Crude Oil Price Forecasting Based On CEEMDAN-GWO-KELM Model

As the world’s most important industrial raw material and strategic reserve,crude oil is an important guarantee for the smooth operation of the national economy of modern countries.The violent fluctuations in crude oil prices will have a serious impact on a country’s political,economic,military and other aspects.With the advancement of China’s reform and opening up process,the oil dependency of China continues to rise,and the impact of fluctuations in crude oil prices on the operation of China’s national economy is increasing.Therefore,it is significant to seek the potentially influential factors of oil price fluctuations and accurately predict the trend of oil prices for the formulation of China’s crude oil strategy and corporate business strategies.Many factors affect the fluctuation of crude oil prices,such as market supply and demand,geopolitics,international emergencies,and financial speculation,so the crude oil prices show highly complex nonlinear and non-stationary characteristics.Therefore,this thesis proposes a Sparse Bayesian Learning(SBL)model to analyze the major influential factors of crude oil prices,and proposes a method integrating Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN),Grey Wolf Optimizer(GWO),Kernel Extreme Learning Machine(KELM)—namely,CEEMDAN-GWO-KELM—to predict crude oil prices.This is the main research content of this thesis,and it is also the two innovations of this thesis.Firstly,this thesis analyzes the influential factors of crude oil prices from the perspectives of supply and demand factors,financial market factors and geopolitical factors of crude oil,and selects 29 variables as input variables of SBL model to analyze the importance of various factors affecting oil prices.Secondly,the original crude oil price time series has extremely complex nonlinear and non-stationary characteristics,in order to reduce the difficulty of subsequent prediction,this thesis uses CEEMDAN method to decompose West Texas Intermediate(WTI)crude oil spot price time series and decompose it into several frequencies components to achieve accurate oil price forecasting by predicting individual components separately and adding individual predictions.In predicting each component,KELM is used as the predictive model,and the GWO algorithm is introduced to optimize the kernel parameters and regularization coefficients in the KELM model.The GWO-KELM model is constructed as the final component prediction model,and finally,the components are predicted.The results were simply summed to obtain the final forecasting result.After empirical analysis of the factors affecting the spot price of WTI crude oil,the results show that many factors affect the fluctuation of oil prices,such as crude oil demand and supply,financial markets.Among them,the WTI crude oil spot price time series’ own lag period time series data has the greatest impact on oil price fluctuations,and considering that in the field of crude oil price forecasting,most scholars use the lag time series data of oil price time series to predict oil prices,this thesis also uses the lag period time series data of the oil price time series in the empirical analysis of the subsequent forecast of oil prices.At the same time,the SBL model is compared with the regression results of the Ordinary Least Square(OLS)model.It is found that the SBL model proposed in this thesis can capture the nonlinear and non-stationary characteristics of crude oil prices.Under the premise of ensuring the accuracy of regression,the factors affecting the international oil price are more accurately analyzed.After empirical analysis of the single model prediction and decomposition integrated model prediction of WTI crude oil spot price,the results show that the KELM prediction performance is significantly better than the comparison model used in this thesis in the single prediction model,and the KELM model can significantly solve the parameter sensitivity problems existing in some machine learning models,which improves the prediction accuracy of oil prices.In the decomposition integrated prediction model,the prediction result of the CEEMDANKELM model is better than the comparison model used in this thesis.It proves that the decomposition integration model can greatly reduce the prediction difficulty and improve the prediction performance of the model.Then by comparing the prediction results of CEEMDAN-GWO-KELM and CEEMDAN-KELM models,it can be found that the short-term prediction accuracy is improved significantly compared with the latter,and the medium-and long-term prediction accuracy is also slightly improved,which proves that the CEEMDAN-GWO-KELM model proposed in this thesis has excellent predictive ability in predicting crude oil price,and provides a new research method for predicting crude oil prices,which has great theoretical and practical significance.