Dynamic construction site layout optimization using deep reinforcement learning with PPO

Purpose Efficient facility layout planning is a critical component of construction site management, directly impacting productivity, safety and cost-effectiveness. Traditional approaches often face challenges in managing spatial constraints, varying facility shapes and avoiding overlaps. This study aims to address these limitations by employing proximal policy optimization (PPO), a reinforcement learning algorithm, to dynamically optimize facility placement on construction sites. Design/methodology/approach A custom simulation environment was developed to model a realistic construction site layout. PPO was implemented to train the model on optimal facility allocation strategies, focusing on reducing overlaps and maximizing space utilization. Additionally, primary data were collected via a structured survey, and the responses were analyzed using SPSS. Appropriate statistical techniques were applied to validate the findings and inform the model parameters. Pareto analysis was used to identify the most influential factors affecting layout planning, forming the basis for the reinforcement of the learning environment. Findings The PPO-based model demonstrated strong performance in allocating facilities with minimal spatial conflicts and efficient use of available area. The analysis revealed that the model successfully reduced overlap occurrences while improving grid utilization. Statistical validation supported the model's effectiveness and confirmed the significance of the identified factors. Originality/value This study contributes a novel, hybrid approach that integrates data-driven statistical analysis with deep reinforcement learning for dynamic site layout optimization. Unlike conventional static models, this framework allows real-time adaptation to change in facility requirements and site constraints. The integration of survey-based data and PPO enhances the model's relevance to practical construction management scenarios.