基于深度强化学习的模块化集成建造车间实时调度方法研究OA

Modular Integrated Construction(MIC)represents an emerging construction paradigm that has gained widespread a-doption in the production of building components.Given the growing demand for customized component products and the intri-cate,dynamic nature of the construction workshop environment,there is an urgent need to develop advanced real-time scheduling methodologies capable of adapting to novel production modes and responding effectively to dynamic events.A real-time scheduling approach based on Deep Reinforcement Learning(DRL)was proposed for modular integrated construction shop scheduling.First,the production process and characteristics of the modular integrated construction workshop were systematically analyzed,ab-stracted as a hybrid-flow production system,and formalized through a relevant mathematical model.Second,by defining scheduling decision points within the production time series,the scheduling problem was formulated as a Markov Decision Process(MDP).Subsequently,a comprehensive state space encompassing 21 production features,8 action spaces,and reward functions derived from Genetic Programming(GP)complex rules were sequentially designed.Building on this foundation,an algorithm based on Proximal Policy Optimization with Dual Memory Pools(PPO-DMP)was proposed to train scheduling agents,enabling efficient mapping between production states and scheduling strategies,thereby achieving effective optimization of scheduling ob-jectives.Finally,comparative experiments demonstrate that the proposed real-time scheduling algorithm exhibits superior schedu-ling efficiency and dynamic adaptability compared to traditional methods,particularly in scenarios involving new order insertions,where its advantages become even more pronounced.