Transformer partial discharge identification method and system based on mask self-encoding technology

Abstract

The invention relates to the technical field of transformer partial discharge detection, in particular to a transformer partial discharge identification method and system based on a mask self-encoding technology, and the method comprises the steps: collecting a discharge signal during the partial discharge of a transformer through an ultrasonic detection method; the collected data information is drawn into a PRPD atlas, and the PRPD atlas is preprocessed; adding masks to the PRPD map features and training the PRPD map features to obtain an encoder which can be used for feature dimension reduction; and training the PRPD atlas through a convolutional neural network to obtain an optimized model. The method has the beneficial effects that the encoder capable of being used for feature dimension reduction is constructed by performing mask auto-encoder training on the PRPD atlas, the defect of the traditional convolutional network in the aspect of global and long-distance context modeling energy is avoided, and the precision and generalization ability of the model are improved. And by adding the mask on the training set, the recognition accuracy of the model is remarkably improved, and the generalization ability of the model is also enhanced.

Description

Technical Field

[0001] The invention relates to the technical field of transformer partial discharge detection, in particular to a transformer partial discharge identification method and system based on masked self-encoding technology. Background Art

[0002] The power transformer is one of the core equipment of the power system, and its safe and stable operation is crucial to the reliability of the entire power system. Due to the influence of the manufacturing process and the operating environment, there may be insulation defects such as metal burrs and suspended particles inside the transformer. These defects will lead to uneven electric field distribution, which will accelerate the degradation of the insulation material and may cause partial discharge. The partial discharge phenomenon seriously threatens the insulation performance of the transformer, and different types of discharges cause different degrees of damage to the equipment.

[0003] Partial discharge detection technology is a key means to obtain the internal insulation status of the transformer. It plays a fundamental role in fault diagnosis, status assessment and even life prediction. When partial discharge occurs, it is usually accompanied by various phenomena such as electricity, light, sound, temperature and gas. The detection of these phenomena has spawned a variety of detection methods, such as pulse current method, high-frequency pulse current method, ultra-high frequency detection method, optical detection method and gas chromatography. Although these methods can effectively detect the presence of partial discharge, they each have some limitations and challenges. Therefore, in practical applications, it is necessary to select the most appropriate detection technology based on the specific situation of the equipment and the characteristics of the fault. Summary of the invention

[0004] In view of the above problems or problems existing in the prior art, the present invention is proposed.

[0005] Therefore, the purpose of the present invention is to provide a transformer partial discharge identification method based on masked self-encoding technology, which can realize an efficient data processing algorithm and improve the processing efficiency and pattern recognition accuracy of the existing PRPD method.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a transformer partial discharge identification method based on masked self-encoding technology, which includes collecting the discharge signal when the transformer has partial discharge by ultrasonic detection method;

[0007] Plot the collected data information into a PRPD map, and pre-process the PRPD map;

[0008] Add a mask to the PRPD graph features and train them to obtain an encoder that can be used to reduce the dimension of the features;

[0009] The PRPD atlas is trained through a convolutional neural network to obtain an optimized model.

[0010] As a preferred solution of the transformer partial discharge identification method based on mask self-encoding technology of the present invention, signal collection includes collecting four types of discharge signals: suspended discharge, corona discharge, insulation discharge, and free particle discharge.

[0011] As a preferred solution of the transformer partial discharge identification method based on mask self-encoding technology of the present invention, the PRPD spectrum is a two-dimensional grid data with the partial discharge discharge phase and discharge amount as coordinate axes and the number of discharges as numerical values.

[0012] As a preferred solution of the transformer partial discharge identification method based on mask self-encoding technology of the present invention, the PRPD spectrum preprocessing includes:

[0013] Compress the resolution of the map;

[0014] Convert to grayscale image and perform normalization;

[0015] 80% of the dataset is used as the training set and 20% as the test set.

