Partial discharge diagnosis method and device for C4F7N / CO2 / O2 mixed gas

Abstract

The invention relates to a partial discharge diagnosis method and device for C4F7N / CO2 / O2 mixed gas. The method comprises the following steps: firstly, based on a pre-constructed partial discharge experiment platform, obtaining a partial discharge signal of a typical defect in C4F7N / CO2 / O2 mixed gas to obtain a partial discharge identification data set; based on the partial discharge identification data set, using the trained C4F7N / CO2 / O2 mixed gas GIS partial discharge diagnosis model to diagnose a to-be-diagnosed signal, and determining a corresponding partial discharge type; wherein the C4F7N / CO2 / O2 mixed gas GIS partial discharge diagnosis model is jointly constructed based on a CNN (Convolutional Neural Network) and a Transformer. Compared with the prior art, the C4F7N / CO2 / O2 mixed gas GIS partial discharge diagnosis method has the advantages of effectively improving the accuracy and practicability of C4F7N / CO2 / O2 mixed gas GIS partial discharge diagnosis and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of GIS insulation defect detection, and in particular to a local discharge diagnosis method and device for C4F7N / CO2 / O2 mixed gas. Background Art

[0002] With the expansion of power grid scale and the continuous development of power grid technology, gas insulated switchgear (GIS) has been widely used in the power field due to its excellent reliability, miniaturized design and low maintenance requirements. Although sulfur hexafluoride (SF6) has long been used as the main gas insulation medium of GIS, as a strong greenhouse gas, it has serious negative impacts on global warming and ozone layer depletion. Perfluoroisobutyronitrile (C4F7N), as a potential substitute for SF6, has excellent insulation properties and relatively low global warming potential (GWP) value. In practical applications, C4F7N is usually mixed with gases such as CO2 and N2 to optimize its insulation properties and reduce the liquefaction temperature. Existing studies have shown that C4F7N / CO2 / O2 mixed gas has shown good performance in electrical insulation and arc extinguishing, and has begun to be used in actual equipment currently in operation.

[0003] Although great progress has been made in the study of the physical, chemical and electrical insulation properties of C4F7N / CO2 / O2 mixed gas, there is relatively little research on partial discharge (PD), the most common fault type in GIS applications, especially the related technology based on ultrasonic or ultra-high frequency electromagnetic wave (UHF) detection is still blank. In the field of GIS partial discharge fault diagnosis, partial discharge fault diagnosis is usually achieved by detecting ultrasonic or ultra-high frequency electromagnetic waves generated by partial discharge. Due to the different insulating gas media, the characteristics of ultrasonic or UHF electromagnetic wave signals will be different, which makes the existing detection methods for SF6 no longer applicable. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a method and device for partial discharge diagnosis of C4F7N / CO2 / O2 mixed gas, which can be compatible with ultra-high frequency electromagnetic wave signal and ultrasonic signal input at the same time, and can effectively improve the accuracy and practicality of GIS partial discharge diagnosis for C4F7N / CO2 / O2 mixed gas.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] According to a first aspect of the present invention, a method for partial discharge diagnosis of C4F7N / CO2 / O2 mixed gas is provided, comprising the following steps: based on a pre-constructed partial discharge experimental platform, obtaining partial discharge signals of typical defects in the C4F7N / CO2 / O2 mixed gas to obtain a partial discharge recognition data set; based on the partial discharge recognition data set, using a trained C4F7N / CO2 / O2 mixed gas GIS partial discharge diagnosis model to diagnose the signal to be diagnosed and determine the corresponding partial discharge type; wherein the C4F7N / CO2 / O2 mixed gas GIS partial discharge diagnosis model is jointly constructed based on CNN and Transformer.

[0007] As a preferred technical solution, the partial discharge experimental platform is constructed based on a real GIS.

[0008] As a preferred technical solution, the typical defects include needle tip defects, suspension defects, particle defects and surface discharge defects.

