FiLM-Conditioned Cross-Attention Multi-Objective Reinforcement Learning Algorithm for Dynamic Seru Production System Reconfiguration

Seru production systems (SPSs), enhancing flexibility and responsiveness compared to traditional assembly lines, face challenges in addressing demand uncertainty characterized by unpredictable and time-varying customer orders. This study proposes a novel bi-objective optimization model for the dynamic seru reconfiguration problem, aiming to simultaneously minimize tardiness and operational cost. We develop a multi-objective reinforcement learning algorithm incorporating a FiLM-conditioned cross-attention mechanism. Computational results demonstrate that our method significantly improves the handling of demand uncertainty, advancing practical production decision-making in dynamic SPS environments.