CN-121981499-A - Branch topology priority dynamic sequencing and collaborative allocation method and device based on graph neural network

CN121981499ACN 121981499 ACN121981499 ACN 121981499ACN-121981499-A

Abstract

The embodiment of the application provides a method and a device for dynamic sequencing and collaborative distribution of branch topology priority based on a graph neural network, which are used for assembling the connection relation between node attributes and branch segments into a branch topology directed graph through skeleton extraction and branch node geometric attribute calculation, so as to realize accurate expression of a tree space structure. And constructing a graph attention risk reasoning mechanism, carrying out weighted fusion on the fusion intrusion distance, the growth rate and the trunk involvement weight to obtain node risk scores, and generating a priority ranking table according to descending order. The method has the advantages that the double-constraint cluster task cooperative allocation is introduced, the real-time position and the load state of the clusters are combined to generate a task allocation table according to the near constraint and the load balance constraint, and the task allocation table is issued and executed, so that the defects of the traditional technology in the aspects of branch topology modeling, multi-factor risk sequencing, cluster cooperative scheduling and the like are effectively overcome, and technical guarantee is provided for the hidden danger disposal of the power transmission corridor tree obstacle.

Inventors

HU YARONG
ZHU HONG
Yang Xunfa
MA XIAOLIN
LIN XUMING
MA HONGXIA
YANG WENNA
LIU YANG
MA QIAN
WU YANHUA
HAN ZHONGFU

Assignees

国网甘肃省电力公司临夏供电公司

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (10)

