CNN-BiLSTM grounding fault phase selection method, system and device fused with CBAM, and medium

Abstract

The invention discloses a CNN-BiLSTM grounding fault phase selection method, system and device fused with CBAM, and a medium, and relates to the technical field of computer platform load balancing, and the method comprises the steps: generating a two-dimensional grayscale image through a signal image conversion method according to collected distribution network grounding fault data, and carrying out the RGB image generation processing of the two-dimensional grayscale image; constructing a CNN-BiLSTM phase selection model fused with CBAM, and performing iterative training to obtain a distribution network grounding fault phase selection model; and inputting the RGB image after the two-dimensional grayscale image processing into a distribution network grounding fault phase selection model, and carrying out fault phase selection on a grounding fault. The method improves the expression capability of the model for current change characteristics, and improves the discrimination precision of the model. Cooperative extraction of spatial local features and time sequence features is realized by constructing a CNN-BiLSTM network architecture fused with a CBAM attention mechanism.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of computer platform load balancing, and particularly relates to a CNN-BiLSTM grounding fault phase selection method, system, equipment and medium fusing CBAM. BACKGROUND

[0002] With the expansion of the distribution network scale and the improvement of the power supply reliability requirement, the rapid detection and accurate positioning of the distribution network fault become the key to ensure the stable operation of the power system; the fault phase selection method can realize fault isolation and recovery in a timely manner after the fault occurs, reduce the power outage time, and improve the reliability and operation efficiency of the power grid.

[0003] How to realize efficient and accurate fault phase selection under complex distribution network topology and various types of fault conditions has become the focus of current research. However, when facing the increasing complexity of the distribution network and the challenge of massive data, the performance of the traditional power system protection method based on a simple linear model and artificial detection is limited. With the improvement of the digitalization and intelligentization level of the distribution network, it is increasingly easy to obtain massive fault information. Methods based on artificial intelligence theory are increasingly applied to the field of fault diagnosis, including support vector machines, artificial neural networks, fuzzy neural networks, convolutional neural networks, convolutional deep belief networks and other methods. The construction of the feature vector in the existing machine learning method excessively depends on the artificial expert experience, and it is difficult to meet the phase selection accuracy requirement under the complex distribution network topology and various types of fault conditions. The deep learning algorithm can adaptively extract the features of the historical fault data and learn, which makes up for the weakness of the traditional machine learning feature extraction ability. However, such methods are often combined with signal processing methods. Since the fault features extracted by the signal processing method lack completeness, it is difficult to adapt to all fault conditions. In addition, the existing deep learning method lacks the mining and attention of key features, which leads to the need to improve the phase selection accuracy of the existing deep learning method. SUMMARY

[0004] In view of the above problems, the application is proposed.

[0005] Therefore, the technical problem solved by the application is that the existing fault signal feature expression ability is weak, the model lacks a key feature attention mechanism, the accuracy and robustness are not high under complex working conditions, and how to realize a deep learning phase selection method fusing image modeling and attention mechanism.

[0006] To solve the above technical problems, the application provides the following technical scheme: a CBAM-fused CNN-BiLSTM grounding fault phase selection method, comprising: generating a two-dimensional gray image through a signal image conversion method according to collected distribution network grounding fault data, and performing RGB image generation processing on the two-dimensional gray image; constructing a CBAM-fused CNN-BiLSTM phase selection model and performing iterative training to obtain a distribution network grounding fault phase selection model; inputting the RGB image processed from the two-dimensional gray image into the distribution network grounding fault phase selection model to perform grounding fault phase selection.

[0007] As a preferred scheme of the CBAM-fused CNN-BiLSTM grounding fault phase selection method, the signal image conversion method comprises: when the distribution network has a grounding fault, collecting three-phase current signals in two power frequency cycles after the fault occurs.

[0008] The three-phase current signals are converted into three two-dimensional gray images through a conversion algorithm by using the signal image conversion method.

[0009] As a preferred scheme of the CBAM-fused CNN-BiLSTM grounding fault phase selection method, the RGB image generation processing comprises: mapping the three two-dimensional gray images obtained through the signal image conversion method to the channels of the RBG image to obtain the RGB image.

[0010] As a preferred scheme of the CBAM-fused CNN-BiLSTM grounding fault phase selection method, the CBAM-fused CNN-BiLSTM phase selection model comprises: an input layer, a local feature extraction layer, a CBAM fusion layer, a time sequence feature extraction layer, and an output layer.

