Real-Time Contingency Screening for Converter-Dominated Power Systems: A Graphical-DeepONet Approach

Existing contingency screening (CS) methods focus on the transient dynamics of synchronous generators (SGs) but overlook the transient behaviors of power electronic converters, leaving a difficult task of comprehensively assessing transient stability of both SGs and converters. In addition, the transition from traditional SG-dominated power systems to converter-dominated power systems has led to the emergence of higher order system models, placing higher computational requirements on real-time CS. To address the above-mentioned challenges, a Graphical-DeepONet is developed to predict transient dynamics trajectories of SGs and converters. Compared to the traditional time-domain simulation (TDS) requiring the iterative computation of differential–algebraic equations or the DL-based methods learn from the initial trajectory generated by TDS, the developed model is a TDS-independent solution, which learns to approximate the solution operators that map prefault measurements to the transient dynamics trajectories, thus significantly improving the computational efficiency. To learn the spatial–temporal features related to$N-1$contingency, which leads to diverse system topologies, a graph neural network-aided branch net is designed to improve learning capability. Comparisons are conducted on the IEEE 39 Bus System and IEEE 118 Bus System to validate the effectiveness and efficiency of the developed approach.