Multi-agent coordination and uncertainty adaptation in deep learning-assisted hierarchical optimization for...

Area of Science:

• Power Systems Engineering
• Optimization Theory
• Machine Learning

Background:

• Increasing renewable energy integration introduces complex uncertainties into power system operations.
• Conventional optimization methods struggle to manage multi-layered uncertainties from sources like solar and wind power.
• Existing frameworks face challenges in balancing economic efficiency, operational reliability, and renewable energy utilization.

Purpose of the Study:

• To propose a novel Deep Learning-assisted Distributionally Robust Optimization (Deep-DRO) framework for enhanced power system operation.
• To improve economic efficiency and operational reliability in power grids with high renewable energy penetration.
• To dynamically manage uncertainty by integrating deep learning for probabilistic inference and robust optimization for system-wide resilience.

Main Methods:

• Developed a hierarchical coordination architecture with deep learning modules for inferring uncertain variable distributions (solar, wind, load).
• Employed a multi-agent dispatch structure (county, feeder, DER layers) with reinforcement learning for adaptive coordination.
• Integrated deep networks for scenario distribution estimation and a robust optimization core to minimize costs and reliability penalties under ambiguity.

Main Results:

• The Deep-DRO model reduced operational costs by 11.0-13.5% and improved reliability indices from 0.864 to 0.911.
• Renewable energy utilization increased from 85.6% to 89.7%, with resilience maintained under 30% higher uncertainty variance.
• Achieved a 28.6% reduction in carbon emissions, demonstrating a balance between economic, environmental, and reliability objectives.

Conclusions:

• The Deep-DRO framework offers a generalizable paradigm for intelligent, risk-aware energy management in power systems.
• Hierarchical learning and robust optimization effectively enhance adaptive coordination and performance consistency under uncertainty.
• The study provides theoretical and practical implications for future smart grid autonomy, sustainable dispatch, and power system restoration.