Real-Time Transient Stability Prediction in Power Systems Using Ensemble Learning and Dynamic State Estimation

Abstract

This work presents a novel approach for real-time transient stability prediction in power systems, combining synchrophasor data with model-driven dynamic state estimations. The proposed method integrates dynamic state information, obtained through the Square Root Unscented Kalman Filter (SR-UKF), with an ensemble of machine learning (ML) algorithms in a Voting Classifier framework. The ensemble leverages the strengths of Multilayer Perceptron, Random Forest, Support Vector Machine, Convolutional Neural Network, and LightGBM to enhance predictive accuracy. By combining Phasor Measurement Unit (PMU) measurements with dynamic state estimations, the framework provides a more comprehensive view of transient stability, leading to improved prediction performance. Evaluated using a dataset from the NPCC 48-machine 140-bus system, the proposed method demonstrated accuracy gains ranging from 0.06% to 8.30%, with peak performance reaching 99.96% when dynamic states were included. Furthermore, the method exhibited strong generalization across varying system topologies and renewable penetration levels, maintaining high accuracy even under line outage conditions. This approach not only enhances prediction accuracy but also proves robust in challenging scenarios, such as noisy environments and short post-fault observation windows, making it well-suited for real-time applications in modern power systems.