A Fault Diagnosis Method for Smart Switchgear Based on Multimodal Data Analysis

By analyzing multimodal data, identifying steady-state data segments, eliminating redundant modes, constructing associated transmission paths, and combining physical transmission mechanisms and joint interpretability to screen fault characteristics, the problem of redundant abnormal information interference in smart switchgear was solved, and high-precision fault source location and diagnosis were achieved.

CN123069028APending Publication Date: 2026-11-10SHANDONG HUAXIN ELECTRIC
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611249140.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-18
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing multimodal fault diagnosis methods are susceptible to electromagnetic interference and data drift in smart switchgear, resulting in excessive redundant abnormal information, making it difficult to accurately identify real faults and tracing fault sources with low accuracy. This fails to meet the requirements of high reliability and high accuracy in intelligent operation and maintenance.

Method used

By analyzing multimodal data, we can identify steady-state data segments, eliminate redundant modes, construct associated transmission paths, and filter effective fault features by combining physical transmission mechanisms and joint explanatory power to trace the fault source in reverse.

Benefits of technology

It improves the accuracy and reliability of fault diagnosis, reduces missed and false diagnoses, enables precise fault location in complex operating scenarios of switchgear, and enhances equipment safety and stability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123069028A_ABST
    Figure CN123069028A_ABST
Patent Text Reader

Abstract

This invention discloses a fault diagnosis method for intelligent switchgear based on multimodal data analysis, belonging to the field of fault diagnosis technology. The method includes: collecting multimodal operating data, identifying steady-state data segments and determining a steady-state baseline; normalizing the data, dividing it into redundant groups and filtering effective modes to form a non-redundant subset; extracting time-series features to construct an association matrix, filtering association relationships that conform to physical mechanisms to form a complete transmission path; calculating the fault contribution of each node, determining the fault source feature nodes through reverse tracing, matching corresponding physical quantities with monitoring points, and completing fault location. This invention achieves accurate fault diagnosis of switchgear, can eliminate the influence of interference signals, improve diagnostic accuracy and efficiency, and can complete the entire process of automated diagnosis without complex operations.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, specifically to a fault diagnosis method for intelligent switchgear based on multimodal data analysis. Background Technology

[0002] With the development of smart grids and digital operation and maintenance technologies, smart switchgear has widely adopted multimodal data acquisition and online analysis technologies. Through comprehensive processing of multi-dimensional data such as electrical parameters, environmental parameters, and status parameters, it enables equipment operation status monitoring and fault diagnosis. Existing multimodal fault diagnosis methods typically perform independent discrimination or simple fusion processing of each modality's data, identifying anomalies and locating faults based on parameter exceeding limits, feature correlation, or trend changes.

[0003] In actual operation, multimodal monitoring data of switchgear is easily affected by factors such as electromagnetic interference, short-term load fluctuations, and data acquisition drift, generating a large amount of discrete pseudo-anomaly information. Existing diagnostic methods do not effectively screen multimodal anomaly information, cannot determine the minimum data combination with sufficient interpretability from redundant anomaly data, and lack a verification mechanism for the internal logical consistency of anomaly combinations. This makes diagnostic results susceptible to interference from non-fault anomalies, resulting in insufficient ability to distinguish between true and false faults and low accuracy in tracing fault sources, making it difficult to meet the high reliability and high accuracy intelligent operation and maintenance requirements of smart switchgear. Summary of the Invention

[0004] The purpose of this invention is to provide a fault diagnosis method for intelligent switchgear based on multimodal data analysis, so as to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A fault diagnosis method for intelligent switchgear based on multimodal data analysis includes the following steps: S1. Collect multi-modal operation data of the smart switchgear, analyze the multi-modal operation data, identify steady-state data segments and determine steady-state reference values; calculate the relative deviation of the current data from the steady-state reference value, mark data whose relative deviation exceeds the preset limit as abnormal modes, and form a complete set of abnormal modes; S2. Normalize the entire set of anomalous modes, calculate the trend consistency coefficient between anomalous modes, and divide redundant mode groups according to the trend consistency coefficient; based on the signal fluctuation stability of modes within each group, select and retain the mode with the highest stability, and remove the remaining redundant modes to form a subset of anomalous modes without redundancy. S3. Extract the temporal features of the non-redundant abnormal mode subset, construct the potential correlation between modes and eliminate false correlations to obtain the correlation propagation path; the temporal features include the abnormal start time and the abnormal duration; when the non-redundant abnormal mode subset contains only a single mode, it is determined to be an interference signal; otherwise, the joint explanatory power is calculated by combining the stability of each mode with the preset weight. When the joint explanatory power reaches the preset quantization threshold and there is only one correlation propagation path, the corresponding non-redundant abnormal mode subset is determined as a valid fault feature set; S4. Tracing back along the unique associated propagation path of the effective fault feature set, calculating the fault contribution of each node, selecting the node with the smallest fault contribution as the fault source feature node, locating the fault source by combining the fault propagation relationship, and outputting the diagnostic results containing fault source information.

