Multiple machine learning models for predicting annual energy consumption and demand of office buildings in subtropical monsoon climate

Abstract

Reducing a building’s energy use has many real-world applications. An early-stage design could have a quantitative foundation for energy-saving designs if energy consumption could be predicted quickly and accurately. The main issue that designers are currently dealing with is the incompatibility of building modelling and energy simulation software. In order to realize the flexibility of building energy systems, accurate and timely thermal load prediction for buildings is essential. Here, three machine learning (ML) models – Artificial Neural Network (ANN), Random Forest (RF) and Extreme Gradient Boosting (XGBoost) were used, for forecasting an office building’s load demand and energy usage. A case study building was selected and analysed via Autodesk Revit and Green Building Studio. For the modelling of ANN, 438 simulated data samples were created based on different design parameters considering different window, wall, roof materials and window to wall ratio, and meteorological conditions considering dew point, dry bulb, wet bulb temperature and relative humidity of seven major cities in Bangladesh. The findings show that the ANN model performs best the best in predicting annual electricity use with an R² value of 0.991 and annual load demand with an R² value of 0.995. The RMSE values ranged between 3.83 and 5.10 showing high accuracy of prediction between the three ML models. Afterwards SHAP analysis was used to analyse the input features effect on the energy consumption. Findings show that relative humidity, dry bulb temperature and pressure significantly affects the energy consumption.