A power communication network physical path automatic planning system and method

Abstract

The present invention belongs to the technical field of electric power communication network systems. Addressing the serious gaps in automatic physical path planning for electric power communication networks and the inability of existing road network automatic planning technologies to be directly applied to physical path planning for electric power communication networks, the present invention proposes an automatic physical path planning system for electric power communication networks. The system extracts visual features from terrain and geographic images using a convolutional neural network, and spatial topological features from road network graphs using a graph convolutional neural network. The system effectively integrates and enhances these features through a convolutional attention mechanism to generate a physical path for the electric power communication network. The system also utilizes the calculated image loss in the physical path to continuously update model parameters until the optimal physical path for the electric power communication network is obtained. The physical path generated by the present invention has good planning accuracy and usability.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric power communication systems, and in particular relates to a system and method for automatic planning of physical paths in an electric power communication network. Background Art

[0002] Power communication networks are the communications backbone of the power grid, providing crucial support for the stability, efficiency, and reliability of modern power systems. These networks support a variety of critical functions, including energy management, equipment fault alarms, and video surveillance. Properly planned physical paths enable effective management and maintenance of power communication networks, ensuring rapid restoration in the event of emergencies. For example, when fires, landslides, or floods damage underground tunnels or utility poles carrying fiber-optic communication cables, new physical paths must be replanned and deployed to repair communication disruptions and restore network services.

[0003] Planning the physical path of power communication networks requires comprehensive consideration of multiple factors, including terrain variations and existing urban infrastructure (including buildings, roads, and bridges). Currently, this task relies primarily on manual labor, requiring the expertise and collaboration of power engineers, construction teams, and urban management experts. This process is not only labor-intensive and time-consuming, but also struggles to cope with the increasingly complex planning requirements of communication networks. Therefore, there is an urgent need for an automated and efficient physical path planning method to improve the management and maintenance efficiency of power communication networks.

[0004] However, automated planning for power and telecommunications networks remains a technological gap. Research has revealed that automated generation technology exists in the area of street networks, a field closely related to power and telecommunications network applications. Some research has developed deep learning-based models. For example, StreetGAN generates new street network layouts by learning from the characteristics of existing street networks. Another model, DeepStreet, predicts future street network expansion patterns within a specific area based on surrounding street networks. Furthermore, these technologies have been expanded to incorporate diverse geographic information to generate more realistic predictions.

[0005] While these techniques have achieved significant success in the field of street network generation, they typically treat street networks as image data, rely on extracting visual features from images, and treat the approximate network generation task as an image inpainting task. However, in practical applications, physical path planning for power communication networks is highly dependent on the spatial layout and topological structure of the street network. This fundamental network information is typically encoded in the form of graph data, which existing methods cannot directly process. Therefore, these successful street network generation techniques cannot be directly applied to physical path planning for power communication networks. Summary of the Invention

[0006] The purpose of the present invention is to overcome the deficiencies of the prior art and to provide a system and method for automatically planning physical paths in a power communication network.

[0007] In this paper, we establish a hybrid model (AutoEPCNRoute) that combines convolutional neural networks (CNN) and graph convolutional neural networks (GCN) to predict and plan the physical path of the power communication network.

[0008] A first object of the present invention is to provide an automatic planning system for a physical path of an electric power communication network, comprising:

[0009] A convolutional neural network branch is used to extract multi-scale visual features from input images of terrain and geographic information, providing environmental perception information for physical path planning in the power communication network;

[0010] The graph convolutional neural network branch is used to extract spatial topological features from the input road network graph that is closely related to the power communication network, capturing the spatial correlation between the road network and the power communication network;

[0011] The feature fusion and enhancement module is used to fuse and enhance the extracted visual features and spatial topological features to generate the physical path of the power communication network in the area to be planned;

[0012] A discriminator is used to evaluate the generated physical path of the power communication network, generate an evaluation result, and dynamically adjust and update the parameters of the convolutional neural network branch and the parameters of the graph convolutional neural network branch according to the image loss value obtained by the loss calculation unit until the image loss value converges to obtain the optimal physical path of the power communication network;

[0013] Among them, the loss calculation unit is used to compare and calculate the generated evaluation result with the real path data to obtain the image loss value.

[0014] Preferably, the convolutional neural network branch includes:

[0015] The auxiliary information pre-extraction module is used to convert the input terrain and geographic information images into feature maps with resolutions of 256x256, 64x64, and 16x16 respectively through the pooling layer. At the same time, the features of the feature maps with resolutions of 64x64 and 16x16 are gradually added to the feature map with resolution of 256x256, and all the features are fused to obtain a multi-scale feature map rich in multi-scale information.

