IGBT inverter fault diagnosis method and device, equipment and storage medium

Abstract

The invention relates to a fault diagnosis method and device for an IGBT inverter, equipment and a storage medium, and the method comprises the following steps: diagnosing whether the IGBT inverter has a fault or not based on original electrical working parameters, and carrying out the probability distribution calculation of the original electrical working parameters if the IGBT inverter has the fault, and obtaining a fault feature probability density curve; carrying out topological structure analysis on the IGBT inverter based on the fault feature probability density curve to obtain a fault feature topological graph; performing graph attention network analysis on the fault feature topological graph to obtain a fault propagation path set; and performing causal reasoning analysis on the IGBT inverter based on the fault propagation path set to obtain a fault type identification result, and formulating a corresponding fault repair strategy based on the fault type identification result, thereby solving the problems that many methods are lack of dynamic response capability to electrical parameter change, cannot accurately capture early fault signals in real time, and are poor in reliability. And therefore, the hysteresis quality of fault diagnosis is caused.

Description

Technical Field

[0001] The present invention relates to the technical field of IGBT inverters, and in particular to a fault diagnosis method, device, equipment and storage medium for an IGBT inverter. Background Art

[0002] In modern industrial applications, insulated gate bipolar transistor (IGBT) inverters are widely used as key components in power electronic systems, such as renewable energy generation, electric vehicles, and industrial drives. These applications place extremely high demands on the reliability of inverters, as any failure may lead to reduced system efficiency, equipment damage, and even casualties. However, traditional fault diagnosis methods often rely on manual experience and simple threshold judgments, which are difficult to cope with fault conditions in complex and changeable actual working environments. In addition, with the increasing integration and complexity of power electronic systems, the effectiveness and accuracy of traditional diagnostic methods have been severely challenged.

[0003] In the face of the above challenges, although existing fault diagnosis technologies have made some progress, there are still some significant problems. For example, many methods lack the ability to dynamically respond to changes in electrical parameters and cannot accurately capture early fault signals in real time, resulting in a lag in fault diagnosis. In addition, since different types of faults may exhibit similar electrical characteristics, it is particularly difficult to accurately identify the type of fault. At the same time, existing diagnostic models usually require a lot of manual intervention for parameter adjustment and model training, which not only consumes a lot of manpower and material resources, but also limits its wide application in practical engineering.

[0004] In order to solve these problems, researchers began to explore more intelligent and efficient fault diagnosis strategies, aiming to improve the speed and accuracy of fault diagnosis while reducing the need for external intervention. Against this background, "A Fault Diagnosis Method for IGBT Inverters" came into being. This method combines a variety of advanced technologies such as high-frequency sampling, adaptive kernel density estimation, topological structure analysis, and graph attention networks to try to overcome the limitations of existing technologies and achieve rapid, accurate diagnosis and effective repair of IGBT inverter faults. This method not only improves the sensitivity and specificity of fault detection, but also can automatically identify fault propagation paths and formulate corresponding repair strategies, greatly improving the safety and stability of inverter operation. Summary of the invention

[0005] The main purpose of the present invention is to provide a fault diagnosis method, device, equipment and storage medium for an IGBT inverter, which solves the technical problem that many methods lack the ability to dynamically respond to changes in electrical parameters and cannot accurately capture early fault signals in real time, thus leading to a lag in fault diagnosis.

[0006] To achieve the above object, the present invention provides a fault diagnosis method for an IGBT inverter, comprising the following steps: Perform high-frequency sampling on the electrical operating parameters of the IGBT inverter to obtain the original electrical operating parameters; Diagnose whether the IGBT inverter has a fault based on the original electrical operating parameters, and if so, perform probability distribution calculation on the original electrical operating parameters through adaptive kernel density estimation to obtain a fault feature probability density curve; Performing a topological structure analysis on the IGBT inverter based on the fault feature probability density curve to obtain a fault feature topological map; Performing graph attention network analysis on the fault feature topology map to obtain a set of fault propagation paths; A causal reasoning analysis is performed on the IGBT inverter based on the fault propagation path set to obtain a fault type identification result, and a corresponding fault repair strategy is formulated based on the fault type identification result.

[0007] Furthermore, the probability distribution calculation of the original electrical working parameters is performed by adaptive kernel density estimation to obtain a fault feature probability density curve, including: Performing sliding window segmentation on the original electrical working parameters to obtain an electrical parameter subsequence set; and performing local density evaluation on the electrical parameter subsequence set using a preset adaptive kernel density estimation algorithm to obtain an electrical parameter subsequence density distribution; Performing feature space mapping on the electrical parameter subsequence density distribution by using a preset multi-dimensional Gaussian kernel function to obtain a feature space mapping result, and performing kernel density calculation based on the feature space mapping result to obtain a fault feature kernel density distribution; Performing marginal density integration on the fault feature kernel density distribution to obtain a marginal density distribution set, and performing multi-dimensional combination on the marginal density distribution set through probability density fusion to obtain a combined probability density sequence, and performing density curve fitting based on the combined probability density sequence to obtain a fault feature density curve set; A time series correlation analysis is performed on the set of fault feature density curves to obtain a time series correlation density curve, and fault feature extraction is performed on the time series correlation density curve to obtain a fault feature probability density curve.

[0008] Furthermore, performing marginal density integration on the fault feature kernel density distribution to obtain a marginal density distribution set includes: Performing dimensionality reduction projection on the fault feature kernel density distribution to obtain a multidimensional density projection set, and performing integral variable separation on the multidimensional density projection set to obtain an integral variable matrix; Performing multiple integral calculations on the integral variable matrix through variable conversion to obtain a marginal probability body, and performing isosurface extraction on the marginal probability body to obtain a marginal density isosurface set; The marginal density isosurface set is fitted based on the mixed Gaussian distribution to obtain the marginal Gaussian component, and the density function is constructed based on the marginal Gaussian component to obtain the marginal density distribution set.

[0009] Furthermore, the topological structure analysis of the IGBT inverter is performed based on the fault feature probability density curve to obtain a fault feature topological map, including: The fault feature probability density curve is segmented and parsed to obtain a fault feature segmentation sequence, and a circuit topology mapping is performed based on the fault feature segmentation sequence to obtain a topology node association set; wherein the topology node association set includes a power tube node, a drive circuit node, and a protection circuit node; Reconstructing the spatial structure of the topological node association set by a preset adjacency matrix construction algorithm to obtain a fault topology adjacency matrix, and calculating the node degree based on the fault topology adjacency matrix to obtain a node degree feature set; Performing centrality analysis on the node degree feature set to obtain a node centrality matrix, and performing fault correlation calculation based on the node centrality matrix to obtain a fault correlation feature set; Performing a propagation dynamic analysis on the fault correlation feature set by using a preset fault propagation path algorithm to obtain a fault propagation dynamic sequence, and performing topological subgraph division on the fault propagation dynamic sequence to obtain a topological subgraph set; Fault feature fusion is performed on the topological subgraph set to obtain a fault feature topological description set, and a graph is generated based on the fault feature topological description set to obtain a fault feature topological graph.

[0010] Furthermore, the graph attention network analysis is performed on the fault feature topology map to obtain a set of fault propagation paths, including: Performing a graph domain representation conversion on the fault feature topology map to obtain a graph domain feature tensor, and performing attention weight calculation on the graph domain feature tensor to obtain a node attention weight matrix; wherein the matrix includes power tube node weights, drive circuit node weights, and protection circuit node weights; Performing a weight distribution analysis on the node attention weight matrix through a preset multi-head attention mechanism to obtain a multi-head attention feature set, and performing feature aggregation based on the multi-head attention feature set to obtain an aggregated feature matrix; Performing graph structure encoding on the aggregated feature matrix to obtain a graph structure encoding sequence, and performing node message transmission based on the graph structure encoding sequence to obtain a node message transmission matrix; Performing feature residual fusion on the node message transfer matrix to obtain a residual feature matrix, and performing path importance evaluation based on the residual feature matrix to obtain a path importance sequence; Performing path screening on the path importance sequence to obtain a critical propagation path set, and performing path feature extraction on the critical propagation path set to obtain a fault propagation path set.

