Lifetime Citizen Portal Access — AI Briefings, Alerts & Unlimited Follows
Researchers outline differentially private synthetic grid data to enable analysis while limiting cyber risk
Loading...
Summary
University of Michigan researcher Vladimir Devorkin presented methods to release synthetic grid datasets that preserve OPF feasibility and cost consistency using differential privacy and post‑processing optimization, and added a cyber‑resilience objective to limit usefulness for adversaries.
Vladimir Devorkin of the University of Michigan presented algorithms for producing synthetic power‑system datasets that preserve statistical and operational properties needed for research while providing differential‑privacy guarantees to protect original data sources.
Devorkin described a pipeline that first perturbs original parameters (loads, line limits) with a randomized mechanism (Laplace/exponential mechanisms), then applies a bilevel post‑processing optimization to restore feasibility for power‑flow and OPF analyses and to match cost statistics. He noted that simple noise addition can break feasibility; the optimization step finds the closest feasible instance that retains cost consistency.
Because synthetic datasets can also aid adversaries, Devorkin added a cyber‑resilience term in the post‑processing optimization that reduces an attacker's ability to use the synthetic data to calibrate load‑redistribution attacks. He showed experiments across test systems (from small 5‑bus to IEEE reliability cases) where naive synthetic releases could enable effective attacks but his privacy‑preserving, cyber‑aware post‑processing substantially reduced attacker gains while keeping analytical value.
Devorkin suggested that differentially private synthetic data can enable controlled transparency — operators can choose privacy parameters to moderate how much operational information is released — and recommended further study on topology obfuscation and large‑scale deployment.
Questions focused on whether topology‑only approaches can create realistic datasets without original data; Devorkin said his method assumes access to original data and that topology‑only synthesis is a related but separate problem.