[0016] The geographic merging module is used to multiply the obtained multi-scale feature map with the physical path image of the power communication network to obtain the power communication network feature map ;

[0017] The pattern learning module is used to encode the input terrain and geographic information images and the power communication network physical path images with masks to obtain high-order features and combine the high-order features with the power communication network feature map. Combined to obtain a coarse feature map , and then the coarse feature map Decode and extract multi-scale visual features.

[0018] Preferably, the pattern learning module includes:

[0019] The encoding and decoding module based on gated convolution is used to encode the input terrain and geographic information images and the physical path images of the power communication network with masks to obtain high-order features and to generate coarse feature maps. Decode and extract multi-scale visual features;

[0020] Dilated gated convolution module, used to combine the obtained high-order features with the power communication network feature map Combined to obtain a coarse feature map .

[0021] Preferably, the graph convolutional neural network branch includes:

[0022] A graph construction module is used to construct a local graph of the road network in the image sample area based on an input road network graph closely related to the power communication network for each image sample;

[0023] The ROI construction and initialization module is used to expand each road node in the constructed local map into a pixel region of interest and initialize the features of the pixel region to obtain the initialized local map;

[0024] The spatial feature extraction and updating module is used to update and extract the features in the initialized local graph to obtain spatial topological features.

[0025] Preferably, the spatial feature extraction and updating module includes:

[0026] The graph convolutional network module is used to update and extract features from the initialized local graph based on the graph convolutional neural network to obtain updated node information;

[0027] Feature mapping module, used to map the updated node information back to the coarse feature map according to the position coordinates In the above example, we can get the spatial topological features.

[0028] Preferably, the feature fusion and enhancement module includes:

[0029] A Unet-style learning module is used to learn visual features through the encoder to obtain feature maps at different scales. In the decoding stage, the decoder uses skip connections to splice feature maps of corresponding sizes to enhance feature information and obtain visual feature maps.

[0030] Convolutional attention module, used to obtain visual feature maps and power communication network feature maps The spatial topological features are integrated and enhanced to generate the physical path of the power communication network in the area to be planned.

[0031] Preferably, the convolutional attention module includes:

[0032] Channel attention module, which is used to adaptively assign weights to different channels based on visual feature maps, power communication network feature maps As well as the importance of each channel in the spatial topological features, enhance relevant features and suppress irrelevant or redundant features;

[0033] Spatial attention module for adaptively capturing visual feature maps and power communication network feature maps By calculating the weight distribution in the spatial dimension, the perception of key locations is enhanced, the feature expression is optimized, and the features are integrated to generate the physical path of the power communication network in the area to be planned.

[0034] A second object of the present invention is to provide a method for automatically planning a physical path in a power communication network, comprising the following steps:

[0035] S1. Extract multi-scale visual features from input terrain and geographic information images to provide environmental perception information for physical path planning of power communication networks;

[0036] S2, extracting spatial topological features from the input road network graph that is closely related to the power communication network, and capturing the spatial correlation between the road network and the power communication network;

[0037] S3, fusing and enhancing the extracted visual features and spatial topological features to generate a physical path for the power communication network in the area to be planned;

[0038] S4. Evaluate the generated physical path of the power communication network and generate an evaluation result;

[0039] S5. Compare and calculate the generated evaluation result with the real path data to obtain the image loss value;

[0040] S6. Dynamically adjust and update the parameters of the convolutional neural network branch and the graph convolutional neural network branch according to the obtained image loss value until the image loss value converges to obtain the optimal physical path of the power communication network.

[0041] Preferably, step S1 comprises the following steps:

[0042] S11, converting the input terrain and geographic information images into feature maps with resolutions of 256x256, 64x64, and 16x16 respectively through a pooling layer, and gradually adding features from the feature maps with resolutions of 64x64 and 16x16 to the generated feature map with resolution of 256x256 to fuse them, thereby obtaining a multi-scale feature map rich in multi-scale information;

[0043] S12, for performing a dot multiplication of the obtained multi-scale feature map and the physical path image of the power communication network to obtain a power communication network feature map ;

[0044] S13, encode the input terrain and geographic information image and the power communication network physical path image with mask to obtain high-order features, and combine the high-order features with the power communication network feature map Combined to obtain a coarse feature map , and then the coarse feature map Decode and extract multi-scale visual features.

