Toward grid-interactive and low-carbon buildings: A comparative analysis of artificial intelligence-driven optimization of renewable sizing and demand-side control

As buildings account for nearly one-third of global energy consumption, improving their energy performance and renewable integration is essential for achieving sustainability targets. Traditional building energy management systems (BEMS), often rule-based and static, struggle to adapt to fluctuating demands, variable tariffs, and the intermittency of solar resources. This study introduces an integrated, artificial intelligence (AI)-driven BEMS framework that jointly optimizes rooftop photovoltaic (PV) sizing and adaptive demand-side management (DSM) using reinforcement learning (Q-Learning) and benchmarks its performance against two established deterministic tools: HOMER Pro for techno-economic PV sizing and particle swarm optimization (PSO) for DSM load scheduling. Using realistic hourly building loads, meteorological data, and time-of-use pricing, the Q-Learning model converged to a PV–inverter configuration closely aligned with HOMER Pro’s optimum, achieving a slightly lower net present cost (–1.85%) and a modest increase in renewable fraction (+3.1%). In DSM applications, Q-Learning consistently outperformed PSO by shifting a larger share of flexible loads and securing higher daily cost reductions. Under grid-only conditions, Q-Learning reduced energy costs by 7.58% in winter and 8.27% in summer, while PV-integrated scenarios achieved savings of 35.14% and 26.89%, respectively. These results demonstrate that reinforcement learning can effectively enhance the performance of conventional BEMS approaches by providing more adaptive scheduling aligned with tariff structures and solar availability. The proposed framework supports more efficient, flexible, and sustainable building operations, highlighting the practical potential of AI-driven energy management in modern grid-interactive environments.