A harmonic source location strategy for distribution networks using group-sparse Bayesian learning under an edge-computing architecture

Conventional harmonic source location methods based on single-shot sampling or short-term measurements offer limited reliability. Therefore, this paper proposes a harmonic source location strategy for distribution networks using group-sparse Bayesian learning under an edge-computing architecture. First, an edge-computing enabled hierarchical and partitioned framework for harmonic source location in distribution networks is established. The functions of edge nodes at various levels and their collaborative workflow are explicitly defined, enabling local processing of harmonic data and preliminary location. Second, a federated conditional diffusion model is proposed to generate missing harmonic measurements, enabling the effective reconstruction of incomplete harmonic measurement data. Then, a method for locating local harmonic source based on grouped sparse Bayesian learning is proposed. It constructs a typical-day confidence-interval measurement set from long-horizon harmonic monitoring data, applies group-sparse Bayesian learning to estimate the harmonic state interval, and further enables initial harmonic source location under limited measurement devices. Finally, the effectiveness of the proposed method is validated in a simulation environment. The simulation results demonstrate that, even under sparse and partially missing measurements, the proposed strategy can achieve accurate harmonic source location.