Abstract:

For the cascaded hybrid flow shop scheduling problem with AGV transportation, a multi-objective optimization model is built to minimize the makespan and total energy consumption. A hybrid algorithm framework integrating the LLM, covariance matrix adaptation evolution strategy, and RL strategy selection mechanism is proposed. LLMs are employed to generate heuristic priority rules and construct high-quality initial solutions. Covariance matrix adaption evolution strategy is adopted to optimize the search in continuous priority space, and an enhanced scalar fitness function integrating hypervolume gain and solution diversity is introduced. An RL strategy selection module based on the upper level information boundary is designed to adaptively adjust the search starting point and operator selection and thus enhance the algorithm's exploration and development capabilities. Simulation results show that the proposed algorithm outperforms the comparative algorithms in terms of evaluation metrics.

Key words: cascaded hybrid flow shop, AGV, LLM, evolutionary computation, RL