Power transmission line reliability prediction method, system and equipment based on SENet and OffNet

Abstract

The invention discloses a power transmission line reliability prediction method, system and device based on SENet and EffNet, and the method comprises the steps: firstly obtaining the image data of a target region, constructing a neural network model of a power transmission line, and carrying out the reliability measurement and calculation through the neural network model; according to the method, multi-source data such as images, electricity and weather are integrated through a comprehensive data analysis method, multi-dimensional features are extracted and fused through a convolutional layer, a fusion module, a pooling layer and the like of a convolutional neural network, a comprehensive feature vector is formed, the vector is integrated through a full-connection layer, nonlinear transformation and normalization processing are combined, and the multi-dimensional features are obtained. The method enhances the training efficiency and stability of the model, adjusts the feature scale through the composite scaling layer, and finally outputs the reliability prediction of the power transmission line through the decision layer, can effectively reflect the real-time state of the power transmission line, and improves the accuracy of reliability evaluation. The problem that a traditional method neglects real-time environment and external factors and depends on historical experience is solved.

Description

Technical Field

[0001] The present invention relates to a transmission line reliability prediction method, a system and a device, which belong to the field of transmission line safety technology and in particular to a transmission line reliability prediction method, a system and a device based on SENet and EffNet. Background Art

[0002] Transmission lines are an important part of the power system, and their reliability directly affects the stable supply of the power system. Therefore, it is necessary to predict the reliability of transmission lines.

[0003] Transmission line reliability prediction is the process of evaluating and calculating the reliability of transmission lines in power systems. Its purpose is to determine the probability that transmission lines can operate normally and without failure within a certain period of time. At present, in the existing reliability prediction process, the failure rate of transmission lines is usually used as a prediction indicator of transmission line reliability. It relies on historical failure data and statistical analysis, but ignores the real-time environment and external factors, such as the impact of weather conditions on transmission line reliability. When a transmission line fails and is repaired, the corresponding cause of the failure is usually solved, and the probability of the same failure occurring repeatedly is small. Therefore, using the failure rate as the only indicator for transmission line reliability prediction cannot fully reflect the overall health status and reliability of the transmission line, resulting in low accuracy in transmission line reliability calculation. Therefore, there is an urgent need for a more accurate prediction method to more comprehensively reflect the reliability of transmission lines. Summary of the invention

[0004] The purpose of the present invention is to overcome the above-mentioned defects and problems existing in the prior art and to provide a method, system and device for predicting the reliability of a power transmission line based on SENet and EffNet with higher accuracy.

[0005] To achieve the above objectives, the technical solution of the present invention is: a transmission line reliability prediction method based on SENet and EffNet, comprising:

[0006] S1. Acquire image data of a target area of ​​a transmission line; the image data includes conductors, insulators, towers, line fittings, guy wires, and grounding devices;

[0007] S2. Construct a transmission line CNN model based on the Keras framework; the CNN model includes a convolution layer, a fusion module, a pooling layer, a fully connected layer, an activation function, a batch normalization layer, a compound scaling layer and a decision layer; the fusion module is obtained by fusing the CNN-SENet module with the EfficientNet module, the CNN-SENet module includes the SE module, and the EfficientNet module includes the MBConv module; the Keras framework includes a splicing layer;

[0008] S3, input the image data into the convolution layer to obtain the image feature map, and input the image feature map into the fusion module to fuse the SE module and the MBConv module, and output the enhanced features of the image feature map, and then input the enhanced features into the pooling layer to obtain the global features;

[0009] The steps of fusing the SE module and the MBConv module are as follows:

[0010] S31, input an image feature map, expand the initial number of channels of the image feature map through the MBConv module, and perform depthwise separable convolution to obtain a feature map after depthwise separable convolution;

[0011] S32, embedding the SE module into the MBConv module, and recalibrating and projecting the number of channels of the feature map after the depthwise separable convolution;

[0012] S33, performing point-by-point convolution on the output feature map after the projection convolution, and then performing residual connection on the feature map after the point-by-point convolution when the number of channels of the feature map after the point-by-point convolution is the same as the initial number of channels and the step size is 1, so as to realize the fusion of the SE module and the MBConv module;

[0013] S4, real-time acquisition of electrical parameters, environmental parameters and weather data of the transmission line and the target area, and splicing the electrical parameters and weather data through the splicing layer, and then inputting the spliced ​​data and the global features into the fully connected layer for fusion to obtain a comprehensive feature vector;

[0014] S5, performing nonlinear transformation on the comprehensive feature vector by the activation function to obtain the feature vector after activation processing, and then inputting it into the batch normalization layer to obtain the normalized feature vector, and then inputting the normalized feature vector into the composite scaling layer to adjust the scale of the feature vector to obtain the final feature vector;

[0015] S6. Input the final feature vector into the decision layer for prediction to obtain the reliability prediction result of the transmission line;

[0016] S7. Map the reliability prediction results to corresponding locations on the map, and display the reliability of the transmission line corresponding to the visual elements.

[0017] The step S32 specifically includes:

[0018] S321, performing global average pooling on the feature map after the depth-wise separable convolution to obtain the first global feature of each channel;

[0019] S322, processing the first global feature through the first fully connected layer, the second fully connected layer and the activation function to generate a channel attention weight;

[0020] The expression of the first fully connected layer is as follows:

[0021] X se1 =ReLU(Dense(X gap , W se1 ));

[0022] Where: X se1 is the feature processed by the first fully connected layer Dense and ReLU activation function, X gap is the first global feature, W se1 is the weight of the first connection layer;

[0023] The expression of the second fully connected layer is as follows:

[0024] X se2 =σ(Dense(X se1 , W se2 ));

[0025] Where: X se2 is the channel attention weight after processing by the second fully connected layer Dense and σ activation function, W se2 is the weight of the second connection layer;

[0026] S323, multiply the generated channel attention weight by the feature map after the depth-separable convolution, recalibrate the number of channels, and obtain the recalibrated feature map, which is expressed as follows:

[0027] X′=X⊙X se2 ;

[0028] Where: X′ is the recalibrated feature map, X is the feature map after depth-wise separable convolution, and ⊙ represents the element-by-element multiplication operation;

[0029] S324, projecting the number of channels of the recalibrated feature map to the initial number of channels to complete the projection convolution, which is expressed as follows:

[0030] Y = W·X′;

[0031] Among them: Y is the output feature map after projection convolution, and W is the weight matrix.

[0032] The step S6 specifically includes:

[0033] S61, inputting the final feature vector into the decision layer to minimize the objective function in the decision layer, and determining the weight vector and bias term of the CNN model according to the constraint conditions of the objective function when the objective function is minimized;

[0034] The expression of the objective function is as follows:

[0035]

[0036] Where: w is the weight vector, ||W|| 2 is the coefficient of the weight vector, C is the penalty parameter, ξ i is the slack variable, n is the number of samples;

[0037] The constraint condition is expressed as follows:

[0038] y i (w·x i +b)≥1-ξ i ,ξ i ≥0, i=1, 2, ..., n;

[0039] Among them: w is the weight vector, y i is the label of sample i, x i is the i-th final eigenvector, b is the bias term;

[0040] S62, calculating a decision function in a decision layer based on the weight vector and the bias term to obtain a transmission line reliability prediction value;

[0041] The expression of the decision function is as follows:

[0042] y = f(w·x+b);

[0043] Among them: y is the reliability prediction value, f is the activation function, w is the weight vector, x is the input of the decision layer, and b is the bias term;

[0044] S63. The reliability prediction value is statistically converted into a probability distribution through a Softmax function to express the probabilities of different reliability levels.

