value of the preceding node," resulting in an overall ordered structure propagating from high-entropy nodes to low-entropy nodes. The construction results of the entropy reduction propagation network intuitively reflect the diffusion path and propagation direction of the circulating fault signal in the pipe gallery sheath grounding system, providing a structured propagation graph basis for subsequent location of candidate fault source nodes based on topological centrality indices. In practical engineering applications, the number of nodes in the pipe gallery may be large, and the number of iteration layers may reach dozens. Entropy reduction constraints can effectively prune the network, retaining only the core path of ordered information propagation, significantly reducing the computational complexity of subsequent analysis.

[0116] like Figure 2 As shown, Figure 2 This is a flowchart illustrating the candidate fault source node identification process in an embodiment of the present invention.

[0117] The weighted sum of the out-degree centrality and betweenness centrality of each node in the entropy reduction propagation network is calculated as the topological centrality index. The node corresponding to the maximum value of the topological centrality index is selected as the candidate fault source node, including:

[0118] Extract all outgoing edges of each node from the entropy reduction propagation network, identify the node entropy difference between the subsequent node pointed to by the outgoing edge and the current node, and count the number of outgoing edges whose node entropy difference exceeds the entropy reduction threshold as the out-degree centrality of each node.

[0119] Extract the propagation paths between all node pairs from the entropy-reducing propagation network, select the paths in which the entropy values ​​between adjacent nodes continuously decrease as monotonic entropy-reducing paths, and count the number of times each node appears as an intermediate node in the monotonic entropy-reducing path as the betweenness centrality of each node.

[0120] Extract the dominant harmonic order corresponding to each node in the entropy reduction propagation network, identify nodes whose dominant harmonic order is the same as the dominant harmonic order of the starting node as co-frequency nodes, set the first weight coefficient group for the out-degree centrality and betweenness centrality of co-frequency nodes, and set the second weight coefficient group for the out-degree centrality and betweenness centrality of non-co-frequency nodes.

[0121] The out-degree centrality and betweenness centrality of each node are weighted and summed according to the corresponding weight coefficient group. The weighted sum is used as the topological centrality index of each node. The maximum value is extracted from the topological centrality index of all nodes, and the node corresponding to the maximum value of the topological centrality index is marked as a candidate fault source node.

[0122] After the entropy reduction propagation network is constructed, it is necessary to identify the most likely source node of the fault. To this end, a topological centrality index is introduced, which comprehensively considers the out-degree centrality and betweenness centrality of a node, and can effectively reflect the driving force and pivotal position of a node in the fault propagation process.

[0123] Out-degree centrality is calculated based on the outgoing edges of a node. All outgoing edges of each node are extracted from the entropy reduction propagation network. For each outgoing edge, the difference in node entropy between the starting node of that outgoing edge and the subsequent nodes it points to is calculated. Specifically, if a node... There exists an outgoing edge pointing to a subsequent node. Then calculate the difference between the node entropy values ​​of the two nodes. .when Exceeding the preset entropy reduction threshold At that time, it is considered that the entropy reduction corresponding to the outgoing edge is significant, and has substantial significance for fault propagation. Statistical Nodes All satisfied The number of outgoing edges of a node is denoted as the out-degree centrality of that node. Entropy reduction threshold The setting can be determined based on the statistical distribution of node entropy values ​​under normal operating conditions in the actual utility tunnel environment. It is usually taken as a certain multiple of the standard deviation of the entropy values ​​of all nodes in the network to eliminate small entropy reduction fluctuations caused by signal noise. The higher the out-degree centrality, the more downstream nodes can be driven to undergo significant entropy reduction jumps, playing an important source role in the fault propagation chain.

