Locally activated neural networks and stable neural controller design for nonlinear dynamic systems

Area of Science:

• * Control Engineering
• * Artificial Intelligence
• * Nonlinear Dynamics

Background:

• * Nonlinear dynamic systems present significant control challenges.
• * Traditional control methods often struggle with real-time adaptation and complex dynamics.
• * Locally activated neural networks offer a promising approach for localized system representation.

Purpose of the Study:

• * To propose a stable neural control scheme for nonlinear dynamic systems.
• * To leverage locally activated neural networks for efficient and real-time control.
• * To ensure global stability and tracking error convergence in the control system.

Main Methods:

• * Employing a locally activated neural network with a CMAC-like content addressing mechanism.
• * Implementing a two-stage learning process: prediction error-based and tracking error-based learning.
• * Integrating sliding mode control as a safety mechanism for stability during learning.

Main Results:

• * The locally activated neural network effectively learns nonlinear basis functions of plant inverse dynamics.
• * Global stability and tracking error convergence were proven under specific assumptions.
• * Computer simulations validated the efficacy of the proposed control scheme.

Conclusions:

• * The proposed neural control scheme provides a stable and efficient method for controlling nonlinear dynamic systems.
• * Localized learning and sliding mode control contribute to robust real-time performance.
• * The approach facilitates global control through the cooperation of local control efforts.