[0045] Preferably, step S2 comprises the following steps:

[0046] S21. For each image sample, construct a local map of the road network within the image sample area based on an input road network map closely related to the power communication network;

[0047] S22, expanding each road node in the constructed local graph into a pixel region of interest, and initializing features of the pixel region to obtain an initialized local graph;

[0048] S23. Update and extract the features in the initialized local graph to obtain spatial topological features.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] The present invention combines the convolutional neural network branch and the graph convolutional neural network branch to extract not only visual features but also spatial topological features simultaneously. That is, the present invention can simultaneously process heterogeneous data in both image and graphic modes through the design of the above two branch structures, so that the extracted data can be highly relevant to the power communication network. The present invention fuses and enhances the extracted visual features and spatial topological features through a feature fusion and enhancement module, and continuously updates the network parameters using the calculated image loss value, so that the final generated physical path of the power communication network can be highly consistent with the real path. The simulation test data verifies the correctness of the path planned by the method of the present invention.

[0051] Therefore, the method provided by the present invention can significantly improve the planning accuracy and practicality of the physical path. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a diagram of the architecture of the automatic physical path planning system for the power communication network proposed in Example 1 of the present invention;

[0053] Figure 2 This is a workflow diagram of the function fusion and enhancement unit in Example 1 of the present invention;

[0054] Figure 3 Demonstration diagram of the dataset constructed for the present invention;

[0055] Figure 4 Qualitative comparison of example results generated using different models for the dataset; Figure 4 Middle (a) - Figure 4 In the middle, (j) are samples randomly selected from the dataset. DETAILED DESCRIPTION

[0056] The following is a summary of the embodiments of the present invention. Figures 1 to 4 The technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0057] Example 1:

[0058] like Figure 1As shown, an embodiment of the present invention provides an automatic physical path planning system for electric power communication networks (AutoEPCNRoute model). This system performs planning based on two types of input data. The first is image data, which includes seven channels of auxiliary urban information: city elevation, slope, aspect, hillshade at four azimuth angles, and urban land use data. It also includes masked electric power communication network image samples and images of expected guide nodes within the samples. The second is graph data, which is topological data obtained by encoding the nodes and edges of the urban road network. The AutoEPCNRoute model consists of multiple modules. First, a convolutional neural network branch extracts visual features from the image data, while a graph convolutional neural network branch extracts spatial topological features from the graph data. The extracted features are then integrated through a feature fusion and enhancement module to generate a planned electric power communication network within the masked area. Finally, a discriminator and a loss calculation unit evaluate the difference between the generated result and the actual result. This information is used to update the parameters of the convolutional neural network branch and the graph convolutional neural network branch, continuously optimizing planning performance until the optimal physical path for the electric power communication network is obtained.

[0059] The automatic physical path planning system for a power communication network provided by an embodiment of the present invention specifically includes:

[0060] Convolutional neural network (CNN) branch units are used to extract multi-scale visual features from input images of terrain and geographic information, while simultaneously learning the patterns of the power communication network to provide environmental perception information for physical path planning of the power communication network;

[0061] In an embodiment of the present invention, the convolutional neural network branch includes:

[0062] The auxiliary information pre-extraction module is used to convert the input terrain and geographic information images into feature maps with resolutions of 256x256, 64x64 and 16x16 respectively through the pooling layer, and gradually add the features of the feature maps with resolutions of 64x64 and 16x16 to the generated feature map with resolution of 256x256, and fuse all the features to obtain a multi-scale feature map rich in multi-scale information; the terrain and geographic information images It consists of image layers of elevation, slope, aspect, hill shadows, lakes, buildings and other urban infrastructure. The auxiliary information pre-extraction module in the embodiment of the present invention is designed based on HrNet, and uses parallel multi-resolution subnetworks and multi-scale fusion mechanisms for subnetwork connection and information exchange. Specifically, the module converts the input channels into feature maps of sizes 256x256, 64x64 and 16x16 through a pooling layer. It then incrementally adds and fuses features from lower resolution subnetworks to produce the final multi-scale feature map rich in multi-scale information. This structure effectively solves the data dispersion problem caused by sampling errors and can directly guide result generation. It retains the spatial representation of terrain and regional information, while learning additional focus areas from images of terrain and geographic information through feature fusion.

[0063] The geographic merging module is used to explore the correlation between the physical path of the power communication network and the auxiliary information. Specifically, it is used to multiply the obtained multi-scale feature map with the physical path image of the power communication network that does not need to be repaired and has a complete physical path, and obtain a power communication network feature map that can guide the generation of the power communication network. ;

[0064] The pattern learning module (power communication network pattern learning module) is used to encode the input terrain and geographic information images and the power communication network physical path images with masks (i.e., areas where power communication is actually interrupted and the power communication network physical path needs to be planned, i.e., areas to be planned), obtain high-order features, and compare the high-order features with the power communication network feature map. Combined to obtain a coarse feature map , and then the coarse feature map Decode and extract multi-scale visual features. This module mainly uses an Encoder-Decoder structure neural network to capture the features of the physical path image of the power communication network and generate preliminary results. The input channel of the pattern learning module in this embodiment is composed of the mask channel M mask Processed Network Context I context and the key guiding point P in the mask area guide composition.

