Moving Sustainable Building Operations Toward Carbon Neutrality with Deep Reinforcement Learning: Balancing Energy Savings, Multi-Dimensional Indoor Comfort, and Carbon Permit Revenue Performance

The concept of carbon-neutral buildings encompasses not only carbon emission reductions but also sustainability. Building sustainability includes the physical durability of the structure, the health and safety of its tenants, and harmony with the surrounding environment. The achievement of these goals requires alignment among diverse stakeholders associated with buildings; however, such alignment is limited by economic (cost), environmental (global warming), and social (institutions and policies) factors. This study proposes an operation model that integrates buildings, the carbon permit market, and deep reinforcement learning (DRL) to address these limitations. The DRL model reduces energy consumption while maintaining indoor comfort, generates carbon permits equivalent to the amount of energy saved, and creates a new revenue stream by selling them. To achieve more precise comfort management, the model incorporates a policy that combines predicted mean vote (PMV) and Humidex. In the context of a privately owned commercial office building, the DRL model achieved indoor comfort levels of 98.51% for PMV and 97.22% for Humidex, while reducing energy consumption by 34,376 kWh, lowering carbon emissions by 26,607 kgCO2eq, and generating USD 176 in carbon permit revenue. These results translated into a total reduction in operating costs of 7.5%, amounting to USD 2951. Consequently, the proposed approach provides cost reductions for building owners, comfort for tenants, efficiency for managers, and carbon emission reductions that contribute to carbon neutrality.