Time series prediction with autoencoding LSTM networks

Nowadays, solving prediction problems in green computing is an open and challenging task, for which solutions based on deep learning are studied. In this work, we present a forecasting algorithm based on Long Short-Term Memory networks applied to renewable energy sources time series prediction. We make use of an encoder-decoder structure to extract useful representative sequence data, employing a stacked LSTM architecture for data embedding and successive prediction. By comparing the performance of the proposed forecasting scheme with a classical twolayer LSTM structure, we are able to asses the performance of the former as a robust tool for solving prediction problems in the green computing framework.