摘要
智能化安全评估是统一坚强电网建设的关键内容,是互联输电网的神经中枢,是维系电力生产过程的基础,是保障电网运行和发展的重要手段。随着并行计算技术和网格计算技术在我国电力系统中的广泛应用,使得利用系统分析特点和先进计算技术开展智能化安全评估系统研究与设计工作成为可能。本文基于同构并行技术、异构并行技术研究了暂态稳定分析并行算法,基于网格技术研究了电力系统动态安全评估。主要成果如下:
1、结合了现代计算机多核心的特点,对电力系统暂态稳定分析的算法及并行化设计进行了研究。在引入基于因子表路径树的网络分割方法上,将电力网络系数矩阵写成对角块加边形式,结合并行环境几乎零延迟的特点,发展了并行求解电力网络方程的算法,并在共享内存型的多核并行计算机上实现了暂态稳定的并行计算。结果显示,这种并行化技术有效地提高了运行效率。
2、针对现代CPU和GPU多核心的特点,对电力系统暂态稳定分析及其并行算法进行了研究。结合电力网络矩阵稀疏化的特征,与计算统一设备架构(CUDA: Compute Unified Device Architecture)技术,提出了基于CPU网络矩阵划分及运算调度和GPU细粒度并行的电力系统暂态稳定分析的异构并行算法,并对稀疏矩阵的处理进行了优化。利用3000节点和1256节点案例,对该算法进行了对比仿真和测试。仿真结果表明异构并行算法可以获得较高的加速比和很好的并行计算效率。另外针对多GPU系统提出基于OpenMP + CUDA的混合并行算法,拓展了异构并行算法的应用范围。
3、针对电网的实际特点,开发了基于网格计算的安全评估体系。系统的设计本着低成本、可靠性、适应性的原则,利用虚拟化的环境构架为其在电力系统中的应用提供了性能优化和竞争优势仿真结果显示:基于网格计算的安全评估系统,能够充分利用现有资源,使批量计算任务达到线性加速比,计算结果准确可靠,为电网运行方式人员提供强有力的工具。
通过不同规模系统的算例分析,证明了本文所提网格与并行仿真算法的正确性和有效性。
As the nerve center of interconnection grid, smart security assessment is a unified key element of a strong power grid, which is the basis for the maintenance of power production process and the way to protect power grid operation and development. With the parallel computing technology and grid computing technology in China’s power system widely used, making use of these information for intelligent security assessment system research and design work possible. The main results of this thesis are shown below.
1. In modern multi-core computers, data exchange between different cores can be realized more easily with reliable high bandwidth and low-latency communication due to the architecture of shared memory and cache, thus forming a native parallel computing environment. Since the transient stability analysis of power systems involves much computing time, this paper focuses on the parallel algorithm and design for transient stability analysis and presents an improved algorithm suitable for implementation in the multi-core computer. To maximize the efficiency of parallel computing, OpenMP is adopted to transform most of the computing work into parallel computing. The results show that, compared to traditional parallel technology, parallel algorithm based on multi-core processor achieved apparent improvements in computation.
2. For modern multi-core CPU and GPU features, a research is made on power system transient stability analysis and its parallel algorithm. Considering the sparse feature of power network grid and Compute Unified Device Architecture technology, a method of power system transient stability heterogeneous parallel algorithm based on CPU network grid partition and calculation dispatch, together with GPU fine-grained parallelism is proposed. Make special optimization for sparse matrix dispose. For test, 3000-bus and 1256-bus power system is used, and through numerical simulation the proposed algorithm has been compared with the other transient stability analysis program. The proposed algorithm has good convergence rate, and high speedup ratio and parallel calculation efficiency can be obtained. A mixed parallel method based on OpenMP and CUDA for multi-GPU system is also proposed, which helps to expand the applied range of heterogeneous parallel algorithm.
3. For the actual characteristics of Shanghai power grid, a security assessment system based on grid computing is developed. Except the feature of low-cost, reliability and adaptability, the use of virtual environment framework provides a performance optimization and competitive advantage. The integration of Globus and SHPG makes the grid power system technical architecture more flexible and reliable. The design of proxy service models, geographical templates and scalable grid group so that the scalability and adaptability of power grid more powerful. The simulation results of Shanghai power grid show that: based on grid computing for dynamic security assessment system can make full use of existing resources to achieve linear speedup and the calculation results are accurate and reliable, which provides a powerful tool.
The simulation results obtained from different tests validate the advantages of the proposed methods.
