Predicting Building Energy Consumption Using Stacking-Based Ensemble Model

Abstract

With the development of urbanization, the smart city is gradually receiving widespread attention. The construction industry accounts for nearly half of China's total carbon emissions, which is one of the main contributors to the greenhouse effect. Accurate prediction of building energy consumption is of vital significance for energy conservation and smart city development. However, the traditional statistical methods for building energy consumption prediction have some problems such as low fitting accuracy and inaccurate prediction results. Machine learning algorithms are developing rapidly, which have significantly improved the predictive accuracy of building energy consumption. In this study, a stacking-based ensemble learning model based on architectural features is proposed. The building energy consumption mainly consists of the heating load and cooling load, which are used as target variables to predict the building energy consumption. Firstly, the normalized preprocessing is performed on the raw data. Subsequently, the ensemble model is obtained by integrating the optimal base predictors using stacking-based ensemble learning method. In the experiments, the dataset from the building energy area is tested with three metrics to evaluate the performance of the proposed model in building energy consumption prediction. The experimental results show that the proposed ensemble model outperforms the base predictors in solving the building energy consumption prediction problem.