Transient Stability in Power Systems Using a Convolutional Neural Network

Monitoring power systems using intelligent systems for post-fault transient stability assessment (TSA) is critical for the grid to avoid cascading instability. Machine learning methods with synchrophasor measurements have been adopted widely for TSA due to the gradual deployment of wide area protection and control systems. TSA methods are crucial to alert operators or controls not only that a fault has occurred, but also that the power system has lost or will lose stability and therefore action must be taken. The two conditions that are important in real time application for a power system are accuracy and response time. Artificial Neural Networks and Support Vector Machines have been implemented in the past and have achieved satisfying classification accuracy. In this paper, a Convolutional Neural Network (CNN) system is proposed for TSA to maximize accuracy while minimizing response time to distinguish from older methods of machine learning. The proposed methodology is implemented on a Brazilian 7-Bus System and the WSCC 9 bus system. Power World is used for simulating the 3-phase short circuit faults on all buses in the case studies and MATLAB is used for simulating the CNN algorithm.