[0065] In an embodiment of the present invention, the pattern learning module includes:

[0066] The encoding and decoding module based on gated convolution is used to encode the input images of terrain and geographic information and the power communication network images with masks to obtain high-order features and to generate coarse feature maps. Decoding is performed to extract multi-scale visual features; this module uses gated convolutional layers to learn the effective distribution within the channel, which is crucial for exploiting contextual details in rugged or sparsely populated areas.

[0067] Dilated gated convolution module, used to combine the obtained high-order features with the power communication network feature map Combined, further update to get rough features .

[0068] The graph convolutional neural network branch is used to extract spatial topological features from the input road network graph that is closely related to the power communication network, capturing the spatial correlation between the road network and the power communication network; the goal of this graph convolutional neural network branch is to utilize the rich spatial topological structure of the road network to enhance the feature representation of the power communication network.

[0069] In an embodiment of the present invention, the graph convolutional neural network branch includes:

[0070] The graph construction module is used to construct a local graph of the road network in the image sample area based on the input road network graph closely related to the power communication network for each image sample; the input is the node set V global and edge E global The global graph of the urban road network G obtained by encoding road At the same time, the image channel R is consistent with the resolution of the power communication network in the area to be studied. index As a guide, for each image sample, in R index Find the corresponding latitude and longitude range, find the existing road nodes within the range, and add them to the node set V of the local graph global At the same time, for each node, find all its roads in the region and add them to the edge set E global A local graph of the road network within the image sample area is constructed. To facilitate subsequent processing, the present invention converts the data into the graph data format defined in the Python geometri library, which defines the graph using node and edge sets, and uses the Euclidean distance between the two endpoints of each edge as the edge weight.

[0071] The ROI construction and initialization module is used to expand each road node in the constructed local map into a pixel region of interest (ROI) and initialize the features of the pixel region to obtain the initialized local map; for the initialization of the local map, the present invention uses the rough features obtained by the above pattern learning module Specifically, the position of each road node in the local graph is assigned from the corresponding coarse feature by the ROI align algorithm. Extract one Pixel-sized feature blocks are used as the initial features of nodes. The algorithm can ensure accurate spatial alignment between road nodes and feature maps and retain the correct spatial relationship. This accuracy is crucial for maintaining the integrity of the topological structure in the graph.

[0072] The spatial feature extraction and updating module is used to update and extract the features in the initialized local graph to obtain spatial topological features.

[0073] In an embodiment of the present invention, the spatial feature extraction and updating module includes:

[0074] The graph convolutional neural network module is used to update and extract the features in the initialized local graph according to the graph convolutional neural network to obtain updated node information. Given that the road network has a more complex and diverse structure than the power communication network, and the spatial influence of different road nodes is different, the present invention uses a graph attention network to dynamically identify the influence of adjacent nodes on the current node and optimize the feature aggregation process. This mechanism can improve key road structure information when node features are updated, while suppressing noise and irrelevant information. Specifically, GAT calculates the attention coefficient To measure the node v With neighbors u The influence between them is calculated as follows:

[0075] ;

[0076] in, W is a learnable weight matrix, is a weight vector with the same dimension as the node feature, || represents concatenation, Leaky Re Lu is the activation function, For nodes v The relevant characteristics of For nodes u related features.

[0077] Next, the attention coefficient is used to calculate the weighted average of the adjacent node features to update the node v The characteristics of are calculated as follows:

[0078] ;

[0079] in, is the attention coefficient, N ( v )express v The set of neighbor nodes of soft max is to normalize the attention coefficient to the attention weight, W is a learnable weight matrix, For nodes u The relevant characteristics of is the node updated according to the neighbor node characteristics v New features.

[0080] Feature mapping module, used to map the updated node information back to the coarse feature map according to the node position In the above example, we can get the spatial topological features.

