基于改进BP神经网络的地层划分方法

Stratigraphic Division Method Based on Improved BP Neural Network

测井曲线分层是地球物理勘探利用测井资料了解地层情况首先要完成的基础工作。针对测井曲线自动分层问题,提出了一种基于改进BP神经网络的地层划分方法。首先针对三层BP神经网络模型,设计了改进的L-M算法以提升其逼近能力。然后设计了基于BP神经网络的地层划分方法。该方法精选了描述地层岩性类别的六个特征,将这些特征进行数据滤波和归一化后构造训练样本,实施网络训练,训练后的网络即可用于同类地区的地层划分。最后以辽河油田某区块的测井资料为基础数据进行地层划分,实验结果表明,与普通L-M算法比较,基于改进L-M算法的BP神经网络,地层划分结果的准确率大约提升3~5个百分点。因此,提出的基于改进BP神经网络的地层划分方法为测井曲线的自动划层提供了新思路。

Well logging curve stratification is the first basic work for geophysical exploration to understand the stratigraphic condition by using well logging data.In view of the problem of logging curves’automatic stratification,a method of stratigraphic division based on improved BP neural network is proposed.Firstly,in the light of the three layer BP neural network model,the improved L-M algorithm is designed to improve its approximation.Then the method of stratigraphic division based on BP neural network is designed.In this method,six features describing the lithologic classification of strata are selected.After constructing training samples out of the filtered and normalized data,the training of the neural network that can be used for stratigraphic classification in similar areas is carried out.Finally,the stratigraphic division based on the logging data of a block in Liaohe Oilfield is carried out.The experiment shows that compared with the common L-M algorithm,the BP neural network based on the improved L-M algorithm improves the accuracy of stratigraphic division by about 3%~5%.Therefore,the proposed method provides a new idea for logging curves’automatic stratification.