[0011] Further, the causal reasoning analysis is performed on the IGBT inverter based on the fault propagation path set to obtain a fault type identification result, and a corresponding fault repair strategy is formulated based on the fault type identification result, including: Constructing a causal graph for the set of fault propagation paths to obtain a fault causal relationship graph, and performing node dependency analysis based on the fault causal relationship graph to obtain a fault dependency matrix; The fault dependency matrix is ​​conditionally calculated by a preset Bayesian inference algorithm to obtain a fault condition probability table, and a fault propagation sequence analysis is performed based on the fault condition probability table to obtain a set of fault propagation chains; Performing causal strength evaluation on the fault propagation chain set to obtain a causal strength matrix, and locating the fault source based on the causal strength matrix to obtain a fault source feature set; The fault source feature set is mapped to a fault type by a preset multi-layer causal analysis algorithm to obtain a fault type feature matrix, and fault mode identification is performed based on the fault type feature matrix to obtain a fault type identification result; wherein the fault type identification result includes a short circuit fault type, an open circuit fault type, and a breakdown fault type; Generating a repair plan for the fault type identification result to obtain a set of fault repair strategies, and sorting the repair priorities based on the set of fault repair strategies to obtain a repair priority sequence, wherein the repair priority sequence includes an emergency repair strategy, a temporary repair strategy, and a preventive repair strategy; Repair instructions are generated based on the repair priority sequence to obtain a fault repair instruction set; wherein the fault repair instruction set includes a power tube replacement instruction, a drive circuit adjustment instruction and a protection parameter configuration instruction.

[0012] Furthermore, the generation of repair instructions based on the repair priority sequence to obtain a fault repair instruction set includes: Decomposing the repair priority sequence into tasks to obtain a repair task sequence, and performing instruction timing planning on the repair task sequence to obtain a repair instruction timing table; Performing dependency analysis on the repair instruction timing table to obtain an instruction dependency matrix, and setting execution conditions based on the instruction dependency matrix to obtain an instruction execution condition set; The set of instruction execution conditions is defined by using preset security constraint rules to obtain an instruction constraint rule table, and an execution process is designed based on the instruction constraint rule table to obtain a repair instruction flow chart; Instruction codes are generated for the repair instruction flow chart to obtain a fault repair instruction set.

[0013] The present invention also provides a fault diagnosis device for an IGBT inverter, comprising: A sampling module is used to perform high-frequency sampling on the electrical operating parameters of the IGBT inverter to obtain the original electrical operating parameters; A calculation module, used for diagnosing whether the IGBT inverter has a fault based on the original electrical operating parameters, and if so, performing probability distribution calculation on the original electrical operating parameters by adaptive kernel density estimation to obtain a fault characteristic probability density curve; A first analysis module, configured to perform a topological structure analysis on the IGBT inverter based on the fault feature probability density curve to obtain a fault feature topological map; The second analysis module is used to perform a graph attention network analysis on the fault feature topology map to obtain a set of fault propagation paths; The third analysis module is used to perform causal reasoning analysis on the IGBT inverter based on the fault propagation path set to obtain a fault type identification result, and formulate a corresponding fault repair strategy based on the fault type identification result.

[0014] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0015] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0016] The present invention provides a fault diagnosis method for an IGBT inverter, comprising the following steps: performing high-frequency sampling on electrical working parameters of the IGBT inverter to obtain original electrical working parameters; diagnosing whether the IGBT inverter has a fault based on the original electrical working parameters, and if so, performing probability distribution calculation on the original electrical working parameters through an adaptive kernel density estimation algorithm to obtain a fault feature probability density curve; performing topological structure analysis on the IGBT inverter based on the fault feature probability density curve to obtain a fault feature topological map; performing graph attention network analysis on the fault feature topological map to obtain a set of fault propagation paths; performing causal reasoning analysis on the IGBT inverter based on the set of fault propagation paths to obtain a fault type identification result, and formulating a corresponding fault repair strategy based on the fault type identification result, thereby solving the technical problem that many methods lack the ability to dynamically respond to changes in electrical parameters and cannot accurately capture early fault signals in real time, thereby causing the lag of fault diagnosis, and adopting a graph attention network to analyze the fault feature topological map to effectively extract a set of key fault propagation paths. By assigning different importance weights to different nodes and edges, this method can accurately locate the critical path that is most likely to cause fault spread among many possible fault paths, greatly improving the efficiency and accuracy of fault location. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 1 is a schematic diagram of the steps of a fault diagnosis method for an IGBT inverter according to an embodiment of the present invention; Figure 2 is a structural block diagram of a fault diagnosis device for an IGBT inverter in one embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0018] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0020] like Figure 1 As shown, Figure 1 It is a schematic diagram of the steps of a fault diagnosis method for an IGBT inverter in one embodiment of the present invention; An embodiment of the present invention provides a fault diagnosis method for an IGBT inverter, comprising the following steps: Step S1, high-frequency sampling of electrical operating parameters of the IGBT inverter is performed to obtain original electrical operating parameters.

[0021] Specifically, in the fault diagnosis method of IGBT inverter, high-frequency sampling of electrical working parameters is a key step to obtain the original electrical working parameters, which is directly related to the accuracy and efficiency of subsequent fault diagnosis. Specifically, by installing high-precision sensors on the key nodes of the IGBT inverter, the changes of key electrical parameters such as voltage and current can be monitored and recorded in real time, and these data form the basis for subsequent analysis. High-frequency sampling means collecting data points multiple times in a very short time interval to ensure that small fluctuations and transient changes in electrical parameters are captured, which is crucial for early detection of potential faults. For example, in a wind power generation system, the IGBT inverter is responsible for converting the AC power generated by the generator into stable DC or AC power suitable for grid transmission. Any slight change in electrical parameters may indicate aging of the internal components of the equipment or an impending failure. In order to achieve this process, it is first necessary to select a suitable sampling frequency, which needs to be determined according to the specific application of the inverter and the type of fault expected to be detected. Generally, the higher the frequency, the more detailed information can be captured, but it also increases the complexity and cost of data processing. Next, through a precisely synchronized data acquisition system, it is ensured that the data collected by each sensor can accurately reflect the electrical status under actual working conditions. These original electrical operating parameters include not only instantaneous values, but also multi-dimensional information such as waveform characteristics and spectral components, providing rich data support for subsequent probability distribution calculations based on adaptive kernel density estimation. Taking the drive system of an electric vehicle as an example, high-frequency sampled data can help identify the increase in resistance caused by temperature changes during motor operation, thereby providing early warning of possible overheating risks and determining through further analysis whether cooling measures or other maintenance operations are needed. In short, through high-frequency sampling of the electrical operating parameters of the IGBT inverter, its operating status can be fully and carefully understood, laying a solid foundation for subsequent fault diagnosis.

[0022] Step S2, diagnose whether the IGBT inverter has a fault based on the original electrical operating parameters, and if so, perform probability distribution calculation on the original electrical operating parameters through an adaptive kernel density estimation algorithm to obtain a fault feature probability density curve.

[0023] Specifically, the process of diagnosing whether an IGBT inverter is faulty based on the original electrical working parameters is to identify abnormal conditions in the system operation by analyzing the collected data. If a fault is detected, the adaptive kernel density estimation algorithm is further used to calculate the probability distribution of these data to obtain the fault characteristic probability density curve. This process first requires a comprehensive analysis of a large number of original electrical working parameters obtained from high-frequency sampling, and by comparing them with the benchmark data under normal operating conditions, it is determined whether there is any behavior that deviates from the expected. Once the fault is confirmed, the adaptive kernel density estimation algorithm begins to work. The algorithm can dynamically adjust its kernel function width according to the actual data to more accurately estimate the probability distribution of electrical parameters under the fault state. The advantage of this method is that it not only considers the overall distribution characteristics of the data, but also adapts to local changes within the data, providing a more accurate description of the fault characteristics. For example, in the application scenario of a wind power generation system, it is assumed that the fluctuation of the inverter output current in a certain period of time is significantly increased and exceeds the normal range. This may be due to performance degradation or short circuit of a certain IGBT module. At this time, the adaptive kernel density estimation algorithm will be used to analyze the current data during this period. The algorithm will automatically adjust the kernel function parameters to adapt to this new data pattern and generate a detailed fault feature probability density curve. This curve not only shows the probability of the fault, but also reveals the various possible fault modes and their relative frequency of occurrence. For example, if the curve shows a high probability density value in a certain interval, this may mean that the abnormal electrical parameters corresponding to the interval are the key factors causing the current fault. In this way, not only can the fault source be quickly located, but also reliable data support can be provided for subsequent topology analysis and fault propagation path identification, and ultimately help formulate effective repair strategies. In short, through the above steps, accurate diagnosis and in-depth analysis of IGBT inverter faults can be achieved to ensure the stable operation of the power electronics system.

[0024] Step S3, performing a topological structure analysis on the IGBT inverter based on the fault feature probability density curve to obtain a fault feature topological map.

