Intelligent power equipment fault diagnosis system and method based on image recognition and partial discharge feature extraction

Abstract

The invention discloses an intelligent power equipment fault diagnosis system and method based on image recognition and partial discharge feature extraction. The system comprises the following modules: a data acquisition module, a signal preprocessing and time-frequency image generation module, a hybrid deep learning network module and a fault classification and diagnosis module. By combining the CNN and the Transform network, local and global features of a partial discharge signal can be captured at the same time, the detection precision and reliability are improved, a self-attention mechanism of the Transform network enables the system to still accurately identify an effective signal under a high-noise background, the false alarm rate is reduced, the system can monitor the operation state of the power equipment in real time, and the system is suitable for popularization and application. The real-time fault early warning and diagnosis report is provided, the method is suitable for being applied to on-line monitoring and remote diagnosis scenes, and accurate recognition and real-time fault early warning of complex partial discharge signals are achieved. Compared with a traditional method, the detection accuracy, the anti-interference capability and the real-time performance are remarkably improved, and the method has a wide application prospect.

Description

Technical Field

[0001] The present invention relates to the technical field of power fault diagnosis, and in particular to an intelligent power equipment fault diagnosis system and method based on image recognition and partial discharge feature extraction. Background Art

[0002] Partial discharge (PD) is one of the main signs of insulation failure in high-voltage power equipment, which can cause serious equipment failure and power outage. Therefore, accurate and timely detection and diagnosis of PD is crucial to the stable operation of the power system.

[0003] Traditional partial discharge detection methods mainly rely on frequency domain analysis and time domain signal processing. The limitation of frequency domain analysis is that it is difficult to process non-stationary signals. Frequency domain analysis is mainly based on the assumption of stationary signals, while partial discharge signals often have non-stationary characteristics. In complex power environments, noise may cover or interfere with the spectral characteristics of partial discharge signals. The limitation of time domain signal processing is that it requires processing a large amount of data, which may lead to insufficient real-time performance; high noise background may reduce the accuracy of time domain signal processing and affect the accurate detection of partial discharge. Therefore, frequency domain analysis and time domain signal processing often cannot guarantee sufficient accuracy and real-time performance when dealing with complex power environments and high noise backgrounds. Summary of the invention

[0004] The purpose of the present invention is to provide an intelligent power equipment fault diagnosis system and method based on image recognition and partial discharge feature extraction to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: an intelligent power equipment fault diagnosis system based on image recognition and partial discharge feature extraction, comprising the following modules:

[0006] A data acquisition module, the data acquisition module is electrically connected to the power equipment to collect partial discharge signals from the power equipment in real time;

[0007] A signal preprocessing and time-frequency image generation module, which processes the collected partial discharge signal and converts it into a time-frequency image;

[0008] A hybrid deep learning network module, wherein the hybrid deep learning network module includes a CNN module and a Transformer module, wherein the hybrid deep learning network module extracts local features and scale features of the time-frequency image and captures global feature dependencies in the time-frequency image;

[0009] The fault classification and diagnosis module classifies the local features, scale features, and global feature dependency fully connected layers, generates a probability distribution of local discharge signals, and determines the discharge type.

[0010] Furthermore, the signal preprocessing and time-frequency image generation module converts the discharge signal into a time-frequency image based on short-time Fourier transform, wavelet transform, and Hilbert-Huang transform.

[0011] Furthermore, the CNN module receives the time-frequency image, and the convolution layer and pooling layer of the CNN module are used to extract multi-scale features of the time-frequency image step by step, and through batch normalization and ReLU activation function.

[0012] Furthermore, the Transformer module receives the multi-scale features extracted by the CNN module, and the Transformer module captures the global feature dependencies in the time-frequency image based on the self-attention mechanism.

