A method for partial discharge pattern recognition of superconducting cables based on residual network Resnet50

Abstract

The present invention discloses a method for identifying partial discharge patterns of superconducting cables based on a residual network Resnet50, comprising: step one: using a partial discharge detection system to collect partial discharge fault signals of different types of superconducting cables, and using the collected fault signals to construct a phase distribution pattern PRPD spectrum data set of superconducting cable faults; step two: classifying the PRPD spectra in the data set according to different discharge types, and dividing them into a training set, a validation set, and a test set according to a certain ratio, and then preprocessing the images; step three: constructing a Resnet50 network model, using training samples and test samples as inputs, and training and testing the network model. By using residual blocks, a deeper network can be trained, which effectively solves the problems of gradient vanishing and gradient explosion caused by the deepening of the network depth, especially when training a deeper network, it shows good performance, which is helpful to realize the pattern recognition of superconducting cable faults.

Description

Technical Field

[0001] The present invention relates to the technical field of superconducting cable fault identification, and in particular to a superconducting cable partial discharge pattern recognition method based on a residual network Resnet50. Background Art

[0002] Different types of partial discharges correspond to different insulation defects. If the characteristics of partial discharges can be determined, the insulation condition of power superconducting cables can be better judged. To understand the insulation condition inside power superconducting cables, the most effective solution is to detect partial discharges of superconducting cables and perform pattern recognition on discharge types. Based on long-term experience, the output and changes of partial discharge signals of power superconducting cables can reflect the insulation condition of superconducting cables. Therefore, pattern recognition can make effective judgments on the insulation state and degree of damage of power superconducting cables, and provide more scientific information for accurately evaluating the insulation performance and operating conditions of superconducting cables.

[0003] With the emergence of computers and the rise of artificial intelligence technology, pattern recognition has rapidly developed into a discipline in the 1960s and has been widely used in many fields. At present, pattern recognition methods are constantly developing in the direction of intelligence, and higher requirements are put forward for the system's adaptive ability, learning ability, and fault tolerance. Since the 1990s, pattern recognition methods have been gradually promoted and applied to the field of partial discharge. Compared with the traditional judgment of partial discharge type based on the actual engineering experience and theoretical knowledge of relevant experts, pattern recognition methods have significantly improved the scientificity and effectiveness of recognition results.

[0004] However, the research on partial discharge pattern recognition technology is still in its infancy. On the one hand, many complex and uncertain factors affect the acquisition of partial discharge signals, and the interference of the surrounding environment and other reasons cause the partial discharge signals to be mixed with many interference signals. It is necessary to find a better signal processing method to obtain more accurate partial discharge signals. On the other hand, the information contained in the partial discharge signal is relatively complex, and its connotation and rules are not yet fully understood. Therefore, it is necessary to conduct more in-depth research on the principle of partial discharge and its practical problems.

[0005] To this end, we propose a superconducting cable partial discharge pattern recognition method based on the residual network Resnet50. The present invention realizes the recognition of the type of superconducting cable partial discharge by collecting different types of superconducting cable partial discharge phase distribution pattern PRPD spectra, thereby identifying the type of partial discharge corresponding to each spectrum. This pattern recognition method can replace the traditional work of judging the type of superconducting cable partial discharge based on the actual engineering experience of relevant experts in the industry. At the same time, the Resnet50 network model is adopted, and the training process of the convolutional neural network is accelerated through the idea of ​​residual learning, which effectively avoids the problems of gradient vanishing and gradient explosion, and can complete the model training in a shorter time and achieve a higher recognition accuracy. Summary of the invention

[0006] The purpose of the present invention is to solve the defects existing in the prior art, and propose a superconducting cable partial discharge pattern recognition method based on the residual network Resnet50, which is used to solve the problems of gradient disappearance and gradient explosion, and can also ensure good performance when training deeper networks.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] A method for recognizing partial discharge patterns of superconducting cables based on a residual network Resnet50 comprises the following steps:

