Transformer partial discharge intelligent diagnosis method based on improved transformation network

Abstract

The invention discloses a transformer partial discharge intelligent diagnosis method based on an improved transformation network, and relates to the technical field of transformer fault diagnosis, and the method comprises the steps: collecting a transformer partial discharge signal, and carrying out the preprocessing of the partial discharge signal, and obtaining a time sequence; respectively converting the time sequence into feature images by adopting a Markov conversion field (MTF) algorithm, a GASF algorithm and a GADF algorithm; processing the feature image through an improved transformation network model to obtain a fusion feature vector; and inputting the fusion feature vector into a fault diagnosis model, and outputting a transformer partial discharge fault diagnosis result. According to the method, the oil gap discharge recognition rate is improved in the test by combining an interpretable diagram reasoning mechanism of environmental parameters and signal characteristics, so that fault diagnosis spans from a traditional judgment mode based on a threshold value to an intelligent mode based on causal logic reasoning.

Description

Technical Field

[0001] The invention relates to the technical field of transformer fault diagnosis, in particular to an intelligent diagnosis method for transformer partial discharge based on an improved conversion network. Background Art

[0002] As the scale of power grid continues to expand, the health of transformers, as the core equipment of the power system, directly affects the safe and stable operation of the power grid. Partial discharge is one of the important manifestations of insulation defects inside transformers. Long-term partial discharge may cause insulation material degradation and eventually lead to equipment failure. Accurately identifying partial discharge patterns and performing intelligent diagnosis are crucial to ensuring the reliability of the power grid.

[0003] At present, the detection and diagnosis methods of partial discharge mainly include: 1) Method based on mathematical model: The partial discharge characteristics are analyzed by establishing a physical model of the transformer. However, due to the complex structure of the transformer, it is difficult to accurately model the transformer, which limits the diagnostic accuracy. In addition, the mathematical model relies on a large number of physical parameters and assumptions and is easily affected by environmental changes.

[0004] 2) Signal processing-based methods: Time-frequency analysis methods (such as wavelet transform, short-time Fourier transform, etc.) are used to extract discharge characteristics, but they are sensitive to noise and lack the ability to effectively model complex patterns. Existing signal processing methods are often difficult to adapt to complex partial discharge signals under various working conditions of transformers.

[0005] 3) Machine learning-based methods: Neural networks or deep learning are used for fault classification. However, traditional deep learning methods have high computational complexity, strong dependence on large-scale labeled data, and insufficient generalization ability. In addition, many deep learning methods lack interpretability, making it difficult for diagnostic results to directly guide maintenance and fault prevention.

[0006] In view of the above problems, the present invention proposes an intelligent diagnosis method for transformer partial discharge based on an improved transformation network, which can improve the feature extraction capability of the discharge signal and enhance the generalization performance of the model. Summary of the invention

[0007] In view of the fact that existing transformer partial discharge diagnosis methods generally have the problems of insufficient feature extraction capability, single feature fusion method and low diagnosis accuracy, the present invention is proposed.

[0008] Therefore, the problem to be solved by the present invention is how to enhance the feature extraction and feature fusion capabilities of partial discharge signals by improving the transformation network model, thereby achieving efficient and accurate partial discharge fault diagnosis under complex working conditions.

[0009] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a transformer partial discharge intelligent diagnosis method based on an improved conversion network, which comprises: Collecting partial discharge signals of the transformer, and preprocessing the partial discharge signals to obtain a time series; The time series are converted into feature images by using a Markov transformation field MTF algorithm, a Gram angular domain field GASF algorithm and a Gram angular difference field GADF algorithm respectively; Processing the feature image through an improved transformation network model to obtain a fused feature vector; The fused feature vector is input into a fault diagnosis model, and a fault diagnosis result of partial discharge of the transformer is output.

