DenseMobileHybrid for Transient Stability Prediction of Power Systems

Transient stability assessment of power systems is essential to ensure the secure operation of power grids. The traditional time-domain simulation has high computational cost and is difficult to meet the online monitoring demand, while the existing data-driven methods have limitations in feature expression capability and computational efficiency. In this study, we propose a novel hybrid deep learning model, DenseMobileHybrid, to achieve efficient prediction of transient stability margins by fusing the dense connectivity mechanism of DenseNet, the lightweight design of MobileNetV2, and the nonlinear regression capability of the fully connected layer. Simulation experiments based on the IEEE 39-node system show that the proposed prediction model has a lower root-mean-square error of 2.2821, a lower average absolute percentage error of 0.7739, a lower average absolute error of 0.4793, and a higher coefficient of determination of 0.0069 than the optimal comparative model, and the results show that the DenseMobileHybrid prediction model has a significant performance advantage in the prediction of the transient stability of the power system, which can provide more efficient and accurate prediction support.