CN-121997507-A - Optimal path planning method and system for communication line of multi-source heterogeneous data

CN121997507ACN 121997507 ACN121997507 ACN 121997507ACN-121997507-A

Abstract

The invention relates to a method and a system for planning an optimal path of a communication line of multi-source heterogeneous data, and belongs to the technical field of communication line planning. The method comprises the steps of constructing an undirected graph reflecting node adjacency relations by obtaining a building function classification thematic graph and communication node space data, extracting node topological features and space semantic features by using an environment perception encoder in a pre-training model, fusing and generating environment features, constructing a selected node sequence by taking a source node as a starting point in a planning process, extracting path history features by using a path memory encoder, fusing the environment features and the path features by using a low-rank attention module, outputting guide scores of candidate nodes by a path decision decoder, selecting a next-hop node dynamic updating sequence according to the scores, and continuously iterating until reaching a target node. By means of multi-source feature depth fusion and time sequence path memory, path redundancy and loop winding phenomena are reduced, and line planning efficiency and accuracy in a complex environment are improved.

Inventors

LIANG TIANZENG
Lai Qinling
ZHANG ZHOU
YAO KEJUN
CAO RONGLIN
SHAN WEI
LIN MIN
LI QI
Mai Jiandong
HUANG SHENCHENG
YAN JILING
LU CHENGWEN
HUANG XIAOYAN
LIN XINYI
DENG WENBIAO
JIANG GUOJUN
CHEN WEN

Assignees

厦门亿力吉奥信息科技有限公司
国网思极位置服务有限公司
国网福建省电力有限公司福州供电公司

Dates

Publication Date: 20260508
Application Date: 20260116

Claims (10)