1. The utility model provides a branch topology priority dynamic ordering and collaborative allocation method based on a graph neural network, which is characterized by comprising the following steps: Collecting point cloud data of a target tree in a power transmission corridor, performing skeleton extraction to obtain branch skeleton data, performing node identification on the branch skeleton data to obtain a branch node set and branch segment connection relation, calculating space coordinates, branch angles and branch diameter parameters for each node in the branch node set to obtain a node geometric attribute set, taking the node geometric attribute set as a graph node, and assembling the branch segment connection relation as a directed edge to obtain a branch topology directed graph; Inputting the branch topological directed graph into a pre-trained branch situation reasoning network, performing neighborhood aggregation on the node geometric attribute set through a graph attention mechanism to obtain a node characteristic vector set, performing weighted fusion on the node characteristic vector set, the intrusion distance parameter, the growth rate parameter and the trunk traction weight parameter to obtain a node risk score set, and arranging the node risk score set in a descending order of values to obtain a branch priority ranking table; Obtaining real-time positions and load states of currently available unmanned aerial vehicle clusters to obtain cluster resource data, performing matching operation on the cluster resource data and the branch priority ranking table according to nearby constraint and load balancing constraint to obtain a cluster task allocation table, and issuing the cluster task allocation table to each unmanned aerial vehicle to drive and execute branch pruning operation.
2. The method for dynamically sequencing and cooperatively distributing branch topology priorities based on a graph neural network according to claim 1, wherein the steps of collecting point cloud data of a target tree in a power transmission corridor, performing skeleton extraction to obtain branch skeleton data, performing node recognition on the branch skeleton data to obtain a connection relationship between a bifurcation node set and branch segments comprise the steps of: Scanning and collecting target trees in a power transmission corridor through a laser radar sensor carried on an unmanned aerial vehicle to obtain original point cloud data, performing noise reduction filtering and ground point elimination on the original point cloud data to obtain tree point cloud data, downsampling the tree point cloud data according to a preset voxel scale to obtain simplified point cloud data, and performing skeleton extraction based on Laplace shrinkage on the simplified point cloud data to obtain branch skeleton data; Traversing each skeleton point in the branch skeleton data, calculating the neighborhood connection number of each skeleton point to obtain a connection degree value set, marking the skeleton points with the values larger than a preset connection degree threshold in the connection degree value set as bifurcation nodes to obtain a bifurcation node set, extracting connection paths aiming at adjacent bifurcation node pairs in the bifurcation node set to obtain branch segment connection relations, and coding the branch segment connection relations into directed connection according to the space positions of a starting node and a terminating node.
3. The method for dynamically ordering and cooperatively distributing branch topology priorities based on a graph neural network according to claim 1, wherein the calculating spatial coordinates, bifurcation angles and branch diameter parameters for each node in the bifurcation node set to obtain a node geometrical attribute set, using the node geometrical attribute set as a graph node and the branch segment connection relationship as a directed edge, and assembling to obtain a branch topology directed graph comprises: Reading three-dimensional positions of nodes in branch skeleton data of a bifurcation node set to obtain space coordinates, calculating included angles between adjacent branch sections of the nodes to obtain bifurcation angles, extracting cross section contours along a neighborhood skeleton path of the nodes, calculating contour diameters to obtain branch diameter parameters, and carrying out association packaging on the space coordinates, the bifurcation angles and the branch diameter parameters according to node identification to obtain a node geometric attribute set; And taking each node in the node geometric attribute set as a graph node, taking the space coordinates, the bifurcation angle and the branch diameter parameters as node initial feature vectors, reading the connection relation of branch sections, encoding each branch section into a directed edge pointing to a child node from a parent node according to the growth direction, calculating the intrusion growth situation of the corresponding branch section for each directed edge to obtain an edge weight parameter, and assembling the graph node, the directed edge and the edge weight parameter to obtain a branch topology directed graph.
4. The method for dynamically sequencing and cooperatively distributing branch topology priorities based on a graph neural network according to claim 1, wherein the step of inputting the branch topology directed graph into a pre-trained branch situation inference network and performing neighborhood aggregation on the node geometrical attribute set through a graph attention mechanism to obtain a node feature vector set comprises the following steps: Loading a pre-trained branch situation reasoning network, reading a graph attention layer parameter set in the branch situation reasoning network, inputting initial feature vectors and edge weight parameters of each graph node in a branch topological directed graph into an attention calculation unit corresponding to the graph attention layer parameter set, and calculating attention coefficients between each graph node and neighborhood nodes to obtain node attention weight distribution; And carrying out weighted summation on neighborhood node characteristics of each graph node based on the node attention weight distribution to obtain neighborhood aggregation characteristics, splicing the neighborhood aggregation characteristics with initial characteristic vectors of each graph node, carrying out nonlinear activation transformation to obtain updated node characteristics, and carrying out neighborhood aggregation on the updated node characteristics according to a preset aggregation layer number iteration until convergence conditions are reached to obtain a node characteristic vector set.