[0016] As a preferred solution of the transformer partial discharge identification method based on the masked self-encoding technology of the present invention, wherein: adding a mask to the PRPD spectrum feature includes adding a mask and segmenting the feature, and training the mask includes training with a masked self-encoder to obtain an encoder that can be used to reduce the dimension of the feature;

[0017] Training the mask using a mask autoencoder involves linear projection, adding a positional encoding head, Transformer encoder encoding, and decoder reconstruction.

[0018] As a preferred solution of the transformer partial discharge identification method based on masked self-encoding technology of the present invention, after the optimized model is established, the optimized model is evaluated, and the evaluation process includes inputting the test set into the trained model for classification, and evaluating and analyzing the classification results through the confusion matrix to quantify the recognition effect of various discharge types and evaluate the overall performance of the model.

[0019] In order to solve the above technical problems, the present invention also provides the following technical solutions: a transformer partial discharge identification system based on masked self-encoding technology, which includes a signal acquisition module for collecting and preprocessing partial discharge signals;

[0020] A data processing module, which is used to store the original data and the processed PRPD atlas data, and to run the data processing algorithm;

[0021] Safety module, used to protect the system from overload damage and reduce the impact of static electricity on electronic equipment.

[0022] As a preferred solution of the transformer partial discharge identification system based on masked self-encoding technology of the present invention, the signal acquisition module includes an ultrasonic sensor, a data acquisition card, and a signal amplifier;

[0023] Ultrasonic sensor, used to collect ultrasonic signals generated by partial discharge of transformer;

[0024] Data acquisition card, used to convert the analog signal of the sensor into a digital signal;

[0025] Signal amplifier increases the strength of the sensor signal to improve the signal-to-noise ratio.

[0026] As a preferred solution of the transformer partial discharge identification system based on masked self-encoding technology of the present invention, wherein: the data processing module includes an industrial computer and a data storage device;

[0027] Industrial computers, used to run data processing software and store data;

[0028] Data storage devices, such as hard disks or solid-state drives, are used to store raw data and processing results for a long time.

[0029] As a preferred solution of the transformer partial discharge identification system based on mask self-encoding technology of the present invention, wherein: the safety module includes an overload protector and an anti-static storage cabinet;

[0030] Overload protector, used to protect the system from overload damage;

[0031] Anti-static storage cabinets are used to reduce the effects of static electricity on sensitive electronic equipment.

[0032] Beneficial effects of the present invention: The present invention constructs an encoder that can be used for feature dimensionality reduction by training a masked autoencoder on the PRPD spectrum, thereby avoiding the deficiencies of traditional convolutional networks in global and long-distance context modeling energy and improving the accuracy and generalization ability of the model. By adding masks to the training set, the recognition accuracy of the model is significantly improved, and the generalization ability of the model is also enhanced. In addition, the present invention proposes to reduce dependence on manual experience by randomly masking the PRPD spectrum, extracting features, and globally reconstructing based on high-dimensional mapping, thereby improving the automation level and accuracy of transformer partial discharge identification, and has significant practical value. This method is suitable for online monitoring and maintenance of power systems, and is of great significance for improving the stable operation of transformer equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] 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:

[0034] Figure 1 The figure is a schematic diagram of the overall process of the transformer partial discharge identification method based on masked self-encoding technology.

[0035] Figure 2 Schematic diagram of the PRPD spectrum processing model architecture of the transformer partial discharge identification method based on masked self-encoding technology.

[0036] Figure 3 Schematic diagram of the network structure of the convolutional neural network for transformer partial discharge identification method based on masked self-encoding technology.

[0037] Figure 4 This is a schematic diagram showing how the recognition accuracy of the transformer partial discharge recognition method based on mask self-encoding technology varies with the mask ratio.

[0038] Figure 5 Schematic diagram of the confusion matrix of the autoencoding technology for transformer partial discharge identification method based on masked autoencoding technology. DETAILED DESCRIPTION

[0039] 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.

[0040] 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.

[0041] 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.

