Research On The Forecasting Of Fuzzy Time Series Models In The Seasonal And Non-seasonal Time Series

Compared with the traditional time series model,fuzzy time series(FTS)models do not require the time series to satisfy any statistical assumptions.Even less observed data or data set contains uncertainty,the model can also effectively forecast.Therefore it has been widely studied.The key in fuzzy time series analysis is to partition the fuzzy interval and deal with the fuzzy relations.Focusing on the problem,Vovan and Lethithu^[53]recently proposed adaptive clustering analysis algorithm to automatically determine the optimal number of a fuzzy set on argument and the specific elements.This method is very good to improve forecasting accuracy.But for how to select parameters in adaptive clustering algorithm did not give details.At the same time,the method is only used for non-seasonal data modeling and forecast,but there is no answer to whether the data with seasonal fluctuation characteristics are equally valid.To solve the above problems,this thesis studies the seasonal and non-seasonal time series data.For the non-seasonal time series,based on Vovan and Lethithu's^[53]algorithm,this thesis proposed the optimization of forecasting error method to choose the parameter of the adaptive algorithm.Through numerical simulation and two real data analysis(Alabama university enrollment and the Shanghai stock composite index),we found that,compared with the traditional time series model,the method has a good forecasting effect.As for the seasonal time series,the thesis presents an eliminated fluctuate seasonally of fuzzy time series analysis method.Through numerical simulation and empirical data analysis(monthly wind power),we also found that compared with the traditional method of the non-fuzzy time series,the Vovan and Lethithu^[53]and the proposed method are good to improve the forecasting accuracy.At the same time,our method of forecasting accuracy is higher than the Vovan and Lethithu's^[53].This shows for the seasonal time series data,it is necessary to eliminate seasonal effect before fuzzy time series analysis.