Study On Advanced Prediction Methods Of Natural Gas Demand

The natural gas mid- & long- term demand prediction is discussed in this dissertation. This dissertation starts with the development of natural gas and the indispensability of its demand prediction, and then it methodically introduces the theories of prediction and the categories of traditional prediction methods in Section Two. Moreover, the transition from the traditional ways to modern methods is also articulated in this section. Three different models, the advanced GM(1,1) , the BP Neural Networks (BPNN) optimized by Genetic Algorithms (GA) and the optimized Least Square Support Vector Machines (LS-SVMs) are respectively discussed in Section Three, Section Four and Section Five respectively, all of which could effectively deal with the natural gas mid- & long- term demand forecasting. Next, in order to use the data gained form diverse methods comprehensively, to prevent defects of a single method that loses some useful information, to reduce the randomness and to improve the prediction accuracy, the prediction with optimization combination is applied in Section Six and the final prediction value of natural gas demand from 2005-2010 is obtained. In the last section, the conclusions are acquired and some disadvantages are indicated. The future research aspects are also pointed out at the end of this dissertation.