基于BP神经网络的页岩静弹性模量预测研究

页岩的静弹性模量是页岩油气资源勘探开发整个过程的重要参数,现阶段页岩静弹性模量的预测往往是先使用岩芯纵波时差及密度计算出动态弹性模量,再寻找动、静态弹性模量之间的关系。岩石矿物组成的差异常常导致常规思路得到的动、静态弹性模量的相关性较差,预测结果难以满足工程需求。为完成对研究区块岩芯的静弹性模量预测研究,首先对岩芯进行密度及纵波时差的测量;而后运用全岩矿物分析、黏土矿物分析及三轴压缩试验的方法对岩芯静弹性模量进行对比分析;并由三轴压缩试获取岩芯静弹性模量。按输入变量的不同建立并训练了三个BP神经网络预测系统;并对三个预测系统的应用效果加以对比分析。分析结果表明:只以岩芯密度和纵波时差为输入变量的BP神经网络的预测效果较差;以岩芯密度、纵波时差、石英含量及伊利石含量为输入变量时的BP神经网络预测效果较好,以岩芯密度、纵波时差、石英含量、方解石含量、伊利石含量及伊/蒙混层含量为输入变量时的BP网络预测效果最好。

Prediction of shale static elastic modulus based on BP Neural Network

Static elastic modulus of shale is an important parameter in the whole process of exploration and development of shale oil and gas, shale-static elastic modulus prediction at this stage tend to use core compressional wave slowness and the density to calculate the dynamic modulus and look for relationships between dynamic and static elastic modulus. Differences rock mineral composition often leads to poor correlation between dynamic elastic modulus, which according to conventional thinking, and static elastic modulus, prediction result is difficult to meet project requirements.To complete the prediction of static elastic modulus of cores belong to our research block, density and compressional wave slowness of cores were measured firstly, then, the static modulus was comparatively analyzed by using the whole rock mineral analysis and clay mineral analysis were done the three axis compression test, and we can obtain the elastic modulus of cores by doing three axis compression test. Three BP neural network prediction systems were established and trained according to different input variables, and there is a comparative analysis about the applied effect of the three systems. The results of analysis are: the prediction effect of the BP network, which only make core density and compressional wave slowness as input variables, is relatively poor, and it is better when core density, compressional wave slowness, content of quartz and content of illite were made as input variables. And the prediction effect of the BP network, which make core density, compressional wave slowness, content of quartz, content of calcite, content of illite and content of illite / smectite as input variables, is the best.