Application of Machine Learning in Forecasting Energy Usage of Building Design

As rapid movement of Information technology, the amount of data and unexpected factors impact on forecasting are gradually uncontrollable, thus, traditional method may be not enough efficient to deal with this issue. Therefore, the development of AI is advantageous for forecasting when data extends significantly. In this paper, machine learning model was developed to estimate energy load based on the characteristics of building design. This determination helps business and engineer reduce energy consumption cost and environmental impact. The data set was collected from energy simulation in the study of Tsanas, A., & Xifara, A. (2012) including 768 observations with 8 inputs and 2 outputs (heating load and cooling load). The energy loads were achieved through innovative methods such as Artificial Neural Network, Support Vector Machine or Random Forest (nonlinear) and Multilinear Regression (linear) with the support of Interactions. The performance of result from neural network technique was quite overfitting with dataset and better than linear as 20% in RMSE. So, it is proposed that using ANN forecasting combine with the predictor variable interaction of Multilinear Regression helps the user analyze the predictive value coordinate with input adjustment. Generally, the research supports the feasibility of machine learning in building energy forecast based on historical data and the building design parameter as well as the possibility to apply to another dataset for prediction purpose.