A method and device for detecting arc faults in electric vehicles based on multidimensional networks

Abstract

The present invention provides an electric vehicle arc fault detection method and device based on a multidimensional network. The method includes obtaining arc fault current data and converting it into a two-dimensional image data set using a GADF algorithm; a multi-scale hierarchical information integration module is further developed to perform preliminary feature extraction on the current image data. At the same time, a parallel interactive network module is designed and combined with a multi-scale hierarchical information integration module to form a comprehensive neural network detection algorithm; the module includes local arc and global arc feature extraction branches and a branch network specifically for extracting horizontal and vertical features to achieve multi-dimensional feature extraction. Finally, an accurate fault diagnosis result can be output through an electric vehicle arc fault neural network detection algorithm based on a multi-dimensional network. The present method aims to solve the problems in the prior art of insufficient extraction of electric vehicle arc fault features, ignoring multi-scale features and the correlation between different pixels, resulting in a high classification error rate.

Description

Technical Field

[0001] The present invention relates to the technical field of arc fault detection, and in particular to a method and device for detecting arc fault of an electric vehicle based on a multi-dimensional network. Background Art

[0002] With the rapid development of electric vehicle technology, arc fault detection has become one of the key technologies to ensure the safe operation of electric vehicles. Among them, arc fault refers to the phenomenon that the voltage between different electrodes in the circuit breaks through the insulating medium and produces luminous discharge. During the use of electrical equipment, arc faults are easily caused by loose and aging lines. Studies have shown that when an arc fault occurs in the line, a high temperature of 20000K can be instantly generated, which is very easy to burn combustible materials and cause fires. Therefore, the development of an efficient and accurate arc fault detection method is crucial for the safe operation of electric vehicles.

[0003] At present, traditional arc fault detection methods mainly rely on the analysis of current time series data, but these methods often have the problem of insufficient feature extraction, especially when dealing with multi-scale features and correlations between different pixels, resulting in a high classification error rate. In addition, traditional detection algorithms often ignore the differences in the performance of arc faults under different working conditions and the complexity of arc features in spatial distribution, which further limits the accuracy and generalization ability of detection algorithms.

[0004] In view of this, this application is filed. Summary of the invention

[0005] The present invention provides an electric vehicle arc fault detection method and device based on a multi-dimensional network, which can at least partially improve the above-mentioned problems.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] A method for detecting arc faults in electric vehicles based on a multidimensional network, comprising:

[0008] Acquire the current time series data of the arc fault collected by the collection device, and transform the current time series data to obtain two-dimensional image current data X;

[0009] Using a multi-scale hierarchical information integration module to perform preliminary feature extraction processing on the two-dimensional image current data X to generate a hierarchical information integration result;

[0010] A parallel interactive network module is used to perform feature extraction preprocessing on the hierarchical information integration result to obtain an interactive result Y, wherein the parallel interactive network module includes a local arc feature and a global arc feature extraction branch network, a horizontal direction feature extraction branch network, and a vertical direction feature extraction branch network;

[0011] The interaction result Y is adjusted and fused for preprocessing to obtain an arc fault diagnosis result.

[0012] The present invention also provides an electric vehicle arc fault detection device based on a multi-dimensional network, which comprises:

[0013] A conversion unit, used to obtain the current time series data of the arc fault collected by the collection device, and perform conversion processing on the current time series data to obtain two-dimensional image current data X;

[0014] A preliminary feature extraction unit, used to perform preliminary feature extraction processing on the two-dimensional image current data X using a multi-scale hierarchical information integration module to generate a hierarchical information integration result;

[0015] A feature extraction unit, used for performing feature extraction preprocessing on the hierarchical information integration result by using a parallel interactive network module to obtain an interactive result Y, wherein the parallel interactive network module includes a local arc feature and a global arc feature extraction branch network, a horizontal direction feature extraction branch network, and a vertical direction feature extraction branch network;

[0016] The result generating unit is used to adjust and fuse the interaction result Y to obtain an arc fault diagnosis result.

