Insulation defect detection method and system based on improved bidirectional gating circulation network

Abstract

The invention discloses an insulation defect detection method and system based on an improved bidirectional gating circulation network. The method comprises the following steps: acquiring a direct current system branch current data sample; performing multi-dimensional data feature extraction and standardization processing on the DC system branch current data sample to obtain standardized DC system branch current feature parameter data; introducing an improved gating circulation network unit, and constructing an improved bidirectional gating circulation network model; and performing insulation defect detection processing on the standardized DC system branch current characteristic parameter data based on the improved bidirectional gating circulation network model to obtain a DC system branch insulation defect detection result. According to the invention, historical information and future information of current characteristic parameter data can be captured, so that the speed and precision of DC system branch insulation defect detection are improved. The insulation defect detection method and system based on the improved bidirectional gating circulation network can be widely applied to the technical field of insulation defect diagnosis.

Description

Technical Field

[0001] The invention relates to the technical field of insulation defect diagnosis, and in particular to an insulation defect detection method and system based on an improved bidirectional gated cyclic network. Background Art

[0002] With the rapid development of my country's economy and society, the requirements for power grid construction are constantly increasing, and the DC system has gradually become an important part of the power grid. However, with the widespread application of DC systems, the problem of insulation defect diagnosis has gradually become prominent. Insulation defects in DC systems not only threaten the safety and stability of the system, but may also cause serious equipment damage and safety accidents. Therefore, effective diagnosis of insulation defects in DC systems is of great significance. The existing insulation defect diagnosis methods mainly include insulation resistance test method and bridge method. The insulation resistance test method mainly evaluates the health of insulation by measuring the resistance value of the insulating material. This method is simple and easy to operate, but it has low sensitivity to initial defects and often needs to be used in combination with other methods. The bridge method uses the bridge method to measure the insulation resistance of each part of the system to help identify and locate insulation defects. This method is more effective in detecting the insulation status of large equipment or systems, but the operation is complicated and requires high skills of the operator. Although the existing insulation defect diagnosis methods provide technical support for ensuring system safety, they still face many challenges in practical applications, such as difficulty in defect identification, strong limitations of detection technology, and significant impact of complex environment. Summary of the invention

[0003] In order to solve the above technical problems, the purpose of the present invention is to provide an insulation defect detection method and system based on an improved bidirectional gated cyclic network, which can capture the historical information and future information of the DC system branch current characteristic parameter data, thereby improving the speed and accuracy of DC system branch insulation defect detection.

[0004] The first technical solution adopted by the present invention is: an insulation defect detection method based on an improved bidirectional gated cyclic network, comprising the following steps:

[0005] Acquire a DC system branch current data sample, wherein the DC system branch current data sample includes a DC system branch current in normal operation and a DC system branch current with insulation defects;

[0006] Perform multi-dimensional data feature extraction and standardization processing on DC system branch current data samples to obtain standardized DC system branch current characteristic parameter data;

[0007] Introduce an improved gated recurrent network unit and construct an improved bidirectional gated recurrent network model;

[0008] Based on the improved bidirectional gated cyclic network model, the insulation defect detection processing is performed on the standardized DC system branch current characteristic parameter data to obtain the DC system branch insulation defect detection result.

[0009] Furthermore, the step of extracting multi-dimensional data features and standardizing the DC system branch current data samples to obtain standardized DC system branch current characteristic parameter data specifically includes:

[0010] Performing time domain feature extraction processing on the DC system branch current data samples to obtain the time domain features of the DC system branch current data samples;

[0011] Performing Fourier transformation and frequency domain feature extraction processing on the DC system branch current data samples to obtain the frequency domain features of the DC system branch current data samples;

[0012] The autoencoder based on the long short-term memory network performs deep feature extraction processing on the DC system branch current data samples to obtain the deep features of the DC system branch current data samples;

[0013] Integrate the time domain characteristics of the DC system branch current data samples, the frequency domain characteristics of the DC system branch current data samples and the depth characteristics of the DC system branch current data samples to obtain DC system branch current characteristic parameter data;

[0014] The DC system branch current characteristic parameter data is standardized to obtain standardized DC system branch current characteristic parameter data.