[0009] S1: Use the partial discharge detection system to collect different types of fault signals of superconducting cables, and use the collected signals to construct a PRPD spectrum dataset of partial discharge phase distribution patterns of superconducting cable insulation defects;

[0010] S2: Classify the PRPD spectra in the data set according to different fault types, perform image preprocessing and data augmentation, and then divide them into training set, validation set and test set according to a certain ratio;

[0011] S3: Use transfer learning to build a Resnet50 residual network model, complete the configuration of various model parameters, take training samples and verification samples as input, and train and verify the network model;

[0012] S4: Perform convolution, pooling, activation, and regression operations on the network model to update the network parameters, and then train the network model. When the loss rate function no longer decreases and the accuracy function no longer increases, the trained network model is obtained, and the obtained model is used for testing;

[0013] S5: The optimal Resnet50 network model is used to identify the fault type of superconducting cables, thus constructing a superconducting cable fault pattern recognition system based on Resnet50.

[0014] Furthermore, in the step S2, the specific steps are as follows:

[0015] S21: Collect PRPD spectra of various superconducting cable partial discharge phase distribution modes through the partial discharge detection system and classify them according to the partial discharge type;

[0016] S22: performing image preprocessing on the collected partial discharge signal;

[0017] S23: performing data augmentation processing on the preprocessed image;

[0018] S24: The processed images are used as the data set of the model, and are divided into a training set, a validation set, and a test set according to a certain ratio.

[0019] Furthermore, in the step S21, the local discharge of the superconducting cable is mainly manifested in three categories through the PRPD spectrum: air gap discharge, suspension discharge and normal external noise interference.

[0020] Furthermore, in the step S22, in order to ensure that the data has the same specifications, size and format, the image needs to be preprocessed, wherein the preprocessing method includes an image sample normalization preprocessing method and an image pixel value processing preprocessing method.

[0021] Furthermore, in the step S23, the number of partial discharge samples is increased by data augmentation, and the data is augmented by methods such as image translation, chromaticity change, brightness change, random cropping, etc.

[0022] Furthermore, in the step S24, 70% of the images in the processed partial discharge phase distribution pattern PRPD spectrum data set are extracted as a training set, 10% of the images are extracted as a verification set, and 20% of the images are extracted as a test set.

[0023] Furthermore, in the step S3, the specific steps are as follows:

[0024] S31: Use the Resnet50 model in the transfer learning approach;

[0025] S32: configure and adjust various parameters of the model;

[0026] S33: Take the training samples and validation samples as input to train and validate the network model.

[0027] Furthermore, in the step S4, the specific steps are as follows:

[0028] S41: Define the loss function and use the cross entropy function as the loss function. The calculation formula is as follows:

[0029]

[0030] Where x is the sample value of the random distribution, p is the expected probability distribution of sample x, and q is the actual probability distribution of sample x;

[0031] S42: Define the optimization function. To ensure that the model training can achieve the best speed and accuracy, adaptive adjustment of simulated annealing is implemented based on cosine decay. The cosine decay formula is as follows:

[0032]

[0033] Among them, α0 represents the initial learning rate, α t represents the learning rate at the tth iteration, and T represents the total number of iterations;

[0034] S43: Set the parameters of model training, the image resolution is set to 224×224, the initial learning rate is set to 0.001, and the number of training rounds is set to 100 rounds;

[0035] S44: Train the network model. When the loss rate function no longer decreases and the accuracy function no longer increases, a trained network model is obtained, and the obtained model is used for testing test samples.

[0036] Furthermore, the superconducting cable partial discharge pattern recognition system based on Resnet50 in the S5 step includes: a partial discharge detection unit, a discharge spectrum construction unit, a data diversity unit, a model training unit, a model testing unit and an identification unit, wherein the partial discharge detection unit is connected to the data diversity unit through the discharge spectrum construction unit, the data diversity unit is respectively connected to the model training unit and the model testing unit, the model training unit is connected to the model testing unit, and the partial discharge detection unit is connected to the model testing unit through the identification unit.

