چكيده به لاتين
Nowadays, not only development but entire human’s life needs energy. One of the most important energies which is necessary in our lives is natural gas. During recent years, natural gas has become one of the most important energy resources in the world and from a local fuel which was used in limited markets has turned to an international commercial fuel that is transported to far distances for consumption in diverse economic sectors. Natural gas has specific features which distinguish it from other resources. Predictions show that there is a substantial increase in the natural gas demand in the country. Examining natural gas demand in residential sector seems significant in many aspects. Residential and commercial natural gas are major sectors of energy consumption in the country. Another primary advantage of residential sector is that natural gas demand experiences many fluctuations owing to seasonal changes of temperature. On the one hand, recognizing sufficient amount of storing and on the other hand, higher risks as a result of increasing the amount of storing, are more reasons for examining natural gas demand in residential sector. In this research, 5 variables have been chosen in order to determine the power of diverse models in forcasting monthly natural gas demand in residential sector of Tehran province. The variables are natural gas price, number of consumers, temperature, humidity and amount of wind speed. The data for presented variables was collected monthly from Farvardin 1382 to Esfand 1399. Two neural network models which are MLP and SVM and also two econometric models which are MGARCH and VAR were used to forecast the amount of monthly consumption of natural gas in residential sector in Tehran province for 24 months from Farvardin 1398 to Esfand 1399. Besides, RMSE was used to compare results. Research findings confirm that between two presented neural network models, MLP was more accurate compared to SVM and also between two presented econometric models, MGARCH was more accurate compared to VAR. In order to find the most accurate model in this research, the results of MLP and MGARCH were compared and MLP was more powerful and more accurate.