[0011] The local feature extraction layer is composed of two convolution layers, a Relu layer, a BN layer, and a maximum pooling layer.

[0012] The CBAM fusion layer is composed of channel attention and spatial attention.

[0013] The time sequence feature extraction layer is composed of a global average pooling layer, a folding layer, a full connection layer, and a BiLSTM layer.

[0014] The output layer is composed of a full connection layer, a Dropout layer, and a SoftMax layer.

[0015] As a preferred scheme of the CBAM-fused CNN-BiLSTM grounding fault phase selection method, the CBAM fusion layer comprises: taking an intermediate feature map as input, and the channel attention performs a pooling operation on the spatial region of the feature map.

[0016] The feature map after the pooling operation is transmitted to the shared network and merged by element-wise summation to obtain a final channel attention map.

[0017] For spatial attention, the channel region of the feature map is subjected to a pooling operation, and a final spatial attention map is obtained through cascading and convolution of the convolution layer.

[0018] The channel attention and the spatial attention are sequentially stacked to form a final CBAM.

[0019] As a preferred scheme of the CNN-BiLSTM grounding fault phase selection method fused with CBAM, the output layer includes a full connection layer that expands the features learned by the BiLSTM into a column vector column by column and inputs into a Softmax layer, calculates the probability of each type of the current sample, and takes the label with the maximum probability as the prediction output of the CNN-BiLSTM phase selection model.

[0020] As a preferred scheme of the CNN-BiLSTM grounding fault phase selection method fused with CBAM, the fault phase selection includes, when a grounding fault occurs in the distribution network, performing a signal image conversion method on the fault data, performing RGB image generation processing on the generated two-dimensional gray image,

[0021] The model parameters are set, and the CNN-BiLSTM phase selection model fused with CBAM is trained until the training condition is met, and the phase selection model is output.

[0022] The RGB image X of the fault data is input into the trained CNN-BiLSTM phase selection model fused with CBAM to obtain the fault phase.

[0023] Another object of the present application is to provide a CNN-BiLSTM grounding fault phase selection system fused with CBAM, which can convert the fault current signal of the distribution network into an RGB image and input the CNN-BiLSTM model fused with channel attention and spatial attention mechanisms for modeling and classification, thereby solving the technical problems of weak feature expression ability, insufficient attention to key features, and poor stability of the model under network topology changes and signal interference in the current distribution network fault phase selection technology.

[0024] As a preferred scheme of the fusion CBAM CNN-BiLSTM grounding fault phase selection system, wherein: comprising a fault data image generation module, a phase selection model processing module, and a fault phase selection module; the fault data image generation module is used for generating a two-dimensional gray image through a signal image conversion method according to the collected distribution network grounding fault data, and performing RGB image generation processing on the two-dimensional gray image; the phase selection model processing module is used for constructing a fusion CBAM CNN-BiLSTM phase selection model, and performing iterative training to obtain a distribution network grounding fault phase selection model; and the fault phase selection module is used for inputting the RGB image processed from the two-dimensional gray image into the distribution network grounding fault phase selection model to perform fault phase selection on the grounding fault.

[0025] A computer device comprises a memory and a processor, the memory stores a computer program, and the processor executes the computer program to implement the steps of the fusion CBAM CNN-BiLSTM grounding fault phase selection method.

[0026] A computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the fusion CBAM CNN-BiLSTM grounding fault phase selection method.

[0027] The fusion CBAM CNN-BiLSTM grounding fault phase selection method provided by the present application encodes time domain waveform information into an image through a signal image conversion method and RGB mapping processing, and inputs the image into a deep neural network, thereby improving the expression ability of the model to current change characteristics and improving the discrimination accuracy of the model. BRIEF DESCRIPTION OF DRAWINGS

[0028] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0029] Figure 1 The present application provides a fusion CBAM CNN-BiLSTM grounding fault phase selection method.

[0030] Figure 2 A signal image conversion method of a CNN-BiLSTM grounding fault phase selection method fusing CBAM provided for an embodiment of the present application is shown in the figure.

[0031] Figure 3 A CBAM fusion layer process diagram of a CNN-BiLSTM grounding fault phase selection method fusing CBAM provided for an embodiment of the present application is shown in the figure.

