Graph Convolutional Network-Based Multi-harmonic Source Localization Method Without the Need for Synchronous Measurement Data

Harmonic source localization is essential for effective harmonic management. Traditional methods depend on the synchronous measurement of harmonic voltage and current phases, posing challenges in real-world engineering situations where only steady-state statistical values are often available. This paper proposes a novel multi-harmonic source localization approach utilizing Graph Convolutional Neural Networks (GCN). By exploiting GCN capabilities, it constructs an electrical distance adjacency matrix reflecting the power grid's topology and employs measurable parameters like harmonic voltage, current amplitude, and related data as input features. This innovative technique addresses the issue of missing synchronous phase information in engineering data, offering a new perspective on harmonic source localization. Demonstrated through comprehensive simulations, this method showcases its efficiency and superior performance, presenting a viable solution for practical implementations.