A Crude Oil Price Forecasting Model Based on Local Mean Decomposition, Marine Predators Algorithm and Least Squares Support Vector Regression

Crude oil is the blood of modern industry. Therefore, accurate forecasting of crude oil prices is crucial. In order to improve the forecasting accuracy of the crude oil price, this paper proposes a hybrid forecasting method based on local mean decomposition (LMD), contribution coefficient, marine predators algorithm (MPA) and least squares support vector regression (LSSVR). This method consists of four steps: The crude oil data are decomposed by LMD and the decomposed components are reconstructed by contribution coefficients to reduce the computational cost, forecasts each individual reconstructed component by LSSVR, and the parameters of LSSVR are optimized by marine predators algorithm incorporating generalized opposition-based learning strategy, finally, the predicted values of each component are simple addition as the final forecasting result. WTI and Brent crude oil are used to verify the proposed hybrid method, and three evaluation criteria are used to compare the forecast accuracy with other methods. The results show that for WTI crude oil data, the RMSE of the proposed method is 2.4285, MAE is 1.8747, and MAPE is 2.0711. For Brent crude oil data, the RMSE of the proposed method is 2.4117, MAE is 1.8033, and MAPE is 1.8322, which is significantly better than other methods, can accurately predict crude oil prices. In addition, statistical tests were also conducted to further verify the superiority of the proposed method. Adaptive decomposition methods and parameter optimization can effectively improve the performance of crude oil price forecasting models. This method has good application prospects in crude oil price forecasting.