[0006] Furthermore, S1 includes the following: Multimodal operation data of the smart switchgear is collected according to a preset cycle to form a continuous time-series sampling dataset; the multimodal operation data refers to different types of monitoring data that can reflect the operating status of the smart switchgear. A sliding window approach is used to identify the steady state of time-series sampled datasets. When the range of data within the window is less than a preset steady-state range threshold, and the change in values ​​of adjacent sampling points is less than a preset single-step change threshold, the corresponding window data is determined to be a steady-state data segment. The steady-state range threshold is determined based on the historical fluctuation amplitude statistics of various modal data under normal operating conditions, and the single-step change threshold is determined based on engineering experience of the maximum allowable fluctuation of adjacent sampling points. The steady-state baseline value of the multimodal operating data is determined based on the steady-state data segment. The steady-state baseline value is obtained by averaging the measured values ​​of all valid sampling points within the steady-state data segment. The relative deviation of the multimodal operating data from the steady-state baseline value at the current moment is calculated. The relative deviation is obtained by dividing the absolute value of the difference between the current measured value and the steady-state baseline value by the steady-state baseline value and converting it into a percentage. The preset deviation limit is determined based on the statistical distribution of the deviation of each modal data under normal operating conditions or industry standards. Modal data categories whose relative deviation exceeds the preset deviation limit are marked as abnormal categories. All abnormal categories are summarized to form a complete abnormal modal set.

[0007] Furthermore, S2 involves dividing redundant mode groups and filtering out anomalous modes, specifically including: Two or more abnormal modes with a trend consistency coefficient greater than the trend consistency judgment threshold are classified into the same redundant mode group. Calculate the deviation coefficients of various abnormal modes within each redundant mode group. The deviation coefficients are used to measure the stability of the modal signal and the sensitivity of fault characterization. They are obtained by dividing the standard deviation of the normalized deviation characteristics of various abnormal modes by the mean. Within each redundant mode group, retain the mode with the smallest deviation coefficient and eliminate the remaining redundant modes; Abnormal modes with a trend consistency coefficient less than or equal to the trend consistency judgment threshold are judged as independent of each other and directly retained; after all abnormal modes are screened, all retained abnormal modes together constitute a subset of abnormal modes without redundancy.

[0008] Furthermore, S2 performs normalization processing and calculates the trend consistency coefficient for the entire set of abnormal modes, specifically including: Normalize the time series deviation sequences corresponding to various anomalous modes in the full set of anomalous modes, and map all types of time series deviation sequences to a dimensionless interval to eliminate the dimensional differences between different modes. The normalization process is obtained by subtracting the minimum value of the sequence from the measured value of the time series deviation from various abnormal modes, and then dividing by the difference between the maximum and minimum values ​​of the sequence. The trend consistency coefficient between any two abnormal modes is calculated based on the normalized deviation characteristics. The magnitude of the coefficient reflects the degree of synchronization of the deviation trends of the two abnormal modes. The trend consistency judgment threshold is determined based on the statistical distribution of the multimodal correlation coefficient in the historical operation data of the switchgear. The corresponding quantile of the trend consistency coefficient distribution under normal operating conditions is taken as the basis for the threshold value.

[0009] Furthermore, S3 includes: Extract the temporal feature parameters corresponding to the abnormal modes, including the abnormal start time and the abnormal duration; construct a mode association matching matrix based on the temporal feature parameters to identify the association relationship between modes; determine the potential association relationship based on the difference between the abnormal start time and the overlap rate of the abnormal duration, and determine the propagation direction corresponding to the association relationship by combining the fault propagation time. The preset time threshold is determined based on the fault propagation time parameter of the switchgear, and the overlap rate of abnormal duration is determined based on the extreme value ratio of the duration of the two abnormal modes.

[0010] Furthermore, S3 includes: Using abnormal modes as nodes and potential correlations as edges, correlation edge filtering is implemented by combining the physical quantity attributes corresponding to the modes; correlation edges that do not conform to the physical transmission mechanism are removed, and correlation edges that conform to the fault transmission law are retained to form the correlation transmission relationship between modes; Based on whether there is a direct causal transmission relationship between physical quantities, the associated edge judgment and elimination operation is carried out; the earliest abnormal mode at the start time and without other mode pointing is determined as the initial node, and the associated transmission path is formed by the initial node along the retained associated edges. All associated transmission paths are obtained by traversing all nodes and edges; a single abnormal mode does not form an associated transmission path and is judged as an interference signal.

[0011] Furthermore, S3 includes: The joint explanatory power is calculated based on the stability of the abnormal modes and the weighting coefficients. The weighting coefficients are determined based on the sensitivity of the modes to the fault source and are normalized. The sum of the weighting coefficients of all modes remains consistent. The joint explanatory power threshold is determined based on the statistical distribution of the effective feature set in historical fault data. The number of associated transmission paths and the joint explanatory power are judged in a dual manner. When the joint explanatory power reaches the corresponding threshold and the number of associated transmission paths is unique, the corresponding subset is determined as the effective fault feature set. The subset that does not meet the conditions is judged as interference signal and is removed.

