A GPU-based Newton–Raphson Algorithm for Power Flow under Topology Changes

Fast and reliable power-flow computations are essential for power-system operation and planning, and they are increasingly challenged by the variability and structural complexity of modern power grids. Applications such as contingency analysis and topology optimization require solving many related AC power-flow problems that differ by localized topology changes. While Newton-Raphson methods provide robust solutions, repeated solves under topology variations pose computational challenges. We present a GPU-oriented Newton-Raphson power-flow solver that enables parallel computation across many grid topologies. Topology-induced changes are represented as low-rank corrections, allowing sparse factorizations with fixed sparsity patterns to be executed efficiently on GPUs. Numerical results on standard test systems illustrate scaling behavior and current limitations of sparse direct solvers.