Partial discharge long-time accidental fault diagnosis method, device, equipment and medium

Abstract

The invention belongs to the technical field of fault diagnosis, and particularly relates to a partial discharge long-time accidental fault diagnosis method, device and equipment and a medium. According to the method, a time-frequency-space three-dimensional feature matrix is obtained through a dynamic adaptive sampling strategy, and feature dimension reduction optimization is realized in combination with a nonlinear sparse coding model. And constructing a hierarchical causal transfer matrix and a self-adaptive dynamic topological graph, and introducing a second derivative phase feature to enhance the causal chain recognition capability, thereby realizing high-precision spatial positioning of a fault source. And a generation-discrimination collaborative network is designed to carry out multi-step trend prediction, a causal aggregation matrix containing an attention mechanism is established, and the early warning false alarm rate is significantly reduced. And a lightweight edge computing architecture is adopted, so that the model complexity is remarkably reduced, and efficient real-time processing is realized. Through multi-dimensional feature fusion and an intelligent reasoning mechanism, the fault detection efficiency is remarkably improved, the positioning time is effectively shortened, and an efficient solution is provided for state maintenance of high-voltage equipment.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault diagnosis, and particularly relates to a method, device, equipment and medium for diagnosing long-term occasional partial discharge faults. Background Art

[0002] With the wide application of power systems and the increase in the operation years of high-voltage electrical equipment, the reliability and safety issues of power equipment have gradually become important concerns in engineering technology. Partial discharge is an early signal of the deterioration of the internal insulation performance of high-voltage electrical equipment, which can cause the gradual deterioration of the insulation system and ultimately lead to equipment failure. Therefore, accurately and efficiently monitoring partial discharge signals, real-time identifying the fault source and predicting the trend are important technical means to ensure the long-term safe operation of power equipment. At present, the monitoring and diagnosis methods for partial discharge have been widely applied to the fault prediction and condition maintenance of power equipment, but there are still a series of technical defects in the existing technology, which are difficult to fully and effectively meet the requirements of high-precision long-term monitoring.

[0003] The existing partial discharge monitoring mainly relies on single sensors or low-dimensional feature acquisition methods, which cannot fully obtain multi-dimensional signal features, making it difficult to distinguish complex signals generated by different fault sources. When encountering occasional discharge signals, the single sensor method is often limited by its detection sensitivity, unable to accurately capture weak discharge events and prone to missing early fault signals. At the same time, most of the existing signal processing technologies are based on linear analysis, showing weak processing effects on non-linear, sudden and random discharge signals, resulting in the difficulty of effectively removing noise interference in monitoring data and affecting the accurate discrimination of fault signals. Traditional signal denoising and feature extraction methods often lack pertinence, resulting in a high false alarm rate and low diagnostic accuracy.

[0004] In terms of fault source discrimination, the existing technology generally adopts discrimination methods based on static models or simple causal associations, but these methods do not fully utilize dynamic structure generation and causal association modeling. In the case of complex multi-source faults, it is easy to cause misjudgment due to the poor adaptability of the model. In addition, traditional causal association analysis generally lacks a hierarchical causal reasoning mechanism, making it difficult to meet the discrimination requirements of complex multi-level fault signals in high-voltage electrical equipment and affecting the accurate positioning of the fault source.

[0005] In terms of trend prediction and early warning, the existing methods mainly rely on simple trend models or prediction algorithms based on static rules. These methods are difficult to dynamically capture the change trend of discharge signals and cannot effectively cope with the non-linear changes of signals during the operation of equipment. Especially in high-voltage electrical equipment, due to the complex changes of fault signals in multiple dimensions such as time, frequency and space, traditional trend prediction methods cannot adaptively analyze the dynamic transfer of signal features, resulting in an unsatisfactory early warning effect.

[0006] In terms of data processing and edge computing, traditional monitoring systems often transmit all data to a central server for centralized analysis, resulting in transmission delays and increasing the computational burden on central processing, making it difficult to achieve real-time response to fault events. For fault detection that requires real-time response, the data processing ability at the edge is particularly important. However, traditional edge computing lacks in real-time performance and efficiency, making it difficult to meet the requirements of real-time monitoring. Therefore, how to provide a high-precision long-term occasional fault diagnosis system for partial discharge is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0007] The purpose of the present invention is to provide a method, device, equipment and medium for diagnosing long-term occasional faults of partial discharge to solve the problems in the prior art such as difficult identification of multi-source signals, low capture rate of weak signals, and insufficient prediction accuracy of dynamic trends in the monitoring of partial discharge of high-voltage electrical equipment.

