Reinforcement Learning-Based Estimation of Distribution Algorithm for Energy-Efficient Distributed Heterogeneous Flexible Job Shop Scheduling Problem

The energy-efficient distributed heterogeneous flexible job shop scheduling problem (DHFJSP), incorporating green objectives and multi-factory production models, is a widespread but challenging problem in the manufacturing industry. A reinforcement learning-based estimation of distribution algorithm (RLEDA) is proposed to solve the energy-efficient DHFJSP while minimizing the makespan and total energy consumption (TEC). A hybrid heuristic initialization method is devised to obtain a high-quality population. Two probabilistic models are employed to generate new solutions based on the characteristics of the sub-problems to avoid premature convergence. The Q-learning-based population learning rate adaptive mechanism adjusts the degree of learning information from dominant individuals to improve the distribution of the population. Thirty instances of different scales are utilized to evaluate the effectiveness of the RLEDA. The experimental results show that the RLEDA outperforms the comparison algorithms in solving energy-efficient DHFJSP.