Application Of Nonlinear Forecasting Methods On Analyses Of Natural Gas Demand

At present, the demand forecasting of natural gas mostly adopts branch energy intensity model method. This method uses energy intensity and the real energy value to produce sum. So it is rough this method to forecast, which educes approximate forecasting value and the error is bigger comparatively. This method can't be satisfied when calling for good precision. This paper uses mathematics algorithm, to forecast the demand of natural gas and acquired some achievements. This paper makes an all-sided research for several commonly-use forecasting methods, such as time serial method, multi-variant regression method, gray system method, artificial neutral network method. The author analyzes merits and demerits of these methods and exerts these methods to forecast the demand of natural gas in Sichuan. Moreover the author put forward some effective better methods aiming at prediction for demand of natural gas. The BP network of ANN is the best among these methods comparing from the result. BP network has good precision and lower error, which shows BP network has powerful advantage in dealing with nonlinear problem absolutely.But it is difficult to carry though accurate and credible forecasting by only using one method when we settle the forecasting problem of demand of natural gas which has many influential factors. Then the author makes use of combination forecasting method and integrates these common-use methods. This paper makes in-depth research for combination forecasting.This paper brings forward applying combination forecasting, combination ANN, combination gray ANN on forecasting the demand of natural gas and gain good effect. Combinations forecasting, to be brief, adopt different forecasting methods and assemble them appropriately. Then we may utilize fully useful information provided by different methods in order to improve forecasting precision. These combination methods simulate the complex relations among serial data by using three-layer ANN can approach any rational function and training the network. Its principle is that use the forecasting value as ANN input stylebook. Combination ANN uses the results of combination forecasting as the input of BP network. Combination gray-ANN uses the results of gray system as the input of BP network and the true value as the output of the network. Then we can takes definite framework, train the ANN and use the BP network to forecast.The author simulates these methods by computer technology and acquires good forecasting result.