[0008] To achieve the above purpose, the present invention adopts the following technical solutions: In the first aspect of the present invention, a method for diagnosing long-term occasional faults of partial discharge is provided, including the following steps: Obtain partial discharge signals in the detection area of high-voltage electrical equipment; Sample abnormal signals from the partial discharge signals based on preset event trigger conditions to generate a feature data set; Construct a feature matrix based on the feature data set, use a non-linear sparse coding model to optimize the dimensionality reduction of the feature matrix to obtain a low-dimensional feature matrix; perform multi-dimensional adaptive mapping on the low-dimensional feature matrix to obtain a mapping matrix; screen weak signals in the mapping matrix to obtain an optimized feature matrix; Hierarchically construct a hierarchical feature matrix for the optimized feature matrix; based on the hierarchical feature matrix, create an adaptive topology map through a dynamic structure generation network; construct a hierarchical causal transfer matrix based on the adaptive topology map, and screen the optimal causal chain based on the constructed hierarchical causal transfer matrix; based on the optimal causal chain, dynamically update the topology weights in combination with the positioning probability to generate a fault source discrimination model, and generate a fault source discrimination result based on the fault source discrimination model; Construct a multi-level causal association matrix based on the fault source discrimination result; for the multi-level causal association matrix, use a hierarchical aggregation method to generate a causal aggregation matrix; Construct a real-time signal feature matrix based on the causal aggregation matrix, calculate the node anomaly score on the signal feature matrix and screen key abnormal nodes to obtain a diagnosis result.

[0009] Further, in the step of sampling abnormal signals from the partial discharge signals based on preset event trigger conditions, the sampling frequency is dynamically adjusted according to the amplitude change trend of the partial discharge signals, so that the sampling frequency is proportional to the change rate of the signal amplitude.

[0010] Further, a non-linear sparse coding model is used to optimize the dimensionality reduction of the feature matrix to obtain a low-dimensional feature matrix, including: Based on the non-linear sparse coding model, the feature matrix is represented as a dictionary matrix and a sparse coefficient matrix; the sparse coefficient matrix The sparse coding model is optimized through an objective function, and the objective function is as follows:

[0011] Among them, represents the Frobenius norm, is the sparse regularization coefficient, is the smoothing regularization coefficient, is the time dimension, is the frequency dimension; D is the dictionary matrix; S is the sparse coefficient matrix; X is the original feature matrix; S i,j is an element in the sparse coefficient matrix S; The optimized sparse coefficient matrix is mapped through a non-linear compression operation to generate a low-dimensional feature matrix.

[0012] Further, in the adaptive topology graph, the edge weight is calculated in the following way:

[0013] Among them, and are adjustment factors, represents the time dimension, represents the frequency dimension; is the hierarchical feature matrix node; represents the th eigenvalue of the

[0014] Further, after the step of generating the causal aggregation matrix by using the hierarchical aggregation method, it also includes: The causal aggregation matrix is subjected to trend prediction through a generative-discriminative collaborative network to construct an output matrix of the generative network; among them, the output matrix of the generative network is discriminated by using the input matrix of the discriminative network; Based on the generative-discriminative error, the generative network and the discriminative network are optimized, and the weights and hierarchical structures of the causal aggregation matrix are adjusted; a trend prediction signal is generated for the optimized causal aggregation matrix, and the early warning probability for the future time period is calculated.

[0015] Further, calculating the node anomaly score on the signal feature matrix and screening key anomaly nodes includes: Calculating a node anomaly score matrix on the signal feature matrix; marking anomaly nodes according to the anomaly score threshold; Among the marked abnormal nodes, key abnormal nodes are further screened based on the fault source location probability, and an aggregated abnormal matrix is constructed for the set of key abnormal nodes.

[0016] Further, local discharge signals in the detection area of high-voltage electrical equipment are obtained, including: By arranging highly sensitive sensors in the detection area of high-voltage electrical equipment, the amplitude, frequency, and spatial position characteristic information of local discharge signals are collected.

[0017] In a second aspect of the present invention, a device for diagnosing long-term occasional faults of local discharge is provided, including: A data acquisition module for obtaining local discharge signals in the detection area of high-voltage electrical equipment; A data collection module for sampling abnormal signals from local discharge signals based on preset event trigger conditions to generate a feature data set; A first matrix construction module for constructing a feature matrix based on the feature data set, using a non-linear sparse coding model to optimize the dimensionality reduction of the feature matrix to obtain a low-dimensional feature matrix; performing multi-dimensional adaptive mapping on the low-dimensional feature matrix to obtain a mapping matrix; screening weak signals in the mapping matrix to obtain an optimized feature matrix; A second matrix construction module for hierarchically constructing a hierarchical feature matrix for the optimized feature matrix; based on the hierarchical feature matrix, creating an adaptive topology graph through a dynamic structure generation network; constructing a hierarchical causal transfer matrix based on the adaptive topology graph, screening the optimal causal chain based on the constructed hierarchical causal transfer matrix; based on the optimal causal chain, dynamically updating the topology weights in combination with the location probability to generate a fault source discrimination model, and generating a fault source discrimination result based on the fault source discrimination model; A third matrix construction module for constructing a multi-level causal association matrix based on the fault source discrimination result; for the multi-level causal association matrix, using a hierarchical aggregation method to generate a causal aggregation matrix; A fourth matrix construction module for constructing a real-time signal feature matrix based on the causal aggregation matrix, calculating node abnormality scores on the signal feature matrix, and screening key abnormal nodes to obtain a diagnosis result.

[0018] In a third aspect of the present invention, an electronic device is provided, including a processor and a memory, and the processor is configured to execute a computer program stored in the memory to implement the method for diagnosing long-term occasional faults of local discharge as described above.