[0013] A method for intelligent power equipment fault diagnosis is applied to the above-mentioned intelligent power equipment fault diagnosis system based on image recognition and partial discharge feature extraction, and the method comprises the following steps:

[0014] Step 1: Data preprocessing and time-frequency image generation: Install sensors on the power equipment to collect partial discharge signals, perform denoising and time domain signal normalization on the partial discharge signals, and generate corresponding time-frequency images;

[0015] Step 2: Hybrid deep learning network design, extract local features from time-frequency images based on CNN, use the outputs of different convolutional layers of CNN to fuse features, form a multi-scale feature map, add position encoding to the multi-scale feature map, and then capture the global features of the multi-scale feature map based on the Transformer network. The processed global features are fused with the local features to form a feature vector;

[0016] Step 3: Fault classification and diagnosis: Input the fused feature vector into the fully connected layer for classification;

[0017] Step 4: Online monitoring of substation equipment: install sensors on electrical equipment to collect partial discharge signals in real time and perform diagnosis.

[0018] Furthermore, in step 1, the sensor includes a TEV sensor and an ultrasonic sensor, and the partial discharge signal includes an electromagnetic wave signal, an ultrasonic signal, and a high-frequency electromagnetic wave signal;

[0019] In step 1, the denoising process is based on bandpass filter, low-pass filter and wavelet denoising to remove high-frequency noise and low-frequency interference in the partial discharge signal;

[0020] The corresponding time-frequency images are generated based on short-time Fourier transform, wavelet transform, and Hilbert-Huang transform.

[0021] Furthermore, in step 2, CNN extracts local features of the time-frequency image based on convolution operation, pooling operation, batch normalization, and ReLU activation function;

[0022] The formation of multi-scale feature maps is based on feature concatenation, cross-layer connection or residual connection;

[0023] The global feature capture of the Transformer network is based on feature block segmentation and multi-head self-attention mechanism.

[0024] Furthermore, in step 3, the design of the classifier for classification includes the following steps:

[0025] Fully connected layer: The multi-layer fully connected neural network maps the fused feature vector to different classification outputs, and uses the Softmax function to calculate the final partial discharge type probability;

[0026] Fault type classification: The output result is the probability distribution of each type of fault;

[0027] The fault type, possible cause, and recommended measures are generated based on the output results. At the same time, an early warning is triggered when the partial discharge signal exceeds the preset threshold.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] (1) Multi-scale feature fusion: By combining CNN and Transformer networks, it is possible to simultaneously capture the local and global features of local discharge signals, thereby improving the accuracy and reliability of detection.

[0030] (2) Strong anti-interference ability: The self-attention mechanism of the Transformer network enables the system to accurately identify valid signals even in a high-noise background, reducing the false alarm rate.

[0031] (3) Real-time diagnostic capability: The system can monitor the operating status of power equipment in real time and provide immediate fault warnings and diagnostic reports, making it suitable for online monitoring and remote diagnosis scenarios.

[0032] (4) Through multi-scale feature extraction and global feature capture, accurate recognition of complex partial discharge signals and real-time fault warning are achieved. Compared with traditional methods, the detection accuracy, anti-interference ability and real-time performance are significantly improved, and it has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 The figure is a flow chart of the system of the present invention. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0035] Example:

[0036] See also Figure 1 The present invention provides a technical solution: an intelligent power equipment fault diagnosis system based on image recognition and partial discharge feature extraction, comprising the following main modules:

[0037] 1. Data acquisition module

[0038] This module is responsible for collecting partial discharge signals from power equipment in real time. The collected signals include TEV (transverse electromagnetic wave), ultrasonic signals, etc. These signals are used to reflect the insulation status inside the power equipment.

[0039] 2. Signal preprocessing and time-frequency image generation module

[0040] After preprocessing steps such as filtering and denoising, the collected partial discharge signals are converted into time-frequency images using time-frequency analysis methods such as short-time Fourier transform (STFT) and wavelet transform (WT). These time-frequency images contain the time domain and frequency domain characteristics of the discharge signals, which are suitable for processing and analysis by deep learning models.

[0041] 3. Hybrid Deep Learning Network Module

[0042] CNN module: First, the time-frequency image is input into the convolutional neural network (CNN) to extract local features of the image, such as texture, edge, and local pattern. Multiple convolutional layers and pooling layers are used to extract multi-scale features step by step, and batch normalization and ReLU activation function are used to enhance the nonlinear expression ability of the network.

