Forecasting daily natural gas consumption with regression, time series and machine learning based methods

An effective short-term natural gas forecasting method contributes to social contributions and allows industrial chain elements to function effectively and minimize economic losses. We dealt with a comparative framework on the applicability of different methods in daily natural gas service (NGS) consumption forecasting. In this context, time series, machine learning, evolutionary and population-based approaches, and their hybrid versions are applied to the NGS data. Hybridized approaches are tested in the scope of NGS consumption forecasting for the first time in the literature in this study. The case of Turkey is handled, and its NGS data is used to demonstrate the comparative framework's applicability. The comparative study is assessed in the lights of common forecasting accuracy measures of mean absolute percentage error (MAPE), R-squared (R2), and mean squared error (MSE). According to each method's results, the seasonal autoregressive integrated moving average with exogenous regressors (SARIMAX) and artificial neural network (ANN) hybrid model provides the most dominant performance with respect to MAPE. The lowest error was obtained with a MAPE value of 0.357 in this hybrid model constructed under seven neurons in its ANN structure. This model is followed by another hybrid model, autoregressive integrated moving average (ARIMA)-ANN, with a MAPE value of 0.5 under nine neurons in terms of accuracy performance. The worst performance value belongs to the Genetic algorithm-ANN hybrid model with a MAPE value of approximately 26%.