[0019] In a fourth aspect of the present invention, a computer-readable storage medium is provided, and the computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by a processor, the method for diagnosing long-term occasional faults of local discharge as described above is implemented.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention proposes a high-precision acquisition method for partial discharge signals based on multi-sensor arrangement. By using high-sensitivity sensors to collect multi-dimensional characteristic information such as the amplitude, frequency, and spatial position of partial discharge signals in real time, a comprehensive foundation for equipment status monitoring is formed, solving the problem in the prior art of capturing sporadic and weak signals in high-voltage electrical equipment, and effectively improving the sensitivity and accuracy of fault signal recognition.

[0021] The present invention uses a method combining non-linear compressive sparse coding and dynamic feature mapping to denoise and separate features of signals. Low-dimensional features are extracted through compressive sparse coding, and weak signal features are adaptively recognized through dynamic feature mapping, eliminating redundant noise and high-frequency interference, and significantly improving the reliability of fault feature extraction in a complex signal environment.

[0022] The present invention introduces a dynamic structure generation network and a hierarchical causal transfer inference model, generates an adaptive topology map and establishes a hierarchical causal relationship chain, effectively differentiates complex sporadic signal sources, accurately discriminates the location of fault sources, realizes multi-level dynamic analysis of fault sources, and significantly improves the fault location accuracy in the case of multi-source faults in complex high-voltage electrical equipment.

[0023] The present invention proposes a method combining hierarchical causal aggregation and generative-discriminative collaborative prediction. By causally aggregating to dynamically adjust the causal association between signals, and combining with a generative-discriminative collaborative feedback mechanism to optimize trend prediction, hierarchical analysis and accurate early warning of signal change trends are realized, effectively improving the timeliness and accuracy of early warning responses and enhancing the fault prevention ability of the system.

[0024] Through real-time anomaly detection at the edge computing end and data transmission on the Internet of Things platform, the present invention realizes real-time monitoring of multi-region devices and cross-region centralized monitoring, eliminates the delay problem of centralized processing, improves the real-time performance of fault response, and ensures the operation safety and stability of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The schematic diagrams in the specification forming a part of this application are used to provide a further understanding of the present invention. The illustrative embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings: Figure 1 is a flowchart of a method for diagnosing long-term sporadic faults of partial discharge according to an embodiment of the present invention; Figure 2 is a flowchart of a method for diagnosing long-term sporadic faults of partial discharge according to another embodiment of the present invention; Figure 3 is a schematic diagram of a method for diagnosing long-term sporadic faults of partial discharge according to another embodiment of an embodiment of the present invention; Figure 4Structural block diagram of a partial discharge long-term occasional fault diagnosis device according to an embodiment of the present invention; Figure 5 Structural block diagram of an electronic device according to an embodiment of the present invention. Detailed implementation manners

[0026] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments. It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0027] The following detailed descriptions are all exemplary descriptions, aiming to provide further detailed descriptions of the present invention. Unless otherwise specified, all technical terms adopted by the present invention have the same meaning as commonly understood by those of ordinary skill in the art to which the present application belongs. The terms used in the present invention are only for describing specific implementation manners, and are not intended to limit the exemplary embodiments according to the present invention.

[0028] Embodiment 1 As Figure 1 shown, a partial discharge long-term occasional fault diagnosis method includes the following steps: S100. Obtain partial discharge signals in the detection area of the high-voltage electrical equipment; Specifically, obtaining partial discharge signals in the detection area of the high-voltage electrical equipment includes: Collect the amplitude, frequency, and spatial position characteristic information of the partial discharge signals through highly sensitive sensors arranged in the detection area of the high-voltage electrical equipment.

[0029] S200. Sample abnormal signals from the partial discharge signals based on a preset event trigger condition to generate a feature data set; Specifically, dynamically adjust the sampling frequency according to the amplitude change trend of the partial discharge signals, so that the sampling frequency is proportional to the change rate of the signal amplitude.

[0030] S300. Construct a feature matrix based on the feature data set, use a non-linear sparse coding model to optimize the dimensionality reduction of the feature matrix to obtain a low-dimensional feature matrix; perform multi-dimensional adaptive mapping on the low-dimensional feature matrix to obtain a mapping matrix; screen weak signals in the mapping matrix to obtain an optimized feature matrix; Specifically, using a non-linear sparse coding model to optimize the dimensionality reduction of the feature matrix to obtain a low-dimensional feature matrix includes: Based on the non-linear sparse coding model, represent the feature matrix as a dictionary matrix and a sparse coefficient matrix; the sparse coefficient matrix Optimize the sparse coding model through an objective function, and the objective function is as follows:

[0031] Wherein, denotes the Frobenius norm, is the sparse regularization coefficient, is the smoothing regularization coefficient, is the time dimension, is the frequency dimension; D is the dictionary matrix; S is the sparse coefficient matrix; X is the original feature matrix; S i,j is an element in the sparse coefficient matrix S; The optimized sparse coefficient matrix is mapped through a non-linear compression operation to generate a low-dimensional feature matrix.

