Performance evaluation of LSTM neural networks for consumption prediction

Energy consumption and energy efficiency are topics that have attracted the attention of researchers in recent years, in order to seek sustainable solutions for energy production and reduction of costs, aiming to provide a balance between development and protection of natural resources. One of the alternatives that have obtained satisfactory results is the use of technologies based on Internet of Things (IoT) and Deep Learning systems. Based on this, we assessed the performance of Long Short-Term Memory (LSTM) neural networks in time series electric energy consumption prediction, for a forecasting module of an IoT system. Three time series were used and we compared LSTM to the algorithms Extreme Boost Gradient and Random Forest. Computational results indicate that the LSTM model showed a tendency of better RMSE performance in the first data set, and statistically significant better results in other two data sets, according to the Kruskal-Wallis test (p