Fault location method and system based on decomposition noise reduction and novel convolutional neural network

Abstract

The invention discloses a fault location method and system based on decomposition noise reduction and a novel convolutional neural network, and belongs to the technical field of power systems, and the method comprises the steps: collecting a three-phase current signal of a line as an original current signal; carrying out decomposition and noise reduction processing on the collected original current signal; the entropy value of each IMF component is calculated, the IMF components with the entropy values larger than a threshold value are screened out, the other IMF components are reconstructed, and a denoised signal is obtained; converting the denoised signal into a two-dimensional feature image by using a Grubrum angle field algorithm, and performing data enhancement processing on the obtained two-dimensional feature image; and inputting the generated two-dimensional feature image data set into the novel convolutional neural network model to obtain a distance measurement result. The system comprises the computer readable storage medium and the processor, and the processor is used for reading the executable instructions stored in the computer readable storage medium to execute the steps of the method. The noise reduction method and the novel convolutional neural network are combined to improve the accuracy and reliability of fault distance measurement.

Description

Technical Field

[0001] The present invention belongs to the technical field of power systems, and in particular, relates to a fault distance measurement method and system based on decomposition noise reduction and a novel convolutional neural network. Background Art

[0002] As people's living standards continue to improve, their reliance on electricity in daily life is increasing, which puts higher requirements on the safety and reliability of distribution networks. One of the measures to ensure the safe and stable operation of distribution networks is to be able to quickly locate faults and quickly restore power supply to non-fault areas.

[0003] The existing distribution network fault location methods mainly include the following methods:

[0004] 1. Using the time difference characteristics of traveling waves to measure fault distance, this method is intuitive in principle but is susceptible to external interference and has high requirements on equipment.

[0005] 2. Use the hybrid measurement state fault method to measure the fault distance of the distribution network. This method combines the data collected by measurement equipment such as smart meters in the power grid and has high reliability. However, it requires the installation of multiple collection devices in the application, which increases the cost.

[0006] 3. Using the BP neural network algorithm, the collected electrical quantities are input into the BP neural network, which can effectively process the data. However, due to the lack of preprocessing of the original data, the final distance measurement result may be disturbed.

[0007] Most of the above methods cannot guarantee the accuracy of fault location, and the above methods do not take anti-interference measures, making the final result susceptible to interference. Therefore, it is necessary to study the fault location algorithm based on the above problems. Summary of the invention

[0008] In view of the shortcomings of the above-mentioned background technology, the purpose of the present invention is to propose a fault ranging method and system based on decomposition denoising and a new convolutional neural network, which is used to solve the problems of low accuracy, poor anti-interference ability and difficult reliability of existing fault ranging methods.

[0009] The technical solution adopted by the present invention to achieve the above-mentioned purpose is:

[0010] According to the first aspect of the present invention, a fault distance measurement method based on decomposition noise reduction and a novel convolutional neural network is provided. The present invention uses signal decomposition and reconstruction to perform noise reduction processing on the original signal, and introduces a noise reduction method to improve the anti-interference ability of the system. The collected data is converted into an image using the Gram angle field, the time series signal is converted into a two-dimensional image signal, and the signal is data enhanced. A novel convolutional neural network is introduced, the collected signal is input into the neural network for feature learning, and finally the fault distance is output.

[0011] The method specifically comprises the following steps:

[0012] Step 1: Collect the three-phase current signal of the line as the original current signal according to the set sampling frequency;

[0013] If the sampling frequency is too high, the data size will increase and the calculation efficiency will be reduced. If the sampling frequency is too low, the features will be lost, affecting the ranging accuracy. The sampling frequency adopted by the present invention in the experimental distribution network model is 5kHz.

[0014] Step 2: Use the ICEEMDAN algorithm to decompose and reduce noise on the collected original current signal;

[0015] The specific process is as follows:

[0016] First, take the original current signal x[n] and add Gaussian white noise to construct a total of I groups of noise-added signals;

[0017] x (i) =x+β 0 E 1 (w (i) ) (1)

[0018] In the formula, x (i) is the i-th current signal after adding Gaussian white noise, x is the original current signal, w (i) represents the added i-th group of white noise, where i=1,2,…,I, let w (i) is Gauss white noise with mean 0 and unit variance; operator E 1 (·) represents the first-order IMF component of the EMD decomposition of a signal, β 0 The weight coefficient when adding noise components represents the ratio of the signal-to-noise ratio of the added noise to the standard deviation of the noise component;

