Batched Fast Decoupled Load Flow for Large-Scale Power System on GPU

Load flow (power flow) solution is the most important calculation in power system applications. It can also be applied to multi-scenarios study, including static voltage stability analysis, Monte-Carlo-based probabilistic load flow and etc. However, such solution relies on batched load flow calculations, which involves high amount of calculations, especially for large-scale power system. As a result, it is necessary to accelerate batched load flow calculations for large-scale systems using the state-of-the-art computing techniques. In this paper, a Layered Directed Acyclic Graph (LDAG) based method is developed to construct the batched load flow algorithm, which exploits both the sparsity of the power system and the fine-grained parallelism of the load flow algorithm. By taking the advantages of the data multiplicity among the scenarios, the proposed batched solution significantly accelerates the multi-scenario load flow study on a Graphics Processing Unit (GPU). With a batch size of 1024, it can achieve over 24x speedup for a 13659-bus system compared with implementations on CPU.