Unmanned aerial vehicle scheduling method and device, electronic equipment, storage medium and program product

Abstract

The invention provides an unmanned aerial vehicle scheduling method and device, electronic equipment, a storage medium and a program product, and relates to the technical field of unmanned aerial vehicles. Time features and spatial features are extracted from a first topological graph and a second topological graph through a stop point prediction model; according to the method, double topological graphs with different correlations of the same geographic reference system are innovatively superposed and synthesized into a fusion feature graph, so that a parking site selection result is obtained, and the parking site deployment efficiency and accuracy of the unmanned aerial vehicle are improved. According to the method, the first topological graph and the second topological graph are preprocessed to obtain the preprocessing features, so that the accuracy of the parking site selection result output by the parking site prediction model is improved.

Description

Technical Field

[0001] The present invention relates to the field of drone technology, and in particular to a drone scheduling method, device, electronic equipment, storage medium, and program product. Background Art

[0002] As global warming continues, extreme weather events are becoming more frequent, intense, and widespread, posing increasing risks and challenges to humanity. Emergency response tasks in scenarios such as earthquakes, geological disasters, floods, urban waterlogging, snowstorms, forest and grassland fires, fires in special urban settings, hazardous chemical accidents, mine (tunnel) accidents, and emergency life-saving efforts require drone rescue platforms capable of fulfilling the following missions: Focusing on the urgent needs of emergency rescue operations in extreme situations such as major earthquakes and disasters, the "three blackouts," and complex environments like high mountain valleys and forested areas, the platform addresses weaknesses in key core capabilities such as emergency command and communications and disaster reconnaissance. The platform aims to establish a drone service area with a two-hour radius to the disaster site, accelerate the development of a highly responsive drone emergency rescue system that integrates large and small drones, coordinates high and low-level capabilities, and fosters new rescue capabilities based on drone rescue platforms, including communication support, disaster reconnaissance, material delivery, and personnel search and rescue.

[0003] In today's emergency scenarios, how to quickly and appropriately deploy drone landing sites and how to deploy limited drone resources where they are most needed have become pressing challenges. Currently, the selection of drone landing sites for emergency situations relies on manual judgment, which is inefficient, error-prone, and requires high skills. Summary of the Invention

[0004] The present invention provides a drone scheduling method for solving the problem of low deployment efficiency of drone parking points in the prior art, thereby improving the deployment efficiency of drone parking points.

[0005] In a first aspect, the present invention provides a drone scheduling method, including: obtaining a first topological map of a drone's stopping point and a second topological map of an area to which the stopping point belongs; preprocessing the first topological map and the second topological map to obtain preprocessing features; inputting the first topological map, the second topological map and the preprocessing features into a stopping point prediction model to obtain a stopping point selection result output by the stopping point prediction model, so as to schedule the drone based on the stopping point selection result; wherein the stopping point prediction model extracts spatial features and temporal features of the first topological map based on the preprocessing features to obtain first topological map features; extracts spatial features and temporal features of the second topological map based on the preprocessing features to obtain second topological map features; fuses the first topological map features and the second topological map features to obtain a fused feature map, and obtains a stopping point selection result based on the fused feature map.

[0006] In one embodiment, the first topological map is composed of stop nodes, and the second topological map is composed of regional nodes. The preprocessing features include a first adjacency matrix, a second adjacency matrix, a first feature matrix, and a second feature matrix. The first topological map and the second topological map are preprocessed to obtain the preprocessing features, including: in the first topological map, extracting the first connection relationship of the stop nodes, and obtaining a first adjacency matrix based on the stop nodes and the first connection relationship; obtaining a first feature matrix based on the stop nodes and the first target features; the first target features include the attributes of the stop nodes, the number of drones that can be called by the stop nodes, the location information of the stop nodes, and disaster feature data of the area corresponding to the stop nodes in a historical time period; in the second topological map, extracting the second connection relationship of the regional nodes, and obtaining a second adjacency matrix based on the regional nodes and the second connection relationship; obtaining a second feature matrix based on the regional nodes and the second target features; the second target features include the location information of the regional nodes and disaster feature data of the area corresponding to the regional nodes in a historical time period.