[0043] Transformer module: Then, the feature image extracted by CNN is input into the Transformer network. Transformer captures the global feature dependencies in the image through the self-attention mechanism, thereby improving the recognition ability of complex partial discharge patterns. The multi-head self-attention mechanism allows the network to process different features in different subspaces in parallel, enhancing the overall perception ability of the model.

[0044] 4. Fault classification and diagnosis module

[0045] The features output by the hybrid network are classified through the fully connected layer to generate the probability distribution of partial discharge signals and determine the specific discharge type (such as internal discharge, surface discharge, corona discharge, etc.). Finally, the system generates a fault diagnosis report for the power equipment based on the classification results and puts forward corresponding maintenance suggestions.

[0046] An intelligent power equipment fault diagnosis method based on image recognition and partial discharge feature extraction, the specific implementation steps are as follows:

[0047] 1. Data preprocessing and time-frequency image generation

[0048] Step 1.1: Signal acquisition and multimodal data fusion

[0049] The system collects partial discharge signals through various sensors (such as TEV sensors, ultrasonic sensors, etc.) installed on equipment such as transformers, switch cabinets or high-voltage cables. These signals can include transverse electromagnetic wave signals, ultrasonic signals and high-frequency electromagnetic wave signals.

[0050] During the signal acquisition process, synchronous acquisition technology is used to ensure that data from different sensors can be aligned in time and space, thus providing a basis for subsequent data fusion.

[0051] Step 1.2: Signal preprocessing

[0052] Preprocess the collected multimodal signals, including:

[0053] De-noising: Use bandpass filters, low-pass filters and wavelet denoising technology to remove high-frequency noise and low-frequency interference in the signal and retain key features related to partial discharge.

[0054] Time domain signal normalization: Normalize the data from different signal sources in the time domain to eliminate the inconsistency caused by the difference in response time of different sensors.

[0055] Step 1.3: Time-frequency image generation

[0056] Perform time-frequency analysis on the preprocessed signal to generate corresponding time-frequency images. These time-frequency images visualize the time and frequency characteristics of the partial discharge signal, including the following common methods:

[0057] (STFT): Generates a time-frequency image with a balanced time and frequency resolution, which is suitable for analyzing relatively stable partial discharge signals.

[0058] Wavelet transform (WT): It is particularly suitable for processing non-stationary signals and can capture the frequency changes of partial discharge signals at different times.

[0059] Hilbert-Huang Transform (HHT): It uses empirical mode decomposition and Hilbert spectrum to generate high-resolution time-frequency images, which is suitable for processing complex partial discharge patterns.

[0060] 2. Hybrid Deep Learning Network Design

[0061] Step 2.1: CNN for local feature extraction

[0062] 2.1.1 Convolutional layer design

[0063] Design a multi-layer convolutional neural network (CNN) to extract local features from time-frequency images. The convolutional layer can be divided into multiple stages, each of which includes the following:

[0064] Convolution operation: Use convolution kernels of different sizes (e.g. 3x3, 5x5) to extract local features of different scales. The number of convolution kernels in each layer is set according to the complexity of the features, for example, 32 convolution kernels in the first layer and 64 convolution kernels in the second layer.

[0065] Pooling operation: A pooling layer (Max Pooling) is added after each convolutional layer to reduce the spatial dimension of the feature map, reduce computational complexity, and retain key information.

[0066] Batch Normalization: Add a batch normalization layer after each convolutional layer to balance the distribution of data, reduce the gradient disappearance or explosion phenomenon, and accelerate training convergence.

[0067] Activation function: Use ReLU (Rectified Linear Unit) activation function to improve the nonlinear expression ability of the network.

[0068] 2.1.2 Multi-scale feature extraction

[0069] The outputs of different convolutional layers are used for feature fusion to form a multi-scale feature representation (multi-scale feature map). The specific methods include:

[0070] Feature splicing: The feature maps of different layers are spliced ​​in the depth direction to form a multi-scale feature map.

[0071] Cross-layer connection: Through skip connection or residual connection, the fine-grained features of the shallow layer are combined with the high-abstract features of the deep layer to retain more detailed information.

