并行计算在环境工程中的应用研究

发布时间：2018-03-19 08:17

本文选题：并行计算　切入点：预处理共轭梯度法　出处：《中国地质大学(北京)》2017年硕士论文　论文类型：学位论文

【摘要】：从古至今,地下水资源都是人类最珍贵的财富,它是我们人类生活用水和饮用水的主要来源,然而近些年来,由于社会的不断进步,工业化的迅猛发展,人类对水资源开始过度开采,加之不能合理的规划和利用,近些年来水污染的现象屡见不鲜,频繁发生,使本来就不充裕的地下水资源变得越来越稀缺。我们必须及时采取相应措施,合理利用地下水资源。地下水流动数值模拟是实现对某一区域水资源评价和监测的有效手段,从而使地下水资源真正得到周期性的补充和恢复。随着对地下水研究的不断深入,复杂性持续增加,研究规模的日益扩大,普通的串行算法在计算时间上无法满足研究人员的需求,加之庞大的数据量极大的消耗计算机内存。并行处理技术的提出很好的解决了这个问题。主要研究工作是用并行处理技术优化地下水流动数值模拟方法。为了高效的求解在地下水模拟中出现的上百万个甚至更多的偏微分方程组,我们结合当今数值领域先进的求解方程组的理论知识,用并行计算方式,解决地下水仿真模拟的现有问题,从而极大的缩短数值模拟的时间及内存消耗问题。以modflow原始模型为基础,以求解大规模线性方程组为主要研究内容,以尽可能缩短运行时间为主要目的,着重介绍主流的并行结构体系和算法思想。主要工作内容包括:1)研究modflow关于三维有限差分公式数学模型的实际意义和理论依据,了解其整体的模型架构以及各个基本子程序包的概念和程序表达,并对其进行程序化编译和运行。2)研究该模型解大型线性方程组所用到的迭代算法,深入研究预处理共轭梯度法和参数估值法的算法思想和原理,并用不同的并行编程模型进行优化实现。3)分别介绍不同的并行编程模型,有:MPI并行、OpenMP并行以及基于GPU的CUDA并行的并行原理,并对前两种不同的并行方法进行性能测试,测试结果可知,两种并行算法均能在一定程度上加快运行速率,且在一定范围内,效率随着规模增加而愈加明显。
[Abstract]:Since ancient times, groundwater resources have been the most precious wealth of mankind. They are the main sources of water and drinking water for human beings. However, in recent years, due to the continuous progress of society, the rapid development of industrialization, Human beings began to over-exploit water resources, coupled with unreasonable planning and utilization. In recent years, the phenomenon of water pollution is common and frequent. We must take appropriate measures in time to make rational use of groundwater resources. Numerical simulation of groundwater flows is an effective means to realize the evaluation and monitoring of water resources in a certain region. So that groundwater resources can be replenished and restored periodically. With the deepening of groundwater research, the complexity and scale of groundwater research are increasing. Ordinary serial algorithms can not meet the needs of researchers in computing time, In addition, the huge amount of data consumes computer memory. The parallel processing technology has solved this problem very well. The main research work is to optimize the numerical simulation method of groundwater flow with parallel processing technology. Effective solutions to millions or more of the partial differential equations that occur in groundwater simulation, Combining with the advanced theoretical knowledge of solving equations in the field of numerical value, we solve the existing problems of groundwater simulation by means of parallel calculation. Therefore, the time and memory consumption of numerical simulation are greatly shortened. Based on the original modflow model, the main research content is to solve large scale linear equations, and the main purpose is to shorten the running time as far as possible. The main work includes: 1) studying the practical significance and theoretical basis of modflow's mathematical model of 3D finite difference formula. Understand its whole model architecture, the concept and program expression of each basic subpackage, and program compile and run. 2) study the iterative algorithm used in solving large linear equations. The idea and principle of preprocessing conjugate gradient method and parameter estimation method are studied in depth, and different parallel programming models are introduced respectively by using different parallel programming models. The parallel principle of CUDA parallelism based on GPU and the parallel principle of CUDA parallel based on GPU are used to test the performance of the first two different parallel methods. The test results show that the two parallel algorithms can speed up the running speed to a certain extent and within a certain range. Efficiency increases with scale.
【学位授予单位】：中国地质大学(北京)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P641.8;TP338.6

【相似文献】