[0007] In one embodiment, the stop point prediction model includes a graph convolution layer and a long short-term memory layer. The stop point prediction model is used to obtain first topological map features: the graph convolution layer is used to extract spatial features of the first topological map based on the first adjacency matrix to obtain first spatial features; the long short-term memory layer is used to extract temporal features of the first topological map based on the first feature matrix to obtain first temporal features; and the first topological map features are obtained based on the first spatial features and the first temporal features.

[0008] In one embodiment, the stop prediction model is used to obtain the second topological map features: the graph convolution layer is further used to extract spatial features of the second topological map based on the second adjacency matrix to obtain second spatial features; the long short-term memory layer is further used to extract temporal features of the second topological map based on the second feature matrix to obtain second temporal features; and the second topological map features are obtained based on the second spatial features and the second temporal features.

[0009] In one embodiment, the stop point prediction model further includes a multi-graph convergence layer and a fully connected layer. The stop point prediction model is used to obtain a stop point selection result: the multi-graph convergence layer is used to fuse the first topology map features and the second topology map features based on stop nodes or regional nodes to obtain a fused feature map; the fully connected layer is used to extract and activate features from the fused feature map to obtain a stop point selection result.

[0010] In one embodiment, the graph convolution layer includes multiple layers of neural networks connected in sequence, and the graph convolution layer is used to obtain the first spatial feature: the current layer neural network is used to obtain the output result of the current layer neural network based on the symmetric normalized matrix of the first adjacency matrix, the parameters of the current layer neural network and the output result of the previous layer neural network; wherein, the output result of the first layer neural network is obtained based on the symmetric normalized matrix of the first adjacency matrix, the parameters of the first layer neural network and the first topological graph; based on the output result of the last layer neural network, the first spatial feature is obtained.

[0011] In a second aspect, the present invention also provides a drone scheduling device, including: an acquisition module, used to obtain a first topological map of the drone's stop and a second topological map of the area to which the stop belongs; a preprocessing module, used to preprocess the first topological map and the second topological map to obtain preprocessing features; a prediction module, used to input the first topological map, the second topological map and the preprocessing features into a stop prediction model, obtain the stop selection result output by the stop prediction model, and schedule the drone based on the stop selection result; wherein the stop prediction model extracts spatial features and temporal features of the first topological map based on the preprocessing features to obtain first topological map features; extracts spatial features and temporal features of the second topological map based on the preprocessing features to obtain second topological map features; fuses the first topological map features and the second topological map features to obtain a fused feature map, and obtains the stop selection result based on the fused feature map.

[0012] In a third aspect, the present invention also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, any one of the above-mentioned drone scheduling methods is implemented.

[0013] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-mentioned drone scheduling methods.

[0014] In a fifth aspect, the present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-mentioned drone scheduling methods.

[0015] The drone scheduling method, device, electronic device, storage medium, and program product provided by the present invention utilize a stop-point prediction model to extract temporal and spatial features from a first topological map and a second topological map. This innovatively overlays and synthesizes dual topological maps with different correlations within the same geographic reference system into a fused feature map, thereby obtaining stop-point selection results and improving the efficiency and accuracy of drone stop-point deployment. The present invention also preprocesses the first and second topological maps to obtain preprocessed features, which helps improve the accuracy of the stop-point selection results output by the stop-point prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 This is one of the flow charts of the UAV scheduling method provided by the present invention.

[0018] Figure 2 It is a schematic diagram of the first topology diagram and the second topology diagram provided by the present invention.

[0019] Figure 3 This is the second flow chart of the UAV scheduling method provided by the present invention.

[0020] Figure 4 This is the third flow chart of the drone scheduling method provided by the present invention.

[0021] Figure 5 It is a structural schematic diagram of the UAV dispatching device provided by the present invention.

[0022] Figure 6 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0023] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0024] The following combination Figures 1-6 The present invention describes the drone dispatching method, device and electronic equipment.

[0025] Figure 1 This is one of the flow charts of the UAV scheduling method provided by the present invention, such as Figure 1 and Figure 3 As shown, the drone scheduling method includes steps S100 to S300, and the details of each step are as follows.

[0026] S100: Obtain a first topological map of the drone and a second topological map of the area to which the parking point belongs.

