A novel adaptive multi-critic based separated-states neuro-fuzzy controller: Architecture and application to chaos control. (May 2021)

• Record Type: Journal Article
• Authors: Eqra, N.; Vatankhah, R.; Eghtesad, M.
• Is Part Of: ISA transactions. Volume 111(2021)
• Page Start: 57
• Page End: 70
• Publication Date: 2021-05
• Subjects: Neuro-fuzzy -- PD critic -- Reinforcement learning -- Rule-based controller -- Chaos control; Engineering instruments -- Periodicals; Engineering instruments; Periodicals; Electronic journals; 629.805
• DOI: 10.1016/j.isatra.2020.11.020
• Languages: English
• ISSNs: 0019-0578

Abstract

Abstract: The design process of a novel adaptive critic based neuro-fuzzy controller is described. The concept is based on separating state variables into groups, assigning the groups to a multi-layer structure, and employing individual neuro-fuzzy controllers for each layer. The correlation procedure between layers is then defined and a proportional–derivative critic is generalized to be used with this structure. The controller's structure leads to a high reduction in the number of tunable parameters. For easier tuning, different gains are considered in the structure and an approach for the synthesis of the networks' initial parameters is given. The effectiveness of the controller is then verified by investigating two case studies: 1. Ball and beam regulation 2. Stabilizing the chaotic spinning disk's lateral vibrations. Simulations proved the feasibility and robustness of the proposed controller. Highlights: A novel multi-critic neuro-fuzzy controller is designed based on the concept of reinforcement learning. Proportional–derivative critic is generalized to be used with dynamic systems of any form. High reduction in the algorithm complexity is resulted using the proposed control scheme. Stabilization of the lateral vibrations of the chaotic spinning disk is done successfully.