[0015] Furthermore, the step of performing deep feature extraction processing on the DC system branch current data samples by the long short-term memory network-based autoencoder to obtain deep features of the DC system branch current data samples specifically includes:

[0016] Based on the long short-term memory network, an autoencoder is constructed, wherein the autoencoder includes an input module, a hidden module and an output module;

[0017] Adding Gaussian noise and dividing the DC system branch current data samples to obtain the DC system branch current input data vector;

[0018] Inputting the DC system branch current input data vector into the autoencoder;

[0019] Based on the input module of the autoencoder, the DC system branch current input data vector is preprocessed in time step to construct a preliminary DC system branch current output data vector;

[0020] Based on the hidden module of the autoencoder, the preliminary DC system branch current output data vector is encoded and decoded to obtain the DC system branch current output data vector;

[0021] Based on the output module of the autoencoder, the DC system branch current output data vector is reconstructed in time steps to obtain the deep features of the DC system branch current data samples.

[0022] Furthermore, it also includes training the autoencoder based on the long short-term memory network by the gradient descent method, and the cost function of the autoencoder is specifically as follows:

[0023]

[0024] In the above formula, L represents the cost function of the autoencoder, N represents the number of samples, μ represents the mean of the latent space variable, σ represents the standard deviation of the latent space variable, i represents the i-th data sample, and x i Represents the input data of the autoencoder, y i Represents the output data of the autoencoder.

[0025] Furthermore, the improved bidirectional gated recurrent network model includes an input layer, a forward gated recurrent network layer, a backward gated recurrent network layer, a fully connected layer and an output layer, wherein the input layer, the forward gated recurrent network layer, the backward gated recurrent network layer, the fully connected layer and the output layer are connected in sequence, wherein the basic unit of the forward gated recurrent network layer and the basic unit of the backward gated recurrent network layer are both improved gated recurrent network units, the improved gated recurrent network unit includes a reset gate and an update gate, the internal activation function of the improved gated recurrent network unit adopts a softsign function, and the threshold function of the improved gated recurrent network unit adopts a sigmoid function.

[0026] Furthermore, the expression of the improved bidirectional gated recurrent network model is specifically as follows:

[0027] r t =σ(W r [h t-1 ,x t ]+b r )

[0028] z t =σ(W z [h t-1 ,x t ]+b z )

[0029]

[0030] In the above formula, rt 、z t , h t Represent the reset gate, update gate, candidate hidden state and current hidden state respectively, W r , W z and W h Represent the reset gate, update gate and the weight of the candidate hidden layer at the current moment, b r , b z and b h They represent the reset gate, update gate and the bias of the candidate hidden layer at the current moment, respectively, and σ(·) represents the sigmoid function.

[0031] Further, the step of performing insulation defect detection processing on the standardized DC system branch current characteristic parameter data based on the improved bidirectional gated cyclic network model to obtain the DC system branch insulation defect detection result specifically includes:

[0032] The standardized DC system branch current characteristic parameter data is input into the improved bidirectional gated recurrent network model;

[0033] Based on the input layer of the improved bidirectional gated recurrent network model, the standardized DC system branch current characteristic parameter data is acquired to obtain the acquired DC system branch current characteristic parameter data;

[0034] Based on the forward gated recurrent network layer of the improved bidirectional gated recurrent network model, sequential information capture is performed on the acquired DC system branch current characteristic parameter data to obtain the DC system branch current characteristic parameter data history information;

[0035] Based on the backward gated recurrent network layer of the improved bidirectional gated recurrent network model, the acquired DC system branch current characteristic parameter data is captured in reverse order to obtain the future information of the DC system branch current characteristic parameter data;

[0036] Integrate the historical information of the DC system branch current characteristic parameter data with the future information of the DC system branch current characteristic parameter data to obtain bidirectional information of the DC system branch current characteristic parameter data;

[0037] Based on the full-link layer of the improved bidirectional gated cyclic network model, the bidirectional information of the DC system branch current characteristic parameter data is mapped to obtain the bidirectional information of the DC system branch current characteristic parameter data after mapping;

[0038] Based on the output layer of the improved bidirectional gated cyclic network model, the bidirectional information of the mapped DC system branch current characteristic parameter data is output to obtain the DC system branch insulation defect detection result.

