Power-grid stability predictions using transferable machine learning.

Area of Science:

• Complex network analysis
• Power systems engineering
• Computational physics

Background:

• Numerical models improve power grid stability but are computationally expensive.
• Estimating dynamic properties like frequency synchronization is challenging.
• High computational cost hinders complex network analysis for power grids.

Purpose of the Study:

• Investigate machine learning (ML) techniques for power grid synchronization stability.
• Compare ML model performance using homogeneous versus heterogeneous synthetic power grids.
• Assess the transferability of ML models trained on synthetic data to real-world power grids.

Main Methods:

• Tested three ML algorithms: random forest, support vector machine, and artificial neural network.
• Trained models on two types of synthetic power grids: homogeneous and heterogeneous input-power distribution.
• Validated model performance on real-world power grids from Great Britain, Spain, France, and Germany.

Main Results:

• ML models achieved better synchronization stability prediction when trained on heterogeneous power grids compared to homogeneous ones.
• The developed ML algorithms demonstrated transferability from synthetic to real-world power grid stability prediction.
• Heterogeneous power grid data enhances the predictive power of ML models for synchronization stability.

Conclusions:

• Machine learning offers a computationally efficient approach to estimate power grid synchronization stability.
• Heterogeneous input-power distribution in training data is crucial for accurate ML-based stability prediction.
• ML models trained on synthetic data show significant potential for real-world power grid stability analysis.