Dynamic event-triggered controller design for nonlinear systems: Reinforcement learning strategy.

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Nonlinear Dynamics

Background:

• Optimal control for discrete-time nonstrict-feedback nonlinear systems presents challenges in computational burden and communication frequency.
• Existing static-event-triggered strategies may not be optimal for reducing controller-actuator communication.
• Reinforcement learning and neural networks offer potential solutions for complex control problems.

Purpose of the Study:

• To develop an optimal control strategy for discrete-time nonstrict-feedback nonlinear systems.
• To introduce a dynamic-event-triggered control approach to minimize communication frequency.
• To enhance system performance and stability using advanced AI techniques.

Main Methods:

• Application of reinforcement learning-based backstepping technique with actor-critic neural networks.
• Development of a neural network weight-updated algorithm to mitigate computational load and local optima.
• Introduction of a novel dynamic-event-triggered control strategy.

Main Results:

• The dynamic-event-triggered strategy significantly outperforms static-event-triggered methods.
• A neural network weight-updated algorithm effectively reduces computational burden.
• Lyapunov stability theory confirms semiglobal uniform ultimate boundedness of all closed-loop system signals.

Conclusions:

• The proposed reinforcement learning-based backstepping control with a dynamic-event-triggered strategy is effective for nonlinear systems.
• The approach successfully balances control performance with reduced communication frequency.
• Numerical simulations validate the practicality and superiority of the developed control algorithms.