[0039] The second technical solution adopted by the present invention is: an insulation defect detection system based on an improved bidirectional gated cyclic network, comprising:

[0040] The first module is used to obtain a DC system branch current data sample, wherein the DC system branch current data sample includes a DC system branch current in normal operation and a DC system branch current with insulation defects;

[0041] The second module is used to extract multi-dimensional data features and perform standardization processing on the DC system branch current data samples to obtain standardized DC system branch current characteristic parameter data;

[0042] The third module is used to introduce an improved gated recurrent network unit and construct an improved bidirectional gated recurrent network model;

[0043] The fourth module is used to perform insulation defect detection processing on the standardized DC system branch current characteristic parameter data based on the improved bidirectional gated cyclic network model to obtain the DC system branch insulation defect detection result.

[0044] The beneficial effects of the method and system of the present invention are as follows: the present invention can accurately obtain the defect feature information of the DC system branch by acquiring the DC system branch current data samples and then performing multi-dimensional data feature extraction and standardization processing on the DC system branch current data samples, further introduce an improved gated cyclic network unit, construct an improved bidirectional gated cyclic network model, and perform insulation defect detection processing on the standardized DC system branch current characteristic parameter data based on the improved bidirectional gated cyclic network model, so as to capture all historical information of the DC system branch current characteristic parameter data from the beginning to the current time step and all future information of the DC system branch current characteristic parameter data from the current time step to the end of the sequence, thereby improving the speed and accuracy of the DC system branch insulation defect detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flowchart of the steps of the insulation defect detection method based on the improved bidirectional gated cyclic network of the present invention;

[0046] Figure 2 It is a structural block diagram of an insulation defect detection system based on an improved bidirectional gated cyclic network of the present invention;

[0047] Figure 3 is a flowchart of the steps of insulation defect detection provided by a specific embodiment of the present invention;

[0048] Figure 4 is a schematic diagram of a long short-term memory network neuron structure provided by a specific embodiment of the present invention;

[0049] Figure 5is a schematic diagram of a GRU neuron structure provided by a specific embodiment of the present invention;

[0050] Figure 6 is a schematic diagram of a bidirectional gated recurrent network structure provided by a specific embodiment of the present invention;

[0051] Figure 7 It is a schematic diagram of a bidirectional gated recurrent network model provided by a specific embodiment of the present invention. DETAILED DESCRIPTION

[0052] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The step numbers in the following embodiments are only provided for the convenience of explanation and description, and the order between the steps is not limited in any way. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0053] Reference Figure 1 The present invention provides an insulation defect detection method based on an improved bidirectional gated cyclic network, the method comprising the following steps:

[0054] S100, obtaining a DC system branch current data sample, wherein the DC system branch current data sample includes a DC system branch current in normal operation and a DC system branch current with insulation defects;

[0055] Specifically, the actual DC distribution system is usually an ungrounded system with small leakage current and difficult to detect. Therefore, when an insulation defect occurs, the current waveform only changes slightly. The branch current signals in the DC system are collected and preprocessed, including multiple sets of current signal samples when the system is operating normally and when different degrees of insulation defects are simulated. Different degrees of insulation defects include: 20KΩ high-resistance grounding, 40KΩ high-resistance grounding, 60KΩ high-resistance grounding, 80KΩ high-resistance grounding, and 100KΩ high-resistance grounding, etc. The collected data set is recorded as D = {x1, x2, ..., x i ...,x n}, where x i is the i-th sample, and n is the number of samples.

