Transformer multi-mode fault diagnosis method and device

Abstract

The invention relates to a multi-mode fault diagnosis method and device for a transformer. The method comprises the following steps: acquiring multi-source heterogeneous data of the transformer and carrying out standardized preprocessing on the multi-source heterogeneous data; according to data characteristics of different modes in the multi-source heterogeneous data, hierarchical feature extraction is carried out on the multi-source heterogeneous data after standardization preprocessing, and feature vectors are obtained; fusing the feature vectors based on a gated cross attention mechanism and the information entropy of the feature vectors to obtain fused features; and obtaining a fault diagnosis result according to the fusion feature. The reliability of transformer fault diagnosis can be improved.

Description

Technical Field

[0001] This application relates to the field of power equipment condition monitoring technology, and in particular to a method and apparatus for diagnosing multi-mode faults in transformers. Background Technology

[0002] Transformers operate under complex conditions, and their potential faults are characterized by changes in various physical and chemical parameters. Current transformer diagnostic techniques largely rely on a limited number of data points (such as oil chromatography and partial discharge analysis). However, the reliability of current transformer fault diagnosis methods is insufficient. Summary of the Invention

[0003] Therefore, it is necessary to provide a method and apparatus for multimodal transformer fault diagnosis that can improve the reliability of transformer fault diagnosis, in order to address the above-mentioned technical problems.

[0004] Firstly, this application provides a method for diagnosing multi-mode faults in transformers, the method comprising:

[0005] Acquire multi-source heterogeneous data of transformers and perform standardized preprocessing;

[0006] Based on the data characteristics of different modalities in multi-source heterogeneous data, hierarchical feature extraction is performed on the standardized preprocessed multi-source heterogeneous data to obtain feature vectors;

[0007] Based on the gated cross-attention mechanism and the information entropy of feature vectors, feature vectors are fused to obtain fused features;

[0008] Based on the fusion characteristics, obtain the fault diagnosis results.

[0009] In one embodiment, based on a gated cross-attention mechanism and the information entropy of the feature vectors, the feature vectors are fused to obtain fused features, including:

[0010] Retrieve the query vector, key vector, and value vector of feature vectors for different modalities;

[0011] Based on a gated cross-attention mechanism, cross-modal features are obtained from query vectors, key vectors, and value vectors;

[0012] The information entropy of feature vectors from different modalities is obtained, and dynamic weighted fusion is performed based on the information entropy and cross-modal features to obtain fused features.

[0013] In one embodiment, acquiring multi-source heterogeneous data of the transformer and performing standardized preprocessing includes:

[0014] Acquire multi-source heterogeneous data of the transformer and divide the multi-source heterogeneous data according to different modes; different modes include at least graphic modes, time-series modes and structured modes;

[0015] The data of the graph modality is subjected to graph standardization, the data of the time-series modality is subjected to time-series standardization, and the data of the structured modality is subjected to missing value imputation and the generation of derived features.

[0016] In one embodiment, based on the data characteristics of different modalities in the multi-source heterogeneous data, hierarchical feature extraction is performed on the standardized preprocessed multi-source heterogeneous data to obtain feature vectors, including:

[0017] Graph modality feature extraction is performed on the graph modality data to obtain graph modality feature vectors;

[0018] Temporal modality features are extracted from the temporal modal data to obtain temporal fusion feature vectors;

[0019] Structured modal features are extracted from the structured modal data to obtain structured feature vectors.

[0020] In one embodiment, obtaining diagnostic results based on fusion features includes:

[0021] The fused features are input into the multi-task learning head to perform multi-task diagnostic output, and the diagnostic results are output. The diagnostic results include fault type, fault severity, and fault location prediction.

[0022] In one embodiment, the method further includes:

[0023] Based on the decision-level DS evidence theory, the confidence level of the diagnostic results is obtained; the confidence level is used to represent the accuracy of the diagnostic results.

[0024] Secondly, this application provides a transformer multi-mode fault diagnosis device, the device comprising:

[0025] The data acquisition module is used to acquire multi-source heterogeneous data of the transformer and perform standardized preprocessing.

[0026] The feature extraction module is used to perform hierarchical feature extraction on the standardized preprocessed multi-source heterogeneous data based on the data characteristics of different modalities in the multi-source heterogeneous data, and obtain feature vectors.

