基于能量守恒方法的重力场反演快速异构并行算法

在利用卫星跟踪卫星资料解算重力场模型位系数时，其海量观测数据处理以及大型方程组解算过程存在计算效率低下、对平台硬件要求高等问题。针对以上问题，提出一种基于能量守恒方法的重力场反演快速异构并行算法，基于CUDA 在GPU端实现并行计算设计矩阵，结合MKL库与分区平差法、预处理共轭梯度法在CPU端完成低内存消耗下的法方程快速构建与求解，实现重力场模型反演的异构并行计算。运用该算法处理GRACE-FO卫星2020-01-01～06-30期间观测数据，反演获得120阶重力场模型GM-GraceFO2020h；与现有模型以及算法对比分析表明，该算法所得模型与现有GRACE重力场模型精度相当，且相较于传统的串行算法，反演耗时减少98.479%，内存消耗减小1个数量级。

Heterogeneous Parallel Fast Gravity Recovery Algorithm Based on Energy Conservation Method

In the process of gravity recovery using satellite-satellite tracking data, there are some problems caused by processing massive observation data and solving large equations, such as low computational efficiency and high requirements for hardware level of computing platform. In view of the above problems, we propose a fast heterogeneous parallel algorithm for gravity recovery using energy conservation method. CUDA is used to realize parallel computation of design matrix on GPU, MKL library is used in partition adjustment method and preconditioned conjugate gradient method for constructing and solving normal equation on CPU fast, thus realizing heterogeneous parallel computation of gravity recovery using satellite data. A 120×120 gravity field model, named GM-GraceFO2020h, is obtained by processing the observation data of GRACE-FO satellite from January 1, 2020 to June 30, 2020 using this algorithm. Compared with the existing models and algorithms, the result shows that the accuracy of the model derived by the proposed algorithm is equivalent to that of existing GRACE gravity field models, and compared with the traditional serial algorithm, the inversion time is reduced by 98.479%, and the memory consumption is an order of magnitude smaller.