[0009] As a preferred technical solution, the needle tip defect is simulated by using a metal tip, the suspended defect is simulated by fixing a metal bolt between the upper and lower plates with epoxy resin, the particle defect is simulated by using steel balls, and the surface discharge defect is simulated by making the metal bolts leading out of the upper plate contact the upper and lower bottom surfaces of the epoxy resin cylinder with the lower plate respectively.

[0010] As a preferred technical solution, the C4F7N / CO2 / O2 mixed gas GIS partial discharge diagnosis model includes an encoder and a classifier, the encoder includes a first convolutional layer, a ResCNN Block module, a second convolutional layer and a Transformer Block module connected in sequence, the classifier includes an average pooling layer and a linear layer connected in sequence, and the output of the Transformer Block module is connected to the input of the average pooling layer.

[0011] As a preferred technical solution, each of the ResCNN Block modules includes multiple layers of one-dimensional convolutional layers, wherein the first one-dimensional convolutional layer is zero-filled to extract low-level features of the mixed gas partial discharge signal, and the remaining convolutional layers are used to gradually extract deep features. Each convolutional layer is nonlinearly processed through a ReLU activation function, and finally the corresponding input features and output features are added through a residual connection.

[0012] As a preferred technical solution, the Transformer Block module includes a multi-head self-attention layer, a feedforward neural network layer and a normalization layer. The multi-head self-attention layer is used to capture the global characteristics of the mixed gas partial discharge signal, and the feedforward neural network layer is used for further nonlinear mapping. Each layer is added to the corresponding input feature through a residual connection, and the normalization layer is used to perform normalization processing after the residual connection.

[0013] As a preferred technical solution, when training the C4F7N / CO2 / O2 mixed gas GIS partial discharge diagnosis model, the AdamW optimizer is used to optimize the model parameters, and the cross entropy loss is used to calculate the classification loss.

[0014] As a preferred technical solution, the expression of the cross entropy loss is:

[0015]

[0016] In the formula, represents the classification loss, N is the number of samples, M is the number of categories, y ic Indicates whether sample i belongs to category c, 1 if it does, 0 if it does not, y′ ic is the probability that the model predicts that sample i belongs to category c.

[0017] According to a second aspect of the present invention, there is provided a partial discharge diagnostic device for a C4F7N / CO2 / O2 mixed gas, comprising a memory, a processor, and a program stored in the memory, wherein the method described is implemented when the processor executes the program.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] 1. The present invention jointly constructs an environmentally friendly C4F7N / CO2 / O2 mixed gas GIS partial discharge diagnosis model based on CNN and Transformer, and uses the collected complete partial discharge identification data set for training and testing to determine the corresponding partial discharge type. The comparative experiment with other existing methods shows that the model can accurately diagnose the type of partial discharge with a small number of parameters, and its accuracy is significantly better than other existing methods. It has high precision and real-time performance in partial discharge diagnosis applications, and can be compatible with ultra-high frequency electromagnetic wave signal and ultrasonic signal input at the same time;

[0020] 2. The C4F7N / CO2 / O2 mixed gas GIS partial discharge diagnosis model in the present invention is designed specifically for the characteristics that partial discharge diagnosis requires a close combination of local features and global features. The CNN part uses a larger convolution kernel to improve the receptive field, which can capture more subtle features in the partial discharge signal and accurately simulate the local features of the actual discharge process. At the same time, the global attention mechanism of the Transformer corresponds to the global characteristics of the partial discharge signal in the actual scene, and captures the overall distribution of the signal from the time and frequency domains, thereby enhancing the recognition ability of different types of discharge features.

