Impact of extreme heat on building cooling energy demand: a hybrid model based on the coupling of indoor and outdoor climate variations

Cities are complex, multi-scale systems where the built environment, urban microclimate, and human behavior interact dynamically. Among these interactions, the response of building energy demand to extreme heat is a critical feedback loop that impacts urban functional stability and energy security. However, quantifying these cross-sectoral feedbacks—specifically how outdoor thermal variations translate into indoor cooling behavior and energy demand—remains a significant modeling challenge. To address this, we propose a hybrid modeling framework that integrates machine learning with a physics-based building energy balance model to bridge the gap between urban microclimate and building energy consumption. Our approach estimates the power consumption of air conditioning (AC) systems by distinguishing operational states based on the coupling and decoupling of indoor and outdoor climate variations. The framework employs an XGBoost model to identify AC operation within optimal time windows, followed by the Pelt algorithm to detect state transition points and pinpoint exact operational periods. Subsequently, a Resistance-Capacitance (R-C) model is parametrized using coupled indoor-outdoor climate data during AC-off periods, which is then utilized to estimate real-time AC power.

The model was validated against data from a residential building in Beijing, demonstrating good accuracy in both predicting AC operating status and estimating power loads. The hybrid model was then applied to real-world urban scenarios to quantify the impact of extreme heat on cooling demand using only monitored climate variations, independent of direct energy metering data. This research provides a robust quantitative tool for climate-adaptive planning, advancing our ability to model the complex dependencies between urban energy systems and a changing climate.