Forecasting Building Energy Consumption Using Statistical Models Incorporating Operational and Environmental Factors

As global and local efforts tackle energy consumption and environmental sustainability, it is crucial to conduct detailed studies on energy demand. This study investigated the effects of wind, relative humidity, temperature, precipitation, and the number of operating days on the monthly energy consumption of a specific building using statistical techniques such as Pearson correlation analysis and time series modeling. Seasonal-trend decomposition using LOESS (STL) was utilized to model the deterministic component in the data and seasonal autoregressive integrated moving average with exogenous variables (SARIMAX) models to further capture the seasonality of energy consumption while taking account of the external effects of weather and operational factors. The forecasting accuracy of the models was benchmarked to naive modeling in terms of normalized Root Mean Squared Error (nRMSE) and Mean Absolute Error (nMAE), Mean Absolute Percentage Error (MAPE), and Skill Score (SS). The results indicate that among the exogenous variables, only the number of operating days significantly correlates with the target variable. Ensemble technique and inclusion of operating days, wind speed, ambient temperature, and total precipitation in the models significantly enhanced the forecasting accuracy. Consequently, the STL-Ensemble 2 model provides optimal forecasting accuracy in predicting building energy consumption with 8.65% nRMSE, 6.84% nMAE, and 7.92% MAPE, which is far superior to the naive model with 27.45% nRMSE, 24.07% nMAE, and 27.75% MAPE, and STL-SARIMA with 10.03% nRMSE, 8.67% nMAE, and 10.21% MAPE. Future research can use more granular data resolution and further explore advanced forecasting methods such as machine learning techniques to achieve improved model performance and realized effects of operational and weather variables.