Data-based decentralized control of nonlinear-constrained interconnected systems using reinforcement learning.

Area of Science:

• Control Theory
• Nonlinear Systems
• Artificial Intelligence

Background:

• Designing controllers for interconnected nonlinear systems with asymmetric input constraints is challenging.
• Existing methods often require precise system models, which are not always available.
• Decentralized control strategies are crucial for large-scale systems like power grids.

Purpose of the Study:

• To develop a data-based decentralized controller for mismatched interconnected nonlinear systems with asymmetric input constraints.
• To leverage reinforcement learning for solving complex optimal control problems without explicit system models.
• To validate the controller's effectiveness using a practical example.

Main Methods:

• A data-based policy iteration (PI) algorithm within a reinforcement learning framework was employed.
• An actor-critic structure using neural networks (NNs) was implemented for the PI algorithm.
• The weighted residuals' method and Monte Carlo integration were combined to train the actor and critic NNs.

Main Results:

• The decentralized controller was designed based on solutions to unconstrained-input optimal control problems of auxiliary subsystems.
• The proposed method successfully determined the weight parameters for actor and critic NNs simultaneously.
• The controller demonstrated effectiveness in stabilizing an interconnected power system.

Conclusions:

• The developed data-based decentralized controller effectively handles mismatched interconnected nonlinear systems with asymmetric input constraints.
• The reinforcement learning-based policy iteration algorithm provides a viable approach for solving the associated Hamilton-Jacobi-Bellman equations.
• The study validates the practical applicability of the proposed control strategy in complex systems.