PHD-EE RESEARCH SEMINAR "MULTICLASS SHORT-TERM VOLTAGE STABILITY ASSESSMENT USING PHYSICS-INFORMED 3D CONVOLUTIONAL NEURAL NETWORK IN SMART GRIDS"
Date: October 08, 2026 Time: 11:00 AM Location: EE Reading Room
Speaker: Afsheen Bibi Supervisor: Dr. Raheel Zafar & Dr. Zubair Khalid
The assessment of short-term voltage stability (STVS) in real time is crucial for monitoring frameworks to guarantee the secure operation of smart grids under large disturbances. The first application of a sliding three-dimensional convolutional neural network (3D-CNN) for short-term voltage stability assessment is proposed in this paper. Compared with a 2D-CNN, a 3D-CNN enables joint extraction of spatial and temporal dependencies from spatiotemporal data cubes by performing convolutions across the depth dimension. The proposed physics-informed preprocessing strategy integrates optimal bus order with voltage dynamic feature matrix construction. A novel data cube representation is introduced for 3D-CNN training, emphasizing temporal evolution while preserving spatial features. The proposed physics-informed 3D-CNN enables accurate multiclass STVS event classification (i.e., normal, alert, emergency, and unstable) directly from raw voltage measurements. A 3DCNN-based regressor is also developed to estimate a wide-area dynamic voltage index for quantifying the severity of STVS events. Simulation studies are performed on a modified IEEE 10-generator 39-bus system with integrated photovoltaic (PV) penetration modeled in DIgSILENT PowerFactory. Empirical results establish the superior performance of the proposed 3D-CNN over existing 2D-CNN with a classification accuracy of 99.89% in real-time short-term voltage stability assessment.
08
Oct
Date: October 08, 2026
Time: 11:00 AM
Location: EE Reading Room
Speaker: Afsheen Bibi
Supervisor: Dr. Raheel Zafar & Dr. Zubair Khalid