[0027] The feature fusion module is used to fuse feature vectors based on the gated cross-attention mechanism and the information entropy of feature vectors to obtain fused features;

[0028] The diagnostic module is used to obtain diagnostic results based on fusion features.

[0029] Thirdly, this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0030] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0031] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method.

[0032] The aforementioned transformer multimodal fault diagnosis method and apparatus acquires multi-source heterogeneous data of the transformer and performs standardized preprocessing; based on the data characteristics of different modes in the multi-source heterogeneous data, hierarchical feature extraction is performed on the standardized preprocessed multi-source heterogeneous data to obtain feature vectors; based on the gated cross-attention mechanism and the information entropy of the feature vectors, the feature vectors are fused to obtain fused features; and based on the fused features, the fault diagnosis results are obtained, thereby improving the reliability of transformer fault diagnosis. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a diagram illustrating the application environment of a transformer multi-mode fault diagnosis method in one embodiment.

[0035] Figure 2 This is a flowchart illustrating a transformer multi-mode fault diagnosis method in one embodiment;

[0036] Figure 3 This is a schematic diagram of a multimodal information fusion model framework in one embodiment;

[0037] Figure 4 This is a schematic diagram of a temporal modality multi-branch feature extraction network structure in one embodiment;

[0038] Figure 5 This is a schematic diagram of a three-level fusion processing flow based on a gated cross-attention mechanism in one embodiment;

[0039] Figure 6 This is a schematic diagram illustrating the fusion of multi-task learning and DS evidence theory decision-making in one embodiment;

[0040] Figure 7 This is a structural block diagram of a transformer multimode fault diagnosis device in one embodiment;

[0041] Figure 8 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0043] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0044] Transformers operate under complex conditions, and their potential faults are characterized by changes in various physical and chemical parameters. Traditional transformer diagnostic techniques rely heavily on a limited number of data points (such as oil chromatography and partial discharge), which presents the following bottlenecks:

[0045] Information silos: Different monitoring systems (such as partial discharge monitoring, oil chromatography monitoring, and online temperature measurement) operate independently, and there is a lack of correlation analysis between the data, making it difficult to form an overall understanding of the status.

[0046] One-sided characteristics: Data of a single type can only reflect one aspect of the fault. For example, oil chromatography is sensitive to overheating and discharge, but not to mechanical loosening; vibration signals are sensitive to mechanical faults, but are easily affected by load and environmental interference.

[0047] Early faults are often missed: Many faults initially exhibit only weak characteristics in specific types of signals, making them easy to miss using single-mode analysis methods.

[0048] Misjudgment of complex faults: For complex faults caused by multiple factors, methods based on single modes or simple fusion are prone to misjudgment. For example, it is difficult to distinguish whether the UHF signal is caused by internal discharge or external interference.

[0049] Although the concept of multimodal fusion has emerged in traditional technologies, the lack of a systematic, layered fusion architecture makes it impossible to fully explore the deep value of data when faced with such massive, heterogeneous (including graph, time series, and structured data) and diverse sampling frequencies of transformer data.

[0050] Based on the aforementioned traditional technologies, embodiments of this application provide a method and system for joint analysis and diagnosis of multimodal operation data of transformers by combining deep learning and information fusion technologies. It is understood that this application relates to the field of power equipment condition monitoring and intelligent fault diagnosis technology.

[0051] Specifically, this application aims to address the shortcomings of traditional technologies by providing a diagnostic method and system that can systematically integrate a full range of transformer status parameters, including acoustic (ultrasonic), electrical (ultra-high frequency, high frequency), insulation status (core grounding current), magnetic (leakage magnetic field), thermal (oil temperature, environment), chemical (oil chromatography), mechanical (vibration), and operating conditions (load, voltage, power). By constructing a three-level fusion architecture of "data level - feature level - decision level," it achieves comprehensive perception of transformer status, accurate early warning of faults, and precise diagnosis of complex faults.

[0052] It should be noted that the beneficial effects or technical problems solved by the embodiments of this application are not limited to this one, but may also be other implicit or related problems. For details, please refer to the description of the embodiments below.