[0010] As a preferred solution of the transformer partial discharge intelligent diagnosis method based on the improved conversion network described in the present invention, the self-supervision-graph reasoning dual-drive framework is introduced to enhance the generalization ability of the improved conversion network model to form a fault diagnosis model, including: Randomly masking a preset proportion of time-frequency blocks on the feature image to obtain a masked feature image; The masked feature image is reconstructed through the Transformer decoder to obtain a reconstructed feature image; Random noise and scale transformation are applied to the same partial discharge signal to generate positive sample pairs, and partial discharge signals of different categories are used as negative samples to construct a comparative learning sample set. Based on the comparative learning sample set, the improved transformation network model is trained to obtain a fault diagnosis model; The fused feature vector is input into the fault diagnosis model, and a dynamic causal graph network is constructed in the graph reasoning stage; Set the discharge pulse characteristics, time-frequency energy distribution and environmental parameters as graph nodes; Calculate the causal relationship strength between the graph nodes through the attention mechanism to obtain the node relationship weight; Based on the node relationship weights, the graph nodes are iteratively updated, a fault propagation path is established, and a fault diagnosis result is output.

[0011] As a preferred solution of the transformer partial discharge intelligent diagnosis method based on the improved transformation network of the present invention, the feature image is processed by the improved transformation network model to obtain a fused feature vector, including: Inputting the feature image into the improved transform network model, wherein the improved transform network model has a time-frequency analysis branch and a time-domain analysis branch; the time-frequency analysis branch uses a spatial attention mechanism to extract three-dimensional time-frequency features; the time-domain analysis branch captures the time-domain features of the local discharge pulse through a deep separable dilated convolutional network, and uses BiLSTM temporal attention to construct an improved transform network model; Based on the three-dimensional time-frequency features and the time domain features, a heterogeneous feature alignment fusion mechanism is adopted to transform the three-dimensional time-frequency features into The spatial dimension is mapped to the time domain features The temporal dimension of A multi-head cross-attention mechanism is used to establish cross-modal feature association, and the dynamic weighted fusion of the three-dimensional time-frequency features and time-domain features is realized through attention weight calculation; The cross-attention features are residually connected with the dimensionally aligned time-frequency features, and the fused feature vector is obtained through layer normalization.

[0012] As a preferred solution of the transformer partial discharge intelligent diagnosis method based on the improved conversion network described in the present invention, wherein: the characteristic image includes an MTF characteristic image, a GASF characteristic image and a GADF characteristic image; the conversion method of the MTF characteristic image is, Discretize the time series into Q states, calculate the state transition probability, and generate the state transition matrix P, where the elements of the state transition matrix are represents the probability of transitioning from state i to state j; Arrange all state transition probabilities according to state numbers and construct the Markov transition matrix.

[0013] As a preferred solution of the transformer partial discharge intelligent diagnosis method based on the improved conversion network described in the present invention, the conversion method of the GASF characteristic image is: For the timing sequence signal Perform angle mapping to obtain the first angle mapping signal ; Mapping the signal based on the first angle , construct the Gram sum matrix, where the calculation formula of the Gram sum matrix elements is as follows: ; ·in, is the element in the ath row and bth column of the Gram angle and field matrix, For time point The corresponding angle, For time point The corresponding angle, is the timestamp of the i-th data point in the original time series, is the timestamp of the jth data point in the original time series.

[0014] As a preferred solution of the transformer partial discharge intelligent diagnosis method based on the improved conversion network described in the present invention, the conversion method of the GADF feature image is: The signal of the timing sequence Converting into a second angle mapping signal through an arc cosine function; Based on the second angle mapping signal, a Gram difference matrix is ​​constructed, wherein the calculation formula of the Gram difference matrix elements is as follows: ; in, is the element in the ath row and bth column of the Gram angle difference field matrix, For time point The corresponding angle, For time point The corresponding angle, is the timestamp of the i-th data point in the original time series, is the timestamp of the jth data point in the original time series.