[0025] Specifically, the process of topological structure analysis of IGBT inverter based on fault feature probability density curve to obtain fault feature topological map is to transform the probability distribution characteristics of electrical working parameters into intuitive graphical representation to reveal the relationship between components inside the system and potential fault paths. First, the fault feature probability density curve generated by adaptive kernel density estimation algorithm provides detailed information about abnormal behavior of electrical parameters, which not only reflects the possibility of fault occurrence, but also contains the spatial distribution characteristics of fault mode. Next, by further processing and parsing these probability density data, a topological structure diagram reflecting the connection relationship and health status of each component inside the IGBT inverter can be constructed. Specifically, this process involves identifying which nodes or connections exhibit abnormal behavior under fault conditions and mapping them to the actual physical structure of the inverter. For example, in the application scenario of wind power generation system, it is assumed that after analysis, it is found that the probability density curve of the inverter output current in a certain period of time shows an area that deviates significantly from the normal value, which indicates that some electrical parameters in this period of time have abnormal fluctuations. In order to deeply understand the reasons behind this anomaly, it is necessary to map these probability density data to the specific topological structure of the inverter. In this way, specific IGBT modules or circuit branches associated with abnormal data can be identified, and a fault feature topology map can be drawn. This map not only shows how the various components are connected, but also marks possible fault points and affected paths. For example, if a certain IGBT module is found to have a higher fault probability density value than other modules, this may mean that there is a potential problem with the module or a problem with its connection line. In this way, not only can the impact range and propagation path of the fault be intuitively seen, but also clear input data can be provided for subsequent graph attention network analysis. Ultimately, this series of steps helps to achieve a comprehensive diagnosis of IGBT inverter faults and ensure the stability and reliability of power electronic systems. In summary, by performing topological structure analysis based on the fault feature probability density curve, the fault mode and its propagation mechanism inside the IGBT inverter can be effectively revealed, providing a solid foundation for formulating accurate repair strategies.

[0026] Step S4, performing graph attention network analysis on the fault feature topology map to obtain a set of fault propagation paths.

[0027] Specifically, the process of performing graph attention network analysis on the fault feature topology map to obtain a set of fault propagation paths is to use advanced graph neural network technology to identify and quantify the mutual influence between the various components inside the IGBT inverter and their mechanism of action when the fault occurs. First, based on the fault feature topology map generated in the previous step, the map details the connection relationship between the components inside the inverter and the location of potential fault points. Then, this information is input into the graph attention network, which is a deep learning model specially designed to process graph structured data. The graph attention network can automatically learn and assign different weights to different nodes and edges, thereby highlighting those parts that play a key role in the fault propagation process. Specifically, this process is achieved by calculating the importance score of each node relative to its neighboring nodes, which reflects the influence of the node in the fault propagation. For example, in the application scenario of a wind power generation system, it is assumed that a certain IGBT module has been determined to be abnormal after the previous step, and the corresponding fault feature topology map has been generated. At this time, applying the graph attention network to analyze the map can further reveal which other modules or circuit branches may be affected by the fault. The graph attention network iteratively updates the representation vector of each node and adjusts its importance score according to the characteristics of adjacent nodes, and finally outputs a set of high-confidence fault propagation paths. For example, if an IGBT module shows a high importance score, it indicates that it may be the main source of the fault or plays a key role in its propagation. In addition, if multiple nodes on a path have obtained high scores, this path is likely to be one of the main fault propagation paths. In this way, not only can the source of the fault and its impact range be accurately located, but also strong support can be provided for subsequent causal reasoning analysis. In short, by performing graph attention network analysis on the fault feature topology map, a complex set of fault propagation paths can be effectively identified, helping to formulate more accurate and effective repair strategies to ensure the stable operation of power electronic systems. This series of steps together constitutes a comprehensive and detailed fault diagnosis process, which greatly improves the efficiency and reliability of system maintenance.

[0028] Step S5, performing causal reasoning analysis on the IGBT inverter based on the fault propagation path set to obtain a fault type identification result, and formulating a corresponding fault repair strategy based on the fault type identification result.

[0029] Specifically, causal reasoning analysis of IGBT inverters based on a set of fault propagation paths to obtain fault type identification results and formulate corresponding fault repair strategies based on these results are key steps in the entire fault diagnosis process. First, the set of fault propagation paths generated by graph attention network analysis provides detailed information about how faults propagate within the system. These paths not only reveal the location of the fault source, but also show how the fault spreads through various components. Next, through causal reasoning analysis, the causal relationship between the nodes on each path can be further understood to determine the specific fault type. This process usually involves comprehensive consideration of factors such as electrical parameter anomalies, component aging or damage, and combines historical data and expert knowledge to infer the most likely cause of the fault. For example, in the application scenario of a wind power generation system, it is assumed that a major fault propagation path has been determined through the previous steps, and this path involves multiple IGBT modules and their connection lines. At this time, by performing causal reasoning analysis on each node on the path, the specific factors that caused the fault can be identified. For example, if the temperature sensor reading of a certain IGBT module is found to be abnormally high, and other components on this path also show similar trends, this may mean that the module has an overheating problem, which may be caused by a failure of the cooling system or excessive load. Further analysis can also reveal whether there are design defects, material aging, or other potential problems. Based on these analysis results, specific fault type identification conclusions can be drawn, such as "performance degradation caused by overheating of the IGBT module" or "local failure caused by short circuit of the circuit board." Once the fault type is determined, the next step is to develop a corresponding fault repair strategy. This usually includes short-term emergency measures and long-term preventive maintenance plans. For example, for the above overheating problem, short-term measures may include temporarily reducing the output power of the inverter or adding cooling equipment; while long-term strategies may involve replacing aging components, optimizing heat dissipation design, or improving monitoring systems to warn of similar problems in advance. In short, by performing causal reasoning analysis based on a set of fault propagation paths, not only can the fault type be accurately identified, but also a scientific basis can be provided for the formulation of effective repair strategies to ensure the efficient and stable operation of power electronic systems. This comprehensive and detailed fault diagnosis and repair method greatly improves the reliability and maintenance efficiency of the system.

[0030] In a specific embodiment, the probability distribution calculation of the original electrical operating parameters is performed by an adaptive kernel density estimation algorithm to obtain a fault feature probability density curve, including: Performing sliding window segmentation on the original electrical working parameters to obtain an electrical parameter subsequence set; and performing local density evaluation on the electrical parameter subsequence set using a preset adaptive kernel density estimation algorithm to obtain an electrical parameter subsequence density distribution; Performing feature space mapping on the electrical parameter subsequence density distribution by using a preset multi-dimensional Gaussian kernel function to obtain a feature space mapping result, and performing kernel density calculation based on the feature space mapping result to obtain a fault feature kernel density distribution; Performing marginal density integration on the fault feature kernel density distribution to obtain a marginal density distribution set, and performing multi-dimensional combination on the marginal density distribution set through probability density fusion to obtain a combined probability density sequence, and performing density curve fitting based on the combined probability density sequence to obtain a fault feature density curve set; A time series correlation analysis is performed on the set of fault feature density curves to obtain a time series correlation density curve, and fault feature extraction is performed on the time series correlation density curve to obtain a fault feature probability density curve.