[0017] In summary, the electric vehicle arc fault detection based on multidimensional network obtains the current sequence data of the arc fault, and converts these data into a two-dimensional current image data set using the GADF algorithm. Next, a multi-scale hierarchical information integration module is designed to perform preliminary extraction on the input two-dimensional current image data of electric vehicle arc fault detection; and a parallel interactive network module is also designed to perform multi-branch feature extraction, which includes branch networks for extracting local arc features and global arc features, as well as branch networks for extracting horizontal and vertical features respectively. The multi-scale hierarchical information integration module is integrated with the parallel interactive network module to construct an electric vehicle arc fault detection algorithm based on a multi-dimensional neural network; finally, the two-dimensional current image data of the arc fault is processed according to the detection algorithm to obtain the arc fault diagnosis result.

[0018] In short, the electric vehicle arc fault detection method based on multidimensional network is a detection method that can fully extract arc fault features, consider multi-scale features and correlation between different pixels, and has the advantages of reducing classification error rate and improving the accuracy and reliability of electric vehicle arc fault detection. It aims to solve the problem of insufficient extraction of electric vehicle arc fault features, ignoring multi-scale features and correlation between different pixels in the prior art, resulting in a high classification error rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flowchart of a method for detecting arc faults in electric vehicles based on a multi-dimensional network provided by the first embodiment of the present invention;

[0020] Figure 2 is a flow chart of a multi-scale hierarchical information integration module provided by the first embodiment of the present invention;

[0021] Figure 3 is a flow chart of a parallel interactive network module provided by the first embodiment of the present invention;

[0022] Figure 4 It is a module schematic diagram of an electric vehicle arc fault detection device based on a multi-dimensional network provided by the second embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0024] refer to Figure 1 As shown, the first embodiment of the present invention discloses an electric vehicle arc fault detection method based on a multi-dimensional network, which can be executed by an electric vehicle arc fault detection device based on a multi-dimensional network (hereinafter referred to as a detection device), and in particular, executed by one or more processors in the detection device to implement the following method:

[0025] S1, acquiring current time series data of an arc fault collected by a collection device, and converting the current time series data to obtain two-dimensional image current data X;

[0026] Specifically, step S1 includes: acquiring current time series data of arc faults of electric vehicles under different working conditions collected by a preset collection device;

[0027] The GADF algorithm is used to encode and transform the current time series data to obtain two-dimensional image current data X. , where R is the multidimensional tensor of the two-dimensional image current data, C is the number of channels of the two-dimensional image current data, H is the number of pixels of the two-dimensional image current data in the vertical direction, and W is the number of pixels of the two-dimensional image current data in the horizontal direction.

[0028] In this embodiment, the current time series data of arc faults under different working conditions of electric vehicles are collected by using a preset collection device. These data are the basis of arc fault characteristics and are crucial for subsequent fault detection. The collection device can be a current sensor, which is installed in the battery management system of the electric vehicle to monitor the change of current in real time.

[0029] Subsequently, the GADF algorithm is used to encode and transform the collected current time series data. The GADF algorithm is an efficient data conversion method, which can convert one-dimensional time series data into two-dimensional image current data X. The advantage of this is that it can more intuitively display the changing trend of the current and provide richer information for subsequent feature extraction and fault diagnosis. In this embodiment, the parameter values ​​of the GADF algorithm can be adjusted according to the actual current data characteristics to obtain the best conversion effect. Such two-dimensional image current data X not only retains the integrity of the signal, but also maintains the signal's dependence on time, providing high-quality input data for the subsequent multidimensional neural network.

[0030] S2, using a multi-scale hierarchical information integration module to perform preliminary feature extraction processing on the two-dimensional image current data X to generate a hierarchical information integration result;

[0031] Specifically, step S2 includes: performing residual processing on the two-dimensional image current data X;

[0032] use The convolution layer performs convolution processing on the two-dimensional image current data X, and uses the Sig activation function to calculate the two-dimensional image current data X after the convolution processing to obtain a first extraction feature;

[0033] use The convolution layer performs convolution processing on the two-dimensional image current data X, and uses the tanh activation function to calculate the two-dimensional image current data X after the convolution processing to obtain a second extraction feature, wherein, Convolutional layers and Convolutional layers are parallelized;

[0034] The two-dimensional image current data X is calculated according to the sw result function, and the The convolution layer performs convolution processing on the calculated two-dimensional image current data X to obtain the third extracted feature;

[0035] The first extracted features, the second extracted features and the third extracted features are fused, and the fused features and the two-dimensional image current data X after residual processing are input into the global average pooling layer GAP to obtain a hierarchical information integration result.