引文
[1]石恒初.基于PC机群的电力系统暂态稳定评估[J].电力系统保护与控制,2009,37(10):5-10.
[2]王锡凡.现代电力系统分析[M].北京:科学出版社,2003,03.
[3]夏俊峰,杨帆,李静,郑秀玉.基于GPU的电力系统并行潮流计算的实现[J].电力系统保护与控制.2010,38(18):100-103.
[4]汪芳宗,何一帆.基于多级高阶辛Runge-Kutta方法的暂态稳定性并行计算方法[J].电力系统保护与控制,2011,39(11):22-26.
[5] Calvin Lin, Lawrence Snyder. Principles of Parallel Programming[M]. Beijing: China Machine Press, 2009.
[6]李传栋,房大中,杨金刚,袁世强,鄂志君.大规模电网并行潮流算法[J].电网技术,2008,32(7):34-39.
[7]张步涵,王凯,方华亮,毛承雄.基于网络分割的电力系统潮流分解协调计算[J].高电压技术,2007,33(7) :173-176.
[8]左墨.电力系统暂态稳定及潮流计算的并行算法研究[D] .陕西:陕西科技大学,2009,06.
[9]苏新民,毛承雄,陆继明.对角块加边模型的并行潮流计算[J].电网技术,2002,26(1):22-25.
[10] H.H哈普.分块法及其在电力系统中的应用[M].北京:科学出版社,1987,03.
[11]杨建林.电力系统暂态稳定空间并行仿真[D].天津:天津大学,2006,01.
[12] K.W.Chan. Efficient heuristic partitioning algorithm for parallel processing of large power systems network equations[J]. IEEE Proceedings-Generation Transmission and Distribution, 1995, 142(6): 625-630.
[13]洪潮,沈俊明.电力系统暂态稳定计算的一种空间并行算法[J].电网技术,2000,24(5):20-24.
[14] Univ.de.Sevilla. An Effective Ordering Algorithm to Improve Sparse Vector Methods[J]. IEEE Transactions on Power Systems, 1988, 3(4): 1538-1544.
[15]韩志伟,刘志刚,鲁晓帆,周登登.基于CUDA的高速并行小波算法及其在电力系统谐波分析中的应用[J].电力自动化设备,2010,30(1):98-101.
[16]许彦芹,陈庆奎.基于SMP集群的MPI+CUDA模型的研究与实现[J].计算机工程与设计,2010,31(15):3408-3412.
[17]张韵.基于CUDA的并行空间计算[J].测绘科学,2010,35(6):26-28.
[18]董荦,葛万成,陈康力.CUDA并行计算的应用研究[J].信息技术,2010,4:11-15.
[19]周伟明.多核计算与程序设计[M].武汉:华中科技大学出版社.2009.03
[20]薛巍等.电力系统暂态稳定仿真并行算法的研究进展[J].清华大学学报.2002,42(9): P1192-1195
[21]卢锦玲等.PC机集群系统在电力系统暂态稳定分析中的应用[J].电力自动化设备.2006, 26(4):P36-38
[22]胡晓力等.多粒度并行计算集群研究与应用[J].电力学报.2007,22(4):P436-438
[23]赵文恺等.电力系统并行计算的嵌套分块对角加边形式划分算法[J].中国电机工程学报.2010,30(25):P66-73
[24] W.F.Tinney. Sparse vector methods[J].IEEE Transaction on Power System.1985,PAS 104,Issue 2:P295‐301
[25]柳洋.基于网格的电力系统并行计算的研究[D].北京:华北电力大学.2008.12
[26] Electric Power Research Institute. Utility Software Operation on Parallel Computers. Project 3103‐06 Final Report.1996.01
[27] 奎因等.MPI与OpenMP并行程序设计[M].北京:清华大学出版社.2004
[28] M.Montagna, G.p.Granelli, et al. Levelwise algorithms for vector processing of sparse power system matrices.IEEE Trans. on PWRS,1996,11(1):239‐245
[29] Fernando L.Alvarado, David C.Yu, Ramon Betancourt. Partitioned sparse A‐1 methods IEEE Trans. on PWRS,1990,5(2):452‐459
[30] A.Padilha, A.Morelato. A W‐matrix methodology ofr solving sparse network equations on multiprocessor computers.IEEE Trans. On PWRS,1992,7(3):1023‐1029
[31] G.P.Granelli, M.Montaagna, G.L.Pasini, et al..A W‐matrix based fast decoupled load flow for contingency studies on vector computers. IEEE Trans. on PWRS, 1993, 8(3):946‐953
[32] H.S.Huang,C.N.Lu. Efficient storage scheme and algorithms for W‐matrix vector multiplication of vector computers.IEEE Tran.on PWRS,1994,9(2):1083‐1091
[33] Jun Qiang Wu, Anjan Bose. A new successive relaxation scheme for the W‐matrix solution method on a shared memory parallel computer.IEEE Trans. On PWRS, 1996,11(1):233‐238
