The Forecasting Natural Gas demand in a region with seven models and evaluating their accuracy using as criteria five types of errors/residuals

The modeling of Natural Gas (NG) demand differs significantly from the demand for electricity in terms of the determinants that affect it, as all fields of economic activities in a modern economy are directly related to electricity but not to NG. But NG is the second energy type after electricity used in all countries in percentages greater than 10% in average terms. NG is going to be installed in the Region of East Macedonia-Thrace (REMTH) the next years. So, we consider it is worth to predict the NG demand in REMTH using eight deterministic forecasting models. In order to do it we used a dataset of 20 years concerning two Greek regions to which the NG is used that period and through them we built the eight forecasting models aiming to find the NG demand in the REMTH. In order to evaluate the reliability and accuracy of them we used four types of statistical errors, Mean Error (ME), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE) and Mean Absolute Scale Error (MASE). These are the most widely used measures of evaluating the accuracy of deterministic predictive models, due to their advantages of scale-independency and interpretability. When  each of them is used alone has the significant disadvantage to produce infinite or undefined values for zero or close-to-zero actual values. In order to address this disadvantage, we propose a way to use the same time all of them measuring the accuracy of a model used to forecast the demand of Natural Gas in the Greek region EMTH. The innovation of this paper is that for NG demand forecasting were used seven different models and they are evaluated regarding their reliability /accuracy using five types of residuals or statistical errors