A GPU-based Distributed Algorithm for Linearized Optimal Power Flow in Distribution Systems

Abstract: We propose a GPU-based distributed optimization algorithm, aimed at controlling optimal power flow in multi-phase and unbalanced distribution systems. Typically, conventional distributed optimization algorithms employed in such scenarios rely on parallel execution with multiple CPUs. However, this often leads to significant computation time primarily due to the need for optimization solvers to solve subproblems for every iteration of the algorithms. To address this computational challenge, we propose a distributed optimization algorithm that eliminates solver dependencies and harnesses GPU acceleration. The central idea involves decomposing networks to yield subproblems with closed-form solutions based on matrix operations that GPU can efficiently handle. We demonstrate the computational performance of our approach through numerical experiments on four IEEE test instances ranging from 13 to 8500 buses. Our results affirm the scalability and superior speed of our GPU-based approach compared to the CPU-based counterpart.