A Non Linear Autoregressive Neural Network Model for Forecasting Appliance Power Consumption

Abstract

Forecasting the electrical appliance power consumption is a necessary and important part of the management of electrical power system, in order to assess people's penchant for using electricity. Even though several studies are focused on forecasting building consumption, less attention is given to forecasting the use of single appliances. Indeed, some of the energy needs of consumers may be relatively delayed or anticipated to obtain a better consumption profile while maintaining consumer comfort. This paper focuses on forecasting appliance power consumption using a non-linear autoregressive (NAR) neural network model. The results obtained on the UK-DALE public dataset demonstrate that NAR models are suitable for forecasting of energy consumption with a good accuracy. The proposed model obtained the best Mean Absolute Errors, compared with the LSTM, Autoencoder, Combinatory optimization, FHMM, and Seq2point techniques.