[0021] 3. The present invention adopts AdamW optimizer to optimize the hyperparameters of the partial discharge diagnosis model. The initial learning rate of the optimizer is 0.0001, the hyperparameter α is set to 0.2, and the hyperparameter β is set to 0.3. These parameters regulate the update amplitude of the model during the training process, affect the model convergence speed and the final diagnostic accuracy, and help maintain the stability and generalization ability of the model in complex partial discharge diagnosis tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A flow chart of the method is provided for the present invention;

[0023] Figure 2 It is the partial discharge experimental platform architecture in Example 1 of the present invention;

[0024] Figure 3 Schematic diagram of four typical defects in Example 1 of the present invention;

[0025] Figure 4 It is a schematic diagram of the structure of the C4F7N / CO2 / O2 mixed gas GIS partial discharge diagnostic model in Example 1 of the present invention;

[0026] Figure 5 Schematic diagram of the structure of the ResCNN Block module in Example 1 of the present invention;

[0027] Figure 6 Schematic diagram of the structure of the Transformer Block module in Example 1 of the present invention. DETAILED DESCRIPTION

[0028] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0029] Example 1

[0030] In view of the problem that the existing detection methods for SF6 are difficult to apply to the local discharge detection of new environmentally friendly gas GIS, the present invention provides a local discharge diagnosis method for C4F7N / CO2 / O2 mixed gas. The method is based on CNN and Transformer to build a neural network model to achieve accurate diagnosis of local discharge in GIS, and has at least the following advantages in practical applications:

[0031] 1) High-precision diagnosis: In practical applications, C4F7N / CO2 / O2 mixed gas is used to replace SF6 in GIS equipment. The online monitoring of partial discharge through the model can greatly improve the accuracy of fault diagnosis. Experimental results show that the diagnostic accuracy of this method on the test set is as high as 99.99%, far exceeding the diagnostic effect of traditional methods;

[0032] 2) Compatible with multiple signal inputs: This method is compatible with both UHF electromagnetic waves and ultrasonic signals, capturing electromagnetic and acoustic wave characteristics respectively, so as to more comprehensively monitor GIS partial discharge phenomena, which has important practical value for fault diagnosis of different types of partial discharge sensors in actual power systems;

[0033] 3) Fewer parameters and high real-time performance: The model has a small number of parameters (about 3.3M), which can achieve fast processing and real-time monitoring in actual operation. It is suitable for large-scale deployment in power systems and realizes continuous monitoring and diagnosis of GIS equipment.

[0034] Therefore, the method provided by the present invention is of great significance for online monitoring of partial discharge of new environmentally friendly gas GIS. Figure 1 As shown, the method comprises the following steps:

[0035] Step S1, based on a pre-built partial discharge experimental platform, obtaining partial discharge signals of typical defects in a C4F7N / CO2 / O2 mixed gas to obtain a partial discharge identification data set;

[0036] Step S2, based on the partial discharge identification data set, diagnose the signal to be diagnosed using the trained C4F7N / CO2 / O2 mixed gas GIS partial discharge diagnosis model to determine the corresponding partial discharge type; wherein the C4F7N / CO2 / O2 mixed gas GIS partial discharge diagnosis model is jointly constructed based on CNN and Transformer.

[0037] The specific implementation process is as follows:

[0038] (1) A typical GIS partial discharge model was built, and an environmentally friendly C4F7N / CO2 / O2 mixed gas GIS partial discharge experimental platform was constructed based on the real GIS. A partial discharge experiment was carried out on the platform to obtain the GIS partial discharge signal.

[0039] Specifically, if Figure 2 As shown, this embodiment builds a partial discharge experimental platform based on 252kv GIS to collect partial discharge signals of typical defects in C4F7N / CO2 / O2 mixed gas. The high-voltage power supply is a corona-free AC power supply. This embodiment uses a UHF electromagnetic wave sensor placed on the inner side of the handhole cover and an ultrasonic sensor attached to the outer wall of the GIS to capture the UHF electromagnetic wave signal and ultrasonic signal of the PD, and uses the Haefely DDX 9121b PD detector to detect the initial voltage of the partial discharge. Among them, the signal acquisition frequencies of UHF electromagnetic waves and ultrasonic waves are set to 5GHz and 500kHz respectively. The UHF sensor captures the ultra-high frequency electromagnetic wave signal, reflecting the high-frequency electromagnetic wave characteristics generated during the partial discharge process; the ultrasonic sensor captures the sound waves generated during the discharge. These two sets of data together constitute the feature data input into the diagnostic model based on CNN and Transformer, helping the model to simultaneously identify the characteristics of electromagnetic waves and sound waves during the discharge process.

