A Fuzzy Similarity-based Forecasting Model for Fuzzy Time Series

• Main Authors: Guan-JyunLong, 龍冠君
• Other Authors: Sheng-Tun Li
• Format: Others
• Language: en_US
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/45616496209057685552

碩士 === 國立成功大學 === 資訊管理研究所 === 98 === In our daily life, vague and incomplete data described as linguistic variables massively exists in various areas. Therefore, fuzzy time series forecasting plays an important role for uncertain situations. Various forecasting models have been proposed with an emphasis on improving forecasting accuracy or reducing computation cost. Generally speaking, the framework of a fuzzy time series forecasting model is constructed of four major steps: (1) determining and partitioning the universe of discourse, (2) defining the fuzzy sets on the universe of discourse and fuzzifying the time series, (3) constructing fuzzy logical relationships existing in the fuzzified time series, and (4) forecasting and defuzzifying of its outputs. However, most of the researches derive the fuzzy logical relationships in step (3) only using the exact-match IF-THEN rules. Their forecasting models ignore the fuzzy character in forecasting step, and have two shortcomings in the following: (1) If the data is not enough to generate sufficient fuzzy logical rules, the low rate of rule matching may occur in forecasting process. (2) If the order of a fuzzy logical relationship is quite high, the previous models will become hardly to find the matching rules in the forecasting process. Therefore, in this study, we propose the fuzzy similarity-based forecasting model for fuzzy time series to solve the two shortcomings and raise the forecasting accuracy.