神经网络在测井解释中的应用

The application of the artificial neural network in the log interpretation

本文利用人工神经网络来确定储层参数：孔除度和渗透率；识别油（气）、水、干层，以便对它们进行评价和开采。其过程是：首先选取适当的网络模型和学习算法，然后利用关键井的岩芯分析数据、试油结果和相应的一组测井数据训练网络，建立研究地区的测井解释模型，最后运用已建立的模型对未知井进行预测，绘制出各井的解释成果图。本文简述了这项技术的基本原理、特点和改进措施；进入学习前的一系列预处理过程，包括深度归位、测井曲线标准化、全区测井数据统一的归一化、样本选取及其原则、自动分层和取特征值方法等；最后通过多种检验手段对学习和预测效果进行评价。通过检验和在多个地区的应用，证实了这一技术的优越性，其地质效果是明显的，大大提高了测井解释精度和速度，为测井解释开辟了一条新途径。

In this paper, thc artificial neural nctwork is used to estimate porosity and permcability, toidentify oil-bcaring, gas-bearing, water-bearing and dry beds' The procedurce are: first of all,selccting an appropriate network model and an appropriate learning algorithm ; thcn trainingnetworks by use of the core analysis data, test oil results and a group of relevant logging data ofkey wells and building the model in the exploratory area for logging interpretation ; finally,pridicting unknown wells by means of thc cstablished model and plotting interprctation rcsultsThis paper gives a brief description of the basic principle, the features and the improvedmeasures about this technique. The preliminary processing before entering the learning stageincludes: depth migration, standardization of logging curves, unified normalization of all loggingdata on the whole area, sample selection according to selection principles, auto-stratification, andeigenvalue selection. At the end, the effect of learning and pridieting are commented throughmultiple means of detection.The results of the detection and its application in many exploratory areas confirm that thetechnique is execllent, which greatly inereases the accuracy of interpretation, greatly shortens theprocess of interpretation, and contributes a new way of logging interprctation.