Multi-Seasonal Modeling of Energy Consumption in a Building: Integration of Sub-Models with Switching

This paper introduces a novel approach for predicting electric energy consumption in buildings throughout the year using meteorological variables such as outdoor temperature, indoor temperature, atmospheric pressure, and outdoor humidity. Specifically, the method involves the development of a comprehensive seasonal Random Forest (RF) model that dynamically adjusts to different seasons-winter, summer, spring, and autumn-based on the calendar date. Furthermore, the sub-models demonstrate robust performance; for instance, the summer model achieves a root mean square error (RMSE) of 59 Wh during testing, while the winter model reaches a coefficient of determination (R2) of 95 % and 61 % in the training and testing phases respectively. Ultimately, this multi-input modeling approach provides significant improvements in reconstructing seasonal energy consumption patterns. To evaluate the performance and effectiveness of the proposed approach, we employed cross-validation, ensuring a comprehensive analysis of our sub-models across all data samples. This method led to reliable results across the four folds for all seasons, with an R2 of 65% and 60% in fold 4 for the winter and autumn seasons, respectively.