[0036] In this embodiment, first, residual processing is performed on the converted two-dimensional image current data X. Residual processing is an effective technique that allows the network to learn the direct mapping relationship of the data while alleviating the gradient vanishing problem in deep network training. In this way, important information of the original data can be retained, laying a solid foundation for subsequent feature extraction. Next, two parallel convolutional layers are used to perform convolution processing on the two-dimensional image current data X. The first convolutional layer uses The convolution kernel of the second convolution layer uses The two convolution layers correspond to different receptive fields and can capture features of different scales. The processed two-dimensional image current data X is calculated by the Sig activation function and the th activation function to obtain the first extracted feature and the second extracted feature. The Sig activation function can limit the output between 0 and 1, which helps to control the output range of the model, while the th activation function limits the output between -1 and 1, which helps to enhance the expressiveness of the model. Among them, the benefit of parallel processing is that it can capture local and global features at the same time, which is crucial to understanding the complexity of arc faults. In addition, this parallel structure also improves the computational efficiency of the model because it allows multiple feature extraction tasks to be processed simultaneously.

[0037] Subsequently, the two-dimensional image current data X is calculated according to the sw result function. The sw result function is a conditional control function that can dynamically adjust the calculation process according to the characteristics of the data. The calculated data is then passed through a The convolution layer is convolved to obtain the third extracted feature. This step further refines the feature extraction process so that more refined features can be captured. Finally, the first extracted feature, the second extracted feature and the third extracted feature are fused. The fusion process is achieved through weighted summation, where the weights are automatically learned based on the characteristics of the data to ensure that the most important features receive more attention. The fused features are input into the global average pooling layer GAP together with the two-dimensional image current data X after residual processing. The global average pooling layer GAP can aggregate the features of the entire image to obtain a global feature representation, which helps the model capture global and abstract features, thereby improving the accuracy of fault detection.

[0038] Through the above steps, a hierarchical information integration result is obtained, which is the result of preliminary feature extraction processing of the two-dimensional image current data X. This result not only contains rich local and global arc features, but also retains the original information of the data, providing strong support for subsequent multi-branch feature extraction and fault diagnosis.

[0039] S3, using a parallel interactive network module to perform feature extraction preprocessing on the hierarchical information integration result to obtain an interactive result Y, wherein the parallel interactive network module includes a local arc feature and a global arc feature extraction branch network, a horizontal feature extraction branch network, and a vertical feature extraction branch network;

[0040] Specifically, step S3 includes: performing residual processing on the hierarchical information integration result;

[0041] The local arc feature and the global arc feature extraction branch network are used to extract the hierarchical information integration result, and the hierarchical information integration result after extraction is input into the average pooling layer for interaction to obtain the weight ;

[0042] The horizontal feature extraction branch network is used to extract and process the hierarchical information integration result to generate weights ;

[0043] The vertical feature extraction branch network is used to extract and process the hierarchical information integration result to generate weights ;

[0044] Weight and weight Multiply and fuse, and multiply and fuse the weights and weights , weight , weight , and the hierarchical information integration result after residual processing is input into the Re-w module, and the Re-w operation is performed to generate the interaction result Y. The calculation formula of the interaction result Y is: .

[0045] Preferably, the local arc feature and global arc feature extraction branch network is composed of a local arc feature extraction module and a global arc feature extraction module after arc convolution modulation fusion, wherein the local arc feature extraction module is composed of an arc feature convolution modulation module and a convolution block, and the global arc feature extraction module is composed of an unfold folding module, multiple arc key feature attention extraction modules, and a Fold folding module.

[0046] Preferably, the calculation formula of the arc feature convolution modulation module is: , , , where Z is the calculation result of the arc feature convolution modulation module, A is the similarity score matrix, V is the arc feature value vector, is the Hadamard product operator, The kernel size is The arc feature deep convolution modulation block, is the weight matrix of the linear layer of branch 1, is the weight matrix of the linear layer of branch 2.

[0047] Preferably, the horizontal feature extraction branch network is used to extract and process the hierarchical information integration result to generate weights , specifically:

[0048] The two-dimensional image current data X is input into the global maximum pooling layer along the horizontal direction of space to obtain the pooled two-dimensional image current data , the calculation formula is: , where Gmp represents the global maximum pooling layer;

[0049] The pooled two-dimensional image current data is sequentially Perform Co convolution processing, th activation function processing, and FC full connection layer processing to obtain the intermediate layer mapping result in the horizontal direction , the calculation formula is: ;

[0050] The intermediate layer mapping result in the horizontal direction is performed according to the gating mechanism Process and obtain weight .

