Aperiodic adaptive control for neural-network-based nonzero-sum differential games: A novel event-triggering...

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Game Theory

Background:

• Event-triggering control conserves resources in networked systems by using aperiodic sampling.
• Implementing event-triggering in neural-network-based nonlinear systems is an emerging research area.
• Nonzero-sum differential games with nonlinear dynamics present significant control challenges.

Purpose of the Study:

• To design novel event-triggering rules for neural-network-based nonlinear differential games.
• To reduce communication load between controllers and neural networks.
• To improve computational efficiency and learning accuracy in adaptive critic algorithms.

Main Methods:

• Design of event-triggering rules tailored for nonlinear dynamics and quadratic cost functions.
• Application of an adaptive critic algorithm for online Nash equilibrium learning.
• Introduction of an alarm sampling period to enhance learning accuracy.

Main Results:

• Successful design of event-triggering rules for the specified game class.
• Demonstration of reduced communication and computational load.
• Validation of approximate-optimal control performance through simulations.

Conclusions:

• The proposed event-triggering strategy effectively mitigates communication and computation in neural-network-based differential games.
• The adaptive critic algorithm with alarm sampling improves learning accuracy.
• The approach offers a promising direction for resource-efficient control of complex nonlinear systems.