[0045] The step S7 specifically includes:

[0046] S71, counting the reliability prediction values ​​of all target areas, and mapping the reliability prediction values ​​of the power transmission lines in all target areas to corresponding positions on the map through the GIS interface; each power transmission line in the map corresponds to a geographical path on the map;

[0047] S72. Based on different reliability prediction values, the transmission lines corresponding to the different reliability prediction values ​​are displayed with different visual elements.

[0048] The activation function is the Leaky ReLU function.

[0049] A transmission line reliability prediction system based on SENet and EffNet, the system comprising:

[0050] An image data acquisition module is used to acquire image data of a target area of ​​a transmission line; the image data includes conductors, insulators, towers, line fittings, guy wires, and grounding devices;

[0051] A transmission line CNN model construction module is used to construct a transmission line CNN model based on the Keras framework; the CNN model includes a convolution layer, a fusion module, a pooling layer, a fully connected layer, an activation function, a batch normalization layer, a compound scaling layer and a decision layer; the fusion module is obtained based on the fusion of the CNN-SENet module and the EfficientNet module, the CNN-SENet module includes the SE module, and the EfficientNet module includes the MBConv module; the Keras framework includes a splicing layer;

[0052] The feature enhancement module is used to input the image data into the convolution layer to obtain the image feature map, and input the image feature map into the fusion module to fuse the SE module and the MBConv module, and output the enhanced features of the image feature map, and then input the enhanced features into the pooling layer to obtain the global features;

[0053] The steps of fusing the SE module and the MBConv module are as follows:

[0054] S31, input an image feature map, expand the initial number of channels of the image feature map through the MBConv module, and perform depthwise separable convolution to obtain a feature map after depthwise separable convolution;

[0055] S32, embedding the SE module into the MBConv module, and recalibrating and projecting the number of channels of the feature map after the depthwise separable convolution;

[0056] S33, performing point-by-point convolution on the output feature map after the projection convolution, and then performing residual connection on the feature map after the point-by-point convolution when the number of channels of the feature map after the point-by-point convolution is the same as the initial number of channels and the step size is 1, so as to realize the fusion of the SE module and the MBConv module;

[0057] The feature fusion module is used to obtain the electrical parameters, environmental parameters and weather data of the transmission line in real time, and to splice the electrical parameters and weather data through the splicing layer. The spliced ​​data and the global features are then input into the fully connected layer for fusion to obtain a comprehensive feature vector.

[0058] The activation normalization module is used to perform nonlinear transformation on the comprehensive feature vector by the activation function to obtain the feature vector after activation processing, and then input it into the batch normalization layer to obtain the normalized feature vector, and then input the normalized feature vector into the composite scaling layer to adjust the scale of the feature vector to obtain the final feature vector;

[0059] The decision prediction module is used to input the final feature vector into the decision layer for prediction and obtain the transmission line reliability prediction result;

[0060] The result display module is used to map the reliability prediction results to the corresponding positions on the map and display the reliability of the transmission line with visual elements.

[0061] The feature enhancement module performs recalibration and projection convolution according to the following steps:

[0062] S321, performing global average pooling on the feature map after the depth-wise separable convolution to obtain the first global feature of each channel;

[0063] S322, processing the first global feature through the first fully connected layer, the second fully connected layer and the activation function to generate a channel attention weight;

[0064] The expression of the first fully connected layer is as follows:

[0065] X se1 =ReLU(Dense(X gap , W se1 ));

[0066] Where: X se1 is the feature processed by the first fully connected layer Dense and ReLU activation function, X gap is the first global feature, W se1 is the weight of the first connection layer;

[0067] The expression of the second fully connected layer is as follows:

[0068] X se2 =σ(Dense(X se1 , W se2 ));

[0069] Where: X se2 is the channel attention weight after processing by the second fully connected layer Dense and σ activation function, W se2 is the weight of the second connection layer;

[0070] S323, multiply the generated channel attention weight by the feature map after the depth-separable convolution, recalibrate the number of channels, and obtain the recalibrated feature map, which is expressed as follows:

[0071] X′=X⊙X se2 ;

[0072] Where: X′ is the recalibrated feature map, X is the feature map after depth-wise separable convolution, and ⊙ represents the element-by-element multiplication operation;

[0073] S324, projecting the number of channels of the recalibrated feature map to the initial number of channels to complete the projection convolution, which is expressed as follows:

[0074] Y = W·X′;

[0075] Among them: Y is the output feature map after projection convolution, and W is the weight matrix.

[0076] The decision prediction module performs reliability prediction of the transmission line according to the following steps:

[0077] S61, inputting the final feature vector into the decision layer to minimize the objective function in the decision layer, and determining the weight vector and bias term of the CNN model according to the constraint conditions of the objective function when the objective function is minimized;

[0078] The expression of the objective function is as follows:

[0079]

[0080] Where: w is the weight vector, ||W|| 2 is the coefficient of the weight vector, C is the penalty parameter, ξ i is the slack variable, n is the number of samples;

[0081] The constraint condition is expressed as follows:

[0082] y i (w·x i +b)≥1-ξ i ,ξ i ≥0, i=1, 2, ..., n;

[0083] Where: w is the weight vector, y i is the label of sample i, x i is the ith final eigenvector, and b is the bias term;

[0084] S62, calculating a decision function in a decision layer based on the weight vector and the bias term to obtain a transmission line reliability prediction value;

[0085] The expression of the decision function is as follows:

[0086] y = f(w·x+b);

[0087] Among them: y is the reliability prediction value, f is the activation function, w is the weight vector, x is the input of the decision layer, and b is the bias term;

[0088] S63. The reliability prediction value is statistically converted into a probability distribution through a Softmax function to express the probabilities of different reliability levels.

[0089] The result display module displays the reliability according to the following steps:

[0090] S71, counting the reliability prediction values ​​of all target areas, and mapping the reliability prediction values ​​of the power transmission lines in all target areas to corresponding positions on the map through the GIS interface; each power transmission line in the map corresponds to a geographical path on the map;

[0091] S72. Based on different reliability prediction values, the transmission lines corresponding to the different reliability prediction values ​​are displayed with different visual elements.

[0092] A transmission line reliability prediction device based on SENet and EffNet, the device comprising a processor and a memory;

[0093] The memory is used to store computer program code and transmit the computer program code to the processor;

[0094] The processor is used to execute the above-mentioned power transmission line reliability prediction method based on SENet and EffNet according to the instructions in the computer program code.

