Multi-unmanned aerial vehicle logistics distribution path planning method, device and system, and storage medium

Abstract

The invention discloses a multi-unmanned aerial vehicle logistics distribution path planning method, device and system, and a storage medium. The method comprises the following steps: S1, constructing an unmanned aerial vehicle low-altitude logistics path planning data set; s2, performing heterogeneous graph modeling according to the unmanned aerial vehicle low-altitude logistics path planning data set; wherein one graph comprises a heterogeneous graph of flight areas, unmanned aerial vehicles, tasks and environment information; s3, constructing a heterogeneous graph convolutional network according to the heterogeneous graph; and S4, carrying out multi-unmanned aerial vehicle logistics distribution path optimization and decision making based on the heterogeneous graph neural network. By adopting the technical scheme of the invention, the problems of data isomerism modeling deficiency, dynamic response insufficiency, weak multi-objective optimization capability and the like in the traditional technology are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicle logistics distribution, and in particular relates to a multi-unmanned aerial vehicle logistics distribution path planning method and device, system, and storage medium. Background Art

[0002] As one of the important technologies to promote the development of low-altitude economy, UAV logistics and delivery is gaining increasing attention. With the rapid growth of logistics demand and the complexity of delivery scenarios, traditional path planning methods can no longer meet the needs of modern logistics and delivery. UAV delivery path planning involves the integration and optimization of multi-source heterogeneous data (such as geographic information, UAV status, environmental data, and task requirements), which is a key link to ensure the efficient completion of delivery tasks. However, existing technologies often rely on a single data source or traditional optimization algorithms, which are difficult to effectively respond to dynamic environmental changes and multi-objective optimization requirements. Especially in complex airspace environments, although methods such as A* algorithms and genetic algorithms can provide certain path planning capabilities, their computational efficiency is low and their adaptability is poor, making it difficult to meet the path optimization requirements under real-time and multi-constraint conditions. Therefore, the study of low-cost, high-efficiency UAV logistics and delivery technology that can achieve accurate path planning and comprehensive dynamic optimization has become an important issue that needs to be urgently addressed in the field of low-altitude economy.

[0003] The patent application number CN202210987654.1 proposes a UAV logistics path planning method based on an improved A* algorithm, which optimizes the path length and obstacle avoidance by introducing a dynamic weight adjustment mechanism. Its technical solution mainly relies on static elevation data, building coordinates and other prior information provided by the GIS geographic information system, lacks the fusion processing of real-time environmental parameters, and does not integrate real-time environmental data (such as wind speed, airspace flow) and task requirements (such as cargo weight and timeliness), resulting in insufficient adaptability of path planning in dynamic scenarios.

[0004] The patent with application number CN202110123456.7 proposes a UAV path planning method based on a homogeneous graph neural network (Graph Neural Network, GNN). The core idea of this method is to construct a unified node feature to represent the environment and the UAV state, and use the powerful graph structure modeling capability of GNN to achieve path planning. Specifically, this method encodes information such as obstacles in the environment, UAV status (such as position, speed, power, etc.) and mission requirements into nodes in the graph, and transmits and aggregates information through the graph neural network to generate the optimal path. The advantage of this method is that it can utilize the global information transmission mechanism of GNN, comprehensively consider the dynamic changes of the environment and the UAV, and achieve efficient path planning.

[0005] The patent with the application number CN202310555666.8 proposes a multi - UAV path planning method based on the Genetic Algorithm (GA). The core idea of this method is to gradually optimize the flight paths of UAVs by simulating the crossover, mutation, and selection operations in the biological evolution process, so as to generate a globally optimal path. Specifically, this method first initializes a set of random paths as the initial population, then combines the characteristics of different paths through the crossover operation, introduces randomness through the mutation operation to avoid falling into local optima, and finally retains the paths with higher fitness through the selection operation. After multiple rounds of iteration, the algorithm can converge to a globally optimal or near - optimal path solution. The advantage of this method lies in its strong global search ability, which can effectively avoid the local optimum problem and shows good robustness in complex environments.

