State-Space Modeling with Type-2 Fuzzy Logic: An Evolving Neural-Fuzzy Model for Handling Uncertain Experimental Data

• Published in: Journal of control, automation & electrical systems Vol. 36; no. 4; pp. 704 - 721
• Main Authors: Evangelista, Anderson P. F., Serra, Ginalber L. O.
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.08.2025; Springer Nature B.V
• Subjects: Accuracy; Artificial intelligence; Clustering; Complex systems; Control; Control algorithms; Control and Systems Theory; Dynamical systems; Electrical Engineering; Engineering; Filtration; Fuzzy logic; Fuzzy sets; Identification; Inference; Mechatronics; Neural networks; Noise levels; Nonlinear systems; Parameter estimation; Process controls; Robotics; Robotics and Automation; State space models; Uncertainty; Incremental fuzzy learning; Interval type-2 fuzzy sets; Type-2 fuzzy instrumental variable; Type-2 fuzzy state-space modeling
• ISSN: 2195-3880; 2195-3899
• DOI: 10.1007/s40313-025-01183-4

In real-world identification problems, dynamic systems are typically nonlinear, complex, and subject to uncertainty. Additionally, experimental data can be corrupted by stochastic noise with varying statistical properties, such as uniform, Gaussian, or autocorrelated noise. To address these challenges, this paper proposes a novel evolving Interval Type-2 Neural-Fuzzy Model. The proposed methodology has two key contributions. First, a filtering layer performs data filtering, mitigating noise and ensuring robustness in both antecedent and consequent estimation. The filtered data enhance the accuracy of the model. Second, an Interval Type-2 fuzzy inference engine computes a confidence region, where the degree of uncertainty is dynamically adjusted based on the noise level in the experimental data. The model is structured into five layers: (1) a filtering layer that applies a recursive moving-average filter; (2) an evolving antecedent estimation layer that partitions the data space using an evolving type-2 fuzzy clustering algorithm; (3) a rule activation layer that computes the degree of rule activation; (4) a recursive submodel estimation layer that updates model parameters using interval type-2 instrumental variables method; and (5) a type-2 fuzzy inference engine that estimates the confidence region. Experimental results on nonlinear SISO systems and rocket trajectory estimation demonstrate the competitive model performance in handling noise and achieving accurate identification. The eIT2NFSO provides a robust and adaptable framework for modeling complex systems in noisy environments.