Comparative Analysis of Univariate Forecasting Techniques for Industrial Natural Gas Consumption

This paper seeks to evaluate the appropriateness of various univariate forecasting techniques for providing accurate and statistically significant forecasts for manufacturing industries using natural gas. The term "univariate time series" refers to a time series that consists of single observation recorded sequentially over an equal time interval. A forecasting technique to predict natural gas requirement is an important aspect of an organization that uses natural gas in form of input fuel as it will help to predict future consumption of organization.We report the results from the seven most competitive techniques. Special consideration is given to the ability of these techniques to provide forecasts which outperforms the Naive method. Naïve method, Drift method, Simple Exponential Smoothing (SES), Holt method, ETS(Error, trend, seasonal) method, ARIMA, and Neural Network (NN) have been studied and compared.Forecasting accuracy measures used for performance checking are MSE, RMSE, and MAPE. Comparison of forecasting performance shows that ARIMA model gives a better performance.