[0053] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0054] The transformer multi-mode fault diagnosis method provided in this application can be applied to, for example... Figure 1 The application environment shown is a fault diagnosis system, which includes a data acquisition layer, an edge computing layer, a cloud-based intelligent analysis layer, and a human-computer interaction layer. Optionally, the functions of each layer are as follows:

[0055] Data acquisition layer: Composed of various sensors and monitoring devices deployed on and around the transformer body, used to synchronously collect multi-source heterogeneous data. Specifically, this includes a full range of status parameters such as transformer acoustic (ultrasonic), electrical (ultra-high frequency, high frequency), insulation status (core grounding current), magnetic (leakage magnetic field), thermal (oil temperature, environment), chemical (oil chromatography), mechanical (vibration), and operational (load, oil level).

[0056] Edge computing layer: Composed of edge computing gateways deployed within the substation. These gateways have built-in preprocessing and feature extraction modules, responsible for: data reception and caching (receiving raw data from the data acquisition layer); data preprocessing (performing data alignment, invalid data removal, and standardization); lightweight feature extraction (running a lightweight deep learning model to perform preliminary feature extraction on the data, reducing network transmission pressure); and data uploading (uploading the preprocessed data or feature vectors to the cloud-based intelligent analysis layer).

[0057] Cloud-based intelligent analysis layer: Deployed on cloud servers or data centers, it is the runtime environment for core algorithms and includes the following software functional modules:

[0058] Data governance module: responsible for receiving and storing data from the edge computing layer, and performing deeper data cleaning, archiving and management.

[0059] The hierarchical feature extraction module includes a map feature extraction submodule (using a deep CNN to process partial discharge maps), a temporal feature extraction submodule (using a multi-branch 1D-CNN-BiGRU hybrid network to process temporal data), and a structured feature extraction submodule (using a feature-enhanced MLP to process structured data such as oil chromatography).

[0060] Multimodal fusion module: It includes a gated cross-attention calculation submodule and an information entropy dynamic weighted fusion submodule, which are used to realize deep interaction and adaptive fusion of features from the three modalities.

[0061] Multi-task diagnostic module: Receives the fused features and, through a multi-task learning network, synchronously outputs preliminary diagnostic results of fault type, severity, and fault location.

[0062] Decision fusion module: Using DS evidence theory, the output of the multi-task diagnosis module is fused with the output of multiple single-modal expert models at the decision level to generate the final diagnostic conclusion and confidence level.

[0063] Human-computer interaction layer: This includes a web application server and client, used to display real-time data, diagnostic results, historical trends, alarm information to users, and to provide system configuration interfaces.

[0064] Optionally, the above-mentioned fault diagnosis system can further form such as Figure 3 This paper presents a hierarchical, multi-granular, multimodal information fusion model framework.

[0065] In one exemplary embodiment, such as Figure 2 As shown, a multi-mode fault diagnosis method for transformers is provided, which is applied to... Figure 1 Taking the fault diagnosis system in the example, the explanation includes the following steps 202 to 208. Wherein:

[0066] Step 202: Obtain multi-source heterogeneous data of the transformer and perform standardized preprocessing.

[0067] Among them, multi-source heterogeneous data can refer to transformer data of different modes.

[0068] For example, multi-source heterogeneous data of transformers are acquired and different standardization preprocessing is performed according to the different modes of these multi-source heterogeneous data.

[0069] Step 204: Based on the data characteristics of different modalities in the multi-source heterogeneous data, perform hierarchical feature extraction on the standardized preprocessed multi-source heterogeneous data to obtain feature vectors.

[0070] For example, based on the data characteristics of different modalities in multi-source heterogeneous data, high-dimensional feature extraction, i.e. hierarchical feature extraction, is performed on the multi-source heterogeneous data of different modalities after standardization and preprocessing to obtain feature vectors.

[0071] Step 206: Based on the gated cross-attention mechanism and the information entropy of the feature vector, the feature vector is fused to obtain the fused feature.

[0072] Specifically, by fusing feature vectors based on a gated cross-attention mechanism and the information entropy of feature vectors, fused features are obtained, thereby achieving adaptive and refined information interaction between different modalities.

[0073] Step 208: Obtain the fault diagnosis results based on the fusion features.

[0074] For example, a multi-task learning mechanism is used to obtain fault diagnosis results based on fusion features.

