Event-triggered control for input-constrained nonzero-sum games through particle swarm optimized neural networks.

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Nonlinear System Dynamics

Background:

• Increasing system scale necessitates efficient control strategies.
• Traditional methods struggle with computational and communication resource demands.
• Obtaining Nash equilibrium solutions for complex systems is challenging.

Purpose of the Study:

• To develop an effective method for solving nonzero-sum games in partially unknown nonlinear systems.
• To improve system operation success rates and conserve resources.
• To address the need for efficient Nash equilibrium solutions in large-scale systems.

Main Methods:

• Integral reinforcement learning to eliminate drift dynamics requirements.
• Extended adaptive particle swarm optimization for neural network weight tuning.
• Event-triggering mechanism for reduced computational and communication loads.
• Lyapunov's direct method for closed-loop system stability verification.

Main Results:

• A novel integral reinforcement learning-based event-triggered control scheme was developed.
• The extended adaptive particle swarm optimization improved neural network training.
• The proposed method demonstrated superior performance compared to gradient descent, nonlinear programming, and standard particle swarm optimization.
• Reduced computational and communication burdens were achieved.

Conclusions:

• The developed control scheme effectively solves nonzero-sum games in partially unknown nonlinear systems.
• The combination of integral reinforcement learning, extended adaptive particle swarm optimization, and event-triggering offers significant advantages in performance and resource efficiency.
• The method ensures closed-loop system stability and outperforms existing algorithms.