Research On Dynamic Scheduling Method Of Multi-AGV System For Deadlock Avoidance

In recent years,multi-AGV system has attracted wide attention in intelligent manufacturing.By using AGV instead of manual material distribution,distribution efficiency can be improved and distribution automation can be realized.The safe and stable operation of multi-AGV system depends on effective scheduling policy.However,at present,most scheduling strategies have poor flexibility and lack of realtime performance,which can not cope with dynamic environment.Therefore,this paper takes the multi-AGV system of an automated factory as the research object,conducts in-depth research on the multi-AGV dynamic scheduling strategy.Firstly,this paper introduces the overall structure of the multi-AGV system,sets up the environment model and operation model for the regional control problem of the multi-AGV system,proposes the dynamic scheduling framework of the multi-AGV system including deadlock avoidance and scheduling efficiency optimization,and builds the simulation platform of the multi-AGV system based on the discrete event framework Simpy.It provides a verification environment for deadlock avoidance algorithm and performance optimization algorithm.Then,this paper proposes a more flexible deadlock avoidance algorithm based on graph theory to solve the problem that the current deadlock avoidance algorithm is too constrained,which leads to the action selection of AGV is excessively limited and the potential performance optimization space is compressed.Firstly,the system state is represented as a mixed graph and the relationship between its structure and state security is analyzed.By defining macro-ring structure,a macro-ring compression graph is established to extract AGV which may lead to deadlock.Finally,the state of these AGVs is analyzed to evaluate the security of the system.Based on this,a deadlock avoidance algorithm is designed,and the performance of the proposed deadlock avoidance algorithm is evaluated from two aspects of time complexity and behavior tolerance.The simulation results show that the proposed algorithm is more flexible than the classical AGV Banker’s algorithm and its variants,and improves the potential of system performance optimization.Finally,on the basis of ensuring the safe operation of multi-AGV system through deadlock avoidance algorithm,this paper further proposes a scheduling optimization method of multi-AGV system based on event-driven reinforcement learning algorithm.Firstly,the scheduling problem of multi-AGV system is modeled as event-driven decision process by introducing macro action to solve the strategy optimization problem of multi-agent asynchronous decision and action of different duration.Secondly,state,macro action,cooperative reward and punishment mechanism and model framework are designed from the perspective of minimizing the average execution time of tasks.Finally,the training is carried out on the aforementioned multiAGV system simulation platform.The simulation results show that the proposed optimization method can effectively train and converge in the cooperative environment where multiple agents make asynchronous decisions and the duration of actions is variable,and better action strategies can be learned.In addition,combining the advantages of traditional strategy and ED-RL strategy,this paper further proposes a hybrid strategy based on the number of non-idle AGVs,which has better operational performance than a single strategy system.All in all,based on the system modeling and simulation platform construction,this paper conducts in-depth research on deadlock avoidance and performance optimization methods,and proposes a relatively complete set of dynamic scheduling control strategies for multi-AGV systems,which provides a solution for automatic material distribution scenarios and has certain reference significance.