Dynamic Event-Triggering Neural Learning Control for Partially Unknown Nonlinear Systems.

Area of Science:

• Control Systems Engineering
• Machine Learning
• Nonlinear Dynamics

Background:

• Adaptive learning algorithms are crucial for controlling complex systems.
• Event-triggered strategies reduce communication load but can suffer from static triggering limitations.

Purpose of the Study:

• To develop a novel dynamic event-triggering strategy for integral reinforcement learning in partially unknown nonlinear systems.
• To enhance learning performance and reduce data transmission through event-based control updates.

Main Methods:

• Utilized a policy iteration technique with two neural networks (actor and critic).
• Introduced a dynamic triggering rule based on a first-order filter to govern event occurrences.
• Implemented event-sampled integral reinforcement learning for adaptive control.

Main Results:

• The proposed dynamic triggering algorithm successfully avoids Zeno behavior.
• Theoretical analysis confirmed asymptotic stability of the event-driven system and convergence of network weights.
• Demonstrated significant reductions in data samples and transmissions compared to existing methods.

Conclusions:

• The dynamic event-triggering strategy offers an effective approach for adaptive learning in nonlinear systems.
• This method enhances efficiency by minimizing data transmission while maintaining robust learning performance and stability.
• The algorithm provides a promising solution for resource-constrained control applications.