1. The optimal path planning method for the communication line of the multi-source heterogeneous data is characterized by comprising the following steps of: preprocessing the building function classification thematic map, constructing an undirected map based on the communication node spatial data, and obtaining a source node and a target node according to a preset path planning task; Inputting the undirected graph and the preprocessed building function classification thematic graph into a pre-trained path planning model environment perception encoder to obtain node topological features and spatial semantic features, and outputting environment features after splicing; the method comprises the steps of taking a source node as an initial node, constructing a selected node sequence, inputting the selected node sequence into a path memory encoder of a pre-trained path planning model to obtain path characteristics, determining a plurality of candidate nodes to be selected according to the connection relation of end nodes in the selected node sequence in an undirected graph, inputting the environmental characteristics and the path characteristics into a characteristic fusion module to obtain fusion characteristics, inputting the fusion characteristics into a path decision decoder of the pre-trained path planning model, and outputting the prediction probability scores of the candidate nodes; Selecting a candidate node with the maximum predictive probability score as a next node according to the predictive probability score of the candidate node, adding the candidate node into the selected node sequence, inputting the selected node sequence into a pre-trained path planning model path memory encoder, and continuously iterating until reaching a target node, and outputting an optimal path of a communication line; Wherein the pre-trained path planning model comprises an environment-aware encoder, a path memory encoder and a path decision decoder.
2. The optimal path planning method for the communication line of the multi-source heterogeneous data according to claim 1, wherein the method inputs an undirected graph and a preprocessed building function classification thematic graph to a pre-trained path planning model environment perception encoder to obtain node topological features and spatial semantic features, specifically: Calculating the attention weights of the current node and the adjacent nodes according to Euclidean distance and dispersion by using a graph attention network, and carrying out preliminary aggregation on the adjacent nodes; Performing secondary convolution on the basis of primary aggregation by using a graph convolution network, and outputting node topological characteristics; Extracting high-level semantic information of the building function classification thematic map through multi-layer convolution operation and downsampling operation, and fusing features of different levels through jump connection to output space semantic features; And splicing the node topological features and the space semantic features, and performing activation function processing to obtain environmental features.
3. The optimal path planning method for a communication line of multi-source heterogeneous data according to claim 1, wherein the method uses a source node as an initial node, constructs a selected node sequence, and inputs the selected node sequence to a pre-trained path planning model path memory encoder to obtain path characteristics, specifically: the path memory encoder adopts a long-term and short-term memory network; Sequentially inputting node characteristics in the selected node sequence into a long-short-period memory network memory unit according to the access sequence, performing time sequence modeling on the sequence by utilizing a gating mechanism of the long-short-period memory network, updating the hidden state, and taking the final hidden state output by the long-short-period memory network as the path characteristics representing the path history information.
4. The method for planning an optimal path of a communication line of heterogeneous data according to claim 1, wherein the feature fusion module is a low-rank attention module.
5. The optimal path planning method for a communication line of heterogeneous multi-source data according to claim 4, wherein the method inputs the environmental characteristic and the path characteristic into a low-rank attention module to obtain a fusion characteristic, specifically: carrying out one-dimensional convolution operation on the environmental features and the path features respectively to generate inquiry, keys, values and output gating feature vectors corresponding to the environmental features and the path features; Respectively splicing the query, key, value and output gating feature vector corresponding to the environmental feature and the path feature to obtain the query, key, value and output gating feature vector of the fusion feature; performing activation function processing on the query and key feature vectors of the fusion features, adding bias items, and applying rotary position codes to obtain the query and key feature vectors with position information; Importance weighting is carried out on the key feature vector with the position information, and a weighted key feature vector of the fusion feature is obtained; The traditional attention matrix calculation is split into two low-rank matrices, and the linear characteristics output through an attention mechanism are calculated and expressed as follows by a formula: ; in the formula, Representing the linear characteristics of the output through the attention mechanism, Representing the query feature vector after the position encoding has been applied, Representing the key feature vector after the position encoding has been applied, Representing the weighted key feature vector after the position encoding is applied, Representing the height of the feature vector, The width of the feature vector is represented, Indicating the operation of the transpose, A value feature vector representing the fusion feature; linear features to be output via the attention mechanism And an output gating feature vector of the fused feature Performing element-by-element multiplication operation to obtain fusion characteristics with environment and path information, wherein the fusion characteristics are expressed as follows: ; in the formula, The characteristics of the fusion are represented and, Representing an operation of changing the shape of the array, Representing an element-by-element multiplication operation, An output gating feature vector representing the fused feature, Representing a fully connected operation.
6. The optimal path planning method for a communication line of multi-source heterogeneous data according to claim 1, wherein the method inputs the fusion characteristics to a pre-trained path planning model path decision decoder, and outputs a predictive probability score of each candidate node, specifically: The path decision decoder comprises a first full connection layer and a second full connection layer; inputting the fusion characteristics into a first full-connection layer for dimension reduction to obtain intermediate characteristics; splicing the intermediate features with the environmental features to obtain fusion features; Inputting the fusion characteristics into a second full-connection layer, and outputting guide values corresponding to adjacent nodes of the current node; and carrying out normalization processing on the guide value to obtain the prediction probability score of each candidate node.
7. The method for optimal path planning for a communication line of multi-source heterogeneous data according to claim 1, wherein the pre-trained path planning model is obtained by training the following loss function: the loss function includes a binary cross entropy loss and a shortest path ratio loss, wherein: the binary cross entropy loss is used for measuring the difference between the prediction probability distribution output by the model and the manually marked optimal path node label; The shortest path ratio loss is used for measuring the proportional relation between the complete predicted path length generated by the model and the manually marked optimal path length, and constraining the path redundancy; the total loss of the path planning model is a weighted sum of the shortest path ratio loss and the shortest path ratio loss.
8. A system for optimizing a path of a communication line of multi-source heterogeneous data, comprising the following modules: The preprocessing module is used for preprocessing the building function classification thematic map of the target area, constructing an undirected map based on the communication node spatial data and obtaining a source node and a target node according to a preset path planning task; The node prediction module is used for inputting the undirected graph and the preprocessed building function classification thematic graph into a pre-trained path planning model environment perception encoder to obtain node topological features and spatial semantic features, and outputting environment features after splicing; the method comprises the steps of taking a source node as an initial node, constructing a selected node sequence, inputting the selected node sequence into a path memory encoder of a pre-trained path planning model to obtain path characteristics, determining a plurality of candidate nodes to be selected according to the connection relation of end nodes in the selected node sequence in an undirected graph, inputting the environmental characteristics and the path characteristics into a characteristic fusion module to obtain fusion characteristics, inputting the fusion characteristics into a path decision decoder of the pre-trained path planning model, and outputting the prediction probability scores of the candidate nodes; And the path result searching module is used for selecting a candidate node with the maximum predictive probability score as a next node according to the predictive probability score of the candidate node, adding the candidate node into the selected node sequence, inputting the selected node sequence into the pre-trained path planning model path memory encoder, and continuously iterating until reaching a target node, and outputting the optimal path of the communication line.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any one of claims 1 to 7 when the program is executed by the processor.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the method according to any one of claims 1 to 7.

