Reinforcement learning-based optimization framework for carbon emission reduction and circularity enhancement in buildings

The increasing demand to reduce carbon emissions and resource consumption in the construction sector has necessitated the integration of circular economy (CE) principles with climate change mitigation (CM) strategies. This study proposes a novel framework that integrates the CE and CM objectives by identifying optimal policies using reinforcement learning (RL), specifically Q-learning. The environmental impacts (EIs) were assessed across the entire building life cycle using a cradle-to-cradle (C2C) system boundary. The effectiveness of the framework was demonstrated through a case study involving three alternatives for exterior materials, three for window systems, and two for structural systems. By incorporating the actions related to the Rs framework—repair, replacement, reuse, and recycling—into the RL model, the study derives optimal maintenance and replacement strategies throughout the building life cycle. When compared with baseline scenarios—none (no repair actions) and 10_minor (repairs made every 10 years)—the RL-based policy reduces the global warming potential by up to 42.9 % and 23.1 %, with average reductions of 12.5 % and 5.7 %, respectively. It also decreases the total waste generation by 29.6 % and 15.5 %, respectively. This research promotes methodological approaches to sustainable building analysis by providing a data-driven framework that integrates RL with life cycle environmental assessment. It supports informed decision-making by identifying strategies that enhance durability and circularity while minimizing the environmental burden.