The Linguistic Forecasting of Time Series based on Fuzzy Cognitive Maps

Most researches of time series forecasting mainly focus on the aspect of pursuing the numerical forecasting precision by constructing the quantitative model. But in the real world, precision is sometimes not necessary for perceiving and reasoning of human, and the qualitative forecasting of time series is able to satisfy requirement of some decision problems. In this paper, a new qualitative forecasting method is proposed, which combines the fuzzy c-means clustering algorithm, fuzzy cognitive map (FCM) and the real-coded genetic algorithm (RCGA). The fuzzy c-means clustering algorithm is used to extract linguistic label, transform the original time series into the fuzzy time series and construct the framework of FCM, automatically. The RCGA algorithm is adopted to learn weights of constructed FCM for modeling the formed fuzzy time series. Finally, a fully learned fuzzy cognitive map is exploited to carry out linguistic forecast by iterations. The proposed forecasting method is applied to forecast the enrollments of university of Alberta on the linguistic level, whose results show the feasibility and effectiveness of proposed method.