Short-term Voltage Stability Assessment in Power Systems based on Deep Learning

The growing integration of renewable energy resources within distribution networks has intensified the coupling and interactions between transmission and distribution systems. Such a shift introduces additional uncertainty, further complicating voltage stability assessments and highlighting the urgent need for enhanced situational awareness to mitigate blackout risks. This paper therefore proposes a convolutional neural network (CNN) based short-term voltage stability (STVS) assessment method for power systems. Following a description of the STVS problem, a time domain simulation (TDS) model of power systems is formulated. This model is subsequently solved repeatedly to generate sufficient training data in an offline manner. Using this training dataset, a CNN-based STVS assessment method is developed and trained to capture the intrinsic relationship between system measurements and STVS indices. To reduce the volume of training data and simplify the learning process, inputs and outputs of the CNN are extracted based on critical voltage regions. Once trained, the neural networks are deployed online to enable efficient and accurate STVS assessment. The proposed approach is evaluated on a modified New England 39-bus power system, with simulation results confirming its effectiveness in both assessing voltage stability and quantifying levels of voltage instability.