[0081] The feature fusion and enhancement module is used to fuse and enhance the extracted visual features and spatial topological features to generate a physical path for the power communication network in the area to be planned. The architecture diagram of the feature fusion and enhancement module in the embodiment of the present invention is as follows: Figure 2 As shown, specifically including:

[0082] A Unet-style learning module is used to learn visual features through an encoder-decoder, generating feature maps at different scales. During the decoding phase, the decoder uses skip connections to concatenate feature maps of corresponding sizes to enhance feature information, resulting in a visual feature map. The final physical path of the power communication network is derived from this visual feature map. Because the extracted visual and spatial topological features contain both high-dimensional and low-dimensional information, directly processing the original image will result in suboptimal results. Therefore, this encoder-decoder structure enhances feature reconstruction capabilities through skip connections, preserving more original information and high-resolution details while significantly improving the quality of feature prediction and generation, ensuring the model's adaptability and accuracy in complex scenarios.

[0083] Convolutional attention module, used to fuse features from different sources, specifically visual feature maps, power communication network feature maps The proposed method fuses and enhances spatial topological features to generate a physical path for the power communication network in the planned area. Because these features originate from different sources and have different focuses, directly concatenating them can lead to semantic and spatial misalignment, making it difficult for the model to learn effective feature representations. To address this issue, the present invention employs a convolutional attention module (CBAM module). The CBAM module dynamically highlights relevant features while suppressing less important ones using its channel and spatial attention mechanisms. This approach mitigates misalignment and redundancy, ensuring effective feature fusion and reducing noise.

[0084] In an embodiment of the present invention, the convolutional attention module includes:

[0085] Channel attention module, which is used to adaptively assign weights to different channels based on visual feature maps, power communication network feature maps As well as the importance of each channel in the spatial topological features, enhance relevant features and suppress irrelevant or redundant features, thereby improving the feature expression ability between channels;

[0086] Spatial attention module for adaptively capturing visual feature maps and power communication network feature maps By calculating the weight distribution in the spatial dimension, the perception of key locations is enhanced, the feature expression is optimized, and the features are integrated to generate the physical path of the power communication network in the area to be planned.

[0087] In this embodiment of the present invention, the feature map output by the convolutional attention module is calculated as follows:

[0088] ;

[0089] in, M c represents the channel attention weight, M s represents the spatial attention weight, ⊙ represents element-wise multiplication, F concat It is the aggregated feature obtained by directly concatenating feature maps from different sources. F output is the feature map obtained after enhancement by the channel attention module.

[0090] The use of the CBAM module can effectively refine features, reduce noise and improve feature representation.

[0091] The discriminator is used to evaluate the generated physical path of the power communication network and generate an evaluation result in accordance with the general process of the generative adversarial network. It dynamically adjusts and updates the parameters of the convolutional neural network branch and the graph convolutional neural network branch according to the image loss value calculated by the loss calculation unit until the image loss value converges to obtain the optimal physical path of the power communication network; and continuously improves the accuracy and rationality of path generation.

[0092] The loss calculation unit is used to compare and calculate the generated evaluation results with the real path data (that is, data that does not require repair and has a complete physical path) to obtain the image loss value.

[0093] The overall loss function used in the loss calculation unit of the present invention is a weighted combination of three partial losses, specifically including reconstruction loss, reverse loss and structural similarity index loss. The overall loss function is expressed as:

[0094] ;

[0095] in, , and are the hyperparameters that control the importance of each component. Through multiple experimental adjustments, these parameters are set to , and , For the overall loss, L recon To rebuild losses; L adv To combat losses, L ssim is the structural similarity index loss;

[0096] Reconstruction loss ( L recon ) is mainly used to measure the generated image ( I result ) and the ground truth image ( I truth ). It is calculated using the mean absolute error (MAE), and the calculation formula is as follows:

[0097] ;

[0098] Where N is the total number of pixels in the image, I result,i and I truth,i are the pixel values of the generated result image and the actual real image respectively.

[0099] Adversarial Loss ( L adv ) evaluates the authenticity of the generated image through the Generative Adversarial Network (GAN). Its calculation formula is:

[0100] ;

[0101] in, I truth represents the real image, I gen Indicates the generated image, D (·) is the discriminator.

[0102] Structural similarity index loss ( L ssim ) is used to evaluate the generated image ( I result ) and actual real images ( I truth ), this loss function helps to ensure the structural integrity of the generated image. The structural similarity index loss is defined as:

[0103] ;

[0104] in, I result andI truth are the generated images and the real images respectively, SSIM is the calculation function of the structural similarity index.