[0032] Figure 4 A whole flow chart of a CNN-BiLSTM grounding fault phase selection system fusing CBAM provided for an embodiment of the present application is shown in the figure.

[0033] Figure 5 A phase selection accuracy result diagram of different phase selection methods of a CNN-BiLSTM grounding fault phase selection method fusing CBAM provided for an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0034] In order to make the above objectives, characteristics and advantages of the present application more apparent, obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the present application.

[0035] Embodiment 1, refer to Figure 1-Figure 3 For an embodiment of the present application, a CNN-BiLSTM grounding fault phase selection method fusing CBAM is provided, which comprises:

[0036] S1: According to the collected grounding fault data of the distribution network, a two-dimensional gray image is generated by a signal image conversion method, and the two-dimensional gray image is subjected to RGB image generation processing.

[0037] Further, the signal image conversion method comprises: when the grounding fault occurs in the distribution network, collecting three-phase current signals within two power frequency cycles after the fault occurs; using a signal image conversion method to convert the three-phase current signals into three two-dimensional gray images through a conversion algorithm, and the conversion method is represented as:

[0038]

[0039] Wherein, P(i,j) is the pixel intensity of the i-th row and j-th column in the two-dimensional gray image, N represents the size of the converted image, X is the value of the phase current signal, and the length is N 2X(i) = round(255 * (X(i) - X(i-1)) / (X(i-1) - X(i-2))), X(i-1) is the value of the i-1th data point in X, and the role of round is to take the data value to ensure that the converted data value is an integer between 0 and 255.

[0040] It should be noted that, as shown in the embodiments of the present application, the RGB image generation processing includes mapping three two-dimensional gray scale images obtained by the signal image conversion method to the channels of the RBG image to obtain the RGB image. Figure 2

[0041] In an alternative embodiment, the intensity proportion of each input gray scale image affecting the final image red, green and blue channels is defined in advance. Generally, the contribution proportion of each image to all channels is ensured to add up to 100%, preventing color values from exceeding the normal range. For each pixel position on the final RGB image, the channel value of the three channels is obtained by taking the gray scale value of image A at the position multiplied by its red proportion, adding the gray scale value of image B at the position multiplied by its red proportion, and adding the gray scale value of image C at the position multiplied by its red proportion. The three products are added to obtain the final red intensity of the point. The intensity values of the red, green and blue channels of all pixel positions obtained are combined to form the final RGB color image.

[0042] In another alternative embodiment, for each pixel position on the image, the gray scale values of the position on the three gray scale images (image A, image B, image C) are simultaneously viewed, and one or more characteristic values reflecting the relationship among the three are calculated. Common features include maximum difference, dominant phase, average intensity, etc. According to the calculated main characteristic value, a pre-defined color mapping rule table is looked up. The rule table specifies what color corresponds to different characteristic value ranges, and the color (composed of red, green and blue components) of each pixel position obtained according to the characteristic value mapping rule is combined to form the final RGB color image.

[0043] S2: construct a CBAM-fused CNN-BiLSTM phase selection model and perform iterative training to obtain a distribution network grounding fault phase selection model;

[0044] Further, the CBAM-fused CNN-BiLSTM phase selection model includes an input layer, a local feature extraction layer, a CBAM fusion layer, a time sequence feature extraction layer, and an output layer.

[0045] The local feature extraction layer is composed of two convolution layers, a Relu layer, a BN layer, and a maximum pooling layer; the CBAM fusion layer is composed of channel attention and spatial attention; the time sequence feature extraction layer is composed of a global average pooling layer, a folding layer, a fully connected layer, and a BiLSTM layer; and the output layer is composed of a fully connected layer, a Dropout layer, and a SoftMax layer. ​

[0046] The local feature extraction layer includes a convolution layer for the feature map M l The convolution operation is performed to obtain the feature map M l , which is expressed as:

[0047]

[0048] wherein, is the convolution kernel matrix of the lth layer, and * represents the sliding convolution process in the CNN; the BN layer performs batch normalization on the output feature map M l , which is expressed as:

[0049]

[0050] wherein, represents the reconstruction parameter, E represents the expectation of the input sample, and Var represents the variance of the input sample; the Relu layer activates the output of the BN layer, and the activation function adopts the Relu function, which is expressed as:

[0051]

[0052] wherein, δ represents the activation function; the max-pooling layer performs down-sampling, which has the advantages of reducing learning parameters and avoiding overfitting.