[0012] Furthermore, S4 includes: A reverse tracing operation is performed along the unique associated propagation path in the set of valid fault features. Starting from the end abnormal mode node of the associated propagation path, all nodes included in the path are traversed sequentially in the direction of the initial abnormal mode node. During the tracing process, the physical quantity type corresponding to the abnormal mode and the abnormal propagation order are recorded sequentially to form a reverse tracing sequence opposite to the direction of the forward propagation path. Calculate the fault contribution degree corresponding to the abnormal mode in the reverse tracing sequence. The fault contribution degree is determined based on the signal stability of the abnormal mode and the hierarchical index of the abnormal mode node in the associated propagation path. Sort the fault contribution degrees of the abnormal modes in the reverse tracing sequence, determine the abnormal mode node with the smallest fault contribution degree, and identify this abnormal mode node as the feature node corresponding to the fault source.

[0013] Furthermore, S4 includes: A one-to-one correspondence between abnormal mode types and physical quantity types is established in advance to match fault source feature nodes to the corresponding physical quantity types; a correspondence between physical quantity types and switchgear monitoring points is established in advance to match physical quantity types to the corresponding monitoring points. The equipment location corresponding to the monitoring point is determined as the fault source location of the smart switchgear. A fault diagnosis conclusion is generated based on the fault source location and the corresponding abnormal mode type. The fault diagnosis conclusion includes the fault source location, abnormal mode type, and fault propagation process. The fault source location information and the fault diagnosis conclusion are summarized to form a complete diagnosis result.

[0014] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention achieves anomaly identification through multimodal operational data collaborative analysis, employs a sliding window combined with dual thresholds to determine steady-state data segments, improves the accuracy of steady-state benchmark value calculation, and avoids misjudgments caused by operating condition fluctuations; through time-series deviation sequence normalization and trend consistency coefficient analysis, it accurately classifies redundant modes from the same source, combines deviation coefficients to screen stable characterization signals, eliminates interference information, and improves the effectiveness of abnormal modes; based on the anomaly start time and duration, it constructs correlation relationships and combines physical transmission mechanisms to eliminate false correlations, forming real fault transmission paths, avoiding meaningless correlation matching, and improving path reliability; through joint interpretability and path uniqueness dual judgment to screen effective fault feature sets, it distinguishes effective features from interference signals, enhancing the stability of fault judgment; it uses reverse tracing combined with fault contribution ranking to locate the fault source, matches physical quantity types and monitoring points to achieve precise positioning, and completes the entire process diagnosis without manual intervention; the overall solution takes into account data reliability, logical rigor, and positioning accuracy, can adapt to complex switchgear operating scenarios, reduces missed and false judgments, improves fault diagnosis efficiency and accuracy, and provides reliable technical support for the safe and stable operation of equipment. Attached Figure Description

[0015] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a smart switchgear fault diagnosis method based on multimodal data analysis according to the present invention. Detailed Implementation

[0016] 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.

[0017] Please see Figure 1 The present invention provides the following technical solution: A fault diagnosis method for intelligent switchgear based on multimodal data analysis includes the following steps: S1. Collect multi-modal operation data of the smart switchgear, analyze the multi-modal operation data, identify steady-state data segments and determine steady-state reference values; calculate the relative deviation of the current data from the steady-state reference value, mark data whose relative deviation exceeds the preset limit as abnormal modes, and form a complete set of abnormal modes; S2. Normalize the entire set of anomalous modes, calculate the trend consistency coefficient between anomalous modes, and divide redundant mode groups according to the trend consistency coefficient; based on the signal fluctuation stability of modes within each group, select and retain the mode with the highest stability, and remove the remaining redundant modes to form a subset of anomalous modes without redundancy. S3. Extract the temporal features of the non-redundant abnormal mode subset, construct the potential correlation between modes and eliminate false correlations to obtain the correlation propagation path; the temporal features include the abnormal start time and the abnormal duration; when the non-redundant abnormal mode subset contains only a single mode, it is determined to be an interference signal; otherwise, the joint explanatory power is calculated by combining the stability of each mode with the preset weight. When the joint explanatory power reaches the preset quantization threshold and there is only one correlation propagation path, the corresponding non-redundant abnormal mode subset is determined as a valid fault feature set; S4. Tracing back along the unique associated propagation path of the effective fault feature set, calculating the fault contribution of each node, selecting the node with the smallest fault contribution as the fault source feature node, locating the fault source by combining the fault propagation relationship, and outputting the diagnostic results containing fault source information.

[0018] S1 includes the following: Multimodal operation data of the smart switchgear is collected according to a preset cycle to form a continuous time-series sampling dataset; the multimodal operation data refers to different types of monitoring data that can reflect the operating status of the smart switchgear. A sliding window approach is used to identify the steady state of time-series sampled datasets. When the range of data within the window is less than a preset steady-state range threshold, and the change in values ​​of adjacent sampling points is less than a preset single-step change threshold, the corresponding window data is determined to be a steady-state data segment. The steady-state range threshold is determined based on the historical fluctuation amplitude statistics of various modal data under normal operating conditions, and the single-step change threshold is determined based on engineering experience of the maximum allowable fluctuation of adjacent sampling points. The steady-state baseline value of the multimodal operating data is determined based on the steady-state data segment. The steady-state baseline value is obtained by averaging the measured values ​​of all valid sampling points within the steady-state data segment. The relative deviation of the multimodal operating data from the steady-state baseline value at the current moment is calculated. The relative deviation is obtained by dividing the absolute value of the difference between the current measured value and the steady-state baseline value by the steady-state baseline value and converting it into a percentage. The preset deviation limit is determined based on the statistical distribution of the deviation of each modal data under normal operating conditions or industry standards. Modal data categories whose relative deviation exceeds the preset deviation limit are marked as abnormal categories. All abnormal categories are summarized to form a complete abnormal modal set.