[0072] Step 2.2: Transformer for global feature capture

[0073] 2.2.1 Self-Attention Mechanism

[0074] Based on the multi-scale feature map extracted by CNN, the Transformer network is used to capture global features. The Transformer network consists of a multi-layer self-attention mechanism and a feedforward neural network:

[0075] Feature block segmentation: The feature map output by CNN is divided into multiple small blocks (patches), and each block is mapped to a feature vector of fixed dimension through linear transformation.

[0076] Multi-Head Self-Attention: Through the multi-head self-attention mechanism (Multi-HeadAttention), the network can focus on different signal features in different feature subspaces in parallel and capture global dependencies.

[0077] 2.2.2 Positional Encoding

[0078] In order to preserve the spatial position information of each feature block in the time-frequency image, positional encoding is added before the feature block is input into the Transformer network. This enables the Transformer network to capture global features while retaining spatial structure information, further improving classification accuracy.

[0079] 2.2.3 Global Feature Fusion

[0080] The global features processed by the Transformer network are fused with the local features extracted by the CNN, and the final feature vector is formed by feature weighting, splicing or superposition. This vector contains the global dependency information and local feature details of the partial discharge signal, providing rich input information for the classifier.

[0081] 3. Fault classification and diagnosis

[0082] Step 3.1: Classifier Design

[0083] The fused feature vector is input into the fully connected layer for classification. The classifier design includes the following steps:

[0084] Fully connected layer: The multi-layer fully connected neural network maps the fused feature vector to different classification outputs, and uses the Softmax function to perform the final partial discharge type probability calculation.

[0085] Fault type classification: The output result is the probability distribution of various faults, such as internal discharge, surface discharge, corona discharge, etc. The corresponding fault type is output according to the highest probability.

[0086] Step 3.2: Generate diagnostic report

[0087] According to the classification results, the system automatically generates a fault diagnosis report for the power equipment. The report includes:

[0088] Fault type: Determine the fault type based on the highest probability classification result.

[0089] Possible cause: Combine historical data and feature library to speculate the possible cause of the fault.

[0090] Suggested measures: Provide targeted maintenance and repair suggestions to help operation and maintenance personnel deal with equipment problems in a timely manner.

[0091] Step 3.3: Real-time monitoring and early warning

[0092] The system supports real-time data monitoring. When the local discharge signal exceeds the preset threshold, it automatically triggers an early warning to notify the operation and maintenance personnel:

[0093] Real-time alarm: Real-time notification of relevant personnel via sound and light alarms, text messages, emails, etc.

[0094] Data recording and backtracking: The system records all monitoring data and supports data backtracking analysis after a fault occurs, helping to gain a deeper understanding of the process of the fault.

[0095] 4. System application scenarios

[0096] Step 4.1: Online monitoring of substation equipment

[0097] In substations, the system collects partial discharge signals in real time and performs intelligent diagnosis through sensors installed on equipment such as transformers and switch cabinets. The system can detect equipment insulation problems in a timely manner and prevent serious accidents caused by partial discharge.

[0098] Step 4.2: Remote fault detection of high voltage cables

[0099] The system can be applied to remote fault detection of high-voltage cables. By deploying monitoring equipment along the cable, the changes in partial discharge signals can be remotely monitored. Combined with the intelligent diagnosis system, operation and maintenance personnel can remotely grasp the health status of the cable in real time.

[0100] Application scenarios:

[0101] Real-time monitoring of substation equipment: After the system is deployed in the substation, it can prevent serious faults and improve the stability and safety of the power system by real-time monitoring and diagnosis of partial discharge signals of transformers, switchgear and other equipment.

[0102] Remote diagnosis and maintenance: The system can be combined with the big data platform of the smart grid to provide remote diagnosis and maintenance support, and realize unified management of distributed power equipment.

[0103] in conclusion:

[0104] This embodiment proposes a partial discharge intelligent detection and fault diagnosis system based on a hybrid deep learning architecture of CNN and Transformer. The system achieves accurate recognition of complex partial discharge signals and real-time fault warning through multi-scale feature extraction and global feature capture. Compared with traditional methods, this embodiment significantly improves the accuracy, anti-interference ability and real-time performance of detection, and has broad application prospects.

