High-precision Transient Stability Assessment Using Convolution Neural Network

Abstract

Transient stability assessment (TSA) is one of the major and essential tools for power system operators. It is of utmost importance from the perspective of power system operation and control to develop a fast and accurate TSA that assists system operators in preventing cascading failures, instabilities, and large area blackouts. This paper proposes a convolution neural network (CNN)-based TSA which exploits transient data collected during a fault to make a reliable prediction of system stability. Temporal voltage phasors, which are captured through Phasor Measurement Units (PMUs) simulated on Real-time Digital Simulators (RTDSs), are used as an input feature to the CNN. Real-time simulation cases are modeled and automated for simulating fault scenarios to generate training and testing data. Furthermore, a laboratory-scale data acquisition and storage framework is developed with the integration of RTDSs, virtual Phasor Data Concentrator (OpenPDC), and MySQL. The significance of the proposed work is to provide a TSA tool for power system operators that can be incorporated with contingency analysis tools. The proposed approach is demonstrated through a comprehensive study on the reduced WECC 9-Bus System. Compared to other machine learning (ML) algorithms, the proposed CNN demonstrates the highest accuracy.