[0105] Example 2:

[0106] An embodiment of the present invention provides a method for automatically planning a physical path in a power communication network, comprising the following steps:

[0107] S1. Extract multi-scale visual features from the input terrain and geographic information images to provide environmental perception information for the physical path planning of the power communication network. The specific steps include:

[0108] S11, converting the input terrain and geographic information images into feature maps with resolutions of 256x256, 64x64, and 16x16 respectively through a pooling layer, and gradually adding the features of the feature maps with resolutions of 64x64 and 16x16 to the generated feature map with resolution of 256x256, fusing all the features to obtain a multi-scale feature map rich in multi-scale information;

[0109] S12. Multiply the obtained multi-scale feature map with the physical path image of the power communication network to obtain the power communication network feature map ;

[0110] S13, encode the input terrain and geographic information image and the power communication network physical path image with mask to obtain high-order features, and combine the high-order features with the power communication network feature map Combined to obtain a coarse feature map , and then the coarse feature map Decode and extract multi-scale visual features;

[0111] S2. Extracting spatial topological features from the input road network graph that is closely related to the power communication network to capture the spatial correlation between the road network and the power communication network. Specifically, the following steps are included:

[0112] S21. For each image sample, construct a local map of the road network within the image sample area based on an input road network map closely related to the power communication network;

[0113] S22, expanding each road node in the constructed local graph into a pixel region of interest, and initializing features of the pixel region to obtain an initialized local graph;

[0114] S23. Update and extract the features in the initialized local graph to obtain spatial topological features.

[0115] S3, fusing and enhancing the extracted visual features and spatial topological features to generate a physical path for the power communication network in the area to be planned;

[0116] S4. Evaluate the generated physical path of the power communication network and generate an evaluation result;

[0117] S5. Compare and calculate the generated evaluation result with the real path data to obtain the image loss value;

[0118] S6. Based on the calculated image loss value, continuously update the parameters of the convolutional neural network branch and the graph convolutional neural network branch, and dynamically adjust and update the parameters of the convolutional neural network branch and the graph convolutional neural network branch until the image loss value converges to obtain the optimal physical path of the power communication network.

[0119] The following is a study on the performance of the automatic planning system for the physical path of the electric power communication network (AutoEPCNRoute) provided in Example 1 of the present invention.

[0120] 1. Experimental setup:

[0121] (1) Dataset construction:

[0122] The experimental dataset of this invention was created in cooperation with the State Grid Corporation of China (SGCC) for the specific needs of Xi'an, Shaanxi Province, China. The AutoEPCNRoute method proposed in this invention was evaluated using this self-made dataset. The structure of the dataset is as follows: Figure 3 shown.

[0123] from Figure 3 It can be seen that the dataset mainly includes the following contents:

[0124] Terrain information: Contains a digital elevation model (DEM) image with a resolution of 12.5 meters, a slope image, a point, and four shadow images at different angles.

[0125] Geographic information: images of urban land use distribution including buildings, lakes, and other urban infrastructure.

[0126] Power communication network information: including images of the physical paths of the urban power communication network and images of guide points marking intersections.

[0127] This information was synthesized into a high-dimensional, multi-channel image, which, along with graphical data on the road network structure within the corresponding area, formed a dataset. All images and graphics were geospatially aligned. During the sampling process, 594 samples were randomly cropped from the high-dimensional, multi-channel image, each measuring 256×256 pixels. A graph representing the road network was created for each sample using a graph construction method. Data augmentation techniques such as rotation and flipping were used to expand the sample size to 2,376 for model training and validation, fully evaluating the model's performance.

[0128] (2) Experimental conditions setting:

[0129] The present invention randomly divides the 2376 samples generated above into a training set and a validation set, of which 1900 samples are used for training and 476 samples are used for validation. During the training process, the Adam optimizer is used for parameter optimization. The training is carried out for a total of 50 epochs with an initial learning rate of , weight decay is also set to , with a batch size of 12. After the 10th epoch, the learning rate is decayed by a factor of 10. The entire training process is completed on a GTX 4090 GPU with 24GB of video memory, ensuring efficient computing performance and model training efficiency.

[0130] Currently, there is no existing model developed for the research task of the present invention. Therefore, the AutoEPCNRoute model proposed in Example 1 of the present invention is compared with some of the most advanced street network generation and image restoration solutions, which are closely related to the research of the present invention, including DeepStreet, Topo-aware, DeepFill, and AOT-GAN. At the same time, since the Topo-aware model has the same CNN branch as the AutoEPCNRoute model of the present invention, the present invention regards the Topo-aware model as the baseline model (Base) of the present invention, and conducts three ablation experiments, specifically by adding or removing the GCN branch, the Unet-style learning module and the CBAM module to evaluate the effectiveness of the method provided by the present invention. The three models corresponding to the three ablation experiments are:

[0131] Base + GCN-Branch: adds a GCN branch to the baseline model, fuses features from the graph structure through simple concatenation, and processes them through a basic encoder-decoder architecture;

[0132] Base + GCN branch + Unet-style learning module (Base+GCN-Branch+Unet-style): This variant modifies the encoder-decoder architecture to a Unet-style structure, while still maintaining simple splicing for feature fusion;

[0133] Base + GCN-Branch + CBAM module: In this variant, the basic encoder-decoder architecture is retained, but a CBAM module is added for feature fusion.

