电力系统全纯嵌入潮流的并行计算  

Parallel Computing Method of Power System Holomorphic Embedded Power Flow

作　　者：李雪[1] 高翔 姜涛[1] 王长江[1] 李国庆[1] Li Xue;Gao Xiang;Jiang Tao;Wang Changjiang;Li Guoqing(Key Laboratory of Modern Power System Simulation and Control&Renewable Energy Technology Ministry of Education Northeast Electric Power University,Jilin 132012 China)

机构地区：[1]现代电力系统仿真控制与绿色电能新技术教育部重点实验室(东北电力大学),吉林132012

出　　处：《电工技术学报》2024年第18期5839-5854,共16页Transactions of China Electrotechnical Society

基　　金：国家自然科学基金资助项目(52077029,52377083,U2066208,U23B20131)。

摘　　要：潮流计算是电力系统规划和运行的基础,全纯嵌入潮流计算方法(HELM)因无需初值且具有全局收敛性,因而在电力系统潮流计算中受到极大关注。然而,采用HELM求解大规模电力系统潮流时,高维幂级数系数线性方程组求解和节点电压的幂级数有理的逼近计算量大、耗时久,是制约HELM计算效率提升的关键。为此,该文提出一种基于稳定双正交共轭梯度(BICGSTAB)和Aitken差分的电力系统全纯嵌入潮流并行计算方法,该方法首先采用近似逆预处理的BICGSTAB法并行迭代求解HELM的高维幂级数系数线性方程组,以快速计算节点电压的各阶幂级数系数;其次,借助Aitken差分法实现所有节点电压幂级数有理逼近值的并行计算;然后,基于CPU-GPU异构平台设计所提算法的并行流程,以实现大规模电力系统潮流的快速求解;最后,通过节点在1 354~13 802的不同规模测试系统对所提方法进行分析、验证。结果表明,所提电力系统潮流全纯嵌入并行计算方法可实现电力系统潮流的准确、快速求解。Power flow calculation is the foundation of power system analysis and control.At present,the Newton-Raphson(NR)method is still the most widely used,and its convergence depends on the initial values election.When the power flow is overloaded,the Jacobian matrix of the NR method is prone to singularity,resulting in divergence of the power flow calculation,making it difficult to meet the efficiency and robustness requirements of complex power flow calculation.Unlike the NR method,the holomorphic embedded power flow(HELM)is a recursive rather than an iterative process.This method fixes the center of the expansion at the initial point and expands the power series term to obtain the target solution.Accordingly,the initial value is no longer to be set,ensuring convergence to the power flow solution and providing a clear signal when there is no power flow solution.However,when solving large-scale power flow,HELM requires multiple solutions to high-dimensional power series coefficient linear equations and Padéapproximation,and its computational efficiency is lower than the NR method.Therefore,improving the computational efficiency of HELM is crucial.This paper proposes a power system holomorphic embedded power flow parallel calculation method based on stable bi-conjugate gradient stabilized(BICGSTAB)and Aitken difference.The BICGSTAB method is used to solve high-dimensional power series coefficient linear equations iteratively.A sparse matrix approximate inverse preprocessor is constructed based on the coefficient matrix of the equation system to improve the convergence effect of the BICGSTAB method and quickly obtain the coefficients of each order of the voltage power series.Then,based on the traditional Aitken difference method,a new Aitken difference calculation for solving voltage is constructed to rapidly calculate rational approximation values of power series of voltage.Next,with the CPU-GPU heterogeneous computing platform,the large-scale matrix vector in the BICGSTAB method is calculated using the GPU,and the log

关 键 词：全纯嵌入法 潮流计算 Aitken差分法 CPU-GPU异构运算平台 预处理器 

分 类 号：TM744[电气工程—电力系统及自动化]