[0031] Specifically, in the fault diagnosis method of the IGBT inverter, the process of calculating the probability distribution of the original electrical working parameters by the adaptive kernel density estimation algorithm to obtain the probability density curve of the fault characteristics is a complex process with multiple levels and steps. First, the original electrical working parameters are processed by sliding window segmentation to generate multiple electrical parameter subsequence sets. This process is to set a sliding window of fixed length, move step by step on the time axis, and intercept a section of data as a subsequence each time. For example, in a wind power generation system, assuming that the monitored current and voltage data are collected once per second, a 10-second sliding window can be set, which moves once every second to generate a series of 10-second data segments. These subsequences not only contain the specific values ​​at each time point, but also reflect the trend of data changes over time. Next, the preset adaptive kernel density estimation algorithm is used to perform local density evaluation on these electrical parameter subsequence sets to obtain the electrical parameter subsequence density distribution. Adaptive kernel density estimation is a non-parametric method that can automatically adjust the width of the kernel function according to the local characteristics of the data, thereby more accurately reflecting the true distribution of the data. In this process, the algorithm assigns a density value to each subsequence, and these density values ​​together constitute the electrical parameter subsequence density distribution. For example, in the above-mentioned example of the wind power generation system, if the current fluctuation is large in a certain period of time, the corresponding subsequence may obtain a higher density value, which indicates that there is a high possibility of abnormal phenomena in this period. Then, the electrical parameter subsequence density distribution is mapped to the feature space by the preset multidimensional Gaussian kernel function to obtain the feature space mapping result, and the kernel density is calculated based on this result to obtain the fault feature kernel density distribution. The multidimensional Gaussian kernel function can map one-dimensional or multi-dimensional data to a higher-dimensional space in order to better capture the intrinsic connection between the data. Specifically, this step converts the previously obtained density distribution into a new representation, making the similarities and differences between different subsequences more obvious. For example, in the wind power generation system, by mapping the current and voltage data with a multidimensional Gaussian kernel function, it can be revealed which time periods show similar fluctuation patterns and which time periods have significantly different data. Furthermore, the marginal density integration is performed on the fault feature kernel density distribution to obtain a marginal density distribution set, and the marginal density distribution set is combined in multiple dimensions through probability density fusion to obtain a combined probability density sequence, and density curve fitting is performed based on the combined probability density sequence to obtain a fault feature density curve set. Marginal density integration aims to reduce multidimensional data to one-dimensional or low-dimensional data for subsequent analysis. In this process, the algorithm integrates the density distribution on each dimension to obtain a series of marginal density distributions. Then, these marginal density distributions are combined through probability density fusion technology to form a comprehensive probability density sequence.For example, in a wind power generation system, after the marginal density integration of the current and voltage data, two independent marginal density distributions can be obtained, and then they are combined through the probability density fusion technology to generate a comprehensive probability density sequence containing the information of both. Finally, the fault feature density curve set is subjected to time series correlation analysis to obtain the time series correlation density curve, and the fault feature is extracted from the time series correlation density curve to obtain the fault feature probability density curve. Time series correlation analysis can reveal the correlation between data at different time points, which is crucial for understanding the development process of the fault. At this stage, the algorithm analyzes the correlation between the density curves at various time points and finds those parts with significant correlation. For example, in a wind power generation system, by performing time series correlation analysis on the current and voltage data over a period of time, it can be found that there is a significant correlation between the data in certain time periods, which may be caused by the failure of a specific component. Finally, by performing fault feature extraction on these correlation curves, a probability density curve that accurately describes the fault characteristics can be obtained, thereby providing a solid foundation for subsequent topological structure analysis and causal reasoning. In summary, through the above series of steps, starting from the original electrical working parameters, after complex processes such as sliding window segmentation, local density evaluation, feature space mapping, kernel density calculation, marginal density integration, probability density fusion, density curve fitting and time series correlation analysis, the fault feature probability density curve is finally obtained. This curve not only depicts in detail the distribution of electrical parameters of each component inside the IGBT inverter under normal and fault conditions, but also provides a key basis for subsequent fault diagnosis. For example, in the application scenario of wind power generation system, this method can accurately identify the fault source and its propagation path inside the inverter, so as to formulate an effective repair strategy to ensure the stable operation of the power electronic system. This comprehensive and detailed fault diagnosis method greatly improves the reliability and maintenance efficiency of the system.

[0032] In a specific embodiment, performing marginal density integration on the fault feature kernel density distribution to obtain a marginal density distribution set includes: Performing dimensionality reduction projection on the fault feature kernel density distribution to obtain a multidimensional density projection set, and performing integral variable separation on the multidimensional density projection set to obtain an integral variable matrix; Performing multiple integral calculations on the integral variable matrix through variable conversion to obtain a marginal probability body, and performing isosurface extraction on the marginal probability body to obtain a marginal density isosurface set; The marginal density isosurface set is fitted based on the mixed Gaussian distribution to obtain the marginal Gaussian component, and the density function is constructed based on the marginal Gaussian component to obtain the marginal density distribution set.

[0033] Specifically, in the fault diagnosis method of IGBT inverter, the process of performing marginal density integration on the fault feature kernel density distribution to obtain a marginal density distribution set is a complex and multi-level technical process. First, it is necessary to perform dimensionality reduction projection on the fault feature kernel density distribution to generate a multidimensional density projection set, and perform integral variable separation on these projection sets to obtain an integral variable matrix. This process aims to simplify high-dimensional data into a more easily processable form for subsequent analysis. For example, in a wind power generation system, assuming that the monitored current and voltage data are three-dimensional or higher-dimensional data sets, they can be converted into two-dimensional or one-dimensional data representations through dimensionality reduction projection technology. Specifically, dimensionality reduction projection can be achieved through principal component analysis (PCA) or other dimensionality reduction methods, so that the original high-dimensional data can be better understood and processed in a low-dimensional space. Next, by performing multiple integral calculations on the integral variable matrix, a marginal probability body is obtained, and isosurface extraction is performed on the marginal probability body to form a marginal density isosurface set. This process involves complex mathematical operations, especially multiple integral calculations, which can integrate information on different dimensions of multidimensional data. Variable transformation plays a key role in this process. The complexity of integral calculation can be simplified through appropriate transformation. For example, in the application scenario of wind power generation system, if a three-dimensional data set of current, voltage and temperature has been obtained, the marginal probability volume in each dimension can be calculated by performing variable transformation and multiple integration on these data. Then, through the isosurface extraction technology, marginal density isosurfaces can be drawn at different probability levels. These isosurfaces intuitively show the distribution of data in various dimensions and their mutual relationship. Subsequently, the marginal density isosurface set is distributedly fitted based on the mixed Gaussian distribution to obtain marginal Gaussian components, and the density function is constructed based on these marginal Gaussian components to finally obtain the marginal density distribution set. The mixed Gaussian distribution is a flexible and powerful statistical model that can fit multimodal data well. In this step, the algorithm automatically identifies and fits multiple Gaussian components, each of which represents a specific data mode or fault type. For example, in a wind power generation system, assuming that the current and voltage data in a certain period of time show a clear bimodal distribution, this may mean that there are two different operating states or fault modes within the system. Through the mixed Gaussian distribution model, these two modes can be accurately captured and the corresponding Gaussian components can be fitted respectively. Then, the density function is constructed based on these Gaussian components to form the final marginal density distribution set. To illustrate this process more specifically, we continue to take the wind power generation system as an example. Assume that the monitored current and voltage data show abnormal fluctuations in a certain period of time, and the fault feature kernel density distribution is generated through the previous steps. First, these high-dimensional data are converted into a two-dimensional representation through dimensionality reduction projection technology to obtain a multidimensional density projection set. Then, these projection sets are separated by integral variables to generate an integral variable matrix.Then, the integral variable matrix is ​​multi-integrated by variable transformation to obtain the marginal probability volume, and the marginal density isosurface set is further obtained by isosurface extraction technology. These isosurfaces intuitively show the distribution of current and voltage data at different probability levels. Next, the marginal density isosurface set is distributed fitted based on the mixed Gaussian distribution to obtain several marginal Gaussian components. Assume that two significant Gaussian components are identified in this process, one corresponding to the normal operating state and the other corresponding to the potential fault state. Based on these Gaussian components, the density function is constructed to form the final marginal density distribution set. These marginal density distributions not only reveal the distribution characteristics of the data in various dimensions, but also provide an important basis for the subsequent fault feature extraction. Through the above series of steps, starting from the fault feature kernel density distribution, after a series of complex operations such as dimensionality reduction projection, integral variable separation, multiple integral calculation, isosurface extraction, mixed Gaussian distribution fitting and density function construction, the marginal density distribution set is finally obtained. This set not only describes in detail the electrical parameter distribution of each component inside the IGBT inverter under normal and fault conditions, but also provides key data support for subsequent topological structure analysis, causal reasoning analysis, etc. For example, in the application scenario of wind power generation system, this method can accurately identify the fault source and its propagation path inside the inverter, so as to formulate an effective repair strategy and ensure the stable operation of the power electronic system. This comprehensive and detailed fault diagnosis method greatly improves the reliability and maintenance efficiency of the system.

[0034] In a specific embodiment, the topological structure analysis of the IGBT inverter is performed based on the fault feature probability density curve to obtain a fault feature topological map, including: The fault feature probability density curve is segmented and parsed to obtain a fault feature segmentation sequence, and a circuit topology mapping is performed based on the fault feature segmentation sequence to obtain a topology node association set; wherein the topology node association set includes a power tube node, a drive circuit node, and a protection circuit node; Reconstructing the spatial structure of the topological node association set by a preset adjacency matrix construction algorithm to obtain a fault topology adjacency matrix, and calculating the node degree based on the fault topology adjacency matrix to obtain a node degree feature set; Performing centrality analysis on the node degree feature set to obtain a node centrality matrix, and performing fault correlation calculation based on the node centrality matrix to obtain a fault correlation feature set; Performing a propagation dynamic analysis on the fault correlation feature set by using a preset fault propagation path algorithm to obtain a fault propagation dynamic sequence, and performing topological subgraph division on the fault propagation dynamic sequence to obtain a topological subgraph set; Fault feature fusion is performed on the topological subgraph set to obtain a fault feature topological description set, and a graph is generated based on the fault feature topological description set to obtain a fault feature topological graph.

