Electrical equipment partial discharge defect type and fault level evaluation method based on convolutional neural network

Abstract

The invention provides an electrical equipment partial discharge defect type and fault level evaluation method based on a convolutional neural network. The method comprises the following steps: carrying out a partial discharge test; establishing a partial discharge information base; training a convolutional neural network, and outputting a defect type and a fault level corresponding to the defect type; and sending the to-be-detected partial discharge grey-scale map into the convolutional neural network to carry out evaluation to obtain the defect type and the fault level of the to-be-detected partial discharge grey-scale map. The partial discharge defect images are classified and diagnosed through the convolutional neural network, the traditional convolutional neural network is subjected to feature fusion, the discharge types and the discharge intensity of various partial discharges can be effectively identified, and a convenient and effective method is provided for the field of electrical equipment partial discharge mode identification.

Description

Technical Field

[0001] The present invention belongs to the technical field of power cable fault diagnosis, and in particular relates to a method for evaluating partial discharge defect types and fault levels of electrical equipment based on a convolutional neural network. Background Art

[0002] During the design, manufacture, transportation, installation, long-term operation and maintenance of electrical equipment, certain insulation defects often appear inside the equipment. These defects interact with the surrounding air to produce different types of partial discharge. Different types of partial discharge represent different insulation degradation mechanisms, and partial discharges can be strong or weak. The type and strength of the discharge have an absolute effect on the degree of damage to the insulation capacity of the equipment. Therefore, research on partial discharge diagnostic technology for electrical equipment is of great significance to the long-term stable operation of electrical equipment. Summary of the invention

[0003] In order to solve the technical problems in the prior art, the present invention provides a method for evaluating the type and fault level of partial discharge defects of electrical equipment based on a convolutional neural network, comprising the following steps:

[0004] Conduct partial discharge tests;

[0005] Establishment of partial discharge information database;

[0006] Train the convolutional neural network to output defect categories and fault levels corresponding to the defect types;

[0007] The grayscale image of the partial discharge to be detected is sent to the convolutional neural network for evaluation to obtain its defect type and fault level.

[0008] Furthermore, the partial discharge test includes:

[0009] Simulate typical defects of electrical equipment, collect and obtain PRPD spectra of different typical defects and different defect levels for multiple times.

[0010] Furthermore, the establishment of a partial discharge information database includes:

[0011] The collected PRPD spectra are preprocessed to establish a partial discharge information database, including several defect types and partial discharge grayscale images under different defect levels and the corresponding defect types and levels.

[0012] Furthermore, the convolutional neural network is trained and the output defect categories include:

[0013] The first convolutional neural network module is built and trained for defect type recognition, which is called the first module.

[0014] Furthermore, the convolutional neural network is trained to output the fault levels corresponding to the defect types, including:

[0015] A second submodule is constructed for each defect type, each submodule is formed based on a second convolutional neural network, and the second convolutional neural network is trained to output a fault level corresponding to the defect type.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] The present invention classifies and diagnoses partial discharge defect images through convolutional neural networks and performs feature fusion on traditional convolutional neural networks, which can effectively identify the discharge type and discharge intensity of various types of partial discharges, and provides a convenient and effective method for the field of partial discharge pattern recognition of electrical equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of the scheme of the present invention;

[0019] Figure 2 It is a flow chart of the first convolutional neural network model of the present invention;

[0020] Figure 3 This is the flow chart of the second convolutional neural network model of the present invention. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described below in conjunction with the accompanying drawings and specific implementation methods.

[0022] like Figure 1 As shown, the present application proposes a method for evaluating the type and fault level of partial discharge defects of electrical equipment based on a convolutional neural network, which specifically includes the following steps:

[0023] 1) Simulate typical defects of electrical equipment, collect and obtain PRPD spectra of different typical defects and different defect levels for multiple times;

[0024] 2) Preprocess the collected PRPD spectra to establish a partial discharge information database, including partial discharge grayscale images of several defect types and their different defect levels and the corresponding defect types and levels;

[0025] 3) Building and training a first convolutional neural network module for defect type recognition, which is referred to as the first module;

[0026] 4) Constructing a second submodule for each defect type, each submodule is formed based on a second convolutional neural network. Training the second convolutional neural network to output a fault level corresponding to the defect type;

[0027] 5) Send the grayscale image of the partial discharge to be detected to steps 3) and 4) for evaluation to obtain its defect type and fault level.

[0028] Simulate and create four typical insulation defect models of electrical equipment: tip discharge, surface discharge, internal discharge, and suspended discharge. Use partial discharge detection instruments and pulse current method to obtain partial discharge signals.

