并行算法在矿体储量估值中的应用

发布时间：2018-05-11 19:24

本文选题：矿体化模型 + CUDA　；参考：《华北水利水电大学》2017年硕士论文

【摘要】：在现代科学研究和工程中,随着问题的规模越来越大,对数据运算速度的要求越来越高。传统的串行算法已经不能满足问题求解的需求,因此研究数据的并行运算已越来越重要。利用计算机将所获得的原始地质资料进行整合,估算资源储量,已成为现代化矿产勘查工作的一项重要步骤。但是在实际工程中,由于矿床及其围岩的空间尺度宏大,延展范围常以千米计,造成矿体数据存储需求与计算效率之间的矛盾问题。近些年来,由于计算机图形处理器的高速发展,人们能够利用GPU进行通用计算,可以较好地提高运算速度。据此,本文采用GPU并行计算技术来解决这一问题。本论文主要是采用并行算法实现体矿化模型的构建,主要研究内容如下:(1)基于矿山复合场理论,分析了体矿化模型构建过程中的问题,并基于其特点论证了并行计算的可能性;(2)针对体矿化建模过程中的关键问题,分析了品位插值技术的串行算法(包括泰森多边形插值算法和距离幂次反比插值算法),设计了两种算法的并行计算流程;(3)采用了CUDA并行计算技术,面向三维空间数据,在GPU上分别实现了泰森多边形插值算法和距离幂次反比插值算法的并行化,并分析了两个算法的时间空间复杂度;(4)通过一个工程实例,将工程里的原始钻孔数据的品位属性作为已知数据,利用两种插值算法计算得到规则的矿体品位属性,为后期的矿体体绘制作准备,并且针对不同规模的随机数据测试其算法的运行时间、误差及加速比,运算结果表明当数据规模比较小的时候,由于内存复制导致访问延时加大,并行算法并没有实现有效的加速,但当数据规模不断增大的时候,加速比不断变大,加速效果明显。本文主要对两种插值算法进行交叉验证并行算法的可行性,加快体矿化构模,对于进一步算法优化和算法自适应采样点均匀与否的情况是下一步要研究的问题。
[Abstract]:In modern scientific research and engineering, with the increasing scale of the problem, the demand for the speed of data operation becomes higher and higher. The traditional serial algorithm can not meet the demand of solving the problem, so it is more and more important to study the parallel operation of data. It has become an important step for modern mineral exploration to integrate the original geological data and estimate the resources and reserves by computer. However, in practical engineering, because of the large spatial scale of ore deposit and its surrounding rock, the extended range is often measured by kilometer, which results in the contradiction between the storage demand of ore body data and the calculation efficiency. In recent years, due to the rapid development of computer graphics processors, people can use GPU to carry out general computing, which can improve the speed of calculation. Therefore, this paper adopts GPU parallel computing technology to solve this problem. In this paper, the parallel algorithm is used to construct the body mineralization model. The main research contents are as follows: (1) based on the mining complex field theory, the problems in the process of building the body mineralization model are analyzed. Based on its characteristics, the possibility of parallel computing is demonstrated. The serial algorithm of grade interpolation (including Tyson polygon interpolation algorithm and distance power inverse proportional interpolation algorithm) is analyzed, and the parallel computing flow of the two algorithms is designed. The parallel computing technology of CUDA is used to face the 3D spatial data. The parallelization of Tyson polygon interpolation algorithm and distance power inverse proportional interpolation algorithm is realized on GPU, and the time and space complexity of the two algorithms are analyzed. The grade attribute of the original borehole data in the project is taken as the known data, and the regular ore body grade attribute is calculated by using two interpolation algorithms, so as to prepare for the later drawing of the orebody. At the same time, the running time, error and speedup ratio of the algorithm are tested for random data of different scales. The results show that when the data scale is small, the memory replication results in the increase of access delay. The parallel algorithm does not achieve effective acceleration, but when the data scale increases, the speedup ratio becomes larger, and the acceleration effect is obvious. In this paper, the feasibility of the parallel algorithm is cross-validated for the two interpolation algorithms. The further optimization of the algorithm and the homogeneity of the adaptive sampling points are the problems to be studied in the next step.
【学位授予单位】：华北水利水电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P624.7;TP338.6

【参考文献】