[0015] As a preferred solution of the transformer partial discharge intelligent diagnosis method based on the improved conversion network described in the present invention, the method for obtaining the time series is as follows: Ultrasonic detection, ultra-high frequency (UHF) detection and electromagnetic wave detection methods are used to synchronously collect partial discharge signals of transformers, and Gaussian filtering and adaptive denoising are performed on the partial discharge signals in turn to obtain denoised time domain signals; Dynamic threshold segmentation is performed on the time domain signal to extract the discharge pulse fragments, and a time series is generated through wavelet transformation.

[0016] In a second aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the intelligent diagnosis method for partial discharge of a transformer based on an improved conversion network as described in the first aspect of the present invention are implemented.

[0017] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the intelligent diagnosis method for partial discharge of a transformer based on an improved conversion network as described in the first aspect of the present invention are implemented.

[0018] Compared with the prior art, the present invention has the following beneficial effects: The multimodal time-frequency feature joint characterization method is adopted, and the three characterization mechanisms of Markov transition field, Gram angle domain field and Gram angle difference field are integrated to deeply analyze the signal from the dimensions of state transition probability, phase change law and energy difference, thus realizing the robust characterization of partial discharge signal under strong noise interference, effectively improving the signal discrimination ability and providing more discriminative input features for subsequent diagnosis. An improved heterogeneous transformation network structure is designed. The time-frequency analysis branch based on the spatial attention mechanism is processed in coordination with the time-domain analysis branch that combines the deep separable dilated convolution and the temporal attention mechanism. The branch features are aligned using deformable convolution, and then dynamic weighted fusion is achieved through a multi-head cross-attention mechanism. This structure can fully capture the complex coupling characteristics of partial discharge signals in the spatial and temporal domains, significantly improving the model's ability to recognize fine-grained features and the accuracy of overall diagnosis. Self-supervised learning and graph reasoning mechanisms are introduced, masking reconstruction and contrastive learning are used to enhance the discriminability of features, and a dynamic causal graph network is constructed to iteratively update the fault propagation path. This dual-drive mechanism not only improves the model's generalization ability for unknown devices or scenarios, but also significantly enhances the precise positioning of the local discharge source, effectively supporting cross-domain diagnosis needs in a multi-device environment. At the engineering application level, multiple signal preprocessing modules such as Gaussian filtering, adaptive denoising and wavelet transform are integrated to enhance the adaptability to complex background noise; when facing actual ultra-high frequency signals and strong noise ratio interference, the system can still maintain a high recognition rate and low false alarm rate, ensuring the stable operation of the system in a real power grid environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them: Figure 1 This is a flow chart of the intelligent diagnosis method for partial discharge of transformer based on improved conversion network in Example 1. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0021] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0023] Example 1 Reference Figure 1 , which is the first embodiment of the present invention, and provides a transformer partial discharge intelligent diagnosis method based on an improved conversion network, comprising: S1: Collect the transformer partial discharge signal and pre-process the partial discharge signal to obtain a time series.

[0024] Specifically, ultrasonic detection, ultra-high frequency (UHF) detection and electromagnetic wave detection methods are used to synchronously collect partial discharge signals of transformers, and Gaussian filtering and adaptive denoising are performed on the partial discharge signals in turn to obtain denoised time domain signals.

[0025] Furthermore, the time domain signal is segmented by dynamic threshold, the discharge pulse fragments are extracted, and the time series is generated by wavelet transform.

[0026] It should be noted that the timing sequence includes energy distribution information of the signal in the frequency band of 0.1-100 MHz.

[0027] S2: The Markov transition field MTF algorithm, the Gram angular domain field GASF algorithm and the Gram angular difference field GADF algorithm are used to convert the time series into feature images respectively.

[0028] Specifically, the characteristic images include MTF characteristic images, GASF characteristic images and GADF characteristic images.

[0029] Furthermore, the method for transforming the MTF feature image is to discretize the time series into Q states, calculate the state transition probability, and generate the state transition matrix P, where the elements of the state transition matrix are represents the probability of transitioning from state i to state j.

