基于GPU的大规模配电网电磁暂态并行仿真技术

随着分布式电源等复杂设备的接入,针对配电网暂态过程的分析逐步依赖详细建模和电磁暂态仿真。图形处理器(GPU)等细粒度并行计算设备可显著提升配电网的电磁暂态仿真效率。提出了一种基于GPU的并行仿真技术,在GPU中加速了对大规模配电网系统的电磁暂态仿真。首先,将配电网电磁暂态仿真的计算过程分为异构计算、同构计算和网络求解三部分,并分别建立了细粒度并行计算模型。其次,针对上述三种计算模型,分别设计了基于分层有向图的异构计算核函数,基于积和熔加计算的同构计算核函数以及基于矩阵运算的网络求解核函数,最终实现了完全基于GPU的配电网电磁暂态仿真。对大规模配电网算例的仿真结果表明,所提出的细粒度计算模型和仿真算法可在保证仿真精度的前提下,提升在GPU中进行大规模配电网仿真的效率。

GPU-based Techniques of Parallel Electromagnetic Transient Simulation for Large-scale Distribution Network

With the integration of distributed generators and other complex devices, analyses of the transient state process for distribution systems and electromagnetic simulation are gradually becoming dependent on detailed modeling and electromagnetic transient(EMT)simulation. As EMT simulation can be appreciably accelerated by fine-grained parallel devices, such as the graphics processing unit(GPU). GPU-based parallel electromagnetic transient simulation techniques for large-scale distribution system are proposed. First, the computational models are formulated by categorizing the overall computations during simulation into three parts, that is, heterogeneous, homogeneous and network solution part. Then, each part of computation is accelerated using fine-grained parallel strategy on the GPU. The heterogeneous part is processed in a layered-directed-acyclic-graph based kernel, the homogeneous part is handled in a fused multiply-add kernel, and the network solution part is done by linear algebraic solvers on the GPU. Finally, large-scale cases are provided for tests. Simulation results indicate that the proposed techniques can significantly accelerate EMT simulation for distribution networks.