[0032] S400. Hierarchically construct a hierarchical feature matrix for the optimized feature matrix; based on the hierarchical feature matrix, create an adaptive topology graph through a dynamic structure generation network; construct a hierarchical causal transfer matrix based on the adaptive topology graph, and screen the optimal causal chain based on the constructed hierarchical causal transfer matrix; based on the optimal causal chain, dynamically update the topology weights in combination with the positioning probability to generate a fault source discrimination model, and generate a fault source discrimination result based on the fault source discrimination model; Specifically, in the adaptive topology graph, the edge weight is calculated in the following way:

[0033] where and are adjustment factors, represents the time dimension, represents the frequency dimension; is a node of the hierarchical feature matrix; represents the th eigenvalue of the

[0034] S500. Construct a multi-level causal association matrix based on the fault source discrimination result; for the multi-level causal association matrix, use a hierarchical aggregation method to generate a causal aggregation matrix; Specifically, after the step of using the hierarchical aggregation method to generate a causal aggregation matrix, it further includes: Perform trend prediction on the causal aggregation matrix through a generation-discrimination collaborative network to construct an output matrix of the generation network; among them, use the input matrix of the discrimination network to discriminate the output matrix of the generation network; Optimize the generation network and the discrimination network based on the generation-discrimination error, adjust the weights and hierarchical structure of the causal aggregation matrix; generate a trend prediction signal for the optimized causal aggregation matrix, and calculate the early warning probability for the future time period.

[0035] S600. Construct a real-time signal feature matrix based on the causal aggregation matrix, calculate the node anomaly score on the signal feature matrix and screen key anomaly nodes to obtain a diagnosis result.

[0036] Specifically, calculating node anomaly scores on the signal feature matrix and screening key anomaly nodes includes: Calculating a node anomaly score matrix on the signal feature matrix; marking anomaly nodes according to the anomaly score threshold; Among the marked anomaly nodes, further screening to obtain key anomaly nodes based on the fault source location probability, and constructing an aggregated anomaly matrix for the set of key anomaly nodes.

[0037] As Figure 2 and Figure 3 shown, in another optional embodiment, a method for diagnosing long-term occasional partial discharge faults is also provided, including the following steps: S1. Arrange highly sensitive sensors in the detection area of high-voltage electrical equipment to collect the amplitude, frequency, and spatial position characteristic information of partial discharge signals; S2. Through an event trigger and an adaptive sampling mechanism, perform real-time acquisition and recording of sensor signals; S3. Use the non-linear compressive sparse coding and dynamic feature mapping method to denoise and separate features of the collected signals; among them, extract low-dimensional signal features based on compressive sparse coding, and achieve adaptive separation of weak signal features through dynamic feature mapping; S4. Based on the fault source discrimination method of a dynamic structure generation network and hierarchical causal transfer inference, generate an adaptive topology map for the signals after feature extraction, and discriminate the fault source of complex occasional signals through a hierarchical causal chain; S5. Adopt a hierarchical causal aggregation combined with a generation-discrimination collaborative prediction method to perform hierarchical analysis on the signal trend, use the causal aggregation mechanism to dynamically adjust the causal relationship between signals, and use the generation-discrimination collaborative feedback mechanism to optimize trend prediction and generate warning information; S6. Perform real-time anomaly detection at the edge computing end and transmit the results to the remote control center; S7. Transmit the monitoring data and warning information to the remote monitoring terminal through the Internet of Things platform to achieve cross-regional centralized monitoring of power equipment.

[0038] In this embodiment, S2 includes the following steps: S21. Based on the amplitude, frequency, and spatial position characteristics of partial discharge signals, set event trigger conditions for screening out abnormal signal events, and perform real-time sampling on the signal data that meets the trigger conditions.

[0039] Among them, the adaptive sampling mechanism is used to adjust the sampling frequency and sampling interval.

[0040] Specifically, during the sampling process, the sampling frequency is dynamically adjusted according to the changing trend of the signal amplitude, such that the sampling frequency is proportional to the change rate of the signal amplitude. For example, during the sampling process, the signal amplitude change rate is used to control the sampling interval, and a threshold is set. When , the sampling frequency is doubled; where represents the signal amplitude, is time, is a preset threshold.

[0041] In the above solution, during the sampling process, the amplitude, frequency, and spatial position of the signal are recorded. Through the adaptive sampling strategy, signals with high-frequency changes are collected densely, and signals with low-frequency changes are collected sparsely.

[0042] Optionally, for sporadic signal events, the trigger window mechanism is used to extend the sampling before and after the event trigger period to ensure a complete signal record before and after the event occurs.

[0043] S22. The collected partial discharge signal data is recorded in real time and sent to the data storage module, and saved as a feature data set for subsequent discrimination and analysis of the fault source.

[0044] In this embodiment, S3 includes the following steps: S31. Construct a feature matrix for the collected partial discharge signal data ; where is the number of sampling points, is the feature dimension, is the value of the th sampling point in the th feature dimension; S32. Based on the non-linear sparse coding model, represent the feature matrix as a dictionary matrix and a sparse coefficient matrix , satisfying ; where is the error matrix.

