The Use of Machine Learning for Prediction of Post-Fault Rotor Angle Trajectories
Xinlin Ye, Ana Radovanović, Jovica V. Milanović
Abstract
This paper proposes a machine learning-based method for predicting generator rotor angle responses (trajectories) following large disturbance in power system. A Long Short-Term Memory (LSTM)-based Recurrent Neural Network (RNN) is used to predict responses at any time instant after the fault inception by designing the input and output of the network with predefined sliding time windows. The numbers of neurons in the LSTM and Fully-Connected (FC) layers are optimised with the Particle Swarm Optimisation (PSO) algorithm, which was proved to be effective in similar tasks in past research. A wide range of realistic constraints associated with the use of the Phasor Measurement Unit (PMU) data has been considered, to demonstrate the feasibility of the proposed method when applied in real systems. Results obtained using modified IEEE 68 bus test system show that the proposed method can predict the future rotor angle responses accurately, and that is highly robust towards the imperfections of the realistic PMU data.