[0027] When a region needs to plan drone stops based on emergency events to facilitate drone dispatch, the UAV emergency dispatch platform collects static and dynamic feature data in real time. Static feature data includes information about drone stops and the areas they belong to (for example, the township to which they belong). Dynamic feature data includes the number of drones available at each stop. Dynamic feature data also includes geological, meteorological, and water disaster warnings for the stops and their respective areas over a historical period.

[0028] The unmanned aerial emergency dispatch platform constructs the first topology map (UAV parking site topology map) and the second topology map (regional administrative geographic location topology map) based on the acquired static feature data and dynamic feature data. Figure 2 As shown, the first topology graph is composed of stop nodes, and the edges of the first topology graph represent the connection relationship between the stop nodes. The stop nodes in the first topology graph include temporary points and base points. Figure 2 As shown, the second topology graph is composed of regional nodes, and the edges of the second topology graph represent the connection relationship between the regional nodes.

[0029] S200: Preprocess the first topology map and the second topology map to obtain preprocessing features.

[0030] Perform preliminary feature extraction on the first topological graph and the second topological graph, construct a preprocessing feature matrix (including a first adjacency matrix and a first feature matrix) of the first topological graph based on the preliminary extracted features, and construct a preprocessing feature matrix (including a second adjacency matrix and a second feature matrix) of the second topological graph based on the preliminary extracted features.

[0031] S300: Inputting the first topology map, the second topology map, and the pre-processed features into a stop point prediction model, obtaining a stop point selection result output by the stop point prediction model, and dispatching the drone based on the stop point selection result.

[0032] Among them, the stop point prediction model extracts spatial features and temporal features of the first topology map based on the preprocessing features to obtain the first topology map features; extracts spatial features and temporal features of the second topology map based on the preprocessing features to obtain the second topology map features; fuses the first topology map features and the second topology map features to obtain a fused feature map, and obtains the stop point selection result based on the fused feature map.

[0033] The stop prediction model is trained and validated based on a pre-defined model, using the sample's first and second topological maps, sample pre-processing features, and the labels of the sample stop selection results (e.g., selected nodes are labeled 1 and unselected nodes are labeled 0). The pre-defined model includes a spatiotemporal multi-graph convolution network (ST-MGCN) model. The ST-MGCN model learns the spatiotemporal relationship between the first and second topological maps, capturing the temporal and spatial dependencies between stop nodes and regional nodes, thereby deriving features from the first and second topological maps. The features of the first and second topological maps are fused to form a fused feature graph, from which the stop selection results are derived. For example, in the first topological map, selected stop nodes are labeled 1, and unselected stop nodes are labeled 0. For example, the stop selection results may include selecting stop 1, stop 2, stop 3, and stop 4 as the stop points for this emergency mission.

[0034] Furthermore, the unmanned aerial emergency dispatch platform dispatches the drone to the selected stop node based on the stop selection results to handle emergency tasks.

[0035] The drone scheduling method provided by the present invention utilizes a stop-point prediction model to extract temporal and spatial features from a first and second topological map. This innovatively overlays and synthesizes dual topological maps with different correlations within the same geographic reference system into a fused feature map, thereby obtaining stop-point selection results and improving the efficiency and accuracy of drone stop-point deployment. This method also preprocesses the first and second topological maps to obtain preprocessed features, which helps improve the accuracy of the stop-point selection results output by the stop-point prediction model.

[0036] Based on the above embodiment, the first topological map is composed of stop nodes, and the second topological map is composed of regional nodes. The preprocessing features include a first adjacency matrix, a second adjacency matrix, a first feature matrix, and a second feature matrix. The first topological map and the second topological map are preprocessed to obtain the preprocessing features, including: in the first topological map, extracting the first connection relationship of the stop nodes, and obtaining a first adjacency matrix based on the stop nodes and the first connection relationship; obtaining a first feature matrix based on the stop nodes and the first target features; the first target features include the attributes of the stop nodes, the number of drones that can be called by the stop nodes, the location information of the stop nodes, and the disaster feature data of the area corresponding to the stop nodes in the historical time period; in the second topological map, extracting the second connection relationship of the regional nodes, and obtaining a second adjacency matrix based on the regional nodes and the second connection relationship; obtaining a second feature matrix based on the regional nodes and the second target features; the second target features include the location information of the regional nodes and the disaster feature data of the area corresponding to the regional nodes in the historical time period.

