Predicting energy consumption in multiple buildings using machine learning for improving energy efficiency and sustainability

Buildings must be energy efficient and sustainable because buildings have contributed significantly to world energy consumption and greenhouse gas emission. Predicting energy consumption patterns in buildings is beneficial to utility companies, users, and facility managers because it can help to improve energy efficiency. This work proposed a Random Forests (RF) - based prediction model to predict the short-term energy consumption in the hourly resolution in multiple buildings. Five one-year datasets of hourly building energy consumption were used to examine the effectiveness of the RF model throughout the training and test phases. The evaluation results presented that the RF model exhibited a good prediction accuracy in the prediction. In four evaluation scenarios, the mean absolute error (MAE) values ranged from 0.430 to 0.501 kWh for the 1-step-ahead prediction, from 0.612 to 0.940 kWh for the 12-steps-ahead prediction, and from 0.626 to 0.868 kWh for the 24-steps-ahead prediction. The RF model was superior to the M5P and Random Tree (RT) models. The RF was better about 49.21%, 46.93% in the MAE and mean absolute percentage error (MAPE) than the RT model in forecasting 1-step-ahead building energy consumption. The RF model approved the outstanding performance with the improvement of 49.95% and 29.29% in MAE compared to the M5P model in the 12-steps-ahead, and 24-steps-ahead energy use, respectively. Thus, the proposed RF model was an effective prediction model among the investigated machine learning (ML) models. This study contributes to (i) the state of the knowledge by examining the generalization and effectiveness of ML models in predicting building energy consumption patterns; and (ii) the state of practice by proposing an effective tool to help the building owners and facility managers in understanding building energy performance for enhancing the energy efficiency in buildings. (C) 2020 Elsevier Ltd. All rights reserved.