Household Electricity Consumption Prediction Under Multiple Behavioural Intervention Strategies Using Support Vector Regression

Abstract

Household electricity consumption influenced by various behavioural intervention strategies is difficult to predict due to the uncertainty arises from involved human behaviours. Based on an energy conservation experiment conducted in Hangzhou, China, this paper firstly proposes a variable selection approach to determine the best subset of consumption predictors using Akaike Information Criterion (AIC). 18 of the 48 initial variables have been considered as the critical predictors including energy behaviours, personality trait, demographic/building features, weather indicators and the last month consumption in this research. Moreover, this research also introduces the interaction effect between the energy behaviour predictors and other variables to the prediction model. The study has developed an energy behaviour based Support Vector Regression (SVR) model that is capable of predicting household electricity consumption under multiple intervention strategies. In particular, Gaussian radial basis function (RBF) is applied as the kernel function of SVR model. The result shows that the proposed model has the best and robust performance on the next month prediction and time-series forecasting.