[0030] It should be noted that the value range of the time series is evenly divided into Q intervals, each interval represents a state. Assuming Q=4, the time series will be divided into four intervals, each interval corresponds to a state {1,2,3,4}.

[0031] Preferably, by traversing all time series, the number of times of transferring from state i to state j is counted, and the transition probability of each state is calculated. The specific formula for the state transition probability is as follows: ; Among them, is the probability of transferring from state i to state j, N is the total length of the time series, that is, the number of observed data points, and are two discrete states in the state space respectively, is the state of the system at time point t, is an indicator function, which takes the value of 1 when the condition is true and 0 otherwise.

[0032] Exemplarily, for the numerator: calculate the number of times of transferring from state to state in the partial discharge signal, that is, the number of all transfers in the time series that satisfy "being in state i at the current moment and transferring to state j at the next moment"; for the denominator: calculate the number of times of state appearing in the partial discharge signal, that is, count the total number of times of state i appearing in the time series (excluding the last moment t = N because there is no subsequent transfer).

[0033] Furthermore, arrange all state transition probabilities according to the state serial numbers to construct a Markov transition matrix. The elements of the Markov transition matrix are calculated by the following formula: ; Among them, is the state transition probability from coding time point i to coding time point j, forming a two-dimensional feature image, is the basic state transition probability, obtained by statistical analysis of historical data, is the discrete state of the i-th transition state in the partial discharge signal, is the discrete state of the j-th transition state in the partial discharge signal.

[0034] Specifically, the conversion method of the GASF feature image is to perform angle mapping on the signal of the time series to obtain the first angle mapping signal , and the specific formula is as follows: .

[0035] Furthermore, based on the first angle mapping signal , construct a Gram sum matrix, and the calculation formula for the elements of the Gram sum matrix is as follows: ; · Among them, is the element in the ath row and bth column of the Gram angle and field matrix, For time point The corresponding angle, For time point The corresponding angle, is the timestamp of the i-th data point in the original time series, is the timestamp of the jth data point in the original time series.

[0036] Specifically, the method for transforming the GADF feature image is to transform the signal of the time series Converted into the second angle mapping signal through the arc cosine function , the specific formula is as follows: .

[0037] Furthermore, based on the second angle mapping signal, a Gram difference matrix is ​​constructed, wherein the calculation formula of the Gram difference matrix elements is as follows: ; in, is the element in the ath row and bth column of the Gram angle difference field matrix, For time point The corresponding angle, For time point The corresponding angle, is the timestamp of the i-th data point in the original time series, is the timestamp of the jth data point in the original time series.

[0038] Furthermore, the MTF feature image, GASF feature image and GADF feature image are weightedly fused through the spatial attention mechanism to output a three-dimensional time-frequency-time domain joint feature image.

[0039] S3: Process the feature image through the improved transformation network model to obtain a fused feature vector.

[0040] Specifically, the feature image is input into an improved transformation network model, wherein the improved transformation network model has a time-frequency parsing branch and a time domain parsing branch.

[0041] Preferably, the time-frequency analysis branch uses a spatial attention mechanism to extract three-dimensional time-frequency features; the time-domain analysis branch captures the time-domain features of local discharge pulses through a deep separable dilated convolutional network, and uses BiLSTM temporal attention to construct an improved transformation network model.

[0042] Furthermore, based on the three-dimensional time-frequency features and time domain features, a heterogeneous feature alignment fusion mechanism is adopted to transform the three-dimensional time-frequency features into The spatial dimension is mapped to the time domain features The time dimension.

[0043] It should be noted that the query matrix is ​​calculated , the bond matrix Sum Matrix ,in , , is the weight matrix, which is optimized through end-to-end training.

[0044] Furthermore, a multi-head cross-attention mechanism is used to establish cross-modal feature associations, and the dynamic weighted fusion of three-dimensional time-frequency features and time domain features is realized through attention weight calculation.

