Towards Reliable Transient Stability Prediction of Power Systems: A CNN-Based Deep Ensemble Model with Optimized...

Area of Science:

• Electrical Engineering
• Computer Science
• Data Science

Background:

• Phasor measurement units (PMUs) enable data-driven transient stability prediction (TSP) in power systems.
• Reliability remains a key challenge for practical data-driven TSP applications.
• Existing methods struggle to guarantee the accuracy and trustworthiness of TSP results.

Purpose of the Study:

• To develop a reliable data-driven transient stability prediction (TSP) method.
• To enhance the practical applicability of TSP in power systems.
• To address the challenge of ensuring reliability in data-driven TSP.

Main Methods:

• A convolutional neural network (CNN)-based deep ensemble model was developed for TSP.
• Time-series generator data was transformed into image-like inputs for the CNN.
• A confidence-based, class-specific thresholding rule was optimized using a multi-objective model.
• The optimization aimed for 100% true unstable rate (TUR), minimized rejection rate, and maximized true stable rate (TSR).

Main Results:

• The proposed method achieved a 100% true unstable rate (TUR).
• A true stable rate (TSR) of at least 99% was maintained.
• Approximately 10% of samples were rejected, balancing reliability and prediction coverage.
• The model demonstrated effectiveness and scalability on benchmark power systems.

Conclusions:

• The CNN-based deep ensemble model with optimized thresholds significantly improves TSP reliability.
• The method provides accurate confidence estimation crucial for power system stability assessment.
• This approach offers a scalable and effective solution for reliable, data-driven transient stability prediction.