Adaptive reinforcement learning framework for sustainable microgrid optimization in arid urban environments

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Energy Management
• Urban Sustainability

Background:

• Urban microgrids face challenges managing variable renewable energy and dynamic demand, especially in arid climates with extreme temperatures.
• Conventional energy management systems (EMS) lack the adaptability for fluctuating renewable generation and demand.
• Inefficiencies persist in coordinating diverse energy sources within urban microgrids.

Purpose of the Study:

• To develop a simulation-based, adaptive reinforcement learning (RL) framework for optimizing energy management in urban microgrids.
• To address inefficiencies in coordinating solar, wind, diesel, and battery resources.
• To enable real-time energy sharing between microgrids through inter-microgrid communication.

Main Methods:

• Developed a simulation platform integrating EnergyPlus with Python/TensorFlow RL agents.
• Implemented dynamic optimization for dispatching solar, wind, diesel, and battery resources.
• Integrated MQTT protocols for inter-microgrid communication and real-time energy sharing.

Main Results:

• Achieved high predictive accuracy for hourly and annual energy consumption (R² = 0.94 and 0.90).
• Reduced CO2 emissions by 14%, SO2 emissions by 13.6%, and primary energy consumption by 10% compared to baseline methods.
• Validated the framework in a case study simulating Riyadh's climatic conditions, showing improved operational and environmental performance.

Conclusions:

• The adaptive RL framework effectively manages energy in urban microgrids, particularly in arid regions.
• The integration of simulation, RL, and inter-microgrid communication offers a robust solution for energy management challenges.
• The proposed system demonstrates significant potential for enhancing the efficiency and sustainability of urban energy systems.