Building the forecasting model for time series based on the improvement of fuzzy relationships

This study builds a new forecasting model for time series based on some important improvements.  First, we choose  the universal set to be the  percentage variation of the series. This universal set is divided to clusters by the automatic algorithm.  The suitable  number of cluster  depends on the similar level of elements in the universal set.   Second, a principle to find the relationship of each element in the series to the found clusters is established. Finally, we propose the forecasting rule from the established fuzzy relationships. The proposed model is illustrated in detail by the numerical examples, and can be quickly applied to real data by the established Matlab procedure.  Comparing many series with the differences about the number of elements, fields, and characteristics, the proposed model has shown the outstanding advantages.  Using the proposed model, we forecast the salty peak  for a coastal province  in Vietnam to illustrate for application of this study.

Keywords:

Cluster analysis , forecasting model , Fuzzy relation , Time series