Intelligent Systems for Space Conditioning and Shading in Buildings: A Systematic Review of Artificial Intelligence Techniques and Their Energy Impact

This is the authors' abstract. We don't add key points for this paper.

Artificial intelligence (AI) is increasingly applied to heating, ventilation, and air-conditioning (HVAC) control, whereas solar-radiation management still relies mainly on reactive strategies; whether AI capabilities validated in HVAC could be transferred to solar control has not been systematically assessed. This review compares both domains in terms of experimental maturity, climatic coverage, and input variables. Web of Science studies (2015–2026) reporting quantitative energy-demand variations were systematically reviewed following PRISMA 2020 and synthesized descriptively, with a sensitivity search adding reinforcement-learning and predictive-control terms. Fifty-six studies were included (29 HVAC, 28 solar, 1 shared). HVAC reductions ranged from 3.8% to 62.4% (median 22.1%), a spread associated more with heterogeneous baselines, durations, and metrics than with techniques. Real-world studies represented 31% in HVAC versus 14% in solar management, where AI-based predictive control appeared in one real-world study. Anticipatory variables were scarce but coincided with some of the highest savings, a non-causal association. Model-level patterns, such as Random Forest replication, were not preserved when 18 additional studies were considered. Transferring predictive control from HVAC to solar management is a promising line that requires long-term real-world validation in underrepresented climates. Limitations include a single database and single-reviewer screening. The review was not registered or externally funded.