[0037] The first topological graph can be expressed as G1 (V1, E1). V1 is the set of the drone’s stop nodes, V1={v 11 , v 12 , v 13 ,...,v 1N}. E1 is the set of edges in the first topological graph. The second topological graph can be represented as G2(V2, E2). V2 is the set of the drone’s stop nodes, V2={v 21 , v 22 , v 23 ,...,v 2N E2 is the set of edges of the second topological graph.

[0038] The first topological graph is preprocessed to obtain a first adjacency matrix and a first feature matrix. The first adjacency matrix represents the connectivity between each stop node in the first topological graph. All first target features are obtained from the dynamic feature data and the static feature data. Assuming there are M first target features, for each stop node, the attribute values ​​of the M first target features corresponding to the stop node are obtained. The first target features include the attributes of the stop node, the number of drones that can be deployed at the stop node, the location information of the stop node (e.g., the latitude and longitude of the stop node), and disaster characteristic data for the area corresponding to the stop node over a historical period (e.g., geological, meteorological, and water disaster warning characteristics for the area corresponding to the stop node).

[0039] ; in, It is a stop node; Stop node and stop nodes The first connection relationship between them; and When there is a route between is 1, otherwise 0; is the first adjacency matrix, is the first characteristic matrix, Stop node No. The attribute value of the first target feature.

[0040] The second topology graph is preprocessed to obtain a second adjacency matrix and a second feature matrix. The second adjacency matrix represents the connection relationship between the regional nodes in the second topology graph. All second target features are obtained from the dynamic feature data and the static feature data. Assuming that there are M second target features in total, for each regional node, the attribute values ​​of the M second target features corresponding to the regional node are obtained. The second target features include the location information of the regional node (for example, the latitude and longitude location information of the administrative area of ​​the regional node) and the disaster feature data of the region corresponding to the regional node within a historical time period.

[0041] ; in, is a regional node; For regional nodes and regional nodes The second connection relationship between and When there is a road connection between is 1, otherwise 0; is the second adjacency matrix, is the second characteristic matrix, For regional nodes No. The attribute value of the second target feature.

[0042] The embodiment of the present invention obtains the first adjacency matrix, the second adjacency matrix, the first feature matrix, and the second feature matrix, and converts the main features of the first topology map, the second topology map, and the dynamic feature data into data that is easily recognized by the model, which is beneficial to the efficiency of the subsequent stop prediction model in performing feature processing on the first topology map and the second topology map.

[0043] Based on the above embodiment, the stop point prediction model includes a graph convolution layer and a long short-term memory layer. The stop point prediction model is used to obtain first topological map features: the graph convolution layer is used to extract spatial features of the first topological map based on the first adjacency matrix to obtain first spatial features; the long short-term memory layer is used to extract temporal features of the first topological map based on the first feature matrix to obtain first temporal features; and the first topological map features are obtained based on the first spatial features and the first temporal features.

[0044] The graph convolution layer includes multiple layers of neural networks connected in sequence, and the graph convolution layer is used to obtain the first spatial feature: the current layer of neural network is used to obtain the output result of the current layer of neural network based on the symmetric normalized matrix of the first adjacency matrix, the parameters of the current layer of neural network and the output result of the previous layer of neural network; wherein, the output result of the first layer of neural network is obtained based on the symmetric normalized matrix of the first adjacency matrix, the parameters of the first layer of neural network and the first topological graph; based on the output result of the last layer of neural network, the first spatial feature is obtained.

[0045] The current neural network layer is connected to the previous neural network layer. The graph convolution layer includes multiple layers of neural networks and activation functions connected in sequence, for example, the activation function is ReLU.

[0046] ; Among them, the input data of the first layer of the neural network in the graph convolution layer (the first topological graph) is , is the first convolutional layer of the graph The output of the layer (the previous layer of neural network), is the first convolutional layer of the graph The output of the layer (current layer neural network), is the first convolutional layer of the graph The parameters of the layer (current layer neural network), is the first adjacency matrix The symmetric normalized matrix of , is the identity matrix, for The node degree diagonal matrix of .

