Q -learning hyperheuristic genetic algorithm for integrated worker overtime and multi-mode instant transportation in seru production scheduling

In volatile markets, order delivery delays and inefficient logistics present major challenges for the Seru production system (SPS). Existing studies frequently overlook worker overtime capacity and flexible departure times, limiting operational agility. To overcome these limitations, this paper investigates the integrated worker overtime and multi-mode instant transportation in Seru production scheduling (WOMT-SPS). A mixed-integer linear programming (MILP) model is developed to simultaneously minimize total batches delay time (BDT) and total costs (TC). To solve this model, a Q-learning hyper-heuristic genetic algorithm (QHHGA) is designed. QHHGA combines a hybrid initialization scheme, a low-level heuristic (LLH) pool comprising eight heuristic operators and tabular Q-learning acting as a high-level strategy (HLS) to adaptively guide search actions. Comprehensive comparisons across 27 test instances demonstrate that QHHGA outperforms five state-of-the-art algorithms in convergence, diversity and solution quality.