Time Series Prediction Based On Hybrid ARIMA And FTS Model

Time series are ordinal data recorded in chronological order.As a special data type with time attribute,time series has long been attached importance to in the development process of social history.Researchers usually make a series of observations and studies on time series data and find out some of the rules that are hidden behind their data to predict their future trends.Time series prediction is a way to predict the future through known data,which makes the decision-maker have the ability to stand up and make a more reasonable choice.Therefore,the study of time series prediction is of great significance.As time series prediction is gradually applied to all aspects of social life,the complexity of the data is usually greatly beyond the imagination of the researchers in the face of practical problems.At the same time,the characteristics of the data are difficult to be completely determined.The use of a single model cannot get more satisfactory results,and the use of mixed models is used.The aforementioned shortcomings can be avoided at a certain level.And because the hybrid model can integrate the strengths of different models,it is also better to improve the modeling effect,so that the hybrid model can be applied to a wider range of fields.In this way,based on the advantages of ARIMA model and FTS model,this paper proposes a hybrid modeling method for these two models,and then improves the prediction results of time series.In this paper,inspired by the idea of Zhang(2003)hybrid model construction,the ARIMA model and the FTS model are mixed to get the ARIMA-FTS model,and the model estimation and prediction methods are proposed.In order to verify the validity of the proposed model,a single ARIMA model,a single traditional FTS model and a ARIMA-FTS model are used to compare the simulated data.The simulation results show that the prediction accuracy of the hybrid model is higher than that of the single model.Then,on the basis of the mixed model proposed in this paper,the improved idea of the traditional isometric domain partition method in the FTS model is modified to apply the non-equal length method based on the K-medoids algorithm.At the same time,the median(rather than the mean of the traditional method)is put forward to replace the interval center value table.The information of a fuzzy interval.In addition,this paper also proposes to improve the processing method of fuzzy logic relations in the FTS model,which is based on the SVM algorithm of machine learning to predict the two-order fuzzy relation,and the improved method is more effective in improving the prediction accuracy of the fuzzy time series.Finally,the improved ARIMA-FTS model is simulated to demonstrate the significance of its improvement.By comparing the numerical results of different models,it is not difficult to find that the prediction ability of the improved ARIMA-FTS model is much better than that of the improved model.Finally,this paper sets up the ARIMA model,FTS model,unimproved and improved ARIMA-FTS mixed model of Shanghai and Shenzhen 300 index data of each year of 2015-2017 year,and obtains the model evaluation by analyzing the empirical results of different models,and uses the corresponding model to predict the end of the year data using the corresponding model.In theory,this paper puts forward a new exploration of the ARIMA hybrid model,attempts to introduce FTS model to model and optimize the nonlinear information that the ARIMA model fails to capture,proposes the model estimation and prediction method of the ARIMA-FTS mixed model,and proposes that the K-medoids algorithm and the SVM algorithm be added to the FTS modeling process,so as to further the process of FTS modeling.The ARIMA-FTS model is improved.In the practical application,the mixed model is used to predict the Shanghai and Shenzhen 300 index data,and the application range of the mixed model is widened.It is proved that the mixed model can be better adapted to the time series data under different trends.