[0047] The graph convolution layer extracts spatial features from the first topological graph based on the first adjacency matrix to obtain the first spatial features. The output of the last neural network layer in the graph convolution layer is obtained ( ), obtaining a first topological graph carrying the first spatial features. The Long Short-Term Memory (LSTM) layer extracts temporal features from the first topological graph carrying the first spatial features based on the first feature matrix, obtaining first temporal features. The LSTM layer also fuses and activates the first temporal features and the first spatial features to obtain first topological graph features. For example, the LSTM layer has 64 neurons, and the activation function is set to ReLU.

[0048] ; in, is the first topological feature, is the first topological graph carrying the first spatial feature (the output result of the last layer of neural network in the graph convolution layer).

[0049] The present invention performs spatial feature extraction and temporal feature extraction on the first topological map through a graph convolution layer and a long short-term memory layer, thereby fusing the first temporal feature and the first spatial feature into one map and realizing comprehensive information extraction of the first topological map.

[0050] Based on the above embodiment, the stop point prediction model is used to obtain the second topological map features: the graph convolution layer is further used to extract spatial features of the second topological map based on the second adjacency matrix to obtain second spatial features; the long short-term memory layer is further used to extract temporal features of the second topological map based on the second feature matrix to obtain second temporal features; and the second topological map features are obtained based on the second spatial features and the second temporal features.

[0051] The current layer neural network is used to obtain the output result of the current layer neural network based on the symmetric normalized matrix of the second adjacency matrix, the parameters of the current layer neural network and the output result of the previous layer neural network; wherein, the output result of the first layer neural network is obtained based on the symmetric normalized matrix of the second adjacency matrix, the parameters of the first layer neural network and the second topological graph; based on the output result of the last layer neural network, the second spatial feature is obtained.

[0052] ; Among them, the input data of the first layer of the neural network of the graph convolution layer (the second topological graph) is , is the first convolutional layer of the graph The output of the layer (the previous layer of neural network), is the first convolutional layer of the graph The output of the layer (current layer neural network), is the first convolutional layer of the graph The parameters of the layer (current layer neural network), is the second adjacency matrix The symmetric normalized matrix of , is the identity matrix, for The node degree diagonal matrix of .

[0053] The graph convolution layer extracts spatial features from the two topological graphs based on the second adjacency matrix to obtain the second spatial features. Get the output of the last neural network layer in the graph convolution layer ( ), obtaining a second topological map carrying the second spatial features. The long short-term memory (LSTM) layer extracts temporal features from the second topological map carrying the second spatial features based on the second feature matrix, obtaining second temporal features. The LSTM layer also fuses and activates the second temporal and spatial features to obtain second topological map features. For example, the LSTM layer has 64 neurons, and the activation function is set to ReLU.

[0054] ; in, is the second topological feature, is the second topological graph carrying the second spatial features (the output result of the last layer of neural network in the graph convolution layer).

[0055] The present invention performs spatial feature extraction and temporal feature extraction on the second topological map through a graph convolution layer and a long short-term memory layer, thereby fusing the second temporal feature and the second spatial feature into one map and realizing comprehensive information extraction of the second topological map.

[0056] The stop point prediction model also includes a multi-graph convergence layer and a fully connected layer. The stop point prediction model is used to obtain the stop point selection result: the multi-graph convergence layer is used to fuse the first topology map features and the second topology map features based on the stop node or regional node to obtain a fused feature map; the fully connected layer is used to extract and activate features from the fused feature map to obtain the stop point selection result.

[0057] The multi-graph convergence layer is used to fuse the features of the first topology graph and the second topology graph based on the stop nodes or regional nodes to obtain a fused feature graph.

[0058] ; in, is the first topological feature, is the second topological feature, is the fusion feature map, is an aggregation function, and the aggregation method can be sum, maximum, or average.

[0059] The fully connected layer extracts and activates the fused feature map to generate the stop selection result. For example, the number of neurons in the fully connected layer is set to N (N is the number of stop nodes in the first topology graph), the activation function is set to sigmoid, and the output is the stop selection result (selected as a stop is marked as 1, and not selected as a stop is marked as 0).

