Generation of representative meteorological years through anomaly-based detection of extreme events

Abstract

Typical Meteorological Years (TMYs) have long supported the building sector by integrating local climate into building design for energy, thermal comfort, and peak load assessments. As climates shift, past heat waves and cold spells signal future conditions requiring greater adaptability. This study proposes a new file generation method that preserves TMY properties while embedding extreme events. We combine three anomaly-detection methods—temperature thresholds, Graph Neural Networks (GNNs), and Extreme Value Theory (EVT)—to capture climatic deviations, detect anomalies, and model statistical extremes. An integrated hierarchical method forms the new Representative Meteorological Year (RMY) file. RMY files for six ASHRAE climate-zones consistently capture past extremes, producing worst-case scenarios for key metrics, including peak loads, indoor thermal stress, natural ventilation and outdoor comfort. The largest deviation between the TMY and RMY was a doubling of indoor thermal stress hours across all climates, while average energy use remained aligned, with a deviation of 6%.