As an efficient and clean burning fuel, natural gas provides one quarter of the global energy supply. It has a significant impact on national energy production, consumption, economic growth, and our daily lives. Many studies have been conducted to predict natural gas consumption based on certain parameters. In this paper, artificial neural networks (ANN) are used to predict future hourly natural gas loads with up to 95% accuracy. The ANN model is trained based on multiple parameters, for example, gas load historical pattern, weather pattern, calendar effect, etc. for different seasonal profiles; and it can then be used to create the future gas load forecast. Sensitivity study results from different ANN models will be presented and the parameters analyzed with a view to improve the ANN model and give improved forecasting accuracy.
There are many customers using gas load forecasting results on a daily basis for daily analysis and long range planning on a yearly basis, such as gas planning groups, storage planning, gas control, field managers, gas rental and marketing groups, and even Federal and State regulators. The cost of an inaccurate gas load forecast can be millions of dollars for a large operating corporation. A good forecasting model with higher accuracy is needed to save costs to the gas company for additional charges: e.g. oversupplying gas, parking, line packing, bringing additional supplies on at a higher price (demand charge), and withdrawing from storage to make up shortfalls. A user case study is presented on how this ANN approach can be used to improve the efficiency of gas transmission operation and provide economic benefits from improved gas load forecasting results which are derived from the proper selection of training data set and training parameters.