【作者】 张玉；

【导师】 梁昌洪；

【作者基本信息】 西安电子科技大学 ， 电磁场与微波技术， 2004， 博士

【摘要】 开发独立自主知识产权的并行电磁计算软件，对典型的电磁场数值算法进行系列的并行化研究，无论是从解决工程问题出发，还是就提高复杂大型电磁计算能力的战略目标而言，都迫切需要。因此对典型数值方法的新技术以及其并行计算方法进行研究具有现实的、深远的意义。 MPI已经取代其它信息传递系统成为目前并行计算中最广泛采用的通信标准。本文研究了几种具有代表意义的数值算法(几何绕射理论、时域有限差分法、矩量法)的关键技术及其基于MPI编程环境的并行计算实现，有效的解决了机载天线特性计算时间过长、波导宽边缝隙天线阵局部特性难以准确建模、整体阵列分析慢、复杂PBG结构电磁特性难以仿真等几个典型的工程问题。 本文首先提出并应用了按照计算区域进行任务分解的并行UTD方法，缩短了机载平台天线特性的分析时间，解决了耗时太长的问题。这为开发小型机系统中大型EMC软件奠定了坚实的基础。 本文接着研究了FDTD的几种新技术在波导缝隙阵以及PBG结构中的应用。为此，本文构造了一种适合于分析波导圆头缝隙特性的共形网格FDTD方案，并应用该方案分析了倾斜组合裂缝、耦合缝隙结构的特性；在此基础上，实现了并行共形网格FDTD方法，并将这部分研究工作扩展到微波光子晶体结构的电磁特性仿真，为这类问题提供了一个有效的仿真手段。一个千万级网格量的FDTD算例，表明本文并行共形网格FDTD程序的有效性与健壮性。本文还结合一种非均匀网格FDTD方法，对非均匀网格共形FDTD算法做了试探性研究。 为了提高矩量法的计算效率，本文深入地研究了求解矩阵方程的预条件CGN方法，提出了一种以基函数邻居为基础的预条件方法——BFN预条件方法。把该预条件方法应用于矩量法分析大型波导缝隙阵时矩阵方程的求解，将收敛速度提高一个数量级，这一结果为数值优化设计大型裂缝阵带来了巨大的益处。此外，还将这种具有物理含义的预条件方法应用于其它几种典型基函数的矩量法问题，数值结果均表明BFN预条件方法稳定、高效。为扩展矩量法计算规模，论文还研究了并行矩量法的矩阵填充与求解方法，并以某飞行器散射特性分析为例进行了计算。 最后，论文概括的指出了尚未展开的相关研究工作的实现途径。更多还原

【Abstract】 The developing of complicated software on Computational Electromagnetics with parallel techniques and the studying of typical parallelization algorithms are of both technical and strategical significance in increasing the domestic capacity for analyzing EM characteristics for targets in complicated EM environments.For technical computing, MPI has displaced most other message-passing systems. It is a library specification for message-passing, proposed as a standard by a broadly based committee of vendors, implementors, and users. The work concerning parallelizatibn done in this thesis is practically based upon MPI.In this thesis, key techniques and parallelization of several widely used numerical computational electromagnetic methods are studied. Three problems known as typical in engineering are: 1) too much time is needed for analyzing the radiating pattern of airborne antennas; 2) curved boundary structures in slot array can’t be easily modeled; 3) complicated EBG structures are difficulty to simulate. These problems are efficiently resolved to some degree in this thesis.Firstly, the parallel UTD algorithm is presented based on the Domain Decomposition Method. Numerical results show that parallel UTD algorithm can drastically reduce the computing time. The parallel UTD method provides a solid foundation for producing complicated EMC software on small computer systems.Then, a modified locally conformal FDTD (MLC-FDTD) scheme suitable for analyzing bulletheaded slots is presented. The MLC-FDTD is also parallelized for analyzing EBG structures. The Parallelized Conformal FDTD is continuously run on a PC cluster for several days to solve a Thousand-Million-Grid FDTD problem. This manifests the efficiency, robustness and accuracy of the Parallel MLC-FDTD. Furthermore, the MLC-FDTD is studied to save computational resources by combining it with a type of non-uniform FDTD scheme, which we refer to as NUMLC-FDTD.Finally, a Basis Function Neighbor (BFN) preconditioning method is presented to solve the equation system obtained from the application of the moment methods. It is a physically based general-purposed preconditioning method for any type of basis function. Numerical results show that the proposed preconditioner can approximately improve the convergence of an iterative solution by several times and in some cases, even an order of magnitude. To enable MoM to analyze electrical large targets, Parallel MoM is studied at the end of the thesis.To sum Up, the systematical study in this thesis of the three widely used computational electromagnetic methods, namely UTD, FDTD, and MoM, are reviewed. The schemes for parallelizing the Hybrid numerical method concerning the three methods are also presented.更多还原