[0019] In this embodiment, multimodal operation data of the smart switchgear is collected according to a preset period to form a continuous time-series sampling dataset. The multimodal operation data refers to different types of monitoring data that can reflect the operating status of the smart switchgear, specifically including electrical operation data, equipment status data, and environmental monitoring data. Steady-state identification is carried out on the time-series sampling dataset using a sliding window method. It is determined whether the range of the data within the window is less than a preset steady-state range threshold, and whether the numerical change of adjacent sampling points is less than a preset single-step change threshold. The window data that meets both conditions is the steady-state data segment. The steady-state range threshold is determined based on the historical fluctuation amplitude of various modal data under normal operating conditions, and the single-step change threshold is determined based on engineering experience of the maximum allowable fluctuation of adjacent sampling points. The steady-state baseline value of multimodal operating data is calculated based on the steady-state data segment. The calculation formula is: Xb i =(1 / N)∑ k∈[1,N] X ik , among which, Xb i X is the steady-state reference value for the i-th modal data, with dimensions consistent with the corresponding category data; ik is the measured value of the kth sampling point of the i-th modal data; N is the total number of valid sampling points in the steady-state data segment; The deviation of the current multimodal operating data from the corresponding steady-state reference value is calculated using the following formula: P i =|(Xr i -Xb i ) / Xb i |×100%, where P i Xr represents the relative deviation of the i-th modal data, dimensionless, expressed as a percentage; the preset deviation limit is determined based on the statistical distribution of the deviation under normal operating conditions or industry standards for each modal data; i The current measured value of the i-th modal data is used; the relative deviation is compared with the preset deviation limit, and the modal data categories that exceed the preset deviation limit are marked as abnormal categories. All abnormal categories are summarized to form a full abnormal modal set.

[0020] S2 involves dividing redundant mode groups and filtering out anomalous modes, specifically including: Two or more abnormal modes with a trend consistency coefficient greater than the trend consistency judgment threshold are classified into the same redundant mode group. Calculate the deviation coefficients of various abnormal modes within each redundant mode group. The deviation coefficients are used to measure the stability of the modal signal and the sensitivity of fault characterization. They are obtained by dividing the standard deviation of the normalized deviation characteristics of various abnormal modes by the mean. Within each redundant mode group, retain the mode with the smallest deviation coefficient and eliminate the remaining redundant modes; Abnormal modes with a trend consistency coefficient less than or equal to the trend consistency judgment threshold are judged as independent of each other and directly retained; after all abnormal modes are screened, all retained abnormal modes together constitute a subset of abnormal modes without redundancy.

[0021] S2 performs normalization processing and calculates the trend consistency coefficient for the entire set of anomalous modes, specifically including: Normalize the time series deviation sequences corresponding to various anomalous modes in the full set of anomalous modes, and map all types of time series deviation sequences to a dimensionless interval to eliminate the dimensional differences between different modes. The normalization process is obtained by subtracting the minimum value of the sequence from the measured value of the time series deviation from various abnormal modes, and then dividing by the difference between the maximum and minimum values ​​of the sequence. The trend consistency coefficient between any two abnormal modes is calculated based on the normalized deviation characteristics. The magnitude of the coefficient reflects the degree of synchronization of the deviation trends of the two abnormal modes. The trend consistency judgment threshold is determined based on the statistical distribution of the multimodal correlation coefficient in the historical operation data of the switchgear. The corresponding quantile of the trend consistency coefficient distribution under normal operating conditions is taken as the basis for the threshold value.

