Research On Oil Price Forecast Based On Hybrid Algorithm

It is well known that petroleum is an important strategic chemical raw material and is vital to the development of the country.With China's increasing dependence on oil,fluctuations in oil prices may affect China's economic development and social stability.Therefore,it is of practical significance to dig deep into the international oil market and design scientific methods to reasonably predict the trend of oil prices and deal with fluctuations in oil prices.In this paper,after combing the research literature of related oil price forecasting models,this paper attempts to construct a Lasso-Adaboost-BP combination forecasting model in view of the complexity and non-linearity of oil price fluctuations.A total of 20 variables affecting oil price fluctuations were selected.Based on the characteristics and applicability of the Lasso method,the Lasso method was selected as a screening technique for the factors affecting oil prices,and the significant factors affecting oil prices were explored from many variables.Combined with the Adaboost-BP integration model,a combined prediction based on the Lasso regression model and the Adaboost-BP integration model is proposed to realize the complementary advantages of the two models and improve the prediction accuracy.The two main ways of constructing the Lasso-Adaboost-BP combination in this paper are as follows :The first way is to combine Lasso regression and Adaboost-BP integration model in a tandem way.Specifically,first,the main index that affects oil price fluctuations is selected by the Lasso variable selection method,and then the selected significant factors are used as input variables of the Adaboost-BP integrated model.After learning and training,the network outputs the predicted value of oil prices for the corresponding period.In the second method,Lasso regression and Adaboost-BP integration model are combined in parallel.Specifically,Lasso regression and Adaboost-BP integrated models are used to predict the WTI oil price,and then the optimal combination weights are obtained based on the conditions with the smallest mean square error.Finally,the prediction results of the two models are calculated by weighting the final Predictive value.The monthly data from 1994.01 to 2019.05 were taken as samples for empirical analysis.The two combined models constructed in this paper were compared with the prediction methods of Lasso,BP,Adaboost-BP,and traditional ARIMA,which showed that the combined model of Lasso-Adaboost-BP was predictive Compared with the single model;compared with the linear ARIMA,the average absolute percentage error of the combined model has increased by 86.39% and 75.54%,respectively,and the direction of the price of oil has been slightly improved.Lasso-Adaboost-BP effectively improves the interpretability of forecast results,overcomes the weakness of traditional data-driven methods that lack economic interpretation,and improves the forecast accuracy of international oil prices.