重磁反演的拟人工神经网络BP算法及其应用
来源期刊：中南大学学报(自然科学版)1994年第3期
论文作者：朱自强 黄国祥
文章页码：288 - 293
关键词：人工神经网络; BP算法; 反演; 重磁场
Key words：artificial neural networks; BP algorithm; inversion; gravity and magnetic field
摘 要：将神经网络反推算法(BP算法)引入重、磁场反演,建立了基于神经网络原理描述反演过程的模型,并在通用数字计算机上实现了一种拟BP算法,实践表明,拟BP算法作重磁资料反演,具有快速,大维数,容错能力强和便于充分利用已知信息等特点,用BP算法对湘南地区作重磁参数反演,给出了本区的磁性,密度三维分布特征,进而对本区地壳结构和地质构造得出了新的认识,取得较好效果,神经网络算法的引入,可能带来地球物理场反演方法的革新.
Abstract: The Back Propagation algorithm(BP algorithm)of artificial neural networks was intro-duced in the inversion of gravity and magnetic data,and at the same time the models of inver-sion process based on the principle of BP algorithm was established.We design an inversionprogram which can be executed on the Neumann computer,and by using a fast forward calcu-lating method in this program,the inversion speed is much quicker than traditional method.As a result,using this inversion algorithm the interpretation of gravity and magnetic data canbe done more precisely,more quickly and more flexibly.We used the proposed method insouthern Hunan area,and the 3-D physical properties(Densities and magnetizabilities)distri-butions in this area were calculated from the 1 ：200000 gravity and magnetic data,The map-ping depth is 10 km,3-Dunits size is 4 km×2 km×0.1 km.1200 observation station datawere fitted,whose fitting accuracy is 8.0×10-6m/s2 for gravity and 3.5nT for magnetic.With the geology interpretation of the 3-D gravity and magnetic inversion,the Predevonian u-plifts and concealed rock bodies were successfully delineated,different types granites were i-dentified,and a new rock zone with west-east trend was confirmed in 1992.