[0045] The relevant formula for excellent, cross-modal feature association is as follows: ; in, is the output of the attention mechanism, is the query matrix, which is used to retrieve information related to the current task in the input sequence. is the key matrix, which is matched with the query matrix to determine the weight distribution of different positions. is the transposed form of the key matrix, used for matrix multiplication with the query matrix, is a value matrix that stores the actual information content of the input sequence, is the key vector The dimension size of is a normalized exponential function that maps a real vector to a probability distribution.

[0046] Specifically, the cross-attention features are residually connected with the dimensionally aligned time-frequency features, and the fused feature vector is obtained through layer normalization to achieve a three-dimensional joint representation of time, frequency and space.

[0047] Preferably, the specific formula for fusing feature vectors is as follows: ; in, is the fusion feature vector, is the layer normalization operation, is the three-dimensional time-frequency feature, It is a cross-modal or cross-level cross feature.

[0048] S4: Input the fused feature vector into the fault diagnosis model and output the fault diagnosis result of transformer partial discharge.

[0049] Specifically, the method for constructing the fault diagnosis model is to introduce a self-supervision-graph reasoning dual-drive framework to enhance the generalization ability of the improved transformation network model to form a fault diagnosis model.

[0050] Furthermore, a preset proportion of time-frequency blocks is randomly masked on the feature image to obtain a masked feature image; the masked feature image is reconstructed through the Transformer decoder to obtain a reconstructed feature image.

[0051] Preferably, the original signal is reconstructed through the Transformer decoder, forcing the network to learn the relevant formula of the inherent pattern of the discharge signal as follows: ; in, is the loss of the model’s ability to reconstruct the input data, is the total number of samples in the training data set, is the original time-frequency feature vector of the i-th sample, is the reconstructed time-frequency feature vector of the i-th sample.

[0052] Furthermore, random noise and scale transformation are applied to the same partial discharge signal to generate positive sample pairs, and partial discharge signals of different categories are used as negative samples to construct a comparative learning sample set to optimize feature discriminability. The relevant formula is as follows: ; in, is the comparison loss function value, B is the total number of samples in the training batch, Generate the projection feature vector of the i-th positive sample pair after different data enhancement for the same original sample, Generate the projection feature vector of the jth positive sample pair after different data enhancement for the same original sample, is the projected feature vector of other samples in the current training batch, is the temperature hyperparameter, which is used to adjust the distribution of similarity scores.

[0053] Specifically, based on the contrastive learning sample set, the improved transformation network model is trained to obtain a fault diagnosis model.

[0054] Furthermore, the fused feature vector is input into the fault diagnosis model, and a dynamic causal graph network is constructed in the graph reasoning stage; the discharge pulse characteristics, time-frequency energy distribution and environmental parameters are set as graph nodes.

[0055] Furthermore, the causal relationship strength between graph nodes is calculated through the attention mechanism to obtain the node relationship weight. The relevant formula is as follows: ; in, is the attention score of the i-th element and the j-th element in the input sequence, is the permutation of the eigenvector of the i-th element in the input sequence, is the learnable weight matrix, is the feature vector of the jth element in the input sequence, For vector and Dimension.

[0056] Specifically, based on the node relationship weights, the graph nodes are iteratively updated, the fault propagation path is established, and the fault diagnosis results are output.

[0057] Preferably, the fault diagnosis result includes a fault type probability distribution and a fault location result; the node representation is iteratively updated to capture the fault propagation path, and the specific formula is as follows: ; in, is the feature representation of the node at the l+1 layer after the l-th layer graph convolution operation, is a nonlinear activation function, is the degree matrix, which is a diagonal matrix. is the adjacency matrix, is a learnable parameter matrix used to linearly transform node features.

