Adaptive multi-objective optimization of microgrid energy management using deep reinforcement learning considering...

Area of Science:

• Electrical Engineering
• Computer Science
• Renewable Energy Systems

Background:

• Microgrids enhance resilience but need advanced energy management systems (EMS) to balance cost, renewable integration, and component lifespan.
• Traditional EMS struggle with real-time adaptation and complex trade-offs, especially under forecast uncertainty and battery degradation concerns.

Purpose of the Study:

• Develop a deep reinforcement learning (DRL) based EMS for microgrid optimization.
• Address conflicting objectives: operational cost minimization, maximized renewable energy utilization, and battery degradation mitigation.
• Enhance microgrid operational robustness under forecast uncertainty.

Main Methods:

• Implemented a deep Q-network (DQN) agent for microgrid energy flow management in a simulated environment.
• Integrated solar PV, battery storage, controllable loads, and grid connection.
• Designed a reward function encompassing costs, battery degradation, and renewable utilization.

Main Results:

• Achieved a 12.01% reduction in operational costs compared to model predictive control.
• Reduced battery degradation by 8.19% and increased renewable energy utilization by 10.39%.
• Demonstrated robust performance under uncertainty, with only an 8.9% cost increase during severe forecast errors.

Conclusions:

• Deep reinforcement learning (DRL) effectively manages microgrids, balancing economic, environmental, and operational longevity objectives.
• The DRL-based EMS provides adaptive control superior to conventional methods, particularly under uncertain conditions.
• This approach shows significant potential for intelligent and resilient microgrid operation.