[0042] Example 1

[0043] Refer to 1~ Figure 5, which is the first embodiment of the present invention, provides a transformer partial discharge identification method based on masked self-encoding technology, which can implement an efficient data processing algorithm and improve the processing efficiency and pattern recognition accuracy of the existing PRPD method.

[0044] Specifically, the discharge signal when partial discharge occurs in the transformer is collected by ultrasonic detection method;

[0045] Plot the collected data information into a PRPD map, and pre-process the PRPD map;

[0046] Add a mask to the PRPD graph features and train them to obtain an encoder that can be used to reduce the dimension of the features;

[0047] The PRPD atlas is trained through a convolutional neural network to obtain an optimized model.

[0048] It should be noted that the network structure of the convolutional neural network used in this scheme is as follows Figure 3 As shown in the figure, there are 18 layers in total, including 5 convolutional layers, 5 pooling layers, 5 non-linear activation layers, and 2 fully connected layers. A Dropout layer is added in between, where the pooling layer selects the maximum pooling function and the activation function selects the ReLU function.

[0049] Among them, the convolution layer is the core of CNN, responsible for extracting the features of the input data. In the convolution layer, the convolution kernel (or filter) slides on the input data, calculates the dot product between the convolution kernel and the local area of ​​the input data, and generates feature maps;

[0050] The pooling layer is used to reduce the spatial dimensions (height and width) of the feature map, extract important features, and reduce the amount of computation and the risk of overfitting;

[0051] The nonlinear activation layer introduces nonlinear factors, enabling CNN to learn and simulate complex function mappings;

[0052] The fully connected layer is the output part of CNN, which is responsible for converting the feature map into the final output result, such as the classification label;

[0053] The Dropout layer is a regularization technique used to prevent overfitting of neural networks. During the training process, the Dropout layer randomly sets the output of some neurons to 0 to reduce the complex co-adaptation relationship between neurons.

[0054] The full name of the ReLU function is "Rectified Linear Unit", which is an activation function widely used in deep learning. Its main features are simple calculation, fast convergence speed, and can effectively alleviate the overfitting problem. The specific calculation formula of the ReLU function is as follows:

[0055] ReLU(x)=max(0,x)

[0056] Its meaning is that when the input x is greater than 0, the output is x itself; when the input x is less than or equal to 0, the output is 0.

[0057] In this scheme, the main hyperparameter configuration settings of the model are as follows: batch_size is set to 64, the learning rate is set to 0.001, the maximum number of iterations is 50, the classification loss function uses the Cross-Entropy loss function, the optimization algorithm is the stochastic gradient descent with momentum (SGDM) algorithm, the stochastic gradient descent momentum is 0.9, and the back-propagation algorithm is used for supervised fine-tuning to optimize the model parameters.

[0058] Among them, the Cross-Entropy Loss function is often used in multi-classification problems to measure the difference between the model prediction value and the true value. For a classification problem with C categories, the formula of the cross entropy loss function is:

[0059]

[0060] Among them, y is the one-hot encoding of the true label, y i is the true label of the i-th category (0 or 1); p is the probability distribution predicted by the model, p i is the probability that the model predicts the i-th category.

[0061] The SGDM algorithm is an improved gradient descent algorithm that accelerates convergence and reduces oscillation by introducing momentum terms. The update formula of SGDM is:

[0062]

[0063] θ=θ-αv t

[0064] Where: θ is the model parameter; J(θ) is the loss function; is the gradient of the loss function with respect to the parameter θ; α is the learning rate; β is the momentum coefficient, which usually takes a value between 0.8 and 0.999; v t is the momentum at time step t; v t-1 is the momentum at time step t-1.

