【作者】 郭延祥；

【导师】 周建军；

【作者基本信息】 清华大学 ， 水利工程， 2010， 博士

【摘要】 复杂问题模拟和计算机多核心的发展趋势要求改变传统数学模型单一串行模型程序的开发方式。本文针对工程领域常见的时间过程模拟模型,基于多进程和网络技术,提出了一种通用的并行组合数学模型开发方式(PMDA)。将复杂过程模拟变成简单模型的进程级并行计算与数据交换。论文主要工作与成果如下:1、分析了数学模型要求和计算机技术发展趋势,得出PMDA是解决数学模型发展瓶颈的关键技术之一的结论。在此基础上提出了PMDA研究的目标、任务和技术路线。2、针对时间过程数学模型提出一种通用的PMDA方法。可从时间步长层面比较方便地组合不同尺度和性质的模型并行计算、模拟一个复杂过程。该方法采用中央控制的框架结构,只需在模型的时间循环内插入一个过程函数即可将模型组件化,采用网络作为数据传输的基本手段,采用“物理量—位置—时间”来描述交互数据,采用分层方式来管理模型间组合信息,采用扩展权重法实现了不同数据之间的转换,设计了“按时间,有节奏”的时间调度方法、显式和半隐式两种耦合方式和空间与时间的平滑处理方法。3、利用C++和Fortran混合编程技术对本文所提的PMDA进行了具体实现,开发了相关软件系统。提供了模型组件化中间件和组件化编程模板、组合控制平台和用于分布式计算的远程模型管理客户端,极大地简化了模型组合时的难度,使PMDA很容易被一般科研人员接受和使用。4、利用本文的PMDA方法和系统对几种典型的组合问题进行了研究,并对模型的组合特性进行了初步分析。初步检验表明,并行组合方式在模型间物理量交互、逻辑控制和分区并行计算方面具有广泛的应用前景。对三峡水库城陵矶补偿调度研究表明三峡水库对长江中游来洪有很大的调蓄作用。5、提出了并行组合数学模型标准(纲要),从子模型开发和PMDA系统开发两个方面指出了PMDA开发中应该注意的问题,为今后提出正式标准奠定了基础。PMDA充分利用现代计算机多计算核心技术和网络化技术趋势避免数学模型向无限复杂化的方向发展,为多计算核心环境下的复杂大系统的数学模型开发探寻更简单、有效和合理的途径。更多还原

【Abstract】 The numerical simulation for complicated problems and advances of computers in recent years made in multi-core and network demand changing traditional developing approach of numerical models that resolve all problems in one single program with serial computing. Based on multi-processing and network, a parallelized models development approach (PMDA) is proposed, specially being applied for the time dependent models, to change the simulation of complicated processes into parallel computing of many simple models and exchanging data with each other. The main work and achievements in the thesis are as follows:1. The analysis of the demands of models and technical trend of computer result in the conclusion that PMDA is one of key issue to simulate complex problems evolved with different dimensions, scales and multi-physic processes under the multi-core hardware environment. Then, the aim, task and technique routing of PMDA are presented.2. A way of PMDA aiming at time dependent models is proposed to simulate a complicated process easily by assembling models with different dimensions and characters from each other work together as one model and parallel computing in time step level. Under the framework architecture with a central controller, data can be indirectly exchanged between involved models only by inserting a probe function into each model source code at main time loop. Network and TCP/IP protocol is basic technique to transmit data between models and central controller. Exchanged data is represented in the form of“quantity-location-time”. Assembling information is organized in four level-model, quantity, grid and data transition. A weight and an adder matrix are introduced to change the data from source model into that the target model need. A schedule mode named as“by time, with rhythm”, the explicit and half implicit time coupling approach, and spatial and temporal polishing method by average are presented to coordinate the parallel running of models.3. The PMDA is implemented by mixed programming of C++ and Fortran. The correlated software including the middleware and code template used to change a model to model-component, assembling and control platform and client to manage remote models are developed to decrease the difficulty of developing parallel assembling model so that PMAD may be used easily and admitted by extensive researchers.4. Some kinks of assembling cases are studied using PMDA and software built in this paper. What happened when models are assembled is brought to light and preliminarily analyzed. Simple examples indicate extensive application of PMDA in data exchanging and logical controlling between models. The study of Three Gorge Reservoir (TGR) compensative operation according flood at Chenglingji makes clear that TGR has an important effect on the flood from Mid Reach of Yangtze River.5. What should be paid attention to when developing sub-models and PMDA system aspect is listed in the standard of PMDA (outline). It can be the base for the formal standard set in the future.In total, PMDA can be used to make the best of the capability of multi-core computer and network and avoid of developing ultra-complex models. More simple, efficient and reasonable simulation route for complicated problems in the multi-core computing environment is explored out by using PMDA.更多还原