Statistically Evolving Fuzzy Inference System for Non-Gaussian Noises

Non-Gaussian noises always exist in the nonlinear system, which usually lead to inconsistency and divergence of regression identification applications. The conventional evolving fuzzy systems (EFSs) common sense have succeeded conquer uncertainties external disturbance employing specific variable structure characteristic. However, non-Gaussian would trigger frequent changes under transient criteria, severely degrades performance. Statistical criterion provides an informed choice strategies evolution, utilizing approximation uncertainty as observation model sufficiency. can be decomposed into term noise term, is suitable for condition, especially relaxing traditional Gaussian assumption. In this article, a novel incremental statistical inference system (SEFIS) proposed, has capacity updating parameters, components integrate new knowledge process characteristic, behavior, operating conditions with noises. generates rule based on sufficiency gives so insight whether models are reliable their approximations trusted. nearest presents inactive current data stream further deleted without losing any information accuracy subsequent trained when satisfied. our adaptive maximum correntropy extend Kalman filter derived update parameters rules cope problems improve robustness parameter process. shares estimate criteria make computation less burden dramatically. simulation studies show that proposed SEFIS faster learning speed more accurate than existing case free noisy conditions.