Hybrid AI-based comparative forecasting of building heating and cooling demand across various envelope configurations: implications for energy consumption, CO₂ emissions, and costs
Аннотация
Forecasting building Heating/Cooling demand is essential for designing low-carbon, energy-efficient, and resilient buildings. Yet, it remains challenging to select the most accurate forecast model. Efficient evaluation and understanding of how the building envelope affects performance metrics depend on this selection. The present study investigates three different forecast models and their hybridization. These include a statistical SARIMA (Seasonal Autoregressive Integrated Moving Average) model, a deep learning LSTM (Long Short-Term Memory) model, and the decomposable Prophet time series model. A selected hybrid forecasting model is also examined for predicting energy demand. The proposed models were applied to a real monthly energy consumption dataset for different residential buildings, namely A, B, C, and D, each with 80 m 2 area, active/passive factors, and operational and climatic conditions, with the only difference being the thermophysical properties of their envelopes. Based on the results, the SARIMA model shows the highest precision for heating energy forecasting, with the lowest scattering index (SI) of 0.03; otherwise, the hybrid SARIMA-Prophet model achieves the highest accuracy for cooling energy forecasting, with an SI of 0.18. The ten year forecasted scenarios are applied to evaluate the long-term impact of the envelope’s thermophysical properties on buildings’ energy efficiency, CO2 emissions, and operating costs. Compared to the reference building (Building A), which is insulated only by clay brick wall insulation, the building insulated by an airgap layer (Building D) demonstrates a 27% reduction in energy usage, a 28.67% reduction in CO2 emissions, and a 55.17% reduction in operating costs. Overall, these findings clearly indicate that the comparative application of statistical, artificial intelligence-based, and hybrid models provides strong evidence for accurately predicting energy demand, assessing building-envelope performance, and well-informing future energy management decisions.