[0037] Furthermore, the partial discharge detection unit is used to collect several common partial discharge types of superconducting cables and partial discharge signals of these typical defects; the discharge spectrum construction unit is used to construct a PRPD spectrum data set of local discharge phase distribution patterns of superconducting cables; the data diversity unit is used to divide the PRPD spectra in the data set into training sets, test sets and verification sets according to a certain ratio; the model training unit is used to use training samples and verification samples as input to train and verify the Resnet50 network model; the model testing unit is used to test the trained network model on the test samples, and the individual recognition rates and average recognition rates of various local discharge types can be obtained from the test results; the recognition unit is used to use the optimal Resnet50 network model to perform pattern recognition on the fault types of local discharge of superconducting cables.

[0038] Compared with the prior art, the present invention has the following advantages:

[0039] 1. The present invention solves the problem that there are many types of noise interference and difficulty in judging the type of partial discharge in the process of collecting partial discharge signals of superconducting cables. The Resnet50 network model is used to realize the recognition of the type of partial discharge of superconducting cables. The recognition accuracy is greater than 97%, which meets the requirements of partial discharge pattern recognition of superconducting cables and has great application prospects.

[0040] 2. The present invention provides a method for identifying a local discharge pattern of a superconducting cable, comprising: using a local discharge detection system to collect local discharge signals of different types of superconducting cables, and using the collected local discharge signals to construct a PRPD spectrum data set of a local discharge phase distribution pattern of insulation defects of superconducting cables; classifying the PRPD spectra in the data set according to different discharge types, and performing image preprocessing and data augmentation on the PRPD spectra, and then dividing them into a training set, a verification set and a test set according to a certain ratio; adopting a transfer learning method to construct a Resnet50 residual network model, complete the configuration of various parameters of the model, and use the training samples and the verification samples as input to train and verify the network model; performing convolution, pooling, activation and regression operations on the network model to update the network parameters, and then training the network model, and when the loss rate function no longer decreases and the accuracy function no longer increases, a trained network model is obtained, and the obtained model is used for testing; using the optimal Resnet50 network model to identify the fault type of local discharge of superconducting cables, thereby constructing a local discharge pattern recognition system for superconducting cables based on Resnet50.

[0041] 3. The present invention adopts data augmentation to increase the number of partial discharge samples of superconducting cables. According to the internal structure of power superconducting cables and the partial discharge signal characteristics of typical defects, image translation, chromaticity change, brightness change, random cropping and other methods are used to augment the data, avoiding the problem of poor learning effect caused by a small number of samples, thereby further improving the effect of model training.

[0042] 4. The present invention adopts a simulated annealing algorithm with adaptive adjustment of learning rate parameters based on cosine decay. Through this algorithm, at the beginning of model training, a larger learning rate is used to ensure that the model has a faster convergence speed. When the model converges to the vicinity of the optimal point, the learning rate is automatically reduced to avoid back and forth oscillations. This can effectively ensure that the model training can achieve the optimal speed and accuracy.

[0043] 5. The present invention has made further improvements on the basis of the Resnet50 network model. The Resnet50 network is a residual network. There are two different connection relationships between input and output in the residual network. One is that the input data is output after passing through multiple neural layers, and the other is that the input data is directly connected to the output. The use of the Resnet50 network model can effectively solve the problems of gradient explosion and gradient disappearance. In addition, the residual network converts the deep structure of the network into a parallel structure during forward propagation. In this way, even if the capacity of the deep neural network model becomes larger, the model will not have overfitting and affect the training effect of the model. At the same time, the results show that it is feasible to apply the Resnet50 model to the classification and identification of local discharge fault types of superconducting cables through the method of transfer learning. This pattern recognition method can complete the training of the model in a relatively short time and can obtain a higher recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0045] Figure 1 is a flow chart of the steps of the present invention;

