摘要
The density inversion of gravity gradiometry data has attracted considerable attention;however,in large datasets,the multiplicity and low depth resolution as well as efficiency are constrained by time and computer memory requirements.To solve these problems,we improve the reweighting focusing inversion and probability tomography inversion with joint multiple tensors and prior information constraints,and assess the inversion results,computing efficiency,and dataset size.A Message Passing Interface(MPI)-Open Multi-Processing(OpenMP)-Computed Unified Device Architecture(CUDA)multilevel hybrid parallel inversion,named Hybrinv for short,is proposed.Using model and real data from the Vinton Dome,we confirm that Hybrinv can be used to compute the density distribution.For data size of 100×100×20,the hybrid parallel algorithm is fast and based on the run time and scalability we infer that it can be used to process the large-scale data.
利用重力梯度数据进行密度反演一直是重力勘探中的研究热点。然而大规模数据的反演中不仅存在着多解性和纵向分辨率低等不足,而且大量时间与内存的消耗也严重制约了计算效率。针对这些问题,本文利用多分量联合、先验信息约束等手段改进了重加权聚焦反演与相关成像反演;分析了两种方法在反演效果、数值计算效率与规模上的特点,提出一种MPI-OpenMP-CUDA多级混合并行反演方法以综合二者的优势。理论模型与文顿盐丘数据试验证明,改进的反演方法能够有效计算目标体的密度分布;当数据量为100×100×20时,混合并行算法的加速比可达140以上;在性能分析中,提出了一种仅利用并行程序运行时间分析可扩展性的方法,说明混编算法能够处理大规模位场数据。
基金
support by the China Postdoctoral Science Foundation(2017M621151)
Northeastern University Postdoctoral Science Foundation(20180313)
the Fundamental Research Funds for Central Universities(N180104020)
NSFCShandong Joint Fund of the National Natural Science Foundation of China(U1806208)