复杂火山岩油藏岩性智能识别及预测

Intelligent Recognition and Prediction of Lithology of Complex Volcanic Reservoir

摘要: 目前火山岩油气藏正在引起广泛关注，其复杂程度超过其他油藏类型。针对二连盆地、准噶尔盆地复杂火山岩油气藏，其岩性具有复杂多变的特点，常规方法难以准确识别的问题，本文提出利用机器学习的方法对研究区岩性进行智能识别，获得了良好效果。研究中，在分析研究区火山岩储层地质特点基础上，根据取芯描述、薄片分析、成像测井等信息，分析不同岩性的测井响应特征。并根据测井信息，构造M，N两个对火山岩岩性极为敏感的参数，确定了识别岩性的8个敏感特征参数为：GR，DT，RHOB，CNL，RT，RI，M，N。根据测井特征参数和岩性标签，利用机器学习中的决策树、随机森林、梯度提升树、贝叶斯四种不同方法，建立了四种岩性识别预测模型。对不同模型进行了对比评价研究，优选出最优的随机森林岩性识别模型，岩性识别的准确率达到0.9以上，为火山岩油气藏评价奠定了基础。

Abstract: At present, volcanic rock oil and gas reservoirs are attracting widespread attention, and their complexity exceeds other reservoir types. Aiming at the complex volcanic reservoirs in Erlian Basin and Junggar Basin, their lithology is complex and changeable, and conventional methods are difficult to accurately identify the problem. This paper proposes to use machine learning methods to intelligently identify the lithology in the study area, and obtain good results. In the study, based on the analysis of the geological characteristics of the volcanic reservoir in the study area, the logging response characteristics of different lithologies were analyzed based on information such as coring description, thin section analysis, and imaging logging. And according to the logging information, the two parameters of structure M and N that are extremely sensitive to volcanic rock lithology, eight sensitive characteristic parameters for identifying lithology are determined: GR, DT, RHOB, CNL, RT, RI, M, N. According to logging feature parameters and lithology tags, four different methods of decision tree, random forest, gradient boosting tree, and Bayesian in machine learning are used to establish four lithology recognition and prediction models. Different models were compared and evaluated, and the best random forest lithology recognition model was selected. The accuracy of lithology recognition was above 0.9, which laid the foundation for the evaluation of volcanic reservoirs.