Predicting dynamic stability from static features in power grid models using machine learning.

Area of Science:

• Electrical Engineering
• Network Science
• Data Science

Background:

• Reliable electric power is essential for society.
• Transmission line failures threaten power grid stability, potentially causing fragmentation.
• Existing simulation models need complementary assessment methods.

Purpose of the Study:

• To develop and evaluate a novel approach for predicting power grid desynchronization events.
• To integrate network science metrics with machine learning for enhanced stability assessment.
• To identify key network properties influencing grid robustness and vulnerability.

Main Methods:

• Utilized network science metrics (e.g., redundancy, centrality) to characterize transmission lines.
• Employed machine learning models for feature selection and prediction of desynchronization.
• Trained and tested models on simulated data from synthetic power grids.

Main Results:

• Achieved an average precision greater than 0.996 in predicting desynchronization events post-line failure.
• Demonstrated the capability of learning transfer between different datasets with minimal performance degradation.
• Identified a few critical network metrics governing power grid desynchronization.

Conclusions:

• The integrated network science and machine learning approach effectively predicts power grid desynchronization.
• Network metrics quantifying rerouting capacity and static line loading are key factors.
• This method offers a promising tool for enhancing power grid stability and reliability.