[0075] In the aforementioned multimodal fault diagnosis method for transformers, multi-source heterogeneous data of the transformer is acquired and standardized preprocessed. Based on the data characteristics of different modes in the multi-source heterogeneous data, hierarchical feature extraction is performed on the standardized preprocessed multi-source heterogeneous data to obtain feature vectors. Based on the gated cross-attention mechanism and the information entropy of the feature vectors, the feature vectors are fused to obtain fused features. Based on the fused features, the fault diagnosis results are obtained. Thus, through a systematic hierarchical fusion architecture, the deep value of the data is fully explored, and the transformer fault diagnosis results are obtained on the basis of information fusion of transformer data, thereby further improving the reliability of transformer fault diagnosis.

[0076] In an exemplary embodiment, step 202, acquiring multi-source heterogeneous data of the transformer and performing standardization preprocessing, includes:

[0077] Acquire multi-source heterogeneous data of the transformer and divide the multi-source heterogeneous data according to different modes; different modes include at least graphic modes, time-series modes and structured modes;

[0078] The data of the graph modality is subjected to graph standardization, the data of the time-series modality is subjected to time-series standardization, and the data of the structured modality is subjected to missing value imputation and the generation of derived features.

[0079] Among them, the multi-source heterogeneous data of transformers can also be referred to as raw data.

[0080] For example, multi-source heterogeneous data of transformers is acquired, and multi-source heterogeneous data preprocessing and mode segmentation are performed. Specifically, this includes: dividing the collected raw data into three modes, and performing standardization preprocessing on each mode, wherein:

[0081] Graphical Modality (X~G~): Includes ultrasonic partial discharge patterns, ultra-high frequency partial discharge PRPD patterns, and high frequency partial discharge waveform patterns. Preprocessing includes: pattern normalization to a uniform size (e.g., 128x128), background noise removal, and color / grayscale normalization.

[0082] Time-Series Modality (X~T~): Includes vibration signals (x, y, z axes), core grounding current, load current, oil temperature, and ambient temperature and humidity. Preprocessing includes: data alignment, resampling to a unified frequency, detrending, and Z-score normalization.

[0083] Structured Modality (X~S~): Includes oil chromatographic data (H2, CH4, C2H2, C2H4, C2H6, CO, CO2 content and various ratios, such as C2H2 / C2H4, CH4 / H2, etc.) and oil level. Preprocessing includes: missing value imputation, generation of derived features based on the three-ratio method or David's triangle method, etc.

[0084] In an exemplary embodiment, step 204 involves performing hierarchical feature extraction on the standardized preprocessed multi-source heterogeneous data based on the data characteristics of different modalities in the multi-source heterogeneous data to obtain feature vectors. This includes: performing graph modality feature extraction on the graph modality data to obtain graph modality feature vectors; performing time-series modality feature extraction on the time-series modality data to obtain time-series fusion feature vectors; and performing structured modality feature extraction on the structured modality data to obtain structured feature vectors.

[0085] For example, high-dimensional feature extraction is performed using a dedicated deep neural network, specifically for data characteristics of different modalities, including:

[0086] Atlas Modal Feature Extraction: A deep convolutional neural network is used to extract deep spatial features from the localized atlas. Specifically, an EfficientNet-B3 model pre-trained on ImageNet is used, its top-level classifier is removed, and the output of the final global average pooling layer is used as the feature vector, i.e., the atlas modal feature vector h. g R1536. This network can effectively capture key information such as discharge phase, discharge quantity distribution, and mode shape of partial discharge patterns.

[0087] Temporal modality feature extraction (its network structure is as follows) Figure 4 As shown): A multi-branch one-dimensional CNN-BiGRU hybrid network is employed. Independent 1D-CNN branches are established for vibration, electrical / current (core grounding current, load), and thermal (oil temperature, ambient temperature) to extract their respective local temporal patterns. Each branch is followed by a BiGRU network to capture long-term dependencies. Finally, the final hidden states of the BiGRU branches are concatenated and fused through a fully connected layer to obtain the temporal fusion feature vector h. t R512.

[0088] Structured Modal Feature Extraction: A Feature-Enhanced Multilayer Perceptron (MLP) is employed. First, n features from the original oil chromatogram, including gas content, ratio, and oil level, are used to construct an input vector. Then, a deep MLP with residual connections is used for nonlinear transformation and feature interaction to learn complex feature relationships beyond the traditional ratio method, outputting a structured feature vector h. s R256.

