THE IMPACT OF IMPUTED WEATHER DATA DERIVED FROM ADVANCED MACHINE LEARNING MODELS ON BUILDING ENERGY SIMULATION

Accurate weather data are fundamental in building energy simulations, as they directly affect predictions of energy needs and efficiency. Localized weather data from site-specific weather monitoring stations are especially critical as they ensure that simulations accurately reflect the true environmental conditions each building will face. Localized weather data are also essential for the calibration of building energy simulation models, which are important for analyzing the specific characteristics of the building, thereby enabling more precise and reliable predictions of energy performance. This accuracy can be affected by equipment malfunctions, environmental factors, and occasional human errors. While broader meteorological models offer valuable data, they typically do not provide the detail required for site-specific simulations. In this current study, the author focused on the impact of imputing missing weather data for building energy simulations. A comprehensive meta model combining different machine learning algorithms was developed to improve data imputation accuracy. In the study, the author assessed the impact of various levels of imputed weather data, ranging from 5% to 30% in 5% increments, on the accuracy of heating and cooling load simulations using EnergyPlus software. In addition, the author used a detailed sensitivity analysis, using 30% imputed weather data, to examine its influence on building energy simulations and identify key weather parameters affecting total heating and cooling load computations. This approach emphasizes machine learning’s potential in enhancing building energy simulations by effectively imputing weather data with partial gaps. The findings suggest that the imputed data levels critically affect building energy modeling, making the choice of machine learning models and their training important.