[0045] The sparse coding model is optimized through an objective function, and the objective function is as follows:

[0046] where represents the Frobenius norm, is the sparse regularization coefficient, is the smoothing regularization coefficient, is the time dimension, is the frequency dimension; D is the dictionary matrix; S is the sparse coefficient matrix; X is the original feature matrix; F is; S i,jis an element in the sparse coefficient matrix S; S33. Map the sparse coefficient matrix through a non-linear compression operation to generate a low-dimensional feature matrix ; S34. Construct a dynamic feature mapping matrix , and perform multi-dimensional adaptive mapping on the low-dimensional feature matrix . The calculation method for each element is as follows:

[0047] where is the time dimension, is the frequency dimension, and are the spatial position coordinates, and are the adjustment factors; S35. Set the weak signal detection threshold . Screen for weak signal features in the mapping matrix . When is satisfied, mark the feature as a weak signal; S36. Denoise the marked weak signal features, remove background noise and interference components, and retain the signal features related to diagnosis to generate an optimized feature matrix ; S37. Output the optimized feature matrix to the fault source discrimination module for subsequent fault analysis.

[0048] In this embodiment, S4 includes the following steps: S41. Perform multi-level feature node stratification on the optimized matrix after feature extraction to construct a stratified feature matrix , where represents the number of stratified nodes, is the feature dimension, represents the th th eigenvalue of the S42. Based on the stratified feature matrix , use a dynamic structure generation network to generate an adaptive topological graph , where represents the node set, represents the edge set, represents the edge weight, and the weight is calculated by the following method:

[0049] Among them, and are adjustment factors, represents the time dimension, represents the frequency dimension; S43. Based on the generated adaptive topological map , construct a hierarchical causal transfer matrix , where represents the causal relationship strength from the th layer to the th layer of nodes. The calculation formula for the causal relationship is:

[0050] Among them, and are regularization coefficients, and are the spatial coordinate dimensions; t represents the time dimension, and f represents the frequency dimension.

[0051] S44. Set a causal strength threshold in the causal transfer matrix , and filter out the causal paths whose strength satisfies to form an optimal causal chain , which is used to mark the causal paths directly associated with the fault source; S45. Based on the optimal causal chain , calculate the localization probability for the candidate fault node ;

[0052] S46. Feed back the result of the localization probability to the dynamic structure generation network to update the node weights and the causal chain structure in the topological map, and generate an optimized fault source discrimination model; S47. Output the optimized fault source discrimination result to the monitoring and early warning module to provide basic data for subsequent trend prediction. Based on the localization probability P( ) of the candidate fault node ), construct a fault source localization matrix F∈R^(p×q) for subsequent anomaly detection.

[0053] In this embodiment, S5 includes the following steps: S51. Based on the fault source discrimination result, construct a multi-level causal association matrix , where is the number of levels, is the feature dimension, and the matrix element represents the The causal association strength between layers and the feature dimension; S52. Using the hierarchical causal aggregation method for the causal association matrix to establish a causal aggregation matrix , where each element represents the aggregated causal strength between the layer and the layer nodes:

[0054] Among them, and are adjustment factors, is the time dimension, used to adjust the causal strength and the rate of change; represents the time decay factor.

[0055] S53. Through the generative-discriminative collaborative network, perform trend prediction on the causal aggregation matrix to construct the generative network output matrix = , satisfying:

[0056] Among them, and are adjustment parameters in the generative network, is the frequency dimension, and are spatial coordinates; is the neighborhood of node i; is the edge weight defined in S42.

[0057] S54. Construct the discriminative network input matrix , and perform discrimination on the output of the generative matrix ; S55. Based on the generative-discriminative error optimize the generative network and the discriminative network, and adjust the weights and hierarchical structure of the causal aggregation matrix ; S56. Generate a trend prediction signal for the optimized causal aggregation matrix , and calculate the early warning probability for the future time period; S57. Transmit the trend prediction result and the early warning information to the monitoring and early warning module to provide a prediction basis for fault protection.

[0058] In this embodiment, S6 includes the following steps: S61. Based on the foregoing causal aggregation matrix and the fault source location matrix , build a real-time anomaly detection model and define the detection signal feature matrix at the edge computing end , where is the number of node levels, is the feature dimension, represents the real-time signal feature value of the th layer and the th dimension; S62. Calculate the node anomaly score matrix on the signal feature matrix , where each element represents the anomaly degree of the th layer and the th node; Use statistical Z-score to calculate the anomaly score:

[0059] where, μ j and σ j are the historical mean and standard deviation of the jth dimension feature respectively, z ij represents the real-time signal feature value of the jth dimension of the ith layer.

[0060] S63. Set the anomaly score threshold . When , mark the node as an abnormal node and record the signal feature value of the abnormal node at the same time; S64. Among the marked abnormal nodes, further screen the key abnormal nodes based on the fault source location probability , and aggregate the signal features of the key abnormal node set ; max(P(F)) is the maximum location probability, is the ratio threshold.

[0061] S65. Build an aggregated anomaly matrix for the key abnormal node set ; S66. Convert the aggregated anomaly matrix into a real-time anomaly detection signal and transmit it to the remote control center for monitoring and analysis; S67. Real-time record and transmit the detection signal features and anomaly score information to form an edge-side anomaly detection data stream, providing basic data for fault monitoring.