[0006] However, the above - mentioned technology has the following limitations:

[0007] Lack of data heterogeneity modeling:

[0008] When the existing technology uses a homogeneous graph structure for UAV path planning, it fails to effectively distinguish the heterogeneous characteristics among UAVs, tasks, and geographical nodes. In a homogeneous graph, all nodes are regarded as the same type, and the relationships between nodes are also simplified to homogeneous relationships, resulting in the inability to accurately model the complex interactions among UAVs, tasks, and the geographical environment. The power consumption of UAVs is closely related to the flight distance and load weight, and the task timeliness may require UAVs to give priority to executing some high - priority tasks. Since the homogeneous graph cannot distinguish these heterogeneous relationships, the path planning results often cannot accurately reflect the actual needs, leading to problems such as insufficient power and task timeout in the actual execution of the generated paths.

[0009] Insufficient dynamic environment response ability:

[0010] The existing methods mainly rely on static environment data in the path planning process and fail to effectively integrate real - time meteorological data (such as wind speed, rainfall) and airspace traffic data (such as the positions and speeds of other aircraft). This reliance on static data leads to poor performance of path planning in the face of sudden environmental changes. Due to the lack of real - time response ability to the dynamic environment in the existing methods, the generated paths may fail in actual execution, resulting in task interruption or UAV damage.

[0011] Weak multi - objective optimization ability:

[0012] Traditional path planning algorithms usually optimize only a single objective (such as path length), while in practical application scenarios, multiple optimization objectives often need to be considered simultaneously, such as path length, flight time, energy consumption, and mission timeliness. There are usually conflicting relationships among these objectives. For example, the shortest path may have high energy consumption, while the path with the lowest energy consumption may not meet the requirements of mission urgency. Traditional single-objective optimization methods cannot balance these contradictions, resulting in the generated paths being difficult to meet the needs of complex logistics scenarios in practical applications.

[0013] Low computational efficiency:

[0014] The path planning method based on the genetic algorithm (GA) has the problem of low computational efficiency in the scenario of large-scale unmanned aerial vehicle (UAV) swarms. With the increase in the number of UAVs and the environmental complexity, the path search space grows exponentially, resulting in a significant increase in the computational time of the algorithm. This high computational complexity makes the existing methods difficult to meet the requirements of real-time path planning, especially in a dynamic environment where path planning needs to be completed within seconds or minutes. Summary of the Invention

[0015] The technical problem to be solved by the present invention is to provide a multi-UAV logistics distribution path planning method, device, system, and storage medium, which solve the problems of missing data heterogeneity modeling, insufficient dynamic response, and weak multi-objective optimization ability in traditional technologies. By constructing a heterogeneous graph containing geographical, UAV, task, and environmental nodes, the complex interaction relationships between nodes (such as the dynamic association between power consumption and flight distance) are accurately characterized, and the path is dynamically updated in combination with real-time meteorological and airspace traffic data. A collaborative optimization algorithm is introduced to jointly optimize multiple objectives such as path length, energy consumption, and timeliness, and the conflicting requirements are balanced through weight adjustment.

[0016] To achieve the above object, the present invention adopts the following technical solutions:

[0017] A multi-UAV logistics distribution path planning method includes:

[0018] Step S1, constructing a UAV low-altitude logistics path planning data set;

[0019] Step S2, performing heterogeneous graph modeling according to the UAV low-altitude logistics path planning data set; wherein, a heterogeneous graph including flight area, UAV, task, and environmental information;

[0020] Step S3, constructing a heterogeneous graph convolutional network according to the heterogeneous graph;

[0021] Step S4, optimizing and making decisions on the multi-UAV logistics distribution path based on the heterogeneous graph neural network.

[0022] Preferably, in step S1, the UAV low-altitude logistics path planning data set includes: geographic information data, UAV status data, environmental data, and task data.

