Research On Multi-task Dynamic Assignment Method Based On UAV Cooperation

With the continuous advancement of technology,UAVs(Unmanned Aerial Vehicles)have been used in various fields due to their strong mobility,easy deployment,and low cost.When the number of tasks increases and task constraints have become more and more complex,a single UAV can no longer meet the needs of the task.The execution mode of this single drone has gradually evolved to a coordinated execution mode of UAVs.Therefore,how to design an effective multi-UAV and multi-task cooperation system has become a hot research direction nowadays.When designing a multi-UAV cooperative system,factors such as the number of UAVs,the type and number of tasks,and the constraints between drones and tasks must be considered.This makes the task assignment of multiple drones a part of the multi-UAV cooperative system.In the existing task assignment research,there is usually a lack of a complete multi-UAV multi-task assignment system,and when performing task assignment,only the initial constraint conditions of the task are considered for static assignment of tasks,and the emergency situations such as newly issued tasks during task execution is ignored.Therefore,in this thesis,the static initial allocation of multi-tasks and dynamic reallocation of multi-tasks are studied to solve the problem of multi-UAV multi-task coordination.The specific research works are as follows:First,a hierarchical and distributed system architecture is proposed due to the task coordination inefficiency of the multi-UAV multi-task dynamic coordination system,the system is divided into three layers in total,which respectively assume different roles in the multi-UAV multi-task coordination system.The first layer is the control layer,which is responsible for task generation and distribution;the second layer is the processing layer,which is responsible for the static initial assignment of tasks,and divides the tasks into multiple subtasks and distributes the subtasks to the third layer according to the optimization goal.The third layer is the execution processing layer,which is responsible for the execution of tasks.At the same time,the UAVs in the execution layer are independent and can communicate with each other,and they are responsible for the dynamic task redistribution during the task execution process.Therefore,the proposed hierarchical and distributed system architecture can completely realize the multi-UAV and multi-task collaborative work.Second,in the static initial assignment of tasks,due to the large task scale and multiple task characteristics,the problem of slow convergence of task assignment and easy to fall into local optimality,a method of segmented task initial assignment is proposed.This method divides the initial assignment of tasks into two steps,one is the division of subtasks,and the other is the distribution of subtasks.When subtasks are divided,the k-means mean clustering algorithm is used to classify the subtask regions,and then the path planning is performed on the task target points within the subtasks through the genetic algorithm.So,the division of subtasks in the initial assignment of tasks is realized through dimensionality reduction processing.When distributing subtasks,an improved genetic algorithm is used to achieve the shortest overall distance and balanced UAV resources after subtasks are distributed.The improved genetic algorithm selects chromosomes through the combination of optimal retention strategy and roulette.It not only retains the best samples,but also improves the diversity of samples.And it performs the genetic operation of classifying and crossing the chromosomes according to the fitness function value during the crossover.At the same time,the crossover probability and mutation probability are dynamically adjusted according to the fitness function value of the current chromosome which speeds up the algorithm convergence speed.The experimental results show that the static initial task assignment method reduces the overall flight distance by 33% compared with the traditional task assignment method,and achieves the balance of UAV resources.Third,aiming at the problem of low task completion caused by ignoring time constraints in the scenario where tasks are randomly assigned,a dynamic task redistribution method based on deep reinforcement learning is proposed.In the process of mission execution,the drones exchange information and quantify the start time,coordinates,task volume and other characteristics of the global mission in real time to form the global mission information shared by the drones.At the same time,each UAV generates new priority characteristics of the tasks performed based on real-time global information,and makes decisions based on deep reinforcement learning algorithms.Then UAVs dynamically adjust the execution order of subtasks under time constraints and completes dynamic redistribution of multiple tasks.The experimental results show that the dynamic redistribution method improves the system task completion degree of 30% with time constraints in the scenario of issuing new tasks at any time.