[0124] Betweenness centrality calculation focuses on the pivotal role of a node in the global propagation path. Directed propagation paths between all node pairs are extracted from the entropy-decreasing propagation network, and a monotonicity test is performed on each path: the change in node entropy values ​​between adjacent nodes is examined segment by segment along the path direction. If the entropy value of each adjacent node pair on the path is strictly less than the entropy value of the preceding node, then the path is marked as a monotonically decreasing entropy path. A monotonically decreasing entropy path represents a continuous and orderly energy dissipation pattern of the fault signal along the propagation direction, and is the most physically meaningful fault propagation trajectory. After screening all monotonically decreasing entropy paths, the statistics of each node are... The number of times a node appears as an intermediate node (i.e., a node that is neither the start nor the end of a path) in these monotonically decreasing paths is denoted by the betweenness centrality of that node. The higher the betweenness centrality, the more critical the node is in a relay position in multiple fault propagation channels, and the wider the impact of its failure or anomaly on the entire propagation network.

[0125] After obtaining the out-degree centrality and betweenness centrality, the weight coefficients need to be configured differently based on the frequency characteristics of the nodes. The dominant harmonic order of each node in the entropy reduction propagation network is extracted, and the dominant harmonic order of each node is compared with that of the starting node. The dominant harmonic order is compared with that of the starting node. Nodes with the same dominant harmonic order as the starting node are identified as co-frequency nodes, and the remaining nodes are non-co-frequency nodes. Co-frequency nodes and the starting node have consistent frequency characteristics, which means that they belong to the response chain of the same fault excitation source in terms of electrical coupling mechanism. Their topology centrality index should be given higher weight to highlight their fault correlation. A first weight coefficient is set for the out-degree centrality of co-frequency nodes. To set another coefficient in the first weighted coefficient group for the betweenness centrality of nodes with the same frequency. A second weighting coefficient is set for the out-degree centrality of non-co-frequency nodes. A second weighting coefficient is set for the betweenness centrality of non-co-frequency nodes. ,in , The basis for this differentiated weighting design is that: grounding loop faults in the sheath of high-voltage cable tunnels usually manifest as abnormal enhancement of specific harmonic orders. Nodes with the same dominant harmonic characteristics as the fault source are more likely to be on the same fault propagation path, so their topological status should be given higher recognition in the comprehensive evaluation.

[0126] The topological centrality index of each node is obtained by weighting and summing the results according to its corresponding weight coefficient set. For nodes with the same frequency, For nodes that are not in sync with each other, In practical engineering applications, and The proportional relationship can be adjusted according to the topological characteristics of the utility tunnel route: for utility tunnels with longer routes and a larger number of nodes, betweenness centrality contributes more significantly to fault location, and can be appropriately increased. The value of is important; for pipe corridors with dense nodes and many branches, the out-degree centrality has stronger distinguishability and can be appropriately increased. The value of .

[0127] The topological centrality index of all nodes in the entropy reduction propagation network is traversed, and the maximum value is extracted. The corresponding node is marked as a candidate fault source node. The node with the highest topological centrality index has both strong fault propagation driving ability (high out-degree centrality) and a wide influence coverage (high betweenness centrality). Its frequency characteristics are also consistent with the starting node, comprehensively reflecting its highest probability as a fault source. When multiple nodes have similar topological centrality indices, a relative difference threshold can be set. When the difference between the second highest value and the highest value is less than At the same time, these nodes are included in the candidate fault source node set for further screening in the subsequent frequency feature matching step, thereby avoiding the omission of candidate nodes due to slight differences in a single indicator and improving the robustness of overall fault location.

[0128] This method of calculating topological centrality index closely combines the classical centrality measure in graph theory with the physical characteristics of cable sheath grounding loop faults. It considers the structural position of nodes in the propagation network and introduces domain knowledge through the consistency constraint of frequency characteristics, effectively reducing the probability of misjudgment caused by network topology complexity and providing reliable candidate node inputs for subsequent fault mode matching.

