Uncertainty-Aware Adaptive Control for Human-Centric Cyber-Physical Indoor Spaces *
Year: 2026
Type: conference-paper
Abstract: This paper presents an uncertainty-aware adaptive control framework for indoor HVAC operation under occupancy uncertainty. The method combines a quantile-regression occupancy predictor with conformal calibration, a PMV-based Random Forest thermal-comfort classifier, and a risk-averse interval MPC operating on an identified RC thermal/IAQ model. The occupancy module achieves an MAE of 0.0326 persons with empirical 95% prediction-interval coverage of 0.9516 and mean interval width of 0.3463 persons. The comfort classifier achieves 0.8387 accuracy, 0.8557 balanced accuracy, and 0.8345 Macro-F1. In a 7-day closed-loop evaluation using real operational building data, the proposed UA-MPC reduces temperature violations from 47 to 30 and CO2 violations from 47 to 29 relative to a certainty-equivalent MPC, at a 23.5% energy overhead versus CE-MPC but with 92.7% lower energy use than a rule-based controller. Additional analysis over five representative two-day windows suggests that these gains are not limited to a single operating window. These results position UA-MPC as a robust comfort- and IAQ-oriented alternative to deterministic building control under uncertain occupancy.
Authors: Zahra Elmi, Soheila Elmi, Semira Elmi
Institutions: Beykoz Üniversitesi, Koç University, Beykent University
Language: English
Subfield: Building and Construction
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7201833112