[0065] The back-propagation algorithm is a method for training neural networks by calculating the gradient of the loss function with respect to each parameter and using these gradients to update the parameters. The basic steps of the back-propagation algorithm are as follows:

[0066] Forward propagation: calculate the network output and loss function;

[0067] Calculate gradients: Use the chain rule to calculate the gradient of the loss function with respect to each parameter;

[0068] Back propagation: Starting from the output layer, calculate the gradient backward layer by layer;

[0069] Parameter update: Update parameters using gradients and optimization algorithms such as SGDM;

[0070] The gradient calculation of back propagation can be expressed as:

[0071]

[0072] δ(l)=((w(l+1)) T b(l+1))⊙σ′(z(l))

[0073]

[0074] Where C is the loss function; δ(l) is the error term of the lth layer; w(l) and b(l) are the weight and bias of the lth layer respectively; a(l) is the activation output of the lth layer; z(l) is the weighted input of the lth layer; σ is the activation function and σ′ is its derivative; ⊙ represents element-by-element multiplication.

[0075] Furthermore, signal collection includes collecting four types of discharge signals: suspended discharge, corona discharge, insulating discharge, and free particle discharge.

[0076] Furthermore, the PRPD spectrum is a two-dimensional grid data with the local discharge discharge phase and discharge amount as coordinate axes and the number of discharges as numerical values.

[0077] It should be noted that the PRPD spectrum (Polarization Phase-Pulse Strength Distribution) is a graphical tool for analyzing and identifying partial discharge (PD) phenomena. PD refers to the local electrical breakdown phenomenon that occurs in the insulating material of electrical equipment. This discharge will not immediately lead to complete breakdown of the insulating material, but long-term existence will accelerate the aging of the insulating material and may eventually lead to equipment failure.

[0078] The PRPD spectrum helps engineers and technicians identify and analyze the characteristics of partial discharge in a visual way, thereby evaluating the insulation status of the equipment. The PRPD spectrum includes:

[0079] Phase axis , which represents the phase relative to the supply voltage cycle when the partial discharge occurs. This axis is usually in degrees (°) and represents the position of the discharge event in the voltage cycle;

[0080] The amplitude axis (q) represents the amplitude of the local discharge, that is, the strength of the discharge pulse. This axis is usually expressed in units of charge (such as pico-coulomb, pC), indicating the size of the discharge;

[0081] The number of discharges (n) indicates the number of discharge events that occur in a specific area on the phase-amplitude plane within a certain period of time (such as one power frequency cycle).

[0082] PRPD spectra are usually used for discharge type identification, that is, different types of partial discharges (such as suspended discharge, corona discharge, insulation discharge, etc.) show different distribution characteristics on the PRPD spectra, which can be identified through the spectra;

[0083] Fault diagnosis: By analyzing the PRPD spectrum, the insulation status of the equipment can be determined and potential faults can be discovered in time;

[0084] Condition monitoring: In the operation and maintenance of power systems, PRPD spectra can be used to monitor the insulation status of equipment online to prevent faults.

[0085] Life prediction: By tracking the changes in the PRPD spectrum over a long period of time, the service life of the equipment can be predicted.

[0086] Furthermore, PRPD spectrum preprocessing includes:

[0087] Compress the resolution of the map;

[0088] Convert to grayscale image and perform normalization;

[0089] 80% of the dataset is used as the training set and 20% as the test set.

[0090] It should be noted that this scheme uses ultrasonic detection method to collect information on partial discharge conditions inside the transformer, and a total of 4832 pieces of raw data are collected. The collected raw data information is converted into PRPD atlas, with a total of 4748 quality data, 1237 suspended discharge data, 1158 corona discharge data, 1184 insulation discharge data, and 1169 free particle discharge data. The resolution of the drawn PRPD atlas is further compressed to 200×200×3, and the compressed PRPD atlas is converted into a grayscale image and normalized, and then 80% of the preprocessed PRPD atlas data set is used as a training set and 20% as a test set.

[0091] Further, adding a mask to the PRPD graph feature includes adding a mask and segmenting the feature, and training the mask includes training with a masked autoencoder to obtain an encoder that can be used to reduce the dimension of the feature;

[0092] Training the mask using a mask autoencoder involves linear projection, adding a positional encoding head, Transformer encoder encoding, and decoder reconstruction.

