Analysis of intelligent construction scheduling optimization based on reinforcement learning and its carbon emission reduction effect

Construction scheduling must balance project duration, resource allocation, and environmental sustainability under dynamic constraints. This study proposes a carbon-aware reinforcement learning–based scheduling framework using a Proximal Policy Optimization actor–critic model with task-level action masking to optimize makespan, resource feasibility, and carbon emissions simultaneously. Experimental results demonstrate that the proposed framework reduces total project duration to 600 days, achieving a 20.3% improvement over the Critical Path Method (CPM), while delivering a 43.6% reduction in total CO2 emissions compared with CPM, Genetic Algorithm, and Earliest Start methods. The framework also eliminates resource conflicts entirely, ensuring fully feasible scheduling decisions. These findings confirm that integrating carbon emission modeling directly into reinforcement learning enables adaptive, multi-objective construction scheduling that improves operational efficiency while significantly reducing environmental impact. The study establishes a practical and sustainable decision-support framework for intelligent construction planning.