[0062] During the maintenance of high-voltage switchgear in large substations, due to the risks of early faults such as aging of insulating materials and partial discharge inside the equipment, it is easy to cause equipment failures, resulting in unstable power supply of the power grid and even large-scale power outages. In order to achieve timely detection and prediction warning of early faults, high-voltage switchgear in a substation in a certain area is selected as the application scenario for testing partial discharge monitoring and long-term occasional fault diagnosis. The substation is located in a coastal area with humid climate and large temperature fluctuations, and the equipment ages relatively fast, which is a typical application scenario for the method of the present invention.

[0063] In this scenario, by arranging a variety of high-sensitivity sensors in the high-voltage switch detection area of the substation, including ultra-high frequency sensors, ultrasonic sensors and fiber optic sensors, which are distributed on the insulators, busbars and switch cabinets of the equipment, responsible for real-time collection of signal feature information of partial discharge. Since the fault characteristic frequencies in this area are in the high-frequency range and are highly sporadic, traditional monitoring methods often have difficulty in accurately capturing tiny discharge phenomena. Therefore, the method of the present invention, through high-sensitivity sensors, combined with an event-triggered and adaptive sampling mechanism, collects features such as signal amplitude, frequency and spatial position, and collects signals in real time and automatically records abnormal event data. The trigger threshold for data collection is set at 0.1 μA. When the detected signal change rate exceeds the threshold, the system automatically adjusts the sampling frequency to twice the normal frequency, so as to effectively capture key signal features.

[0064] During the data collection process, in order to ensure the accuracy of the signal, the nonlinear compression sparse coding and dynamic feature mapping methods are applied to the original data. Nonlinear compression sparse coding is used to remove background noise and irrelevant interference signals, making tiny discharge signals more prominent; dynamic feature mapping then adaptively maps the compressed low-dimensional signal features to obtain a signal matrix after separating weak signal features. After signal processing, the system can identify the fault features contained in the tiny signals and provide high-precision data input in the subsequent fault source discrimination process.

[0065] The system inputs the processed signal features into a dynamic structure generation network and a hierarchical causal transfer inference module to generate an adaptive topology map containing causal associations and establish a hierarchical causal chain. The advantage of this module is that it can dynamically update the causal relationship map between signals. Through multi-level hierarchical association modeling of the discharge source of the switchgear, the system can infer and analyze the source location of the discharge event from multiple dimensions. During the test, through continuous monitoring and analysis for 30 days, the system captured abnormal changes in the discharge signal many times and accurately located the discharge fault source near the insulator. The traditional method failed to identify this fault source within the same time, while the method of the present invention obtained a probability of up to 89% for the internal fault source of the insulator through hierarchical causal analysis and generated a fault alarm message.

[0066] For trend prediction and early warning, the present invention uses a hierarchical causal aggregation combined with a generation-discrimination collaborative prediction method within the substation to perform hierarchical analysis on the changing trend of fault signals. The generation-discrimination collaborative prediction network is based on the partial discharge data collected in the early stage, and generates prediction results by dynamically adjusting the causal association model. Through data observation for two consecutive months, the system issued 3 fault early warnings, and finally confirmed that the matching rate between the early warning fault location and the actual fault location reached more than 92%, and the early warning time was advanced by more than 10 days on average. Traditional monitoring methods usually can only provide abnormal prompts and cannot issue accurate early warning information days in advance.

[0067] To ensure the real-time data transmission and centralized monitoring, a real-time anomaly detection module is added at the edge of the system, and the transmission delay of anomaly detection is kept within 0.5 seconds. The detected data is synchronously transmitted to the remote monitoring center, and centralized monitoring of cross-regional devices is realized through the Internet of Things platform. The on-site data record shows that during the entire 30-day test cycle, the system maintained 99.5% data integrity at a data transmission speed of 200 data per second on average, and was able to achieve synchronous monitoring of cross-regional devices and real-time response to faults.

[0068] Table 1 Data Table

[0069] From the above data, it can be seen that the diagnostic method of the present invention maintains stable abnormal capture times in different time periods, the accuracy of fault source location gradually increases, the early warning lead time increases with the optimization of trend prediction and collaborative feedback, and the integrity and real-time performance of data transmission are excellent, which can meet the requirements of fault diagnosis and early warning in complex environments.

