基于深度学习的重力梯度事后误差补偿方法  

Post mission error compensation method of gravity gradiometry based on deep learning

作　　者：程一 李桐林[1] 张镕哲 周帅[1] CHENG Yi;LI Tong-lin;ZHANG Rong-zhe;ZHOU Shuai(College of Geo-exploration Science and Technology,Jilin University,Changchun 130026,China)

机构地区：[1]吉林大学地球探测科学与技术学院,长春130026

出　　处：《世界地质》2021年第3期671-679,共9页World Geology

基　　金：国家重点研发计划航空重力梯度仪研制项目(2017YFC0601600);中国博士后科学基金特别资助(站前)项目(2020TQ0114);中国博士后科学基金项目(2020M681036)资助。

摘　　要：旋转式重力梯度仪动态测量的误差补偿是航空重力梯度测量的关键技术。因深度学习直接提取数据特征,有望补偿以往通过建模难以补偿的误差,从而进一步提高误差补偿效果,笔者提出了基于深度学习的航空重力梯度测量事后误差补偿方法,利用航空试验实测数据建立数据集,搭建并训练由梯度仪运动参数映射到梯度仪输出噪声的神经网络,利用该网络预测梯度仪输出噪声,以此对其他实测数据进行误差补偿。结果表明,利用深度学习预测的梯度仪输出噪声与实际输出噪声非常接近,补偿后梯度仪输出噪声下降超过一个数量级,补偿的误差水平与梯度仪样机指标接近。The error compensation of dynamic measurement of rotating accelerometer gravity gradiometer is the key technology of airborne gravity gradiometry.Because of the direct extraction of data features by deep learning,it is expected to compensate the error which is difficult to be compensated by modeling in the past,so as to further improve the effect of error compensation.The authors put forward a post mission error compensation method for airborne gravity gradiometry based on deep learning.The dataset is established by using the measured data of the aerial test,and the neural network is built and trained to map the motion parameters of the gradiometer to the output noise of the gradiometer,and the network is used to predict the output noise of the gradiometer.It is used to compensate the errors of other measured data.The results show that the output noise of the gradiometer predicted by deep learning is very close to the actual output noise.After compensation,the output noise of the gradiometer decreases by more than an order of magnitude,and the error level of compensation is close to the target accuracy of the prototype of the gradiometer.

关 键 词：航空重力梯度测量 旋转式重力梯度仪 动态测量误差 误差补偿 深度学习 

分 类 号：P631.1[天文地球—地质矿产勘探]