[0058] In an optional implementation, accurate determination of fault diagnosis results is achieved through a dynamic causal graph network, including: input is a 512-dimensional fused feature vector, and output is the probability distribution of five types of faults (such as insulation degradation, winding deformation, etc.); during the processing, the fused features are decomposed into three types of graph nodes: discharge pulses (256-dimensional time domain features), time-frequency energy distribution (128-dimensional frequency band energy features), and environmental parameters (128-dimensional sensor data); by constructing an initial node matrix, these nodes are used as input for subsequent processing; the multi-head attention mechanism is used to calculate the causal relationship weights between nodes, and the edges with weights greater than 0.5 are screened out as key causal paths; after iterative updates of three layers of graph convolutional layers, the node features are gradually optimized; the updated node features are spliced ​​with the global features and input into the classification layer to generate a fault probability output.

[0059] For example, when a 220kV transformer is working under high temperature (45°C), the fusion feature analysis shows that the high-frequency energy accounts for 85% and the pulse amplitude is 120mV; through dynamic causal graph network analysis, the fault propagation path is obtained as: "temperature → oil gap insulation drop → high-frequency energy surge → pulse mutation"; after the path is analyzed, the output oil gap discharge probability is 85%, and the false alarm rate is less than 2%, and the positioning error is only ±5cm, which is three times the positioning accuracy of the traditional method (±15cm).

[0060] In summary, the present invention achieved an oil gap discharge recognition rate of 98% in the test by combining the interpretable graph reasoning mechanism of environmental parameters and signal characteristics. This technology enables fault diagnosis to leap from the traditional threshold-based judgment mode to an intelligent mode based on causal logic reasoning, effectively improving the system's fault identification capability and diagnostic efficiency.

[0061] This embodiment also provides a computer device, which is applicable to the case of an intelligent diagnosis method for partial discharge of a transformer based on an improved conversion network, and includes a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to implement the intelligent diagnosis method for partial discharge of a transformer based on an improved conversion network as proposed in the above embodiment.

[0062] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device 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 and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0063] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the following steps are implemented: collecting transformer partial discharge signals, and preprocessing the partial discharge signals to obtain a time series; using a Markov transition field MTF algorithm, a Gram angle domain field GASF algorithm, and a Gram angle difference field GADF algorithm to convert the time series into a feature image respectively; processing the feature image through an improved transformation network model to obtain a fused feature vector; inputting the fused feature vector into a fault diagnosis model to output a fault diagnosis result of transformer partial discharge.

[0064] Example 2 Referring to Tables 1 to 4, which are the second embodiment of the present invention, this embodiment provides an intelligent diagnosis method for partial discharge of transformers based on an improved conversion network. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0065] Specifically, the partial discharge diagnosis of transformers in 500kV substations was used as an example for verification; the experiment used three detection methods (ultrasonic, ultra-high frequency UHF, and electromagnetic waves) to synchronously collect partial discharge signals to form a multimodal data set, including 1,200 groups of samples from a 500kV substation of the State Grid (covering five types of faults: bubbles in oil, suspended discharge, cone corona, creepage discharge, and surface discharge) and 3,500 groups of samples from the IEEE PES public data set (eight types of faults); the signal preprocessing stage uses Gaussian filtering (cutoff frequency 15kHz) combined with adaptive empirical mode decomposition (EEMD) to achieve noise suppression, and extracts time-frequency features through wavelet transform. When converting the signal into a feature image, the MTF algorithm uses 32 discrete states for division, and both GASF and GADF use 256×256 resolution.

[0066] Furthermore, the time-frequency analysis branch of the improved transformation network model uses 3 layers of attention-enhanced deep separable convolution (convolution kernel size is 3×3, expansion rate is 1, 2, 4), and the time domain analysis branch uses 2 layers of BiLSTM (hidden unit number is 128); the training adopts the self-supervision-graph reasoning dual-drive framework, the masking ratio is set to 15%, and the positive sample pair is constructed using Gaussian noise (σ=0.1) and random scaling (0.8-1.2 times) perturbations. 21 nodes are set in the dynamic causal graph network, including 12 discharge pulse feature nodes, 6 time-frequency energy distribution nodes, and 3 environmental parameter nodes (temperature, humidity, oil pressure), and multi-head attention (head number = 8) is used to calculate the relationship weights between nodes.

