Interval-valued Time Series Modeling Method Research Based On Granular Computing

As a kind of time series,interval-valued time series(ITS)collects interval-valued data in time sequence,and it can describe the uncertainty and variability of variables.It has important practical value in many fields for early warning,control and decision-making,such as economy,society,energy and environment.The conventional ITS modeling methods take numerical value as the center and pursue the accuracy of prediction results.But with the increasing scale of data,paying too much attention to the accuracy of individual data no longer meets the actual needs.It can not completely describe the uncertainty caused by the irregularity of data or the ambiguity of human language,which is not convenient for human's cognition and understanding.As an information processing method,granular computing draws lessons from the thinking of human in dealing with complex problems,abstracts problems with the help of theories such as fuzzy set,rough set and probability set,and divides problems into a series of sub-problems which are easier to manage by information granulation.In the field of time series,this method of simulating human's processing a large amount of complex information pursues the interpretabil-ity of the model,which makes it easier for users to understand,capture and describe the dynamic characteristics between time series data,and to analyze complex information more efficiently.In this background,this paper proposes a method of ITS modeling based on granular com-puting combining with artificial neural network.In this method,each interval is regarded as a granule,which is granulated in the framework of granular computing based on fuzzy sets,and information granules with semantics that can reflect the characteristics of data are established.Data are described by these information granules,and the original ITS is transformed into gran-ular and linguistic time series and is trained by multi-layer perceptron.The output of the model is a set of descriptions of interval-valued data about information granules.The linguistic and numerical prediction results can be obtained by analysis and decoding.A series of experimental results show that the ITS model proposed by this paper has good interpretability as well as the certain guarantee on numerical accuracy.It makes the prediction results easier to be understood and recognized,and provides effective support for users to make reasonable decisions.