[0040] Finally, the captured PD signal is recorded by a LeCroy oscilloscope. In order to be consistent with the actual application scenario, the pressure of the C4F7N / CO2 / O2 mixed gas filled in the GIS is 0.5MPa, and the proportion of C4F7N / CO2 / O2 in the C4F7N / CO2 / O2 mixed gas is 8.5% / 86% / 5.5%. Among them, the parameter setting of 0.5Mpa simulates the working pressure of the GIS insulating medium in actual applications. Changes in pressure will affect the intensity and characteristics of the partial discharge signal. By accurately setting the pressure, the model can better capture the partial discharge behavior of different insulating gases under actual working conditions.

[0041] The following are the settings for four typical defect models. The schematic diagrams for each defect are as follows: Figure 3 shown. Figure 3 Part (a) is the needle tip defect model. The needle tip defect model uses a metal tip with a bottom diameter of 8 mm and a length of 20 mm to simulate the tip defect inside the GIS. The metal tip is 5 mm away from the low-voltage conductor. Figure 3 Part (b) is a suspension defect model. For the suspension defect, this embodiment uses epoxy resin to fix metal bolts between the upper and lower plates, wherein the metal screws are 8 mm away from the high-voltage conductor and 5 mm away from the low-voltage conductor. Figure 3 Part (c) is the particle defect model. For the particle defects, small steel balls with a diameter of about 0.5 mm are used to simulate free metal particles. These small steel balls are placed on the lower plate and surrounded by a cylinder made of epoxy resin to prevent them from rolling down. Figure 3Part (d) is a surface discharge defect model. The surface discharge defect is achieved by making the metal bolts leading out of the upper electrode plate contact the upper and lower bottom surfaces of the epoxy resin cylinder with a diameter of 50 mm and a thickness of 20 mm with the lower electrode plate respectively.

[0042] During the experiment, in order to ensure that local discharge occurs stably, this embodiment uses the PD initial voltage of each defect to increase by 10kV as the applied voltage of the defect PD experiment. For each defect type, this embodiment respectively collects the local discharge signal of the defect in the C4F7N / CO2 / O2 mixed gas. When collecting signals, an oscilloscope is used to simultaneously record the signals of the ultrasonic sensor and the UHF sensor, and 10,000 groups of signals are collected under each experimental condition. Among them, the sampling rate of the ultrasonic signal is set to 500kHz, and the recording is 20ms; the sampling rate of the UHF signal is set to 5GHz, and the recording is 2us. Then, the obtained time domain signal is subjected to fast Fourier transform (FFT) and normalized. Finally, a PD recognition data set is constructed with the processed signal, and the data set is randomly divided into a training set, a validation set, and a test set in a ratio of 6:2:2.

[0043] (2) Construct an environmentally friendly C4F7N / CO2 / O2 mixed gas GIS partial discharge diagnosis model based on CNN and Transformer.

[0044] Specifically, in order to realize automatic recognition of PD signals, this embodiment uses a basic one-dimensional convolutional layer and a Transformer Block module to build a PD signal classification model. The input of the model is the spectrum of the PD signal, and the output is the classification probability. The structure of the model is as follows: Figure 4 shown.

[0045] The model consists of two parts: the encoder and the classifier. The encoder is used to encode the input spectrum to generate high-dimensional features, and the classifier is responsible for regressing the signal category from the high-dimensional features. The encoder includes the first convolutional layer, the ResCNN Block module, the second convolutional layer, and the Transformer Block module connected in sequence. The classifier includes the average pooling layer and the linear layer connected in sequence. The output of the Transformer Block module is connected to the input of the average pooling layer.