[0134] This paper evaluates the performance of generated physical paths for power communication networks from multiple perspectives. First, it uses mean absolute error (MAE) and mean square error (MSE) to measure pixel-level differences. It also uses the structural similarity index (SSIM) to assess image quality. Furthermore, it uses the Dice coefficient (DICE) and Jaccard index (JAC) to measure the degree of overlap between the generated physical paths and the real paths, thereby assessing the accuracy of the generated paths.

[0135] 2. Experimental results:

[0136] The evaluation results of different models are shown in Table 1 below. It can be seen from Table 1 that the AutoEPCNRoute model proposed in the present invention has significant performance. Compared with existing street network generation models, the AutoEPCNRoute model not only achieved the best results in MAE, SSIM, DICE and JAC, but also demonstrated significant performance improvement. In particular, compared with the second best performing model, the AutoEPCNRoute model improved by 5.4% in MAE, 4.5% in SSIM, 25.7% in DICE, and 38.9% in JAC. Although the performance of the AutoEPCNRoute model in MSE was slightly lower than the best score, the focus of the present invention is to generate coherent overall results rather than pixel-level reconstruction. Therefore, this result does not weaken the advantages of the method proposed in the present invention. On the other hand, the ablation experiment results show that the complete AutoEPCNRoute model has been optimized to achieve the best performance in generating physical paths for power communication networks.

[0137] Table 1 Evaluation results of different models:

[0138] ;

[0139] Figure 4The experimental results of the dataset applied to the AutoEPCNRoute model proposed in this invention are shown in the figure. The first column is the actual physical path, and the second column shows the mask area where the physical path needs to be planned to connect the breakpoints. The following columns show the results generated by different models. Figure 4 Middle (a) to Figure 4 It can be clearly seen in (j) that the physical path generated by the AutoEPCNRoute model provided by the present invention is highly consistent with the real path, effectively capturing the complexity of the structure and capable of generating a high-quality physical path for the power communication network.

[0140] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the claims and their equivalents.

Claims

1. An automatic physical path planning system for a power communication network, characterized in that: include: A convolutional neural network branch is used to extract multi-scale visual features from input images of terrain and geographic information, providing environmental perception information for physical path planning in the power communication network; The graph convolutional neural network branch is used to extract spatial topological features from the input road network graph that is closely related to the power communication network, capturing the spatial correlation between the road network and the power communication network; The feature fusion and enhancement module is used to fuse and enhance the extracted visual features and spatial topological features to generate the physical path of the power communication network in the area to be planned; A discriminator is used to evaluate the generated physical path of the power communication network, generate an evaluation result, and dynamically adjust and update the parameters of the convolutional neural network branch and the parameters of the graph convolutional neural network branch according to the image loss value obtained by the loss calculation unit until the image loss value converges to obtain the optimal physical path of the power communication network; The loss calculation unit is used to compare the generated evaluation result with the real path data to obtain the image loss value; The overall loss function used in the loss calculation unit is a weighted combination of three partial losses, specifically including reconstruction loss, adversarial loss, and structural similarity index loss. The overall loss function is expressed as: ； in, , and is a hyperparameter that controls the importance of each component, For the overall loss, To rebuild the losses, For adversarial loss, is the structural similarity index loss; The reconstruction loss is calculated as follows: ； Where N is the total number of pixels in the image, and are the pixel values of the generated result image and the actual real image respectively; The adversarial loss is calculated as: ； in, represents the real image, Indicates the generated image, is the discriminator; The structural similarity index loss is defined as: ； in, and are the generated images and the real images respectively, SSIM is the calculation function of the structural similarity index.

2. The automatic physical path planning system for a power communication network according to claim 1, characterized in that: Convolutional neural network branches include: The auxiliary information pre-extraction module is used to convert the input terrain and geographic information images into feature maps with resolutions of 256x256, 64x64, and 16x16 respectively through the pooling layer. At the same time, the features of the feature maps with resolutions of 64x64 and 16x16 are gradually added to the feature map with resolution of 256x256, and all the features are fused to obtain a multi-scale feature map rich in multi-scale information. The geographic merging module is used to multiply the obtained multi-scale feature map with the physical path image of the power communication network to obtain the power communication network feature map ; The pattern learning module is used to encode the input terrain and geographic information images and the power communication network physical path images with masks to obtain high-order features and combine the high-order features with the power communication network feature map. Combined to obtain a coarse feature map , and then the coarse feature map Decode and extract multi-scale visual features.

