Research On Natural Gas Price Prediction Based On CEEMD Model

Under the background of the vigorous development of the global low-carbon economy,natural gas,as one of the most important clean energy sources,plays an increasingly important role in the global energy system.Fluctuations in natural gas prices affect the development of the energy industry,thereby affecting economic growth and social development.Accurate forecast results of natural gas price can not only help market participants make reasonable decisions and benefit from them,but also provide scientific references for the government to formulate policies such as dealing with the risks of price fluctuation and ensuring the supply of natural gas.Therefore,it is very important to predict the price of natural gas.The world ’s major natural gas markets can be divided into North America,Europe and Asia-Pacific markets.Among them,the Henry hub in North America is the most active and largest natural gas hub in the world,with global influence.Its representative price is more and more widely quoted by spot trading centers and media around the world.Therefore,this paper took the time sequence data of the spot price of natural gas at Henry Hub as the research object.Firstly,the X12 seasonal adjustment method,which is only suitable for decomposing monthly or quarterly data,was used to decompose the monthly natural gas spot price sequence to obtain the trend cycle factor sequence,irregular factor sequence and seasonal factor sequence.The HP filtering method was used to further extract the cycle factor sequence and trend factor sequence from the trend cycle factor sequence,and the characteristics of each factor sequence were analyzed in combination with the development stage of natural gas to clarify the influencing factors and specific characteristics of natural gas spot price fluctuations.Secondly,on the basis of clarifying the nonlinear,multi-scale and other complex characteristics of the daily spot price of natural gas,a decomposition integration combination forecasting model based on CEEMD was constructed.The price sequence was decomposed by CEEMD to obtain the trend component and the IMF components,and the IMF components were reconstructed into high frequency component and low frequency component.The ELM model was used to predict the high frequency component,and the GA-SVR model was used to predict the remaining components.Finally,the prediction results of the two models were integrated to obtain the prediction results of the combined model.The model test was carried out,and the prediction evaluation indexes were compared and analyzed in combination with other comparison models.The results showed that the CEEMD-ELM-GA SVR combined model constructed in this paper had higher fitting degree,lower prediction error and better prediction effect than other prediction models.Compared with the variance reciprocal method,the simple summation method was more suitable for integrating the prediction results of different models in the combined model.Among the two different seasonal factor elimination methods used in this paper,the method of eliminating the arithmetic mean of seasonal factors was more scientific,but eliminating seasonal factors reduced the prediction accuracy of the combined model.