Research And Application Of Two Improved Series Combination Prediction Models Based On Intervention Analysis

Prediction is the basis of decision-making.Most departments in various fields need to make decisions when they are working.Therefore,accurate prediction is an important guarantee for departments to make scientific decisions,prevent adverse situations and prepare feasibility plans or development strategies.In reality,data in various fields often have very complex characteristics,and it is difficult for the traditional single prediction model to comprehensively capture all characteristics of the data.In particular,the occurrence of major emergencies such as COVID-19 has affected data in various fields and brought great challenges to the prediction work.How to effectively predict the data with complex characteristics and improve the accuracy of prediction under the background of emergent events is a research problem with important theoretical and practical significance.In order to improve the accuracy of prediction,this thesis aims at the complex characteristics of data in various fields.Using the advantages of time series ARIMA model to deal with linear problems,Intervention analysis model to deal with emergencies and outliers,GARCH model to deal with volatile data,Support vector regression(SVR)model and BP neural network model to deal with nonlinear problems,the above models are rationally combined.According to the characteristics of data in different fields,two series combination prediction models based on improved Intervention analysis are proposed,namely,the series combination prediction model based on SARIMA-Intervention-SVR/BP neural network and combination prediction model the ARMA-AO-GARCH combination prediction model.The main research contents and results of this thesis are as follows:(1)The SARIMA-Intervention-SVR/BP neural network series model is proposed in view of the linear and nonlinear characteristics of data in many fields in reality and the characteristics of being susceptible to unexpected events and outliers.Through the empirical analysis of the air passenger mileage of the United States affected by 911,Beijing inbound tourism affected by SARS and China civil aeronautic cargo capacity affected by COVID-19,it is indicated that this novel prediction model is more effective than the single model,the SARIMA-Intervention series model as well as some derivative series or parallel combination models when the data are affected by certain intervention.(2)For data in the often more complex financial field,it is more susceptible to sudden events and outliers.In addition,financial data are mostly conditional heteroscedasticity.In view of its data characteristics,this thesis proposes a combination prediction model of ARMA-AO-GARCH-BP neural network,and conducts empirical research and prediction on the CSI 300 index series data set.Results shows that the ARMA-AO-GARCH-BP neural network combined model proposed in this thesis has the highest prediction accuracy under this data set,which proves the effectiveness of the series model compared with the single ARMA model and other combined models under the influence of outliers.