[0023] Preferably, in step S2, the flight area, UAV, and environmental information are integrated into a heterogeneous graph and processed by a graph neural network; each node in the graph represents an entity, and the nodes are connected by edges; each node type has different features, and the weight of the edge represents the relationship between the nodes.

[0024] Preferably, in step S3, the node features and edge features are input into the heterogeneous graph neural network model for training through a graph transformation layer and a graph topology adaptive transformation layer.

[0025] The present invention also provides a multi-UAV logistics distribution path planning device, including:

[0026] A construction module for constructing a UAV low-altitude logistics path planning data set;

[0027] A first processing module for performing heterogeneous graph modeling according to the UAV low-altitude logistics path planning data set; wherein, a heterogeneous graph including flight area, UAV, task, and environmental information;

[0028] A second processing module for constructing a heterogeneous graph convolutional network according to the heterogeneous graph;

[0029] A path planning module for optimizing and making decisions on the multi-UAV logistics distribution path based on the heterogeneous graph neural network.

[0030] Preferably, the UAV low-altitude logistics path planning data set includes: geographic information data, UAV status data, environmental data, and task data.

[0031] Preferably, the first processing module integrates the flight area, UAV, and environmental information into a heterogeneous graph and processes it through a graph neural network; each node in the graph represents an entity, and the nodes are connected by edges; each node type has different features, and the weight of the edge represents the relationship between the nodes.

[0032] Preferably, the second processing module inputs the node features and edge features into the heterogeneous graph neural network model for training through a graph transformation layer and a graph topology adaptive transformation layer.

[0033] The present invention also provides a multi-UAV logistics distribution path planning system, including: a memory and a processor, where a computer program is stored on the memory and run by the processor, and the computer program executes the multi-UAV logistics distribution path planning method when run by the processor.

[0034] The present invention also provides a storage medium, on which a computer program is stored, and the computer program executes the multi-UAV logistics distribution path planning method when running.

[0035] Advantages of the present invention:

[0036] Efficiently handle data heterogeneity and improve path planning accuracy: Through the heterogeneous graph neural network (HGNN) modeling, the present invention effectively distinguishes the heterogeneous features of geographical nodes, UAV nodes, task nodes, and environmental nodes, and accurately depicts the complex interaction relationships between different types of nodes (such as the correlation between power consumption and flight distance, and the correlation between task timeliness and UAV status). Compared with the traditional homogeneous graph method, it can generate paths that better meet the actual needs, significantly reduce problems such as insufficient power and task timeout caused by inaccurate modeling, and improve the success rate of path planning.

[0037] Enhanced real-time response ability in dynamic environments: Integrate real-time meteorological data (wind speed, rainfall) and airspace traffic data (dynamics of other aircraft), and update the environmental model in real time with sensors. Through the adaptive feature update mechanism of the heterogeneous graph convolutional network, dynamically adjust the path planning strategy to reduce the probability of path failure under sudden environmental changes (such as sudden wind speed changes and airspace congestion), and significantly improve the robustness of task execution.

[0038] Outstanding multi-objective collaborative optimization ability: Introduce a collaborative optimization algorithm to jointly optimize multiple objectives such as path length, flight time, energy consumption, and task timeliness. Balance conflicting objectives by dynamically adjusting the objective weights. Under the same constraints, compared with the traditional genetic algorithm (GA) and A* algorithm, the comprehensive cost (path length + energy consumption + timeout penalty) of the present invention is reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0040] Figure 1 It is the flowchart of the multi-UAV logistics distribution path planning method in the embodiment of the present invention;

[0041] Figure 2 It is the flowchart of the construction of the UAV path planning data set;

[0042] Figure 3 It is the flowchart of heterogeneous graph modeling;

[0043] Figure 4 It is the structural diagram of the graph transformation layer unit. Detailed implementation manners

[0044] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0045] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific implementation manners.

