|
Abstract
|
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 sh ortterm
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.
|