[0095] Compared with the prior art, the present invention has the following beneficial effects:

[0096] In a transmission line reliability prediction method based on SENet and EffNet of the present invention, the method first obtains the image data of the target area and constructs a neural network model of the transmission line, and then performs reliability prediction through the neural network model; in the application of this scheme, by comprehensively considering multiple data sources such as image data, electrical data and weather data, the actual operation status of the transmission line and the influence of the external environment are fully reflected, and multi-level features are extracted and fused through structures such as convolutional layers, fusion modules, and pooling layers to obtain richer and more meaningful feature representations, and image features, electrical data and weather data are fused through a fully connected layer to construct a comprehensive feature vector, which can effectively integrate the features of multi-source data to form a transmission line reliability prediction method. A comprehensive description of the state of the power line can capture complex feature relationships through nonlinear transformation, and normalize the features to reduce internal covariate offsets in the model training process and improve the training efficiency and stability of the model. In addition, the scale of the features is adjusted through the composite scaling layer, and the reliability prediction value of the transmission line is output through the decision layer. In summary, the present invention can more comprehensively and accurately evaluate the reliability of the transmission line by introducing multi-source data, multi-level feature extraction and fusion, nonlinear transformation and normalization processing, etc. Compared with the method that only relies on historical fault data and statistical analysis, it can better reflect the overall health status and real-time operating conditions of the transmission line, thereby improving the accuracy of the reliability prediction of the transmission line. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] Figure 1 It is a flow chart of the method steps of the present invention.

[0098] Figure 2 It is an architecture diagram of the CNN model of the present invention.

[0099] Figure 3 It is a schematic diagram of the system structure of the present invention.

[0100] Figure 4 It is a schematic diagram of the device structure of the present invention.

[0101] In the figure: image data acquisition module 1, transmission line CNN model construction module 2, feature enhancement module 3, feature fusion module 4, activation normalization module 5, decision prediction module 6, result display module 7, processor 8, memory 9, computer program code 91. DETAILED DESCRIPTION

[0102] The present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0103] Embodiment 1:

[0104] See also Figure 1, a transmission line reliability prediction method based on SENet and EffNet, comprising:

[0105] S1. Acquire image data of a target area of ​​a transmission line; the image data includes conductors, insulators, towers, line fittings, guy wires, and grounding devices;

[0106] In this embodiment, a high-resolution camera or an unmanned aerial vehicle is used to capture image data of the power transmission lines in the target area. The acquired image data contains various details of the power transmission lines, including conductors, insulators, towers, line fittings, guy wires, grounding devices, etc.

[0107] S2. Construct a CNN model of power transmission lines based on the Keras framework; see Figure 2 , the CNN model includes a convolution layer (preferably at least three convolution layers), a fusion module, a pooling layer, a fully connected layer, an activation function, a batch normalization layer, a compound scaling layer and a decision layer; the fusion module is obtained by fusing a CNN-SENet module with an EfficientNet module, the CNN-SENet module includes an SE module, and the EfficientNet module includes an MBConv module; the Keras framework includes a splicing layer;

[0108] S3, input the image data into the convolution layer to obtain the image feature map, and input the image feature map into the fusion module to fuse the SE module and the MBConv module, and output the enhanced features of the image feature map, and then input the enhanced features into the pooling layer to obtain the global features;

[0109] Furthermore, in this scheme, the image data is input into the convolution layer of the convolutional neural network (CNN), and the convolution layer extracts the features in the image through a series of convolution operations. The principle of the convolution operation is to slide the convolution kernel (filter) on the image and calculate the dot product between the convolution kernel and the local area of ​​the image, so as to extract the features of the area, and multiple convolution kernels can extract different features, such as edges, textures, shapes, etc. After multiple layers of convolution operations, the image features obtained gradually change from low-level features to high-level features, such as from edges and textures to objects and scenes, and then the high-level features are further processed and combined in subsequent layers to form a comprehensive understanding of the image.

[0110] Furthermore, the fusion module in this solution is obtained by fusing the CNN-SENet module and the EfficientNet module. The design purpose of the fusion module is to further enhance these features on the basis of extracting image features to make them more recognizable and representative. Specifically, the CNN-SENet module introduces an attention mechanism, which enhances the focus on important features by weighting the channels of the feature map, while the EfficientNet module improves the performance of the model while maintaining efficient calculation through a compound scaling strategy. After the image features are input into the fusion module, the enhanced features output by the fusion module include both the key features brought by the attention mechanism of the CNN-SENet module and the efficient features optimized by the EfficientNet module.

[0111] Furthermore, the main function of the pooling layer in this scheme is to reduce the dimension and compress the feature map while retaining important feature information. Through the pooling operation, the size of the feature map can be reduced, thereby reducing the computational complexity and enhancing the robustness of the model. In this embodiment, the pooling operation can be maximum pooling or average pooling. In maximum pooling, the maximum value in the pooling window is taken as the output, which can highlight the significant features in the feature map; in average pooling, the average value in the pooling window is taken as the output, which can smooth the noise in the feature map. After the enhanced features are input into the pooling layer, the global features obtained through the pooling operation not only retain the important information in the original feature map, but also reduce the size of the feature map. The global feature is an overall description of the input image, which contains important information of each part of the image; through the processing of the pooling layer, the global feature provides a concise and effective input for subsequent feature fusion and decision-making.

[0112] In this embodiment, the steps of fusing the SE module and the MBConv module are as follows:

[0113] S31, input an image feature map, expand the initial number of channels of the image feature map through the MBConv module, and perform depthwise separable convolution to obtain a feature map after depthwise separable convolution;

[0114] S32, embedding the SE module into the MBConv module, and recalibrating and projecting the number of channels of the feature map after the depthwise separable convolution;

[0115] Further, it specifically includes:

[0116] S321, performing global average pooling on the feature map after the depth-wise separable convolution to obtain the first global feature of each channel;

[0117] Furthermore, global average pooling is performed on the feature map after the depthwise separable convolution to obtain the first global feature of each channel. Global average pooling is a method that compresses the spatial dimension of the feature map into a single value. A global feature vector is generated by averaging all the pixels of each channel. Specifically, assume that the size of the feature map is H×W×C, where H represents the height, W represents the width, and C represents the number of channels. Global average pooling averages the H×W values ​​of each channel to obtain a vector of size 1×1×C, which can compress the spatial information of each channel into a global feature, retaining the overall information of each channel while reducing the amount of data, providing a concise input for subsequent fully connected layer processing.

[0118] S322, processing the first global feature through the first fully connected layer, the second fully connected layer and the activation function to generate a channel attention weight;

[0119] Furthermore, the first global feature is processed through the first fully connected layer, the second fully connected layer and the activation function to generate channel attention weights. First, the first fully connected layer compresses the dimension of the global feature vector from C to a smaller dimension r, where r is a hyperparameter and usually takes a smaller value to reduce the amount of calculation. Then, through the ReLU activation function, nonlinear features are added to enable the model to learn more complex features. Then, through the second fully connected layer, the dimension is restored from r to C, and through the σ (Sigmoid) activation function, the output value is limited to between 0 and 1 to generate channel attention weights. The channel attention weights reflect the importance of each channel. The larger the value, the higher the importance of the channel.

