Distributed workshop scheduling was a hot topic several years ago, particularly in heterogeneous factory settings. At the time, evolutionary algorithms represented the simplest approach to solving such complex problems. With basic encoding and decoding modeling, one could rapidly explore the problem space. However, constrained by the “No Free Lunch Theorem,” evolutionary algorithms often employed variable neighborhood search to enhance convergence. Inspired by Prof. Ye Tian's 2022 TETCI paper, I experimented with using DQN for local search operator recommendation, achieving reasonably good convergence. However, such methods lack efficiency. To reduce the time spent on DRL, agents should be pre-trained for online inference.
Please refer to
@ARTICLE{10242078, author={Li, Rui and Gong, Wenyin and Wang, Ling and Lu, Chao and Dong, Chenxin}, journal={IEEE Transactions on Systems, Man, and Cybernetics: Systems}, title={Co-Evolution With Deep Reinforcement Learning for Energy-Aware Distributed Heterogeneous Flexible Job Shop Scheduling}, year={2024}, volume={54}, number={1}, pages={201-211}, keywords={Production facilities;Job shop scheduling;Search problems;Manufacturing;Statistics;Sociology;Behavioral sciences;Co-evolution;deep Q-networks (DQNs);distributed heterogeneous flexible job shop scheduling (DHFJS) problem;energy-saving;multiobjective optimization}, doi={10.1109/TSMC.2023.3305541}}