A fault recognition method for transmission systems based on independent component analysis and convolutional neural networks

Abstract

Accurate recognition of transmission line (TL) faults plays a vital role in protection systems, avoiding misoperations and false trips. Several studies in this area have revealed the limitations of existing algorithms, particularly when considering more complex electrical power systems (EPS), where some expected operations (capacitor and load switching, for example) can be misinterpreted as fault events. This work addresses the critical issue of fault recognition (FR) in TLs with a special focus on distinguishing faults from other types of disturbances. The primary objective of this study is to introduce an approach based on Convolutional Neural Networks (CNN) and Continuous Wavelet Transform (CWT) for FR in EPS. This method relies only on voltage signals acquired with a reasonable sampling frequency. The proposed method is enhanced with an Independent Component Analysis (ICA)-based preprocessing, making it robust and versatile in handling different types of disturbances, such as adjacent line faults, load switching and capacitor switching. The evaluation of the proposed method is done considering two different EPS, achieving an accuracy up to 99 %, showing the robustness of the method against different scenarios such as fault events and switching operations.