[0120] The expression of the first fully connected layer is as follows:

[0121] X se1 =ReLU(Dense(X gap , W se1 ));

[0122] Where: X se1 is the feature processed by the first fully connected layer Dense and ReLU activation function, X gap is the first global feature, W se1 is the weight of the first connection layer;

[0123] The expression of the second fully connected layer is as follows:

[0124] X se2 =σ(Dense(X se1 , W se2 ));

[0125] Where: X se2is the channel attention weight after processing by the second fully connected layer Dense and σ activation function, W se2 is the weight of the second connection layer;

[0126] S323, multiply the generated channel attention weight by the feature map after the depth-separable convolution, recalibrate the number of channels, and obtain the recalibrated feature map, which is expressed as follows:

[0127] X′=X⊙X se2 ;

[0128] Where: X′ is the recalibrated feature map, X is the feature map after depth-wise separable convolution, and ⊙ represents the element-by-element multiplication operation;

[0129] Furthermore, the generated channel attention weight is multiplied with the feature map after the depthwise separable convolution to recalibrate the number of channels. Specifically, the channel attention weight is a vector of size 1×1×C, which is multiplied with each channel of the feature map to adjust the eigenvalue of each channel. Assuming that the feature map is H×W×C and the channel attention weight is 1×1×C, the recalibrated feature map is: X′=X⊙X se2 ; Through the above method, important channels will be amplified and unimportant channels will be suppressed, thereby realizing the recalibration of the importance of feature map channels. The recalibrated feature map retains important feature information, enhances the feature expression ability of the model, and helps to improve the recognition accuracy of the model.

[0130] S324, projecting the number of channels of the recalibrated feature map to the initial number of channels to complete the projection convolution, which is expressed as follows:

[0131] Y = W·X′;

[0132] Among them: Y is the output feature map after projection convolution, and W is the weight matrix.

[0133] Furthermore, the number of channels of the recalibrated feature map is projected to the initial number of channels. The projection convolution is implemented through 1×1 convolution, and the number of channels of the recalibrated feature map is projected from C back to the initial number of channels C0. 1×1 convolution is an efficient convolution operation, which integrates and compresses the information of multiple channels by performing linear transformation on each pixel. Assuming that the recalibrated feature map is H×W×C and the initial number of channels is C0, the output of the projection convolution is H×W×C0, and the calculation formula of 1×1 convolution is Y=W·X′. Through this operation, the number of channels of the feature map is compressed to the initial number of channels, while retaining the important feature information after recalibration. The effect of projection convolution is to reduce the number of channels of the feature map and reduce the computational complexity, while maintaining the size and important features of the feature map, ensuring the computational efficiency and feature expression ability of the subsequent network layers.

[0134] In this embodiment, the global features of each channel are extracted through global average pooling, the channel attention weights are generated using the fully connected layer and activation function, the channels of the feature map are recalibrated and optimized, and finally the number of channels of the feature map is adjusted through projection convolution. This not only effectively enhances the expressive power of the feature map and enables the model to better focus on important features, but also optimizes the structure and amount of information of the feature map while maintaining computational efficiency, providing more valuable input for subsequent network layers.

[0135] S33, performing point-by-point convolution on the output feature map after the projection convolution, and then performing residual connection on the feature map after the point-by-point convolution when the number of channels of the feature map after the point-by-point convolution is the same as the initial number of channels and the step size is 1, so as to realize the fusion of the SE module and the MBConv module.

[0136] In this embodiment, the feature map is obtained through the convolution operation of the first few layers, which contains various feature information of the image. The initial number of channels of the feature map indicates the depth of the feature map, that is, how many independent feature channels are contained in the feature map. The input of the feature map is the basis of the entire fusion process. Subsequent operations will be processed and optimized based on these input feature maps. The process of inputting feature maps can be understood as taking the output of the previous layer of the network as the input of the current layer. These feature maps contain various feature information extracted by the previous layers of the network, such as edges, textures, etc.

[0137] The MBConv module is used to expand the initial number of channels of the feature map, and the feature map is subjected to a depthwise separable convolution. The MBConv module is the core component of EfficientNet. It increases the expressive power of the feature map by expanding the number of channels of the feature map. The depthwise separable convolution is an efficient convolution operation that decomposes the standard convolution into depthwise convolution and pointwise convolution, thereby greatly reducing the amount of calculation. The depthwise convolution performs convolution operations independently on each channel, while the pointwise convolution integrates the features of each channel through 1x1 convolution. After the depthwise separable convolution, the size and number of channels of the feature map are optimized, and more useful features are extracted. Specifically, the depthwise convolution processes each channel independently and retains the spatial features of each channel, while the pointwise convolution fuses the information of each channel through a 1x1 convolution operation, thereby further extracting high-level features. The advantage of the depthwise separable convolution is that it can extract richer feature information while maintaining computational efficiency.

[0138] After the depthwise separable convolution, the SE module is embedded in the MBConv module, and the number of channels of the feature map after the depthwise separable convolution is recalibrated and projected convolution is performed. The SE module recalibrates the number of channels of the feature map by calculating the importance weight of each channel, that is, adjusting the weight of each channel so that important feature channels are enhanced and unimportant feature channels are suppressed. The projection convolution is used to project the number of channels of the recalibrated feature map back to the initial number of channels, thereby maintaining the consistency of the size and number of channels of the feature map. The projection convolution is implemented through 1x1 convolution, which compresses and integrates the recalibrated feature map so that the number of channels of the feature map is consistent with the initial number of channels. The important features are enhanced through the recalibration of the SE module, and the number of channels of the feature map is guaranteed to remain unchanged through the projection convolution.

[0139] The feature map after projection convolution is convolved point by point. When the number of channels of the feature map after point by point convolution is the same as the initial number of channels and the step size is 1, the feature map after point by point convolution is residually connected to realize the fusion of MBConv module and SE module. The point by point convolution integrates the channels of the feature map through 1x1 convolution, so that the number of channels of the feature map is consistent with the initial number of channels. Residual connection is an effective feature fusion method. By adding the input feature map to the processed feature map, the information in the input feature map is retained, and the expression ability of the processed feature map is enhanced. Through the residual connection, the features of the MBConv module and the SE module are effectively fused. The output feature map contains both the rich features extracted by the deep separable convolution and the important features after the SE module is recalibrated, thereby improving the expression ability and recognition of the feature map. The advantage of residual connection is that it can alleviate the gradient vanishing problem, promote the training and convergence of the network, and ensure the effective transmission and fusion of feature maps between different modules.

[0140] The MBConv module and SE module of this embodiment are fused so that the output feature map contains both the rich features extracted by the deep separable convolution and the important features after recalibration by the SE module, thereby effectively improving the expressiveness and recognition of the feature map.

[0141] S4, real-time acquisition of electrical parameters, environmental parameters and weather data of the transmission line and the target area, and splicing the electrical parameters and weather data through the splicing layer, and then inputting the spliced ​​data and the global features into the fully connected layer for fusion to obtain a comprehensive feature vector;

[0142] Furthermore, electrical parameters and environmental parameters can be obtained in real time through sensors installed on transmission lines. Electrical parameters include current, voltage, etc., and environmental parameters include temperature, humidity, etc.; weather data can be obtained through meteorological stations or online weather services, including temperature, humidity, wind speed, rainfall, etc.

[0143] Sensor data and weather data are important external factors that affect the reliability of transmission lines. In this scheme, the above data are input into the fully connected layer together with the global features, and the data from different sources are fused through the linear transformation of the fully connected layer to obtain a comprehensive feature vector. The comprehensive feature vector contains comprehensive information of image features, sensor data and weather data, which can comprehensively reflect the status of the transmission line and the impact of the external environment. Through feature fusion, the prediction ability of the reliability of the transmission line is enhanced.