[0053] As shown in Figure 3 , the CBAM fusion layer includes an intermediate feature map as input, the channel attention performs a pooling operation on the spatial region of the feature map; the feature map after the pooling operation is transmitted to the shared network and merged through element-wise summation to obtain the final channel attention map , which is expressed as:

[0054]

[0055] wherein, respectively represent the features on the channel axis after the average pooling and max-pooling operations, W1 and W2 are respectively the learning weights of the fully connected layer, σ represents the sigmoid function, and δ represents the activation function; for the spatial attention, the channel region of the feature map is subjected to a pooling operation, and the final spatial attention map is obtained through cascading and convolution of the convolution layer , which is expressed as:

[0056]

[0057] wherein, respectively represent the features on the spatial axis after the average pooling and max-pooling operations, is a cascade operator, f represents the convolution operation; the channel attention and spatial attention are stacked in sequence to form the final CBAM; the output F′ of CBAM is expressed as:

[0058]

[0059] in, Represents element-wise multiplication.

[0060] The temporal feature extraction layer includes the global average pooling layer in the temporal feature extraction layer, which averages each channel of the feature map F′ output by the CBAM layer, expressed as:

[0061]

[0062] Among them, H×W represents the feature map F′ i,j The pixel value at position (i, j), ∑∑ represents the sum of the pixel values ​​at all positions.

[0063] The folding layer expands the previous output into a column and inputs it into the fully connected layer and BiLSTM layer. The BiLSTM layer combines the forward and the latter The information in both directions provides a more comprehensive understanding of the characteristic connections of the data in the time series, thereby improving the model's utilization of data and prediction accuracy, which can be expressed as:

[0064]

[0065] The output layer includes a fully connected layer that expands the features learned by BiLSTM into column vectors column by column and inputs them into the Softmax layer. The probability of each type of the current sample is calculated and the label with the highest probability is taken as the predicted output of the CNN-BiLSTM phase selection model, which is expressed as:

[0066]

[0067] in, Predict the probability of one type for the current sample.

[0068] In summary, by building a CNN-BiLSTM phase selection model integrated with CBAM, we combine local feature extraction, attention mechanisms, and temporal feature extraction to form a multi-layered feature learning framework. This process enables deep learning and feature extraction of input data, enabling the model to automatically capture key information. Furthermore, the CBAM module, through channel and spatial attention mechanisms, enhances the model's sensitivity to important features, effectively reducing interference from redundant information, and significantly improving the accuracy of ground fault detection and prediction.

[0069] S3: input the RGB image processed by the two-dimensional gray image into the grounding fault phase selection model of the distribution network to perform grounding fault phase selection.

[0070] Further, the fault phase selection includes, when the grounding fault occurs in the distribution network, performing a signal-image conversion method on the fault data, performing RGB image generation processing on the generated two-dimensional gray image, and inputting the RGB image X of the fault data into the trained CNN-BiLSTM phase selection model fused with CBAM to obtain the fault phase.

[0071] It should be noted that the parameter setting before model training includes that the initial learning rate is 0.001, the training rounds are 40 times, the data is randomly shuffled in each training round to prevent model overfitting, and the model uses Adam optimizer for solution.

[0072] It should also be noted that the training process of the CNN-BiLSTM network model fused with CBAM includes the forward propagation and backward propagation processes.

[0073] The training data is compared with the target label value, the error between the predicted value and the target value is calculated, if the training condition is not met, the weight and threshold of the network are updated, the error value between the predicted value and the target value is repeatedly calculated until the training condition is met, and the network model is output.

[0074] When the grounding fault occurs in the distribution network, the RGB image generation and the CNN-BiLSTM phase selection model fused with CBAM are executed, the fault data X is input into the CNN-CBAM-BiLSTM fault phase selection model y=f(X), wherein the value of y is [1, 2,..., N], indicating the phase of the line fault in the distribution network, corresponding to 8 fault phases AG, BG, CG, ABG, ACG, BCG, ABCG and Non-Faulty respectively; the data label y is in the form shown in Table 1:

[0075] Table 1 Data label y form

[0076]

[0077] For fault phase selection, when the grounding fault occurs in the distribution network system, set the fault time as TS, collect the three-phase current of the fault line within two power frequency cycles after the fault occurs

[0078]

[0079] The signal-image conversion method and the RGB image generation are performed on X to obtain the image X *'

[0080] The image X is input into the CNN-CBAM-BiLSTM fault phase selection model *'The input is fused into the CNN-BiLSTM phase selection model, that is, y=f(X *' ), and the fault phase is determined according to the y value.

