Short-Term Dynamic Voltage Stability Status Estimation Using Multilayer Neural Networks

The power grid stability is significantly impacted by the exponentially growing electrical demand and the complex electrical systems modernization projects. This intensifies the urgent need and yet challenging Dynamic Security Assessment (DSA) to withstand high-probability severe contingencies. This paper proposes an effective machine-learning solution for Short-Term Voltage Stability (STVS) detection and classification. This work also addresses fault detection and classification into line faults or bus faults under different operating conditions as a supplementary warning system to boost power system protection and resiliency with fast remedial actions. The proposed approach combines three necessary steps for high accuracy: feature subset selection, hyperparameter optimization, and critical bus identification. The efficiency of the proposed forecasting model is assessed using the IEEE New England 39-bus test case with the CLOD composite model. The generated N-1 contingency test cases data from dynamic Power System Simulator/Engineering (PSS/E) time domain simulations for fault-induced voltage events include the measured post-disturbance voltage magnitude, angle, frequency, and active and reactive power trajectories of the system buses. Numerical results from the proposed classifier confirm a high classification accuracy of 94.24% in identifying the post-disturbance stability state. The proposed method will be outperforming traditional shallow learning-based approaches. Further, the robustness of classifiers is demonstrated by evaluating the computational time, accuracy, precision, recall, and F-measure.