A Research And Application On Discourse Of Fuzzy Time Series Models

For the concept of fuzzy,to accurately grasp it is often very difficult.Similarly,for the inaccurate and fuzzy data,we want to find out the rule to predict the future trend is also very difficult.In the experiments,which are on the background of fuzzy data of prediction.Fuzzy time series model with its precision,simplicity,applicability,referred to as the leader in the fuzzy prediction.It is agreed by experts at home and abroad.But,ambigulity means it must have errors of prediction.How to improve the prediction precision,reduce the deviation with the real value,becomes widely in the field of fuzzy time series research focus.In this paper,the definition and partition of the universe and the uncertainty of clustering number of FCM clustering are put forward.And the classical experiment data and the data collected in the society are used to carry on the experiment,proves the correctness and the wide applicability of the optimization scheme.The theoretical basis of fuzzy time series prediction is discourse domain interval.Therefore,based on the definition of discourse domain interval and the division of discourse domain,Nile calibration method is proposed to find outlier data and the method algorithms for dividing ratio are proposed.In the domain definition,different methods of defining the universe of discourse are proposed.Respectively,from these aspects of the model to be improved.The experimental results of the fuzzy time series model are compared with the traditional forecasting results,and the results show that the modified model has a significant improvement on the result of the fuzzy logarithm model of the classical Alamaba state register.Before the domain definition and division,the sample data need to be clustered.The number of clustering becomes the division of the whole universe,which directly affects the accuracy of prediction.The traditional FCM clustering algorithm can not determine the number of clusters when it is clustering.Therefore,it must be subjectively given according to the specific expermential model requirements.The Number of Clustering Parameters C.a too large a number of clusters will make the calculation more complicated,too small number of clusters will not have the same characteristics of the sample data clustering together,thus affecting the universe division accuray.In general,wrong clustering leads to errors in the definition of large universe,which directly affect the determination of upper and lower bounds of domain and local partition of domain.Based on this.In the paper,the NVR algorithm is presented,and the formula for how to find the number of cluseters is given,which eliminates the subjectivity of the traditional FCM algorithm.Through the experiment of the traffic flow in the Internet cafes of Guanggu predestrian street,it is proved that the improvement can greatly reduce the prediction error of the model.