[0022] In this embodiment, the time-series deviation sequences corresponding to various anomalous modes in the full set of anomalous modes are normalized, mapping all types of time-series deviation sequences to a dimensionless interval of 0 to 1, eliminating dimensional differences between different modes. The normalization calculation formula is: U it =(P it -P i_min ) / (P i_max -P i_min ), where U it P represents the normalized deviation characteristic of the i-th anomalous mode at the t-th sampling time, which is dimensionless; it P represents the relative deviation of the i-th anomalous mode at the t-th sampling time; i_min P is the minimum value in the time series deviation sequence of the i-th type of anomalous mode; i_max It represents the maximum value in the time series deviation sequence of the i-th type of anomalous mode; The trend consistency coefficient between anomalous modes is calculated based on the normalized deviation characteristics, and the formula is as follows: , Among them, R ij U is the trend consistency coefficient between the i-th and j-th anomalous modes, dimensionless, ranging from 0 to 1, and its value reflects the degree of synchronization between the two anomalous modes in deviating from the trend; T is the total number of sampling times within the anomalous period; U jt The deviation characteristics of the j-th anomalous mode after normalization at the t-th sampling time; The trend consistency judgment threshold is determined based on the statistical distribution of multimodal correlation coefficients in the historical operation data of the switchgear. The 95th quantile of the trend consistency coefficient distribution under normal operating conditions is taken as the basis for the threshold value. Multiple abnormal modes with trend consistency coefficients greater than the trend consistency judgment threshold are judged as having information redundancy. These modes are caused by the same fault source and the information is highly overlapping. Only one type needs to be retained as a representative of the redundancy group. Calculate the deviation coefficient of the abnormal modes within each group. The magnitude of the deviation coefficient measures the stability of the modal signal and the sensitivity of fault characterization. The formula for calculating the deviation coefficient is: Di = σUi / μUi, where Di is the deviation coefficient of the i-th type of abnormal mode, which is dimensionless; σUi is the standard deviation of the normalized deviation feature of the i-th type of abnormal mode; and μUi is the mean of the normalized deviation feature of the i-th type of abnormal mode. In the redundant mode group, retain the mode with the smallest deviation coefficient and remove the remaining redundant modes. Extract the temporal feature parameters corresponding to the abnormal modes, including the abnormal start time and the abnormal duration; construct a mode association matching matrix based on the temporal feature parameters to identify the association relationship between modes; determine the potential association relationship based on the difference between the abnormal start time and the overlap rate of the abnormal duration, and determine the propagation direction corresponding to the association relationship by combining the fault propagation time. The preset time threshold is determined based on the fault propagation time parameter of the switchgear, and the overlap rate of abnormal duration is determined based on the extreme value ratio of the duration of the two abnormal modes.

[0023] Using abnormal modes as nodes and potential correlations as edges, correlation edge filtering is implemented by combining the physical quantity attributes corresponding to the modes; correlation edges that do not conform to the physical transmission mechanism are removed, and correlation edges that conform to the fault transmission law are retained to form the correlation transmission relationship between modes; Based on whether there is a direct causal transmission relationship between physical quantities, the associated edge judgment and elimination operation is carried out; the earliest abnormal mode at the start time and without other mode pointing is determined as the initial node, and the associated transmission path is formed by the initial node along the retained associated edges. All associated transmission paths are obtained by traversing all nodes and edges; a single abnormal mode does not form an associated transmission path and is judged as an interference signal.

[0024] The joint explanatory power is calculated based on the stability of the abnormal modes and the weighting coefficients. The weighting coefficients are determined based on the sensitivity of the modes to the fault source and are normalized. The sum of the weighting coefficients of all modes remains consistent. The joint explanatory power threshold is determined based on the statistical distribution of the effective feature set in historical fault data. The number of associated transmission paths and the joint explanatory power are judged in a dual manner. When the joint explanatory power reaches the corresponding threshold and the number of associated transmission paths is unique, the corresponding subset is determined as the effective fault feature set. The subset that does not meet the conditions is judged as interference signal and is removed.

[0025] In this embodiment, temporal feature parameters corresponding to each type of abnormal mode are extracted. These temporal feature parameters include the abnormal start time and the abnormal duration. A modal association matching matrix is ​​constructed based on the temporal feature parameters to complete the preliminary identification of the correlation between modes. When the difference between the abnormal start times of two types of abnormal modes is less than a preset time threshold, and the overlap rate of the abnormal duration is greater than a preset overlap threshold, it is determined that there is a potential correlation between the two types of abnormal modes, and the corresponding position in the association matching matrix is ​​marked as 1; otherwise, it is marked as 0. For two types of abnormal modes with potential correlation, the propagation direction is determined according to the time of abnormal occurrence, with the mode that occurs first pointing to the mode that occurs later. The preset time threshold is determined based on the typical time constant of switchgear fault propagation, and the average modal propagation time in historical fault data is taken as the basis for the preset time threshold value. The abnormal duration overlap rate is calculated by the formula: C ij =min(T i ,T j ) / max(T i ,T j ),C ij Let be the overlap rate of the abnormal duration between the i-th and j-th anomalous modes, dimensionless, and ranging from 0 to 1; Ti is the abnormal duration of the i-th anomalous mode; Tj is the abnormal duration of the j-th anomalous mode. Using various anomalous modes as nodes and the potential associations marked in the association matching matrix as edges, the association edges are screened according to the conventional logic of fault propagation, taking into account the physical quantity attributes corresponding to each mode. During the screening process, false association edges that do not conform to the physical mechanism are removed, while true association edges that conform to the fault propagation law are retained, ultimately forming the association propagation relationship between modes. If the physical quantities corresponding to two anomalous modes have no direct causal propagation relationship, the corresponding association edge is determined to be a false association edge and is removed. For example, current anomalies can directly cause temperature anomalies and vibration anomalies, but there is no direct physical causal relationship between temperature anomalies and partial discharge anomalies, so the corresponding association edge is determined to be a false association edge and is removed. The initial anomalous mode node corresponds to the earliest anomalous start time and has no other anomalous modes associated with it; the association propagation path starts from the initial anomalous mode node and propagates sequentially along the real associated edges to other anomalous mode nodes to form a complete link; all nodes and edges in the association propagation relationship are traversed to obtain all association propagation paths; when the non-redundant anomalous mode subset contains only a single anomalous mode, an association propagation path cannot be formed, and such modes are uniformly judged as interference signals; The joint explanatory power of a subset of non-redundant anomalous modes is calculated using the formula: J = ∑ i∈[1,H] [1 / (1+D i )]×W iWhere H is the total number of modes in the subset of non-redundant anomalous modes; Wi is the weight coefficient of the i-th type of anomalous mode, which is dimensionless and its value is determined according to the degree of correlation between the mode and the fault source, and the sum of the weight coefficients of all modes is 1; the weight coefficient is determined based on the sensitivity of different modes to the representation of the fault source in historical fault samples, and is calculated by information gain or fault correlation degree and normalized, and the weights of modes of the same category are kept consistent. A preset quantization threshold for joint explanatory power is set. This threshold is determined based on the statistical distribution of the joint explanatory power of valid fault feature sets in historical fault data, with the minimum value of the joint explanatory power of valid fault feature sets used as the basis for setting the preset quantization threshold. A dual judgment is made on the number of associated transmission paths and the joint explanatory power. When the joint explanatory power is greater than or equal to the preset quantization threshold, and there is only one associated transmission path, the corresponding subset of non-redundant abnormal modes is determined as a valid fault feature set. When the joint explanatory power is less than the preset quantization threshold, or there are two or more associated transmission paths, the corresponding subset of non-redundant abnormal modes is determined as interference signals and is removed. The joint explanatory power characterizes the overall signal stability and fault characterization reliability of the non-redundant abnormal mode subset. A higher value indicates a more accurate characterization of the switchgear's abnormal state and a more stable signal. Determining the uniqueness of associated transmission paths avoids fault tracing bias caused by multi-path interference, ensures the accuracy of subsequent reverse fault source tracing, and improves the overall reliability of fault diagnosis.

