Reinforcement Learning Assisting Artificial Bee Colony Algorithm for Scheduling Distributed Assembly Flowshops With Batch Delivery

Abstract

In response to escalating market demands, we extend the distributed assembly flowshop problems (DAFSPs) by incorporating batch delivery, optimizing both total energy consumption (TEC) and total completion time, simultaneously. First, a mathematical model for DAFSP with batch delivery is constructed. Second, the artificial bee colony (ABC) algorithm is enhanced to solve the concerned problems. Two dispatch rules are designed to enhance the quality and diversity of initial solutions. Third, seven local search operators tailored to problem characteristics and two objective-oriented machine speed adjustment strategies are designed for improving the performance of ABC. Two reinforcement learning (RL) algorithms, SARSA and Q-learning, are used to select the appropriate local search operators and speed adjustment strategies during iterations. Two pairs of state-action strategies are developed for local search selection and speed adjustment, respectively. Finally, extensive simulation experiments and detailed analysis demonstrate that the SARSA-assisted ABC has a better performance than its peers for DAFSP with batch delivery.