Identification method for fuzzy forecasting models of time series

In this paper, we propose a fuzzy forecasting methodology of time series, which is tested on two series: the price of electricity in New South Wales, Australia; and on the futures market index of Taiwan. The method uses a triangular membership function in a fuzzification process, including an alpha-cut, and applies the extended autocorrelation function. The identification algorithm enables optimization of the number of fuzzy sets to be used, to determine the optimal order for the fuzzy prediction model and estimate its parameters with greater accuracy. The fuzzy prediction models of time series found in the scientific literature are compared using mainly trivalent membership functions (0,0.5 and 1 as membership values), and the proposed method shows more accurate results. (C) 2016 Elsevier B.V. All rights reserved.