A Modified CNN for Detection of Faults During Power Swing in Transmission Lines

Abstract

Fault detection, classification, and identification of location are key to ensure normal operation and healthiness of the power system including extra high voltage (EHV) and ultra-high voltage (UHV) transmission lines, where almost 85% of faults occur. One of the most important feature in distance relays is the power swing blocking function, a key element for preventing unintended tripping of transmission lines. Distinction of faults from power swing should be quick and specific to reduce widespread blackouts, and economic losses. This lead to the introduction of newer methods on fault analysis for much faster and accurate operation of relays. Modern methods use machine learning techniques which are still at its early stage. This paper proposes a modified 1-D Convolutional Neural Network (CNN) architecture to identify power swings and fault cases with high accuracy. An infinite bus system and Western System Coordinating Council (WSCC) 3-machine, 9-bus system are selected for investigating different fault conditions during power swing. A comparative analysis of different traditional machine learning (ML) and deep learning (DL) techniques are also conducted to validate the efficacy of proposed methodology.