[0046] Figure 2 It is a flow chart of the superconducting cable partial discharge pattern recognition system based on Resnet50 in the present invention;

[0047] Figure 3 It is a schematic diagram of the operation logic of the present invention;

[0048] Figure 4 A comparison chart of feedforward neural network and residual neural network;

[0049] Figure 5 This is the Resnet50 network structure diagram. DETAILED DESCRIPTION

[0050] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0051] Example

[0052] Reference Figure 1 and Figure 3 , a method for identifying partial discharge patterns of superconducting cables, comprising the following steps:

[0053] Step 101: using a partial discharge detection system to collect different types of partial discharge signals of a superconducting cable, and using the collected partial discharge signals to construct a PRPD spectrum data set of a partial discharge phase distribution mode of insulation defects of the superconducting cable;

[0054] Specifically, the local discharge phase distribution PRPD mode is a commonly used discharge mode, also known as the φ-qn mode. This mode describes the relationship between the power frequency phase φ (0°~360°) corresponding to the local discharge pulse, the discharge amount q (since the discharge amount cannot be calibrated by standard when using the ultra-high frequency method for detection, q represents the pulse amplitude) and the number of discharges n.

[0055] Step 103: classify the PRPD spectra in the data set according to different discharge types, perform image preprocessing and data augmentation on them, and then divide them into a training set, a validation set, and a test set according to a certain ratio;

[0056] Specifically, the partial discharge of superconducting cables is mainly manifested in three categories through PRPD spectra: air gap discharge, suspension discharge and normal external noise interference;

[0057] Since the original PRPD spectra collected are from different sources, have different sizes and formats, they cannot be put into the same model for training. In order to ensure that the data specifications, sizes and formats are the same, the images need to be preprocessed, mainly including two preprocessing methods: image sample normalization and image pixel value processing.

[0058] In addition, model training requires that the resolution of the input image should be 224×224, while the resolution of the original PRPD spectrogram collected is random, so it is necessary to unify the image size by scaling and cropping. The calculation formula for the image scaling factor is as follows:

[0059]

[0060] Among them, w and h represent the width and height of the image respectively, and scale represents the scaling factor.

[0061] After the image is scaled, the center cropping method is used to crop the image, and the width and height of the cropping are set to 224, thereby obtaining an image with a resolution of 224×224. In order to more effectively improve the effect of model training, it is also necessary to normalize the scaled and cropped image. The present invention uses a zero mean normalization method to normalize the image data. The normalization formula is as follows:

[0062]

[0063] Among them, z represents the normalized value required, x represents the original value, and X0 and σ represent the mean and variance of the original value.

[0064] Different image data has different corresponding image formats. In order to keep the format of image data consistent, it is necessary to process the pixel values ​​of the image, that is, convert the image in BMP format into a matrix vector of RGB three channels and then store it.

[0065] More specifically, due to the small number of original data samples, especially the difficulty in obtaining PRPD spectrum samples of local discharge of superconducting cables, and the small number of samples will greatly affect the learning effect of deep learning, it is necessary to increase the number of local discharge samples by data augmentation. According to the internal structure of power superconducting cables and the local discharge signal characteristics of typical defects, image translation, chromaticity change, brightness change, random cropping and other methods are used to augment the data.

[0066] Furthermore, 70% of the images in the processed partial discharge phase distribution pattern PRPD spectrum dataset are extracted as training sets, 10% of the images are used as validation sets, and 20% of the images are used as test sets.

[0067] Step 105: Use transfer learning to build a Resnet50 residual network model, complete the configuration of various model parameters, take the training sample and the verification sample as input, and train and verify the network model;

[0068] Reference Figure 4 Specifically, the Resnet network used is also called a residual neural network, which refers to a module in which residual learning is added to a traditional convolutional neural network. It can be defined by the formula:

[0069] y=F(x,{Wi})+x

[0070] Among them, y represents the output, F(x, {Wi}) represents the residual part, and x represents the sample.

