Diversity-enhanced hyper-heuristics for multi-objective dynamic flexible job shop scheduling

In the realm of multi-objective dynamic flexible job shop scheduling (MODFJSS), the prevalent reliance on genetic programming based hyper-heuristics (GPHH) has been identified as a bottleneck with quality-limited and redundant heuristics. To deal with these issues, this study introduces a novel approach named Diversity-Enhanced Hyper-Heuristics (DEHH). Our methodology encompasses three strategic thrusts: First, we introduce a multi-grained knowledge (MGK) method to represent knowledge more accurately. Second, we propose an explicit knowledge sharing (EKS) mechanism coupled with surrogate models to discern a diverse set of problem-relevant knowledge. Third, we design a multiple Pareto retrieval (MPR) mechanism to curb the proliferation of duplicate heuristics during evolution. Through comprehensive experimentation, we demonstrate that DEHH achieves superior generalization ability and diversity performance across various scenarios compared with state-of-the-art GPHH algorithms.