[0089] In an exemplary embodiment, step 206, based on the gated cross-attention mechanism and the information entropy of the feature vectors, fuses the feature vectors to obtain fused features, including:

[0090] Retrieve the query vector, key vector, and value vector of feature vectors for different modalities;

[0091] Based on a gated cross-attention mechanism, cross-modal features are obtained from query vectors, key vectors, and value vectors;

[0092] The information entropy of feature vectors from different modalities is obtained, and dynamic weighted fusion is performed based on the information entropy and cross-modal features to obtain fused features.

[0093] Optionally, a three-level fusion process is performed based on a gated cross-attention mechanism, as detailed in the following steps: Figure 5 As shown, it includes the following steps:

[0094] First, generate the query (Q), key (K), and value (V) vectors:

[0095] ;

[0096] in, This refers to the query vector, which is used to match key vectors from other modalities. This refers to the key vector, which is used to calculate similarity with the query vector; This refers to the value vector, which contains the feature information of that modality; , , The learnable weight matrix consists of three independent linear transformation matrices that transform the same feature... Projected into three different "semantic spaces" to assume different roles; This refers to the feature vector of mode m (such as the features extracted from spectrograms, time series, and structured modes); m refers to the mode identifier, g represents spectrograms, t represents time series, and s represents structured mode.

[0097] Then, gated cross-attention calculation is performed. Taking the temporal modality of graph modality query as an example, the specific steps include:

[0098] Calculate attention score:

[0099] ;

[0100] in, This refers to the distribution of attention weights between graphical modes and temporal modes; This refers to the query vector of the graph modality, which is derived from the graph feature vector h. g Obtained through linear transformation; This refers to the key vector of the temporal mode, which is composed of the temporal feature vector h. t Obtained through a linear transformation; T refers to the transpose operator; This refers to the transpose of the key vector in a temporal modality; This refers to the dot product operation, which yields a similarity score that reflects the degree of association between the query of a graph modality and the various keys of a time-series modality. This refers to the dimension of the key vector, used for scaling to prevent gradient explosion; This refers to the scaling factor; This refers to the normalization function.

[0101] Subsequently, a learnable gating vector is introduced through a gating mechanism to control the flow of information:

[0102] ;

[0103] in, This refers to the gating vector; This refers to the gate control weight matrix, whose function is to learn how to calculate the gate value based on the feature combination of two modes; This refers to the gate control bias vector, which provides a baseline bias to ensure a reasonable gate value even when all input features are zero. This refers to the spectral modality eigenvector; This refers to the temporal modality feature vector; This refers to the vector concatenation operation, which joins the beginning and end of two vectors. This is the Sigmoid function.

[0104] Calculate the gating weights: ;

[0105] in, This is element-wise multiplication; This refers to attention output that has been finely modulated through a gating mechanism; subscript This refers to the information obtained by the spectral mode from the temporal mode, which flows from the temporal sequence to the spectral sequence. This refers to the value vector of the temporal modalities. Gating mechanisms can suppress the flow of irrelevant or noisy information, thereby enhancing the robustness of the fusion.

[0106] Generate cross-modal features: For each modality, aggregate gated attention information from all other modalities, specifically including:

[0107] Cross-modal features of the graph: ;

[0108] Cross-modal characteristics of time series: ;

[0109] Cross-modal characteristics of the structure: .

[0110] in, Cross-modal features of a graph modality refer to enhanced features formed after the graph modality absorbs gated complementary information from temporal and structured modalities. [ || || ] represents a triple vector concatenation operation. MLP refers to a multilayer perceptron. The value vector of the spectral mode itself, The value vector of the time-series mode itself. The value vector of the structured mode itself; This refers to the gating attention information flowing from the temporal modality to the graph modality; the same applies to A with different subscripts. The graph mode refers to the information obtained by querying the time-series mode, which is the information flowing from the time series to the graph. The same applies to other subscripts.

[0111] Perform dynamic weighted fusion of information entropy: calculate the initial feature vectors of each modality. Information entropy As a measure of its information content, the higher the entropy value, the more complex the modality data and the greater the amount of information it may contain.

[0112] Information entropy is calculated by estimating the distribution of feature vector values: ;

[0113] in, Shannon information entropy is used to measure feature vectors. The uncertainty or richness of the information contained therein. : Represents the feature vector The probability distribution of each value in the equation.

