Research On MAS Task Allocation Method Based On Reinforcement Learning And Neural Network

Multi-Agent System(MAS)is composed of multiple agents.Problems that cannot be solved by a single agent,that is,large-scale or more complex problems,can be completed through collaboration between the agents in the system.Nowadays,with the increasing complexity of the problem,multi-agent systems have received more and more attention.Task allocation is one of the most important parts in multi-agent system.It solves the problem of allocating complex tasks to each agent.The purpose is to maximize the overall utility or profit of the task.Nowadays,how to efficiently and reasonably allocate complex tasks to various agents has become one of the hot issues in multi-agent systems.Because neural networks can mine the hidden information of data,reinforcement learning can enable agents to learn autonomously through interaction with the environment,and transfer learning can prevent the model from training from scratch and speed up the task allocation process.Therefore,this paper uses the neural network and deep reinforcement learning to deal with the task allocation of the multi-agent system.In addition,transfer learning is adopted to reduce task allocation time.To complex tasks,neural network and deep reinforcement learning can be used to learn the abstract representation of data and achieve task allocation.This paper mainly studies the task allocation of multi-agent system through reinforcement learning and neural network.The main work is as follows:(1)In order to speed up task allocation,this paper proposes a multi-task allocation algorithm based on reinforcement learning and transfer learning.Many traditional algorithms require recalculation and reallocation of tasks when new tasks arrive.This process consumes much more computing resources.In the new task allocation,the algorithm uses transfer experience to accelerate the task allocation speed and improves the efficiency of task allocation.The main idea is to compare the similarity between the target task and the tasks in the source task library,find the source task that is most similar to the target task,and migrate the allocation scheme of the corresponding task in the strategy library to the target task.The algorithm also uses transfer learning to speed up the agent's use of reinforcement learning to learn the optimal path,that is,the strategy of the sub-task previously processed by the agent is migrated to the completion of the target sub-task.Finally,experiments show that the use of transfer learning speeds up the learning of task allocation strategies and optimal paths,reducing computational overhead and time loss(2)In order to reduce the resource consumption of the system while ensuring that the task can be completed,this paper proposes a multi-agent system task allocation algorithm based on deep learning.The first problem is to use a neural network for task allocation.The input of the network is the resources required for the task to be allocated.The output of the network is the type of task allocation strategy.The goal is to maximize the total income of the system Experiments show that the algorithm has a higher accuracy rate and better real-time performance.Feather,a deep reinforcement learning model is built for task allocation.Because task allocation is a complex optimization problem,it is difficult for traditional distributed task allocation algorithms to obtain the optimal strategy,resulting in low benefit or the long solution time.The algorithm uses the idea of deep reinforcement learning to solve this problem Through interaction with the environment,the agents can get actions that can get actions with greater cumulative reward by continuous trial and error,that is,allocation strategies.The main idea is to use replay buffer to storage state and sample it to obtain training samples.After continuous iteration,the final task allocation strategy is obtained.The algorithm is suitable for solving large-scale and complex task allocation problems.Simulation experiments show that the algorithm effectively reduces the resource consumption of the system and increases the total benefit of the system.