Non-iterative procedure incorporated into the fuzzy identification on a hybrid method of functional randomization for time series forecasting models

This study presents a hybrid forecasting method that was designed and fine-tuned using the fuzzy set theory, combined with the classical methodology of time series modeling, to increase the accuracy of the forecasting algorithm. Thus, a fuzzy identification strategy is proposed, which considers the following (among other parameters): the extended autocorrelation function and the number of triangular fuzzy sets. Using a non-iterative procedure based on a fuzzy version of the extended autocorrelation function, the interpolation and generalization capabilities of fuzzy systems were exploited in order to obtain a robust forecast, even when considering a time series with insufficient data. In order to increase the accuracy of the forecasting algorithm, several parameters were tested and optimized using computational simulations. Several parameters inherent to the forecasting algorithm proposed in this study were tested and optimized. To evaluate the performance of the proposed methodology, two case studies were performed on two databases that are available in the literature. The proposed fuzzy forecasting algorithm yielded very good results, given the low forecasting errors that were returned by the identified models, especially when compared with the conventional statistical methodologies and other fuzzy forecasting methods that are established in the literature. (C) 2019 Elsevier B.V. All rights reserved.