【作者】 张硕；

【导师】 姚蜀军；

【作者基本信息】 华北电力大学(北京) ， 电力电子与电力传动， 2018， 硕士

【摘要】 随着电网的不断发展,电力系统规模不断扩大复杂度逐渐提高,因而电磁暂态仿真的计算量也越来越大,基于CPU的传统电磁暂态仿真算法及模型已无法再满足当前的迫切需求。另一方面,近年来图形处理器(Graphics Processing Unit,GPU)迅速发展,已在大数据及神经网络等方面崭露头角,但其在电力系统中的应用较少。因此,研究如何利用GPU强大的计算能力来加速数值计算,从而提高电磁暂态仿真计算效率势在必行。本文以加速电磁暂态仿真计算为目的,分别从GPU并行加速和模型优化两个方面进行研究,设计了基于GPU的并行电磁暂态改进算法,并取得了一定的加速效果。本文开展的主要工作如下:首先,本文研究了旋转降频的基本原理,提出了一种基于三角变换的解析信号构造方法。在此基础上推导了单相无损、单相有损和三相输电线路的改进模型,并编程实现了输电线路分网大步长仿真,通过和传统模型计算结果对比分析,验证了模型的有效性。然后,本文研究了 GPU的架构和CUDA平台的编程模型。通过与CPU进行对比分析,本文对GPU并行计算的优点、不足以及CUDA内核函数的编程方法和特点做出了详细地研究。其次,本文研究了稀疏线性方程组求解的经典方法,分析各个方法的优劣以及在GPU上实现并行的可行性。在此基础上,从LU分解和前代回代求解两个方面进行了并行优化处理,并通过六组典型算例对比分析,验证了该算法的准确性性与效率。最后,在改进输电线路模型研究和稀疏线性方程组GPU并行求解研究的基础上,本文利用C++语言在Visual Stdio 2015平台上编写了简单的电磁暂态仿真程序,实现了线性方程组的并行求解和输电线路的大步长仿真算法。更多还原

【Abstract】 With the development of power grid,the scale of system is expanded contin-uously and the complexity is also increased gradually.This makes the calculation of EMT(Electromagnetic Transient)simulation more and more large.The traditional algorithm and model based on CPU can not meet the urgent needs.Recently,the Graphics Processing Unit(GPU)has developed rapidly and has been widely used in big data and neural networks.But it has less application in power system.Therefore,it is imperative to research how to use GPU’s great computing power to accelerate numerical calculation,thus improving the simulation efficiency.In this paper,two aspects,the GPU parallel acceleration and model optimiza-tion,are researched to improve the efficiency of EMT simulation.The main work of this paper is as follows:Firstly,the basic principle of frequency reduction by rotation transformation is resrarched.And a new construction method of analytic signal is presented which is based on triangulation transformation.On this basis,the improved models of single phase lossless,single-phase lossy and three-phase transmission line is derived,and are realized by C++.The validity of the model is verified by comparing with the simulation results.Secondly,the architecture of GPU and the programming model of CUDA platform are studied.By comparing with CPU,the advantages and disadvantages of parallel computation based on GPU and the programming characteristics of CUDA kernel function are reserached in detail.Thirdly,the classical methods of solving sparse linear equations are studied and the implementation feasibility of them is also analyzed.The LU decomposition and the previous and back generation are optimized parallelly.The feasibility and efficiency of this algorithm are verified by six examples.Finally,on the basis of the research about improving model of transmission line and GPU parallel solving,a EMT simulation program is compiled on Visual Studio 2015 platform by C++.The functions which include parallel solving linear equations and large step simulation of transmission line are realized.The correctness of the program is verified by comparing with the extended IEEE14 node system simulation.更多还原