[0105] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent power equipment fault diagnosis system based on image recognition and partial discharge feature extraction, characterized in that: Includes the following modules: A data acquisition module, the data acquisition module is electrically connected to the power equipment to collect partial discharge signals from the power equipment in real time; A signal preprocessing and time-frequency image generation module, which processes the collected partial discharge signal and converts it into a time-frequency image; A hybrid deep learning network module, wherein the hybrid deep learning network module includes a CNN module and a Transformer module, wherein the hybrid deep learning network module extracts local features and scale features of the time-frequency image and captures global feature dependencies in the time-frequency image; The fault classification and diagnosis module classifies the local features, scale features, and global feature dependency fully connected layers, generates a probability distribution of local discharge signals, and determines the discharge type.

2. According to claim 1, an intelligent power equipment fault diagnosis system based on image recognition and partial discharge feature extraction is characterized in that: The signal preprocessing and time-frequency image generation module converts the discharge signal into a time-frequency image based on short-time Fourier transform, wavelet transform, and Hilbert-Huang transform.

3. The intelligent power equipment fault diagnosis system based on image recognition and partial discharge feature extraction according to claim 1 is characterized in that: The CNN module receives the time-frequency image, and the convolution layer and pooling layer of the CNN module are used to extract the multi-scale features of the time-frequency image step by step, and use batch normalization and ReLU activation function.

4. The intelligent power equipment fault diagnosis system based on image recognition and partial discharge feature extraction according to claim 3 is characterized in that: The Transformer module receives the multi-scale features extracted by the CNN module, and the Transformer module captures the global feature dependencies in the time-frequency image based on the self-attention mechanism.

5. A method for diagnosing faults of intelligent power equipment, characterized in that: The method is applied to an intelligent power equipment fault diagnosis system based on image recognition and partial discharge feature extraction as claimed in any one of claims 1 to 4, and the method comprises the following steps: Step 1: Data preprocessing and time-frequency image generation: Install sensors on the power equipment to collect partial discharge signals, perform denoising and time domain signal normalization on the partial discharge signals, and generate corresponding time-frequency images; Step 2: Hybrid deep learning network design, extract local features from time-frequency images based on CNN, use the outputs of different convolutional layers of CNN to fuse features, form a multi-scale feature map, add position encoding to the multi-scale feature map, and then capture the global features of the multi-scale feature map based on the Transformer network. The processed global features are fused with the local features to form a feature vector; Step 3: Fault classification and diagnosis: Input the fused feature vector into the fully connected layer for classification; Step 4: Online monitoring of substation equipment: install sensors on electrical equipment to collect partial discharge signals in real time and perform diagnosis.

6. A method for diagnosing faults of intelligent power equipment according to claim 5, characterized in that: In step one, the sensor includes a TEV sensor and an ultrasonic sensor, and the partial discharge signal includes an electromagnetic wave signal, an ultrasonic signal, and a high-frequency electromagnetic wave signal.

7. A method for diagnosing faults of intelligent power equipment according to claim 5, characterized in that: In step 1, the denoising process is based on bandpass filter, low-pass filter and wavelet denoising to remove high-frequency noise and low-frequency interference in the partial discharge signal; The corresponding time-frequency images are generated based on short-time Fourier transform, wavelet transform, and Hilbert-Huang transform.

8. A method for diagnosing faults of intelligent power equipment according to claim 5, characterized in that: In step 2, CNN extracts local features of the time-frequency image based on convolution operation, pooling operation, batch normalization, and ReLU activation function; The formation of multi-scale feature maps is based on feature concatenation, cross-layer connection or residual connection; The global feature capture of the Transformer network is based on feature block segmentation and multi-head self-attention mechanism.

9. A method for diagnosing faults of intelligent power equipment according to claim 5, characterized in that: In step 3, the design of the classifier for classification includes the following steps: Fully connected layer: The multi-layer fully connected neural network maps the fused feature vector to different classification outputs, and uses the Softmax function to calculate the final partial discharge type probability; Fault type classification: The output result is the probability distribution of each type of fault; The fault type, possible cause, and recommended measures are generated based on the output results. At the same time, an early warning is triggered when the partial discharge signal exceeds the preset threshold.

Images