[0093] It should be noted that in order to utilize the capabilities of the masked autoencoder, it is necessary to add a mask to the PRPD atlas data. A mask is a random or deterministic cover that hides part of the data, forcing the model to learn and reconstruct the entire data from the visible part. The mask can be randomly distributed or designed based on a certain strategy to simulate data loss and enhance the robustness and feature extraction capabilities of the model.

[0094] Segmentation refers to dividing the PRPD atlas data into multiple small blocks. This segmentation can be a fixed grid division or a dynamic division based on some optimization strategy. Each small block contains a part of the atlas information for subsequent encoder processing.

[0095] The specific steps of the encoder processing include:

[0096] The segmented small blocks of data are fed into the encoder, and features are extracted from them to generate a compressed representation of the entire data, namely the potential representation;

[0097] The encoder processes small chunks of input data through a multi-layer neural network, extracts local features using the power of deep learning, and gradually builds a global understanding of the entire PRPD map;

[0098] In order to preserve spatial information, position encoding (such as sine and cosine position encoding) is added to the encoder so that the model can understand the positional relationship of features in the graph.

[0099] During the training process, the model's loss function usually includes reconstruction loss, which is the difference between the model output and the original PRPD map. Model training minimizes this loss, and the model learns how to better recover complete information from the mask data.

[0100] Furthermore, after the optimized model is established, the optimized model is evaluated. The evaluation process includes inputting the test set into the trained model for classification, evaluating and analyzing the classification results through the confusion matrix to quantify the recognition effect of various discharge types and evaluate the overall performance of the model.

[0101] It should be noted that this scheme inputs the test set into the trained CNN model for classification, and trains with different mask ratios. When the mask ratio is 40%, the recognition accuracy can reach 99.865%. The recognition accuracy changes with the mask ratio as shown in the figure Figure 4 The classification result confusion matrix is ​​shown as Figure 5 As shown, Figure 5 (a) is the confusion matrix diagram without using masked self-encoding technology. Figure 5 (b) is the confusion matrix diagram of using mask self-encoding technology when the mask ratio is 40%.

[0102] The results of partial discharge type recognition show that the highest accuracy is 92.25% without using masked self-encoding technology. After adding the masked self-encoding technology designed by the present invention, the recognition accuracy is significantly improved to 98.865%, indicating that while this operation affects the accuracy, it effectively improves the generalization ability and recognition accuracy of the model, and significantly improves the reliability of transformer partial discharge recognition based on PRPD spectrum.

[0103] In summary, the present invention constructs an encoder that can be used for feature dimensionality reduction by training the PRPD spectrum with a masked autoencoder, avoiding the deficiencies of traditional convolutional networks in global and long-distance context modeling energy, and improving the accuracy and generalization ability of the model. By adding masks to the training set, the recognition accuracy of the model is significantly improved, and the generalization ability of the model is also enhanced. In addition, the present invention proposes to reduce the dependence on manual experience by randomly masking the PRPD spectrum, extracting features, and globally reconstructing based on high-dimensional mapping, thereby improving the automation level and accuracy of transformer partial discharge identification, and has significant practical value. This method is suitable for online monitoring and maintenance of power systems, and is of great significance for improving the stable operation of transformer equipment.

[0104] Example 2

[0105] Refer to 1~ Figure 5, which is the second embodiment of the present invention, provides a transformer partial discharge identification method based on masked self-encoding technology, which can implement an efficient data processing algorithm and improve the processing efficiency and pattern recognition accuracy of the existing PRPD method.

[0106] Specifically, this solution applies masked self-encoding technology in the training set to improve the generalization ability and recognition accuracy of the model. The specific operation process is as follows:

[0107] The preprocessed PRPD spectra were linearly projected to obtain the fragment representation;

[0108] Add position encoding headers before these representations so that these fragments contain the position information of the slices;

[0109] The representation that completes the encoding and adds position information is put into the Transformer encoder for encoding.