[0144] Furthermore, the CNN model in this scheme uses Keras to build and train the deep learning model. Keras includes a concatenation layer (Concatenate layer) to input the sensor data, weather data and global features into the fully connected layer to obtain a comprehensive feature vector. During operation, the sensor data and weather data are concatenated through the Concatenate layer, with the purpose of integrating data from different sources into an overall input. Specifically, the Concatenate layer is a layer in Keras that is used to concatenate multiple tensors along a specified axis. Assuming that the shape of the sensor data is (N, D1) and the shape of the weather data is (N, D2), where N is the number of samples, D1 and D2 are the feature dimensions of the sensor data and weather data, respectively, the Concatenate layer concatenates the two data on the feature dimension to obtain concatenated data with a shape of (N, D1+D2), and the specific implementation method can be implemented through the Concatenate layer of Keras. In this way, the concatenated data contains all the features of the sensor data and weather data, providing a complete input for subsequent feature fusion. This method can integrate data from different sources into a whole, retain their respective feature information, and facilitate subsequent processing and analysis.

[0145] In this solution, the concatenated data is fused with the global features through a fully connected layer to obtain a comprehensive feature vector. A fully connected layer (Dense layer) is a common neural network layer used to map input data to a new feature space.

[0146] In this embodiment, the shape of the global feature is (N, G), where G is the dimension of the global feature. During operation, the spliced ​​data is first processed through one or more fully connected layers to extract high-level features. The output shape of the fully connected layer is (N, H), where H is the output dimension of the fully connected layer. The processed data is then spliced ​​with the global feature, and then further processed through one or more fully connected layers to finally obtain a comprehensive feature vector. The specific implementation method can be implemented through the Dense layer and Concatenate layer of Keras. In this way, the comprehensive feature vector contains the fusion information of sensor data, weather data, and global features.

[0147] In this embodiment, features from different sources are fused to generate a comprehensive feature vector containing all important information, which is convenient for subsequent model training and prediction. Through this feature fusion, the model can better capture the relationship between data and improve the accuracy and robustness of prediction.

[0148] S5, performing nonlinear transformation on the comprehensive feature vector by the activation function to obtain the feature vector after activation processing, and then inputting it into the batch normalization layer to obtain the normalized feature vector, and then inputting the normalized feature vector into the composite scaling layer to adjust the scale of the feature vector to obtain the final feature vector;

[0149] Furthermore, in this scheme, the comprehensive feature vector is subjected to nonlinear transformation through an activation function to obtain a feature vector after activation processing. The role of the activation function is to introduce nonlinear factors so that the neural network can fit complex nonlinear relationships. In this embodiment, the activation function is preferably a Leaky ReLU function, which has the advantages of simple calculation and fast convergence. When the input of the Leaky ReLU function is a non-negative value, the output of the Leaky ReLU is the same as the input. When the input of the Leaky ReLU function is a negative value, the output of the Leaky ReLU is a positive number. The comprehensive feature vector is input into the activation function, and the linear relationship in the feature vector is broken through nonlinear transformation, so that the complex feature relationship can be better fitted. The feature vector after activation processing contains the feature information after nonlinear transformation, which enhances the expression and fitting ability of the model. By introducing nonlinear factors, the model can capture the complex relationship between features and improve the prediction accuracy of the reliability of the transmission line.

[0150] Furthermore, the main function of the batch normalization layer is to standardize the feature vectors so that the distribution of the feature vectors is more stable, thereby accelerating the model training process and improving the stability of the model. Specifically, the batch normalization layer normalizes the feature vectors by calculating the mean and standard deviation of the feature vectors so that the mean of the feature vectors is zero and the standard deviation is one. The normalized feature vectors have a more stable distribution, avoiding the adverse effects of too large or too small feature values ​​on model training. The feature vectors after activation processing are input into the batch normalization layer. Through normalization processing, the normalized feature vectors obtained have a more stable distribution, which is helpful for subsequent feature processing and decision-making, making the distribution of feature vectors more stable and improving the training efficiency and stability of the model.

[0151] Furthermore, the normalized feature vector is input into the compound scaling layer to obtain the final feature vector; the role of the compound scaling layer is to further adjust the feature vector so that the scale of the feature vector is more reasonable, thereby improving the prediction ability of the model. The compound scaling layer adjusts the scale of the feature vector by linearly transforming the feature vector so that each dimension of the feature vector has similar importance at the same scale. Specifically, the compound scaling layer scales and adjusts each dimension of the feature vector by the learned scaling factor so that each dimension of the feature vector has similar importance at the same scale. The normalized feature vector is input into the compound scaling layer. Through scaling and adjustment, the final feature vector obtained has a more reasonable scale, which is helpful for subsequent decision-making. Through scaling and adjustment, each dimension of the feature vector has similar importance at the same scale, which improves the prediction ability of the model.

[0152] S6. Input the final feature vector into the decision layer for prediction to obtain the transmission line reliability prediction result;

[0153] Furthermore, in this scheme, the final feature vector is input into the decision layer to obtain the reliability prediction value of the transmission line; the role of the decision layer is to process the final feature vector to obtain the reliability prediction value of the transmission line; the decision layer uses a classifier to obtain the prediction result by linearly transforming and nonlinearly transforming the final feature vector. Specifically, the decision layer calculates the score of each category by linearly transforming the final feature vector, and performs linear and nonlinear transformations through the activation function to obtain the final prediction value, which reflects the probability that the transmission line can operate normally and without failure within a certain period of time.

[0154] Further, it specifically includes:

[0155] S61, inputting the final feature vector into the decision layer to minimize the objective function in the decision layer, and determining the weight vector and bias term of the CNN model according to the constraint conditions of the objective function when the objective function is minimized;

[0156] The expression of the objective function is as follows:

[0157]

[0158] Where: w is the weight vector, ||W|| 2 is the coefficient of the weight vector, C is the penalty parameter, ξ i is the slack variable, n is the number of samples;

[0159] The constraint condition is expressed as follows:

[0160] y i (w·x i +b)≥1-ξi ,ξ i ≥0, i=1, 2, ..., n;

[0161] Where: w is the weight vector, y i is the label of sample i, x i is the ith final eigenvector, and b is the bias term;

[0162] Furthermore, in this scheme, the final feature vector is input into the decision layer to minimize the objective function in the decision layer. The decision layer is usually the last layer of the neural network and is responsible for generating the final prediction results. The objective function (loss function) is the function that needs to be optimized during the model training process. After the final feature vector is input into the decision layer, it is trained through the back propagation algorithm, the purpose of which is to adjust the parameters of the model (including weight vectors and bias terms) to minimize the value of the objective function. The back propagation algorithm calculates the gradient of the objective function relative to each parameter, gradually updates the parameters, and gradually converges the objective function to the minimum value. In this way, the weight vector and bias term of the model are continuously adjusted during the training process to minimize the value of the objective function. By optimizing the parameters of the model, the model can fit the training data more accurately and improve the accuracy of the prediction.