[0071] The residual network effectively solves the problem of model degradation through short-cut connection. Short-cut connection adds a jump connection on the basis of the standard feed-forward convolutional neural network, thereby skipping some convolutional layers and connecting directly to the output.

[0072] The calculation process of the predicted value H(x) after two layers of feedforward neural network is as follows:

[0073] f(x) = ReLU(bi+x in w1)

[0074] H(x)=ReLU(b2+f(x)·w2)

[0075] Among them, x inis the input feature; w1 and w2 are the weights of the first and second neural layers respectively; b1 and b2 are the biases of the first and second neural layers respectively; ReLU is the activation function.

[0076] Assuming that the functional relationship between the predicted value H(x) and x in the residual network satisfies H(x) = 2x, then

[0077] H(x)=ReLU(b2+f(x)·w2)+x in

[0078] Reference Figure 5 ,The network structure of the Resnet50 model consists of 49 convolutional layers and 1 fully connected layer.

[0079] Step 107: Perform convolution, pooling, activation, and regression operations on the network model to update the network parameters, and then train the network model. When the loss rate function no longer decreases and the accuracy function no longer increases, the trained network model is obtained, and the obtained model is used for testing;

[0080] Specifically, updating the network parameters includes the following steps:

[0081] Step 1: Define the loss function and use the cross entropy function as the loss function. The calculation formula is as follows:

[0082]

[0083] Among them, x is the sample value of the random distribution, p is the expected probability distribution of sample x, and q is the actual probability distribution of sample x.

[0084] Step 2: Define the optimization function. In order to ensure that the model training can achieve the best speed and accuracy, a simulated annealing algorithm with adaptive adjustment of learning rate parameters based on cosine decay is used. This algorithm can use a larger learning rate at the beginning of model training to ensure the convergence speed of the model, and automatically reduce the learning rate when the model converges to the optimal point to avoid oscillation. The cosine decay formula is as follows:

[0085]

[0086] Among them, α0 represents the initial learning rate, α t represents the learning rate at the tth iteration, and T represents the total number of iterations.

[0087] Step 3: Set the parameters for model training, the image resolution is set to 224×224, the initial learning rate is set to 0.001, and the number of training rounds is set to 100.

[0088] Step 109: The optimal Resnet50 network model is used to identify the fault type of partial discharge of the superconducting cable, thereby constructing a superconducting cable partial discharge pattern recognition system based on Resnet50.

[0089] According to an embodiment of the present invention, a superconducting cable partial discharge pattern recognition system based on Resnet50 is also provided.

[0090] Reference Figure 2 A superconducting cable partial discharge pattern recognition system based on Resnet50 includes: a partial discharge detection unit, a discharge spectrum construction unit, a data diversity unit, a model training unit, a model testing unit and an identification unit, wherein the partial discharge detection unit is connected to the data diversity unit through the discharge spectrum construction unit, the data diversity unit is respectively connected to the model training unit and the model testing unit, the model training unit is connected to the model testing unit, and the partial discharge detection unit is connected to the model testing unit through the identification unit.

[0091] in:

[0092] The partial discharge detection unit is used to collect partial discharge signals of several common partial discharge types and typical defects of superconducting cables;

[0093] The discharge spectrum building unit is used to build a superconducting cable partial discharge phase distribution mode PRPD spectrum data set;

[0094] The data collection unit is used to divide the PRPD spectra in the data set into a training set, a test set and a validation set according to a certain ratio;

[0095] The model training unit is used to take the training samples and the verification samples as input to train and verify the Resnet50 network model;

[0096] The model testing unit is used to test the trained network model on the test sample, and the individual recognition rate and average recognition rate of various partial discharge types can be obtained from the test results;

[0097] The identification unit is used to use the optimal Resnet50 network model to perform pattern recognition on the fault type of partial discharge of the superconducting cable.

[0098] In the present embodiment, the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, computer software or a combination of the two, and the programs corresponding to the software modules and method steps can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the technical field. In order to clearly illustrate the interchangeability of electronic hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0099] In order to better understand the technical solution of the present invention, it is further described below in conjunction with comparative examples.

