Online simulation task scheduling in cloud manufacturing with cross attention and deep reinforcement learning

Online simulation task scheduling in a private cloud manufacturing platform usually requires rapid decision-making algorithms because of the characteristics of unpredictability and diversity of tasks. However, the existing approaches face challenges in generating satisfactory scheduling schemes within a limited solving time. Therefore, this paper proposes a dynamic scheduling algorithm for online simulation task scheduling that is based on cross-attention and deep reinforcement learning (DRL). A multichannel DRL-based framework with discrete event triggering is introduced to effectively recognize online scheduling environments. An innovative multistep state feature cross-attention method is proposed to address the challenge of temporal features caused by nonsimultaneous task arrivals. A case study in the semiconductor display industry with 35 diverse scheduling scenarios was conducted to evaluate the efficacy of the proposed algorithm, which was compared with six classic state-of-the-art DRL algorithms and three commonly used priority dispatching rules. The results show that the proposed algorithm maintains superior scheduling performance across multiple scheduling scenarios and outperforms the other algorithms by an average of nearly 30% when the optimization objective is considered.