基于U-Net神经网络的声波测井曲线重构

Reconstruction of Acoustic Curve Based on U-Net Neural Networ

本文提出了一种基于U-Net神经网络的声波测井曲线重构方法。通过编码器提取自然伽马(GR)、密度(RHOB)等测井曲线的数据特征,利用解码器建立数据特征与声波测井曲线之间的映射关系,实现了声波测井曲线的精准重构。实验结果表明,该方法在保留原始声波测井低频信息的基础上,兼顾了输入测井曲线的高频特征,实现了对原始声波测井泥岩层段数据噪音的有效压制,在渤中凹陷东南环测井数据重构中取得了良好的效果,验证了该方法较高的精度和实用性。

Logging data is not only the basis of identifying underground lithologic characteristics,but also the core of well seismic joint inversion.Due to the influence of geological and construction conditions in practical work,the acoustic logging data is distorted or missing,which cannot reflect the change law of formation lithology,affecting the development of subsequent work.This paper presents a reconstruction method of acoustic logging curve based on U-Net neural network.The data characteristics of logging curves(GR,RHOB)are extracted by the encoder.The mapping relationship between data characteristics and the acoustic logging curve is established by using the decoder,then,accurate reconstruction of the acoustic logging curve is realized.The results show that this method retains the low-frequency information of the original acoustic logging,takes into account the high-frequency characteristics of the input logging curve,and realizes the effective suppression of the noise of the original acoustic logging mudstone interval data.It has achieved good results in the logging data reconstruction of the southeast rim in Bozhong sag,which verifies this method’s high accuracy and practicability.