Research On Improved Algorithm Of Fuzzy Time Series Model

Fuzzy time series model, which was proposed to solve the fuzzy problems that can’t be dealt with classical time series analysis methods, is an extensive research subject in the research field of data prediction and analysis, and its main feature is that it can complete the prediction of data with the shortcomings of incompleteness, inaccuracies and vagueness, etc. With the further improvement of the information degree, fuzzy time series model has been widely applied to different areas and the research on how to improve the prediction accuracy of model attracts more and more attention. Based on the classical model, this paper analyses the deficiencies of the existing methods and proposes improved methods from the two aspects which include the division of fuzzy intervals and the establishment of fuzzy rules. At last, the improved methods are extended to model’s form of high-order.Considering the distribution of the historical data and the complexity of the model, this paper proposes an improved automatic clustering method for interval’s division which can solve the problems of poor interpretability with the equal division method and the increase of model’s complexity with too many division intervals by classical automatic clustering method. Then this paper adopts Particle Swarm algorithm to optimize the model and get the optimal fuzzy interval division.Based on the frequency number and priorities of each fuzzy logical relationship, this paper proposes a hybrid method to establish fuzzy rules with weights. This method can realize to set different weights base on fuzzy rules with different importance, and be able to get a more reasonable rules. Considering the subordinated vector of data can impact prediction results in the process of establishing fuzzy rules, this paper presents to increase of the main fuzzy rules and the secondary fuzzy rules which effectively improves the prediction accuracy of the model.Based on the above improved methods for the division of fuzzy intervals and the establishment of fuzzy rules, this paper proposes to establish an improved fuzzy time series model and its form of high-order. Through conducting experiments on3groups of data, the validity of the improved methods is confirmed respectively, and experiments verify the reasonableness of the proposed model and extended model and theirs advantage in prediction.