EVALUATING THE CAPABILITIES OF SURROGATE MODELING TECHNIQUES IN PREDICTING HOURLY BUILDING ENERGY CONSUMPTION

Building energy simulation models have strong energy prediction capabilities but suffer from high computational costs, which could be reduced through surrogate modeling approaches. Existing surrogate models predict energy consumption on an annual resolution, however, for strategizing net-zero measures, granular predictions are necessary. This paper evaluates the ability of four stateof-the-art machine learning algorithms to predict building energy consumption on an hourly basis. The results indicate that Random Forest Regression is the most suitable predictive model due to the high R2 value of 0.94. The proposed framework can be further expanded to test net-zero energy retrofits at minimal computational costs.