[0051] Preferably, the vertical feature extraction branch network is used to extract and process the hierarchical information integration result to generate a weight , specifically:

[0052] The two-dimensional image current data X is input into the global maximum pooling layer along the vertical direction of the space to obtain the pooled two-dimensional image current data , the calculation formula is: , where Gmp represents the global maximum pooling layer;

[0053] The pooled two-dimensional image current data is sequentially Perform Co convolution processing, th activation function processing, and FC full connection layer processing to obtain the intermediate layer mapping result in the vertical direction , the calculation formula is: ;

[0054] The intermediate layer mapping result in the horizontal direction is performed according to the gating mechanism Process and obtain weight .

[0055] In this embodiment, residual processing is performed on the hierarchical information integration result; then the local arc feature and global arc feature extraction branch network are used to extract and process the hierarchical information integration result. This branch network consists of a local arc feature extraction module and a global arc feature extraction module after arc convolution modulation fusion. The local arc feature extraction module consists of an arc feature convolution modulation module and a convolution block, while the global arc feature extraction module consists of an unfold module, multiple arc key feature attention extraction modules, and a Fold module. This design enables the model to capture local details and global contextual information at the same time, which is crucial to understanding the complexity of arc faults.

[0056] In this embodiment, the horizontal feature extraction branch network and the vertical feature extraction branch network are used to extract and process the hierarchical information integration results to generate weights. The two branch networks input the two-dimensional image current data X into the global maximum pooling layer along the horizontal and vertical directions of the space, respectively, to obtain the pooled two-dimensional image current data. Then, the pooled data is processed by convolution, th activation function, and full connection layer to obtain the intermediate layer mapping result. These results are processed according to the gating mechanism to obtain weights. Finally, these weights are multiplied and fused, and the fused weights are input into the Re-w module together with the hierarchical information integration results after residual processing, and the Re-w operation is performed to generate the interaction result Y.

[0057] S4, adjusting and fusion preprocessing the interaction result Y to obtain an arc fault diagnosis result.

[0058] Specifically, step S4 includes: inputting the interaction result Y into Channel number adjustment and feature fusion processing are performed in the convolution layer;

[0059] The processed interaction result Y is input into the network layer to obtain the arc fault diagnosis result, wherein the network layer is composed of a global pooling layer and a linear layer.

[0060] In this embodiment, the interaction result Y is input into the convolution layer for channel number adjustment and feature fusion processing. The purpose of this step is to optimize the representation of features so that the model can focus more on information that is useful for arc fault diagnosis. By adjusting the number of channels, the dimension of the features can be controlled to avoid overfitting while retaining the most critical features. Feature fusion processing involves merging feature information from different sources to form a comprehensive feature representation, which helps to improve the model's ability to identify arc faults. Among them, the design of the convolution layer allows the model to learn the spatial relationship between features, which is crucial for understanding the complex patterns of arc faults. In this way, it can be ensured that the method can not only capture local arc features, but also understand how these features interact at the global level.

[0061] Continue to input the interaction result Y after the convolution layer processing into the network layer to obtain the arc fault diagnosis result, where the network layer consists of a global pooling layer and a linear layer. The role of the global pooling layer is to aggregate the information of the entire feature map into a vector of a fixed size, which helps to capture the global features and reduces the dependence on the input size. The linear layer is responsible for mapping the output of the global pooling layer to the final diagnosis result, which usually involves the prediction of the number of categories. The use of the global pooling layer improves the generalization ability of the algorithm because it no longer depends on a specific input size, but can extract the most representative information from the entire feature map. The linear layer provides a direct mapping from the feature space to the label space, thereby realizing the classification of arc faults.

[0062] In summary, the electric vehicle arc fault detection method based on a multidimensional network achieves comprehensive capture of arc fault characteristics and improves the accuracy of fault detection through the application of a multi-scale hierarchical information integration module and a parallel interactive network module; at the same time, the method saves more memory when processing high-resolution images and improves computational efficiency. In addition, the use of a global pooling layer improves the generalization ability of the model, enabling it to work stably under different working conditions and conditions, thereby enhancing the robustness of the model. In general, the present invention provides an efficient and accurate electric vehicle arc fault detection method, which is of great significance for improving the safety and reliability of electric vehicles.