[0163] S62, calculating a decision function in a decision layer based on the weight vector and the bias term to obtain a transmission line reliability prediction value;

[0164] The expression of the decision function is as follows:

[0165] y = f(w·x+b);

[0166] Among them: y is the reliability prediction value, f is the activation function, W is the weight vector, x is the input of the decision layer, and b is the bias term;

[0167] Furthermore, the decision function in this scheme is a function that the model generates prediction results, which is usually a linear or nonlinear transformation. Assuming that the input of the decision layer is x, the weight vector is W, and the bias term is b, the decision function can be expressed as y=f(w·x+b). By calculating the decision function, the prediction value of the model can be obtained. In the task of transmission line reliability prediction, the prediction value represents the reliability score of the transmission line. In this way, the output of the model is the reliability prediction value of the transmission line.

[0168] S63. The reliability prediction value is statistically converted into a probability distribution through a Softmax function to express the probabilities of different reliability levels.

[0169] Furthermore, the prediction value of the model finally output by the decision function is the probability distribution calculated by the Softmax function. Through the Softmax function, the output of the model is converted into a probability distribution, with each category corresponding to a probability value. For the task of predicting the reliability of transmission lines, these probability values ​​can be expressed as probabilities of different reliability levels. For example, assuming that the model outputs three categories: high reliability, medium reliability, and low reliability, the output of the model may be [0.7, 0.2, 0.1], indicating that the transmission line has a 70% probability of high reliability, a 20% probability of medium reliability, and a 10% probability of low reliability.

[0170] This implementation generates a final prediction result and provides a quantitative assessment of the reliability of the transmission line. The reliability prediction value can be used for preventive maintenance and decision support to help improve the reliability and safety of the transmission line.

[0171] S7. Map the reliability prediction results to corresponding locations on the map, and display the reliability of the transmission line corresponding to the visual elements.

[0172] Further, it specifically includes:

[0173] S71, counting the reliability prediction values ​​of all target areas, and mapping the reliability prediction values ​​of the power transmission lines in all target areas to corresponding positions on the map through the GIS interface; each power transmission line in the map corresponds to a geographical path on the map;

[0174] S72. Based on different reliability prediction values, the transmission lines corresponding to the different reliability prediction values ​​are displayed with different visual elements.

[0175] Furthermore, in this scheme, the output results of the model are counted and sorted to form a data set containing the reliability prediction values ​​of all transmission lines in the target area, and then these reliability prediction values ​​are mapped to the corresponding positions on the map through the GIS (Geographic Information System) interface.

[0176] The GIS interface is a tool for processing and displaying geospatial data. It can combine geographic data with attribute data for visualization. Each transmission line corresponds to a geographic path on the map, and the reliability prediction values ​​of these paths are accurately mapped to the corresponding positions on the map. Complex prediction data can be visualized in geospatial form, allowing users to intuitively view and analyze the reliability of transmission lines in different regions.

[0177] In order to make the reliability information of power transmission lines on the map more intuitive and easy to understand, different colors, line thickness, transparency and other visual elements can be used to represent different reliability levels. For example, power transmission lines with high reliability can be represented by green, medium reliability by yellow, and low reliability by red. In addition, different reliability levels can be further distinguished by adjusting the thickness and transparency of the lines.

[0178] In this implementation, different visual elements are used to represent the reliability of the corresponding transmission lines, so that users can identify at a glance which transmission lines require special attention and maintenance. By distinguishing visual elements, the reliability information of the transmission lines is more intuitive and easy to understand, helping users to make decisions quickly and take corresponding measures.

[0179] Embodiment 2:

[0180] See also Figure 3 , a transmission line reliability prediction method system based on SENet and EffNet, the system comprising:

[0181] Image data acquisition module 1, used to acquire image data of a target area of ​​a transmission line; the image data includes conductors, insulators, towers, line fittings, guy wires, and grounding devices;

[0182] The power transmission line CNN model construction module 2 is used to construct a power transmission line CNN model based on the Keras framework; the CNN model includes a convolution layer, a fusion module, a pooling layer, a fully connected layer, an activation function, a batch normalization layer, a compound scaling layer and a decision layer; the fusion module is obtained based on the fusion of the CNN-SENet module and the EfficientNet module, the CNN-SENet module includes the SE module, and the EfficientNet module includes the MBConv module; the Keras framework includes a splicing layer;

[0183] Feature enhancement module 3, used to input image data into the convolution layer to obtain an image feature map, and input the image feature map into the fusion module to fuse the SE module and the MBConv module, and output enhanced features of the image feature map, and then input the enhanced features into the pooling layer to obtain global features;

[0184] The steps of fusing the SE module and the MBConv module are as follows:

[0185] S31, input an image feature map, expand the initial number of channels of the image feature map through the MBConv module, and perform depthwise separable convolution to obtain a feature map after depthwise separable convolution;

[0186] S32, embedding the SE module into the MBConv module, and recalibrating and projecting the number of channels of the feature map after the depthwise separable convolution;

[0187] Furthermore, the feature enhancement module 3 performs recalibration and projection convolution according to the following steps:

[0188] S321, performing global average pooling on the feature map after the depth-wise separable convolution to obtain the first global feature of each channel;

[0189] S322, processing the first global feature through the first fully connected layer, the second fully connected layer and the activation function to generate a channel attention weight;

[0190] The expression of the first fully connected layer is as follows:

[0191] X se1 =ReLU(Dense(X gap , W se1 ));

[0192] Where: X se1 is the feature processed by the first fully connected layer Dense and ReLU activation function, X gap is the first global feature, X se1 is the weight of the first connection layer;

[0193] The expression of the second fully connected layer is as follows:

[0194] X se2 =σ(Dense(X se1 , W se2 ));

[0195] Where: X se2 is the channel attention weight after processing by the second fully connected layer Dense and σ activation function, W se2 is the weight of the second connection layer;

[0196] S323, multiply the generated channel attention weight by the feature map after the depth-separable convolution, recalibrate the number of channels, and obtain the recalibrated feature map, which is expressed as follows:

[0197] X′=X⊙X se2 ;

[0198] Where: X′ is the recalibrated feature map, X is the feature map after depth-wise separable convolution, and ⊙ represents the element-by-element multiplication operation;

[0199] S324, projecting the number of channels of the recalibrated feature map to the initial number of channels to complete the projection convolution, which is expressed as follows:

[0200] Y = W·X′;

[0201] Among them: Y is the output feature map after projection convolution, and W is the weight matrix.