[0046] Embodiment 1:

[0047] As Figure 1 shown, the embodiment of the present invention provides a multi-UAV logistics distribution path planning method, including:

[0048] S1. Construction of the UAV low-altitude logistics path planning data set

[0049] The construction process of the UAV path planning data set is as Figure 2 shown. The entire construction process involves four types of data: geographic information data, UAV status data, environmental data, and task data. The construction process is as follows:

[0050] S101. Data acquisition and preliminary processing: Extract geographic information data from the three-dimensional map or satellite image of the flight area, and use remote sensing satellites or three-dimensional scanners to obtain the environmental map. Collect real-time status data through the sensors on the UAV, including information such as position, speed, and altitude, and perform time synchronization on the sensor data to ensure data consistency. Collect meteorological data of the flight environment, including wind speed, temperature, humidity, etc., and at the same time obtain airspace traffic information, including the real-time positions and dynamics of other flying objects around. Collect task-related data, such as the starting point and target point, the timeliness requirements of the task, the weight and volume of the goods, etc.

[0051] S102. Data cleaning: Data cleaning includes the processing of missing values and outliers. For the missing parts in the geographic information data, fill in the missing area data by simulating and generating it. For the real-time status data collected by the sensors, such as position and speed, if there are missing values, interpolate and fill them according to the position information of the previous and subsequent moments. If there are missing meteorological information in the environmental data, fill it with historical weather data or meteorological data of the surrounding areas. Outlier processing mainly targets UAV status and environmental data. For unreasonable speed or position outliers in the UAV data, correct or eliminate them. For extreme weather values in the environmental data or incorrect records in the airspace traffic data, remove or correct them to ensure the validity of the data.

[0052] S103. Data Structure Conversion: Convert geographic information data into graph structure data suitable for path planning, and use a two-dimensional image or multi-dimensional grid with a resolution as the standard to represent the map of the flight area. The UAV status data is stored in the form of a time series, and the data at each moment records the flight status of the UAV. The meteorological information and airspace traffic data in the environmental data are converted into numerical features and standardized to meet the model input requirements. The task data is structured according to features such as time, location, and cargo requirements and converted into numerical data.

[0053] S104. Complete the construction of the UAV path planning dataset: Through steps S101 to S103, a high-quality UAV path planning dataset is finally constructed. Each piece of data includes the geographic information image data of the flight area, the real-time data of the UAV status, the environmental meteorological data, and the task requirement data. This dataset can provide rich inputs for the subsequent path planning model and support the path optimization and decision-making process based on the heterogeneous graph neural network.

[0054] S2. Heterogeneous Graph Modeling

[0055] In path planning, the present invention integrates various data such as the flight area, UAVs, and environmental information into a heterogeneous graph and processes it through a graph neural network. Each node in the graph represents an entity, and the nodes are connected by edges. Each type of node has different features, and the weight of the edge represents the relationship between the nodes. The flow chart of heterogeneous graph modeling is as Figure 3 shown:

[0056] S201. Node Features

[0057] In the heterogeneous graph, the present invention defines four types of nodes: geographic information nodes, UAV nodes, task nodes, and environmental nodes. The features of each type of node are represented by a vector, which is specifically as follows.

[0058] The geographic information nodes represent each position in the flight area, including the starting point, the target point, and other relevant areas. The feature vector of the geographic information node is represented as:

[0059] x g =[x, y, z, t, o]

[0060] Where x, y, and z are geographic coordinates, t is the area type (city, mountain area, etc.), and o is the obstacle information. If there is an obstacle at this position, then o = 1; otherwise, o = 0.

[0061] The UAV nodes represent each UAV, and its feature vector includes flight status data such as position, speed, and altitude:

[0062] x u = [x u , y u , z u , v u , θ u , e u

[0063] Among them, x u , y u , z u are the three-dimensional positions of the UAV, v u is the speed, θ u is the flight direction, and e u is the remaining battery power.

