This study deals with the short-term forecasting on the civilian demand of the natural gas in Korea, using Time Series Analysis and Artificial Neural Networks(ANN) that is advocated as an alternative to traditional statistical forecasting methods.
This paper divided the forecasting period into timely, daily, and monthly gas demand. The forecasting models were built using Time Series Analysis and ANN according to each time horizon. After compared the results, this study attempted to combine these two methods in order to attain much more accurate forecasts.
For forecasting the daily gas demand, we used mixed regression/ARIMA model and ANN models. On the other hand, multiplicative decomposition model, Box-Jenkins* univariate ARIMA model, Transfer Function model and ANN models were employed for the monthly gas demand.
The empirical results of this study show as follows;
Time series forecasts produced by ANN, as suggested by theory, outperformed those from the traditional statistical methods in the daily time series. For the monthly forecasting, ANN didn*t show the same results contrary to the expectation.