[0056] S200, performing multi-dimensional data feature extraction and standardization processing on the DC system branch current data sample to obtain standardized DC system branch current characteristic parameter data;

[0057] Specifically, the data samples are subjected to feature extraction in the time and frequency domains, and deep features are extracted using an autoencoder based on a long short-term memory network.

[0058] S210, performing time domain feature extraction processing on the DC system branch current data samples to obtain time domain features of the DC system branch current data samples;

[0059] Specifically, the time domain features extracted from signal samples include mean, absolute mean, variance, standard deviation, maximum, minimum, peak-to-peak value, root mean square, skewness, kurtosis, crest factor, waveform factor, margin coefficient and energy.

[0060] S220, performing Fourier transform and frequency domain feature extraction processing on the DC system branch current data samples to obtain frequency domain features of the DC system branch current data samples;

[0061] Specifically, the sample data is subjected to Fourier transform, and the time domain signal is converted to the frequency domain and then the frequency domain features are extracted. The extracted frequency domain features include: harmonic amplitude, amplitude average frequency, center frequency, root mean square frequency, standard deviation frequency, maximum frequency and minimum frequency, etc., which represent the degree of spectrum dispersion or concentration.

[0062] S230, performing deep feature extraction processing on the DC system branch current data samples based on the long short-term memory network autoencoder to obtain deep features of the DC system branch current data samples;

[0063] Specifically, based on a long short-term memory network, an autoencoder is constructed, and the autoencoder includes an input module, a hidden module and an output module; Gaussian noise is added to and a division process is performed on a DC system branch current data sample to obtain a DC system branch current input data vector; the DC system branch current input data vector is input into the autoencoder; based on the input module of the autoencoder, the DC system branch current input data vector is preprocessed in time step to construct a preliminary DC system branch current output data vector; based on the hidden module of the autoencoder, the preliminary DC system branch current output data vector is encoded and decoded to obtain a DC system branch current output data vector; based on the output module of the autoencoder, the DC system branch current output data vector is reconstructed in time step to obtain the deep features of the DC system branch current data sample.

[0064] In this embodiment, an autoencoder based on a long short-term memory network is constructed to extract deep features. The autoencoder includes an input module, a hidden module, and an output module. The input layer and the output layer are both composed of a long short-term memory network.

[0065] First, in the original sample x i Add Gaussian noise ξ to obtain the current signal sample after adding noise The noised samples Divide into T segments of the same length to form the input data vector

[0066] Among them, Figure 4As shown in the figure, the unit of the long short-term memory network is composed of a forget gate, an input gate, and an output gate. The state update formulas in the long short-term memory network are as follows:

[0067] f t =σ(W f ×(h t-1 ,x t )+b f )

[0068] i t =σ(W i ×(h t-1 ,x t )+b i )

[0069]

[0070] O t =σ(W o ×(h t-1 ,x t )+b o )

[0071] h t =O t ×tanh(C t )

[0072] In the above formula, f t 、i t , O t Respectively represent the values ​​of the forget gate, input gate and output gate at time t, and C t Respectively represent the intermediate quantity and state quantity of the memory unit at time t, W f , W i , W c , W o Respectively represent the weight matrices corresponding to different state units, b f , b i , b c and b o They represent the bias matrices corresponding to different state units, h t-1 and h t Respectively represent the state of the hidden layer at time t-1 and t, x t represents the input data at time t, σ represents the sigmoid activation function, and tanh(·) is the hyperbolic tangent activation function.

[0073] Will Input into the input module, pass through the hidden layer of the long short-term memory network, and output the feature vector z at time step T i ={h i1 ,h i2,......,h iT}.

[0074] The hidden module consists of m fully connected layers, each with n nodes. The hidden module takes the feature vector z output by the input module as input. i Encode and get the encoding result K i '={k i1 ',k i2 ',...,k ii '...,k iN '}, k ii ' is the i-th feature obtained by the i-th sample in the hidden module, and N is the number of features. Then the encoding result K i 'Decode and obtain the i-th output feature vector z of the hidden module' i ={h i1 ',h i2 ',......,h iT '}.

