Tabular Foundation Model Enhances Power System Security Assessment.

Dynamic Security Assessment (DSA) is crucial for evaluating the stability of power systems against potential contingencies. Current data-driven DSA methods typically require extensive labeled datasets for training and often necessitate a separate model for each contingency, leading to high maintenance and poor generalization to new scenarios. This research proposes a novel approach using a tabular foundation model (TFM) to overcome these limitations. The TFM assesses stability through in-context learning, eliminating the need for retraining or hyperparameter optimization for each contingency. A single TFM can handle multiple contingencies simultaneously. The study also explores the use of electrical distance coordinates (EDC) as features, demonstrating how a few labeled samples with EDC encoding can dramatically improve the TFM's generalization to unseen contingencies. Experiments on the IEEE 68-bus system show the TFM achieving high accuracy with significantly fewer labeled samples than conventional methods.