Research And Application Of Fuzzy Time Series

For incomplete data such as vague and inaccurate,fuzzy time series shows the unique advantages with respect to the classical time series. Along with a large number of scholars’ research, fuzzy time series has been applied to travel, stocks, temperature and other fields to predict, and has a higher prediction accuracy.About the research of fuzzy time series, scholars mainly focus on three aspects is that division the universe,establishment fuzzy rules and defuzzification.With the vigorous development of machine learning algorithms in recent years, some scholars had put machine learning algorithms into fuzzy time series and get a good model.Firstly,improving the existing rules in terms of the establishment of fuzzy rules.The present good fuzzy rules are taking the number and the order of fuzzy logic relation into account, righting to assign the internal weights and the weight between the two aspects need extra attention. In this paper, the improvement rules is mainly aimed at the weight distribution of the fuzzy logic relation about its order, and how to reasonably deploy the weights between the frequency and order can achieve better prediction effect.Secondly, the article applying wavelet transform to the fuzzy time series.Using the idea of discrete wavelet decomposition and reconstruction,decomposing the original data to low and high frequency sequence using multi-scale decomposition method with selected wavelet function, and establishing fuzzy time series modeling and forecasting according to their characteristics respectively. Then, using the fuzzy C-means algorithm to divide the universe and the improved rules to establish the fuzzy rules and defuzzification for improving the validity of the model.Finally, selecting the data of the national revenue and the enrollments of Alabama University to do an experiment, the validity of the new model is verified by comparing with the previous fuzzy time series model.The experimental results shows that the improved hybrid model has more advantages compared with the traditional model. The new model can also be further improved, by combining with other complex machine learning algorithms, we believe there will be more accurate prediction results.