Physics-informed line graph neural network for power flow calculation.

Area of Science:

• Electrical Engineering
• Computational Science
• Artificial Intelligence

Background:

• Traditional power flow calculation methods are computationally intensive and struggle with large datasets.
• Existing machine learning methods for power flow lack accuracy due to overlooking transmission line importance and physical constraints.

Purpose of the Study:

• To develop a more accurate and efficient power flow calculation method.
• To incorporate physical mechanisms and transmission line relationships into machine learning models for power systems.

Main Methods:

• Proposed a physics-informed line graph neural network framework.
• Utilized incidence matrix and line graph matrix for information propagation between buses and transmission lines.
• Designed a physics-integrated loss function to ensure adherence to physical laws.

Main Results:

• The proposed model demonstrates enhanced prediction accuracy in power flow calculations.
• Experimental results validate the model's effectiveness across diverse power grid datasets and scenarios.
• The framework achieves improved interpretability compared to traditional methods.

Conclusions:

• The physics-informed line graph neural network framework offers a significant advancement in power flow calculation.
• Focusing on transmission line adjacency and physical principles improves model performance and reliability.
• This approach addresses the limitations of existing methods, paving the way for more robust power system analysis.