Research On Several Problems Of Gravity Field Based On Parallel Computing

With the development of Surveying and mapping observation methods,the amount of surveying and mapping data increases greatly,with which there is also the demand for higher resolution and high precision gravity field in various fields.Therefore,it is an inevitable trend to calculate and analyze the external gravity field in a higher resolution and a wider range.At present,the data processing methods of gravity field related problems still use the conventional serial scheme,which is difficult to adapt to the growth of data volume.Too long calculation time becomes the bottleneck of its development.Based on this,this paper introduces parallel computing technology to parallel the data processing of gravity field related problems,studies and analyzes the applicability and efficiency of homogeneous parallel technology and heterogeneous parallel technology in gravity field problems,and obtains some useful conclusions.The main contents of this paper include.1 In view of the low efficiency of high-order gravity field model to calculate the bulk gravity field elements,an isomorphic parallel computing scheme based on OpenMP is designed,which is suitable for large-scale grid computing and a large number of independent point-to-point computing.The strategy of array dimension increasing and group computing is put forward,which effectively solves the read-write conflict of CPU multithreaded memory.On the basis of this parallel method,the grid calculation of height anomaly and vertical deviation and the whole parallel calculation of disturbing gravity vector of flight path point are realized.The calculation acceleration ratio is more than 5 times.Without changing any hardware,the calculation efficiency is effectively improved.2 In view of the low efficiency of the high-order gravity field model to calculate the single point gravity field element,the heterogeneous parallel computing scheme based on CUDA is designed,and the fast parallel computing of disturbing gravity vector is realized according to this method.Using the geopotential coefficient model to calculate the disturbing gravitational vector can be divided into two parts: the calculation of the associated Legendre function and the calculation of the spherical harmonic series.The former is difficult to be parallelized in the process of single point calculation because of its correlation before and after recursive operation,so it can only be serially calculated in the CPU side,while the spherical harmonic series part can be parallelized in the GPU side.By using the principle of vectorization,the original serial logic formula is transformed into the vector operation formula suitable for parallel logic,and the parallel calculation of single point gravity field element is realized from the formula level.The experimental results show that the acceleration ratio of parallel computation of disturbing gravity vector is 13 times.3 Aiming at the low efficiency of local terrain correction,a heterogeneous parallel scheme based on CUDA is designed to realize the fast calculation of local terrain correction.This paper proposes the scheme of terrain grid recoding and the method of strict prism integral eight component disassembling,which effectively solves the problem of task assignment and overload of each thread calculation in GPU,and makes full use of the computing capacity of the equipment.It takes only 1.5 seconds to calculate the local terrain correction with a resolution of 1 minute and an integral radius of 40 minutes on a computer with a graphics card model of Tesla V100;it takes only 14.6 minutes to calculate the local terrain correction with a resolution of 10 seconds and an integral radius of 40 minutes;it takes 45.7 hours to calculate the terrain correction with a resolution of 3 seconds,while the traditional serial algorithm is difficult to complete the calculation.Under the condition of ensuring the calculation accuracy above the micro gamma level,the maximum acceleration ratio of calculation is over 850 times,which effectively shortens the calculation time and improves the calculation efficiency.4 To solve the problem of low efficiency of equalization compensation correction,a heterogeneous parallel scheme based on CUDA is designed to realize the fast calculation of equalization compensation correction.On the computer with Tesla V100 as the graphics card model,it takes only 1.7 seconds to calculate the local terrain correction with 1 minute resolution and 40 minutes integration radius;it takes only 19.9 minutes to calculate the local terrain correction with 10 seconds resolution and 40 minutes integration radius.Under the condition of ensuring the calculation accuracy above the micro gamma level,the maximum acceleration ratio of calculation is over 730 times,which effectively shortens the calculation time and improves the calculation efficiency.