Inverse Optimal Control in Conjunction With Inverse Reinforcement Learning for Distributed Parameter Systems.

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Optimal control policies may perform poorly in real-world distributed parameter systems (DPSs) due to model bias.
• Predefined reward-weight matrices can lead to performance degradation in optimal control processes.

Purpose of the Study:

• To design an inverse optimal control (IOC) strategy for DPSs with unknown dynamic parameters.
• To address the challenge of transferring optimal control policies to real-world systems using human-behavior learning (HBL).
• To overcome performance degradation issues caused by fixed reward-weight matrices.

Main Methods:

• Utilized human-behavior learning (HBL) to transfer optimal strategies from reference systems to real-world DPSs.
• Employed inverse reinforcement learning (IRL) policy iteration algorithm for IOC in reference systems.
• Solved for equivalent reward-weight matrices and optimal control gains of reference systems.

Main Results:

• Successfully transferred optimal control strategies to DPSs with unknown parameters.
• Developed a method to derive reward-weight matrices and control gains via IRL.
• Demonstrated the effectiveness and superiority of the proposed algorithms through simulations.

Conclusions:

• The proposed IOC design effectively handles unknown dynamic parameters in DPSs.
• HBL enables robust transfer of optimal control policies, mitigating model bias.
• The IRL-based approach successfully determines reward functions and control gains, improving system performance.