[0114] The weights are obtained by normalizing based on the information entropy:

[0115]

[0116] in, For dynamic fusion weights; subscript m: represents modality identifier, m (Graphics, time series, structured representation); The exponential function maps entropy values ​​to the positive real number field;

[0117] Denominator: It is a normalization factor used to ensure that the sum of all weights is 1, thereby forming an effective probability distribution.

[0118] Calculate the final fused feature vector: ;

[0119] in, For the final fused feature vector, For the dynamic fusion weights of the graph modes, For dynamic fusion weights of time-series modalities, For the dynamic fusion weights of structured modes.

[0120] In an exemplary embodiment, step 208, obtaining a diagnostic result based on the fusion features, includes:

[0121] The fused features are input into the multi-task learning head to perform multi-task diagnostic output, and the diagnostic results are output. The diagnostic results include fault type, fault severity, and fault location prediction.

[0122] For example, the process of multi-task learning and decision-level DS evidence fusion can be as follows: Figure 6 As shown, the multi-task diagnostic output may specifically include: Input a multi-task learning head and output: Task 1 Fault Types: Output the probabilities of the following types via Softmax: normal, arc discharge, overheating, partial discharge, insulation dampness, mechanical loosening, etc. Task 2: Fault Severity: Output a regression value representing the severity through a linear layer. (e.g., 0-1). Task 3: Fault Location Prediction: Output the probability of possible fault locations using Softmax, such as windings, core, etc. .

[0123] In one exemplary embodiment, the method further includes: obtaining the confidence level of the diagnostic result based on the decision-level DS evidence theory; the confidence level is used to represent the accuracy of the diagnostic result.

[0124] Optionally, refer to Figure 6 The Decision-level Dempster-Shafer Evidence Theory integrates the following:

[0125] The multi-task output of this model As evidence source A, where P fault This refers to the probability of fault type, S severity For the severity of the fault, P location The probability of the fault location is given. Simultaneously, three independent, high-performance expert models are trained: a CNN model using only graph data, an LSTM model using only time-series data, and a gradient boosting tree model using only structured data. Their outputs are used as evidence sources B, C, and D.

[0126] Dempster's combination rule was used to synthesize four sources of evidence, A, B, C, and D. This rule effectively handles conflicts between pieces of evidence; when multiple sources of evidence agree on a certain fault type, the reliability of that type is significantly improved; when evidence conflicts, it reflects the uncertainty of the diagnosis.

[0127] Finally, the synthesized fault type, severity, and fault location (part) with the highest confidence level are output as the final diagnostic conclusion, and an overall confidence level is given.

[0128] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0129] Based on the same inventive concept, this application also provides a transformer multi-mode fault diagnosis device for implementing the above-described transformer multi-mode fault diagnosis method. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations of one or more embodiments of the transformer multi-mode fault diagnosis device provided below can be found in the limitations of the transformer multi-mode fault diagnosis method described above, and will not be repeated here.

[0130] In one exemplary embodiment, such as Figure 7 As shown, a transformer multi-mode fault diagnosis device 900 is provided, including: a data acquisition module 901, a feature extraction module 902, a feature fusion module 903, and a diagnosis module 904, wherein:

[0131] The data acquisition module 901 is used to acquire multi-source heterogeneous data of the transformer and perform standardized preprocessing.

[0132] The feature extraction module 902 is used to perform hierarchical feature extraction on the standardized preprocessed multi-source heterogeneous data according to the data characteristics of different modalities in the multi-source heterogeneous data, and obtain feature vectors.

[0133] The feature fusion module 903 is used to fuse feature vectors based on the gated cross-attention mechanism and the information entropy of feature vectors to obtain fused features;

[0134] The diagnostic module 904 is used to obtain diagnostic results based on the fusion features.

[0135] In one embodiment, the feature fusion module 903 is further configured to: obtain query vectors, key vectors, and value vectors of feature vectors of different modalities; obtain cross-modal features based on the query vectors, key vectors, and value vectors using a gated cross-attention mechanism; obtain the information entropy of feature vectors of different modalities; and perform dynamic weighted fusion based on the information entropy and cross-modal features to obtain fused features.