[0019] The local operator M(·) is introduced, which represents the local mean of the signal. The calculation can be obtained for each x (i) The corresponding first-order IMF component

[0020]

[0021] Summing the above components, we can get the first-order IMF component D of the original current signal x: 1 ;

[0022]

[0023] Get the first-order residual r 1 :

[0024] r 1 =xD 1 (4)

[0025] The second-order residual r 2 :

[0026]

[0027] Among them, β 1 Refers to the weight coefficient of adding noise to the second-order residual; operator E 2 (·) refers to the second-order IMF component of the signal EMD decomposition;

[0028] The second-order IMF component D can be obtained 2 for:

[0029]

[0030] Repeat the above steps to get the k-th order residual r k , k = 3, 4, ..., K, K is the total decomposition order;

[0031]

[0032] Among them, β k-1 Refers to the weight coefficient of adding noise to the k-1th order residual; operator E k (·) refers to the k-th order IMF component of the signal EMD decomposition;

[0033] The Kth order IMF component D k for:

[0034] D k =r k-1 -r k (8)

[0035] Repeat the above steps to obtain the IMF components decomposed by ICEEMDAN.

[0036] Step 3: Calculate the entropy value of each IMF component, and filter out the IMF components with entropy values ​​greater than the threshold, and reconstruct the remaining IMF components to obtain the denoised signal;

[0037] The present invention adopts linear reconstruction, and the threshold value varies according to different experimental models and conditions. The threshold value adopted in the experiment of the present invention is 0.9.

[0038] Step 4: Use the Gram angular field algorithm to convert the denoised signal into a two-dimensional feature image, and perform data enhancement on the obtained two-dimensional feature image to generate a two-dimensional feature image dataset;

[0039] The data enhancement includes vertical flipping, horizontal flipping, graying, adjusting brightness and adding noise operations.

[0040] Step 5: Input the generated two-dimensional feature image data set into the new convolutional neural network model to obtain the distance measurement result, thus completing the fault distance measurement;

[0041] The novel convolutional neural network model includes:

[0042] Basic module: ConvNeXt Block includes the following parts:

[0043] The core building block of the ConvNeXt network is a ResNet-like structural module, but with many modern tweaks:

[0044] 1) Depthwise Convolution: Perform a depthwise convolution operation to perform convolution calculations on each channel in the input feature map independently. This operation can greatly reduce the amount of calculation and the number of parameters. In this layer, the input feature map is divided into multiple channels, each channel is convolved separately, and the output has the same number of channels as the input.

[0045] 2) Point convolution layer (1x1 Convolution), which is executed immediately after the depth convolution, is responsible for linearly combining the information of different channels. This is similar to the point convolution in ResNet. In this layer, after the feature map of the depth convolution, all channels are linearly combined to generate new channels.

[0046] 3) Normalization layer: Normalize each layer of the feature map to stabilize the training process and improve model performance. Each channel of the feature map in this layer is standardized so that the output mean is 0 and the variance is 1.

[0047] 4) GELU activation function is a nonlinear activation function that introduces nonlinearity into the features after linear combination to increase the expressive power of the model. After being processed by the activation function, the output features retain the nonlinear information of the original features.

[0048] 5) DropPath: randomly discard some paths during training to increase the generalization ability of the model, which is similar to dropout but more suitable for deep networks.

[0049] The overall structure of ConvNeXt is as follows:

[0050] ConvNeXt is composed of multiple ConvNeXt Blocks and adopts a hierarchical structure similar to ResNet, that is, it contains multiple stages, each with a different resolution and a gradually increasing number of channels.

[0051] 1) Initial Convolutional Layer (Stem Layer)

[0052] The original image is initially input, and after passing through this layer, the size is reduced and the channels are expanded. The input image is convolved with a large kernel (usually 4x4) and a large stride (usually 4), and the spatial size of the output feature map is reduced, while the number of channels is expanded.

[0053] 2) Stages

[0054] The structure of ConvNeXt is usually divided into 4 stages, each of which contains several ConvNeXt Blocks. The main function of each stage is to extract features of a specific scale. At the beginning of each stage, the resolution of the feature map decreases (through pooling or convolution operations), while the number of channels increases.

[0055] Before entering each new stage, the input feature map is first processed through a downsampling operation (usually achieved by convolution with a stride of 2) to reduce the resolution. The feature map is then processed through multiple ConvNeXt Blocks in sequence, with the number of channels remaining the same in each stage. At the end of each stage, the output feature map is passed to the next stage.

