Formal privacy guarantees for optimization datasets in power systems

Abstract:  The nexus of algorithmic privacy and optimization theory holds the potential to deliver solutions for decision-making without violating the privacy of underlying optimization datasets. Focusing on constrained optimization problems in the power systems domain, I will overview our work in two directions: privacy-preserving optimization and synthetic data generation. For the former, we develop a stochastic optimization counterpart for deterministic convex optimization problems that approximates optimal solutions without disclosing optimization datasets. In the second part, I will present several algorithms that generate synthetic data from real-world datasets, ensuring feasibility and statistical consistency with respect to optimization on the real data. I will supplement both parts with applications to power systems problems, including electric power grid dispatch and physics-informed wind power curve fitting.

Bio:  Vladimir Dvorkin is an Assistant Professor in the Electrical Engineering and Computer Science Department at the University of Michigan — Ann Arbor. Before moving to Michigan, he was a postdoctoral fellow at the Massachusetts Institute of Technology (Energy Initiative and LIDS) from 2021–2023. He earned his Ph.D. in electrical engineering from the Technical University of Denmark (DTU Elektro) in 2021 and also visited Georgia Tech’s School of Industrial and Systems Engineering during his Ph.D. studies. Vladimir’s research focuses on the energy transition towards a renewable-dominant and low-carbon energy supply, viewed through the lenses of optimization and machine learning, energy economics, and algorithmic data privacy. His work has received numerous recognitions, including the Marie Skłodowska-Curie Actions & Iberdrola Group postdoctoral fellowship and the IEEE Transactions on Power Systems Best Paper Award.