A Unified Online Deep Learning Prediction Model for Small Signal and Transient Stability

This paper proposes a novel unified prediction approach for both small-signal and transient rotor angle stability as opposed to other studies that have only addressed transient rotor angle stability. Deep learning techniques are employed in this paper to train an online prediction model for rotor angle stability (RAS) using the voltage phasor measurements which are collected across the entire system. As a result, the trained model provides a fast yet accurate prediction of the transient stability status when a power system is subjected to a disturbance. Also, if the system is transiently stable, the prediction model updates the power system operator concerning the damping of low-frequency local and inter-area modes of oscillations. Therefore, the presented approach provides information concerning the transient stability and oscillatory dynamic response of the system such that proper control actions are taken. To achieve these objectives, advanced deep learning techniques are employed to train the online prediction model using a dataset which is generated through extensive time domain simulations for wide range of operating conditions. A convolutional neural network (CNN) transient stability classifier is trained to operate on the transient response of the phasor voltages across the entire system and provide a binary stability label. In tandem, a long-short term memory (LSTM) network is trained to learn the oscillatory response of a predicted stable system to capture the step-by-step dynamic evolution of the critical poorly damped low-frequency oscillations. The superior performance of the proposed model is tested using the New-England 10-machine, 39-bus, IEEE 16-machine, 68-bus, 5-area and IEEE 50-machine, 145-bus test systems and is verified with time domain simulation.