[0110] The latent representation and the mask parts are reassembled in the order of cutting and sent to the decoder for reconstruction.

[0111] Furthermore, the traditional original input image R is mapped using the relative total variation in As a self-supervisory term in the pre-training process. The relative total variation model minimizes the input image R in With the expected image R rtv The sum of the energy functional of the Euclidean metric and the pixel total variation regularization makes the output mapping gradually approach the structure of the input image. Relative Total Variation (RTV) is a technique used for image reconstruction and denoising. It optimizes the image by combining the Euclidean distance and total variation regularization to reduce noise while maintaining structural features. The original input image R in is the image to be processed, the expected image R rtv It is the image obtained by relative total variation mapping, which aims to approximate the original input image R in The structure and features of R rtv The operation process is as follows:

[0112]

[0113] Where i represents each pixel of the input and output PRPD maps of the same scale; is the total variation intensity parameter; Represents the total variation regularization, which can be decomposed into the sum of the anisotropy along the long axis coordinate w and the wide axis coordinate h of the image. The formula is as follows:

[0114]

[0115] Furthermore, the input transformer PRPD spectrum data is masked and split into blocks and sent to the encoder to generate the potential representation of the entire data.

[0116] Furthermore, the data is first processed by the multi-head attention layer in the Transformer to complete the identification of important information, and the identified data representation is used to train the multi-layer perception set layer. After multi-layer encoding, the spectral block features of the PRPD spectrum containing position encoding information are encoded to obtain the potential representation of the data.

[0117] It should be noted that in the Transformer model, the Multi-Head Attention layer is one of its core components, which enables the model to focus on different parts of the input sequence at the same time and capture features in different subspaces. The specific steps of the Multi-Head Attention layer to process data are as follows:

[0118] Input representation, i.e., the feature block of the input PRPD graph, which will be represented as a matrix, where each row represents a feature block and each column represents a feature dimension;

[0119] Linear transformation, that is, for each feature block in the input matrix, the multi-head attention layer will perform three linear transformations to generate three matrices: query (Q), key (K), and value (V). Among them, query (Q): represents the information that the current feature block needs to find in other feature blocks; key (K): represents the information contained in other feature blocks; value (V): represents the content or feature corresponding to the key;

[0120] Scaled dot product attention, that is, each head calculates the dot product (ie, inner product) of the query and all keys to obtain an attention score matrix, and then divides it by the scaling factor (usually the square root of the key vector dimension) to prevent the gradient from disappearing or exploding. The specific formula is as follows:

[0121]

[0122] Among them, d k is the dimension of the key vector;

[0123] Apply the Softmax function, that is, apply the Softmax function to the attention score matrix, convert the score into a probability distribution, which represents the attention weight of each head to other feature blocks. Use the attention weight obtained by Softmax to perform weighted summation on the value (V) matrix, and the weighted value matrix formula is as follows:

[0124] Attention Weights=Softmax(Attention Scores)

[0125]

[0126] Weighted Values=Attentuin Weights·V

[0127] Where n is the dimension of the input vector x, that is, the number of elements; e is the base of the natural logarithm, which is approximately equal to 2.71828;

[0128] Multi-head merging,Since the multi-head attention layer contains multiple heads, the attention pattern learned by each head may be different. Each head performs the above operations independently, and then the outputs of all heads are merged through the linear layer to obtain the final multi-head attention output, the formula is as follows:

[0129] Multi-Head Output=Concat(Head1,Head2,…,Head h )W O

[0130] Among them, Concat represents the connection operation; W O is a learnable linear transformation matrix;

[0131] Finally, the output of the multi-head attention layer can be passed to subsequent layers, such as the Feed-Forward Network, or directly output as the final result of the model.

[0132] Furthermore, in the reconstruction process of the decoder, two advanced local attention mechanisms are used: window attention mechanism and drift window attention mechanism, to enhance the model's ability to process the details of the PRPD map. This mechanism divides the PRPD map into small blocks of the same size, each corresponding to a pixel. The drift window attention mechanism makes each divided small block consistent in shape through reasonable planning of the partition window.