[0060] This invention uses multi-graph convolution to perform graph convolution on various spatial correlations, followed by feature fusion, to synthesize different correlation graphs at a given moment into a single fused feature graph. Prediction in the temporal dimension fuses information from T historical time steps into a single graph; prediction in the spatial dimension synthesizes different correlation graphs at a given moment into a single graph. This invention uses a spatiotemporal graph network (including a graph convolution layer and a long-short-term memory layer) to model the two correlations between the first and second topological graphs. Edge values ​​between stop nodes (or regional nodes) are determined based on the correlations in different dimensions, with the nodes being identical across all graphs.

[0061] The present invention fuses the features of the first topology map and the second topology map through a multi-map convergence layer, innovatively superimposing and synthesizing dual topology maps with different correlations in the same geographic reference system into a fused feature map, and then obtains the stop point selection result through a fully connected layer, thereby improving the efficiency and accuracy of drone stop point deployment.

[0062] The drone scheduling device provided by the present invention is described below. The drone scheduling device described below and the drone scheduling method described above can be referenced to each other.

[0063] like Figure 5 As shown, a drone scheduling device includes: an acquisition module 501, which is used to obtain a first topological map of the drone's parking point and a second topological map of the area to which the parking point belongs.

[0064] The preprocessing module 502 is configured to preprocess the first topology map and the second topology map to obtain preprocessing features.

[0065] Prediction module 503 is used to input the first topology map, the second topology map and the preprocessing features into the stop point prediction model, obtain the stop point selection result output by the stop point prediction model, and dispatch the drone based on the stop point selection result; wherein the stop point prediction model extracts spatial features and temporal features from the first topology map based on the preprocessing features to obtain first topology map features; extracts spatial features and temporal features from the second topology map based on the preprocessing features to obtain second topology map features; fuses the first topology map features and the second topology map features to obtain a fused feature map, and obtains the stop point selection result based on the fused feature map.

[0066] The drone dispatching device provided by an embodiment of the present invention uses a stop-point prediction model to extract temporal and spatial features from a first and second topological map. This innovatively overlays and synthesizes dual topological maps with different correlations within the same geographic reference system into a fused feature map, thereby obtaining stop-point selection results and improving the efficiency and accuracy of drone stop-point deployment. This invention also preprocesses the first and second topological maps to obtain preprocessed features, which helps improve the accuracy of the stop-point selection results output by the stop-point prediction model.

[0067] In one embodiment, the first topological graph is composed of stop nodes, the second topological graph is composed of regional nodes, and the preprocessing features include a first adjacency matrix, a second adjacency matrix, a first feature matrix, and a second feature matrix. The preprocessing module 502 is used to: in the first topological graph, extract the first connection relationship of the stop nodes, and obtain a first adjacency matrix based on the stop nodes and the first connection relationship; obtain a first feature matrix based on the stop nodes and the first target features; the first target features include the attributes of the stop nodes, the number of drones that can be called by the stop nodes, the location information of the stop nodes, and disaster feature data of the area corresponding to the stop nodes in a historical time period; in the second topological graph, extract the second connection relationship of the regional nodes, and obtain a second adjacency matrix based on the regional nodes and the second connection relationship; obtain a second feature matrix based on the regional nodes and the second target features; the second target features include the location information of the regional nodes and disaster feature data of the area corresponding to the regional nodes in a historical time period.

[0068] In one embodiment, the stop prediction model includes a graph convolution layer and a long short-term memory layer. The prediction module 503 is used to obtain first topological map features: the graph convolution layer is used to extract spatial features of the first topological map based on the first adjacency matrix to obtain first spatial features; the long short-term memory layer is used to extract temporal features of the first topological map based on the first feature matrix to obtain first temporal features; and the first topological map features are obtained based on the first spatial features and the first temporal features.

[0069] In one embodiment, the prediction module 503 is used to obtain the second topological map features: the graph convolution layer is also used to extract spatial features of the second topological map based on the second adjacency matrix to obtain second spatial features; the long short-term memory layer is also used to extract temporal features of the second topological map based on the second feature matrix to obtain second temporal features; based on the second spatial features and the second temporal features, the second topological map features are obtained.

