Long‐term intuitionistic fuzzy time series forecasting model based on vector quantisation and curve similarity measure

Abstract

In existing fuzzy time series forecasting models, the accuracy of forecasting excessively relies on priori knowledge and output cannot effectively forecast multi values. The forecasting accuracy reduces drastically when time series data deviate from experience boundary in most models. The generalisation performance is insufficient. To overcome defects of traditional methods, this study proposed a long‐term intuitionistic fuzzy time series (IFTS) forecasting model based on vector quantisation and curve similarity measure. In preprocessing of the proposed model, FTS theory is extended to long‐term IFTS forecasting scope, the raw historical data are quantised vectors and optimum clustering centroids are searched by intuitionistic fuzzy c ‐means clustering algorithm. Curve similarity measure algorithm is proposed in procedure of forecasting, which avoids influence of mutation points and overcomes limitation of priori information. Euclidean distance is replaced by Fréchet distance, it is appropriate for such directed time series in vector matching. The proposed model and relevant models are implemented on three different datasets, a synthetic dataset, the monthly total retail sale of social consumer goods and daily mean temperature dataset. The forecasting results, index mean square error and average forecasting error rate indicate that our model performance better in different time series patterns than others.