[0035] Specifically, in the fault diagnosis method of the IGBT inverter, the process of topological structure analysis of the IGBT inverter based on the fault feature probability density curve to obtain the fault feature topology map is a complex and multi-level technical process. First, the fault feature probability density curve needs to be segmented and parsed to generate a fault feature segment sequence, and the circuit topology mapping is performed based on these segment sequences to obtain a topological node association set. This process aims to convert the probability density information into specific circuit components and their connection relationships. For example, in a wind power generation system, it is assumed that the monitored current and voltage data are processed to obtain a detailed fault feature probability density curve. By segmenting and parsing this curve, the fault features in different time periods can be identified and mapped to the actual circuit structure of the inverter. Specifically, this step not only involves key components such as power tube nodes, drive circuit nodes, and protection circuit nodes, but also reveals the interactions between them. Next, the spatial structure of the topological node association set is reconstructed by a preset adjacency matrix construction algorithm to generate a fault topology adjacency matrix, and the node degree is calculated based on the matrix to obtain a node degree feature set. The adjacency matrix is ​​a standard form of representing graph structure, which can clearly show the connection relationship between each node. In this process, the algorithm constructs an adjacency matrix based on the information in the topological node association set, and then calculates the degree of each node (that is, the number of connections with other nodes). For example, in the application scenario of a wind power generation system, if an abnormal fluctuation is detected in a certain IGBT module within a certain period of time, by constructing the adjacency matrix and calculating the node degree of the module and its related components, it can be determined which nodes have a high degree of connection. These high-connectivity nodes may be potential sources of faults or key points of the propagation path. Subsequently, the node degree feature set is analyzed for centrality, a node centrality matrix is ​​generated, and the fault correlation degree is calculated based on the matrix to obtain a fault correlation feature set. Centrality analysis aims to evaluate the importance of each node in the entire network, which is crucial to understanding the fault propagation mechanism. For example, in the above-mentioned wind power generation system, by performing centrality analysis on the node degree feature set, those nodes with high influence in the network can be identified, which may play a key role in the fault propagation process. Then, the fault correlation degree is calculated based on the node centrality matrix, which can further quantify the correlation strength between the nodes, thereby forming a fault correlation feature set. Next, the fault propagation path algorithm is used to perform a propagation dynamic analysis on the fault correlation feature set, generate a fault propagation dynamic sequence, and divide these sequences into topological subgraphs to obtain a topological subgraph set. The fault propagation path algorithm can simulate the diffusion process of faults in the system and help identify the main propagation path.For example, in a wind power generation system, assuming that a fault correlation feature set has been obtained, the fault propagation path algorithm can simulate how the fault propagates from one node to other nodes and generate the corresponding propagation dynamic sequence. Then, by performing topological subgraph division on these sequences, the entire system can be divided into multiple subgraphs, each of which represents a specific fault propagation path or area. Finally, the fault feature fusion is performed on the topological subgraph set to generate a fault feature topological description set, and a graph is generated based on the set to finally obtain a fault feature topological graph. Fault feature fusion aims to integrate the information of each subgraph to form a comprehensive fault description. For example, in the application scenario of a wind power generation system, by performing fault feature fusion on the topological subgraph set, the key information in each subgraph can be integrated to form a detailed fault feature description. Then, a graph is generated based on this description set, which not only shows the connection relationship between the components inside the inverter, but also marks the potential fault points and propagation paths. To illustrate this process more specifically, we continue to take the wind power generation system as an example. Assuming that the monitored current and voltage data show significant abnormal fluctuations, the fault feature probability density curve is generated through the previous steps. First, by segmenting and parsing this curve, the fault features in different time periods are identified and mapped to the actual circuit structure of the inverter to generate a topological node association set. Then, the adjacency matrix construction algorithm is used to reconstruct the spatial structure of these association sets, generate the fault topological adjacency matrix, and calculate the node degree feature set. Then, by performing centrality analysis on the node degree feature set, a node centrality matrix is ​​generated, and the fault correlation feature set is calculated based on the matrix. Next, the fault propagation path algorithm is used to perform propagation dynamic analysis on the fault correlation feature set to generate a fault propagation dynamic sequence, and these sequences are divided into topological subgraphs to obtain a topological subgraph set. Finally, by performing fault feature fusion on these subgraphs, a fault feature topology description set is generated, and a fault feature topology map is generated based on the set. This map not only describes in detail the connection relationship between the components inside the IGBT inverter, but also reveals the potential fault source and its propagation path, providing a key basis for subsequent causal reasoning analysis and repair strategy formulation. This comprehensive and detailed fault diagnosis method greatly improves the reliability and maintenance efficiency of power electronic systems.

[0036] In a specific embodiment, the performing graph attention network analysis on the fault feature topology map to obtain a set of fault propagation paths includes: Performing a graph domain representation conversion on the fault feature topology map to obtain a graph domain feature tensor, and performing attention weight calculation on the graph domain feature tensor to obtain a node attention weight matrix; wherein the matrix includes power tube node weights, drive circuit node weights, and protection circuit node weights; Performing a weight distribution analysis on the node attention weight matrix through a preset multi-head attention mechanism to obtain a multi-head attention feature set, and performing feature aggregation based on the multi-head attention feature set to obtain an aggregated feature matrix; Performing graph structure encoding on the aggregated feature matrix to obtain a graph structure encoding sequence, and performing node message transmission based on the graph structure encoding sequence to obtain a node message transmission matrix; Performing feature residual fusion on the node message transfer matrix to obtain a residual feature matrix, and performing path importance evaluation based on the residual feature matrix to obtain a path importance sequence; Performing path screening on the path importance sequence to obtain a critical propagation path set, and performing path feature extraction on the critical propagation path set to obtain a fault propagation path set.

[0037] Specifically, in the fault diagnosis method of IGBT inverter, the process of performing graph attention network analysis on the fault feature topology map to obtain the set of fault propagation paths is a complex and multi-level technical process. First, the fault feature topology map needs to be converted into a graph domain representation to generate graph domain feature tensors, and the attention weights of these tensors are calculated to obtain the node attention weight matrix. This process aims to convert each node and its connection relationship in the topological structure into a form that can be processed by the graph attention network. For example, in a wind power generation system, assuming that a detailed fault feature topology map has been obtained, each node (such as power tube node, drive circuit node and protection circuit node) and its connection relationship can be represented as a feature tensor by performing graph domain representation conversion on the map. Then, the attention weight of each node is calculated through the attention mechanism to form a node attention weight matrix. These weights reflect the role and importance of each node in the fault propagation process. Next, the node attention weight matrix is ​​analyzed for weight distribution through the preset multi-head attention mechanism to generate a multi-head attention feature set, and feature aggregation is performed based on these feature sets to obtain an aggregated feature matrix. The multi-head attention mechanism is a powerful technology that can capture the interaction between nodes in different dimensions. In this process, the algorithm analyzes the relationship between nodes from multiple perspectives to generate multiple attention feature sets. For example, in the application scenario of a wind power generation system, if an abnormal fluctuation is detected in a certain IGBT module within a certain period of time, by analyzing the multi-head attention mechanism of the module and its related components, it is possible to identify which nodes play a key role in fault propagation from different perspectives. Then, these multi-head attention feature sets are combined through feature aggregation technology to form a comprehensive aggregate feature matrix, which contains detailed information about all nodes and their interactions. Subsequently, the aggregate feature matrix is ​​graph-encoded to generate a graph-encoded sequence, and node message passing is performed based on the sequence to obtain a node message passing matrix. Graph structure encoding aims to convert the information in the graph structure into an easy-to-process form for subsequent analysis. In this process, the algorithm encodes the information of each node and its neighboring nodes and generates a corresponding graph structure encoding sequence. For example, in a wind power generation system, by graph-encoding the aggregate feature matrix, a series of encoding sequences can be generated, which describe the connection relationship and information interaction between each node and its neighboring nodes. Then, node message passing is performed based on these encoding sequences, which can simulate the information propagation process in the graph structure and generate a node message passing matrix. Furthermore, the feature residual fusion is performed on the node message passing matrix to generate a residual feature matrix, and the path importance is evaluated based on the matrix to obtain the path importance sequence. Feature residual fusion aims to integrate information at different levels and improve the performance of the model.In this process, the algorithm combines the original features and the features after multiple message passing to generate a residual feature matrix containing more information. For example, in the application scenario of wind power generation system, by performing feature residual fusion on the node message passing matrix, the subtle changes in the fault propagation process can be better captured. Then, based on the residual feature matrix, the path importance evaluation can be performed to quantify the importance of each possible fault propagation path and generate a path importance sequence. These path importances not only reflect the possibility of faults on each path, but also reveal their impact range in the entire system. Finally, the path importance sequence is screened to generate a set of key propagation paths, and path features are extracted for these key paths to finally obtain a set of fault propagation paths. Path screening aims to select the key paths that are most likely to cause fault propagation from many possible paths. For example, in a wind power generation system, assuming that the path importance sequence has been obtained, by screening these sequences, paths with high importance values ​​can be identified, which may be the main propagation paths of the fault. Then, by extracting features from these key paths, the specific characteristics of each path can be further analyzed to generate a detailed set of fault propagation paths. This set not only shows how faults propagate in the system, but also provides a key basis for subsequent causal reasoning analysis and repair strategy formulation. To illustrate this process more specifically, we continue to take the wind power generation system as an example. Assume that the monitored current and voltage data show significant abnormal fluctuations, and the fault feature topology map is generated after the previous steps. First, the graph domain representation is converted to generate a graph domain feature tensor, and the attention weight of each node is calculated to form a node attention weight matrix. Then, these weights are analyzed using the multi-head attention mechanism to generate a multi-head attention feature set, and an aggregate feature matrix is ​​generated through feature aggregation technology. Then, the aggregate feature matrix is ​​encoded by graph structure to generate a graph structure encoding sequence, and node message passing is performed based on these sequences to generate a node message passing matrix. Next, the node message passing matrix is ​​generated by performing feature residual fusion on the node message passing matrix, and the path importance is evaluated based on the matrix to generate a path importance sequence. Finally, the path importance sequence is screened to generate a set of key propagation paths, and features are extracted from these key paths to finally obtain a set of fault propagation paths. This collection not only depicts the fault propagation paths of the components inside the IGBT inverter in detail, but also reveals the potential fault sources and their propagation mechanisms, providing a key basis for subsequent causal reasoning analysis and repair strategy formulation. This comprehensive and detailed fault diagnosis method has greatly improved the reliability and maintenance efficiency of power electronic systems.

