Deep Reinforcement Learning Scheduling Method for Prefabricated Bridge Construction

Off-site construction has become a widely accepted method due to its advantages in time-saving, rapid erection, and low cost. The rapid growth in this area demands better and more refined construction scheduling methods. Construction scheduling is complicated by the nature of various constraints in different aspects, such as resources and labour. Traditional methods, such as the Critical Path Method (CPM), lack consideration of various constraints, making them less applicable in real-world projects. This study proposes a Deep Reinforcement Learning (DRL) method to generate optimal construction schedules under limited labour and resource constraints. The objective of this study is to minimize the duration of construction projects. The proposed method introduces an improved DRL framework that enables the DRL to handle the scheduling of all construction processes. A case study is conducted on a real-world prefabricated bridge with 9 spans to evaluate the proposed method's performance. The DRL method is compared with traditional methods and the Genetic Algorithm (GA). The results show that DRL outperformed other methods in generating optimal construction schedules and required less running time. Therefore, the proposed method in this study extends the construction scheduling method and can be used in real projects.