并行处理技术在地理信息数据处理中的应用

随着地理信息存储量的飞速增长,传统的单进程、集中式的数据处理方式已不能满足基于网络的地理信息服务的效能要求。分析对比了OpenMP,MPI和MapReduce等主流并行编程模式,将关系型数据库与分布式空间数据管理系统相结合,提出了面向并行处理的地理信息存储模型和数据组织模型,将该模型与传统模型进行了对比分析,并基于MapReduce实现了地理空间数据并行处理框架,选取了矢量数据装载、影像数据装载以及数据切片作为典型数据处理案例开展对比实验,该技术方案的处理效率均数倍于传统技术方案。实验表明,该模型能够很好地支持并行处理框架,可为分布式环境下数据处理中心构建提供一个有效解决方案。     

A parallel algorithm for coverage optimization on multi-core architectures

Location siting is an important part of service provision, with much potential to impact operational efficiency, safety, security, system reliability, etc. A class of location models seeks to optimize coverage of demand for service that is continuously distributed across space. Decision-making and planning contexts include police/fire resource allocation for a community, siting cellular towers to support cell phone signal transmission, locating emergency warning sirens to alert the public of severe weather and other related dangers, and many others as well. When facilities can be sited anywhere in continuous space to provide coverage to an entire region, this is a very computationally challenging problem to solve because potential demand for service is everywhere and there are an infinite number of potential facility sites to consider. This article develops a new parallel solution approach for this location coverage optimization problem through an iterative bounding scheme on multi-core architectures. The developed approach is applied to site emergency warning sirens in Dublin, Ohio, and fire stations in Elk Grove, California. Results demonstrate the effectiveness and efficiency of the proposed approach, enabling real-time analysis and planning. This work illustrates that the integration of cyberinfrastructure can significantly improve computational efficiency in solving challenging spatial optimization problems, fitting the themes of this special issue: cyberinfrastructure, GIS, and spatial optimization.     

A new model of edge-to-edge contact for three dimensional discontinuous deformation analysis

A robust contact theory can be regarded as key to three dimensional discontinuous deformation analyses (3D-DDA). Not only must this theory provide an efficient algorithm to judge the type and location of contacts but also it must be able to present comprehensive formulations for every kind of contact (open, sliding and locked contact). There are six types of contact in three dimensional discontinuous deformation analyses (vertex-to-vertex, vertex-to-edge, vertex-to-face, edge-to-edge, edge-to-face and face-to-face) that can be converted to vertex-to-face and edge-to-edge contacts. This paper presents a new model of edge-to-edge contact to three dimensional discontinuous deformation analyses (3D-DDA). This new model considers both kinds of edge-to-edge contact (cross-over and parallel edge-to-edge contact) and presents a criterion for inter-penetration. Sub matrices of normal and shear spring and friction force are derived by geometrical analysis and penalty method. This new model is implemented in a 3D-DDA computer programme, and the example results demonstrate the validity of the model.     

Parallelization of the flow-path network model using a particle-set strategy

High-performance simulation of flow dynamics remains a major challenge in the use of physical-based, fully distributed hydrologic models. Parallel computing has been widely used to overcome efficiency limitation by partitioning a basin into sub-basins and executing calculations among multiple processors. However, existing partition-based parallelization strategies are still hampered by the dependency between inter-connected sub-basins. This study proposed a particle-set strategy to parallelize the flow-path network (FPN) model for achieving higher performance in the simulation of flow dynamics. The FPN model replaced the hydrological calculations on sub-basins with the movements of water packages along the upstream and downstream flow paths. Unlike previous partition-based task decomposition approaches, the proposed particle-set strategy decomposes the computational workload by randomly allocating runoff particles to concurrent computing processors. Simulation experiments of the flow routing process were undertaken to validate the developed particle-set FPN model. The outcomes of hourly outlet discharges were compared with field gauged records, and up to 128 computing processors were tested to explore its speedup capability in parallel computing. The experimental results showed that the proposed framework can achieve similar prediction accuracy and parallel efficiency to that of the Triangulated Irregular Network (TIN)-based Real-Time Integrated Basin Simulator (tRIBS).     

遥感影像并行处理的数据划分及其路径优化算法

研究了一种基于数据划分的遥感影像并行处理的路径优化算法,用于解决将并行技术应用于海量遥感影像分布式存储和处理领域时其处理模型所具有的多路可达性所引起的路径动态、最优选择问题。在栅格数据可分解性分析及并行模型数据态、元素、相对信息量和映射等8个基本定义和6个性质的基础上,给出并行处理一般数学模型。以该模型为基础获得在一般并行处理情况下,以平均计算代价变量的比值作为控制横向并行与纵向并行选择方式的标志,并进一步给出四叉树索引并行生成、基于四叉树的目标检测并行处理等具体示例。最后,通过试验验证了算法的有效性,分析了算法的特点及影响因素。     

网格环境下缓冲区分析的并行计算

提出了基于网格环境下缓冲区分析的并行计算方法,用以提高缓冲区分析的效率。考虑到大多数地理信息系统中缓冲区分析数据的特点,提出了基于图层和地理空间区域的任务分解方法。利用已经构建的网格计算环境,实现了武汉城市地图中一条道路缓冲区分析的图层式并行计算,从实验结果的分析,可以看出网格环境下缓冲区分析的并行计算在效率方面具有明显的优势。     

大型GNSS观测网的并行计算

CGCS2000精化及其全球拓展需要采用最新的模型,处理长期积累的大型GNSS观测网数据,大型GNSS网联合、快速和协同解算是空间基准精化、维持与服务的重要技术方向.在大数据技术背景下,以并行计算、云计算为代表的高性能计算技术逐渐成为大规模数据处理的首选方法.针对海量、多源、异构GNSS数据在解算处理与平差分析等方面面...     

锥束CT图像重建算法的快速实现

本文基于锥束CT滤波反投影重建的FDK算法,通过两种算法改进并结合基于共享内存的OpenMP并行技术和代码优化,实现了锥束CT图像的快速重建.基于锥束CT实际投影数据的重建结果表明,图像重建速度得到了较大的提高,断层图像重建质量与FDK原型算法相当.     

基于GPU加速的Boussinesq类波浪传播变形数值模型

Boussinesq波浪模型是一类相位解析模型,在时域内求解需要较高的空间和时间分辨率以保证计算精度。为提高计算效率,有必要针对该类模型开展并行算法的研究。与传统的中央处理器(CPU)相比,图形处理器(GPU)有大量的运算器,可显著提高计算效率。基于统一计算设备架构CUDA C语言和图形处理器,实现了Boussinesq模型的并行运算。将本模型的计算结果同CPU数值模拟结果和解析解相比较,发现得到的结果基本一致。同时也比较了CPU端与GPU端的计算效率,结果表明,GPU数值模型的计算效率有明显提升,并且伴随数值网格的增多,提升效果更为明显。     

多视密集匹配并行传播GPU-PatchMatch算法

针对多视密集匹配的效率较低的问题,提出了GPU-PatchMatch多视密集匹配算法。该算法使用GPU提高PatchMatch的计算效率;同时充分利用稀疏场景信息,对深度信息进行规则初始化;为提高传播效率,使用了金字塔红黑板并行传播深度信息。最后在DTU、Strecha和Vaihigen数据集上进行了试验,并与常用的多视密集匹配算法进行对比。试验结果表明,本文算法在重建效率上有较大提高,与CPU算法(PMVS、MVE、OpenMVS)相比有7倍以上提升,与GPU算法相比也有2.5倍以上提升,表明本文算法的有效性。