A Model for Predicting Building Energy Consumption Based on the Stacking of Machine Learning Regression Models
الموضوعات : Journal of Computer & RoboticsMohammadHosein Khodadadi 1 , Ladan Riazi 2 , Samaneh Yazdani 3
1 - Department of Information Technology Management, Science and Research Branch, Islamic Azad University, Tehran, Iran
2 - Department of Information Technology Management, Science and Research Branch, Islamic Azad University, Tehran, Iran
3 - Department of Computer Engineering, North Tehran Branch, Islamic Azad University, Tehran, Iran
الکلمات المفتاحية: Regression, energy consumption, MLP, XGBoost, Stacking,
ملخص المقالة :
In different societies, buildings are considered one of the main energy consumers in the world, and accordingly, they are responsible for a significant percentage of greenhouse gas emissions. Due to the upward growth of the population, the demand for energy consumption is increasing day by day. In such a situation, the prediction of energy consumption has become a vital issue to control the efficiency of energy consumption. To obtain an effective solution to solve this problem, a number of machine learning methods were examined and Xgboost and MLP methods were selected as the best available methods. In order to obtain more suitable results in this research, a system based on stacking was proposed. In the proposed method based on stacking, XGBoost and MLP methods were used in the first level so that the advantages of both methods can be used. The predictions made by each of these methods, in the second level, were used as input to another XGBoost algorithm, which was used as a meta-learner. To obtain better results, the hyperparameters of the basic techniques were optimized using the successive halving search. For a better comparison, machine learning regression techniques were implemented to solve the problem of energy consumption intensity prediction, and the results obtained from them were analyzed on WiDS Datathon. The results showed that the proposed system has improved the MAE, MAPE, and R2 criteria by 0.6, 0.03, and 0.07, respectively, compared to the best existing method.Keywords: Energy Consumption, Stacking, Regression, XGBoost, MLP.