[0070] In one embodiment, the stop prediction model further includes a multi-graph convergence layer and a fully connected layer. The prediction module 503 is used to obtain a stop selection result: the multi-graph convergence layer is used to fuse the first topology graph features and the second topology graph features based on stop nodes or regional nodes to obtain a fused feature graph; the fully connected layer is used to extract and activate features from the fused feature graph to obtain a stop selection result.

[0071] In one embodiment, the graph convolution layer includes multiple layers of neural networks connected in sequence, and the graph convolution layer is used to obtain the first spatial feature: the current layer neural network is used to obtain the output result of the current layer neural network based on the symmetric normalized matrix of the first adjacency matrix, the parameters of the current layer neural network and the output result of the previous layer neural network; wherein, the output result of the first layer neural network is obtained based on the symmetric normalized matrix of the first adjacency matrix, the parameters of the first layer neural network and the first topological graph; based on the output result of the last layer neural network, the first spatial feature is obtained.

[0072] Figure 6 An example of a physical structure diagram of an electronic device is shown below. Figure 6 As shown, the electronic device may include: a processor 610 , a communications interface 620 , a memory 630 and a communication bus 640 , wherein the processor 610 , the communications interface 620 and the memory 630 communicate with each other via the communication bus 640 . The processor 610 can call the logic instructions in the memory 630 to execute the drone scheduling method, which includes: obtaining a first topological map of the drone's stopping point and a second topological map of the area to which the stopping point belongs; preprocessing the first topological map and the second topological map to obtain preprocessing features; inputting the first topological map, the second topological map and the preprocessing features into a stopping point prediction model, obtaining a stopping point selection result output by the stopping point prediction model, and scheduling the drone based on the stopping point selection result; wherein the stopping point prediction model extracts spatial features and temporal features of the first topological map based on the preprocessing features to obtain first topological map features; extracts spatial features and temporal features of the second topological map based on the preprocessing features to obtain second topological map features; fuses the first topological map features and the second topological map features to obtain a fused feature map, and obtains a stopping point selection result based on the fused feature map.

[0073] Furthermore, the logic instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0074] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the drone scheduling method provided by the above methods, the method including: obtaining a first topological map of the drone's stopping point and a second topological map of the area to which the stopping point belongs; preprocessing the first topological map and the second topological map to obtain preprocessing features; inputting the first topological map, the second topological map and the preprocessing features into a stopping point prediction model to obtain a stopping point selection result output by the stopping point prediction model, so as to schedule the drone based on the stopping point selection result; wherein the stopping point prediction model extracts spatial features and temporal features from the first topological map based on the preprocessing features to obtain first topological map features; extracts spatial features and temporal features from the second topological map based on the preprocessing features to obtain second topological map features; fuses the first topological map features and the second topological map features to obtain a fused feature map, and obtains the stopping point selection result based on the fused feature map.

[0075] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the drone scheduling method provided by the above-mentioned methods, the method comprising: obtaining a first topological map of the drone's stopping point and a second topological map of the area to which the stopping point belongs; preprocessing the first topological map and the second topological map to obtain preprocessing features; inputting the first topological map, the second topological map and the preprocessing features into a stopping point prediction model to obtain a stopping point selection result output by the stopping point prediction model, so as to schedule the drone based on the stopping point selection result; wherein the stopping point prediction model extracts spatial features and temporal features from the first topological map based on the preprocessing features to obtain first topological map features; extracts spatial features and temporal features from the second topological map based on the preprocessing features to obtain second topological map features; fuses the first topological map features and the second topological map features to obtain a fused feature map, and obtains the stopping point selection result based on the fused feature map.

[0076] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0077] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A drone scheduling method, characterized in that: include: Obtain a first topological map of the parking point of the drone and a second topological map of the area to which the parking point belongs; Preprocessing the first topological map and the second topological map to obtain preprocessing features; The first topology map, the second topology map, and the preprocessing features are input into a stop point prediction model to obtain a stop point selection result output by the stop point prediction model, so as to dispatch the UAV based on the stop point selection result; wherein, the stop point prediction model performs spatial feature extraction and temporal feature extraction on the first topology map based on the preprocessing features to obtain first topology map features; performs spatial feature extraction and temporal feature extraction on the second topology map based on the preprocessing features to obtain second topology map features; the first topology map features and the second topology map features are fused to obtain a fused feature map, and the stop point selection result is obtained based on the fused feature map.

