唐坤杰, 董树锋, 宋永华. 一种GPU-CPU异构运算框架加速的实时N-1交流潮流计算方法[J]. 中国电机工程学报, 2018, 38(15): 4329-4338,4633. DOI: 10.13334/j.0258-8013.pcsee.172148
引用本文: 唐坤杰, 董树锋, 宋永华. 一种GPU-CPU异构运算框架加速的实时N-1交流潮流计算方法[J]. 中国电机工程学报, 2018, 38(15): 4329-4338,4633. DOI: 10.13334/j.0258-8013.pcsee.172148
TANG Kunjie, DONG Shufeng, SONG Yonghua. A Real-time N-1 AC Power Flow Calculation Method Based on GPU-CPU Heterogeneous Computing Framework[J]. Proceedings of the CSEE, 2018, 38(15): 4329-4338,4633. DOI: 10.13334/j.0258-8013.pcsee.172148
Citation: TANG Kunjie, DONG Shufeng, SONG Yonghua. A Real-time N-1 AC Power Flow Calculation Method Based on GPU-CPU Heterogeneous Computing Framework[J]. Proceedings of the CSEE, 2018, 38(15): 4329-4338,4633. DOI: 10.13334/j.0258-8013.pcsee.172148

一种GPU-CPU异构运算框架加速的实时N-1交流潮流计算方法

A Real-time N-1 AC Power Flow Calculation Method Based on GPU-CPU Heterogeneous Computing Framework

  • 摘要: 随着电力系统规模的扩大,为了适应N-1安全校验日益上升的实时性和精确性的需求,提出一种图形处理单元—中央处理单元(graphics processing unit-central processing unit,GPU-CPU)异构运算框架加速的实时N-1交流潮流计算方法。算法中设计一种N-1潮流问题的拼接求解方法,将原本多个独立的潮流问题组合为一个。雅可比矩阵的拼接生成采用并行化处理,线性方程组的求解根据规模大小选择直接法或迭代法处理,其中迭代法采用并行化处理。算法整体分为CPU处理部分和GPU处理部分,CPU处理迭代初值的设定、节点导纳矩阵的形成、校验集合的形成、迭代值的修正、收敛性判断等步骤,GPU处理雅可比矩阵的拼接生成等步骤,修正方程组的求解根据其规模选择CPU求解或GPU求解,以达到快速求解的目的。算例表明,所提算法效率和精度高、空间占用小,与传统N-1潮流算法相比具有明显优势,能够满足电网实时N-1潮流计算的需求,具有工程应用价值。

     

    Abstract: With the increment of scale of power system, in order to satisfy the increasing real-time and accuracy requirements of N-1 security verification, a real-time N-1 AC power flow calculation algorithm based on graphics processing unit-central processing unit(GPU-CPU) heterogeneous computing framework was proposed. In the algorithm, a method for solving the N-1 power flow problem was designed, which concatenated several independent power flow problems into one. The concatenated Jacobi matrix was generated by parallel processing, and the linear equations were solved by the direct method or the iterative method according to the scale of equations, in which the iterative method was parallelized. The algorithm was divided into CPU processing part and GPU processing part. CPU processed iterative initial value set, node admittance matrix formation, check set formation, iterative value correction and convergence judgement while GPU processed generation of concatenated Jacobi matrix. Correction equations solution was processed by CPU or GPU depended on the scale of equations, in order to achieve fast solution. The case analysis shows that, the proposed algorithm has high efficiency, high accuracy and small storage space. It has advantages compared with traditional N-1 power flow algorithm, which can meet the real-time requirement of N-1 power flow calculation and has engineering application value.

     

/

返回文章
返回