[0029] First, a partial discharge test platform for the model is built. The pulse current method is used to collect the discharge signal of the discharge model. The sampling frequency of the sampler is 100MHz. The partial discharge PRPD spectrum is obtained through the experiment. The PRPD spectrum here refers to the phase, discharge amplitude, and discharge number data, which are used to represent the phase distribution in each discharge amplitude interval. During each defect model test, three voltage measurement points are evenly taken in the starting voltage and the highest test voltage, and the data of 5 different levels of test voltages are recorded respectively. The data includes the starting voltage, the highest voltage, and the three middle test voltages under each discharge type. These 5 different levels of test voltages represent different severity of discharge. The number of samples for each discharge type at each test voltage is 100, and a total of 2000 groups of samples are obtained.

[0030] The collected PRPD spectra are preprocessed, the amplitude q of the partial discharge pulse signal is divided into A intervals, the power frequency phase φ is divided into B intervals, and the q-φ plane is divided into A×B intervals; the number of partial discharge pulses n in each interval is counted a,b (a∈A, b∈B), a three-dimensional q-φ-n spectrum is obtained; for the number of local discharge pulses n a,b Normalization is performed: Among them, n a,b ′ is the normalized pulse number, n a,b is the actual number of pulses, n max is the maximum number of pulses in all intervals of the q-φ-n spectrum; set n a,b ′The gray value of each pixel: K a,b =(1-n a,b ′)×255, where the maximum and minimum values ​​of the normalized pulse times correspond to the minimum grayscale level and the maximum grayscale level in the grayscale image, respectively, and then a grayscale image is obtained to characterize the partial discharge defect type and fault level.

[0031] The grayscale images of each defect and the labels of the corresponding discharge types are used as the input and output of the first convolutional neural network, respectively, to build a model for training discharge defect types. The first convolutional neural network model includes four convolutional layers, four pooling layers, and two fully connected layers, and the parameters of each layer are initialized. The process is as follows: Figure 2The ReLU activation function is used between the convolution layer and the pooling layer, and the cross entropy cost function is used to update the parameters of the model using the Adam method. Finally, the output layer uses the softmax function to output the probability that the sample belongs to each discharge type, and the discharge type with the highest probability is taken as the result of the input image. At the same time, the conclusion of the first convolutional neural network is used as the basis for determining the input channel of the second convolutional neural network.

[0032] The second convolutional neural network model has four input channels, which represent the classification of the fault levels of the four types of discharges. First, the input channel of the second convolutional neural network is selected according to the judgment results of the first convolutional neural network. The grayscale images of different test voltage levels of each discharge type are used as the input of the convolutional neural network in the input channel, and the output is the fault level. For example, if it is determined to be a tip discharge by the first convolutional neural network, the fault level of the tip discharge is determined by the model in the first channel of the second convolutional neural network. Since the grayscale images under different test voltage levels of the same discharge type are mainly due to the differences in the power frequency phase distribution, their local features should be highlighted. A convolutional neural network with multi-layer feature fusion is used here to screen its local features. Each channel of the second convolutional neural network model has three convolutional layers, three pooling layers and two fully connected layers. The output feature maps of the three pooling layers are input into the fully connected layer. The process is as follows Figure 3 ,After feature fusion, the Droput method is used to reduce the amount of calculation and suppress overfitting.,The second convolutional neural network module uses the cross entropy cost function, the SGD method is used to update the model parameters, the activation function uses ReLU, and finally the Softmax classifier.

[0033] The partial discharge grayscale image of the electrical equipment to be tested is sent to the model in the third step to classify its discharge type, and then sent to the model in the fourth step to evaluate its discharge level.

[0034] In summary, the above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited thereto. All equivalent changes and modifications made according to the patent scope of the present invention and the contents of the specification are within the scope covered by the patent of the present invention.

Claims

1. A method for evaluating the type and fault level of partial discharge defects in electrical equipment based on convolutional neural networks, characterized in that: The following steps are involved: Conduct partial discharge tests; Establishment of partial discharge information database; Train the convolutional neural network to output defect categories and fault levels corresponding to the defect types; The grayscale image of the partial discharge to be detected is sent to the convolutional neural network for evaluation to obtain its defect type and fault level.

2. The method according to claim 1, characterized in that Partial discharge testing includes: Simulate typical defects of electrical equipment, collect and obtain PRPD spectra of different typical defects and different defect levels for multiple times.

3. The method according to claim 2, characterized in that Establishing a partial discharge information database includes: The collected PRPD spectra are preprocessed to establish a partial discharge information database, including several defect types and partial discharge grayscale images under different defect levels and the corresponding defect types and levels.

4. The method according to claim 3, characterized in that Train a convolutional neural network to output defect categories including: The first convolutional neural network module is built and trained for defect type recognition, which is called the first module.

5. The method according to claim 4, characterized in that Train the convolutional neural network and output the fault levels corresponding to the defect types, including: A second submodule is constructed for each defect type, each submodule is formed based on a second convolutional neural network, and the second convolutional neural network is trained to output a fault level corresponding to the defect type.

Images