Identification method for fuzzy forecasting models of time series

• Published in: Applied soft computing 2017-01, Vol.50, p.166-182
• Main Authors: Carvalho, J.G., Costa, C.T.
• Format: Article
• Language: English
• Subjects: Autoregressive seasonal fuzzy model; Electrical energy price; Futures market index; Fuzzy prediction; α-cut
• ISSN: 1568-4946; 1872-9681
• DOI: 10.1016/j.asoc.2016.11.003

[Display omitted] •An algorithm for forecasting models with hybrid structure in time series was implemented.•A triangular fuzzy number incorporated the random behavior of the variable with efficiency.•The performance of the suggested forecasting method was evaluated and optimized after simulation.•The proposed method in this study was tested experimentally on enrollment data and the TAIFEX index. 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 α-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.