[0129] Calculate the matching distance between the frequency components of the candidate fault source node and a preset fault feature frequency database, select the fault mode corresponding to the minimum matching distance, and output the location identifier and fault mode type of the candidate fault source node, including:

[0130] Frequency components are extracted from the time-varying coupling matrix corresponding to the candidate fault source node. The relationship between the occurrence time of each frequency component and the emergence time of the candidate fault source node is identified. Frequency components whose occurrence time is earlier than the emergence time are marked as leading frequency components, and frequency components whose occurrence time is later than the emergence time are marked as response frequency components. A set of leading frequency components and a set of response frequency components are constructed.

[0131] Extract the standard leader frequency component set and standard response frequency component set for each fault mode from the preset fault characteristic frequency library. Calculate the set similarity between the leader frequency component set and the standard leader frequency component set as the leader matching degree. Calculate the set similarity between the response frequency component set and the standard response frequency component set as the response matching degree. Use the absolute value of the difference between the leader matching degree and the response matching degree as the matching distance.

[0132] Select the fault mode corresponding to the minimum matching distance from all fault modes, extract the node identifier information of the candidate fault source node as the location identifier, and output the location identifier and fault mode of the candidate fault source node.

[0133] After identifying candidate fault source nodes, it is necessary to further determine the specific fault mode type corresponding to the node in order to provide accurate fault diagnosis basis for maintenance personnel. All frequency components related to the candidate fault source node are extracted from the time-varying coupling matrix, including the dominant frequency and its weight coefficient recorded when the node participates in coupling enhancement as a row or column index. For each extracted frequency component, its first occurrence time is recorded, i.e., the earliest time point during the sliding time window traversal when the frequency component is identified as the dominant frequency of coupling enhancement. This time is compared with the emergence time of the candidate fault source node, which is taken from the earliest coupling enhancement time corresponding to when the node is confirmed as a causal emerging node in the emergence event graph.

[0134] The timing labeling rules for frequency components are as follows: if the first occurrence of a frequency component is earlier than the emergence time of the candidate fault source node, it is labeled as a leading frequency component and included in the leading frequency component set; if the first occurrence of a frequency component is later than the emergence time, it is labeled as a response frequency component and included in the response frequency component set. Leading frequency components reflect the early abnormal frequency characteristics formed in the sheath grounding circulation during the fault initiation stage, and are usually related to partial discharge excitation in the early stages of insulation degradation or harmonic injection caused by nonlinear changes in grounding resistance. Response frequency components reflect the second harmonic response characteristics generated by the coupling drive of surrounding nodes after fault evolution; their frequency components often have higher harmonic orders or wider frequency band distributions. In actual engineering scenarios, common fault modes in high-voltage cable gallery sheath grounding systems include single-point grounding faults, multi-point grounding faults, enhanced inductive coupling caused by insulation aging, and capacitive coupling anomalies caused by external interference. Different fault modes show significant differences in the combination of leading and response frequencies, which provides a physical basis for subsequent matching and classification.

[0135] The pre-defined fault characteristic frequency library is constructed from historical fault data and simulation models. Each fault mode record in the library contains a corresponding set of standard leader frequency components and a set of standard response frequency components. This serves as a fault mode index. The set similarity is calculated using a frequency component weighted overlap rate method: For a leading frequency component set and a standard leading frequency component set, each frequency component in the leading frequency component set is traversed. In the standard leading frequency component set, the standard component whose frequency value is closest to the leading frequency component is found. If the frequency difference between the two falls within a preset tolerance bandwidth, a successful match is considered. The square root of the product of the weight coefficients of the two frequency components is taken as the matching contribution value of that component pair. The sum of the contribution values ​​of all successfully matched component pairs is divided by the mean of the sum of the weight coefficients of all components in both sets to obtain the leading match degree. The response match degree is calculated in the same way, expressed as the weighted overlap rate between the response frequency component set and the standard response frequency component set.