[0067] Furthermore, as shown in Table 1, the present invention is significantly superior to existing technical methods in many key indicators. In terms of diagnostic accuracy, the present invention reaches 98.7% in a 20dB noise environment, which is 20.4 percentage points higher than the traditional network method and 5.2 percentage points higher than the most advanced ViT-Base model; in the more challenging 10dB noise environment, the accuracy of the present invention is 95.3%, showing extremely strong noise resistance, 11.3 percentage points higher than ViT-Base. In particular, in the cross-device generalization ability test, the accuracy of the present invention is 92.3%, which is much higher than the 72.3% of the second place ViT-Base, indicating that the self-supervision-graph reasoning dual-drive framework of the present invention has significant advantages in generalization ability.

[0068] Table 1. Data comparison table of the present invention and the prior art

[0069] Specifically, in terms of fault location accuracy, the average location error of the present invention is only ±4.2cm, which is 77.3% lower than the ±18.5cm of the traditional network method. Specifically for different fault types, as shown in Table 2, the location errors of the present invention for oil-immersed discharge, cone corona and floating discharge are ±2.0cm, ±3.5cm and ±4.8cm respectively, which are significantly lower than the traditional method. This is due to the collaborative reasoning mechanism of environmental parameter nodes and time-frequency energy nodes in the dynamic causal graph network.

[0070] Table 2. Comparison of fault location accuracy

[0071] Furthermore, in terms of real-time performance and resource consumption, as shown in Table 3, the single-sample inference time of the present invention is only 20ms, which is 86.7% lower than the 150ms of the traditional network method and 83.3% lower than the 120ms of CNN-LSTM; the GPU memory occupancy is 1.2GB, which is only 31.6% of CNN-LSTM; the CPU utilization rate is 35%, which is much lower than the 95% of the traditional network method and 80% of CNN-LSTM. This shows that the present invention has successfully achieved model lightweighting through deep separable dilated convolution and BiLSTM temporal attention mechanism, with a parameter volume of only 18.7MB, which is 5.7% of ResNet-50 and 4.5% of ViT-Base.

[0072] Table 3. Comparison of real-time performance and resource consumption

[0073] Furthermore, in terms of public data set verification, as shown in Table 4, compared with the current SOTA method, the accuracy, recall and F1-Score of the present invention have been improved by 5.6%, 5.7% and 5.8% respectively, which fully verifies the advancement and effectiveness of the present method.

[0074] Table 4. Public dataset verification table

[0075] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A transformer partial discharge intelligent diagnosis method based on an improved conversion network, characterized in that: include, Collecting partial discharge signals of the transformer, and preprocessing the partial discharge signals to obtain a time series; The time series are converted into feature images by using a Markov transformation field MTF algorithm, a Gram angular domain field GASF algorithm and a Gram angular difference field GADF algorithm respectively; Processing the feature image through an improved transformation network model to obtain a fused feature vector; The fused feature vector is input into a fault diagnosis model, and a fault diagnosis result of partial discharge of the transformer is output.

2. The transformer partial discharge intelligent diagnosis method based on the improved conversion network as claimed in claim 1 is characterized in that: The self-supervision-graph reasoning dual-driven framework is introduced to enhance the generalization ability of the improved transformation network model and form a fault diagnosis model, including: Randomly masking a preset proportion of time-frequency blocks on the feature image to obtain a masked feature image; The masked feature image is reconstructed through the Transformer decoder to obtain a reconstructed feature image; Random noise and scale transformation are applied to the same partial discharge signal to generate positive sample pairs, and partial discharge signals of different categories are used as negative samples to construct a comparative learning sample set. Based on the comparative learning sample set, the improved transformation network model is trained to obtain a fault diagnosis model; The fused feature vector is input into the fault diagnosis model, and a dynamic causal graph network is constructed in the graph reasoning stage; Set the discharge pulse characteristics, time-frequency energy distribution and environmental parameters as graph nodes; Calculate the causal relationship strength between the graph nodes through the attention mechanism to obtain the node relationship weight; Based on the node relationship weights, the graph nodes are iteratively updated, a fault propagation path is established, and a fault diagnosis result is output.

