ADAPTIVE CONTROL OF RESOURCE FLOW IN CONSTRUCTION PROJECTS THROUGH DEEP REINFORCEMENT LEARNING: A FRAMEWORK FOR ENHANCING PROJECT PERFORMANCE IN COMPLEX ENVIRONMENTS

Abstract

The advancements in Deep Reinforcement Learning (DRL) are transforming construction project management, particularly in resource allocation, scheduling, and risk mitigation. Traditional heuristic-based methods struggle with dynamic project environments, necessitating AI-driven approaches. This systematic review, following PRISMA guidelines, evaluates 482 peer-reviewed studies to assess the effectiveness of DRL in optimizing construction workflows. Findings reveal that DRL-based workforce allocation reduces idle time by 30% and enhances labor productivity by 35%, while DRL-driven equipment utilization improves efficiency by 40% and reduces downtime by 28%. Additionally, material logistics optimization through DRL decreases procurement delays and waste by 30%, significantly improving supply chain management. Risk-sensitive DRL models outperform Monte Carlo simulations, reducing cost overruns by 27% and improving risk prediction accuracy by 30%. Comparative analysis confirms that DRL scheduling frameworks, including Proximal Policy Optimization (PPO), Deep Q Networks (DQN), and Actor-Critic models, improve project efficiency by 32%, surpassing traditional CPM and PERT methods. Simulation-based studies further validate that DRL-driven decision-making reduces discrepancies in resource utilization by 21%, while IoT-integrated DRL improves safety compliance by 38% and reduces accident risks by 35%. Despite computational challenges, DRL offers scalability, adaptability, and superior automation, making it a powerful tool for intelligent construction management. This review highlights gaps in empirical validation, AI adoption frameworks, and multi-agent DRL applications, emphasizing the need for further research and industry integration to enhance efficiency, reduce costs, and mitigate risks in construction projects.