[0100] Comparative Example 1

[0101] A method for identifying partial discharge patterns of superconducting cables comprises the following steps:

[0102] S1: Use the partial discharge detection system to collect different types of fault signals of superconducting cables, and use the collected signals to construct a PRPD spectrum dataset of partial discharge phase distribution patterns of superconducting cable insulation defects;

[0103] S2: Classify the PRPD spectra in the data set according to different fault types, perform image preprocessing and data augmentation, and then divide them into training set, validation set and test set according to a certain ratio;

[0104] S3: Feature extraction: extract features from the annotated PD PRPD spectrum;

[0105] S4: Train through a 3-layer BP neural network to obtain a trained network model.

[0106] Compared with the embodiment, in comparative example 1, on the basis of step S1 and step S2, after performing the "feature extraction" step in comparative example 1, a 3-layer BP network is used for model training.

[0107] Comparative Example 2

[0108] A method for identifying partial discharge patterns of superconducting cables comprises the following steps:

[0109] S1: Use the partial discharge detection system to collect different types of fault signals of superconducting cables, and use the collected signals to construct a PRPD spectrum dataset of partial discharge phase distribution patterns of superconducting cable insulation defects;

[0110] S2: Classify the PRPD spectra in the data set according to different fault types, perform image preprocessing and data augmentation, and then divide them into training set, validation set and test set according to a certain ratio;

[0111] S3: Feature extraction: extract features from the annotated PD PRPD spectrum;

[0112] S4: Train through SVM support vector machine to obtain a trained network model.

[0113] Compared with the embodiment, in comparative example 2, on the basis of step S1 and step S2, comparative example 2 performs the "feature extraction" step and then uses the SVM support vector machine to perform model training.

[0114] Compared with Comparative Examples 1 and 2, the embodiment eliminates the step of “feature extraction” and leaves the feature extraction to the convolution layer. The specific data of the recognition accuracy of the embodiment, Comparative Examples 1 and 2 are shown in Table 1.

[0115]

[0116] Table 1

[0117] As can be seen from the above, the superconducting cable partial discharge pattern recognition method based on the residual network Resnet50 provided by the present invention is aimed at the problem that there are many complex and uncertain factors in the recognition process, and the data collection process will be disturbed by the surrounding environment noise, making the recognition process difficult to control. The Resnet50 network model is used to classify and recognize the various types of partial discharge phase distribution pattern PRPD spectra collected, which can achieve a more ideal classification and recognition effect. In addition, the method of transfer learning is introduced to train the model, and the pre-trained model weights are transferred to the training process of the partial discharge pattern recognition model. Due to the huge image data set in deep learning, transfer learning can accelerate the convergence speed of the network model and can effectively improve the recognition accuracy of the model. The ResNet50 network introduces a residual structure, which can solve the gradient disappearance problem caused by the excessive depth of the neural network, not only enhance the ability of feature extraction, but also improve the accuracy of classification and recognition, and the generalization ability of the model is also improved, which is suitable for use in practical engineering applications, so as to better grasp the type and defect characteristics of internal defects of power superconducting cables.

[0118] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features.