[0038] In a specific embodiment, the causal reasoning analysis is performed on the IGBT inverter based on the fault propagation path set to obtain a fault type identification result, and a corresponding fault repair strategy is formulated based on the fault type identification result, including: Constructing a causal graph for the set of fault propagation paths to obtain a fault causal relationship graph, and performing node dependency analysis based on the fault causal relationship graph to obtain a fault dependency matrix; The fault dependency matrix is ​​conditionally calculated by a preset Bayesian inference algorithm to obtain a fault condition probability table, and a fault propagation sequence analysis is performed based on the fault condition probability table to obtain a set of fault propagation chains; Performing causal strength evaluation on the fault propagation chain set to obtain a causal strength matrix, and locating the fault source based on the causal strength matrix to obtain a fault source feature set; The fault source feature set is mapped to a fault type by a preset multi-layer causal analysis algorithm to obtain a fault type feature matrix, and fault mode identification is performed based on the fault type feature matrix to obtain a fault type identification result; wherein the fault type identification result includes a short circuit fault type, an open circuit fault type, and a breakdown fault type; Generating a repair plan for the fault type identification result to obtain a set of fault repair strategies, and sorting the repair priorities based on the set of fault repair strategies to obtain a repair priority sequence, wherein the repair priority sequence includes an emergency repair strategy, a temporary repair strategy, and a preventive repair strategy; Repair instructions are generated based on the repair priority sequence to obtain a fault repair instruction set; wherein the fault repair instruction set includes a power tube replacement instruction, a drive circuit adjustment instruction and a protection parameter configuration instruction.

[0039] Specifically, in the fault diagnosis method of the IGBT inverter, the process of performing causal reasoning analysis on the IGBT inverter based on the set of fault propagation paths to obtain the fault type identification results and formulating the corresponding fault repair strategy based on these results is a complex and multi-level technical process. First, it is necessary to construct a causal graph for the set of fault propagation paths to generate a fault causal relationship graph, and perform node dependency analysis based on the graph to obtain a fault dependency matrix. This process aims to transform the fault propagation path into a form that can reveal the causal relationship between each node. For example, in a wind power generation system, assuming that a detailed set of fault propagation paths has been obtained, a fault causal relationship graph showing each component and its mutual dependency can be generated by constructing a causal graph for these paths. Then, through the node dependency analysis, the degree of influence of each node on other nodes can be quantified to form a fault dependency matrix. These dependencies not only reflect the direct connection between each node, but also reveal their role in the entire fault propagation process. Next, the fault dependency matrix is ​​conditionally calculated by a preset Bayesian inference algorithm to generate a fault conditional probability table, and a fault propagation sequence analysis is performed based on the table to obtain a set of fault propagation chains. The Bayesian inference algorithm is a powerful statistical method that can make inferences and decisions under uncertain conditions. In this process, the algorithm calculates the conditional probability of each possible fault propagation path based on the information in the fault dependency matrix. For example, in the application scenario of a wind power generation system, if an abnormal fluctuation is detected in a certain IGBT module within a certain period of time, the probability values ​​of various fault propagation paths can be calculated by performing Bayesian inference analysis on the module and its related components. Then, based on these probability values, a fault propagation sequence analysis is performed to generate a series of fault propagation chain sets. These chains not only show how faults propagate from one node to other nodes, but also provide the probability of each path occurring. Further, the causal strength evaluation is performed on the set of fault propagation chains to generate a causal strength matrix, and the fault source is located based on the matrix to obtain a set of fault source characteristics. The causal strength evaluation aims to quantify the importance of each fault propagation path and its impact on the entire system. In this process, the algorithm analyzes the nodes on each chain and their interactions to generate a causal strength matrix. For example, in a wind power generation system, by evaluating the causal strength of the fault propagation chain set, paths with high causal strength values ​​can be identified, which may be the main propagation paths of the fault. Then, based on the causal strength matrix, the fault source is located to determine which nodes or components are potential fault sources and generate a set of fault source features. These features not only describe the specific location of the fault source, but also reveal its possible fault mode. Subsequently, the fault source feature set is mapped to the fault type through the preset multi-layer causal analysis algorithm to generate a fault type feature matrix, and fault mode identification is performed based on the matrix to obtain the fault type identification result.The multi-layer causal analysis algorithm can deeply explore the reasons behind the fault source and help identify the specific fault type. In this process, the algorithm will combine the information in the fault source feature set for multi-layer analysis to generate a fault type feature matrix. For example, in the application scenario of wind power generation system, assuming that the fault source feature set has been obtained, by performing multi-layer causal analysis on the set, different types of fault modes such as short circuit fault, open circuit fault and breakdown fault can be identified. Then, fault mode identification is performed based on the fault type feature matrix, and finally a detailed fault type identification result is obtained. These identification results not only clarify the specific type of fault, but also provide a key basis for the subsequent repair strategy formulation. Next, the repair plan is generated for the fault type identification results, a fault repair strategy set is generated, and the repair priority sequence is obtained based on the set. The repair plan generation aims to formulate corresponding repair measures according to different fault types. In this process, the algorithm will combine the fault type identification results to generate a series of repair strategies, including emergency repair strategies, temporary repair strategies and preventive repair strategies. For example, in a wind power generation system, if a short circuit fault is identified in a certain IGBT module, an emergency repair can be performed through a power tube replacement instruction; if it is a drive circuit problem, a temporary repair can be performed through a drive circuit adjustment instruction; if the protection parameter is improperly configured, a preventive repair can be performed through a protection parameter configuration instruction. Then, priority is sorted based on these repair strategies to generate a repair priority sequence to ensure that the most urgent problems are handled first. Finally, repair instructions are generated based on the repair priority sequence to generate a set of fault repair instructions. This process aims to convert the repair priority sequence into specific operation instructions so as to perform actual repair work. For example, in the application scenario of a wind power generation system, assuming that a repair priority sequence has been obtained, a series of specific operation instructions can be generated by generating repair instructions for the sequence, such as a power tube replacement instruction, a drive circuit adjustment instruction, and a protection parameter configuration instruction. These instructions not only clarify the specific repair measures that need to be taken, but also guide maintenance personnel on how to efficiently implement these measures. For example, for power tube replacement instructions, maintenance personnel can replace damaged power tubes according to the detailed steps of the instructions; for drive circuit adjustment instructions, the inverter performance can be optimized by adjusting the drive circuit parameters; for protection parameter configuration instructions, similar faults can be prevented from happening again by reconfiguring the protection parameters. To illustrate this process more specifically, we continue to take the wind power generation system as an example. Assume that the monitored current and voltage data show significant abnormal fluctuations, and a set of fault propagation paths is generated through the previous steps. First, by constructing a causal graph for these paths, a fault causal relationship graph is generated, and node dependency analysis is performed based on the graph to generate a fault dependency matrix.Next, the Bayesian inference algorithm is used to calculate the conditional probability of the fault dependency matrix to generate a fault conditional probability table, and the fault propagation sequence analysis is performed based on the table to generate a set of fault propagation chains. Then, the causal strength of these chains is evaluated to generate a causal strength matrix, and the fault source is located based on the matrix to generate a set of fault source features. Next, the fault type is mapped to the fault source feature set through a multi-layer causal analysis algorithm to generate a fault type feature matrix, and the fault mode is identified based on the matrix to obtain detailed fault type identification results. For example, it is assumed that a short circuit fault is identified in a certain IGBT module, an open circuit fault is identified in another module, and a breakdown fault is identified in the third module. Then, the repair scheme is generated for these fault type identification results to generate a set of fault repair strategies, and the repair priority is sorted based on the set to generate a repair priority sequence. For example, a short circuit fault is marked as an emergency repair strategy, an open circuit fault is marked as a temporary repair strategy, and a breakdown fault is marked as a preventive repair strategy. Finally, repair instructions are generated based on the repair priority sequence to generate a set of fault repair instructions. For example, a power tube replacement instruction is generated for a short-circuit fault, a drive circuit adjustment instruction is generated for an open-circuit fault, and a protection parameter configuration instruction is generated for a breakdown fault. This comprehensive and detailed fault diagnosis and repair method greatly improves the reliability and maintenance efficiency of the power electronic system.