[0064] The task node represents the delivery task in path planning, and its feature vector includes the start time, timeliness requirement, cargo weight, etc. of the task:

[0065] x t = [t start , t deadline , w, v]

[0066] Among them, t start and t deadline are the start and end times of the task respectively, w is the weight of the cargo, and v is the volume of the cargo.

[0067] The environment node is used to store external environment data related to flight, including meteorological information and airspace traffic. Its features can be expressed as:

[0068] x e = [wind_speed, temperature, humidity, traffic_density]

[0069] Among them, wind_speed, temperature, and humidity are the wind speed, temperature, and humidity respectively, and traffic_density represents the flight traffic density in the airspace.

[0070] S202, Edge Features and Weights

[0071] Nodes are connected by edges, and the weights of the edges are defined according to the relationships between the nodes. The weight of each edge reflects the mutual influence between the nodes. Below, how to calculate the weights of the edges will be introduced according to different node types.

[0072] The edges between the geographic information nodes represent the connections between different geographic regions. Assume there are two geographic nodes and The weight of its edge can be calculated by the distance and flight time between the two points:​

[0073]

[0074] Among them, represents the spatial distance between geographic nodes and the time required for flight, and α and β are adjustment coefficients.

[0075] The edge between the UAV and the geographic node represents the possibility of the UAV passing through a certain geographic area. The weight w of the edge u,g is determined by the flight state of the UAV and the characteristics of the geographic area, and is expressed as:

[0076] w u,g = γ·|z u - z g | + δ·complexity(v g )

[0077] Among them, |z u - z g | represents the height difference between the UAV and the geographic area, and complexity(v g ) represents the complexity of the geographic area (such as the presence of obstacles or flight restrictions), and γ and δ are adjustment coefficients.

[0078] The edge between the UAV and the task node represents the possibility of the UAV performing a specific task. The weight w of the edge u,t depends on the timeliness requirement of the task and the state of the UAV, and the formula is:

[0079] w u,t = η·timeliness(v t ) - ζ·status(v u )

[0080] Among them, timeliness(v t ) is the timeliness requirement of the task, and status(v u ) represents the state of the UAV (such as battery power, flight altitude, etc.), and η and ζ are weight coefficients.

[0081] The edge between the task and the geographic node represents the starting position and the target position of the task. The weight w of the edge t,g can be adjusted according to the timeliness requirement of the task and the distance between the geographical locations:

[0082] w t,g = θ·d(v t , v g ) - κ·timeliness(v t )​

[0083] Among them, d(v t , v g ) represents the spatial distance between the task node and the geographical node, timeliness(v t ) represents the timeliness requirement of the task, and θ and κ are adjustment coefficients.

[0084] By defining node features and edge weights in the above manner, the present invention can integrate multi-dimensional information such as flight areas, drones, tasks, and environments into a heterogeneous graph. In this graph, different types of nodes and edges reflect the complex relationships in path planning through specific features and weights. Finally, through the graph neural network model, the present invention can effectively perform path prediction and optimization to provide support for drone logistics distribution tasks.

[0085] S3. Heterogeneous Graph Convolutional Network

[0086] In the present invention, a heterogeneous graph containing flight area, drone, task, and environmental information has been constructed. In this graph, the node types include geographical nodes, drone nodes, task nodes, and environmental nodes, and each type of node has a corresponding feature vector. The edges in the graph connect different types of nodes, and the weights of the edges represent the relationships between the nodes. For the input data, input features and structures need to be defined for each type of node and edge type respectively. These node features and edge features are input into the heterogeneous graph neural network model for training. Node features: Set the feature vector of each node as x v , where v represents different types of nodes. Adjacency matrix: According to the structure of the heterogeneous graph, define the adjacency matrix A uv of different types of edges, representing the connection relationship between node u and node v.

