3种经典机器学习算法在火山岩测井岩性识别中的对比

岩性识别一直是火山岩油气勘探中的重要课题，基于测井数据的岩性识别也逐渐成为火山岩研究的需要，大数据背景下的机器学习算法为测井岩性识别提供了一个新方向。为提高某研究区火山岩岩性识别符合率，本文采用K近邻、支持向量机和自适应增强3种经典机器学习算法，对研究区内的粗面岩、非致密粗面岩、辉绿岩、辉长岩、玄武岩和非致密玄武岩等6类中基性火山岩进行岩性识别。从常规测井系列中优选对研究区岩性敏感的自然伽马、声波时差、补偿中子、深侧向电阻率和补偿密度等5种测井参数作为岩性识别模型的输入向量，从研究区内5口有岩心样品或薄片鉴定资料的目标层中选取测井数据点1 440个，其中960个作为训练样本，其余480个作为测试样本。以识别符合率和时间作为评价指标，对3种算法的识别结果进行对比分析，实验表明：自适应增强算法的分类准确率最高，6类岩性平均识别符合率达到82.10%；支持向量机算法表现良好，平均识别符合率为81.04%；K近邻算法平均识别符合率为76.04%。

Comparison of Three Classical Machine Learning Algorithms for Lithology Identification of Volcanic Rocks Using Well Logging Data

Lithology identification has always been an important project in oil and gas exploration of volcanic rocks, and based on logging data, it has become a need for volcanic rock research. Machine learning algorithms under the background of big data provide a new direction for logging lithology identification. In order to improve the lithology recognition accuracy of volcanic rocks in the study area, K-nearest neighbor (KNN), support vector machine (SVM), and adaptive boosting (Ada Boosting) various classic machine learning algorithms are used here to identify six types of volcanic rocks,which consist of basalt, non-compacted basalt, trachyte, non-compacted trachyte, gabbro and diabase. Five types of well logging parameters sensitive to the lithology of the study area are selected from the conventional logging series as input vectors. 1 440 logging data points are selected from five wells with core samples or segmented data, 960 of them are used as training samples, and the remaining 480 are used as test samples. Using recognition accuracy and time as evaluation indicators, the recognition results of the three algorithms are compared and analyzed. The experiments show that the classification accuracy of the Ada Boosting algorithm is the highest with an average recognition rate of 82.10% for six types of lithology; The SVM algorithm performs well with an average recognition rate of 81.04%; The recognition rate of KNN algorithm is 76.04%.