The proposed study introduces a long short-term memory architecture integrated with a novel comprehensive dynamic security index to enable multi-domain stability assessment beyond post-fault dependencies. The comprehensive dynamic security index unifies voltage, frequency, and transient stability metrics into a single interpretable scalar, quantifying real-time proximity to instability boundaries while classifying system states into five actionable categories. By prioritizing generator terminal dynamics, the framework operates with reduced PMU coverage through strategic feature engineering. Validated on IEEE 14 and 118-bus systems, the long short-term memory-deep neural network (LSTM-DNN) model outperforms state-of-the-art techniques in both prediction speed and operational granularity. By bridging static and dynamic data streams, a hybrid attention mechanism improves operator confidence by linking model decisions to physical grid components. Results demonstrate robustness to class imbalance.
Shahriyari et al. (Tue,) studied this question.