Forecasting hourly electricity demand of Uruguay for the next day using artificial neural networks

This article presents different models applying computational intelligence to forecast the total hourly electricity demand of Uruguay for the next day. Short term electricity demand forecasting is crucial to optimize the economic dispatch of electricity generation, improving the rational use of resources. It also allows improving energy efficiency and demand response policies related with smart grids. Classical statistical models have been applied to predict electricity demand but with the recent development of computational hardware and the vast amount of data available from various sources, computational intelligence models have emerged as successful methods for prediction. In this article, two artificial neural network architectures are presented and applied to forecast the total electricity demand of Uruguay for the next day. The first architecture combines Long Short Term Memory units (LSTM) with fully connected neural networks layers, and the second architecture improves the first by adding a Convolutional Neural Network as first layer (CNN+LSTM). Both architectures use a dropout technique to avoid overfitting. An ExtraTreesRegressor model is used as benchmark to evaluate both architectures. Three steps of data preprocessing are carried out, including treating missing values, removing outliers, and standardization. Considering the high computing demands of the applied techniques, they are developed and executed on the high performance computing platform provided by National Supercomputing Center (Cluster-UY), Uruguay. Standard performance metrics are applied to evaluate the proposed models. The experimental evaluation reports successful forecasting results: the CNN+LSTM model has a mean absolute percentage error of 4.3% when applied to the prediction of unseen data.