A generalizable deep reinforcement learning-based algorithm for priority rule scheduling in stochastic resource-constrained project scheduling

In practical construction project scheduling, deterministic project information is available at the planning stage, and uncertainty arises during project execution. For the stochastic resource-constrained project scheduling problem (SRCPSP), existing approaches typically infer probability distributions from historical data or expert judgment during planning. However, accurately estimating such distributions is often difficult in practice. Accordingly, this study proposes a deep reinforcement learning (DRL)-based framework for online solution of the SRCPSP. The scheduling process is formulated as a Markov decision problem, and a multi-instance trained DRL agent is developed to dynamically select priority rules (PRs) during execution, based on deterministic information available at the planning stage and stochastic realizations encountered during execution. The computational experiments on a project scheduling problem library (PSPLIB) and a real-world case study show that the proposed method outperforms all traditional PR-based methods for the SRCPSP under five different distributions, validating its effectiveness, practicality and generalizability.