Reinforcement Learning-Based Predefined-Performance Control for Nonlinear Switched Interconnected Systems

Area of Science:

• Control Theory
• Artificial Intelligence
• System Dynamics

Background:

• Nonlinear switched interconnected systems present significant control challenges due to unmeasurable states and complex switching behaviors.
• Existing control methods often struggle with guaranteed performance and adaptability to system variations.

Purpose of the Study:

• To develop a reinforcement learning (RL)-based control framework for nonlinear switched interconnected systems.
• To guarantee predefined performance metrics, including convergence time and accuracy.
• To address challenges posed by unmeasurable states and group average dwell time switching mechanisms.

Main Methods:

• Reconstruction of system equations targeting nonlinear and interconnected terms, approximated by neural networks (NNs).
• Design of an NN-based switching state observer for estimating unmeasurable states.
• Development of a distributed optimal controller using a backstepping framework with a performance transformation function integrated into the cost function.
• Approximation of the control law via an identifier-actor-critic architecture.
• Generalization of group average dwell time-based stability analysis for optimal control.

Main Results:

• The proposed framework effectively handles unmeasurable states and group average dwell time switching.
• Convergence time and accuracy can be preset through parameter configuration.
• The approach demonstrates enhanced extensibility and practicality compared to existing methods.
• Simulation examples validate the effectiveness and superiority of the proposed method.

Conclusions:

• The developed RL-based control framework provides a robust and adaptable solution for nonlinear switched interconnected systems.
• The method achieves guaranteed performance and overcomes key limitations of prior approaches.
• This work offers significant potential for real-world applications requiring precise and reliable control.