【作者】 李新硕；

【导师】 黄卡玛；

【作者基本信息】 四川大学 ， 无线电物理， 2004， 硕士

【摘要】 本文针对复杂结构高功率微波天线的优化设计内存要求高、计算耗时长的缺点，提出了一套改进的并行优化算法，首次将并行FDTD与并行遗传算法相结合，并应用于实际天线设计，实现了更高的计算效率。 与传统“串行FDTD+并行GA”优化算法模式不同的是，本文采用了改进的“三维FDTD的一维并行算法”与“非阻塞式主从式小种群并行遗传算法”相结合的方式，对天线结构进行优化。通过两种高性能并行算法的结合，大幅度提高了天线优化设计的效率，使原本不可能采用普通算法实现的复杂天线的优化得以实施。 在三维FDTD的一维并行实现中，本文灵活运用MPI系列通信调用，对三维FDTD的一维并行程序进行了系列优化：确立了PML层对FDTD并行效率影响因子的计算方式，及PML负载平衡的网格分割公式；提出了改进的并行子节点间的数据传递方式——奇偶并行节点数据传递法；首次将MPI中的重复非阻塞通信方式引入到并行FDTD算法中，从多方面提高了三维FDTD的一维并行算法的通信效率。该算法的并行效率较传统的并行方法提高了大约6％。 在GA优化算法中，实现了非阻塞式、主从式、小种群遗传算法的并行，编写了一套可以动态分配计算任务的并行遗传算法程序，大大提高了整体遗传算法的搜索效率。 本套算法程序是在由16台微机组成的“元谋二号”Beowulf集群计算机系统上采用MPI并行开发环境实现的。采用该并行优化算法，得到了一种由同轴线馈电、工作在2.45GHz的、端射的新型高功率微波天线——同轴外罩圆锥喇叭天线，具有较理想的辐射特性。数值结果表明该并行优化算法的实效性。 该并行优化算法对复杂电磁结构的优化设计具有重要的现实意义。更多还原

【Abstract】 In this paper, parallel finite-difference time-domain (FDTD) method combined with parallel genetic algorithm (GA) is investigated and its application in the optimization of complicated high-power microwave antenna is discussed. The requirements for the memory and computation time in a sophisticated microwave antenna design are dramatically decreased by using that method.Compared to traditional optimization algorithm (serial FDTD and parallel GA), modified One-dimensional Parallel Algorithm of Three-dimensional FDTD Method combined with Non-blocking Master-slave Parallel Micro Genetic Algorithm increases the computation efficiency so greatly that some sophisticated antenna can not be optimized using conventional method can be solved now.Various MPI functions and novel communication methods are used to improve One-dimensional Parallel Algorithm of Three-dimensional FDTD Method: Meanwhile, the impact factor for the PML cells to increase the parallel efficiency, as well as the formula of cells division for computing load balance, is established. And Odd-even Sub-node Communication Mode is studied. In addition, Persistent Communication Request is used to reduce the overhead for communication between the processes. Final results show that the computing efficiency of the improved parallel algorithm is increased about 6%.Non-blocking Master-slave Parallel Micro Genetic Algorithm with dynamic task scheduling is presented. Therefore, the search efficiency is dramatically increased using that optimization algorithm combined with parallel FDTD.The whole set of programs is built on the hardware platform of a cluster of computers system, Beowulf system named YUANMOU II, which is composed of Connection Machine 16. Message Passing Interface library is used as software platform.By utilizing this set of optimization program, a novel boresight-peak high power microwave antenna, Coaxial Radome Conical Horn Antenna, working at 2.45GHz, is designed. The antenna has relative good operating characteristics. Numerical results demonstrate the effectiveness of the scheme.This parallel optimization algorithm has great potential in the optimization design of sophisticated electromagnetic structures.更多还原