[0026] S4 includes the following: A reverse tracing operation is performed along the unique associated propagation path in the set of valid fault features. Starting from the end abnormal mode node of the associated propagation path, all nodes included in the path are traversed sequentially in the direction of the initial abnormal mode node. During the tracing process, the physical quantity type corresponding to the abnormal mode and the abnormal propagation order are recorded sequentially to form a reverse tracing sequence opposite to the direction of the forward propagation path. Calculate the fault contribution degree corresponding to the abnormal mode in the reverse tracing sequence. The fault contribution degree is determined based on the signal stability of the abnormal mode and the hierarchical index of the abnormal mode node in the associated propagation path. Sort the fault contribution degrees of the abnormal modes in the reverse tracing sequence, determine the abnormal mode node with the smallest fault contribution degree, and identify this abnormal mode node as the feature node corresponding to the fault source.

[0027] A one-to-one correspondence between abnormal mode types and physical quantity types is established in advance to match fault source feature nodes to the corresponding physical quantity types; a correspondence between physical quantity types and switchgear monitoring points is established in advance to match physical quantity types to the corresponding monitoring points. The equipment location corresponding to the monitoring point is determined as the fault source location of the smart switchgear. A fault diagnosis conclusion is generated based on the fault source location and the corresponding abnormal mode type. The fault diagnosis conclusion includes the fault source location, abnormal mode type, and fault propagation process. The fault source location information and the fault diagnosis conclusion are summarized to form a complete diagnosis result.

[0028] In this embodiment, a reverse tracing operation is performed along the unique associated transmission path in the effective fault feature set. Starting from the end abnormal mode node of the associated transmission path, all nodes included in the path are traversed sequentially in the direction of the initial abnormal mode node. During the tracing process, the physical quantity type and abnormal transmission order corresponding to each abnormal mode are recorded sequentially to form a reverse tracing sequence opposite to the forward transmission path direction. The fault contribution rate corresponding to each abnormal mode in the reverse tracing sequence is calculated using the formula F. i =D i ×L i , of which F i L represents the fault contribution corresponding to the i-th type of abnormal mode, which is dimensionless; i Let be the hierarchical index of the i-th type of anomalous modality node in the associated propagation path. The hierarchical index is a positive integer, which is incremented sequentially starting from the initial anomalous modality node. The hierarchical index of the initial anomalous modality node is 1. The hierarchical index is incremented by 1 for each node passed along the forward propagation direction. The fault contribution of various abnormal modes in the reverse tracing sequence is sorted to determine the abnormal mode node with the smallest fault contribution value. Among them, the smaller the deviation coefficient Di, the higher the stability of the abnormal signal; the smaller the hierarchical number Li, the closer the abnormal mode node is to the fault source; the node with the smallest fault contribution Fi is the node with the most stable signal and the closest to the fault source. Combined with the physical mechanism of switchgear fault propagation, such abnormal mode nodes are identified as characteristic nodes corresponding to the fault source. Based on the essential attributes of abnormal modes, matching rules are established with physical quantity types. The matching rules are determined according to the abnormal characteristics of switchgear operation reflected by the abnormal modes, that is, the abnormal mode type is consistent with the physical quantity type that caused the abnormality. Combining the matching rules, the feature nodes corresponding to the fault source are matched with the corresponding physical quantity types to clarify the physical quantity types corresponding to the feature nodes. For example, current-type abnormal modes correspond to current-type physical quantities in electrical operation data, temperature-type abnormal modes correspond to temperature-type physical quantities in equipment status data, vibration-type abnormal modes correspond to vibration-type physical quantities in equipment status data, and partial discharge-type abnormal modes correspond to partial discharge-type physical quantities in electrical operation data. Based on the monitoring purpose of the physical quantity type, a matching rule is established with the monitoring point of the switchgear. The matching rule is determined according to the functional part of the switchgear equipment monitored by the physical quantity, that is, the physical quantity type corresponds to the switchgear equipment part used to monitor the physical quantity. In combination with the matching rule, the determined physical quantity type is matched with the corresponding switchgear monitoring point to clarify the monitoring point corresponding to the fault source. For example, current physical quantities correspond to monitoring points such as contacts and busbars in the switchgear that are responsible for current conduction; temperature physical quantities correspond to monitoring points such as moving and stationary contacts and terminals in the switchgear that are prone to abnormal heating; vibration physical quantities correspond to monitoring points such as operating mechanisms and cabinet frames in the switchgear that are prone to vibration; and partial discharge physical quantities correspond to monitoring points such as insulating bushings and busbar compartments in the switchgear that are prone to partial discharge. The equipment location corresponding to the monitoring point is identified as the fault source location of the smart switchgear. A fault diagnosis conclusion is generated based on the fault source location and the corresponding abnormal mode type. The fault diagnosis conclusion includes the fault source location, abnormal mode type, and fault propagation process. The fault source location information and the fault diagnosis conclusion are summarized to form a complete diagnosis result, thus completing the fault tracing and diagnosis process of the smart switchgear. The fault diagnosis result can be directly used for on-site operation and maintenance reference, providing accurate guidance for fault investigation and handling.

