Transient Angle Stability Prediction of Virtual Synchronous Generator Using LSTM Neural Network

Abstract

Virtual synchronous generator (VSG) attracts great attention for mimicking synchronous generators but suffers from transient instability. Predicting the stability is important for protecting the VSG. Unlike synchronous generators, quick and precise prediction is needed for VSG due to the lack of physical inertia. In this paper, a long-short term memory (LSTM) neural network is proposed to predict hundreds of milliseconds in the future of VSG’s synchronousness and stability margin, but only takes dozens of milliseconds. Furthermore, the input and output data of the proposed LSTM is designed based on singular perturbation theory so that quick and accurate prediction is guaranteed. Simulation result shows that proposed LSTM possesses a great potential in online prediction.