AGV Scheduling Method Based On Online Learning On The Automation Container Terminal

With the increasing container throughput of container ports,terminal managers have introduced highly automated integrated loading and unloading equipment,including automatic guided vehicles in order to improve the overall operating efficiency of the terminal and to meet the requirements of low-carbon,environmental protection,intelligence,and energy saving.Automatic Guided Vehicle(AGV)is an automated horizontal transport equipment.The horizontal transportation of automated container terminals is one of the important factors affecting the efficiency of terminal operations.How to reasonably implement active and effective AGV assignment scheduling,and how to ensure the conflict control of AGV operations is one of the problems that must be solved in the automation of container terminals.This paper studies the background of the current development status of automated container terminals,operation procedures,overview of dispatching systems and AGV horizontal transport conflicts.Based on reading related literature at home and abroad,optimize the assignment of AGV in the automated container terminal,and design the AGV conflict avoidance control method.Firstly,an online learning AGV real-time scheduling method is proposed.This method can dynamically adjust the AGV assignment policy according to the conditions of the terminal.The performance evaluation of this dispatching method considers three factors of the average load time of quay crane,the average dead time of AGV and the average waiting time of conflict of AGV.and paired the best operation with other operations to form the training samples.Using the machine learning method,these samples are trained based on preference function assignment rules,rule learning using BP neural network.Pairing best job with the others to form training samples,these samples are used to train dispatching rules based on preference function by machine learning.Rules learning is implemented using BP neural network.Finally,the reliability of the online learning AGV real-time dispatching method is verified through experiments and the validity of this algorithm is verified by comparison with the result of genetic algorithm and the shortest distance strategy.The experimental results show that this method not only obtains good results,but also can gives the real-time AGV dispatching results according to the changing situation of the terminal.Secondly,aiming at the problem of multi-AGV obstacle avoidance in the scheduling process,a multi-AGV avoidance control method for reinforcement learning is proposed.The model considers the cooperative characteristics of AGVs,establishes joint states,joint action sets,and strengthens signal functions.The training of the model was conducted through Qlearning and DQN.And compared with the other two avoidance strategies to verify the effectiveness of the avoidance model.Finally,a horizontal operation system based on B/S architecture is built,and the real-time scheduling optimization algorithm is used as the scheduling method of the system.Through the above research,this paper mainly solves the problem of real-time scheduling and avoidance of AGVs in automated container terminals.The feasibility of the study of machine learning in AGV real-time scheduling and avoidance control is proved.It is hoped that this paper can lead to the discussion of AGV scheduling problems based on machine learning.