[0063] See also Figure 4 The second embodiment of the present invention provides an electric vehicle arc fault detection device based on a multi-dimensional network, which includes:

[0064] The conversion unit 201 is used to obtain the current time series data of the arc fault collected by the collection device, and perform conversion processing on the current time series data to obtain two-dimensional image current data X;

[0065] A preliminary feature extraction unit 202, configured to perform preliminary feature extraction processing on the two-dimensional image current data X using a multi-scale hierarchical information integration module to generate a hierarchical information integration result;

[0066] A feature extraction unit 203 is used to perform feature extraction preprocessing on the hierarchical information integration result by using a parallel interactive network module to obtain an interactive result Y, wherein the parallel interactive network module includes a local arc feature and a global arc feature extraction branch network, a horizontal feature extraction branch network, and a vertical feature extraction branch network;

[0067] The result generating unit 204 is used to perform adjustment and fusion preprocessing on the interaction result Y to obtain an arc fault diagnosis result.

[0068] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for detecting arc faults in electric vehicles based on a multidimensional network, characterized in that: include: Acquire the current time series data of the arc fault collected by the collection device, and transform the current time series data to obtain two-dimensional image current data X; Using a multi-scale hierarchical information integration module to perform preliminary feature extraction processing on the two-dimensional image current data X to generate a hierarchical information integration result; A parallel interactive network module is used to perform feature extraction preprocessing on the hierarchical information integration result to obtain an interactive result Y, wherein the parallel interactive network module includes a local arc feature and a global arc feature extraction branch network, a horizontal direction feature extraction branch network, and a vertical direction feature extraction branch network; Performing adjustment and fusion preprocessing on the interaction result Y to obtain an arc fault diagnosis result; A multi-scale hierarchical information integration module is used to perform preliminary feature extraction processing on the two-dimensional image current data X to generate a hierarchical information integration result, specifically: Performing residual processing on the two-dimensional image current data X; use The convolution layer performs convolution processing on the two-dimensional image current data X, and uses the Sig activation function to calculate the two-dimensional image current data X after the convolution processing to obtain a first extraction feature; use The convolution layer performs convolution processing on the two-dimensional image current data X, and uses the th activation function to calculate the two-dimensional image current data X after the convolution processing to obtain a second extraction feature, wherein, Convolutional layers and Convolutional layers are parallelized; The two-dimensional image current data X is calculated according to the sw result function, and the The convolution layer performs convolution processing on the calculated two-dimensional image current data X to obtain the third extracted feature; The first extracted features, the second extracted features and the third extracted features are fused, and the fused features and the two-dimensional image current data X after residual processing are input into the global average pooling layer GAP to obtain a hierarchical information integration result.

2. The electric vehicle arc fault detection method based on multidimensional network according to claim 1 is characterized in that: The current time series data of the arc fault collected by the collection device is obtained, and the current time series data is converted and processed to obtain the two-dimensional image current data X, specifically: Acquire the current time series data of arc faults of electric vehicles under different working conditions collected by a preset collection device; The GADF algorithm is used to encode and transform the current time series data to obtain two-dimensional image current data X. , where R is the multidimensional tensor of the two-dimensional image current data, C is the number of channels of the two-dimensional image current data, H is the number of pixels of the two-dimensional image current data in the vertical direction, and W is the number of pixels of the two-dimensional image current data in the horizontal direction.

3. The electric vehicle arc fault detection method based on multidimensional network according to claim 2 is characterized in that: The parallel interactive network module is used to perform feature extraction preprocessing on the hierarchical information integration result to obtain the interactive result Y, which is specifically: Performing residual processing on the hierarchical information integration result; The local arc feature and the global arc feature extraction branch network are used to extract the hierarchical information integration result, and the hierarchical information integration result after extraction is input into the average pooling layer for interaction to obtain the weight ; The horizontal feature extraction branch network is used to extract and process the hierarchical information integration result to generate weights ; The vertical feature extraction branch network is used to extract and process the hierarchical information integration result to generate weights ; Weight and weight Multiply and fuse, and multiply and fuse the weights and weights , weight , weight , and the hierarchical information integration result after residual processing is input into the Re-w module, and the Re-w operation is performed to generate the interaction result Y. The calculation formula of the interaction result Y is: .