[0136] Matching distance Defined as the absolute value of the difference between the leader matching degree and the response matching degree, i.e. The physical meaning of this definition is as follows: if the standard leader and standard response frequencies of a certain fault mode highly match the measured frequencies of the candidate nodes, then the leader matching degree and response matching degree are both close to 1, the absolute value of their difference is close to 0, and the matching distance is minimal, indicating that the fault mode best matches the measured signal characteristics. Conversely, if only the leader frequency matches but the response frequency does not, or if neither matches, then the absolute value of the difference is large, and the matching distance is large, indicating that the fault mode deviates significantly from the measured characteristics. By calculating the matching distance for each fault mode in the fault characteristic frequency database, a matching distance sequence is formed. Select the minimum matching distance from them. Corresponding Fault Mode Index The corresponding fault mode type is used as the final diagnostic result.

[0137] During the output phase, the node identification information of candidate fault source nodes is extracted as location identifiers. Node identification information typically includes the pipe gallery section number, grounding box number, and physical coordinates or mileage markers of the sheath grounding point. This information is already bound to the acquisition channels of each monitoring node during the original dataset construction phase and can be directly retrieved from the node metadata table. The location identifier and fault mode type together constitute the final output result, output in the form of a structured report. This report includes a description of the pipe gallery location of the candidate fault source node, its corresponding grounding section interval, the name of the identified fault mode (e.g., "single-point grounding impedance anomaly," "multi-point grounding circulation current superposition," etc.), and the specific frequency values ​​and weight information of the corresponding leader and response frequency components, for maintenance personnel's reference.

[0138] When multiple fault modes in the matching distance sequence have matching distance values ​​close to the minimum, an auxiliary judgment mechanism can be introduced: calculate the relative difference between the second smallest matching distance and the smallest matching distance. If the relative difference is lower than a preset identification threshold, all corresponding fault modes are included in the output result and marked as "suspected composite fault," prompting maintenance personnel to conduct further on-site verification. Furthermore, if the set of leading frequency components or the set of response frequency components is empty (i.e., all frequency components appear after the emergence time, or all frequency components appear before the emergence time), the corresponding matching degree calculation degenerates into one-sided matching. In this case, the matching distance is directly taken as the absolute value of the difference between the corresponding one-sided matching degree and 1, to ensure the integrity and robustness of the calculation process. Through the above complete frequency time-series labeling, set similarity calculation, and matching distance filtering process, accurate mapping from candidate fault source nodes to specific fault mode types is achieved, completing the entire AI localization output of high-voltage cable gallery sheath grounding loop fault sources.

[0139] A second aspect of the present invention provides an AI-based system for locating grounding loop current fault sources in high-voltage cable gallery sheaths, comprising:

[0140] The data acquisition unit is used to collect the circulating current signals of each node in the grounding system of the pipe gallery sheath and construct the original dataset;

[0141] The time-varying coupling unit is used to calculate the mutual information entropy of node pairs in the original dataset within a sliding time window. The time series gradient of the mutual information entropy value is used to identify the moment corresponding to the gradient peak as the coupling enhancement moment. The frequency component corresponding to the coupling enhancement moment is extracted to construct the time-varying coupling matrix.

[0142] Emergent graph units are used to count the number of coupling enhancements generated by each node in the time-varying coupling matrix. Clustering algorithms are used to identify nodes whose number of coupling enhancements is located at the cluster center of the maximum value as causal emergent nodes, and an emergent event graph is constructed.

[0143] Entropy reduction propagation unit is used to select the earliest causal emergence node from the emergence event graph as the starting node, calculate the propagation path entropy value based on the time-varying coupling matrix, and select the propagation path where the entropy value of subsequent nodes is less than that of the previous node to construct the entropy reduction propagation network.

[0144] The topological centrality unit is used to calculate the weighted sum of the out-degree centrality and betweenness centrality of each node in the entropy reduction propagation network as the topological centrality index, and the node corresponding to the maximum value of the topological centrality index is selected as the candidate fault source node.

[0145] The fault matching unit is used to calculate the matching distance between the frequency components of the candidate fault source node and the preset fault characteristic frequency library, select the fault mode corresponding to the minimum matching distance, and output the location identifier and fault mode type of the candidate fault source node.