3. The transformer partial discharge intelligent diagnosis method based on improved conversion network as claimed in claim 2 is characterized by: The feature image is processed by an improved transformation network model to obtain a fused feature vector, including: Inputting the feature image into the improved transform network model, wherein the improved transform network model has a time-frequency analysis branch and a time-domain analysis branch; the time-frequency analysis branch uses a spatial attention mechanism to extract three-dimensional time-frequency features; the time-domain analysis branch captures the time-domain features of the local discharge pulse through a deep separable dilated convolutional network, and uses BiLSTM temporal attention to construct an improved transform network model; Based on the three-dimensional time-frequency features and the time domain features, a heterogeneous feature alignment fusion mechanism is adopted to transform the three-dimensional time-frequency features into The spatial dimension is mapped to the time domain features The temporal dimension of A multi-head cross-attention mechanism is used to establish cross-modal feature association, and the dynamic weighted fusion of the three-dimensional time-frequency features and time-domain features is realized through attention weight calculation; The cross-attention features are residually connected with the dimensionally aligned time-frequency features, and the fused feature vector is obtained through layer normalization.

4. The transformer partial discharge intelligent diagnosis method based on improved conversion network as claimed in claim 3 is characterized by: The characteristic images include MTF characteristic images, GASF characteristic images and GADF characteristic images; the conversion method of the MTF characteristic image is: Discretize the time series into Q states, calculate the state transition probability, and generate the state transition matrix P, where the elements of the state transition matrix are represents the probability of transitioning from state i to state j; Arrange all state transition probabilities according to state numbers and construct the Markov transition matrix.

5. The transformer partial discharge intelligent diagnosis method based on improved conversion network as claimed in claim 4 is characterized by: The conversion method of the GASF characteristic image is: For the timing sequence signal Perform angle mapping to obtain the first angle mapping signal ; Mapping the signal based on the first angle , construct the Gram sum matrix, where the calculation formula of the Gram sum matrix elements is as follows: ； ·in, is the element in the ath row and bth column of the Gram angle and field matrix, For time point The corresponding angle, For time point The corresponding angle, is the timestamp of the i-th data point in the original time series, is the timestamp of the jth data point in the original time series.

6. The transformer partial discharge intelligent diagnosis method based on improved conversion network as claimed in claim 4 is characterized by: The conversion method of the GADF feature image is: The signal of the timing sequence Converting into a second angle mapping signal through an arc cosine function; Based on the second angle mapping signal, a Gram difference matrix is ​​constructed, wherein the calculation formula of the Gram difference matrix elements is as follows: ； in, is the element in the ath row and bth column of the Gram angle difference field matrix, For time point The corresponding angle, For time point The corresponding angle, is the timestamp of the i-th data point in the original time series, is the timestamp of the jth data point in the original time series.

7. The transformer partial discharge intelligent diagnosis method based on improved conversion network as claimed in claim 6 is characterized by: The method for obtaining the time series is: Ultrasonic detection, ultra-high frequency (UHF) detection and electromagnetic wave detection methods are used to synchronously collect partial discharge signals of transformers, and Gaussian filtering and adaptive denoising are performed on the partial discharge signals in turn to obtain denoised time domain signals; Dynamic threshold segmentation is performed on the time domain signal to extract the discharge pulse fragments, and a time series is generated through wavelet transformation.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the intelligent diagnosis method for partial discharge of a transformer based on an improved conversion network as described in any one of claims 1 to 7 are implemented.

9. 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 intelligent diagnosis method for partial discharge of a transformer based on an improved conversion network as described in any one of claims 1 to 7 are implemented.

Images