[0075] Finally, the eigenvector z' i Input to the output module, after passing through the hidden layer of the long short-term memory network, the feature vector is output at time step T, and the i-th current signal sample y is reconstructed i ={y i1 ,y i2 ,......,y iT}.

[0076] S240, integrating the time domain characteristics of the DC system branch current data samples, the frequency domain characteristics of the DC system branch current data samples, and the depth characteristics of the DC system branch current data samples to obtain DC system branch current characteristic parameter data;

[0077] S250, performing standardization processing on the DC system branch current characteristic parameter data to obtain standardized DC system branch current characteristic parameter data.

[0078] Specifically, in order to reconstruct the sample y i Can reconstruct the original sample x as much as possible i , through iteration, the cost function is minimized. The encoding result of the hidden module corresponding to the minimum cost function is taken as the deep feature, denoted as K i ={k i1 ,k i2 ,...,k ii ...,k iN}, k iiis the i-th deep feature of the i-th sample, and N is the number of features. Therefore, by extracting feature parameters, the current signal data set is converted into a feature data set, which is recorded as E = {time domain feature, frequency domain feature, depth feature}.

[0079] Since there are dimensional differences between different feature parameters, each feature parameter in the feature data set is normalized to the interval [0, 1]. The normalization formula of the feature parameter is as follows.

[0080]

[0081] In the above formula, e represents a parameter value in the same characteristic parameter, e min Indicates the minimum value of the characteristic parameter, e max represents the maximum value of the feature parameter, and e' represents the normalized feature parameter value, thereby obtaining the normalized feature data set E'={time domain feature', frequency domain feature', depth feature'}.

[0082] Furthermore, it also includes training the autoencoder based on the long short-term memory network by the gradient descent method, and the cost function of the autoencoder is specifically as follows:

[0083]

[0084] In the above formula, L represents the cost function of the autoencoder, N represents the number of samples, μ represents the mean of the latent space variable, σ represents the standard deviation of the latent space variable, i represents the i-th data sample, and x i Represents the input data of the autoencoder, y i Represents the output data of the autoencoder.

[0085] S300, introducing an improved gated recurrent network unit and constructing an improved bidirectional gated recurrent network model;

[0086] Specifically, the improved bidirectional gated recurrent network model includes an input layer, a forward gated recurrent network layer, a backward gated recurrent network layer, a fully connected layer and an output layer, wherein the input layer, the forward gated recurrent network layer, the backward gated recurrent network layer, the fully connected layer and the output layer are connected in sequence, wherein the basic unit of the forward gated recurrent network layer and the basic unit of the backward gated recurrent network layer are both improved gated recurrent network units, the improved gated recurrent network unit includes a reset gate and an update gate, the internal activation function of the improved gated recurrent network unit adopts a softsign function, and the threshold function of the improved gated recurrent network unit adopts a sigmoid function.

[0087] In this embodiment, the bidirectional gated recurrent network (BiGRU) is a variant of the recurrent neural network (RNN), which combines the characteristics of the gated recurrent network (GRU) and the bidirectional RNN. The bidirectional structure can capture more comprehensive information in the sequence and is suitable for processing time series data. The improved bidirectional gated recurrent network consists of a forward gated recurrent network and a backward gated recurrent network. Its basic unit is an improved GRU unit, including a reset gate and an update gate. The internal activation function uses softsign, and the threshold function uses a sigmoid function. The improved GRU structure is as follows: Figure 5 shown.

[0088] The state update formulas in the improved GRU structure are as follows:

[0089] r t =σ(W r [h t-1 ,x t ]+b r )

[0090] z t =σ(W z [h t-1 ,x t ]+b z )

[0091]

[0092] In the above formula, r t 、z t , h t Represent the reset gate, update gate, candidate hidden state and current hidden state respectively, W r , W z and W h Represent the reset gate, update gate and the weight of the candidate hidden layer at the current moment, b r 、b z and b h They represent the reset gate, update gate and the bias of the candidate hidden layer at the current moment, respectively, and σ(·) represents the sigmoid function.

