Integrated scheduling of multi-objective lot-streaming hybrid flowshop with AGV based on deep reinforcement learning

Abstract

The conventional approach to shop floor scheduling often overlooks handling time, either neglecting it or assimilating it into processing time. This results in inadequate information sharing, hindering operational synergy and impeding overall optimisation. Addressing this issue, this study focuses on the three-stage lot-streaming hybrid flowshop scheduling problem with automated guided vehicles (LSHFSP-AGV). It introduces a multi-objective scheduling model to enhance collaboration between machine production and AGV distribution, three objectives of minimum maximum completion time, minimum total machine idle time, and shortest AGV transportation distance are set. An improved multi-objective double-depth Q learning algorithm (NSGA2-MDDQN) based on NSGA-II is proposed to solve the problem. Taguchi experiments were conducted to determine the optimal parameter combination among alternative parameter combinations, and extensive numerical experiments involving 27 instances demonstrated the superiority of NSGA2-MDDQN over combinatorial scheduling rules, MDDQN, and NSGA-II, proving the superiority of the algorithm. The experimental results show that on the objective of minimising makespan, NSGA2-MDDQN reduces by an average of 23.17% compared to the composite scheduling rule, NSGA-II and MDDQN. And achieving an average reduction of 43.78% in machine idle time, and 9.12% in AGV transport distance.