Contrastive-Active Transfer Learning-Based Real-Time Adaptive Assessment Method for Power System Transient Stability.

Area of Science:

• Electrical Engineering
• Computer Science
• Artificial Intelligence

Background:

• Machine learning for transient stability assessment faces challenges like imbalanced data and poor generalization.
• Existing methods struggle with real-time adaptive assessment when data distributions shift.

Purpose of the Study:

• To propose an intelligent enhancement method for real-time adaptive transient stability assessment.
• To improve model accuracy and adaptability when dealing with unbalanced and shifting data distributions.

Main Methods:

• Utilized a convolutional neural network (CNN) with contrastive learning for offline training to enhance recognition of unbalanced samples.
• Implemented an active transfer strategy with uncertainty-based active sampling for online model updating using new system data.
• Fine-tuned model parameters to reduce update costs and boost adaptability.

Main Results:

• The contrastive learning approach improved the CNN's accuracy in recognizing imbalanced samples.
• The active transfer strategy effectively adapted the model to new data distributions with reduced updating costs.
• Experiments on the IEEE39-node system demonstrated the proposed method's effectiveness.

Conclusions:

• The proposed intelligent enhancement method significantly improves real-time adaptive transient stability assessment.
• The combination of contrastive learning and active transfer learning offers a robust solution for data imbalance and distribution shift challenges.