[0093] The reset gate determines how new input information is combined with historical information and how much historical information is retained. The update gate controls the state of the previous moment and the degree of influence of the new input on the current state. The internal activation function uses the softsign function. Compared with the tanh function, the internal activation function of the existing standard gated recurrent neural network, the softsign function is less likely to saturate. The softsign function has a flatter curve and a slower descent derivative. These characteristics indicate that it has a more efficient learning ability.

[0094] S400, performing insulation defect detection processing on the standardized DC system branch current characteristic parameter data based on the improved bidirectional gated cyclic network model to obtain a DC system branch insulation defect detection result.

[0095] Specifically, the standardized DC system branch current characteristic parameter data is input into the improved bidirectional gated cyclic network model; based on the input layer of the improved bidirectional gated cyclic network model, the standardized DC system branch current characteristic parameter data is acquired to obtain the acquired DC system branch current characteristic parameter data; based on the forward gated cyclic network layer of the improved bidirectional gated cyclic network model, the acquired DC system branch current characteristic parameter data is sequentially captured to obtain the historical information of the DC system branch current characteristic parameter data; based on the backward gated cyclic network layer of the improved bidirectional gated cyclic network model, the acquired DC system branch current characteristic parameter data is sequentially captured to obtain the historical information of the DC system branch current characteristic parameter data. Reverse information is captured to obtain future information of DC system branch current characteristic parameter data; historical information of DC system branch current characteristic parameter data and future information of DC system branch current characteristic parameter data are integrated to obtain bidirectional information of DC system branch current characteristic parameter data; based on the full-link layer of the improved bidirectional gated cyclic network model, the bidirectional information of DC system branch current characteristic parameter data is mapped to obtain the bidirectional information of DC system branch current characteristic parameter data after mapping; based on the output layer of the improved bidirectional gated cyclic network model, the bidirectional information of DC system branch current characteristic parameter data after mapping is output to obtain the DC system branch insulation defect detection result.

[0096] In this embodiment, the network structure of BiGRU is as follows: Figure 6 As shown in the figure, the forward GRU unit processes from the first element to the last element in the input data. It receives the current input and the hidden state of the previous time step at each time step, so that it can capture all historical information from the beginning to the current time step. The backward GRU unit processes from the last element to the first element in the input data. It receives the current input and the hidden state of the subsequent time step at each time step, so that it can capture all future information from the current time step to the end of the sequence. This bidirectional structure can simultaneously utilize past and future information in the input data, thereby providing a more comprehensive internal connection of the input data.

[0097] The output of BiGRU is determined by the states of the forward GRU and the backward GRU. The hidden state of the forward GRU at time t-1 is and the backward GRU hidden state The weighted summation is used to obtain the hidden layer state of BiGRU at time t, and its calculation expression is as follows:

[0098]

[0099] The improved BiGRU model consists of an input layer, two BiGRU network layers, a fully connected layer, and an output layer. Figure 7 As shown in the figure. The input layer is used to receive and organize the input feature data set. The two-layer BiGRU network layer is used to capture the bidirectional information of the input data, deeply understand the complex patterns and dependencies between the data, and generate deep features. The full connection layer is used to integrate the features generated by the BiGRU layer and map them to the classification category space. The output layer uses the activation function to generate the final category probability and output the prediction result. The insulation defect feature vector matrix is ​​used as the input data, and the classification and recognition are performed after the BiGRU network learning. Finally, the output layer outputs the model recognition result.

[0100] Add data labels to the extracted feature data set. Then divide the data set into training set and test set in a ratio of 7:3. Input the training set data samples into the improved BiGRU model for training, and continuously optimize the model parameters such as the number of training iterations, training batch size, and learning rate according to the test results of the test set. Finally, an improved BiGRU model with better performance can be obtained.