2. The drone dispatching method according to claim 1, characterized in that: The first topology graph is composed of stop nodes, the second topology graph is composed of area nodes, the preprocessing features include a first adjacency matrix, a second adjacency matrix, a first feature matrix, and a second feature matrix. Preprocessing the first topology graph and the second topology graph to obtain the preprocessing features includes: Extracting first connection relationships of the stop nodes from the first topological graph, and obtaining a first adjacency matrix based on the stop nodes and the first connection relationships; obtaining a first feature matrix based on the stop nodes and first target features; the first target features including attributes of the stop nodes, the number of drones available for the stop nodes, location information of the stop nodes, and disaster feature data of the areas corresponding to the stop nodes within a historical time period; In the second topological graph, the second connection relationship of the regional nodes is extracted, and the second adjacency matrix is ​​obtained based on the regional nodes and the second connection relationship; the second feature matrix is ​​obtained based on the regional nodes and the second target features; the second target features include the location information of the regional nodes and the disaster feature data of the area corresponding to the regional nodes in the historical time period.

3. The drone dispatching method according to claim 2, characterized in that: The stop prediction model includes a graph convolution layer and a long short-term memory layer, and is used to obtain the first topological graph features: The graph convolution layer is used to extract spatial features from the first topological graph based on the first adjacency matrix to obtain first spatial features; The long short-term memory layer is used to extract time features from the first topological graph based on the first feature matrix to obtain first time features; The first topological map feature is acquired based on the first spatial feature and the first temporal feature.

4. The drone dispatching method according to claim 3, characterized in that: The stop prediction model is used to obtain the second topology map feature: The graph convolution layer is further used to extract spatial features from the second topological graph based on the second adjacency matrix to obtain second spatial features; The long short-term memory layer is further used to extract time features from the second topological graph based on the second feature matrix to obtain second time features; Based on the second spatial feature and the second temporal feature, the second topological map feature is acquired.

5. The drone dispatching method according to claim 2, characterized in that: The stop point prediction model further includes a multi-graph aggregation layer and a fully connected layer, and the stop point prediction model is used to obtain the stop point selection result: The multi-graph convergence layer is used to fuse the first topology graph features and the second topology graph features based on the stop node or the regional node to obtain the fused feature graph; The fully connected layer is used to extract and activate features of the fused feature map to obtain the stop selection result.

6. The drone dispatching method according to claim 3, characterized in that: The graph convolution layer includes a multi-layer neural network connected in sequence, and the graph convolution layer is used to obtain the first spatial feature: Current layer The neural network is used to obtain an output result of the neural network in the current layer based on the symmetric normalized matrix of the first adjacency matrix, the parameters of the neural network in the current layer, and the output result of the neural network in the previous layer; wherein the output result of the first layer of the neural network is obtained based on the symmetric normalized matrix of the first adjacency matrix, the parameters of the first layer of the neural network, and the first topological graph; Based on the output result of the last layer of neural network, the first spatial feature is obtained.

7. A drone dispatching device, characterized in that: include: An acquisition module, configured to acquire a first topological map of the parking point of the drone and a second topological map of the area to which the parking point belongs; A preprocessing module, configured to preprocess the first topology map and the second topology map to obtain preprocessing features; A prediction module is configured to input the first topology map, the second topology map, and the preprocessing features into a stopover point prediction model, obtain a stopover point selection result output by the stopover point prediction model, and dispatch the drone based on the stopover point selection result; wherein the stopover point prediction model performs spatial feature extraction and temporal feature extraction on the first topology map based on the preprocessing features to obtain first topology map features; performs spatial feature extraction and temporal feature extraction on the second topology map based on the preprocessing features to obtain second topology map features; fuses the first topology map features and the second topology map features to obtain a fused feature map, and obtains the stopover point selection result based on the fused feature map.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the drone scheduling method according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the drone scheduling method according to any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the drone scheduling method according to any one of claims 1 to 6 is implemented.

Images