Lyapunov-Based Safe Reinforcement Learning for Microgrid Energy Management.

Area of Science:

• Electrical Engineering
• Computer Science
• Control Theory

Background:

• Renewable energy sources (RESs) integration into microgrids (MGs) necessitates advanced energy management.
• Model-based methods lack efficiency due to reengineering needs, while model-free methods face safety challenges during training.

Purpose of the Study:

• To propose a safe reinforcement learning (SRL) framework for efficient and safe MG energy management.
• To address the safety concerns of model-free approaches in MG operations.

Main Methods:

• Development of a safety assessment optimization model (SAOM) to evaluate and refine energy management schemes.
• Formulation of MG energy management as an assess-based constrained Markov decision process (A-CMDP).
• Application of Lyapunov-based safety policy optimization for agent policy learning to ensure operational safety.

Main Results:

• The proposed SRL framework effectively learns energy management policies for MGs.
• The framework ensures the safety of MG operations throughout the learning process.
• Demonstrated superior performance in the economic operation of MGs.

Conclusions:

• The SRL framework provides a safe and efficient solution for MG energy management.
• The method enhances the economic viability of MGs by optimizing energy operations.
• This approach overcomes the limitations of traditional model-based and model-free methods.