[0056] 3) Final Classification Layer, which is used for image classification tasks. The feature map output from the last stage is flattened and input into a fully connected layer to output the class probability. The output feature map from the last stage is globally pooled into a fixed-size vector, and then passes through a fully connected layer or a linear layer to output the final classification result.

[0057] According to a second aspect of the present invention, there is provided a fault location system based on decomposition noise reduction and a novel convolutional neural network, comprising: a computer-readable storage medium and a processor;

[0058] The computer-readable storage medium is used to store executable instructions;

[0059] The processor is used to read the executable instructions stored in the computer-readable storage medium and execute the steps of the above-mentioned fault distance measurement method based on decomposition denoising and novel convolutional neural network.

[0060] Through the above design scheme, the present invention can bring the following beneficial effects:

[0061] First, improve accuracy: By adopting a new convolutional neural network to perform feature learning on the obtained data, the characteristic information of the signal can be fully mined, thereby improving accuracy.

[0062] Second, enhance reliability: The signal is processed by noise reduction algorithm to enhance the reliability and robustness of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present application. The exemplary embodiments of the present invention and their descriptions are used to understand the present invention and do not constitute improper limitations of the present invention. In the drawings:

[0064] Figure 1 This is a schematic diagram of the experimental distribution network model;

[0065] Figure 2 Example graph for data augmentation processing;

[0066] Figure 3 This is a structural diagram of a new convolutional neural network;

[0067] Figure 4 It is the structure diagram of the convolution block;

[0068] Figure 5 It is the structure diagram of the downsampling layer;

[0069] Figure 6 This is a flow chart of the fault location method based on decomposition denoising and a new convolutional neural network. DETAILED DESCRIPTION

[0070] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical scheme in the present invention is clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the present invention is not limited to the following embodiments, and the specific implementation method can be determined according to the technical scheme of the present invention and the actual situation. In order to avoid confusing the essence of the present invention, the known methods, processes and procedures are not described in detail.

[0071] The present invention proposes a fault distance measurement method based on decomposition denoising and a novel convolutional neural network, and a fault distance measurement system based on decomposition denoising and a novel convolutional neural network, the system comprising a computer-readable storage medium and a processor; the computer-readable storage medium is used to store executable instructions; the processor is used to read the executable instructions stored in the computer-readable storage medium, and execute the steps of the fault distance measurement method based on decomposition denoising and a novel convolutional neural network. Exemplarily, the processor can be a central processing unit (CPU) or other general-purpose processors. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The computer-readable storage medium can be non-volatile or volatile.

[0072] The fault distance measurement method based on decomposition noise reduction and a new convolutional neural network proposed in the present invention improves the anti-interference ability of the system by introducing a noise reduction method. The present invention uses signal decomposition and reconstruction to perform noise reduction processing on the original current signal. The Gram angular field algorithm is used to perform image conversion on the collected data, convert the time series signal into a two-dimensional image signal, and perform vertical flipping, horizontal flipping, graying, brightness adjustment and noise addition operations on the signal. A new convolutional neural network is introduced, and the collected signal is input into the neural network for feature learning, and finally the fault distance is output.

[0073] Figure 1 The schematic diagram of the experimental distribution network model. The model used in the present invention is a radial distribution network with 5 outgoing lines. The voltage on the grid side is 110kV, which is converted to 10kV by the transformer. The distribution network voltage level is 10kV. The model line lengths are 18km, 10km, 12km, 16km, 13km, and 17km respectively. The fault type is set as a single-phase grounding fault in the line, and the grounding phase is phase A.

[0074] The present invention Figure 1 Experimental verification is carried out on the distribution network model shown, batch simulation experiments are carried out, different fault conditions are set for simulation experiments and data is collected.

[0075] Figure 2 This is an example of data enhancement processing. In order to avoid the problem that the data set sample diversity is insufficient, resulting in insufficient description of the complete sample distribution, data enhancement processing is performed to improve the diversity of data set samples and the generalization ability of the model. The present invention uses vertical flipping, horizontal flipping, graying, adjusting brightness and adding noise to enhance the data. The specific enhancement method used in the present invention belongs to the prior art and will not be described in detail here.

[0076] Figure 3This is a structural diagram of a new convolutional neural network. Compared with traditional convolutional neural networks, this network adopts a unique parallel branch structure, introduces multiple branches to extract different features of the data set, and enhances the robustness and learning ability of the model through cross-channel aggregation. Among them, the cross-channel is also a major innovation of the network. The cross-channel connects the various branches, communicates and integrates different information, and effectively improves the performance of the model. Multi-layer convolution and multi-layer pooling are constructed in each branch to form a deep network structure, which can effectively extract feature information from feature images. The model consists of convolution blocks and downsampling layers. The structure of convolution blocks and downsampling layers is as follows: Figure 4 and 5 shown.

