Intelligent forecasting of time series based on evolving distributed Neuro‐Fuzzy network
Selmo Eduardo Rodrigues, Ginalber Luiz de Oliveira Serra · Computational Intelligence · 2020
Abstract An evolving methodology based on Neuro‐Fuzzy Takagi‐Sugeno network (NF‐TS) for distributed forecasting of univariate time series, is proposed. First, the unobservable components, or hidden patterns, are extracted from experimental data of the time series. Then, a distributed forecasting is performed separately for each component, considering an evolving NF‐TS associated with each extracted pattern. The evolving NF‐TS uses components data to adapt and adjust its structure, as the number of fuzzy rules increases or decreases according the behavior of the unobservable components. A recursive version of singular spectral analysis (SSA) technique is formulated, as one of the main contributions of this article, and it is applied to extract the components. The efficiency of proposed methodology is illustrated from results of comparison to others state‐of‐the‐art techniques for forecasting of various univariate time series.