[0081] In summary, by inputting the processed RGB image into the trained phase selection model, the ground fault can be quickly identified and classified. The beneficial effect of this process is that the prediction ability of the model can be used to determine the fault phase in real time, providing timely fault analysis information. When a ground fault occurs in the distribution network, the model output by the training can accurately output the fault phase label, which can effectively guide the subsequent fault processing and recovery work, and reduce the economic loss and power outage time caused by incorrect fault judgment.

[0082] Embodiment 2, refer to Figure 5 An embodiment of the present application provides a CNN-BiLSTM grounding fault phase selection method fusing CBAM. In order to verify the beneficial effects of the present application, scientific demonstration is carried out through experiments.

[0083] The experiment compares ANFIS, HHT-CNN, CNN-BiLSTM and the CNN-BiLSTM fusing CBAM used in the present application, wherein the CNN-BiLSTM and the method proposed in this paper use the same network parameters. The phase selection accuracy results of different phase selection methods are shown in Figure 5 .

[0084] The topology structure of the distribution network changes due to balanced load, reduced network loss, etc. For the distribution network shown in Figure 3 , two kinds of topological structures are set to test the influence of topological adjustment on the model accuracy of the proposed method, that is, deleting line L4 and adding a 5km overhead line L5, and the test results are shown in Table 2.

[0085] Table 2 Topology structure test result 1

[0086] Network topology changes Sample size Phase selection accuracy / % Delete line L4 320 100 New line L5 320 98.75

[0087] The distribution network system in Figure 3 is adjusted to a neutral point ungrounded mode. Line L4 is set to be faulty, and the simulation parameters are shown in Table 3. 224 test samples are obtained after the neutral point ungrounded mode is operated, and the test results are shown in Table 3.

[0088] Table 3 Test results after neutral point ungrounded mode operation

[0089]

[0090]

[0091] The Gaussian white noise with different signal-to-noise ratios is superimposed on the test set to test the noise resistance of the model. The noise is superimposed 10 times randomly, and the average value of 10 times is taken as the evaluation index. The test results of the method are shown in Table 4.

[0092] Table 4 Anti-noise ability test results

[0093] SNR(dB) 30 20 10 Average phase selection accuracy (%) 100 99.99 99.97

[0094] Embodiment 3 is a third embodiment of the application, which is different from the previous embodiment in that:

[0095] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the application essentially or the parts that contribute to the prior art or parts of the technical solutions can be embodied in the form of software products. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device) to execute all or part of the steps of the method described in the various embodiments of the application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.

[0096] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered a list of executable instructions for implementing logic functions, and can be specifically embodied in any computer-readable medium for use by an instruction execution system, apparatus or device, such as a computer-based system, a system including a processor or other system that can fetch and execute instructions from the instruction execution system, apparatus or device, or in conjunction with these instructions. For this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transport programs for instruction execution systems, apparatus or devices, or in conjunction with these instruction execution systems, apparatus or devices.

[0097] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can also be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example, via optical scanning of the paper or other medium, then compiled, interpreted, or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory.

[0098] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the above embodiments, various steps or methods can be implemented in software or firmware that is stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any of the following technologies, known in the art, or their combinations can be used: discrete logic circuitry having logic gates for implementing logic functions on data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), etc.

[0099] Embodiment 4, reference Figure 4 As a fourth embodiment of the present application, the embodiment provides a CBAM-fused CNN-BiLSTM grounding fault phase selection system, comprising a fault data image generation module, a phase selection model processing module, and a fault phase selection module.

[0100] The fault data image generation module is configured to generate a two-dimensional gray image by a signal image conversion method according to the collected distribution network grounding fault data, and perform RGB image generation processing on the two-dimensional gray image; the phase selection model processing module is configured to construct a CBAM-fused CNN-BiLSTM phase selection model, and perform iterative training to obtain a distribution network grounding fault phase selection model; and the fault phase selection module is configured to input the RGB image processed from the two-dimensional gray image into the distribution network grounding fault phase selection model, and perform fault phase selection on the grounding fault.

