A machine-learning-based complementary line pilot protection

• CNN-based pilot protection scheme proposed for high-voltage AC lines. • Enhances selectivity, sensitivity, speed, and reliability of transmission protection. • Outperforms traditional zero-sequence current differential protection methods. • Demonstrates strong fault resistance immunity and rapid fault tripping capabilities. While conventional main protection elements for high-voltage AC transmission lines exhibit satisfactory performance in a majority of scenarios, they still encounter challenges such as limited sensitivity, adverse effects of CT saturation, incapability in evolving fault scenarios, etc. To mitigate these limitations, this paper introduces a pilot protection scheme for transmission lines that leverages the convolutional neural network (CNN) paradigm. This innovative approach explores the integration of advanced data processing techniques with a fusion model rooted in machine learning technology to accurately identify fault classes and find the faulted line. As a complementary enhancement to phase-segregated current differential protection, the proposed complementary protection scheme offers substantial improvements in selectivity, sensitivity, speed, and overall system security. Its performance surpasses that of the widely adopted zero-sequence current differential protection in engineering practices. Furthermore, this study quantitatively assesses the impact of various factors, including noise, loading conditions, fault resistance, and associated faults, on the protection relay’s efficacy.