[0087] S301. Graph Transformation Layer

[0088] The graph transformation layer is the basic unit of the heterogeneous graph convolutional network, as Figure 4 shown:

[0089] In each layer, the present invention updates the node features and adjacency information, and the specific process can be expressed as:

[0090]

[0091] Among them, represents the feature vector of node v in the l-th layer, A uvis the adjacency matrix between nodes. The graph transformation layer combines graph convolution and attention mechanism to calculate the new feature representation of nodes. In GTN, the update of node representation at each layer is based on the similarity and context information between nodes, which is specifically implemented through the self-attention mechanism. Given a node v and its neighbor node set N(v), the node feature update process is as follows:

[0092]

[0093] Among them, is the feature of node v at the l-th layer, is the feature of node u at the l-th layer, and N(v) is the set of neighbor nodes of node v. The core formula of self-attention is:

[0094]

[0095] Among them, W q is a trainable weight matrix, and d is the feature dimension, representing the embedding dimension of each node. This formula calculates the similarity between node v and its neighbor nodes, and updates the node features according to this similarity.

[0096] S302. Graph Topology Adaptive Transformation Layer

[0097] Considering the characteristics of heterogeneous graphs, the present invention designs a graph topology adaptive transformation layer based on the graph transformation layer. Different from the traditional graph convolution layer, the graph topology adaptive transformation layer can combine the structural information of the graph and node features, and adaptively adjust the information propagation path through the local topological information of the graph. In this way, the features of nodes not only depend on the features of their adjacent nodes, but also are affected by the topological structures of various parts in the graph.

[0098] Construct a structure representation matrix S based on the local structure of the graph. This matrix is used to measure the local topological relationship between nodes. The local structure matrix S is calculated by the following formula:

[0099]

[0100] Among them, h v is the feature vector of node v, N(v) is the set of neighbors of node v, and Σ u∈N(v) T(h v , h u ) is a function used to represent the local topological relationship between nodes v and u, and measures the similarity between nodes by calculating the cosine similarity of the feature vectors of nodes v and u. The specific calculation method is as follows:

[0101]

[0102] To enable the model to capture global topological information simultaneously, the global structure matrix G of the graph is further introduced to represent the topological features of the entire graph. G reflects the global connectivity between nodes in the graph. The design goal of the global structure matrix is to summarize the overall structure of the graph so that each node not only considers the local structure during information propagation but also can perceive the global connectivity of the entire graph.

[0103] The degree distribution of the graph can reflect the global connectivity. For each node v, its degree d is calculated v , and then the degrees are normalized so that the global structure information of the nodes can be passed in the convolution operation. The calculation method is as follows:

[0104]

[0105] In this way, the topological information of the graph is effectively guided into the update of node features while maintaining the computational efficiency. The new method for updating node features is as follows:

[0106]

[0107] Among them, α is a hyperparameter that controls the balance between local information and global information. By carefully designing the calculation methods of the local structure matrix S and the global structure matrix G, the graph neural network can not only focus on the features of neighboring nodes but also perceive the topological structure of the graph. This method also provides a richer and more flexible graph information propagation mechanism. By combining local and global information, the expressive power of the network can be enhanced, which is especially suitable for complex graph structures such as heterogeneous graphs.

[0108] S303. Global representation

[0109] After the graph is transformed by multiple heterogeneous graph convolutional networks, each node will have a final representation This representation contains the global information of the node in the graph. To further optimize path planning, the present invention can merge the node embeddings in the graph into a global graph embedding representation through a pooling operation.

[0110]

[0111] Among them, Pool average pooling is the pooling operation. The finally obtained global graph embedding representation is h global , and this representation is used for path prediction.

[0112] S4. Path prediction and loss function

[0113] S401. Path generation and optimization

[0114] The goal of path planning is to find an optimal path from the starting point to the target point, which involves multiple constraints and objectives. In graph neural networks, by learning the relationships between nodes, potential paths can be predicted. This invention is based on the node feature representation h learned by the graph neural network global and generate a path through a series of decisions.