[34] 汪芳宗,电力系统并行计算,北京:中国电力出版社,1998
[35] 贺任睦,周庆捷,郝玉国,电力系统机-网暂态仿真的并行算法,中国电机工程学报,1995,15(3):179‐184
[36] Scala M.L, Sblendorio G, Sbrizzai R.Parallel‐in‐time implemention of transient stability simulations on a transputer network. IEEE Trans on PWRS, 1994,9(2): 1117‐1125
[37] 毛承雄,吴增华,电力系统并行仿真计算的一种新方法,华东电力,1999,3(1):21‐23
[38] M.L.Scala, MicheleBrucoli, FrancescoTorelli, et al. A Gauss‐Jacobi‐Block –Newton method for parallel transient stability analysis. IEEE Trans. on PWRS, 1990,5(4): 1168‐1177
[39] M.L.Scala, Roberto Sbrizzai. A Pipelined‐in‐time parallel algorithm for transient stability analysis. IEEE Trans.on PWRS,1991,6(2):715‐722
[40] Peter Saviz,Omar Wing.Circuit simulation by hierarchical waveform relaxation. IEEE Trans. on CAD ofintegrated Circuits and Systems,1993,12(6): 845‐860
[41] George Diedrich Gristede, Albert E.Ruehli,Charles Albert Zukowski. Convergence properties of waveform relaxation circuit simulation methods.IEEE Trans. on CAS,1998,45(7):726‐738
[42] M.Ilic‐Spong,M.L.Crow, M.A.Pai.Transient stability simulation by waveform relaxation methods.IEEE Trans. on PWRS,1987,2(4):943‐952
[43] Scala M.L,Bose A J,et al. A highly parallel method for transient stability analysis.IEEE Trans on PWRS,1990,5(4):1439‐1446
[44] 姚恩瑜,何勇,陈仕平,数学规划与组合优化,杭州:浙江大学出版社 2001.9
[45] Ning Zhu, Anjan Bose. A dynamic partitioning scheme for parallel transient stability analysie. IEEE Trans. on PWRS,1992,7(2):940‐946
[46] A.Kurita, H.Okubo, et al. Multiple time‐scale power system dynamic simulation. IEEE Trans.on PWRS,1993,8(1):216‐223
[47] M.L.Crow,J.G.Chen. The multirate simulation of FACTS devices in power system dynamics.IEEE Trans. on PWRS,1996,11(1):376‐382
[48] M.H.M.Vale, D.M.Falcao, E.Kaszkurewicz. Electrical power network decomposition for parallel computations.Proceeding of the IEEE Symposium on circuits and systems. San Diego,CA,May 1992.2761‐2764
[49] Banerjee.P.,Jones.M.H., Sargent J.S.. Parallel simulated annealing algorithms for cell placement on hypercube multiprocessors.IEEE Trans,.1990,PDS‐1:91‐106
[50] M.R.Irving, Optimal network tearing using simulated annealing. IEE Proceeding., 1990,137(1):69‐72
[51] 舒继武,薛巍,郑炜民,一种电力系统暂态稳定并行计算的优化分区策略,电力系统自动化,2003,27(19),6‐10
[52] Kirk Schloegel.George Karypis. Vipim Kumar.Graph partitioning for high performances scientific simulations.CRPC Parallel computing handbook.Morgan Kaufmann,2000
[53] Bruce Hendrickson, Robert Leland. An improved spectral graph partitioning algorithm for mapping parallel computations. SIAM J.Sci. Compute., 1995,16(2): 452‐469
[54] H.Simon, S.Teng. How good is recursive bisection? SIAM J.Scientific computing,1997,18(5):1445‐1463
[55] 薛巍,全国联网巨系统的暂态稳定并行计算研究:[博士学位论文].北京:清华大学,2003
[56] Tinney W F,Brandwajn V,Chan S M,Sparse vector methods.IEEE Trans on power apparatus and systems,Feb 1985,PAS‐104(2):295‐301
[57] 张伯明,陈寿孙,高等电力网络分析,北京:清华大学出版社,1996(8)