5. The method for dynamically sorting and cooperatively distributing branch topology priorities based on a graph neural network according to claim 1, wherein the step of performing weighted fusion on the node feature vector set, the intrusion distance parameter, the growth rate parameter and the trunk involvement weight parameter to obtain a node risk score set, and sorting the node risk score set in descending order of values to obtain a branch priority sorting table comprises the steps of: Calculating the shortest distance between the space coordinates and the power transmission wires of each node in the node feature vector set to obtain an intrusion distance parameter, reading historical monitoring data, calculating the unit time elongation of corresponding branch segments of each node to obtain a growth rate parameter, tracing back each node to a trunk path along a branch topological directed graph, counting the number of the passed branch segments to obtain a trunk involvement weight parameter, and carrying out weighted summation on the intrusion distance parameter, the growth rate parameter and the trunk involvement weight parameter according to a preset fusion weight coefficient and the node feature vector set to obtain a node risk score set; And reading the node risk evaluation diversity, performing descending order sorting according to the score values to obtain an initial sorting sequence, performing secondary ascending order sorting according to the intrusion distance parameters for the nodes with the same score values in the initial sorting sequence to obtain a refined sorting sequence, and performing association packaging on node identifiers of all nodes in the refined sorting sequence, the corresponding score values and the priority sequence numbers to obtain a branch priority sorting table.
6. The method for dynamically ordering and cooperatively distributing branch topology priorities based on a graph neural network according to claim 1, wherein the obtaining real-time positions and load states of currently available unmanned aerial vehicle clusters to obtain cluster resource data, and performing matching operation on the cluster resource data and the branch priority ordering table according to a nearby constraint and a load balancing constraint to obtain a cluster task allocation table, comprises: Polling an airborne positioning module of each unmanned aerial vehicle in a currently available unmanned aerial vehicle cluster through a cluster communication link to obtain real-time position data, reading the battery residual capacity and the pruning tool mounting state of each unmanned aerial vehicle to obtain load state data, and carrying out association packaging on the real-time position data and the load state data according to unmanned aerial vehicle identifications to obtain cluster resource data; And reading the space coordinates of each node in the branch priority ranking table, calculating Euclidean distance with the real-time position data of each unmanned aerial vehicle in the cluster resource data to obtain a cluster node distance matrix, screening candidate unmanned aerial vehicle sets corresponding to each node according to nearby constraint based on the cluster node distance matrix to obtain an initial matching relationship, performing balanced adjustment on the task number distributed by each unmanned aerial vehicle according to load balancing constraint aiming at the initial matching relationship to obtain an balanced matching relationship, and carrying out association packaging on each unmanned aerial vehicle identifier, corresponding node identifier and operation sequence in the balanced matching relationship to obtain a cluster task distribution table.
7. The method for dynamically ordering and cooperatively allocating branch topology priorities based on a graph neural network according to claim 1, wherein said issuing the cluster task allocation table to each unmanned aerial vehicle driver to execute branch pruning operation comprises: Reading a group task allocation table, splitting according to unmanned aerial vehicle identifications to obtain independent task instruction sets of all unmanned aerial vehicles, packaging the independent task instruction sets into task issuing data packets according to a preset communication protocol, and sending the task issuing data packets to an onboard control unit of a corresponding unmanned aerial vehicle one by one through a group communication link, and receiving a transmission confirmation receipt to obtain an issuing completion state; And the airborne control units of all unmanned aerial vehicles analyze the received independent task instruction sets, extract the space coordinates of the target nodes according to the operation sequence to obtain a waypoint sequence, plan a flight path based on the waypoint sequence, drive the flight control module to execute waypoint navigation to obtain an arrival state, and trigger the airborne pruning executing mechanism to execute pruning actions on the target branches and record operation completion identifiers when the arrival state meets a positioning accuracy threshold condition.
8. A branch topology priority dynamic ordering and collaborative allocation device based on a graph neural network, the device comprising: The branch topology module is used for collecting point cloud data of a target tree in a power transmission corridor, performing skeleton extraction to obtain branch skeleton data, performing node identification on the branch skeleton data to obtain a branch node set and branch segment connection relation, calculating space coordinates, branch angles and branch diameter parameters for each node in the branch node set to obtain a node geometric attribute set, taking the node geometric attribute set as a graph node and taking the branch segment connection relation as a directed edge to assemble a branch topology directed graph; The priority ranking module is used for inputting the branch topological directed graph into a pre-trained branch situation reasoning network, performing neighborhood aggregation on the node geometric attribute set through a graph attention mechanism to obtain a node feature vector set, performing weighted fusion on the node feature vector set, the intrusion distance parameter, the growth rate parameter and the trunk traction weight parameter to obtain a node risk score set, and ranking the node risk score set according to a numerical descending order to obtain a branch priority ranking table; And the pruning operation module is used for acquiring the real-time position and the load state of the current available unmanned aerial vehicle cluster to obtain cluster resource data, executing matching operation on the cluster resource data and the branch priority ranking table according to the nearby constraint and the load balancing constraint to obtain a cluster task allocation table, and issuing the cluster task allocation table to each unmanned aerial vehicle to drive and execute branch pruning operation.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the graph neural network based branch topology priority dynamic ordering and co-allocation method of any one of claims 1 to 7 when the program is executed by the processor.
10. A computer readable storage medium having stored thereon a computer program, which when executed by a processor, implements the steps of the graph neural network based branch topology priority dynamic ordering and co-allocation method of any of claims 1 to 7.

