| Fuzzy times series model is a powerful tool to deal with the historical data which are linguistic values and uncertain circumstances, more and more researchers focus on fuzzy time series model. At present, fuzzy time series model has been applied to forecast the number of tourists, temperature, stock market, network users, incidence of hemorrhagic fever with renal syndrome etc. In order to get higher precision, different fuzzy time series methods are proposed. Most studies focus on the universe of discourse partition and data fuzzification in fuzzy time series models.Fuzzy time series is evolved from the classic time series, in which data correlation is involved inevitably. The correlation would be not changed through data fuzzification, resulting in the antecedent of fuzzy rules influencing on the consequent of fuzzy rules in a different way, Then the paper proposes a new optimization algorithm to optimize fuzzy inference rules by combining the autocorrelation theory of time series.Firstly, stationarity of data is discussed by virtue of time series theory. A mean non-stationary sequence is transformed into a stationary sequence through appropriate difference. In addition, appropriate variance stabilizing transformation is required for variance non-stationary sequence. Generally, for the mean and variance non-stationary process, the logarithmic transformation is applied firstly and then the difference transformation is applied. Based on above step, fuzzy sets are obtained through traditional fuzzy sets on the universe of discourse U, thereby fuzzy rules are constructed; Secondly, fuzzy rules are optimized by autocorrelation theory. At the same time, an improved method of standardized deviation is established for weight calculation; Finally, through combining with the previous time series model and traditional fuzzy time series model about the forecasting of Alabama university enrollments, the comparative results show that the proposed method is effective. |
PDF Full Text Request