[0133] In summary, the present invention constructs an encoder that can be used for feature dimensionality reduction by training the PRPD spectrum with a masked autoencoder, avoiding the deficiencies of traditional convolutional networks in global and long-distance context modeling energy, and improving the accuracy and generalization ability of the model. By adding masks to the training set, the recognition accuracy of the model is significantly improved, and the generalization ability of the model is also enhanced. In addition, the present invention proposes to reduce the dependence on manual experience by randomly masking the PRPD spectrum, extracting features, and globally reconstructing based on high-dimensional mapping, thereby improving the automation level and accuracy of transformer partial discharge identification, and has significant practical value. This method is suitable for online monitoring and maintenance of power systems, and is of great significance for improving the stable operation of transformer equipment.

[0134] Example 3

[0135] This embodiment is the third embodiment of the present invention, which provides a transformer partial discharge identification method based on masked self-encoding technology, which can realize an efficient data processing algorithm and improve the processing efficiency and pattern recognition accuracy of the existing PRPD method.

[0136] Specifically, a signal acquisition module is used to collect and pre-process partial discharge signals;

[0137] A data processing module, which is used to store the original data and the processed PRPD atlas data, and to run the data processing algorithm;

[0138] Safety module, used to protect the system from overload damage and reduce the impact of static electricity on electronic equipment.

[0139] Further, the signal acquisition module includes an ultrasonic sensor, a data acquisition card, and a signal amplifier;

[0140] Ultrasonic sensor, used to collect ultrasonic signals generated by partial discharge of transformer;

[0141] Data acquisition card, used to convert the analog signal of the sensor into a digital signal;

[0142] Signal amplifier increases the strength of the sensor signal to improve the signal-to-noise ratio.

[0143] Further, the data processing module includes an industrial computer, and a data storage device;

[0144] Industrial computers, used to run data processing software and store data;

[0145] Data storage devices, such as hard disks or solid-state drives, are used to store raw data and processing results for a long time.

[0146] Furthermore, the safety module includes an overload protector and an anti-static storage cabinet;

[0147] Overload protector, used to protect the system from overload damage;

[0148] Anti-static storage cabinets are used to reduce the effects of static electricity on sensitive electronic equipment.

[0149] In summary, the present invention constructs an encoder that can be used for feature dimensionality reduction by training the PRPD spectrum with a masked autoencoder, avoiding the deficiencies of traditional convolutional networks in global and long-distance context modeling energy, and improving the accuracy and generalization ability of the model. By adding masks to the training set, the recognition accuracy of the model is significantly improved, and the generalization ability of the model is also enhanced. In addition, the present invention proposes to reduce the dependence on manual experience by randomly masking the PRPD spectrum, extracting features, and globally reconstructing based on high-dimensional mapping, thereby improving the automation level and accuracy of transformer partial discharge identification, and has significant practical value. This method is suitable for online monitoring and maintenance of power systems, and is of great significance for improving the stable operation of transformer equipment.

[0150] Importantly, it should be noted that the construction and arrangement of the present application shown in a plurality of different exemplary embodiments are only exemplary. Although only a few embodiments are described in detail in this disclosure, it should be readily understood by those who refer to this disclosure that many modifications are possible without substantially departing from the novel teachings and advantages of the subject matter described in the application (e.g., mounting arrangements, use of materials, color, changes in orientation, etc.). For example, the element shown as integrally formed may be composed of a plurality of parts or elements, the position of the element may be inverted or otherwise changed, and the nature or number or position of the discrete element may be altered or changed. Therefore, all such modifications are intended to be included within the scope of the present invention. The order or sequence of any process or method steps may be changed or reordered according to an alternative embodiment. In the claims, any "bracket plus function" clause is intended to cover the structure of the execution function described herein, and is not only structurally equivalent but also equivalent structure. Without departing from the scope of the present invention, other replacements, modifications, changes and omissions may be made in the design, operating conditions and arrangement of the exemplary embodiments. Therefore, the present invention is not limited to a specific embodiment, but extends to a variety of modifications that still fall within the scope of the appended claims.

