Research On Multi AGV Path Planing Based On AnyLogic And Reinforcement Learning Control

In recent years,with the rapid development of intelligent manufacturing,ecommerce,and intelligent warehousing industries,more and more industries have begun to use Automated Guide Vehicles(AGV)to replace manual work in production,transportation,and other aspects.AGV can efficiently complete transportation tasks,with broad application prospects and market prospects.This thesis mainly focuses on the multi-AGV path planning in factory material delivery,the specific research content is as follows:According to the workflow of the factory material delivery,a simulation model was established using AnyLogic’s material handling library,and the physical parameters of the model were set.An improved A-star algorithm was proposed,incorporating the actual travel time of AGVs between nodes into the evaluation function to guide subsequent AGVs in path planning.Compared to the traditional A-star algorithm using Euclidean distance and Manhattan distance,it is more suitable for factory material delivery.The feasibility and effectiveness of the proposed algorithm were verified through simulation experiments.Aiming at the excellent performance of Q-routing in network routing,this thesis introduces the structure of Q-routing into Q(λ)-learning algorithm,and applies it to multi-AGV path planning in factory material delivery system.The feasibility and effectiveness of the proposed method are verified by simulation experiments.An improved QLWBR(λ)algorithm is proposed to solve the problems of large amount of calculation and high load in centralized AGV control system.Combining the actual travel time of AGV with the potential function to set the reward,adopting the Boltzmann policy to keep the balance between exploration and utilization,and relying on the paths stored by AGV to construct the communication network between agents for local communication,the Q table is updated by the information among the neighbors,which optimizes the updating process,reduces the computation and eliminates the invalid updating problem.The simulation results show that the proposed algorithm has lower computational complexity and higher computational efficiency.