[0046] In the encoder part, a one-dimensional convolution layer with a large convolution kernel (i.e., the first convolution layer) is first used to capture the preliminary features of the input spectrum. The convolution kernel size of this convolution layer is 7, the step size is 2, and the output dimension is 32. Here, this embodiment selects a larger convolution kernel to improve the receptive field, which is very important for capturing subtle features in the spectrum. Next, four consecutive ResCNN Block modules are used to increase the dimension of the features and shorten the feature length at the same time. Each ResCNN Block module includes multiple layers of one-dimensional convolution layers, wherein the first one-dimensional convolution layer is zero-filled to keep the input and output sizes consistent. This convolution layer is used to extract low-level features of the mixed gas partial discharge signal, and the remaining convolution layers are used to gradually extract deep features. Each convolution layer is nonlinearly processed by the ReLU activation function, and finally the corresponding input features and output features are added through residual connections to retain the original information and improve the stability of model training.

[0047] like Figure 5 As shown in the figure, taking ResCNN Block,64 as an example, the module first performs preliminary feature extraction through Conv 1D,16,kr 1, then downsamples through Conv 1D,16,kr 3, stride 2, and then expands the features through Conv 1D,64,kr 1, and is connected to Conv 1D,64,kr 1, stride 2 through a bypass, and the results of the two are added; then enter the submodule, which consists of multiple stacked convolutional layers, including Conv 1D,32,kr1, Conv 1D,32,kr 3, and Conv1D,64,kr 1, and repeats N times to achieve deep feature extraction, and finally adds it to the initial input through a residual connection. The output of the last ResCNN Block module is connected to the second convolutional layer, the convolution kernel size of this convolutional layer is 1, and the output dimension is 128. The aforementioned kr represents the convolution kernel size, and stride represents the step size.

[0048] In the final part of the encoder, three Transformer Block modules with global attention are used to capture long-range dependencies within the data to further improve the representation ability of features. Figure 6As shown in the figure, each TransformerBlock module captures the global information in the input features through the Multi-Head Attention layer, then adds the output of the Multi-Head Self-Attention layer to the original input through the residual connection (Add), and then normalizes it through the normalization (Norm) layer; then, the normalized result is input to the feedforward neural network layer, and the output features of the feedforward neural network are added to the input features through the residual connection, and finally the normalization layer is used for secondary normalization to complete the further extraction and processing of the features. Among them, the multi-head self-attention layer is used to capture the global features of the mixed gas partial discharge signal, and the feedforward neural network layer is used for further nonlinear mapping. Each layer is added to the corresponding input features through the residual connection, and the normalization layer is used for normalization after the residual connection to improve the efficiency of feature extraction and the training stability of the model.

[0049] After the features are extracted by the encoder, the classifier part of the network first reduces the dimension of the global features through an average pooling layer, then passes through a linear layer with 64 neurons, which uses the ReLU activation function for nonlinear transformation, and finally passes through a linear layer with 4 output nodes to achieve the final prediction output for the four categories.

[0050] (3) Use the collected partial discharge signals to train the partial discharge diagnosis model.

[0051] All models in this embodiment are implemented based on Pytorch and trained on an NVIDIA GeForce RTX 4060 GPU. During training, the AdamW optimizer is used to optimize the model parameters. The initial learning rate of the optimizer is set to 0.0001, the hyperparameter α is set to 0.2, and the hyperparameter β is set to 0.3. The training data in the training process uses the training set part in the PD recognition data set, and the cross entropy loss (CE Loss) is used to calculate the classification loss during training, which can be expressed as:

[0052]

[0053] In the formula, represents the classification loss, N is the number of samples, M is the number of categories, y ic Indicates whether sample i belongs to category c, 1 if it does, 0 if it does not, y′ ic is the probability that the model predicts that sample i belongs to category c.

[0054] (4) A partial discharge diagnosis experiment was conducted on an environmentally friendly C4F7N / CO2 / O2 mixed gas GIS partial discharge experimental platform and the results were statistically analyzed to verify the effectiveness of the method provided in this embodiment for partial discharge diagnosis of an environmentally friendly C4F7N / CO2 / O2 mixed gas GIS.

