LSTM recurrent neural network for energy demand forecasting

Abstract

Abstract—Recurrent Neural Networks (RNN) of the Long Short Term Memory (LSTM) type provide high accuracy in predicting sequential models in various application domains. As in most process control problems, their dynamics include non manipulated variables that need to be predicted. This paper proposes using an LSTM neural network for energy demand forecasting, which applies to an Economic Model Predictive Control (EMPC) as a forecasting tool. For the training, data are taken from a three-phase intelligent power quality analyser located at the National Technological University, Reconquista Regional Faculty (Santa Fe, Argentina). A recursive strategy is used to update the state of the neural network and forecast over different prediction horizons. The accuracy achieved in training the neural network is measured using the root mean square error (RMSE) metric. Experimental results show that the proposed LSTM neural network has excellent generalisation capability.