基于参数优化AdaBoost算法的酸性火山岩岩性分类

岩性识别是火山岩油气藏勘探的基础，为提高长岭气田火山岩岩性识别的准确率，采用决策树、支持向量机、逻辑回归、AdaBoost-决策树、AdaBoost-支持向量机和AdaBoost-逻辑回归6种算法，对研究区酸性火山岩岩性进行分类与识别。通过分析研究区火山岩不同岩性的测井响应特征，选取了对火山岩岩性、组构和孔隙结构反应灵敏的12种岩石物理测井参数作为分类特征量。选择3口井中岩心分析和岩矿录井资料完整的7 150个测井数据作为数据集，并从中随机选取70 % 的数据作为训练集建立岩性识别模型，剩余30 % 的数据作为测试集。对6种算法建立的模型通过交叉验证进行参数优化及模型评价，对比不同算法与录井剖面的结果表明，AdaBoost-决策树算法可作为长岭气田利用常规测井资料识别火山岩岩性的有效手段，准确率可达90 % 以上。

Lithology classification of acidic volcanic rocks based on parameter-optimized AdaBoost algorithm

Lithologic identification provides a basis for the exploration of volcanic oil and gas reservoirs. In order to improve the accuracy of lithologic identification of volcanic rocks in Changling gasfield, six machine learning algorithms are used, i.e., decision tree, support vector machine, logistic regression, AdaBoost-DTC, AdaBoost-SVC and AdaBoost-LR, to classify and identify the lithologies of acidic volcanic rocks. After summarizing the logging response characteristics of different volcanic rocks in the study area, 12 kinds of petro-physical logging parameters sensitively reflecting the lithology, fabrics and pore structure of volcanic rocks are selected as classification characteristic variables. A total of 7 150 pieces of logging data with complete core analysis and rocklogging data of three wells are taken as the data set, 70 % of which are randomly selected as the training set to create lithology identification model; the remaining 30 % of the data is taken as a test set. The models established for six algorithms are cross-validated for parameter optimization and model evaluation. Through comparing different algorithms and logging profiles, the results indicate that the AdaBoost-decision tree algorithm is an effective tool using the conventional logging data of Changling gas field to identify the lithologies of volcanic rocks with the accuracy more than 90 %.