3. The automatic physical path planning system for a power communication network according to claim 2, characterized in that: The pattern learning modules include: The encoding and decoding module based on gated convolution is used to encode the input terrain and geographic information images and the physical path images of the power communication network with masks to obtain high-order features and to generate coarse feature maps. Decode and extract multi-scale visual features; Dilated gated convolution module, used to combine the obtained high-order features with the power communication network feature map Combined to obtain a coarse feature map .

4. The automatic physical path planning system for a power communication network according to claim 1, characterized in that: Graph convolutional neural network branches include: A graph construction module is used to construct a local graph of the road network in the image sample area based on an input road network graph closely related to the power communication network for each image sample; The ROI construction and initialization module is used to expand each road node in the constructed local map into a pixel region of interest and initialize the features of the pixel region to obtain the initialized local map; The spatial feature extraction and updating module is used to update and extract the features in the initialized local graph to obtain spatial topological features.

5. The automatic physical path planning system for a power communication network according to claim 4, characterized in that: The spatial feature extraction and update module includes: The graph convolutional network module is used to update and extract features from the initialized local graph based on the graph convolutional neural network to obtain updated node information; Feature mapping module, used to map the updated node information back to the coarse feature map according to the position coordinates In the above example, we can get the spatial topological features.

6. The automatic physical path planning system for a power communication network according to claim 2, characterized in that: The feature fusion and enhancement module includes: A Unet-style learning module is used to learn visual features through the encoder to obtain feature maps at different scales. In the decoding stage, the decoder uses skip connections to splice feature maps of corresponding sizes to enhance feature information and obtain visual feature maps. Convolutional attention module, used to obtain visual feature maps and power communication network feature maps The spatial topological features are integrated and enhanced to generate the physical path of the power communication network in the area to be planned.

7. The automatic physical path planning system for a power communication network according to claim 6, characterized in that: The convolutional attention module includes: Channel attention module, which is used to adaptively assign weights to different channels based on visual feature maps, power communication network feature maps As well as the importance of each channel in the spatial topological features, enhance relevant features and suppress irrelevant or redundant features; Spatial attention module for adaptively capturing visual feature maps and power communication network feature maps By calculating the weight distribution in the spatial dimension, the perception of key locations is enhanced, the feature expression is optimized, and the features are integrated to generate the physical path of the power communication network in the area to be planned.

8. A method for automatically planning physical paths in a power communication network, characterized in that: The method is implemented by using the automatic physical path planning system for the electric power communication network according to any one of claims 1 to 7, comprising the following steps: S1. Extract multi-scale visual features from input terrain and geographic information images to provide environmental perception information for physical path planning of power communication networks; S2, extracting spatial topological features from the input road network graph that is closely related to the power communication network, and capturing the spatial correlation between the road network and the power communication network; S3, fusing and enhancing the extracted visual features and spatial topological features to generate a physical path for the power communication network in the area to be planned; S4. Evaluate the generated physical path of the power communication network and generate an evaluation result; S5. Compare and calculate the generated evaluation result with the real path data to obtain the image loss value; S6. Dynamically adjust and update the parameters of the convolutional neural network branch and the graph convolutional neural network branch according to the obtained image loss value until the image loss value converges to obtain the optimal physical path of the power communication network.

9. The method for automatic planning of physical paths in a power communication network according to claim 8, characterized in that: Step S1 includes the following steps: S11, converting the input terrain and geographic information images into feature maps with resolutions of 256x256, 64x64, and 16x16 respectively through a pooling layer, and gradually adding features from the feature maps with resolutions of 64x64 and 16x16 to the generated feature map with resolution of 256x256 to fuse them, thereby obtaining a multi-scale feature map rich in multi-scale information; S12, for performing a dot multiplication of the obtained multi-scale feature map and the physical path image of the power communication network to obtain a power communication network feature map ; S13, encode the input terrain and geographic information image and the power communication network physical path image with mask to obtain high-order features, and combine the high-order features with the power communication network feature map Combined to obtain a coarse feature map , and then the coarse feature map Decode and extract multi-scale visual features.

10. The method for automatic planning of physical paths in a power communication network according to claim 8, characterized in that: Step S2 includes the following steps: S21. For each image sample, construct a local map of the road network within the image sample area based on an input road network map closely related to the power communication network; S22, expanding each road node in the constructed local graph into a pixel region of interest, and initializing features of the pixel region to obtain an initialized local graph; S23. Update and extract the features in the initialized local graph to obtain spatial topological features.

Images