Forecasting natural gas demand: A primary concern for natural gas pipeline companies.

摘要

A backpropagation neural network and a Box-Jenkins model are developed to forecast natural gas demand for a local gas company, also called a local distribution company, LDC.; Natural gas rates, utilized by 84 local distribution companies for the year December 1, 1993 to November 30, 1994, are available for study. In addition to the natural gas rates, temperature and other weather data are also at hand. Preliminary plots of the natural gas rates and temperature data for all 84 local gas companies indicate that almost half of the LDC's natural gas rates are directly related to temperature; i.e., as temperature gets colder, gas rates increase. In other words, the majority of the 84 local gas companies supply natural gas for home, office, and business heating. Although some of the LDC's natural gas rates indicate a marginal relationship to temperature, other unidentified factors are also obvious. A small number of LDC's natural gas rates show no relationship to temperature whatsoever.; The neural network and Box-Jenkins model mentioned above are designed for one of the local distribution companies whose natural gas rates show a strong and direct relationship to temperature. Although both techniques prove to be quite effective at forecasting natural gas demand for the LDC under investigation, the neural network has a lower mean absolute error in forecasting accuracy than the Box-Jenkins model.; The major factor affecting demand for natural gas from local distribution companies considered in this study is temperature. Other important variables, not considered, are those that deal with the economics of supply and demand for natural gas. In particular, price and regulation and their potential effect on sales of natural gas. These economic issues may well need to be evaluated and included in neural network forecasting techniques designed to predict natural gas demand on a local level.