Complex network control and stability through distributed critic-based neuro-fuzzy learning

Abstract

Inspired by advancements in swarm autonomous vehicles and intelligent control systems, this research addresses the issue of frequency synchronization and phase tracking in oscillator networks. A novel distributed consensus protocol and a reinforcement learning algorithm for a multi-agent network with a leader–follower topology, considering stability conditions, are developed. The critic-based neuro-fuzzy learning (CBNFL) method aims to achieve consensus and minimize local tracking errors. Additionally, an explicit synchronization condition for the network using the Lyapunov theorem is derived. Each vehicle tracks its reference phase and frequency. Employing a fuzzy critic to evaluate the current state and generate a stress signal for the controller, the method prompts adaptive parameter adjustments to minimize this signal. The proposed design's versatility and adaptability to various networks demonstrate robustness against dynamic vehicle properties and network parameter uncertainties, ensuring consistent controller performance. This approach exhibits high scalability, accommodating numerous autonomous agents. To validate the proposed learning method's efficacy, numerical simulations are conducted on a network of five oscillators. The outcomes of implementing CBNFL compared with a conventional PI controller underscore the CBNFL method's superior performance and robustness in maintaining network stability and achieving synchronization.