| Title |
A Meta-Analysis of AI·ML HVAC Control Effects for the Development of a Railway Station Energy Control Virtual Testbed (RS-ECVTB) |
| DOI |
https://doi.org/10.5370/KIEE.2025.74.9.1599 |
| Keywords |
Railway Station; RS-ECVTB; Energy Optimization; HVAC; Artificial Intelligence; Machine Learning; Model Predictive Control; Meta-Analysis |
| Abstract |
To secure fundamental data necessary for developing the Railway Station Energy Control Virtual Testbed (RS-ECVTB), this study conducted a systematic meta-analysis of AI·ML-based Heating, Ventilation, and Air Conditioning (HVAC) control effects in large public facilities. Railway stations face limitations with conventional control methods due to unique operational conditions such as 24-hour continuous operation and fluctuating passenger density, necessitating quantitative analysis of implementation effects of AI·ML-based intelligent control techniques. Following PRISMA guidelines, studies published from 2015 to 2024 were collected, and through rigorous selection processes, 267 studies were finally selected for meta-analysis from an initial 1,385 papers. Among these, 134 studies (50.2%) directly targeted railway stations, while the remainder consisted of research on large public facilities with similar characteristics. The meta-analysis revealed that AI·ML-based HVAC control techniques achieved an average energy reduction of 17.4% compared to conventional control methods (95% CI: 15.8-19.0%). Performance analysis by technique revealed that hybrid approaches (RL+MPC) demonstrated the best performance at 34.2%, followed by Model Predictive Control (MPC) at 28.3%, and Deep Reinforcement Learning (DRL) at 26.8%. Notably, above-ground stations demonstrated higher effectiveness than underground stations (24.8% vs 19.3%), and BEMS-integrated systems achieved 28.7% reduction, representing an 8.9 percentage point improvement over non-integrated systems. The average energy reduction of 17.4% and superior performance of hybrid approaches (34.2%) derived from this study can serve as reference data for setting performance targets and selecting algorithms when developing RS-ECVTB. Particularly, the additional 8.9%p effect from BEMS integration and the importance of high-resolution data should be reflected as core requirements in RS-ECVTB system design. The average payback period of 3.2 years from economic analysis provides investment feasibility evidence for responding to the mandatory ZEB certification policy for railway stations in 2025. |