[0077] Overview of the overall data transmission of the neural network proposed in this invention:

[0078] The ConvNeXt-T network is divided into four main stages, each of which contains multiple ConvNeXt Blocks. A parallel branch structure is introduced in each Block, and cross-channel summaries are performed after the output of each branch.

[0079] 1) Stem Layer (initial convolutional layer)

[0080] Input: image of size 224x224x3.

[0081] Convolution kernel: 4x4 convolution with a stride of 4.

[0082] Output: feature map of size 56x56x96.

[0083] Function: Downsample, reduce the resolution of the input image, and expand the number of channels.

[0084] 2) Stage 1

[0085] Structure: 3 ConvNeXt Blocks, each Block contains multiple parallel branches.

[0086] Number of channels: 96.

[0087] Output: After processing by 3 ConvNeXtBlocks, the output feature map is 56x56x96 in size.

[0088] 3) Stage 2

[0089] Downsampling: Reduce the feature map size by half through a 2x2 convolution with a stride of 2.

[0090] Structure: 3 ConvNeXt Blocks, including parallel branching structure.

[0091] Number of channels: 192.

[0092] Output: feature map of size 28x28x192.

[0093] 4) Stage 3

[0094] Downsampling: Through a 2x2 convolution with a stride of 2.

[0095] Structure: 9 ConvNeXt Blocks with multi-branch structure.

[0096] Number of channels:384.

[0097] Output: feature map of size 14x14x384.

[0098] 5) Stage 4

[0099] Downsampling: Again pass 2x2 convolution with stride 2.

[0100] Structure: 3 ConvNeXt Blocks with integrated parallel branches.

[0101] Number of channels:768.

[0102] Output: feature map of size 7x7x768.

[0103] ConvNeXt Block is the core unit of ConvNeXt-T. Each Block contains multiple parallel branches, which may use different convolution kernel sizes or different step sizes to process the input feature map and capture information of different scales.

[0104] Workflow of parallel branching and cross-channel aggregation:

[0105] 1. Input feature map: The input feature map is copied to multiple parallel branches, each branch adopts different convolution operations (such as different convolution kernel sizes).

[0106] 2. Parallel processing: Each branch processes the input features independently and produces different output feature maps.

[0107] 3. Aggregation: The output features of different branches are fused together by channel concatenation or element-by-element addition.

[0108] 4. Post-fusion processing: further compress and fuse channel information through 1x1 convolution to generate the final output features.

[0109] ConvNeXt-T introduces a parallel branch structure and a cross-channel aggregation mechanism based on the traditional convolutional neural network, which enables it to capture features of different scales and generate more expressive feature representations after fusing these features. Through this design, ConvNeXt-T achieves superior performance while keeping the model lightweight.

[0110] Figure 6 This is the overall flow chart of the fault location method based on decomposition denoising and novel convolutional neural network proposed in the present invention. The overall process is as follows:

[0111] 1. First, collect the three-phase current of the distribution network line and set the sampling frequency.

[0112] 2. Decompose and reconstruct the collected current signal to achieve noise reduction.

[0113] 3. Convert the noise-reduced signal into an image and perform enhancement processing.

[0114] 4. Input the processed feature image data as a data set into the new convolutional neural network and output the fault distance.

[0115] In summary, the present invention adopts a decomposition and noise reduction method to perform noise reduction on the collected information, and performs image conversion on the noise-reduced data, and finally enhances the image data. The present invention adopts a new convolutional neural network, which is based on a neural network and obtains a new network through different structural processing and layer changes for feature mining and learning of image data.