[0202] S33, performing point-by-point convolution on the output feature map after the projection convolution, and then performing residual connection on the feature map after the point-by-point convolution when the number of channels of the feature map after the point-by-point convolution is the same as the initial number of channels and the step size is 1, so as to realize the fusion of the SE module and the MBConv module;

[0203] Feature fusion module 4 is used to obtain the electrical parameters, environmental parameters and weather data of the transmission line in real time, and to splice the electrical parameters and weather data through the splicing layer, and then input the spliced ​​data and the global features into the fully connected layer for fusion to obtain a comprehensive feature vector;

[0204] The activation normalization module 5 is used to perform nonlinear transformation on the comprehensive feature vector by the activation function to obtain the feature vector after activation processing, and then input it to the batch normalization layer to obtain the normalized feature vector, and then input the normalized feature vector to the composite scaling layer to adjust the scale of the feature vector to obtain the final feature vector;

[0205] The decision prediction module 6 is used to input the final feature vector into the decision layer for prediction to obtain the transmission line reliability prediction result;

[0206] Furthermore, the decision prediction module 6 performs reliability prediction of the power transmission line according to the following steps:

[0207] S61, inputting the final feature vector into the decision layer to minimize the objective function in the decision layer, and determining the weight vector and bias term of the CNN model according to the constraint conditions of the objective function when the objective function is minimized;

[0208] The expression of the objective function is as follows:

[0209]

[0210] Where: w is the weight vector, ||W|| 2 is the coefficient of the weight vector, C is the penalty parameter, ξ i is the slack variable, n is the number of samples;

[0211] The constraint condition is expressed as follows:

[0212] y i (w·x i +b)≥1-ξ i ,ξ i ≥0, i=1, 2, ..., n;

[0213] Among them: w is the weight vector, y i is the label of sample i, x i is the i-th final eigenvector, b is the bias term;

[0214] S62, calculating a decision function in a decision layer based on the weight vector and the bias term to obtain a transmission line reliability prediction value;

[0215] The expression of the decision function is as follows:

[0216] y = f(w·x+b);

[0217] Among them: y is the reliability prediction value, f is the activation function, w is the weight vector, x is the input of the decision layer, and b is the bias term;

[0218] S63. The reliability prediction value is statistically converted into a probability distribution through a Softmax function to express the probabilities of different reliability levels.

[0219] The result display module 7 is used to map the reliability prediction results to corresponding positions on the map and display the reliability of the transmission line corresponding to the visual elements.

[0220] Furthermore, the result display module 7 displays the reliability according to the following steps:

[0221] S71, counting the reliability prediction values ​​of all target areas, and mapping the reliability prediction values ​​of the power transmission lines in all target areas to corresponding positions on the map through the GIS interface; each power transmission line in the map corresponds to a geographical path on the map;

[0222] S72. Based on different reliability prediction values, the transmission lines corresponding to the different reliability prediction values ​​are displayed with different visual elements.

[0223] Embodiment 3:

[0224] See also Figure 4 , a transmission line reliability prediction device based on SENet and EffNet, the device comprising a processor 8 and a memory 9;

[0225] The memory 9 is used to store computer program code 91 and transmit the computer program code 91 to the processor 8;

[0226] The processor 8 is used to implement the power transmission line reliability prediction method based on SENet and EffNet as described in Example 1 according to the instructions in the computer program code 91.

[0227] Generally speaking, the computer instructions for implementing the method of the present invention may be carried in any combination of one or more computer-readable storage media. Non-transitory computer-readable storage media may include any computer-readable media, except for the signal itself that is temporarily propagating.

[0228] The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EKROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, device, or device.

[0229] Computer program code for performing the operation of the present invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, SMalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages, in particular, Python suitable for neural network computing and platform frameworks based on TensorFlow, PyTorch, etc. can be used. The program code can be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer or to an external computer (for example, using an Internet service provider to connect via the Internet) through any type of network, including a local area network (LAN) or a wide area network (WAN).

[0230] The above-mentioned device and non-transitory computer-readable storage medium can refer to the specific description of a transmission line reliability prediction method based on SENet and EffNet and its beneficial effects, which will not be repeated here.

[0231] Although the embodiments of the present invention have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A transmission line reliability prediction method based on SENet and EffNet, characterized in that: include: S1. Acquire image data of a target area of ​​a transmission line; the image data includes conductors, insulators, towers, line fittings, guy wires, and grounding devices; S2. Construct a transmission line CNN model based on the Keras framework; the CNN model includes a convolution layer, a fusion module, a pooling layer, a fully connected layer, an activation function, a batch normalization layer, a compound scaling layer and a decision layer; the fusion module is obtained by fusing the CNN-SENet module with the EfficientNet module, the CNN-SENet module includes the SE module, and the EfficientNet module includes the MBConv module; the Keras framework includes a splicing layer; S3, input the image data into the convolution layer to obtain the image feature map, and input the image feature map into the fusion module to fuse the SE module and the MBConv module, and output the enhanced features of the image feature map, and then input the enhanced features into the pooling layer to obtain the global features; The steps of fusing the SE module and the MBConv module are as follows: S31, input an image feature map, expand the initial number of channels of the image feature map through the MBConv module, and perform depthwise separable convolution to obtain a feature map after depthwise separable convolution; S32, embedding the SE module into the MBConv module, and recalibrating and projecting the number of channels of the feature map after the depthwise separable convolution; S33, performing point-by-point convolution on the output feature map after the projection convolution, and then performing residual connection on the feature map after the point-by-point convolution when the number of channels of the feature map after the point-by-point convolution is the same as the initial number of channels and the step size is 1, so as to realize the fusion of the SE module and the MBConv module; S4, real-time acquisition of electrical parameters, environmental parameters and weather data of the transmission line and the target area, and splicing the electrical parameters and weather data through the splicing layer, and then inputting the spliced ​​data and the global features into the fully connected layer for fusion to obtain a comprehensive feature vector; S5, performing nonlinear transformation on the comprehensive feature vector by the activation function to obtain the feature vector after activation processing, and then inputting it into the batch normalization layer to obtain the normalized feature vector, and then inputting the normalized feature vector into the composite scaling layer to adjust the scale of the feature vector to obtain the final feature vector; S6. Input the final feature vector into the decision layer for prediction to obtain the reliability prediction result of the transmission line; S7. Map the reliability prediction results to corresponding locations on the map, and display the reliability of the transmission line corresponding to the visual elements.

2. The power transmission line reliability prediction method based on SENet and EffNet according to claim 1, characterized in that: The step S32 specifically includes: S321, performing global average pooling on the feature map after the depth-wise separable convolution to obtain the first global feature of each channel; S322, processing the first global feature through the first fully connected layer, the second fully connected layer and the activation function to generate a channel attention weight; The expression of the first fully connected layer is as follows: X se1 =ReLU(Dense(X gap ,W se1 )): Where: X se1 is the feature processed by the first fully connected layer Dense and ReLU activation function, X gap is the first global feature, W se1 is the weight of the first connection layer; The expression of the second fully connected layer is as follows: X se2 =σ(Dense(X se1 ,W se2 )); Where: X se2 is the channel attention weight after processing by the second fully connected layer Dense and σ activation function, W se2 is the weight of the second connection layer; S323, multiply the generated channel attention weight by the feature map after the depth-separable convolution, recalibrate the number of channels, and obtain the recalibrated feature map, which is expressed as follows: X′=X⊙X se2 ; Where: X′ is the recalibrated feature map, X is the feature map after depth-wise separable convolution, and ⊙ represents the element-by-element multiplication operation; S324, projecting the number of channels of the recalibrated feature map to the initial number of channels to complete the projection convolution, which is expressed as follows: Y = W·X′; Among them: Y is the output feature map after projection convolution, and W is the weight matrix.

