A hybrid transformer-PPO framework for multi-objective energy management in renewable-based microgrids.

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence
• Control Theory

Background:

• Microgrids with high renewable penetration face challenges in managing energy supply and demand due to intermittent generation.
• Intelligent energy management is crucial for optimizing economic profit, environmental impact, and grid stability.
• Existing methods often struggle to balance multiple, conflicting objectives in dynamic microgrid environments.

Purpose of the Study:

• To develop and evaluate a hybrid AI framework for intelligent energy management in renewable-based microgrids.
• To optimize microgrid operations by jointly considering economic, environmental, and grid dependency objectives.
• To assess the framework's flexibility and scalability in adapting to different operational priorities.

Main Methods:

• A hybrid framework combining a Transformer-based forecasting model and a Proximal Policy Optimization (PPO) agent.
• The Transformer model predicts load, PV, wind generation, and electricity prices 24 hours ahead.
• A weighted multi-objective reward function in the PPO agent balances profit maximization, CO₂ emission reduction, and import dependency minimization.

Main Results:

• The proposed framework significantly increased operational profit compared to baseline methods.
• A substantial reduction in CO₂ emissions associated with grid electricity usage was achieved.
• The system demonstrated a notable decrease in dependency on external power imports.
• Sensitivity analysis confirmed the impact of reward weights and input window length on performance.

Conclusions:

• The hybrid Transformer-PPO framework offers a flexible and scalable solution for data-driven microgrid energy management.
• Integrating advanced sequence modeling with reinforcement learning effectively addresses uncertainty in renewable energy systems.
• The approach facilitates sustainable and optimized operation of renewable-based microgrids.