Description

Optimal path planning method and system for communication line of multi-source heterogeneous data Technical Field The invention relates to a method and a system for planning an optimal path of a communication line of multi-source heterogeneous data, and belongs to the technical field of communication line planning. Background Conventional power distribution communication path planning methods typically employ a random sampling based strategy. Such as Probabilistic Roadmaps (PRMs), rapid exploration of random trees (RRTs), and Extended Spatial Trees (ESTs), which explore feasible paths by randomly sampling in continuous space and constructing graphs or tree structures, avoiding complete discrete modeling of the environment. However, they still require a large number of collision detections or probability calculations at the sampling points, which are computationally expensive. And the search result is unstable due to randomness, so that the robustness and the overall optimality of the path are difficult to ensure especially in long-distance or complex geographic environments. With the development of the deep convolutional neural network, the path planning result is greatly improved. For example, based on deep reinforcement learning, in a fixed thematic scene, the scores selected by nodes of different paths are learned, and the node with the highest score is selected as a result. But this approach does not solve the problem of continuous prediction. And although the existing deep learning method has succeeded in specific tasks, topology dependence among nodes is not fully considered, and the capability of interaction between node attributes and geospatial features is also lacking. In areas where the building function types are significantly unevenly distributed, energy supply and load imbalances, such methods tend to have difficulty accurately capturing dynamic associations between local structures and global patterns, resulting in paths that bypass or contain feasible paths of redundant nodes that are not optimal paths. Therefore, a method and a system for intelligently planning an optimal path of a communication line by using multi-source heterogeneous data such as nodes, images and the like are needed. Disclosure of Invention In order to solve the problems in the prior art, the invention provides a method and a system for planning an optimal path of a communication line of multi-source heterogeneous data. The technical scheme of the invention is as follows: In one aspect, the present invention provides a method for planning an optimal path of a communication line of multi-source heterogeneous data, including the following steps: preprocessing the building function classification thematic map, constructing an undirected map based on the communication node spatial data, and obtaining a source node and a target node according to a preset path planning task; Inputting the undirected graph and the preprocessed building function classification thematic graph into a pre-trained path planning model environment perception encoder to obtain node topological features and spatial semantic features, and outputting environment features after splicing; the method comprises the steps of taking a source node as an initial node, constructing a selected node sequence, inputting the selected node sequence into a path memory encoder of a pre-trained path planning model to obtain path characteristics, determining a plurality of candidate nodes to be selected according to the connection relation of end nodes in the selected node sequence in an undirected graph, inputting the environmental characteristics and the path characteristics into a characteristic fusion module to obtain fusion characteristics, inputting the fusion characteristics into a path decision decoder of the pre-trained path planning model, and outputting the prediction probability scores of the candidate nodes; Selecting a candidate node with the maximum predictive probability score as a next node according to the predictive probability score of the candidate node, adding the candidate node into the selected node sequence, inputting the selected node sequence into a pre-trained path planning model path memory encoder, and continuously iterating until reaching a target node, and outputting an optimal path of a communication line; Wherein the pre-trained path planning model comprises an environment-aware encoder, a path memory encoder and a path decision decoder. Preferably, the method inputs the undirected graph and the preprocessed building function classification thematic graph to a pre-trained path planning model environment perception encoder to obtain node topological features and spatial semantic features, specifically: calculating the attention weights of the current node and the adjacent nodes according to Euclidean distance and dispersion by using a graph attention network, and aggregating the neighborhood characteristics; Extracting high-level semantic i