Forecasting natural gas flow using neural and fuzzy-neural networks and time series methods

In transmitting natural gas from refineries to households, there are many equipment to reduce gas flow pressure and prepare it for consumption in the cities entrance. City Gate stations (CGS) are one of the last ones, breaking the pressure to town border stations. The amount of household gas consumption in a region is measurable by the amount of gas transmitted from CGS's. Precise forecasting natural gas flow is a crucial task for scheduling and planning the gas distribution in province gas companies, especially in cold seasons that consumption rises. In order to make an accurate prediction we need to identify relevant data series which independently affect gas consumption. Previous works have selected variables such as previous days consumption amount, previous days minimum, average, effective and maximum temperature, minimum and maximum amount of humidity, previous days sunshine hours, previous days evaporation amount, wind speed, number of gas clients, and weekday. In this article first, we have performed input selection on parameters mentioned above, and then after data normalization, we have predicted future amounts of gas consumption with a group of statistical methods and intelligent algorithms. Finally, the best method has been selected by comparing the results of experiments on data which were collected from a CGS in Qazvin as a case study

کلیدواژه ها:

input selection ، forecasting natural gas flow