[0115] Assume that the goal of path planning is to plan a path between the starting point s and the target point t. The process of path generation can be expressed as:

[0116] π = {v0, v1, …, v n}

[0117] where v0 = s is the starting node and v n = t is the target node. The goal of path generation is to maximize the probability of the path:

[0118]

[0119] where represents the transition probability from node v i to node v i+1 . This transition probability is learned through a fully connected layer to dynamically update the path generation process according to the relationships between nodes and edges in the graph.

[0120] S402. Cooperative Optimization Algorithm

[0121] To further optimize multiple objectives in path planning, such as the total length of the path, flight time, energy consumption, etc., this invention introduces a cooperative optimization algorithm. The goal of the cooperative optimization algorithm is to jointly optimize multiple objective functions to ensure that all constraint conditions are met during the path generation process and to consider the trade - offs of multiple optimization objectives at the same time. In path planning, the following objectives need to be optimized: path length (the shortest path to reduce flight distance and time), flight time (ensuring the flight time is minimized to avoid exceeding the mission timeliness requirements), energy consumption (considering the energy limit of the drone and optimizing the path to extend the flight time), and mission timeliness (ensuring that the path planning meets the mission timeliness requirements and avoiding delays). Define the cooperative optimization loss function as

[0122]

[0123] where f m (π) represents the m - th objective function of path planning, and λ m is a hyperparameter used to adjust the importance of different objectives. Specifically, each objective function f m (π) is as follows.

[0124] The path length f1(π) represents the total length of the path, which is calculated as the cumulative distance between all nodes:

[0125]

[0126] where is the spatial distance between node v i and node v i+1 .

[0127] The flight time f2(π) represents the flight time of the path and is calculated by the flight distance and the speed of the UAV:

[0128]

[0129] where v u is the flight speed of the UAV.

[0130] The energy consumption f3(π): The energy consumption of the path is usually related to factors such as flight distance, flight altitude, and flight speed, and can be expressed as:

[0131]

[0132] where α1 and α2 are energy consumption coefficients related to flight distance and altitude difference, and |z i - z i+1 | represents the flight altitude difference.

[0133] The task timeliness f4(π): The timeliness requirement of the task can be measured by calculating the difference between the completion time of the path and the deadline required by the task:

[0134] f4(π) = max(0, t end - t deadline )

[0135] where t end is the completion time of the path planning, and t deadline is the deadline of the task. The core of the collaborative optimization algorithm lies in weighing different objective functions. By adjusting the weights of each objective function, the present invention can flexibly control the importance of different objectives. The collaborative optimization algorithm can simultaneously optimize multiple aspects of the path (such as path length, flight time, energy consumption, timeliness, etc.), so as to obtain the optimal path that meets the task requirements. Finally, by jointly optimizing the objective functions of the model, better performance can be achieved in complex multi-constrained path planning problems.

[0136] Example 2:

[0137] The embodiment of the present invention further provides a multi-UAV logistics distribution path planning device, including:

[0138] A building module for building a dataset for low-altitude logistics path planning of unmanned aerial vehicles;

[0139] A first processing module for performing heterogeneous graph modeling based on the dataset for low-altitude logistics path planning of unmanned aerial vehicles; among them, a heterogeneous graph including flight area, unmanned aerial vehicle, task and environmental information;

[0140] A second processing module for building a heterogeneous graph convolutional network according to the heterogeneous graph;

[0141] A path planning module for optimizing and making decisions on the multi-unmanned aerial vehicle logistics distribution path based on the heterogeneous graph neural network.

[0142] As an implementation manner of an embodiment of the present invention, the dataset for low-altitude logistics path planning of unmanned aerial vehicles includes: geographic information data, unmanned aerial vehicle status data, environmental data, and task data.

