Daily basis mid-term demand forecast of city natural gas using univariate statistical techniques

Abstract

City distribution companies or companies with high consumption are required to report monthly consumption demand forecasts for the year ahead and year based daily consumption demand forecasts in natural gas sector. This paper studies forecasting daily and monthly demand for mid-term natural gas as contract estimations using statistical methods (time series decomposition, Holt-Winters exponential smoothing, ARIMA/SARIMA), include univariate seasonality. In the study, 365-day forecast is performed on a daily basis and 12-month forecast is performed on a monthly basis at once. Among all statistically appropriate forecasting models, ARIMA(1,0,1)1(0,1,1)(365) model found daily basis year ahead natural gas consumptions the best with the lowest error, highest compliance with 24.6% MAPE and 0.802 R-2, for the year 2014. The coefficients of this model were statistically significant and the residuals were found as white noise. The same model has the lowest error (MAPE - 11.32%) and highest compliance (R-2 - 0.981) in the monthly estimations as well. The results show that seasonal ARIMA models are the most appropriate estimation technique among the univariate techniques. The fact that many predictions can be made at a time and the results are acceptable allow these techniques to be used in the year ahead monthly and daily forecasting.