Enhancing building energy efficiency using a random forest model: A hybrid prediction approach

TLDR

Building envelope design significantly influences building energy consumption. The study proposes a method to predict building energy consumption from envelope design to improve energy efficiency. The authors built a BIM of a university teaching building, simulated energy use for six envelope parameters via orthogonal testing, trained a random‑forest model to predict consumption and rank parameter importance, and compared its performance to BP‑ANN and SVM. The RF model identified exterior wall and window heat transfer coefficients and window‑wall ratio as the most influential parameters and outperformed BP‑ANN and SVM, making it the best predictor of building energy consumption.

Abstract

The building envelope considerably influences building energy consumption. To enhance the energy efficiency of buildings, this paper proposes an approach to predict building energy consumption based on the design of the building envelope. The design parameters of the building envelope include the comprehensive heat transfer coefficient and solar radiation absorption coefficient of exterior walls, comprehensive heat transfer coefficient and solar radiation absorption coefficient of the roof, comprehensive heat transfer coefficient of outer windows, and window-wall ratio. The approach is applied to optimize the design parameters of the building envelope structure of a university teaching building in northern China. First, a building information model of a teaching building is established in Revit and imported into DesignBuilder energy consumption analysis software. Subsequently, a data set of the abovementioned 6 parameters is obtained by performing orthogonal testing and energy consumption simulations. On this basis, an RF model is used to predict building energy consumption and rank the importance of each parameter, and the Pearson function is used to evaluate the corresponding correlations. The results show that the most important parameters with the highest correlations to building energy consumption are the comprehensive heat transfer coefficients of the exterior walls and outer windows and the window-wall ratio. Finally, the RF prediction results are compared to the prediction results of a BP artificial neural network (BP-ANN) and support vector machine (SVM). The findings indicate that the RF model exhibits notable advantages in building energy consumption prediction and is the optimal prediction model among the compared models.