[0136] In one embodiment, the data acquisition module 901 is further configured to: acquire multi-source heterogeneous data of the transformer and divide the multi-source heterogeneous data according to different modes; the different modes include at least graph modes, time-series modes and structured modes; perform graph standardization on the data of graph modes, perform time-series standardization on the data of time-series modes, and perform missing value imputation and generate derived features on the data of structured modes.

[0137] In one embodiment, the feature extraction module 902 is further configured to: extract spectral modality features from the spectral modality data to obtain spectral modality feature vectors; extract temporal modality features from the temporal modality data to obtain temporal fusion feature vectors; and extract structured modality features from the structured modality data to obtain structured feature vectors.

[0138] In one embodiment, the diagnostic module 904 is further configured to: input the fused features into the multi-task learning head, perform multi-task diagnostic output, and output diagnostic results; the diagnostic results include fault type, fault severity, and fault location prediction.

[0139] In one embodiment, the apparatus further includes: a confidence level acquisition module, used to acquire the confidence level of the diagnostic result based on the decision-level DS evidence theory; the confidence level is used to represent the accuracy of the diagnostic result.

[0140] Each module in the aforementioned transformer multimode fault diagnosis device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0141] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 8As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores transformer fault diagnosis data. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a transformer multi-mode fault diagnosis method.

[0142] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0143] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described above.

[0144] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0145] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the method described above.

[0146] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0147] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0148] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for diagnosing multi-mode faults in transformers, characterized in that, The method includes: Acquire the multi-source heterogeneous data of the transformer and perform standardized preprocessing; Based on the data characteristics of different modalities in the multi-source heterogeneous data, hierarchical feature extraction is performed on the standardized preprocessed multi-source heterogeneous data to obtain feature vectors; Based on the gated cross-attention mechanism and the information entropy of the feature vector, the feature vector is fused to obtain fused features; Based on the fusion features, the fault diagnosis results are obtained.

2. The method according to claim 1, characterized in that, The method of fusing the feature vectors based on the gated cross-attention mechanism and the information entropy of the feature vectors to obtain fused features includes: Obtain the query vector, key vector, and value vector of the feature vectors for different modalities; Based on a gated cross-attention mechanism, cross-modal features are obtained according to the query vector, the key vector, and the value vector; The information entropy of the feature vectors of different modalities is obtained, and the information entropy and the cross-modal features are dynamically weighted and fused to obtain the fused features.

3. The method according to claim 1, characterized in that, The process of acquiring multi-source heterogeneous data of the transformer and performing standardized preprocessing includes: The multi-source heterogeneous data of the transformer is acquired, and the multi-source heterogeneous data is divided according to different modes; the different modes include at least graphical modes, time-series modes, and structured modes; The data of the graph modality is subjected to graph standardization, the data of the time-series modality is subjected to time-series standardization, and the data of the structured modality is subjected to missing value imputation and derived feature generation.

4. The method according to claim 3, characterized in that, The step of extracting hierarchical features from the standardized preprocessed multi-source heterogeneous data based on the data characteristics of different modalities in the multi-source heterogeneous data to obtain feature vectors includes: The data of the spectral modes are subjected to spectral mode feature extraction to obtain spectral mode feature vectors; Temporal modality features are extracted from the data of the aforementioned temporal modalities to obtain a temporal fusion feature vector; The structured modality data is subjected to structured modality feature extraction to obtain structured feature vectors.

5. The method according to claim 1, characterized in that, The step of obtaining a diagnostic result based on the fusion features includes: The fused features are input into a multi-task learning head to perform multi-task diagnostic output, and the diagnostic results are output; the diagnostic results include fault type, fault severity, and fault location prediction.

6. The method according to any one of claims 1 to 4, characterized in that, The method further includes: Based on the decision-level DS evidence theory, the confidence level of the diagnostic result is obtained; the confidence level is used to represent the accuracy of the diagnostic result.

7. A transformer multi-mode fault diagnosis device, characterized in that, The device includes: The data acquisition module is used to acquire multi-source heterogeneous data of the transformer and perform standardized preprocessing. The feature extraction module is used to perform hierarchical feature extraction on the standardized preprocessed multi-source heterogeneous data according to the data characteristics of different modalities in the multi-source heterogeneous data, and obtain feature vectors. The feature fusion module is used to fuse the feature vectors based on the gated cross-attention mechanism and the information entropy of the feature vectors to obtain fused features; The diagnostic module is used to obtain diagnostic results based on the fusion features.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images