Research On Integrated Model Of Crude Oil Price Yield Forecast Based On LASSO

Starting from the research background and research significance,this article analyzes the trend of crude oil prices in the past two years,and studies the crude oil price prediction model through three aspects: the formation of crude oil price mechanism,the exploration of crude oil price fluctuation factors,and the study of international crude oil price forecasts.The sequence model can better describe the linear characteristics in the crude oil price series,but only the time series model cannot accurately fit the nonlinear characteristics in the sequence.The artificial intelligence model can better characterize the nonlinear characteristics in the crude oil price sequence,but it cannot describe the linear trend in the sequence.Therefore,it is difficult for a single model to describe the multiple characteristics of crude oil price series,so a hybrid model integrating different models is an effective choice.Traditional model integration mainly includes Bagging,Random Forest,Boosting,AIC and BIC.Among them,Bagging,Random Forest and Boosting integrate different samples of the same type of model through voting.AIC criterion and BIC criterion are based on different variables.Of the same type of model.The above traditional ensemble model studies the integration effect of the same type of nested models.At present,time series models and machine learning models are commonly used in research,which also include linear and nonlinear models,and the model and model are non-nested Relationship,this article attempts to study the use of LASSO to achieve different types of model integration.Use 11 models such as moving average,one-time exponential smoothing,two-time exponential smoothing,three-time exponential smoothing,ARIMA,SARIMA,regression analysis,SVR,LSTM,RNN,and Xgboost to generate 11 prediction sequences of crude oil price returns,and select candidate models Set variableization,by solving the coefficients of each variable,and then assigning the corresponding weight to each model,so as to achieve the integration of the model.There may be two types of problems directly using the LASSO integration model.The first type of problem is that different types of models are used to obtain the crude oil price return prediction sequence that may be highly correlated;the second type of problem is the time series model prediction sequence group and the machine learning model prediction sequence group May present a group structure.In response to the above problems,this paper proposes to use the LASSO integration model and the Group LASSO integration model under the factor structure,and further consider the Group LASSO integration model under the factor structure.The 11 candidate models,the LASSO weight integration model and the simple weight model are evaluated through indicators such as MSE,RMSE,MAE and MAPE.The results show that it is more reasonable to use LASSO to set the weight of the candidate model set,and the LASSO integration under the factor structure The prediction effects of the model and the Group LASSO integrated model are better than those of the LASSO integrated model,and the Group LASSO integrated model under the factor structure has the best prediction effect.