Research On Multi-UAV Cooperative Task Allocation And Route Planning Technology

With the continuous development of hardware and software technology of UAV,UAV has been applied in many fields and has become a research hotspot at home and abroad.A single UAV has some limitations,while multiple UAVs can complete more complex tasks.The core of multi-UAV collaboration problem is that multi-UAV achieves the optimal overall performance through effective collaboration control,which makes the multi-UAV system show its unique advantages in various practical application scenarios.Using the optimization theory and deep learning theory,this thesis studies the task allocation and route planning technology in the multi-UAV collaboration problem.The main work consists of the following two aspects:（1）Multi-UAV collaborative task allocation is an important application of multi-UAV collaborative completion of complex tasks.All tasks are allocated to UAV clusters in a reasonable way,so as to maximize the system performance and give full play to the collaborative work efficiency of UAV clusters.This thesis uses the extended team orientation problem as the model of multi-UAV task allocation problem,and analyzes the characteristics of three benchmark algorithms,namely genetic algorithm,ant colony algorithm and particle swarm optimization algorithm.The benchmark algorithms simply maximize the reward value obtained by the multi-UAV system,which can not make full use of the system resources.In order to solve this problem,based on the advantages of fast convergence and easy expansion of genetic algorithm,this thesis makes three improvements on genetic algorithm.Firstly,by modifying the optimization objective function,this thesis reduces the total flight distance of UAVs,so as to reduce the total energy required to complete the tasks.Secondly,in order to make all UAVs participate in the mission as much as possible and make full use of UAV resources,the flight paths of UAV system need to be balanced as much as possible.This thesis proposes a calculation method of equilibrium coefficient,and improves the equilibrium of UAV flight paths by improving genetic algorithm.Finally,combining simulated annealing algorithm with genetic algorithm can make genetic algorithm get a better solution.（2）Route planning / path planning is an important research field of multi-UAV collaboration.It needs to generate non-conflict paths for UAVs according to waypoints under various environmental and physical constraints.A* algorithm is a path planning algorithm widely used in UAV.Neural A* algorithm is a new path planning method based on datadriven search,which is mainly composed of a convolutional encoder and an A* search module.This thesis uses Neural A* as the path planning algorithm of UAV,deeply studies the encoder and search module,and obtains three methods that can improve the algorithm through a large number of experiments.Firstly,several mainstream convolutional neural networks are replaced as the feature extraction network of the encoder,and it is found that Res Net and DPN are better than the original VGG.Secondly,several mainstream semantic segmentation models are replaced as encoder,and it is found that Deep Lab V3 is more efficient than UNet.Finally,through experiments,it is found that the weight change of A*search module has a great impact on the algorithm,so we can take the weight as a hyperparameter and use the validation dataset to get the best weight value,so as to improve the performance of the algorithm.