[0055] Table 1 shows the comparison results of the partial discharge diagnosis method (denoted by Ours) proposed in this embodiment and other popular partial discharge diagnosis methods. UHF and US in brackets represent the training and test data of the model, UHF represents ultra-high frequency electromagnetic wave data, and US represents ultrasonic wave data. The data given in the table are the recognition accuracy of the model on the test set.

[0056] Table 1 Diagnostic accuracy of different methods for partial discharge in environmentally friendly C4F7N / CO2 / O2 mixed gas GIS

[0057]

[0058] This embodiment selects four currently popular methods for training and testing on the PD recognition data set. It can be seen that the two methods based on convolutional structures, ResNet and GoogLeNet, have low diagnostic accuracy on C4F7N / CO2 / O2 mixed gas data due to poor fitting ability. LSTM has a low diagnostic accuracy of only 49.8% due to the problem of poor fitting ability caused by gradient disappearance when processing long sequences. Transformer can capture high-quality global features due to its long-distance attention mechanism. And global features have strong representation capabilities, so the accuracy of Transformer on C4F7N / CO2 / O2 mixed gas data reaches 84.6%. However, none of the above four methods has made a separate design for the characteristics that local features and global features need to be closely combined for local discharge diagnosis, so the diagnostic accuracy is far lower than the method proposed in this embodiment. This proves that the design of the method proposed in this embodiment is necessary and effective, and successfully solves the diagnosis problem of local discharge in environmentally friendly C4F7N / CO2 / O2 mixed gas GIS.

[0059] In summary, in order to improve the accuracy and practicality of partial discharge diagnosis in GIS, the partial discharge diagnosis method based on CNN and Transformer proposed in this embodiment first builds a partial discharge experimental platform based on a real GIS, and collects partial discharge signals of four typical defects in C4F7N / CO2 / O2 mixed gas based on the platform to build a complete partial discharge identification data set. Then, an environmentally friendly C4F7N / CO2 / O2 mixed gas GIS partial discharge diagnosis model based on CNN and Transformer is constructed, and the model is trained on the partial discharge identification data set. Finally, the trained partial discharge diagnosis model is experimentally tested. The results show that the model proposed in the present invention can accurately diagnose the type of partial discharge with only a small amount of parameters, which is significantly better than the existing methods. Therefore, the environmentally friendly C4F7N / CO2 / O2 mixed gas GIS partial discharge diagnosis method based on CNN and Transformer has high accuracy and real-time performance in partial discharge diagnosis applications, and can be compatible with ultra-high frequency electromagnetic wave signals and ultrasonic signal inputs at the same time. It is an effective diagnostic method with broad application prospects.

[0060] Example 2

[0061] This embodiment provides a partial discharge diagnostic device for C4F7N / CO2 / O2 mixed gas, including a memory, a processor, and a program stored in the memory. When the processor executes the program, the method provided in Example 1 is implemented. The device processor includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or computer program instructions loaded from a storage unit to a random access memory (RAM). In the RAM, various programs and data required for the operation of the device can also be stored. The CPU, ROM, and RAM are connected to each other through a bus. The input / output (I / O) interface is also connected to the bus. Multiple components in the device are connected to the I / O interface, including: input units, such as keyboards, mice, etc.; output units, such as various types of displays, speakers, etc.; storage units, such as disks, optical disks, etc.; and communication units, such as network cards, modems, wireless communication transceivers, etc. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunication networks. The processing unit performs the various methods and processes described above, such as the various steps of the method in the aforementioned embodiment. For example, in some embodiments, the steps of the aforementioned method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via a ROM and / or a communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more steps of the method described above may be performed. Alternatively, in other embodiments, the CPU may be configured to perform the aforementioned method in any other appropriate manner (e.g., by means of firmware). The functions described above may be performed at least in part by one or more hardware logic components. For example, without limitation, the hardware logic components of exemplary types that may be used include: field programmable gate array (FPGA), application specific integrated circuit (ASIC), application specific standard product (ASSP), system on chip (SOC), complex programmable logic device (CPLD), etc.