[0146] A third aspect of the present invention provides an electronic device, comprising:

[0147] processor;

[0148] Memory used to store processor-executable instructions;

[0149] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0150] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0151] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0152] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for AI-based localization of grounding loop current fault sources in high-voltage cable gallery sheaths, characterized in that, include: Collect circulating current signals from each node of the pipe gallery sheath grounding system to construct the original dataset; The mutual information entropy of the node pairs in the original dataset is calculated within the sliding time window. The time series gradient of the mutual information entropy is used to identify the time corresponding to the gradient peak as the coupling enhancement time. The frequency component corresponding to the coupling enhancement time is extracted, and a time-varying coupling matrix is ​​constructed. The number of coupling enhancements generated by each node in the time-varying coupling matrix is ​​counted. The nodes whose number of coupling enhancements is located at the cluster center of the maximum value are identified by the clustering algorithm as causal emergent nodes, and an emergent event graph is constructed. The earliest causal emergence node in the emergence event graph is selected as the starting node. The propagation path entropy value is calculated based on the time-varying coupling matrix. The propagation path with the entropy value of subsequent nodes being less than that of the preceding node is selected to construct an entropy reduction propagation network. The weighted sum of the out-degree centrality and betweenness centrality of each node in the entropy reduction propagation network is calculated as the topological centrality index, and the node corresponding to the maximum value of the topological centrality index is selected as the candidate fault source node. Calculate the matching distance between the frequency components of the candidate fault source node and the preset fault characteristic frequency library, select the fault mode corresponding to the minimum matching distance, and output the location identifier and fault mode type of the candidate fault source node.

2. The method according to claim 1, characterized in that, The mutual information entropy of node pairs in the original dataset is calculated within a sliding time window. The time series gradient of the calculated mutual information entropy is used to identify the moment corresponding to the gradient peak as the coupling enhancement moment. The frequency components corresponding to the coupling enhancement moment are extracted, and the time-varying coupling matrix is ​​constructed, including: For any two nodes in the original dataset, extract the corresponding circulation signal segments of each node within the sliding time window, calculate the mutual information entropy value between the circulation signal segments, and arrange the mutual information entropy values ​​in chronological order to construct a time series of mutual information entropy values. The time series of mutual information entropy values ​​is subjected to first-order difference operation to obtain the time series gradient. The time series gradient is traversed to identify continuous positive gradient intervals. The gradient cumulative sum within the continuous positive gradient interval is calculated. The termination time of the interval corresponding to the maximum value of the gradient cumulative sum is selected as the time corresponding to the gradient peak. The time corresponding to the gradient peak is marked as the coupling enhancement time. Extract the circulating signal segment within the sliding time window where the coupling enhancement moment is located, perform Fourier transform on the circulating signal segment to obtain the spectral distribution, extract the dominant frequency corresponding to the maximum amplitude from the spectral distribution as the frequency component corresponding to the coupling enhancement moment, and calculate the amplitude ratio of the dominant frequency as the weighting coefficient of the frequency component. Using node pairs as matrix row and column indices, and the coupling enhancement time, frequency components, and weight coefficients as matrix element contents, a time-varying coupling matrix is ​​constructed, where each matrix element records the frequency components and weight coefficients of the corresponding node pair at the coupling enhancement time.

3. The method according to claim 1, characterized in that, The number of coupling enhancements generated by each node in the time-varying coupling matrix is ​​statistically analyzed. Nodes whose coupling enhancement numbers are located at the cluster centers of the maximum values ​​are identified as causal emerging nodes using a clustering algorithm. An emerging event graph is then constructed, including: Traverse the time-varying coupling matrix to extract all coupling enhancement moments of each node, calculate the time interval between adjacent coupling enhancement moments of each node, construct a probability distribution for the time interval and calculate the information entropy, and use the information entropy as the number of coupling enhancements generated by each node. Cluster the number of coupling enhancements generated by each node, extract the frequency components and weight coefficients corresponding to the nodes whose number of coupling enhancements is located at the cluster center of the maximum value, calculate the autocorrelation coefficient of the frequency component weighted sequence, use the ratio of the autocorrelation coefficient to the information entropy as the instability index, and select the nodes whose instability index is greater than the median as causal emergence nodes. Extract all coupling enhancement moments and frequency components corresponding to causal emergent nodes in the time-varying coupling matrix. Calculate the cross-correlation function for the coupling enhancement moments of any two causal emergent nodes. Identify the time delay corresponding to the peak value of the cross-correlation function to determine the propagation direction. Calculate the frequency component difference between the two causal emergent nodes as the propagation intensity. An emergent event graph is constructed by using causal emergent nodes as graph nodes, propagation direction as directed edges, propagation intensity as directed edge weights, and the earliest coupling enhancement moment as the temporal attribute of the graph node.