[0101] After obtaining the trained improved BiGRU model, the complete process of DC system insulation defect diagnosis is as follows: Figure 3 First, the current data of insulation defects in the DC system is collected, and the time domain feature parameters, frequency domain feature parameters and depth feature parameters of the collected sample signals are extracted. The extracted feature parameters are normalized and finally input into the trained improved BiGRU model for diagnosis, and finally the diagnosis results are output.

[0102] Reference Figure 2 , an insulation defect detection system based on an improved bidirectional gated recurrent network, comprising:

[0103] The first module 201 is used to obtain a DC system branch current data sample, wherein the DC system branch current data sample includes a DC system branch current in normal operation and a DC system branch current with insulation defects;

[0104] The second module 202 is used to perform multi-dimensional data feature extraction and standardization processing on the DC system branch current data samples to obtain standardized DC system branch current characteristic parameter data;

[0105] The third module 203 is used to introduce an improved gated recurrent network unit and construct an improved bidirectional gated recurrent network model;

[0106] The fourth module 204 is used to perform insulation defect detection processing on the standardized DC system branch current characteristic parameter data based on the improved bidirectional gated cyclic network model to obtain the DC system branch insulation defect detection result.

[0107] The contents of the above method embodiments are all applicable to the present system embodiments. The functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0108] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. An insulation defect detection method based on an improved bidirectional gated cyclic network is characterized in that: The following steps are involved: Acquire a DC system branch current data sample, wherein the DC system branch current data sample includes a DC system branch current in normal operation and a DC system branch current with insulation defects; Perform multi-dimensional data feature extraction and standardization processing on DC system branch current data samples to obtain standardized DC system branch current characteristic parameter data; Introduce an improved gated recurrent network unit and construct an improved bidirectional gated recurrent network model; Based on the improved bidirectional gated cyclic network model, the insulation defect detection processing is performed on the standardized DC system branch current characteristic parameter data to obtain the DC system branch insulation defect detection result.

2. The insulation defect detection method based on the improved bidirectional gated cyclic network according to claim 1 is characterized in that: The step of extracting multi-dimensional data features and standardizing the DC system branch current data samples to obtain standardized DC system branch current characteristic parameter data specifically includes: Performing time domain feature extraction processing on the DC system branch current data samples to obtain the time domain features of the DC system branch current data samples; Performing Fourier transformation and frequency domain feature extraction processing on the DC system branch current data samples to obtain the frequency domain features of the DC system branch current data samples; The autoencoder based on the long short-term memory network performs deep feature extraction processing on the DC system branch current data samples to obtain the deep features of the DC system branch current data samples; Integrate the time domain characteristics of the DC system branch current data samples, the frequency domain characteristics of the DC system branch current data samples and the depth characteristics of the DC system branch current data samples to obtain DC system branch current characteristic parameter data; The DC system branch current characteristic parameter data is standardized to obtain standardized DC system branch current characteristic parameter data.

3. The insulation defect detection method based on the improved bidirectional gated cyclic network according to claim 2 is characterized in that: The step of performing deep feature extraction processing on the DC system branch current data samples by the long short-term memory network-based autoencoder to obtain the deep features of the DC system branch current data samples specifically includes: Based on the long short-term memory network, an autoencoder is constructed, wherein the autoencoder includes an input module, a hidden module and an output module; Adding Gaussian noise and dividing the DC system branch current data samples to obtain the DC system branch current input data vector; Inputting the DC system branch current input data vector into the autoencoder; Based on the input module of the autoencoder, the DC system branch current input data vector is preprocessed in time step to construct a preliminary DC system branch current output data vector; Based on the hidden module of the autoencoder, the preliminary DC system branch current output data vector is encoded and decoded to obtain the DC system branch current output data vector; Based on the output module of the autoencoder, the DC system branch current output data vector is reconstructed in time steps to obtain the deep features of the DC system branch current data samples.