[0040] In a specific embodiment, the generating of repair instructions based on the repair priority sequence to obtain a fault repair instruction set includes: Decomposing the repair priority sequence into tasks to obtain a repair task sequence, and performing instruction timing planning on the repair task sequence to obtain a repair instruction timing table; Performing dependency analysis on the repair instruction timing table to obtain an instruction dependency matrix, and setting execution conditions based on the instruction dependency matrix to obtain an instruction execution condition set; The set of instruction execution conditions is defined by using preset security constraint rules to obtain an instruction constraint rule table, and an execution process is designed based on the instruction constraint rule table to obtain a repair instruction flow chart; Instruction codes are generated for the repair instruction flow chart to obtain a fault repair instruction set.

[0041] Specifically, in the process of fault repair of IGBT inverter, generating a specific set of fault repair instructions based on the repair priority sequence is a systematic and complex task. First, it is necessary to decompose the repair priority sequence into tasks to obtain a more detailed repair task sequence, and perform instruction timing planning on these task sequences to generate a repair instruction timing table. This process aims to refine the high-level repair strategy into specific executable task steps. For example, in the application scenario of wind power generation system, it is assumed that a detailed repair priority sequence has been obtained, in which short-circuit faults are marked as emergency repair strategies. By decomposing this strategy into tasks, a series of specific repair tasks can be generated, such as checking related circuits, locating specific fault points, etc. Then, according to the logical relationship and actual operation sequence between these tasks, instruction timing planning is performed to form a clear repair instruction timing table. The purpose of this is to ensure that each repair task can be executed in the correct order to avoid secondary faults caused by incorrect sequences. Next, the repair instruction timing table is subjected to dependency analysis to generate an instruction dependency matrix, and execution conditions are set based on the matrix to obtain an instruction execution condition set. Dependency analysis is a key step in identifying the interdependencies between various repair tasks. In this process, by analyzing the relationship between each instruction and other instructions in the timing table, it is possible to determine which tasks must be completed before other tasks and which tasks can be executed in parallel. For example, in a wind power generation system, if a task requires replacing a damaged power tube, it may be necessary to turn off the relevant power supply before this to ensure safety. By performing such dependency analysis on all instructions in the repair instruction timing table, an instruction dependency matrix can be generated, which reveals the dependencies between all instructions. Next, the execution conditions are set based on this dependency matrix to generate an instruction execution condition set. This step clarifies the prerequisites that must be met to execute each instruction. For example, before replacing the power tube, it is necessary to confirm that the power supply has been cut off, which is one of the prerequisites for executing this task. Subsequently, the instruction execution condition set is defined by the preset safety constraint rules, and an instruction constraint rule table is generated. Based on this table, the execution process is designed to obtain a repair instruction flow chart. Safety constraint rules are criteria set to ensure the safety and effectiveness of the entire repair process. For example, when handling power equipment, specific safety operating procedures must be followed, including wearing appropriate protective equipment and using insulating tools. In this case, by applying these safety constraints to the set of instruction execution conditions, a detailed instruction constraint rule table can be generated. This table not only lists the operating instructions for each task, but also emphasizes the safety issues that need to be paid attention to during implementation. Then, based on this instruction constraint rule table, the execution process is designed, and finally a repair instruction flow chart is generated. This flow chart graphically shows the entire repair process from start to finish, and each step and its corresponding constraint conditions are clear at a glance.This is crucial for guiding maintenance personnel to complete the repair work accurately and without error. Finally, the repair instruction flow chart is used to generate instruction codes to obtain a set of fault repair instructions. This step involves converting the abstract concepts formed in the previous stages into specific executable instruction codes. In this process, not only the technical implementation details should be considered, but also the code should be ensured to comply with the established safety standards and operating specifications. For example, in the application scenario of a wind power generation system, if the final repair instruction flow chart shows a series of specific steps on how to replace the power tube, then the corresponding instruction code will implement these steps one by one. This includes detailed operations such as how to disconnect the power supply, how to remove the old power tube, and how to install the new power tube. In addition, in order to improve efficiency and accuracy, automated tools or scripts can be used to assist in the generation of some instruction codes. In this way, a complete set of fault repair instructions can be generated quickly and accurately, providing strong support for the actual repair work. To illustrate this process more specifically, we continue to take the wind power generation system as an example. Assume that the fault diagnosis has been completed and a detailed repair priority sequence has been developed. First, the sequence is decomposed into tasks to obtain a specific repair task sequence. For example, for a short-circuit fault of an IGBT module, it is necessary to complete tasks such as checking the circuit connection status, measuring current and voltage parameters, removing faulty components, and installing new components in sequence. Then, the instruction timing is planned according to the actual operation sequence of these tasks to form a repair instruction timing table containing the execution sequence of each task. Next, the dependency analysis is performed on this timing table to generate an instruction dependency matrix, and the execution conditions of each task are set based on the matrix to obtain a set of instruction execution conditions. For example, for the task of removing the faulty component, its execution condition may be "it has been confirmed that the power supply has been cut off". On this basis, these execution conditions are further defined through preset safety constraint rules, and an instruction constraint rule table is generated. Based on this table, the execution process is designed to form a detailed repair instruction flow chart. This flow chart not only contains the specific operation methods of each repair step, but also marks the necessary safety precautions. Finally, the instruction code is generated for this flow chart to obtain the final set of fault repair instructions. These instructions can not only guide on-site maintenance personnel to complete the repair work efficiently, but also ensure the safety and reliability of the entire process. This systematic fault repair solution can greatly improve the maintenance efficiency and safety of power electronic equipment.

[0042] The above describes the fault diagnosis method of the IGBT inverter in the embodiment of the present invention. The following describes the fault diagnosis device of the IGBT inverter in the embodiment of the present invention. Figure 2 , an embodiment of a fault diagnosis device for an IGBT inverter in an embodiment of the present invention comprises: The sampling module 21 is used to perform high-frequency sampling on the electrical operating parameters of the IGBT inverter to obtain the original electrical operating parameters; A calculation module 22, configured to diagnose whether the IGBT inverter has a fault based on the original electrical operating parameters, and if so, to perform probability distribution calculation on the original electrical operating parameters by adaptive kernel density estimation to obtain a fault characteristic probability density curve; A first analysis module 23, configured to perform a topological structure analysis on the IGBT inverter based on the fault feature probability density curve to obtain a fault feature topological map; The second analysis module 24 is used to perform a graph attention network analysis on the fault feature topology map to obtain a set of fault propagation paths; The third analysis module 25 is used to perform causal reasoning analysis on the IGBT inverter based on the fault propagation path set to obtain a fault type identification result, and formulate a corresponding fault repair strategy based on the fault type identification result.