Claims

1. A fault location method based on decomposition denoising and a novel convolutional neural network, characterized in that: include: Step 1: Collect the three-phase current signal of the line as the original current signal according to the set sampling frequency; Step 2: Use the ICEEMDAN algorithm to decompose and denoise the collected original current signal to obtain K IMF components; Step 3: Calculate the entropy value of each IMF component, and filter out the IMF components with entropy values ​​greater than the threshold, and reconstruct the remaining IMF components to obtain the denoised signal; Step 4: Use the Gram angular field algorithm to convert the denoised signal into a two-dimensional feature image, and perform data enhancement on the obtained two-dimensional feature image to generate a two-dimensional feature image dataset; Step 5: Input the generated two-dimensional feature image data set into the new convolutional neural network model to obtain the distance measurement result, thus completing the fault distance measurement; The novel convolutional neural network model includes: 1) Initial convolutional layer The original image is input into the initial convolution layer, after which the size is reduced and the channels are expanded; 2) Stage structure The structure of ConvNeXt is usually divided into four stages, each of which contains several ConvNeXt Blocks. Each stage is used to extract features of a specific scale. At the beginning of each stage, the resolution of the feature map is reduced and the number of channels is increased through pooling or convolution operations. Before entering each new stage, the input feature map first passes through a downsampling operation to reduce the resolution; then, the feature map is processed by multiple ConvNeXt Blocks in sequence, and the number of channels remains unchanged in each stage; at the end of each stage, the output feature map is passed to the next stage; 3) Classification layer: Flatten the feature map output by the last stage of the stage structure and input it into a fully connected layer to output the category probability; the output feature map of the last stage is globally pooled into a vector of a fixed size, and then passes through a fully connected layer or a linear layer to output the final classification result; The ConvNeXt Block includes: The deep convolution layer performs a deep convolution operation to independently perform convolution calculations on each channel in the input feature map. In this layer, the input feature map is divided into multiple channels, each channel is convolved separately, and the output has the same number of channels as the input. The point convolution layer is executed immediately after the depth convolution and is responsible for linearly combining the information of different channels. In this layer, all channels of the feature map after the depth convolution are linearly combined to generate new channels. Normalization layer, which performs normalization on each layer of the feature map. Each channel of the feature map in this layer is standardized so that the output mean is 0 and the variance is 1; GELU activation function, after being processed by the activation function, the output features retain the nonlinear information of the original features; Randomly discard paths DropPath, randomly discard some paths during training to increase the generalization ability of the model.

2. The fault location method based on decomposition denoising and novel convolutional neural network according to claim 1 is characterized in that: The sampling frequency is set to 5kHz.

3. The fault location method based on decomposition denoising and novel convolutional neural network according to claim 1 is characterized in that: Step 2 specifically includes: First, take the original current signal x[n] and add Gaussian white noise to construct a total of I groups of noise-added signals; x (i) =x+β0E1(w (i) ) (1) In the formula, x (i) is the i-th current signal after adding Gaussian white noise, x is the original current signal, w (i) represents the added i-th group of white noise, where i=1,2,…,I, let w (i) is Gauss white noise with mean 0 and unit variance; the operator E1(·) represents the first-order IMF component of the EMD decomposition of a signal, β0 represents the weight coefficient when the noise component is added, and the coefficient represents the ratio of the signal-to-noise ratio of the added noise to the standard deviation of the noise component; The local operator M(·) is introduced, which represents the local mean of the signal. The calculation can be obtained for each x (i) The corresponding first-order IMF component By summing the above components, we can get the first-order IMF component D1 of the original current signal x; Obtain the first-order residual r1: r1=x-D1 (4) The second-order residual r2: Where β1 refers to the weight coefficient of adding noise to the second-order residual; the operator E2(·) refers to the second-order IMF component of the signal EMD decomposition; The second-order IMF component D2 is obtained as: Repeat the above steps to obtain the k-th order residual r k , k = 3, 4, ..., K, K is the total decomposition order; Among them, β k-1 Refers to the weight coefficient of adding noise to the k-1th order residual; operator E k (·) refers to the k-th order IMF component of the signal EMD decomposition; The Kth order IMF component D k for: D k ＝r k-1 -r k (8) Repeat the above steps to obtain the IMF components decomposed by ICEEMDAN.

4. The fault location method based on decomposition denoising and novel convolutional neural network according to claim 1 is characterized in that: Data enhancement processing includes flipping, grayscale, brightness adjustment, and noise addition operations.

5. The fault location method based on decomposition denoising and novel convolutional neural network according to claim 1 is characterized in that: The initial convolutional layer is a 4x4 convolution with a stride of 4.

6. The fault location method based on decomposition denoising and novel convolutional neural network according to claim 1 is characterized in that: Downsampling is usually done via convolution with a stride of 2.

7. A fault distance measurement system based on decomposition noise reduction and a novel convolutional neural network, comprising a computer-readable storage medium and a processor; the computer-readable storage medium is used to store executable instructions; characterized in that: The processor is used to read the executable instructions stored in the computer-readable storage medium and execute the steps of the fault distance measurement method based on decomposition denoising and a new convolutional neural network as described in any one of claims 1-6.

Images