DEDSAC: Centralized microgrid dispatch via dual exploration mechanism enhanced diffusion soft actor-critic.

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Energy
• Power Systems Optimization

Background:

• Microgrid clusters face complex power optimization and scheduling challenges due to internal unit heterogeneity and scale.
• Existing algorithms struggle with the complexity of large-scale, heterogeneous microgrid systems.

Purpose of the Study:

• To propose a novel centralized diffusion-based Soft Actor-Critic (SAC) framework, DEDSAC, for optimizing power scheduling in microgrid clusters.
• To address the limitations of current methods in handling the scale and complexity of microgrid networks.

Main Methods:

• Developed the Dual Exploration Mechanism Enhanced Diffusion Soft Actor-Critic (DEDSAC) framework.
• Incorporated dual probabilistic exploration mechanisms: diffusion-based latent action generation and probability-triggered selective noise injection.
• Utilized SAC-style squashed Gaussian reparameterization for actor optimization and a diffusion model as a robust policy network.

Main Results:

• DEDSAC demonstrated superior performance over standard SAC and other baselines, achieving higher-quality scheduling strategies.
• DEDSAC improved over SAC by 29.5% (small-scale) and 13.2% (large-scale) in converged evaluation rewards.
• DEDSAC performance closely approached the Oracle Model Predictive Control (MPC) upper bound in both small and large-scale scenarios.

Conclusions:

• DEDSAC is highly effective for optimizing power scheduling in large-scale microgrids.
• The framework enables efficient and sustainable energy management despite increasing system complexity and operational uncertainty.
• DEDSAC offers a promising solution for advanced microgrid energy management systems.