Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
SOE: Fachverband Physik sozio-ökonomischer Systeme
SOE 4: Poster Session
SOE 4.6: Poster
Montag, 9. März 2026, 18:00–21:00, P4
GPU-Parallel Load-Flow Solvers with Low-Rank Updates for Contingency Analysis and Topology Optimization — •Marc Hunkemöller1,2, Nico Westerbeck3, Lars Schewe4, and Dirk Witthaut1,2 — 1Forschungszentrum Jülich, Institute of Climate and Energy Systems – Energy System Engineering (ICE-1), 52428 Jülich, Germany — 2Institute for Theoretical Physics, University of Cologne, Köln, 50937, Germany — 3University of Edinburgh, School of Mathematics, Peter Guthrie Tait Road, Edinburgh, EH9 3FD, UK — 4Elia Group, Boulevard de l’Empereur 20, 1000 Brussels, Belgium
Power-flow simulations based on the Newton-Raphson method are key tools for transmission system operators. GPUs may accelerate these computations and thus enable fast contingency analysis and new applications such as transmission topology optimization, which are increasingly important for integrating large shares of renewable energy and reducing dependence on fossil fuels. However, parallelization on GPUs is challenging, as the treatment of different topologies does not align well with the "Single Instruction, Multiple Data" paradigm. We present a GPU-parallel AC load-flow solver designed to overcome the difficulties introduced by changing sparsity patterns in the Jacobian matrix when the network topology varies. Our approach uses low-rank updates to separate the Jacobian’s dependence on the network structure from its dependence on the system state, allowing topology changes to be incorporated efficiently. In combination with iterative linear solvers, we explore different update orders and strategies to identify stable and fast solver configurations suitable for large sets of topology scenarios.
Keywords: GPU; Parallel computing; Load flow; Contingency analysis