Generalized correntropy-based evolving fuzzy system for robust continuous learning

This study presents a generalized correntropy-based evolving fuzzy system (GCEFS) to enhance robustness and adaptability in continual learning under non-Gaussian disturbances. Unlike traditional evolving fuzzy systems that rely on mean squared error optimization, GCEFS utilizes generalized correntropy for simultaneous structural and parametric adaptation, capturing higher-order statistics and providing superior resilience to heavy-tailed outliers. Starting from an empty rule base, GCEFS evolves its architecture online through a combined generalized correntropy and Euclidean distance criterion, while recursively updating consequent parameters via a generalized maximum correntropy criterion (GMCC) algorithm. A Lyapunov-based analysis establishes bounded convergence and system stability, with generalizability to MSE, LMP, and MCC-based variants. Comprehensive evaluations on chaotic time series, time-varying nonlinear systems, regression benchmarks, and aircraft engine prognostics under diverse non-Gaussian noises demonstrate GCEFS’s assured stability, superior accuracy, and enhanced adaptability over state-of-the-art evolving fuzzy systems.