Adaptive Fault-Tolerant Tracking Control for Discrete-Time Multiagent Systems via Reinforcement Learning Algorithm.

Area of Science:

• Control Engineering
• Artificial Intelligence
• Systems Science

Background:

• Multiagent systems face challenges in maintaining tracking control due to actuator faults and system uncertainties.
• Adaptive control strategies are crucial for robust performance in dynamic environments.
• Reinforcement learning offers a powerful framework for developing intelligent control solutions.

Purpose of the Study:

• To investigate adaptive fault-tolerant tracking control for discrete-time multiagent systems.
• To develop a reinforcement learning-based control strategy with reduced computational burden.
• To ensure system stability and performance despite actuator faults and dead zones.

Main Methods:

• Utilizing action and critic neural networks (NNs) for approximating control inputs and cost functions.
• Employing a combination of backstepping technique and reinforcement learning for direct adaptive optimal control.
• Implementing adaptive auxiliary signals to counteract fault effects and dead zones.
• Applying Lyapunov stability theory to guarantee system boundedness.

Main Results:

• The proposed method effectively approximates unknown control signals and estimates the cost function.
• A reduced computational burden is achieved compared to existing reinforcement learning algorithms.
• Adaptive auxiliary signals successfully compensate for actuator faults and dead zones.
• Simulation results demonstrate the effectiveness of the proposed fault-tolerant control approach.

Conclusions:

• The developed reinforcement learning algorithm provides an effective solution for adaptive fault-tolerant tracking control in discrete-time multiagent systems.
• The approach ensures semiglobally uniformly ultimately bounded stability for the closed-loop system.
• The method offers a computationally efficient alternative for complex control problems.