Research On Crude Oil Price Prediction Model Based On EEMD And Local Regression

Crude oil is an important basic energy and strategic material.It has a very important influence on the stability of the country and the development of social economy.China is the world's second largest consumer of crude oil and the largest importer,its economic development has serious reliance on crude oil.The correct prediction of crude oil prices is of great significance to our government,enterprises,and individual investors.Because the crude oil market is a complex nonlinear dynamic system,the crude oil price is essentially a high-noise,non-linear and non-stationary chaotic time series.Therefore,this paper proposes several new"decomposition-ensemble"forecasting framework for crude oil price forecasting based on the efficient and accurate“divide and conquer”principle in time series prediction.Empirical analysis shows that compared to the benchmark model,the new forecasting framework proposed in this paper improves the prediction accuracy of crude oil pricesThis article takes the most important crude oil prices:WTI and Brent crude oil prices in the crude oil market as the research object.First,the EEMD model is used as a decomposition model,and a local regression model is introduced as a forecasting model,seven new“decomposition-ensemble”forecasting frameworks are proposed,including EEMD-LLP,EEMD-LPP,EEMD-LRR,EEMD-PCR,EEMD-PLSR,EEMD-LLASSO and EEMD-LEN to made a one-step-ahead forecast for crude oil prices.The experimental results show that,the new decomposition framework proposed in this paper is better than the benchmark models EEMD-SVR and EEMD-ANN,and EEMD-LLP model is superior to other models in the comprehensive performance of RMSE,D_stat and operating efficiency,illustrating that the EEMD-LLP model has strong predictive ability and prioritized operating efficiency,the LLP model is more suitable for the prediction of the components with simple modes and stationary fluctuations derived from EEMD decomposition than other local regression models and benchmark models.In addition,the paper analyzes the parameters of the EEMD-LLP model,and constructs the EEMD-LLP model with the same parameter structure to further enhance the model's operating speed.The experimental results show that the prediction result of"same parameter"EEMD-LLP is similar to that of EEMD-LLP model using parameter search method,and the operating efficiency is greatly improved,which proves the efficiency of the"same parameter"EEMD-LLP model.Because multi-step prediction has more reference significance than single-step prediction and the difficulty of modeling is higher,this paper also studies the multi-step-ahead forecasting of crude oil price.Based on the“same parameter”EEMD-LLP model,this paper further introduces the“multi-model ensemble”step,constructs the“same parameter,multi-model ensemble”prediction framework EEMD-ELLP,and studies the best forecasting strategy of EEMD-ELLP in multi-step-ahead forecasting.EEMD-ELLP reduces the uncertainty in multi-step predictions through multi-model ensemble and overcomes the limitations of“same parameter”EEMD-LLP.The experimental results show that in the multi-step-ahead prediction,the best prediction strategy of EEMD-ELLP is direct strategy.The EEMD-ELLP model using direct strategy has achieved the best horizontal prediction results in 3,6,9 and 12-step forecasts,and the directional prediction accuracy is also ranked in the top 2,indicating that the EEMD-ELLP model has high prediction accuracy and robustness,can further improve the prediction accuracy of the crude oil price series.