Prediction of Energy Consumption in the Buildings Using Multi-Layer Perceptron and Random Forest

Abstract

© 2017 SERSC. The aim of the paper is to facilitate energy suppliers to make decisions for the provision of energy to different residential buildings according to their demand, which will enable the energy suppliers to manage and optimize the energy consumption in an efficient manner. In this paper, we have used Multi-layer perceptron and Random Forest to classify residential buildings according to their energy consumption. The hourly consumed historical data, of two types of buildings, have been predicted: high power and low power consumption buildings. The prediction consists of three stages: data retrieval, feature extraction, and prediction. In the data retrieval stage, the hourly consumed data based on the daily basis is retrieved from the database. In the feature extraction stage, statistical features; mean, standard deviation, skewness and kurtosis are computed from the retrieved data. In the prediction stage, Multi-Layer Perceptron and Random Forest have been used for the prediction of high power and low power consumption buildings. The hourly consumed historical data of 400 residential buildings have been used for experimentation. The data was divided into 70% (280 buildings) training and 30% (120 buildings) testing. The Multi-Layer Perceptron achieved 95.00% accurate result, whereas the accuracy observed by Random Forest was 90.83%.