Predicting Electric Energy Use of a Low Energy House: A Machine Learning Approach

Abstract

Electricity is one of the greatest inventions of modern science. Now-a-days, without electricity, life is unthinkable. However, the production and distribution of electricity is costly. So, it is necessary to use this facility with great care. Moreover, people in many of the countries using prepaid electric meters in their houses, offices, stores, even in factories, where they have to buy electricity in pre-paid basis. To do so, they have to estimates the future use of electricity in their vicinity. In this study, we have focused on predicting electric energy use of home appliances in a low energy consumption house. Two very popular and computationally effective machine learning algorithms, namely Support vector regression (SVR) and Artificial neural network with back-propagation (BP-ANN) are applied with cross validation approach. Principal component analysis (peA) is used to solve data dimensionality problems to analyze and select input features. In addition, F-test and correlation analysis are also done to check the dependencies of dependent variables (target label) on independent variables (predictors). Finally, the proposed model able to predict electric energy uses with more that 98% accuracy.