机器学习方法对砂砾岩岩屑成分的预测——以西北缘X723井百口泉组为例

发布时间：2018-04-11 15:31

  本文选题：岩屑成分预测 + 砂砾岩　； 参考：《西安石油大学学报(自然科学版)》2017年05期

【摘要】：选择凝灰岩岩屑作为预测对象,对测井数据进行标准化处理,对砂砾岩储层薄片鉴定结果和测井数据进行相关性分析,优选对岩屑敏感的CNL、GR、RT、RI、SP测井参数作为训练学习的对象;分别利用SVM、BP神经网络、CART、BP神经网络-Bagging、CART-Bagging、随机森林等机器学习方法建立岩屑预测模型,对西北缘X723井百口泉组岩屑成分进行预测、对比和分析。结果表明:单个机器学习方法预测效果不佳,而经集成学习方法优化的BP神经网络-Bagging、随机森林取得较好的实验结果,尤其是随机森林的预测效果最好,平均相对误差绝对值为17.17%,证实机器学习方法在本工区预测岩屑成分是有效的,可以进行推广。
[Abstract]:Tuff cuttings are selected as prediction objects, log data are standardized processed, correlation analysis is carried out on the results of sheet identification and logging data of sandy gravel reservoir, and logging parameters are selected as the object of training and learning.The cuttings prediction model was established by using SVMS-BP neural network and Carton-BP neural network (Baggingling CART-Bagginging, random forest), respectively. The cuttings composition of Baikouquan formation in X723 well in northwest margin was predicted, compared and analyzed.The results show that the prediction effect of single machine learning method is not good, but the BP neural network (BP neural network), which is optimized by integrated learning method, obtains better experimental results, especially the prediction effect of random forest is the best.The absolute value of the average relative error is 17.17. It is proved that the machine learning method is effective in predicting cuttings in this area and can be popularized.
【作者单位】： 长江大学地球科学学院;中国石油新疆油田公司勘探开发研究院;
【基金】：国家科技重大专项(2016ZX05027)
【分类号】：P618.13;P631.81
，

本文编号：1736582