Fuzzy Time Series Forecasting With An Improved Adaptive Analysis Window

The time series forecasting investigates the relations between the sequential set of historical data measured over time to forecast the future values, through a variety of mathematical analysis. The purpose of time series forecasting model is to find out the development trends and rules of time series and use of these trends and rules to predict the data in the future.The traditional time series forecasting is based on the classical set theory, and it can predict through the complete historical data. But in reality, there is a large number of historical data which is incomplete or uncertain. In order to solve these problems, Song and Chissom put forward the concept of fuzzy time series prediction, which mainly is based on the traditional time series prediction by introducing fuzzy theory. It establishes the corresponding fuzzy logical relationship to forecast. After years of research and demonstration, the fuzzy time series forecasting has become one useful method of solving the contradiction between rich data and poor knowledge.In this paper, the adaptive fuzzy time series forecasting algorithm is developed on the foundation of domain, fuzzy data and fuzzy relationship. About domain partion, the ratio method is applied to selet internal length. In the aspect of the fuzzy data, the frequency of the fuzzification historical data is used to make the second division. In the aspect of the fuzzy relationship, overdetermined equation is easablished and the approximating solution of equation is given. Finally, based on the above steps, the(?)nal prediction model is established.This model is used to forecaste the five years from 1995 to 1999 shares of stock index. By contrast with the root mean square error rate (RMSE), the higher accuracy can be obtained through this algorithm. The model, proposed in this paper, can be used to deal with uncertainty of the sequence. And it can get the higher accuracy and has a wide range of applications.