USING MACHINE LEARNING TOOLS FOR FORECASTING NATURAL GAS MARKET DEMAND

Providing reliable and accurate forecasts of natural gas consumption can keep supply and demand of natural gas pipelines in balance, which can increase profits and reduce supply risks. In order to accurately predict the short-term load demand of different gas nodes in the natural gas pipeline network, a hybrid optimization strategy of integrated genetic optimization algorithm and support vector machine are proposed. Factors such as holidays, date types and weather were taken into account to build a natural gas daily load prediction model based on GA-SVM was established. A natural gas pipeline network in China includes three gas supply nodes of different user type gas is forecasted, and a variety of error evaluation method, the GA - SVM evaluation index compared with other prediction methods, and through different data set partition is discussed in the periods of peak gas and gas resources in the GA - the applicability of the SVM prediction model, the ends of a natural gas pipeline network in China includes four gas supply nodes of different user type gas is forecasted, and a variety of error evaluation method, the GA - SVM evaluation index compared with other prediction methods, The applicability of the method is also discussed by dividing different data sets. By predicting the gas load forecast of the three nodes, the results show that GA-SVM hybrid prediction model has high prediction accuracy compared with other single models, and the three gas nodes MAPE of GA - SVM is respectively 3.66%, 5.17% and 3.43%. Through further analysis, even with the data samples reduced, the winter gas peak of gas prediction can still maintain good prediction effects. The research shows that the GA-SVM model has high accuracy and strong applicability in predicting gas consumption at different nodes of the natural gas pipeline network. This study can provide a research basis for analysis of gas supply uncertainty and further gas supply reliability evaluation of pipeline network.