Automated construction scheduling using deep reinforcement learning with valid action sampling

The growing demand for buildings and infrastructures requires optimal construction schedules under real-world complexities. This paper presents an automated scheduling optimization method for large construction projects, considering real-world constraints and the need for rescheduling. A Deep Reinforcement Learning (DRL) model with a Valid Action Sampling (VAS) mechanism is proposed to optimize schedules. The method integrates a Graph Convolutional Network (GCN) for feature extraction and includes a reward shaping mechanism to expedite convergence. The proposed method outperforms traditional methods with reduced project duration and runtime in both scheduling and rescheduling cases. This advancement benefits construction managers seeking efficient and flexible project management. The findings inspire future research into broader and more practical construction scheduling solutions utilizing DRL.