文章基本信息

标题：A COMPUTATIONALLY EFFICIENT METHOD FOR HIGH ORDER FUZZY TIME SERIES FORECASTING
本地全文：下载
作者：SIBARAMA PANIGRAHI ; H.S. BEHERA
期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2018
卷号：96
期号：21
出版社：Journal of Theoretical and Applied
摘要：Despite over more than twenty years of research on fuzzy time series forecasting (TSF) and several studies indicating superior performance, an appropriate computationally efficient method have not been developed to predict various time series using fuzzy TSF method. Motivated by this, in this paper a computationally efficient method is proposed to forecast various time series by using a high order fuzzy TSF model. In this method, the fuzzy TSF parameters such as length of intervals, number of intervals and order of the model are determined deterministically. The order of the model is determined by making analysis on the autocorrelation function (ACF) and partial autocorrelation function (PACF) of the fuzzy time series. The length of interval is determined by using single-variable constrained optimization based method and defuzzification is done by using interval average. In addition, motivated by the boost in forecasting performance due to the use of artificial neural network (ANN) for representing FLR, in this paper, a fast learning one-pass neural network called generalized regression neural network (GRNN) is used for representing the FLR. The use of GRNN model avoids the problems of traditional ANN models such as: ad hoc architecture selection and determining large number of weights and other parameters. In order to evaluate the effectiveness of the proposed model, ten univariate time series datasets are considered and three recent fuzzy time series forecasting models using ANN to represent FLR are implemented. Each model is independently executed for fifty times on each time series and extensive statistical analysis is made on the obtained results. Results revealed the robustness and statistical superiority of the proposed model considering its alternatives existing in the recent literature.
关键词：Time Series Forecasting; Fuzzy Time Series; Fuzzy Logical Relationship; Autocorrelation and partial Autocorrelation function; Generalized Regression Neural Network