[0029] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0030] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A fault diagnosis method for intelligent switchgear based on multimodal data analysis, characterized in that: The method includes the following steps: S1. Collect multi-modal operation data of the smart switchgear, analyze the multi-modal operation data, identify steady-state data segments and determine steady-state reference values; calculate the relative deviation of the current data from the steady-state reference value, mark data whose relative deviation exceeds the preset limit as abnormal modes, and form a complete set of abnormal modes; S2. Normalize the entire set of anomalous modes, calculate the trend consistency coefficient between anomalous modes, and divide redundant mode groups according to the trend consistency coefficient; based on the signal fluctuation stability of modes within each group, select and retain the mode with the highest stability, and remove the remaining redundant modes to form a subset of anomalous modes without redundancy. S3. Extract the temporal features of the non-redundant abnormal mode subset, construct the potential correlation between modes and eliminate false correlations to obtain the correlation propagation path; the temporal features include the abnormal start time and the abnormal duration; when the non-redundant abnormal mode subset contains only a single mode, it is determined to be an interference signal; otherwise, the joint explanatory power is calculated by combining the stability of each mode with the preset weight. When the joint explanatory power reaches the preset quantization threshold and there is only one correlation propagation path, the corresponding non-redundant abnormal mode subset is determined as a valid fault feature set; S4. Tracing back along the unique associated propagation path of the effective fault feature set, calculating the fault contribution of each node, selecting the node with the smallest fault contribution as the fault source feature node, locating the fault source by combining the fault propagation relationship, and outputting the diagnostic results containing fault source information.

2. The intelligent switchgear fault diagnosis method based on multimodal data analysis according to claim 1, characterized in that: S1 includes: Multimodal operation data of the smart switchgear is collected according to a preset cycle to form a continuous time-series sampling dataset; the multimodal operation data refers to different types of monitoring data that can reflect the operating status of the smart switchgear. A sliding window approach is used to identify the steady state of time-series sampled datasets. When the range of data within the window is less than a preset steady-state range threshold, and the change in values ​​of adjacent sampling points is less than a preset single-step change threshold, the corresponding window data is determined to be a steady-state data segment. The steady-state range threshold is determined based on the historical fluctuation amplitude statistics of various modal data under normal operating conditions, and the single-step change threshold is determined based on engineering experience of the maximum allowable fluctuation of adjacent sampling points. The steady-state baseline value of the multimodal operating data is determined based on the steady-state data segment. The steady-state baseline value is obtained by averaging the measured values ​​of all valid sampling points within the steady-state data segment. The relative deviation of the multimodal operating data from the steady-state baseline value at the current moment is calculated. The relative deviation is obtained by dividing the absolute value of the difference between the current measured value and the steady-state baseline value by the steady-state baseline value and converting it into a percentage. The preset deviation limit is determined based on the statistical distribution of the deviation of each modal data under normal operating conditions or industry standards. Modal data categories whose relative deviation exceeds the preset deviation limit are marked as abnormal categories. All abnormal categories are summarized to form a complete abnormal modal set.

3. The intelligent switchgear fault diagnosis method based on multimodal data analysis according to claim 1, characterized in that: The normalization process and trend consistency coefficient calculation for the full set of abnormal modes in S2 specifically include: Normalize the time series deviation sequences corresponding to various anomalous modes in the full set of anomalous modes, and map all types of time series deviation sequences to a dimensionless interval to eliminate the dimensional differences between different modes. The normalization process is obtained by subtracting the minimum value of the sequence from the measured value of the time series deviation from various abnormal modes, and then dividing by the difference between the maximum and minimum values ​​of the sequence. The trend consistency coefficient between any two abnormal modes is calculated based on the normalized deviation characteristics. The magnitude of the coefficient reflects the degree of synchronization of the deviation trends of the two abnormal modes. The trend consistency judgment threshold is determined based on the statistical distribution of the multimodal correlation coefficient in the historical operation data of the switchgear. The corresponding quantile of the trend consistency coefficient distribution under normal operating conditions is taken as the basis for the threshold value.