[0143] As an implementation manner of an embodiment of the present invention, the first processing module integrates the flight area, unmanned aerial vehicle, and environmental information into a heterogeneous graph and processes it through a graph neural network; each node in the graph represents an entity, and the nodes are connected by edges; each type of node has different features, and the weight of the edge represents the relationship between the nodes.

[0144] As an implementation manner of an embodiment of the present invention, the second processing module inputs the node features and edge features into the heterogeneous graph neural network model through a graph transformation layer and a graph topology adaptive transformation layer for training.

[0145] Embodiment 3:

[0146] An embodiment of the present invention further provides a multi-unmanned aerial vehicle logistics distribution path planning system, including: a memory and a processor, where a computer program is stored on the memory and run by the processor, and the computer program executes the multi-unmanned aerial vehicle logistics distribution path planning method when run by the processor.

[0147] Embodiment 4:

[0148] An embodiment of the present invention further provides a storage medium, where a computer program is stored on the storage medium, and the computer program executes the multi-unmanned aerial vehicle logistics distribution path planning method when running.

[0149] The above-described embodiments are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for path planning of multi-UAV logistics distribution, characterized in that Including: Step S1: Construct a dataset for low-altitude UAV logistics path planning; Step S2: Perform heterogeneous graph modeling based on the dataset for low-altitude UAV logistics path planning; specifically, a heterogeneous graph including flight area, UAV, task, and environmental information; Step S3: Construct a heterogeneous graph convolutional network based on the heterogeneous graph; Step S4: Optimize and make decisions on the multi-UAV logistics distribution path based on the heterogeneous graph neural network.

2. The multi-UAV logistics distribution path planning method according to claim 1, wherein, In step S1, the dataset for low-altitude UAV logistics path planning includes: geographic information data, UAV status data, environmental data, and task data.

3. The multi-UAV logistics distribution path planning method according to claim 2, wherein In step S2, the flight area, UAV, and environmental information are integrated into a heterogeneous graph and processed through a graph neural network; each node in the graph represents an entity, and the nodes are connected by edges; each node type has different features, and the weight of the edge represents the relationship between the nodes.

4. The multi-UAV logistics distribution path planning method according to claim 3, wherein, In step S3, through the graph transformation layer and the graph topology adaptive transformation layer, the node features and edge features are input into the heterogeneous graph neural network model for training.

5. A multi-UAV logistics distribution path planning device, characterized in that Including: A construction module for constructing a dataset for low-altitude UAV logistics path planning; A first processing module for performing heterogeneous graph modeling based on the dataset for low-altitude UAV logistics path planning; specifically, a heterogeneous graph including flight area, UAV, task, and environmental information; A second processing module for constructing a heterogeneous graph convolutional network based on the heterogeneous graph; A path planning module for optimizing and making decisions on the multi-UAV logistics distribution path based on the heterogeneous graph neural network.

6. The multi-UAV logistics distribution path planning device according to claim 5, characterized in that, The dataset for low-altitude UAV logistics path planning includes: geographic information data, UAV status data, environmental data, and task data.

7. The multi-UAV logistics distribution path planning device according to claim 6, characterized in that, The first processing module integrates the flight area, UAV, and environmental information into a heterogeneous graph and processes it through a graph neural network; each node in the graph represents an entity, and the nodes are connected by edges; each node type has different features, and the weight of the edge represents the relationship between the nodes.

8. The multi-UAV logistics distribution path planning device according to claim 7, characterized in that, The second processing module inputs the node features and edge features into the heterogeneous graph neural network model for training through the graph transformation layer and the graph topology adaptive transformation layer.

9. A multi-UAV logistics distribution path planning system, characterized in that, Including: A memory and a processor, with a computer program stored on the memory and run by the processor, and the computer program, when run by the processor, executes the multi-UAV logistics distribution path planning method according to any one of claims 1-4.

10. A storage medium, characterized in that, A computer program is stored on the storage medium, and the computer program, when running, executes the multi-UAV logistics distribution path planning method according to any one of claims 1-4.

Images