基于多元线性拟合与卷积神经网络的测井曲线环境校正研究——以新和—三道桥地区古近纪—白垩纪地层为例

Research on Environmental Correction of Logging Curves Based on Multiple Linear Fitting and Convolutional Neural Network:A Case Study of the Palaeogene-Cretaceous Strata in Xinhe-Sandaoqiao Area

测井曲线质量是影响岩石物理建模和地震标定等工作的重要因素。多种地质和工程因素都可能导致井壁垮塌,进而造成密度等探测深度较浅的曲线失真。本文提出一种基于多元线性拟合与卷积神经网络相结合面向不同程度的井眼垮塌测井曲线的环境校正方法,即利用多元线性拟合方法,实现井眼垮塌较小井的环境校正并作为扩充样本,训练卷积神经网络以实现井眼垮塌较大井的环境校正。该方法能够针对测井曲线的失真程度和样本多少,充分考虑多元线性拟合与卷积神经网络方法的适应性与技术优势,校正到接近原状地层的测井曲线。本文方法应用于国内西部某油田深层碎屑岩地层,有效地改善了研究层段的岩石物理特征的规律性和提高了井震标定质量,有利于研究人员正确理解地震响应特征,为油气田测井岩石物理分析和地震储层预测提供技术支撑。

The quality of logging curve is an important factor affecting rock physics modeling and seismic calibration.A variety of geological and engineering factors can cause borehole collapse,resulting in shallow curve distortion such as density.In this paper,a method of logging curve environmental correction based on multiple linear fitting and convolutional neural network is proposed.Multiple linear fitting method is used to realize environmental correction of wells with small hole collapse,and the convolutional neural network is trained as an extended sample to realize environmental correction of wells with large hole collapse.This method can fully consider the adaptability and technical advantages of multiple linear fitting and convolutional neural network for logging curve distortion and sample size to be corrected to the logging curves close to the original formation.The method in this paper is applied to the deep clastic rock formation of an oilfield in western China,which effectively improves the regularity of rock physical characteristics and the quality of well seismic calibration and provides technical support for logging rock physical analysis and seismic reservoir prediction.