Claims

1. A method for identifying partial discharge patterns of superconducting cables based on residual network Resnet50, characterized in that: The method for recognizing partial discharge patterns of superconducting cables based on the residual network Resnet50 includes the following steps: S1: Use the partial discharge detection system to collect different types of fault signals of superconducting cables, and use the collected signals to construct a PRPD spectrum dataset of partial discharge phase distribution patterns of superconducting cable insulation defects; S2: Classify the PRPD spectra in the data set according to different fault types, perform image preprocessing and data augmentation, and then divide them into training set, validation set and test set according to a certain ratio; S3: Use transfer learning to build a Resnet50 residual network model, complete the configuration of various model parameters, take training samples and verification samples as input, and train and verify the network model; S4: Perform convolution, pooling, activation, and regression operations on the network model to update the network parameters, and then train the network model. When the loss rate function no longer decreases and the accuracy function no longer increases, the trained network model is obtained, and the obtained model is used for testing; S5: The optimal Resnet50 network model is used to identify the fault type of superconducting cables, thus building a superconducting cable fault pattern recognition system based on Resnet50; In the step S2, the specific steps are as follows: S21: Collect PRPD spectra of various superconducting cable partial discharge phase distribution modes through the partial discharge detection system and classify them according to the partial discharge type; S22: performing image preprocessing on the collected partial discharge signal; S23: performing data augmentation processing on the preprocessed image; S24: The processed images are used as the data set of the model, and are divided into a training set, a validation set, and a test set according to a certain ratio; In the step S22, in order to ensure that the data has the same specifications, size and format, the image needs to be preprocessed, wherein the preprocessing method includes an image sample normalization preprocessing method and an image pixel value processing preprocessing method; In the step S23, the number of partial discharge samples is increased by data augmentation, and the data is augmented by image translation, chromaticity change, brightness change, and random cropping methods; In the step S24, 70% of the images in the processed partial discharge phase distribution pattern PRPD spectrum data set are extracted as a training set, 10% of the images are extracted as a verification set, and 20% of the images are extracted as a test set; In the S3 step, the specific steps are as follows: S31: Use the Resnet50 model in the transfer learning approach; S32: configure and adjust various parameters of the model; S33: taking the training sample and the validation sample as input, training and validating the network model; In the S4 step, the specific steps are as follows: S41: Define the loss function and use the cross entropy function as the loss function. The calculation formula is as follows: Where x is the sample value of the random distribution, p is the expected probability distribution of sample x, and q is the actual probability distribution of sample x; S42: Define the optimization function. To ensure that the model training can achieve the best speed and accuracy, adaptive adjustment of simulated annealing is implemented based on cosine decay. The cosine decay formula is as follows: Among them, α0 represents the initial learning rate, α t represents the learning rate at the tth iteration, and T represents the total number of iterations; S43: Set the parameters for model training, the image resolution is set to 224×224, the initial learning rate is set to 0.001, and the number of training rounds is set to 100; S44: training the network model, and when the loss rate function no longer decreases and the accuracy function no longer increases, obtaining a trained network model, and using the obtained model for testing test samples; The superconducting cable partial discharge pattern recognition system based on Resnet50 in the step S5 includes: a partial discharge detection unit, a discharge spectrum construction unit, a data diversity unit, a model training unit, a model testing unit and an identification unit, wherein the partial discharge detection unit is connected to the data diversity unit through the discharge spectrum construction unit, the data diversity unit is respectively connected to the model training unit and the model testing unit, the model training unit is connected to the model testing unit, and the partial discharge detection unit is connected to the model testing unit through the identification unit.

2. The method for identifying partial discharge patterns of superconducting cables based on residual network Resnet50 according to claim 1 is characterized in that: In the step S21, the local discharge of the superconducting cable is manifested in three categories through the PRPD spectrum: air gap discharge, suspension discharge and normal external noise interference.

3. The method for identifying partial discharge patterns of superconducting cables based on residual network Resnet50 according to claim 1 is characterized in that: The partial discharge detection unit is used to collect the partial discharge types occurring in superconducting cables and the partial discharge signals of these typical defects; the discharge spectrum construction unit is used to construct a PRPD spectrum data set of the partial discharge phase distribution mode of superconducting cables; the data diversity unit is used to divide the PRPD spectra in the data set into a training set, a test set and a verification set according to a certain ratio; the model training unit is used to take the training samples and the verification samples as input to train and verify the Resnet50 network model; The model testing unit is used to test the trained network model on the test sample, and obtain the individual recognition rate and average recognition rate of various partial discharge types from the test results; the recognition unit is used to use the optimal Resnet50 network model to perform pattern recognition on the fault type of partial discharge of superconducting cables.

Images