3. The power transmission line reliability prediction method based on SENet and EffNet according to claim 1, characterized in that: The step S6 specifically includes: S61, inputting the final feature vector into the decision layer to minimize the objective function in the decision layer, and determining the weight vector and bias term of the CNN model according to the constraint conditions of the objective function when the objective function is minimized; The expression of the objective function is as follows: Where: W is the weight vector, ||W|| 2 is the coefficient of the weight vector, C is the penalty parameter, ξ i is the slack variable, n is the number of samples; The constraint condition is expressed as follows: y i (w x i +b)≥1-ξ i ,x i ≥0, i=1, 2,..., n; Where: w is the weight vector, y i is the label of sample i, x i is the i-th final eigenvector, b is the bias term; S62, calculating a decision function in a decision layer based on the weight vector and the bias term to obtain a transmission line reliability prediction value; The expression of the decision function is as follows: y = f(w·x+b); Among them: y is the reliability prediction value, f is the activation function, w is the weight vector, x is the input of the decision layer, and b is the bias term; S63. The reliability prediction value is statistically converted into a probability distribution through a Softmax function to express the probabilities of different reliability levels.

4. The power transmission line reliability prediction method based on SENet and EffNet according to claim 3 is characterized in that: The step S7 specifically includes: S71, counting the reliability prediction values ​​of all target areas, and mapping the reliability prediction values ​​of the power transmission lines in all target areas to corresponding positions on the map through the GIS interface; each power transmission line in the map corresponds to a geographical path on the map; S72. Based on different reliability prediction values, the transmission lines corresponding to the different reliability prediction values ​​are displayed with different visual elements.

5. The power transmission line reliability prediction method based on SENet and EffNet according to claim 1, characterized in that: The activation function is a Leaky ReLU function.

6. A transmission line reliability prediction system based on SENet and EffNet, characterized in that: The system comprises: An image data acquisition module (1) is used to acquire image data of a target area of ​​a power transmission line; the image data includes conductors, insulators, towers, line fittings, guy wires, and grounding devices; A power transmission line CNN model construction module (2), used for constructing a power transmission line CNN model based on a Keras framework; the CNN model includes a convolution layer, a fusion module, a pooling layer, a fully connected layer, an activation function, a batch normalization layer, a compound scaling layer and a decision layer; the fusion module is obtained by fusing a CNN-SENet module with an EfficientNet module, the CNN-SENet module includes an SE module, and the EfficientNet module includes an MBConv module; the Keras framework includes a splicing layer; A feature enhancement module (3) is used to input the image data into the convolution layer to obtain an image feature map, and input the image feature map into the fusion module to fuse the SE module and the MBConv module, and output the enhanced features of the image feature map, and then input the enhanced features into the pooling layer to obtain the global features; The steps of fusing the SE module and the MBConv module are as follows: S31, input an image feature map, expand the initial number of channels of the image feature map through the MBConv module, and perform depthwise separable convolution to obtain a feature map after depthwise separable convolution; S32, embedding the SE module into the MBConv module, and recalibrating and projecting the number of channels of the feature map after the depthwise separable convolution; S33, performing point-by-point convolution on the output feature map after the projection convolution, and then performing residual connection on the feature map after the point-by-point convolution when the number of channels of the feature map after the point-by-point convolution is the same as the initial number of channels and the step size is 1, so as to realize the fusion of the SE module and the MBConv module; A feature fusion module (4) is used to obtain electrical parameters, environmental parameters and weather data of the transmission line in real time, and to splice the electrical parameters and weather data through a splicing layer, and then input the spliced ​​data and the global features into a fully connected layer for fusion to obtain a comprehensive feature vector; An activation normalization module (5) is used to perform nonlinear transformation on the comprehensive feature vector by an activation function to obtain a feature vector after activation processing, and then input it into a batch normalization layer to obtain a normalized feature vector, and then input the normalized feature vector into a composite scaling layer to adjust the scale of the feature vector to obtain a final feature vector; A decision prediction module (6) is used to input the final feature vector into the decision layer for prediction to obtain a transmission line reliability prediction result; The result display module (7) is used to map the reliability prediction results to corresponding positions on the map and display the reliability of the power transmission line in accordance with visual elements.

7. The power transmission line reliability prediction system based on SENet and EffNet according to claim 6, characterized in that: The feature enhancement module (3) performs recalibration and projection convolution according to the following steps: S321, performing global average pooling on the feature map after the depth-wise separable convolution to obtain the first global feature of each channel; S322, processing the first global feature through the first fully connected layer, the second fully connected layer and the activation function to generate a channel attention weight; The expression of the first fully connected layer is as follows: X se1 =ReLU(Dense(X gap ,W se1 )): Where: X sel is the feature processed by the first fully connected layer Dense and ReLU activation function, X gap is the first global feature, W sel is the weight of the first connection layer; The expression of the second fully connected layer is as follows: X se2 =σ(Dense(X se1 ,W se2 )); Where: X se2 is the channel attention weight after processing by the second fully connected layer Dense and σ activation function, W se2 is the weight of the second connection layer; S323, multiply the generated channel attention weight by the feature map after the depth-separable convolution, recalibrate the number of channels, and obtain the recalibrated feature map, which is expressed as follows: X′=X⊙X se2 ; Where: X′ is the recalibrated feature map, X is the feature map after depth-wise separable convolution, and ⊙ represents the element-by-element multiplication operation; S324, projecting the number of channels of the recalibrated feature map to the initial number of channels to complete the projection convolution, which is expressed as follows: Y = W·X′; Among them: Y is the output feature map after projection convolution, and W is the weight matrix.

8. The power transmission line reliability prediction system based on SENet and EffNet according to claim 7, characterized in that: The decision prediction module (6) performs reliability prediction of the power transmission line according to the following steps: S61, inputting the final feature vector into the decision layer to minimize the objective function in the decision layer, and determining the weight vector and bias term of the CNN model according to the constraint conditions of the objective function when the objective function is minimized; The expression of the objective function is as follows: Where: W is the weight vector, ||W|| 2 is the coefficient of the weight vector, C is the penalty parameter, ξ i is the slack variable, n is the number of samples; The constraint condition is expressed as follows: y i (w x i +b)≥1-ξ i ,x i ≥0, i=1, 2,..., n; Where: w is the weight vector, y i is the label of sample i, x i is the i-th final eigenvector, b is the bias term; S62, calculating a decision function in a decision layer based on the weight vector and the bias term to obtain a transmission line reliability prediction value; The expression of the decision function is as follows: y = f(w·x+b); Among them: y is the reliability prediction value, f is the activation function, w is the weight vector, x is the input of the decision layer, and b is the bias term; S63. The reliability prediction value is statistically converted into a probability distribution through a Softmax function to express the probabilities of different reliability levels.

9. The power transmission line reliability prediction system based on SENet and EffNet according to claim 8, characterized in that: The result display module (7) displays the reliability according to the following steps: S71, counting the reliability prediction values ​​of all target areas, and mapping the reliability prediction values ​​of the power transmission lines in all target areas to corresponding positions on the map through the GIS interface; each power transmission line in the map corresponds to a geographical path on the map; S72. Based on different reliability prediction values, the transmission lines corresponding to the different reliability prediction values ​​are displayed with different visual elements.

10. A transmission line reliability prediction device based on SENet and EffNet, characterized in that: The device comprises a processor (8) and a memory (9); The memory (9) is used to store computer program code (91) and transmit the computer program code (91) to the processor (8); The processor (8) is used to execute the power transmission line reliability prediction method based on SENet and EffNet according to any one of claims 1 to 5 according to the instructions in the computer program code (91).

Images