Research On LPG Price Prediction Model And Application In China

Liquefied petroleum gas(LPG)is a widely used and highly market-oriented clean energy.As an important fuel and chemical raw material,LPG is of great significance to the daily life of Chinese residents and national industrial development.LPG futures have been listed in March 2020,and the Chinese market will play an increasingly important role.However,LPG prices are affected by many factors and fluctuate greatly,showing complex nonlinear and non-stationary characteristics,which makes the prediction of LPG prices very difficult,and also brings great risks and challenges to the formulation of relevant national policies,the production and operation management of enterprises and the decision-making of financial investors.At present,there are very few studies on LPG price prediction at home and abroad,Carrying out this research has important theoretical and practical significance.At present,there are two common price prediction ideas,which only consider the single characteristic time series prediction of the historical data of the research object and the multi characteristic time series prediction considering the influencing factor data.This thesis adopts these two ideas respectively,and takes the LPG national price index as the research object.Under the first idea,the prediction accuracy of long-term and short-term memory network(LSTM)and convolution long-term and short-term memory network(ConvLSTM)are compared,and then Empirical Mode Decomposition(EMD),Ensemble Empirical Mode Decomposition(EEMD)and Complete Ensemble Empirical Mode Decomposition With Adaptive Noise(CEEMDAN)are combined on the basis of ConvLSTM with better performance.The experimental results show that the prediction error of CEEMDAN-ConvLSTM combined model is the smallest.Under the second idea,analyze the domestic LPG Industry Chain and pricing mechanism,explore the main influencing factors of domestic LPG price,preliminarily screen the characteristics through correlation analysis,use the time lag cross-correlation(TLCC)method to explore the lag relationship between the influencing factors and the research object,and finally screen out five influencing factors guiding the change of LPG price.The TLCC method is combined with LSTM,ConvLSTM and CEEMDAN-ConvLSTM models respectively,which overcomes the deficiency of using the future value of influencing factors as characteristic input or assuming or estimating the future value of influencing factors in the prediction of multiple characteristic time series,so that the model has the ability of backward prediction.The experimental results show that the prediction error of TLCCConvLSTM combined model is the smallest.In this thesis,an innovative combination model is proposed under the two prediction ideas,which has achieved good prediction results and provided new ideas and methods for the field of LPG price prediction.