[0062] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for partial discharge diagnosis of C4F7N / CO2 / O2 mixed gas, characterized in that: The following steps are involved: Based on the pre-built partial discharge experimental platform, the partial discharge signals of typical defects in the C4F7N / CO2 / O2 mixed gas are obtained to obtain the partial discharge identification data set; Based on the partial discharge identification data set, the trained C4F7N / CO2 / O2 mixed gas GIS partial discharge diagnosis model is used to diagnose the signal to be diagnosed and determine the corresponding partial discharge type; Among them, the C4F7N / CO2 / O2 mixed gas GIS partial discharge diagnosis model is jointly constructed based on CNN and Transformer.

2. The method for partial discharge diagnosis of C4F7N / CO2 / O2 mixed gas according to claim 1, characterized in that: The partial discharge experimental platform is constructed based on a real GIS.

3. The method for partial discharge diagnosis of C4F7N / CO2 / O2 mixed gas according to claim 1, characterized in that: The typical defects include pinpoint defects, suspension defects, particle defects and surface discharge defects.

4. The method for partial discharge diagnosis of C4F7N / CO2 / O2 mixed gas according to claim 3, characterized in that: The needle tip defect is simulated by using a metal tip, the suspension defect is simulated by fixing a metal bolt between the upper and lower plates using epoxy resin, the particle defect is simulated by using a steel ball, and the surface discharge defect is simulated by making the metal bolt leading out of the upper plate contact the upper and lower bottom surfaces of the epoxy resin cylinder with the lower plate respectively.

5. The method for partial discharge diagnosis of C4F7N / CO2 / O2 mixed gas according to claim 1, characterized in that: The C4F7N / CO2 / O2 mixed gas GIS partial discharge diagnosis model includes an encoder and a classifier. The encoder includes a first convolutional layer, a ResCNN Block module, a second convolutional layer and a Transformer Block module connected in sequence. The classifier includes an average pooling layer and a linear layer connected in sequence. The output of the Transformer Block module is connected to the input of the average pooling layer.

6. The method for partial discharge diagnosis of C4F7N / CO2 / O2 mixed gas according to claim 5, characterized in that: Each of the ResCNN Block modules includes multiple layers of one-dimensional convolutional layers, wherein the first one-dimensional convolutional layer is zero-filled to extract low-level features of the mixed gas partial discharge signal, and the remaining convolutional layers are used to gradually extract deep features. Each convolutional layer is nonlinearly processed through the ReLU activation function, and finally the corresponding input features and output features are added through residual connections.

7. The method for partial discharge diagnosis of C4F7N / CO2 / O2 mixed gas according to claim 5, characterized in that: The Transformer Block module includes a multi-head self-attention layer, a feedforward neural network layer and a normalization layer. The multi-head self-attention layer is used to capture the global characteristics of the mixed gas partial discharge signal, and the feedforward neural network layer is used for further nonlinear mapping. Each layer is added to the corresponding input feature through a residual connection, and the normalization layer is used to perform normalization processing after the residual connection.

8. The method for partial discharge diagnosis of C4F7N / CO2 / O2 mixed gas according to claim 1, characterized in that: When training the C4F7N / CO2 / O2 mixed gas GIS partial discharge diagnosis model, the AdamW optimizer is used to optimize the model parameters, and the cross entropy loss is used to calculate the classification loss.

9. The method for partial discharge diagnosis of C4F7N / CO2 / O2 mixed gas according to claim 8, characterized in that: The expression of the cross entropy loss is: In the formula, represents the classification loss, N is the number of samples, M is the number of categories, y ic Indicates whether sample i belongs to category c, 1 if it does, and 0 if it does not. i ' c is the probability that the model predicts that sample i belongs to category c.

10. A partial discharge diagnostic device for C4F7N / CO2 / O2 mixed gas, comprising a memory, a processor, and a program stored in the memory, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 9 is implemented.

Images