4. The electric vehicle arc fault detection method based on multidimensional network according to claim 3 is characterized in that: The local arc feature and global arc feature extraction branch network is composed of a local arc feature extraction module and a global arc feature extraction module after arc convolution modulation fusion, wherein the local arc feature extraction module is composed of an arc feature convolution modulation module and a convolution block, and the global arc feature extraction module is composed of an unfold folding module, multiple arc key feature attention extraction modules, and a Fold folding module.

5. The electric vehicle arc fault detection method based on multidimensional network according to claim 4 is characterized in that: The calculation formula of the arc feature convolution modulation module is: , , , where Z is the calculation result of the arc feature convolution modulation module, A is the similarity score matrix, V is the arc feature value vector, is the Hadamard product operator, D The kernel size is The arc feature deep convolution modulation block, is the weight matrix of the linear layer of branch 1, is the weight matrix of the linear layer of branch 2.

6. The electric vehicle arc fault detection method based on multidimensional network according to claim 3 is characterized in that: The horizontal feature extraction branch network is used to extract and process the hierarchical information integration result to generate weights , specifically: The two-dimensional image current data X is input into the global maximum pooling layer along the horizontal direction of space to obtain the pooled two-dimensional image current data , the calculation formula is: , where Gmp represents the global maximum pooling layer; The pooled two-dimensional image current data is sequentially Perform Co convolution processing, th activation function processing, and FC full connection layer processing to obtain the intermediate layer mapping result in the horizontal direction , the calculation formula is: ; The intermediate layer mapping result in the horizontal direction is performed according to the gating mechanism Process and obtain weight .

7. The electric vehicle arc fault detection method based on multi-dimensional network according to claim 3 is characterized in that: The vertical feature extraction branch network is used to extract and process the hierarchical information integration result to generate weights , specifically: The two-dimensional image current data X is input into the global maximum pooling layer along the vertical direction of the space to obtain the pooled two-dimensional image current data , the calculation formula is: , where Gmp represents the global maximum pooling layer; The pooled two-dimensional image current data is sequentially Perform Co convolution processing, th activation function processing, and FC full connection layer processing to obtain the intermediate layer mapping result in the vertical direction , the calculation formula is: ; The intermediate layer mapping result in the horizontal direction is performed according to the gating mechanism Process and obtain weight .

8. The electric vehicle arc fault detection method based on multi-dimensional network according to claim 1, characterized in that: The interaction result Y is adjusted and fused for preprocessing to obtain an arc fault diagnosis result, which is specifically: Input the interaction result Y into Channel number adjustment and feature fusion processing are performed in the convolution layer; The processed interaction result Y is input into the network layer to obtain the arc fault diagnosis result, wherein the network layer is composed of a global pooling layer and a linear layer.

9. An electric vehicle arc fault detection device based on a multidimensional network, characterized in that: include: A conversion unit, used to obtain the current time series data of the arc fault collected by the collection device, and perform conversion processing on the current time series data to obtain two-dimensional image current data X; A preliminary feature extraction unit, used to perform preliminary feature extraction processing on the two-dimensional image current data X using a multi-scale hierarchical information integration module to generate a hierarchical information integration result; A feature extraction unit, used for performing feature extraction preprocessing on the hierarchical information integration result by using a parallel interactive network module to obtain an interactive result Y, wherein the parallel interactive network module includes a local arc feature and a global arc feature extraction branch network, a horizontal direction feature extraction branch network, and a vertical direction feature extraction branch network; A result generating unit, used for adjusting and fusing the interaction result Y to obtain an arc fault diagnosis result; A multi-scale hierarchical information integration module is used to perform preliminary feature extraction processing on the two-dimensional image current data X to generate a hierarchical information integration result, specifically: Performing residual processing on the two-dimensional image current data X; use The convolution layer performs convolution processing on the two-dimensional image current data X, and uses the Sig activation function to calculate the two-dimensional image current data X after the convolution processing to obtain a first extraction feature; use The convolution layer performs convolution processing on the two-dimensional image current data X, and uses the th activation function to calculate the two-dimensional image current data X after the convolution processing to obtain a second extraction feature, wherein, Convolutional layers and Convolutional layers are parallelized; The two-dimensional image current data X is calculated according to the sw result function, and the The convolution layer performs convolution processing on the calculated two-dimensional image current data X to obtain the third extracted feature; The first extracted features, the second extracted features and the third extracted features are fused, and the fused features and the two-dimensional image current data X after residual processing are input into the global average pooling layer GAP to obtain a hierarchical information integration result.

Images