利用Kmeans-RBF神经网络提高重力格网化数据精度  

Using Kmeans-RBF neural netw ork to improve the accuracy of grid gravity data

作　　者：黄子炎 王庆宾[1] 李锁柱 李亮 冯进凯 HUANG Ziyan;WANG Qingbin;LI Suozhu;LI Liang;FENG Jinkai(Institute of Geospatial information,Strategic Support Force Information Engineering University,Zhengzhou 450001,China;61243 Troops,Urumqi 830002,China)

机构地区：[1]战略支援部队信息工程大学地理空间信息学院,河南郑州450001 [2]61243部队,新疆乌鲁木齐830002

出　　处：《海洋测绘》2021年第4期43-47,共5页Hydrographic Surveying and Charting

基　　金：国家自然科学基金(41774018);地理信息工程国家重点实验室开放基金(SKLGIE2019-M-1-3);信息工程大学科研团队发展基金(f5206)。

摘　　要：针对重力测量数据在格网化过程中精度会被降低的问题,顾及空间重力异常和地形的强相关性,提出了三维Kmeans-RBF神经网络方法,该方法利用神经网络的复杂非线性映射学习能力进行推估建模,并在模型训练和推估时加入地形数据作为物理控制。最后基于美国爱达荷州地区的实测重力数据进行验证,实验结果表明:该方法相对于二维Kmeans-RBF神经网络方法和直接进行拟合推估的Kriging方法,实验区内精度分别提高了24.85%和44.84%。Aiming at the problem that the accuracy of gravity data will be reduced in the process of gridding,considering the strong correlation between spatial gravity anomaly and terrain,this paper proposes a three-dimensional kmeans RBF neural netw ork method,w hich uses the ability of neural netw ork to learn complex nonlinear mapping for modeling,and adds terrain data as physical control in the process of training model and estimation.Finally,the measured gravity data in Idaho were used to verify the experimental results.Compared with the tw o-dimensional KM eans RBF neural netw ork method and the Kriging method,the accuracy of the proposed method in the experimental area is improved by 24.85%and 44.84%,respectively.

关 键 词：重力异常 格网化 径向基函数神经网络 Kmeans聚类 深度学习 

分 类 号：P223[天文地球—大地测量学与测量工程]