基于机器学习的油气水层随钻识别模型优选

Optimization of Model for Identification of Oil/Gas and Water Layers While Drilling Based on Machine Learning

为了解决基于测井数据对油气水层的实时识别这一技术难题,利用计算机科学与现代数学,结合随钻测井技术与机器学习算法进行油气水层的随钻识别。首先,对训练集数据进行相关性分析,剔除弱相关或冗余数据;其次,选择一对多支持向量机、一对一支持向量机以及随机森林算法分别建立油气水层分类识别模型,并使用网格搜索方法及10折交叉验证法对3种分类识别模型参数进行优选;最后,运用参数优选后的各分类识别模型,对随钻测井数据进行油气水层的识别。研究结果表明,3种分类识别模型对研究区块油气水层随钻识别的准确率均达到75%以上。在训练样本较少的情况下,优先选用一对一支持向量机分类识别模型进行油气水层的随钻识别。

In order to solve the technical difficulty of real-time identification of oil/gas and water layers based on logging data,based on computer science and modern mathematics,logging while drilling technology is combined with machine learning algorithm to identify the oil/gas and water layer while drilling.Firstly,the correlation analysis is performed on the data in training set to eliminate weak correlation or redundant data.Secondly,oil/gas and water layers recognition models are established using one-versus-rest support vectormachine,one-versus-one support vector machine,and random forest algorithm respectively,and the parameters of three recognition models are optimized using grid search method and 10-fold crossvalidation method.Finally,based on logging data while-drilling,the oil/gas and water layers are identified using the models whose parameters are optimized.The results show that the recognition accuracy of three models in the study area is higher than 75%.In the case of less training samples,the recognition model established using one-versus-one support vector machine has the best effect to the identification of oil/gas and water layers while drilling.