Data-Driven Joint Voltage Stability Assessment Considering Load Uncertainty: A Variational Bayes Inference Integrated With Multi-CNNs

Few studies have focused on assessing the transient and steady-state voltage stability status of dynamic systems simultaneously. This motivated us to propose a new concept referred to as joint voltage stability assessment (JVSA). Towards this end, this paper proposes a novel data-driven JVSA method considering load uncertainty. It combines multiple convolutional neural networks (multi-CNNs) and a novel variational Bayes (VB) inference for better JVSA accuracy. First, the multi-CNN model is utilized to fast estimate the maximum voltage deviations during the transient and steady-state process. Uncertain load scenarios and system topology under
$N$
-1 contingency with are chosen as inputs of each CNN model. Second, estimated voltage deviations are put into the VB inference to automatically infer the transient and steady-state voltage stability status. To validate its effectiveness, numerical simulations are performed on the modified WECC 179-bus system by comparing with benchmark algorithms. It is demonstrated that the proposed data-driven JVSA method is more accurate and faster than the conventional VSA method.