[0151] Additionally, in an effort to provide a concise description of example embodiments, all features of an actual implementation may not be described.

[0152] It should be understood that in the development of any actual implementation, as in any engineering or design project, numerous implementation-specific decisions may be made. Such a development effort may be complex and time-consuming, but for those of ordinary skill having the benefit of this disclosure, the development effort will be a routine task of design, fabrication, and production without undue experimentation.

[0153] 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 identification method based on masked self-encoding technology, characterized in that: include, The discharge signal of the transformer when partial discharge occurs is collected by ultrasonic detection method; Plot the collected data information into a PRPD map, and pre-process the PRPD map; Add a mask to the PRPD graph features and train them to obtain an encoder that can be used to reduce the dimension of the features; The PRPD atlas is trained through a convolutional neural network to obtain an optimized model.

2. The transformer partial discharge identification method based on masked self-encoding technology according to claim 1 is characterized in that: The signal collection includes collecting four types of discharge signals: suspended discharge, corona discharge, insulation discharge, and free particle discharge.

3. The transformer partial discharge identification method based on masked self-encoding technology according to claim 2 is characterized in that: The PRPD spectrum is two-dimensional grid data with the local discharge discharge phase and discharge amount as coordinate axes and the number of discharges as values.

4. The transformer partial discharge identification method based on masked self-encoding technology according to claim 3 is characterized in that: The PRPD spectrum preprocessing includes: Compress the resolution of the map; Convert to grayscale image and perform normalization; 80% of the dataset is used as the training set and 20% as the test set.

5. The transformer partial discharge identification method based on masked self-encoding technology according to claim 4 is characterized in that: Adding a mask to the PRPD graph feature includes adding a mask and segmenting the feature, and training the mask includes training with a masked autoencoder to obtain an encoder that can be used to reduce the dimension of the feature; The use of mask autoencoders to train masks includes linear projection, adding a position encoding head, Transformer encoder encoding, and decoder reconstruction.

6. The transformer partial discharge identification method based on masked self-encoding technology according to claim 5 is characterized in that: After the optimized model is established, the optimized model is evaluated. The evaluation process includes inputting the test set into the trained model for classification, evaluating and analyzing the classification results through the confusion matrix to quantify the recognition effect of various discharge types and evaluate the overall performance of the model.

7. A transformer partial discharge identification system based on masked self-encoding technology, applicable to the above transformer partial discharge identification method based on masked self-encoding technology, characterized in that: include, Signal acquisition module, used for collecting and preprocessing partial discharge signals; A data processing module, which is used to store the original data and the processed PRPD atlas data, and to run the data processing algorithm; Safety module, used to protect the system from overload damage and reduce the impact of static electricity on electronic equipment.

8. The transformer partial discharge identification system based on masked self-encoding technology according to claim 7, characterized in that: The signal acquisition module includes an ultrasonic sensor, a data acquisition card, and a signal amplifier; The ultrasonic sensor is used to collect ultrasonic signals generated when the transformer is partially discharged; The data acquisition card is used to convert the analog signal of the sensor into a digital signal; The signal amplifier increases the strength of the sensor signal to improve the signal-to-noise ratio.

9. The transformer partial discharge identification system based on masked self-encoding technology according to claim 8, characterized in that: The data processing module includes an industrial computer and a data storage device; The industrial computer is used to run data processing software and store data; The data storage device, such as a hard disk or a solid state drive, is used to store raw data and processing results for a long time.

10. The transformer partial discharge identification system based on masked self-encoding technology according to claim 9, characterized in that: The safety module includes an overload protector and an anti-static storage cabinet; The overload protector is used to protect the system from overload damage; The anti-static storage cabinet is used to reduce the impact of static electricity on sensitive electronic equipment.

Images