声场时域信号预报并行算法研究

发布时间：2018-05-23 20:52

本文选题：并行计算 + MPI　；参考：《哈尔滨工程大学》2012年硕士论文

【摘要】：随着科学技术的发展，单处理机的处理能力正在逐年提升，但是由于受到技术水平的限制，单处理机的计算能力满足不了许多领域中具有挑战性的大规模计算课题对计算资源的需求，所以除了增加处理器本身的计算性能外，并行计算是一种提高计算能力的有效方法。并行计算技术已经应用到许多领域，在难以依靠修改模型本身来减少计算时间的情况下，可以寻求依靠并行算法来解决问题。本文将并行计算技术应用到现有的声场计算模型中，为解决水下声场计算复杂、计算量大、计算速度慢的问题，实现声场计算实时性的要求做了一定的探索。本文主要进行了以下几方面的研究： 1、研究了并行计算的硬件平台，主要包括现有的并行计算机的体系结构，以及基于CUDA语言的GPU并行平台；研究了并行计算的语言环境，主要包括现有的并行语言，MPI编程环境，以及CUDA语言的编程模型、存储模型等。 2、进行声场预报的基础是声场的计算模型，本文对海洋声场的建模方法进行研究，深入研究了简正波KRAKEN模型，计算原理以及串行的宽带KRAKEN的实现原理。介绍了建立时域上的脉冲传播模型的两种方法，详细介绍了频域解的傅里叶合成方法。 3、详细介绍了消息传递模型标准MPI、CUDA的使用以及安装方法，，搭建了并行计算的运行环境；实现了基于MPI的宽带KRAKEN声场并行计算、基于CUDA CUFFT库的并行快速IFFT算法，最后结合两种并行算法实现了宽带声场时域接收信号的并行预报仿真。 4、通过对不同数据量的测试表明，并行程序可以在保证与串行程序一样的计算精度下有效地缩短计算时间，提高运算效率。
[Abstract]:With the development of science and technology , the processing capacity of a single processor is increasing year by year , but because of the limitation of the technical level , the computing power of the single processor can not meet the need of computing resources in many fields . Therefore , in addition to increasing the computational performance of the processor itself , the parallel computing is an effective method to improve the computing power .

Parallel computing technology has been applied to many fields , and it is difficult to rely on parallel algorithm to solve the problem . In this paper , parallel computing technology is applied to the existing sound field calculation model .

1 . The parallel computing hardware platform is studied , which mainly includes the architecture of the existing parallel computer , and the parallel GPU platform of GPU .
This paper studies the parallel computing language environment , including the existing parallel language , MPI programming environment , and the programming model , storage model and so on .

2 . The sound field prediction is based on the sound field calculation model . In this paper , the modeling method of ocean sound field is studied , and the principle of simple wave KRAKEN model , calculation principle and realization principle of serial broadband KRAKEN are studied . Two methods of establishing pulse propagation model in time domain are introduced , and the Fourier synthesis method of frequency domain solution is introduced in detail .

3 . This paper introduces the standard MPI of message passing model , the usage and installation method of the standard MPI , and sets up the running environment of the parallel computing ;
A parallel fast IFFT algorithm based on MPI - based broadband KRAKEN field parallel computation is realized , and a parallel fast IFFT algorithm based on the GPU FFT library is realized , and the parallel prediction simulation of the wideband sound field time domain received signal is realized in combination with two parallel algorithms .

4 . By testing different data volumes , the parallel program can effectively shorten the computation time and improve the operation efficiency under the same computational accuracy as the serial program .
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP338.6

【参考文献】