Applying Long Short-Term Memory Networks for natural gas demand prediction

Abstract

Long Short-Term Memory (LSTM) algorithm encloses the characteristics of the advanced recurrent neural network methods and is used in this research study to forecast the natural gas demand in Greece in the short-term. LSTM is generally recognized by researchers as a key tool for time series prediction problems and has found important applicability in many different scientific domains over the last years. In this study, we apply the proposed LSTM for the purposes of a day-ahead natural gas demand prediction to three distribution points (cities) of Greece’s natural gas grid. A comparative analysis was conducted by different Artificial Neural Network (ANN) structures and the results offer a deeper understanding of the large urban centers characteristics, showing the efficacy of the proposed methodology on predicting natural gas demand in a daily basis.