随机森林分类方法在储层岩性识别中的应用

Random forest classification method in the application of reservoir lithology recognition

为通过测井数据对储层岩性进行精确的识别,选取自然伽马、声波时差、岩石体积密度、中子密度、微球形聚焦测井、深侧向、浅侧向等7种测井参数作为判别指标.对相关性较高的指标进行因子分析,提取公共因子作为随机森林模型的输入,建立基于因子分析和随机森林的储层岩性判别模型.利用20组测井数据作为学习样本进行训练,并采用回代估计法进行检验,误判率为1/10.用另外8组数据作为测试样本进行模型检验.结果表明:所得判别模型泛化误差满足精度要求,检验结果的误判率为1/8.

In order to identify the reservoir lithology accurately according to logging data, this paper selected natural gamma, acoustic time, rock density, neutron density, volume micro spherically focused logging, deep lateral, shallow lateral, etc., seven kinds of logging parameters as indicators, and analyzed indicators of high correlation with factor analysis method to extract the public factors as input of random forest model, then established the discriminant model of reservoir lithology based on factor analysis and random forests. Using 20 groups logging data as learning samples to train, and use back substitution method for testing, the misjudgment rate was 1/10. Taking another 8 sets of data as test samples to verify the model, results show that the generalization error of discriminant model meets requirements of precision and misjudgment rate was 1/8.