A Hybrid Metaheuristic Algorithm for Multiobjective Scheduling in Distributed Heterogeneous Flexible Job Shops

To address the issues of insufficient population diversity and the tendency to fall into local convergence in the NSGA-II algorithm for distributed heterogeneous flexible job shop scheduling, a dual-population improved algorithm, D3QN-NSGA-II, based on D3QN is proposed. By introducing a dual-population evolution strategy, the breeding population and the regular population are divided using a reproductive capacity assessment mechanism, with differentiated crossover and mutation operators designed for each. A hybrid strategy, consisting of 12 types of local search operators across 3 categories, is constructed, and the innovative application of D3QN deep reinforcement learning for dynamic operator selection is introduced. Additionally, an energy-efficient scheduling strategy is applied to optimize machine idle time. Finally, the effectiveness of the D3QN-NSGA-II algorithm is evaluated using 20 instances, with experimental results demonstrating its superiority over several other advanced algorithms in solving the DHFJS problem.