Households Energy Consumption Forecasting with Echo State Network

Abstract

AbstractCareful planning and forecasting of energy consumption not only influences a nation’s environmental and energy sustainability, as well as giving a useful basis for policy makers to make decisions. This paper presents the results of an appropriate deep learning model forecasting for consumption using echo state network (ESN). ESN is a new paradigm that offers an intuitive methodology using for time series prediction. Basically, it is a recurrent neural network (RNN) with a vaguely connected hidden layer, known as a reservoir, which functions in a strange way in the existence of time-series patterns. In this contest, three types of recurrent neural network used for comparison with ESN, aiming to evaluate the accuracy and performance of the model. To train and test the proposed model, we used the historical data of multivariate household consumption. ESN showed an improvement of 0.057% and 0.095% in terms of mean square error (MSE) and mean absolute error (MAE), and it is trained faster than other models, in very short-term energy consumption forecasting.KeywordsESNConsumption energyRNN modelsDeep learningForecasting