4. The method according to claim 3, characterized in that, For any two causal emergent nodes, calculate the cross-correlation function at the moment of enhanced coupling, identify the time delay corresponding to the peak of the cross-correlation function to determine the propagation direction, and calculate the frequency component difference between the two causal emergent nodes as the propagation strength, including: Fourier spectrum decomposition is performed on the frequency components of each causal emergence node to obtain the amplitude at the fundamental frequency and integer multiples of the power frequency harmonics. The harmonic order corresponding to the harmonic frequency with the largest amplitude at each coupling enhancement moment is identified as the dominant harmonic order. For any two causal emergence nodes, select coupling enhancement times with the same dominant harmonic order to construct a set of time pairs with the same frequency, calculate the time difference of each time pair in the set of time pairs with the same frequency, and construct a probability density distribution function as a cross-correlation function by statistically analyzing the occurrence frequency of the time difference. The time difference corresponding to the maximum probability density is extracted as the time delay corresponding to the peak of the cross-correlation function, and the propagation direction is determined based on the sign of the time delay. The dominant harmonic order of node pairs with propagation direction is extracted to construct a time series. The moment when the dominant harmonic order changes is identified as the harmonic transition moment. The standard deviation of the difference between the harmonic transition moments of the two nodes is calculated. The reciprocal of the standard deviation is normalized and used as the frequency component difference degree. The frequency component difference degree is used as the propagation intensity.

5. The method according to claim 1, characterized in that, The earliest causal emergence node in the emergent event graph is selected as the starting node. The propagation path entropy is calculated based on the time-varying coupling matrix. An entropy-reducing propagation network is constructed by selecting propagation paths where the entropy value of subsequent nodes is less than that of their predecessors. Extract the emergence time of all causal emergence nodes from the emergence event graph, and mark the causal emergence node with the earliest emergence time as the starting node; Extract all outgoing edges of the starting node and the subsequent nodes pointed to by the outgoing edges from the emergent event graph. Obtain the frequency components and weight coefficients of the starting node and each subsequent node in the time-varying coupling matrix. Calculate the weighted variance of the frequency components as the frequency domain dispersion. For all coupling enhancement moments in the time-varying coupling matrix, calculate the Shannon entropy of the time interval as the time-domain random entropy. Use the weighted sum of the frequency domain dispersion and the time-domain random entropy as the node entropy value. Calculate the node entropy value of the starting node and the node entropy value of each subsequent node. Filter subsequent nodes whose entropy value is less than that of the starting node, construct a propagation path by connecting the directed edges between the starting node and the filtered subsequent nodes and their weights, and record the entropy values ​​of the subsequent nodes in the propagation path as the propagation path entropy value. Using the selected subsequent nodes as new starting nodes, the process is repeated iteratively until there are no subsequent nodes whose entropy value is less than that of the preceding node. All propagation paths are then merged to construct an entropy reduction propagation network.

