Answers are grounded in this paper and cite their sources.
Probabilistic power flow (PF) analysis is critical for secure operation of distribution networks under uncertainty from distributed energy resources. However, low observability in secondary networks and frequent topology reconfigurations complicate PF estimation and demand large sample sizes. This paper proposes a meta-graph generative autoencoder (Meta-GGAE) to learn probabilistic PF distributions under sparse sensing. Meta-GGAE employs node-level matrix completion within a bilevel meta-learning framework to infer unmeasured states, where physics-guided regularizers are adaptively learned using a graph attention network. A variational generative encoder embeds system physics to model local dependencies, while a conditional variational decoder captures probabilistic PF outputs by integrating latent variables with attention-based features. The proposed framework reconstructs missing measurements, generalizes to unseen topologies, and enables robust PF estimation under low observability. Simulations on IEEE test systems demonstrate improved accuracy and physical consistency under N-1 contingencies and uncertain injections compared with baseline methods.
No takes yet. Share an insight, caveat, or question.
Prashal et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: