Double-Learning-Strategy-Based Evolutionary Algorithm for Scheduling Multiobjective Distributed Assembly Permutation Flowshops With Setup Time

This study addresses an energy-efficient multiobjective distributed assembly permutation flowshop scheduling problem with sequence dependent setup time. The objectives are to minimize the maximum completion time (makespan), mean of earliness and tardiness, and total carbon emission, simultaneously. First, a mathematical model is established. Second, the double-learning-strategy-based Jaya algorithms are developed to address the problems. According to problem-specific nature, one Q-learning state-action strategy is designed to guide nondominated solutions choosing appropriate machine speed adjustment strategies for achieving a satisfactory tradeoff among the three objectives. Third, eight neighborhood structures are designed and embedded in the proposed Jaya algorithms to discover high-quality solutions in local spaces. Fourth, another three novel Q-learning state-action design strategies are proposed to dynamically select the appropriate neighborhood structures during iterations, which introduce the searching directions and improve the convergence of the proposed Jaya. Finally, 81 benchmark instances are solved and the effectiveness of improved strategies is demonstrated. The proposed Jaya algorithm with the best double Q-learning strategies is compared to a solver, Gurobi, to verify the developed mathematical model. The experimental analysis demonstrates that the improved Jaya algorithm with both the Q-learning based machine speed adjustment and the Q-learning-based neighborhood selection strategies shows the best performance.