An Event-Triggered ADP Control Approach for Continuous-Time System With Unknown Internal States.

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Nonlinear System Dynamics

Background:

• Traditional adaptive dynamic programming (ADP) methods often rely on fixed sampling periods, leading to high computational and transmission costs.
• Event-triggered control strategies aim to reduce these costs by updating controllers only when necessary.
• A key challenge is obtaining the system's entire state, which is often infeasible in practical applications.

Purpose of the Study:

• To propose a novel event-triggered adaptive dynamic programming (ADP) control method for nonlinear continuous-time systems with unknown internal states.
• To address the challenge of state estimation in event-triggered systems by integrating a neural-network-based observer.
• To reduce computational cost and transmission load compared to traditional fixed-period ADP methods.

Main Methods:

• Developed an event-triggered adaptive dynamic programming (ADP) control framework for nonlinear systems.
• Integrated a neural-network-based observer to estimate unknown internal states from measurable feedback.
• Designed a triggering condition for aperiodic updates of both the observer and the controller.
• Utilized neural networks for performance index estimation and control action calculation.
• Performed stability analysis using Lyapunov functions for continuous and jump dynamics.

Main Results:

• The proposed event-triggered ADP method effectively controls nonlinear systems with unknown internal states.
• The integrated neural-network-based observer successfully recovers system states from feedback.
• Aperiodic updates significantly reduce computational and transmission burdens.
• Stability of the control system is rigorously proven using Lyapunov analysis.
• Simulation results validate the theoretical findings and demonstrate the method's efficiency.

Conclusions:

• The novel event-triggered ADP control method offers an efficient solution for nonlinear systems with unknown states.
• The integration of a neural-network-based observer overcomes practical limitations of state observability.
• This approach provides significant reductions in computational and communication overhead.
• The method ensures system stability and demonstrates practical applicability through simulations.