[0070] Embodiment 2 As Figure 4 shown, based on the same inventive concept as the above embodiment, the present invention also provides a long-term occasional fault diagnosis device for partial discharge, including: A data acquisition module for acquiring partial discharge signals in the detection area of high-voltage electrical equipment; A data collection module for sampling abnormal signals from the partial discharge signals based on preset event trigger conditions to generate a feature data set; A first matrix construction module for constructing a feature matrix based on the feature data set, using a non-linear sparse coding model to optimize the dimensionality reduction of the feature matrix to obtain a low-dimensional feature matrix; performing multi-dimensional adaptive mapping on the low-dimensional feature matrix to obtain a mapping matrix; screening weak signals in the mapping matrix to obtain an optimized feature matrix; The second matrix construction module is used to hierarchically construct a hierarchical feature matrix for the optimized feature matrix; based on the hierarchical feature matrix, an adaptive topology graph is created through a dynamic structure generation network; a hierarchical causal transfer matrix is constructed based on the adaptive topology graph, and the optimal causal chain is screened based on the constructed hierarchical causal transfer matrix; based on the optimal causal chain, the topology weight is dynamically updated in combination with the positioning probability to generate a fault source discrimination model, and a fault source discrimination result is generated based on the fault source discrimination model; The third matrix construction module is used to construct a multi-level causal association matrix based on the fault source discrimination result; for the multi-level causal association matrix, a causal aggregation matrix is generated by using a hierarchical aggregation method; The fourth matrix construction module is used to construct a real-time signal feature matrix based on the causal aggregation matrix, calculate the node anomaly score on the signal feature matrix, and screen key abnormal nodes to obtain a diagnosis result.

[0071] Embodiment 3 As Figure 5 shown, the present invention also provides an electronic device 100 for implementing a method for diagnosing long-term occasional partial discharge faults; The electronic device 100 includes a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on the at least one processor 102, and at least one communication bus 104.

[0072] The memory 101 can be used to store the computer program 103. The processor 102 realizes the steps of the method for diagnosing long-term occasional partial discharge faults in Embodiment 1 by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101.

[0073] The memory 101 may mainly include a program storage area and a data storage area. Among them, the program storage area may store an operating system, application programs required for at least one function (such as a sound playback function, an image playback function, etc.); the data storage area may store data created according to the use of the electronic device 100 (such as audio data, etc.). In addition, the memory 101 may include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices.

[0074] At least one processor 102 can be a Central Processing Unit (CPU), or can also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor 102 can be a microprocessor or the processor 102 can also be any conventional processor, etc. The processor 102 is the control center of the electronic device 100 and connects various parts of the entire electronic device 100 through various interfaces and lines.

[0075] The memory 101 in the electronic device 100 stores multiple instructions to implement a method for diagnosing long-term occasional partial discharge faults. The processor 102 can execute the multiple instructions to implement: Obtain partial discharge signals in the detection area of high-voltage electrical equipment; Sample abnormal signals from the partial discharge signals based on preset event trigger conditions to generate a feature data set; Construct a feature matrix based on the feature data set, use a non-linear sparse coding model to optimize the dimensionality reduction of the feature matrix to obtain a low-dimensional feature matrix; perform multi-dimensional adaptive mapping on the low-dimensional feature matrix to obtain a mapping matrix; screen weak signals in the mapping matrix to obtain an optimized feature matrix; Construct a hierarchical feature matrix by layering the optimized feature matrix; based on the hierarchical feature matrix, create an adaptive topology map through a dynamic structure generation network; construct a hierarchical causal transfer matrix based on the adaptive topology map, and screen the optimal causal chain based on the constructed hierarchical causal transfer matrix; based on the optimal causal chain, dynamically update the topology weights in combination with the positioning probability to generate a fault source discrimination model, and generate a fault source discrimination result based on the fault source discrimination model; Construct a multi-level causal association matrix based on the fault source discrimination result; for the multi-level causal association matrix, use a hierarchical aggregation method to generate a causal aggregation matrix; Construct a real-time signal feature matrix based on the causal aggregation matrix, calculate the node anomaly score on the signal feature matrix, and screen key anomaly nodes to obtain a diagnosis result.

[0076] Embodiment 4 If the integrated module / unit of the electronic device 100 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, to implement all or part of the processes in the above-described embodiment methods of the present invention, it can also be completed by a computer program instructing relevant hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-described method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, and read-only memory (ROM, Read-Only Memory).

[0077] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, system, or computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0078] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of processes and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0079] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0080] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions for implementing the functions specified in one process or a plurality of processes and / or blocks Figure 1 one process or a plurality of processes and / or blocks Figure 1 steps of the functions specified in one block or a plurality of blocks.

[0081] In the description of this specification, the descriptions referring to the terms "one embodiment", "example", "specific example", etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0082] 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 above embodiments, those of ordinary skill in the art should understand that: it is still possible to modify the specific implementation manners of the present invention or make equivalent replacements, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. A method for diagnosing long-term occasional partial discharge faults, characterized in that, It includes the following steps: Obtain the partial discharge signals in the detection area of high-voltage electrical equipment; Sample abnormal signals from the partial discharge signals based on preset event triggering conditions to generate a feature dataset; Construct a feature matrix based on the feature dataset, and use a non-linear sparse coding model to optimize the dimensionality reduction of the feature matrix to obtain a low-dimensional feature matrix; Perform multi-dimensional adaptive mapping on the low-dimensional feature matrix to obtain a mapping matrix; Screen weak signals in the mapping matrix to obtain an optimized feature matrix; Hierarchically construct a hierarchical feature matrix for the optimized feature matrix; Based on the hierarchical feature matrix, create an adaptive topology graph through a dynamic structure generation network; Based on the adaptive topology graph, construct a hierarchical causal transfer matrix, screen the optimal causal chain based on the constructed hierarchical causal transfer matrix; based on the optimal causal chain, dynamically update the topology weights in combination with the positioning probability to generate a fault source discrimination model, and generate a fault source discrimination result based on the fault source discrimination model; Construct a multi-level causal association matrix based on the fault source discrimination result; for the multi-level causal association matrix, use a hierarchical aggregation method to generate a causal aggregation matrix; Construct a real-time signal feature matrix based on the causal aggregation matrix, calculate the node anomaly score on the signal feature matrix and screen key abnormal nodes to obtain a diagnosis result.

