基于AdaBoost机器学习算法的大牛地气田储层流体智能识别

大牛地气田储层复杂,矿物组分多样、储集空间复杂、非均质性强,导致流体识别困难.为提高该气田复杂储层流体识别的准确率和解释效率,以广泛发育的低阻气藏为主要研究对象,采用Adaboost机器学习算法,分别以逻辑分类、决策树等主流智能算法作为弱分类器,集成了4类强分类器模型.基于低阻气藏成因机理分析优化了模型输入参数,基于常...

Intelligent Fluid Identification Based on the AdaBoost Machine Learning Algorithm for Reservoirs in Daniudi Gas Field

Complex reservoirs in Daniudi Gas Field are characterized by diverse mineral components, complex reservoir space, and strong heterogeneity, which make fluid identification difficult. To improve the accuracy rate and interpretation efficiency of fluid identification in complex reservoirs, Daniudi Gas Field, with its extensively developed low-resistance gas reservoirs, was taken as the main research object. Then, four strong classifier models were formed by the Adaboost machine learning algorithm with mainstream intelligent algorithms (such as logical classification and decision tree) as weak classifiers. The input parameters of the model were optimized based on the analysis of the genesis mechanism of the low-resistance gas reservoir, the parameters were optimized on the basis of conventional well logging, oil testing and production testing data, etc. The above model was applied to the data of 6 actual wells. The results showed that the strong classifier achieved the best identification effect by using the decision tree algorithm as the weak classifier, with the fluid identification accuracy of 86.5% and the F1 value up to 86.6%. The results indicates that this method is effective for identifying fluid with conventional logging data for low-resistance gas reservoirs, and providing new ideas for fluid evaluation.