Forecasting Energy Consumption of Office Building by Time Series Analysis Methods based on Machine Learning Algorithm

Abstract

Building energy consumption can be considered as time series, which are predicted using time series analysis methods. There are lots of traditional time series prediction algorithms, including AR, ARMA and so on. But the building energy consumption series are usually nonlinear and non-stationary. Especially for non-stationary time series the traditional algorithms will not always get good forecasting results. In this paper, we focused on support vector regression algorithm for forecasting time series energy consumption. It was applied to develop prediction models for different types of building energy consumption, including lighting, outlet and air conditioning energy consumption. The optimal model parameters were determined by support vector regression algorithms and experimental results showed the higher prediction accuracy.