2. The method for diagnosing long-term occasional partial discharge faults according to claim 1, characterized in that In the step of sampling abnormal signals from the partial discharge signals based on preset event triggering conditions, dynamically adjust the sampling frequency according to the amplitude change trend of the partial discharge signals, so that the sampling frequency is proportional to the change rate of the signal amplitude.

3. The method for diagnosing long-term occasional partial discharge faults according to claim 1, characterized in that, Using a non-linear sparse coding model to optimize the dimensionality reduction of the feature matrix to obtain a low-dimensional feature matrix, including: Based on the non-linear sparse coding model, represent the feature matrix as a dictionary matrix and a sparse coefficient matrix; sparse coefficient matrix Optimize the sparse coding model through an objective function, and the objective function is as follows: Among them, represents the Frobenius norm, is the sparse regularization coefficient, is the smoothing regularization coefficient, is the time dimension, is the frequency dimension; D is the dictionary matrix; S is the sparse coefficient matrix; X is the original feature matrix; S i,j is an element in the sparse coefficient matrix S; Perform mapping on the optimized sparse coefficient matrix through non-linear compression operations to generate a low-dimensional feature matrix.

4. The method for diagnosing long-term occasional partial discharge faults according to claim 1, characterized in that In the adaptive topology graph, the edge weight is calculated as follows: Among them, and are adjustment factors, represents the time dimension, represents the frequency dimension; is a node of the hierarchical feature matrix; represents the th eigenvalue of the 5. The long-term occasional fault diagnosis method for partial discharge according to claim 1, characterized in that, After the step of using a hierarchical aggregation method to generate a causal aggregation matrix, it further includes: Perform trend prediction on the causal aggregation matrix through a generation-discrimination collaborative network to construct an output matrix of the generation network; among them, use the input matrix of the discrimination network to discriminate the output matrix of the generation network; Optimize the generation network and the discrimination network based on the generation-discrimination error, adjust the weights and hierarchical structure of the causal aggregation matrix; generate a trend prediction signal for the optimized causal aggregation matrix, and calculate the early warning probability for the future time period.

6. The long-term occasional partial discharge fault diagnosis method according to claim 1, characterized in that Calculating the node anomaly score and screening key abnormal nodes on the signal feature matrix, including: Calculate the node anomaly score matrix on the signal feature matrix; mark abnormal nodes according to the anomaly score threshold; Among the marked abnormal nodes, further screen to obtain key abnormal nodes based on the fault source location probability, and construct an aggregated anomaly matrix for the set of key abnormal nodes.

7. The long-term occasional fault diagnosis method for partial discharge according to claim 1, wherein Obtain the partial discharge signals in the detection area of high-voltage electrical equipment, including: Collect the amplitude, frequency and spatial position feature information of the partial discharge signals through highly sensitive sensors arranged in the detection area of the high-voltage electrical equipment.

8. A device for diagnosing long-term occasional partial discharge faults, characterized in that, It includes: A data acquisition module for obtaining the partial discharge signals in the detection area of high-voltage electrical equipment; A data acquisition module, configured to sample abnormal signals from partial discharge signals based on preset event triggering conditions, and generate a feature data set; A first matrix construction module, configured to construct a feature matrix based on the feature data set, and perform dimensionality reduction optimization on the feature matrix by using a non-linear sparse coding model to obtain a low-dimensional feature matrix; Perform multi-dimensional adaptive mapping on the low-dimensional feature matrix to obtain a mapping matrix; Screen weak signals in the mapping matrix to obtain an optimized feature matrix; A second matrix construction module, configured to hierarchically construct a hierarchical feature matrix for the optimized feature matrix; Based on the hierarchical feature matrix, create an adaptive topology map through a dynamic structure generation network; Construct a hierarchical causal transfer matrix based on the adaptive topology map, screen the optimal causal chain based on the constructed hierarchical causal transfer matrix; based on the optimal causal chain, dynamically update the topology weights in combination with the positioning probability to generate a fault source discrimination model, and generate a fault source discrimination result based on the fault source discrimination model; A third matrix construction module, configured to construct a multi-level causal association matrix based on the fault source discrimination result; For the multi-level causal association matrix, adopt a hierarchical aggregation method to generate a causal aggregation matrix; A fourth matrix construction module, configured to construct a real-time signal feature matrix based on the causal aggregation matrix, calculate node anomaly scores on the signal feature matrix, and screen key abnormal nodes to obtain a diagnosis result.

9. An electronic device, characterized in that, It includes a processor and a memory, and the processor is configured to execute a computer program stored in the memory to implement the partial discharge long-term occasional fault diagnosis method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by the processor, it implements the partial discharge long-term occasional fault diagnosis method according to any one of claims 1 to 7.

Images