Q-learning based estimation of distribution algorithm for scheduling distributed heterogeneous flexible flow-shop with mixed buffering limitation

There are growing interests in the distributed shop scheduling research owing to the diversification of market demand. However, most prevailing studies disregard the synergistic influence of mixed buffering and due window on production efficiency. To reduce cost loss caused by delay in due window, this paper studies a distributed heterogeneous flexible flow-shop scheduling with mixed buffering limitation, i.e., finite buffers and no-wait requirements. The motivation of this work is to fill in void and offer practical insights for exploring how to intelligently implement, optimize and deploy a distributed production system. A mathematical model is established, aiming to minimize total weighted earliness and tardiness cost. An innovative Q-learning based estimation of distribution algorithm (QLEDA) is well-designed to address this problem. The QLEDA proposes well-tailored three-stage dynamic decoding and opposition-based learning to decode and promote the job sequence group. To balance global and local searchability of QLEDA, we introduce problem-specific Q-learning and Chebyshev chaotic mapping. To build a probability model of self-adaptation and self-selection, the job sequence group implements discrete actions by interacting with distributed environment and state space through Q-learning. Numerous experiments demonstrate that the QLEDA can generate more satisfactory results over other three well-performing rivals. The finding corroborates the applicability and effectiveness of presented QLEDA in solving the considered problem.