同时反演两个三维密度界面的拟神经网络BP算法

Quasi neural network BP algorithm for simultaneous inversion of two 3-D density interfaces

中南工业大学, 长沙 410083

Central and South University of Technology, Changsha 410083

两个界面或多界面重力异常的显著特点是异常的不可分高性, 即不能简单地根据异常的幅值和水平宽度等数理特点将两个界面或多个界面的异常区分开来。直接用两个三维密度界面的叠加重力异常同时反演两个界面埋深和起伏形态是本文的特色之一。借助于快速正演算法, 对界面进行详细的单元划分, 使反演结果更为精细。由于采用带输出反馈抑制策略的拟神经网络BP算法, 使本反演方法收敛性好, 收敛速度快。拟BP算法借用了神经网络的并行、自适应联想等概念, 使算法便于利用初始控制条件, 从而减少解的非唯一性。理论模型及在某油田勘探区的应用实例说明, 此方法在石油勘探及在与密度界面相关的问题研究中有良好推广前景。

Gravity anomalies caused by two or more interfaces are characterized by unseparatibility, that is, anomalics can not be separated simply according to mathematical indices such as the amplitude and horizontal width of the anomalies. This paper presents a method which simultaneously inverses the depthes and configurations of two 3-D interfaces from the stacked gravity anomalies caused by as many interfaces. The inversion result is more reliable by using fast forward algorithm in the detailed partition of the interfaces. The inversion algorithm has a good performance in convergence and a high convergent rate. Because the concepts such as parallel and adaptive connexion of neural network are preserved in the quasi-BP algorithm, it makes the initial control condition easy to be used and reduces the ambiguity of solutions. Theoretical modeling and application in an oil field indicate that the technique is promising in the research on problems relatcd to densityinterfaces.