[0101] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A CNN-BiLSTM ground fault phase selection method integrated with CBAM, characterized by: include: Based on the collected distribution network grounding fault data, a two-dimensional grayscale image is generated through a signal image conversion method, and the two-dimensional grayscale image is processed into an RGB image; A CNN-BiLSTM phase selection model integrated with CBAM was constructed and iteratively trained to obtain a distribution network grounding fault phase selection model. The RGB image processed by the two-dimensional grayscale image is input into the distribution network grounding fault phase selection model to perform fault phase selection for the grounding fault.

2. The CNN-BiLSTM ground fault phase selection method integrated with CBAM according to claim 1 is characterized by: The signal image conversion method includes, when a ground fault occurs in the distribution network, collecting three-phase current signals within two power frequency cycles after the fault occurs; The signal image conversion method is used to convert the three-phase current signal into three two-dimensional grayscale images through a conversion algorithm.

3. The CNN-BiLSTM ground fault phase selection method integrated with CBAM as claimed in claim 2 is characterized by: The RGB image generation process includes mapping three two-dimensional grayscale images obtained by the signal image conversion method to channels of the RBG image to obtain the RGB image.

4. The CNN-BiLSTM ground fault phase selection method integrated with CBAM as claimed in claim 3 is characterized by: The CNN-BiLSTM phase selection model fused with CBAM includes an input layer, a local feature extraction layer, a CBAM fusion layer, a temporal feature extraction layer, and an output layer; Among them, the local feature extraction layer consists of two convolutional layers, a Relu layer, a BN layer, and a maximum pooling layer; The CBAM fusion layer consists of channel attention and spatial attention; The temporal feature extraction layer consists of a global average pooling layer, a folding layer, a fully connected layer, and a BiLSTM layer; The output layer consists of a fully connected layer, a Dropout layer, and a SoftMax layer.

5. The CNN-BiLSTM ground fault phase selection method integrated with CBAM according to claim 4 is characterized by: The CBAM fusion layer includes taking an intermediate feature map as input and performing a pooling operation on the spatial region of the feature map with channel attention; The feature map after the pooling operation is passed to the shared network and merged by element-by-element summation to obtain the final channel attention map; For spatial attention, the channel area of ​​the feature map is pooled and concatenated through the convolutional layer to obtain the final spatial attention map; The channel attention and spatial attention are stacked in sequence to form the final CBAM.

6. The CNN-BiLSTM ground fault phase selection method integrated with CBAM according to claim 5 is characterized by: The output layer includes a fully connected layer that expands the features learned by BiLSTM into column vectors column by column and inputs them into the Softmax layer to calculate the probability of each type of the current sample and take the label with the highest probability as the predicted output of the CNN-BiLSTM phase selection model.

7. The CNN-BiLSTM ground fault phase selection method integrated with CBAM according to claim 6 is characterized by: The fault phase selection includes executing a signal image conversion method on the fault data when a ground fault occurs in the distribution network, and performing RGB image generation processing on the generated two-dimensional grayscale image. Set model parameters, train the CNN-BiLSTM phase selection model of the fused CBAM until the training conditions are met, and output the phase selection model; The RGB image X of the fault data is input into the trained CNN-BiLSTM phase selection model fused with CBAM to obtain the fault phase.

8. A system using the CNN-BiLSTM ground fault phase selection method integrated with CBAM as described in any one of claims 1 to 7, characterized in that: It includes fault data image generation module, phase selection model processing module, and fault phase selection module; The fault data image generation module is used to generate a two-dimensional grayscale image based on the collected distribution network grounding fault data through a signal image conversion method, and perform RGB image generation processing on the two-dimensional grayscale image; The phase selection model processing module is used to construct a CNN-BiLSTM phase selection model integrated with CBAM and perform iterative training to obtain a distribution network grounding fault phase selection model; The fault phase selection module is used to input the two-dimensional grayscale image after RGB image processing into the distribution network grounding fault phase selection model to perform fault phase selection on the grounding fault.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the CNN-BiLSTM ground fault phase selection method integrated with CBAM according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the CNN-BiLSTM ground fault phase selection method integrated with CBAM according to any one of claims 1 to 7 are implemented.

Images