[0043] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0044] Reference Figure 3 The present invention also provides a computer device in an embodiment, wherein the internal structure of the computer device can be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0045] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0046] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0047] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0048] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0049] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A fault diagnosis method for an IGBT inverter, characterized in that: The following steps are involved: Perform high-frequency sampling on the electrical operating parameters of the IGBT inverter to obtain the original electrical operating parameters; Diagnose whether the IGBT inverter has a fault based on the original electrical operating parameters, and if so, perform probability distribution calculation on the original electrical operating parameters by using an adaptive kernel density estimation algorithm to obtain a fault feature probability density curve; Performing a topological structure analysis on the IGBT inverter based on the fault feature probability density curve to obtain a fault feature topological map; Performing graph attention network analysis on the fault feature topology map to obtain a set of fault propagation paths; A causal reasoning analysis is performed on the IGBT inverter based on the fault propagation path set to obtain a fault type identification result, and a corresponding fault repair strategy is formulated based on the fault type identification result.

2. The fault diagnosis method for an IGBT inverter according to claim 1, characterized in that: The method of performing probability distribution calculation on the original electrical operating parameters by using an adaptive kernel density estimation algorithm to obtain a fault feature probability density curve includes: Performing sliding window segmentation on the original electrical working parameters to obtain an electrical parameter subsequence set; and performing local density evaluation on the electrical parameter subsequence set using a preset adaptive kernel density estimation algorithm to obtain an electrical parameter subsequence density distribution; Performing feature space mapping on the electrical parameter subsequence density distribution by using a preset multi-dimensional Gaussian kernel function to obtain a feature space mapping result, and performing kernel density calculation based on the feature space mapping result to obtain a fault feature kernel density distribution; Performing marginal density integration on the fault feature kernel density distribution to obtain a marginal density distribution set, and performing multi-dimensional combination on the marginal density distribution set through probability density fusion to obtain a combined probability density sequence, and performing density curve fitting based on the combined probability density sequence to obtain a fault feature density curve set; A time series correlation analysis is performed on the set of fault feature density curves to obtain a time series correlation density curve, and fault feature extraction is performed on the time series correlation density curve to obtain a fault feature probability density curve.

3. The fault diagnosis method for an IGBT inverter according to claim 2, characterized in that: The marginal density integration of the fault feature kernel density distribution is performed to obtain a marginal density distribution set, including: Performing dimensionality reduction projection on the fault feature kernel density distribution to obtain a multidimensional density projection set, and performing integral variable separation on the multidimensional density projection set to obtain an integral variable matrix; Performing multiple integral calculations on the integral variable matrix through variable conversion to obtain a marginal probability body, and performing isosurface extraction on the marginal probability body to obtain a marginal density isosurface set; The marginal density isosurface set is fitted based on the mixed Gaussian distribution to obtain the marginal Gaussian component, and the density function is constructed based on the marginal Gaussian component to obtain the marginal density distribution set.

4. The fault diagnosis method for an IGBT inverter according to claim 1, characterized in that: The topological structure analysis of the IGBT inverter is performed based on the fault feature probability density curve to obtain a fault feature topological map, including: The fault feature probability density curve is segmented and parsed to obtain a fault feature segmentation sequence, and a circuit topology mapping is performed based on the fault feature segmentation sequence to obtain a topology node association set; wherein the topology node association set includes a power tube node, a drive circuit node, and a protection circuit node; Reconstructing the spatial structure of the topological node association set by a preset adjacency matrix construction algorithm to obtain a fault topology adjacency matrix, and calculating the node degree based on the fault topology adjacency matrix to obtain a node degree feature set; Performing centrality analysis on the node degree feature set to obtain a node centrality matrix, and performing fault correlation calculation based on the node centrality matrix to obtain a fault correlation feature set; Performing a propagation dynamic analysis on the fault correlation feature set by using a preset fault propagation path algorithm to obtain a fault propagation dynamic sequence, and performing topological subgraph division on the fault propagation dynamic sequence to obtain a topological subgraph set; Fault feature fusion is performed on the topological subgraph set to obtain a fault feature topological description set, and a graph is generated based on the fault feature topological description set to obtain a fault feature topological graph.

5. The fault diagnosis method for an IGBT inverter according to claim 1, characterized in that: The performing graph attention network analysis on the fault feature topology map to obtain a set of fault propagation paths includes: Performing a graph domain representation conversion on the fault feature topology map to obtain a graph domain feature tensor, and performing attention weight calculation on the graph domain feature tensor to obtain a node attention weight matrix; wherein the matrix includes power tube node weights, drive circuit node weights, and protection circuit node weights; Performing a weight distribution analysis on the node attention weight matrix through a preset multi-head attention mechanism to obtain a multi-head attention feature set, and performing feature aggregation based on the multi-head attention feature set to obtain an aggregated feature matrix; Performing graph structure encoding on the aggregated feature matrix to obtain a graph structure encoding sequence, and performing node message transmission based on the graph structure encoding sequence to obtain a node message transmission matrix; Performing feature residual fusion on the node message transfer matrix to obtain a residual feature matrix, and performing path importance evaluation based on the residual feature matrix to obtain a path importance sequence; Performing path screening on the path importance sequence to obtain a critical propagation path set, and performing path feature extraction on the critical propagation path set to obtain a fault propagation path set.

6. The fault diagnosis method for an IGBT inverter according to claim 1, characterized in that: The performing causal reasoning analysis on the IGBT inverter based on the fault propagation path set to obtain a fault type identification result, and formulating a corresponding fault repair strategy based on the fault type identification result, includes: Constructing a causal graph for the set of fault propagation paths to obtain a fault causal relationship graph, and performing node dependency analysis based on the fault causal relationship graph to obtain a fault dependency matrix; The fault dependency matrix is ​​conditionally calculated by a preset Bayesian inference algorithm to obtain a fault condition probability table, and a fault propagation sequence analysis is performed based on the fault condition probability table to obtain a set of fault propagation chains; Performing causal strength evaluation on the fault propagation chain set to obtain a causal strength matrix, and locating the fault source based on the causal strength matrix to obtain a fault source feature set; The fault source feature set is mapped to a fault type by a preset multi-layer causal analysis algorithm to obtain a fault type feature matrix, and fault mode identification is performed based on the fault type feature matrix to obtain a fault type identification result; wherein the fault type identification result includes a short circuit fault type, an open circuit fault type, and a breakdown fault type; Generating a repair plan for the fault type identification result to obtain a set of fault repair strategies, and sorting the repair priorities based on the set of fault repair strategies to obtain a repair priority sequence, wherein the repair priority sequence includes an emergency repair strategy, a temporary repair strategy, and a preventive repair strategy; Repair instructions are generated based on the repair priority sequence to obtain a fault repair instruction set; wherein the fault repair instruction set includes a power tube replacement instruction, a drive circuit adjustment instruction and a protection parameter configuration instruction.

7. The fault diagnosis method for an IGBT inverter according to claim 6, characterized in that: The generating of repair instructions based on the repair priority sequence to obtain a fault repair instruction set includes: Decomposing the repair priority sequence into tasks to obtain a repair task sequence, and performing instruction timing planning on the repair task sequence to obtain a repair instruction timing table; Performing dependency analysis on the repair instruction timing table to obtain an instruction dependency matrix, and setting execution conditions based on the instruction dependency matrix to obtain an instruction execution condition set; The set of instruction execution conditions is defined by using preset security constraint rules to obtain an instruction constraint rule table, and an execution process is designed based on the instruction constraint rule table to obtain a repair instruction flow chart; Instruction codes are generated for the repair instruction flow chart to obtain a fault repair instruction set.

8. A fault diagnosis device for an IGBT inverter, characterized in that: include: A sampling module is used to perform high-frequency sampling on the electrical operating parameters of the IGBT inverter to obtain the original electrical operating parameters; A calculation module, used for diagnosing whether the IGBT inverter has a fault based on the original electrical operating parameters, and if so, performing probability distribution calculation on the original electrical operating parameters by an adaptive kernel density estimation algorithm to obtain a fault characteristic probability density curve; A first analysis module, configured to perform a topological structure analysis on the IGBT inverter based on the fault feature probability density curve to obtain a fault feature topological map; The second analysis module is used to perform a graph attention network analysis on the fault feature topology map to obtain a set of fault propagation paths; The third analysis module is used to perform causal reasoning analysis on the IGBT inverter based on the fault propagation path set to obtain a fault type identification result, and formulate a corresponding fault repair strategy based on the fault type identification result.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images