4. The intelligent switchgear fault diagnosis method based on multimodal data analysis according to claim 3, characterized in that: The process of dividing redundant mode groups and filtering abnormal modes in S2 specifically includes: Two or more abnormal modes with a trend consistency coefficient greater than the trend consistency judgment threshold are classified into the same redundant mode group. Calculate the deviation coefficients of various abnormal modes within each redundant mode group. The deviation coefficients are used to measure the stability of the modal signal and the sensitivity of fault characterization. They are obtained by dividing the standard deviation of the normalized deviation characteristics of various abnormal modes by the mean. Within each redundant mode group, retain the mode with the smallest deviation coefficient and eliminate the remaining redundant modes; Abnormal modes with a trend consistency coefficient less than or equal to the trend consistency judgment threshold are judged as independent of each other and directly retained; after all abnormal modes are screened, all retained abnormal modes together constitute a subset of abnormal modes without redundancy.

5. The intelligent switchgear fault diagnosis method based on multimodal data analysis according to claim 1, characterized in that: S3 includes: Extract the temporal feature parameters corresponding to the abnormal modes, including the abnormal start time and the abnormal duration; construct a mode association matching matrix based on the temporal feature parameters to identify the association relationship between modes; determine the potential association relationship based on the difference between the abnormal start time and the overlap rate of the abnormal duration, and determine the propagation direction corresponding to the association relationship by combining the fault propagation time. The preset time threshold is determined based on the fault propagation time parameter of the switchgear, and the overlap rate of abnormal duration is determined based on the extreme value ratio of the duration of the two abnormal modes.

6. The intelligent switchgear fault diagnosis method based on multimodal data analysis according to claim 5, characterized in that: S3 includes: Using abnormal modes as nodes and potential correlations as edges, correlation edge filtering is implemented by combining the physical quantity attributes corresponding to the modes; correlation edges that do not conform to the physical transmission mechanism are removed, and correlation edges that conform to the fault transmission law are retained to form the correlation transmission relationship between modes; Based on whether there is a direct causal transmission relationship between physical quantities, the associated edge judgment and elimination operation is carried out; the earliest abnormal mode at the start time and without other mode pointing is determined as the initial node, and the associated transmission path is formed by the initial node along the retained associated edges. All associated transmission paths are obtained by traversing all nodes and edges; a single abnormal mode does not form an associated transmission path and is judged as an interference signal.

7. The intelligent switchgear fault diagnosis method based on multimodal data analysis according to claim 6, characterized in that: S3 includes: The joint explanatory power is calculated based on the stability of the abnormal modes and the weighting coefficients. The weighting coefficients are determined based on the sensitivity of the modes to the fault source and are normalized. The sum of the weighting coefficients of all modes remains consistent. The joint explanatory power threshold is determined based on the statistical distribution of the effective feature set in historical fault data. The number of associated transmission paths and the joint explanatory power are judged in a dual manner. When the joint explanatory power reaches the corresponding threshold and the number of associated transmission paths is unique, the corresponding subset is determined as the effective fault feature set. The subset that does not meet the conditions is judged as interference signal and is removed.

8. The intelligent switchgear fault diagnosis method based on multimodal data analysis according to claim 1, characterized in that: S4 includes: A reverse tracing operation is performed along the unique associated propagation path in the set of valid fault features. Starting from the end abnormal mode node of the associated propagation path, all nodes included in the path are traversed sequentially in the direction of the initial abnormal mode node. During the tracing process, the physical quantity type corresponding to the abnormal mode and the abnormal propagation order are recorded sequentially to form a reverse tracing sequence opposite to the direction of the forward propagation path. Calculate the fault contribution degree corresponding to the abnormal mode in the reverse tracing sequence. The fault contribution degree is determined based on the signal stability of the abnormal mode and the hierarchical index of the abnormal mode node in the associated propagation path. Sort the fault contribution degrees of the abnormal modes in the reverse tracing sequence, determine the abnormal mode node with the smallest fault contribution degree, and identify this abnormal mode node as the feature node corresponding to the fault source.

9. The intelligent switchgear fault diagnosis method based on multimodal data analysis according to claim 8, characterized in that: The S4 includes: A one-to-one correspondence between abnormal mode types and physical quantity types is established in advance to match fault source feature nodes to the corresponding physical quantity types; a correspondence between physical quantity types and switchgear monitoring points is established in advance to match physical quantity types to the corresponding monitoring points. The equipment location corresponding to the monitoring point is determined as the fault source location of the smart switchgear. A fault diagnosis conclusion is generated based on the fault source location and the corresponding abnormal mode type. The fault diagnosis conclusion includes the fault source location, abnormal mode type, and fault propagation process. The fault source location information and the fault diagnosis conclusion are summarized to form a complete diagnosis result.