4. The insulation defect detection method based on the improved bidirectional gated cyclic network according to claim 3 is characterized in that: It also includes training an autoencoder based on a long short-term memory network by a gradient descent method, and the cost function of the autoencoder is specifically as follows: In the above formula, L represents the cost function of the autoencoder, N represents the number of samples, μ represents the mean of the latent space variable, σ represents the standard deviation of the latent space variable, i represents the i-th data sample, and x i Represents the input data of the autoencoder, y i Represents the output data of the autoencoder.

5. The insulation defect detection method based on the improved bidirectional gated cyclic network according to claim 4 is characterized in that: The improved bidirectional gated recurrent network model includes an input layer, a forward gated recurrent network layer, a backward gated recurrent network layer, a fully connected layer and an output layer. The input layer, the forward gated recurrent network layer, the backward gated recurrent network layer, the fully connected layer and the output layer are connected in sequence, wherein the basic unit of the forward gated recurrent network layer and the basic unit of the backward gated recurrent network layer are both improved gated recurrent network units, the improved gated recurrent network unit includes a reset gate and an update gate, the internal activation function of the improved gated recurrent network unit adopts a softsign function, and the threshold function of the improved gated recurrent network unit adopts a sigmoid function.

6. The insulation defect detection method based on the improved bidirectional gated cyclic network according to claim 5 is characterized in that: The expression of the improved bidirectional gated recurrent network model is specifically as follows: r t =σ(W r [h t-1 ,x t ]+b r ) z t =σ(W z [h t-1 ,x t ]+b z ) In the above formula, r t 、z t , h t Represent the reset gate, update gate, candidate hidden state and current hidden state respectively, W r , W z and W h Represent the reset gate, update gate and the weight of the candidate hidden layer at the current moment, b r 、b z and b h They represent the reset gate, update gate and the bias of the candidate hidden layer at the current moment, respectively, and σ(·) represents the sigmoid function.

7. The insulation defect detection method based on the improved bidirectional gated cyclic network according to claim 6 is characterized in that: The step of performing insulation defect detection processing on the standardized DC system branch current characteristic parameter data based on the improved bidirectional gated cyclic network model to obtain the DC system branch insulation defect detection result specifically includes: The standardized DC system branch current characteristic parameter data is input into the improved bidirectional gated recurrent network model; Based on the input layer of the improved bidirectional gated recurrent network model, the standardized DC system branch current characteristic parameter data is acquired to obtain the acquired DC system branch current characteristic parameter data; Based on the forward gated recurrent network layer of the improved bidirectional gated recurrent network model, sequential information capture is performed on the acquired DC system branch current characteristic parameter data to obtain the DC system branch current characteristic parameter data history information; Based on the backward gated recurrent network layer of the improved bidirectional gated recurrent network model, the acquired DC system branch current characteristic parameter data is captured in reverse order to obtain the future information of the DC system branch current characteristic parameter data; Integrate the historical information of the DC system branch current characteristic parameter data with the future information of the DC system branch current characteristic parameter data to obtain bidirectional information of the DC system branch current characteristic parameter data; Based on the full-link layer of the improved bidirectional gated cyclic network model, the bidirectional information of the DC system branch current characteristic parameter data is mapped to obtain the bidirectional information of the DC system branch current characteristic parameter data after mapping; Based on the output layer of the improved bidirectional gated cyclic network model, the bidirectional information of the mapped DC system branch current characteristic parameter data is output to obtain the DC system branch insulation defect detection result.

8. The insulation defect detection system based on the improved bidirectional gated cyclic network is characterized in that: Includes the following modules: The first module is used to obtain a DC system branch current data sample, wherein the DC system branch current data sample includes a DC system branch current in normal operation and a DC system branch current with insulation defects; The second module is used to extract multi-dimensional data features and perform standardization processing on the DC system branch current data samples to obtain standardized DC system branch current characteristic parameter data; The third module is used to introduce an improved gated recurrent network unit and construct an improved bidirectional gated recurrent network model; The fourth module is used to perform insulation defect detection processing on the standardized DC system branch current characteristic parameter data based on the improved bidirectional gated cyclic network model to obtain the DC system branch insulation defect detection result.

Images