Human Learning Effect Multi-Period Scheduling for Circular Disassembly Line Balancing Based on IMPALA Reinforcement Learning

Abstract

In recognition of the rapidly increasing demand for recycling, disassembly has already become a core technique. Traditional manual disassembly lines are being phased out. With the help of the reinforcement learning algorithm importance-weighted actor-learner architecture (IMPALA), this work proposes a solution to the multi-period personnel scheduling problem that considers the learning effect of workers, aiming to maximize profits and efficiency by utilizing the learning effect of workers for more reasonable task scheduling in a cyclic disassembly assembly line environment. After comparing its results with those of exact solvers and other intelligent optimization methods (such as "A2C, SAC, DQN, DDPG, "), we conclude its competitive performance in both solution quality and efficiency.