Description

Branch topology priority dynamic sequencing and collaborative allocation method and device based on graph neural network Technical Field The application relates to the field of data processing, in particular to a branch topology priority dynamic ordering and collaborative allocation method and device based on a graph neural network. Background The existing branch pruning operation method has obvious defects. The traditional system has poor performance in the aspect of branch topological structure modeling, usually relies on manual inspection or a simple image recognition means to acquire tree form information, and cannot effectively utilize point cloud data skeleton extraction and bifurcation node geometric attribute construction directed graph topological expression, so that the description precision of the branch space structure and branch segment connection relation is insufficient, and the subsequent refined risk assessment is difficult to support. Furthermore, the prior art has a bottleneck in branch intrusion risk quantification. Most systems lack intelligent scoring mechanisms for fusing multidimensional risk factors such as intrusion distance, growth rate, trunk traction weight and the like, cannot perform self-adaptive aggregation on node neighborhood information by means of a graph attention mechanism, and cannot accurately sort dynamic threat situations of all branch nodes in a power transmission corridor, so that scientificity and instantaneity of pruning operation priority judgment are affected. The existing system has a technical short board in the aspect of unmanned aerial vehicle cluster task cooperative allocation. The scheduling capability of carrying out joint constraint matching on branch priority sequencing results and real-time positions and load states of clusters is lacking, so that optimal allocation of tasks cannot be realized under the constraint of nearby constraint and load balancing, and the overall operation efficiency and resource utilization rate of the unmanned aerial vehicle clusters are affected. The solution of these problems has important significance to promote intelligent perception and automatic handling ability of transmission corridor tree obstacle hidden danger. Disclosure of Invention Aiming at the problems in the prior art, the application provides a method and a device for dynamically sequencing and cooperatively distributing branch topology priorities based on a graph neural network, which can effectively solve the defects of the traditional technology in the aspects of branch topology modeling, multi-factor risk sequencing, cluster cooperative scheduling and the like and provide technical support for intelligent perception of power transmission corridor tree obstacle hidden danger and unmanned aerial vehicle automatic trimming operation. In order to solve at least one of the problems, the application provides the following technical scheme: in a first aspect, the present application provides a method for dynamically ordering and cooperatively allocating branch topology priorities based on a graph neural network, including: Collecting point cloud data of a target tree in a power transmission corridor, performing skeleton extraction to obtain branch skeleton data, performing node identification on the branch skeleton data to obtain a branch node set and branch segment connection relation, calculating space coordinates, branch angles and branch diameter parameters for each node in the branch node set to obtain a node geometric attribute set, taking the node geometric attribute set as a graph node, and assembling the branch segment connection relation as a directed edge to obtain a branch topology directed graph; Inputting the branch topological directed graph into a pre-trained branch situation reasoning network, performing neighborhood aggregation on the node geometric attribute set through a graph attention mechanism to obtain a node characteristic vector set, performing weighted fusion on the node characteristic vector set, the intrusion distance parameter, the growth rate parameter and the trunk traction weight parameter to obtain a node risk score set, and arranging the node risk score set in a descending order of values to obtain a branch priority ranking table; Obtaining real-time positions and load states of currently available unmanned aerial vehicle clusters to obtain cluster resource data, performing matching operation on the cluster resource data and the branch priority ranking table according to nearby constraint and load balancing constraint to obtain a cluster task allocation table, and issuing the cluster task allocation table to each unmanned aerial vehicle to drive and execute branch pruning operation. Further, the method further comprises the steps of scanning and collecting target trees in a power transmission corridor through a laser radar sensor carried on an unmanned aerial vehicle to obtain original point cloud data, performing noise redu