Forecasting Gasoline Consumption in Iran using Deep Learning Approaches

Abstract

Gasoline consumption is one of the challenging issues of energy management in Iran. The deficit of domestic production and the need for imports on one hand, and the impact of its consumption on macro-and micro-economic variables, on the other hand, cause gasoline consumption management has become more important. The more accurate, predicting the trend of gasoline consumption is the more successful consumption management will be. Since gasoline consumption is affected by several parameters and factors, so, forecasting its consumption with high accuracy is difficult. In this paper, one recursive competitive learning method and two deep learning methods are utilized to provide more accurate forecasting of gasoline consumption. Due to the impact of gasoline consumption patterns on the seasonal changes, climate and holidays, different periods are used for training the learning these approaches, and their efficiency is compared in terms of the standard error metrics. The comparison results show the deep learning approaches and the training patterns with 12 months result in more accurate predictions. Finally, using the best approach and obtained setting, the gasoline consumption in Iran is predicted for the next years, which shows that gasoline consumption will grow 22 percent by 2027.