6. The method according to claim 1, characterized in that, The weighted sum of the out-degree centrality and betweenness centrality of each node in the entropy reduction propagation network is calculated as the topological centrality index. The node corresponding to the maximum value of the topological centrality index is selected as the candidate fault source node, including: Extract all outgoing edges of each node from the entropy reduction propagation network, identify the node entropy difference between the subsequent node pointed to by the outgoing edge and the current node, and count the number of outgoing edges whose node entropy difference exceeds the entropy reduction threshold as the out-degree centrality of each node. Extract the propagation paths between all node pairs from the entropy-reducing propagation network, select the paths in which the entropy values ​​between adjacent nodes continuously decrease as monotonic entropy-reducing paths, and count the number of times each node appears as an intermediate node in the monotonic entropy-reducing path as the betweenness centrality of each node. Extract the dominant harmonic order corresponding to each node in the entropy reduction propagation network, identify nodes whose dominant harmonic order is the same as the dominant harmonic order of the starting node as co-frequency nodes, set the first weight coefficient group for the out-degree centrality and betweenness centrality of co-frequency nodes, and set the second weight coefficient group for the out-degree centrality and betweenness centrality of non-co-frequency nodes. The out-degree centrality and betweenness centrality of each node are weighted and summed according to the corresponding weight coefficient group. The weighted sum is used as the topological centrality index of each node. The maximum value is extracted from the topological centrality index of all nodes, and the node corresponding to the maximum value of the topological centrality index is marked as a candidate fault source node.

7. The method according to claim 1, characterized in that, Calculate the matching distance between the frequency components of the candidate fault source node and a preset fault feature frequency database, select the fault mode corresponding to the minimum matching distance, and output the location identifier and fault mode type of the candidate fault source node, including: Frequency components are extracted from the time-varying coupling matrix corresponding to the candidate fault source node. The relationship between the occurrence time of each frequency component and the emergence time of the candidate fault source node is identified. Frequency components whose occurrence time is earlier than the emergence time are marked as leading frequency components, and frequency components whose occurrence time is later than the emergence time are marked as response frequency components. A set of leading frequency components and a set of response frequency components are constructed. Extract the standard leader frequency component set and standard response frequency component set for each fault mode from the preset fault characteristic frequency library. Calculate the set similarity between the leader frequency component set and the standard leader frequency component set as the leader matching degree. Calculate the set similarity between the response frequency component set and the standard response frequency component set as the response matching degree. Use the absolute value of the difference between the leader matching degree and the response matching degree as the matching distance. Select the fault mode corresponding to the minimum matching distance from all fault modes, extract the node identifier information of the candidate fault source node as the location identifier, and output the location identifier and fault mode of the candidate fault source node.

8. A high-voltage cable gallery sheath grounding loop current fault source AI localization system, used to implement the method as described in any one of claims 1-7, characterized in that, include: The data acquisition unit is used to collect the circulating current signals of each node in the grounding system of the pipe gallery sheath and construct the original dataset; The time-varying coupling unit is used to calculate the mutual information entropy of node pairs in the original dataset within a sliding time window. The time series gradient of the mutual information entropy value is used to identify the moment corresponding to the gradient peak as the coupling enhancement moment. The frequency component corresponding to the coupling enhancement moment is extracted to construct the time-varying coupling matrix. Emergent graph units are used to count the number of coupling enhancements generated by each node in the time-varying coupling matrix. Clustering algorithms are used to identify nodes whose number of coupling enhancements is located at the cluster center of the maximum value as causal emergent nodes, and an emergent event graph is constructed. Entropy reduction propagation unit is used to select the earliest causal emergence node from the emergence event graph as the starting node, calculate the propagation path entropy value based on the time-varying coupling matrix, and select the propagation path where the entropy value of subsequent nodes is less than that of the previous node to construct the entropy reduction propagation network. The topological centrality unit is used to calculate the weighted sum of the out-degree centrality and betweenness centrality of each node in the entropy reduction propagation network as the topological centrality index, and the node corresponding to the maximum value of the topological centrality index is selected as the candidate fault source node. The fault matching unit is used to calculate the matching distance between the frequency components of the candidate fault source node and the preset fault characteristic frequency library, select the fault mode corresponding to the minimum matching distance, and output the location identifier and fault mode type of the candidate fault source node.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images