An adaptive power system transient stability assessment method based on shared feature extraction

Machine learning-based power system transient stability assessment (TSA) faces challenges with performance degradation under varying operating scenarios. This paper proposes a robust and transferable adaptive TSA method based on shared feature extraction of the power system. A domain adversarial alignment network is used to train a shared feature extractor, aligning data before and after system variations to capture critical stability features. This reduces the need for extensive labeled data and improves assessment across different scenarios. When the system scenario changes, data and model knowledge are transferred simultaneously, maintaining high accuracy even with significant data loss in new scenarios. Testing on the IEEE 39-bus system and a 2179-node province-level system shows that the method achieves over 96% prediction accuracy with 30% data loss and sustains 97.99% accuracy in continuously changing scenarios, outperforming traditional methods. The results demonstrate the method's potential for real-world application with enhanced generalizability, robustness, and sustainable learning capability.