基于细胞神经网络方法的重力异常分离  

Gravity anomaly separation based on cellular neural network

作　　者：刘展[1] 刘茂诚[1] 魏巍[2] 杜润林[1] 

机构地区：[1]中国石油大学地球资源与信息学院,山东青岛266555 [2]胜利油田物探研究院,山东东营257022

出　　处：《中国石油大学学报（自然科学版）》2010年第4期57-61,66,共6页Journal of China University of Petroleum(Edition of Natural Science)

基　　金：国家油气重大专项课题(2008ZX05020-006);国家'863'重点项目(2009AA062802)

摘　　要：用细胞神经网络(CNN)方法对重力异常进行分离,讨论如何根据目标局部重力异常特征训练模板。采用拟BP学习算法实时训练神经网络模板,为保证全局误差函数最小,使用梯度下降法,用全局误差函数对权值变量求导的方法推导权值修正公式。通过模型的不同组合方式来模拟各种地质条件,并总结出细胞神经网络方法的适用条件。应用结果表明,用细胞神经网络方法提取目标异常是切实可行的,只要选择合适的模板,突出目标异常,就能将水平(横向)叠加异常区分开,具有较强的横向分辨能力,特别是可以突出局部异常的边界,据此可用于圈定火成岩体,寻找局部矿藏。The method of separating gravity anomalies using the cellular neural network and how to train the templates based on target local gravity anomalies were discussed. A pseduo-BP algorithm was used to train the neural network templates. To ensure the minimum of global error function, weight revision formula was deduced by using the gradient descent algorithm and the derivation of global error function by weight. Molding all kinds of geological conditions, the applicable conditions of cellular neural network method were summarized. The results show that using the cellular neural network method to extract target anomalies is feasible. As long as suitable templates are chosen and target anomaly is prominented, lateral superimposed a- nomalies can be separated, which shows that the method has strong lateral resolution, and the boundary of local anomaly can be prominented. Based on this, igneous body can be delimited and local mineral resources can be discovered.

关 键 词：地球物理勘探 重力异常 分离 细胞神经网络 模板 拟BP学习算法 

分 类 号：P312.9[天文地球—固体地球物理学]