Accelerating Unbalanced Distribution Power Flow on GPUs Using Sparse Inverse Factors

This letter presents a GPU-accelerated implementation of the Current Injection Method for Power Flow (CIM-PF) using a Sparse Inverse Factor Matrix-Vector Multiplication (SIF-MVM) linear solver. Unlike conventional LU-based solvers that are inefficient on GPU architectures due to their sequential nature, the proposed method leverages GPU parallelism through precomputed sparse inverse LU factors and parallel matrix-vector operations. The approach is evaluated on large-scale distribution systems with up to 168,435 nodes. For the largest test case, the solver achieves over 4× acceleration in single-scenario power flow computations and up to 8× acceleration in batch simulations involving 1,000 distinct current injection scenarios, compared to CPU-based methods. These results demonstrate the method's suitability for high-throughput applications such as time-series analysis, probabilistic studies, and large-scale planning simulations.