Data-Driven Optimized Output Regulation for Markov Jump Linear Systems and Its Application

Area of Science:

• Control Systems Engineering
• Applied Mathematics
• Machine Learning

Background:

• Markov jump linear systems (MJLSs) present complex dynamics requiring robust control strategies.
• The linear optimized output regulation problem (LOORP) is crucial for achieving desired system outputs under varying conditions.
• Existing methods for solving regulator equations have limitations in handling unknown system dynamics.

Purpose of the Study:

• To develop an effective model-based scheme for solving the LOORP in MJLSs.
• To introduce a reinforcement learning (RL)-based iteration scheme for LOORP, enhancing robustness and applicability.
• To validate the proposed RL scheme's performance in a practical distributed generation system.

Main Methods:

• Improvement of existing methods for solving regulator equations.
• Development of a model-based control scheme for LOORP in MJLSs.
• Application of a reinforcement learning (RL)-based iteration scheme for robust output regulation.

Main Results:

• A novel model-based scheme effectively addresses the LOORP for MJLSs.
• The RL-based iteration scheme successfully solves LOORP even with partially unknown system dynamics and unstable initial control gains.
• The proposed RL scheme demonstrates superior performance in an LCL-coupled inverter-based distributed generation system.

Conclusions:

• The developed model-based and RL-based schemes offer advanced solutions for the LOORP in MJLSs.
• The RL-based approach provides a robust and adaptable method for output regulation under uncertainty.
• The practical application highlights the efficacy and potential of the proposed RL scheme in power systems.