水力压裂作为页岩气储层开采的核心技术,在压裂过程中水力裂缝的扩展会遇到天然裂缝,与天然裂缝相交后压裂缝的扩展特征对缝网的形成有明显影响,从而影响最终压裂改造效果.基于内聚力单元建立了基于断裂力学的页岩气储层渗流-应力-断裂耦合的水力裂缝与多个天然裂缝相交扩展模型,研究了不同天然裂缝倾角、天然裂缝尺寸、应力差、压裂液排量和黏度下水力裂缝与天然裂缝扩展形态规律.采用基于Bagging算法集成支持向量机(support vector machines,SVM)、决策树(decision tree classifier,DTC)、逻辑回归(logistic regression,LR)、K最邻近算法(K-nearest neighbor,KNN)的裂缝形态分类器对裂缝相交扩展形态进行预测,并将Bagging算法预测结果与SVM、DTC、LR、KNN预测结果进行比较.研究结果表明:基于Bagging集成算法对水力裂缝与天然裂缝相交扩展形态预测准确率达到了92.58%,相较于单个算法,最高提升了17.95%,其中应力差越小、天然裂缝倾角、裂缝尺寸越大和压裂液排量、黏度越低越容易产生剪切缝,反之容易形成穿层缝.通过建立的水力压裂预测结果的数据集对裂缝扩展的路径和形态进行智能化预测,为利用人工智能和机器学习算法进行完井压裂优化设计提供了方法参考和对实际压裂过程优化设计提供了重要参考依据.
Intelligent Prediction of Shale Hydraulic Fracturing Network Evaluation Based on Artificial Intelligence Model
Hydraulic fracturing is the core technology of shale gas reservoir exploitation.In the process of fracturing,the expansion of hydraulic fractures will encounter natural fractures.After intersecting with natural fractures,the expansion characteristics of fractures have a significant impact on the formation of fracture network,thus affecting the final fracturing reconstruction effect.Based on cohe-sion unit,a seepage-stress-fracture coupling model of hydraulic fractures intersecting with multiple natural fractures in shale gas reser-voir was established based on fracture mechanics.The propagation patterns of hydraulic fractures and natural fractures under different natural fracture inclination,natural fracture size,stress difference,fracturing fluid displacement and viscosity were studied.The frac-ture morphology classifier based on Bagging algorithm integrated support vector machine(SVM),decision tree classifier(DTC),logis-tic regression(LR)and k-nearest neighbor algorithm(KNN)was used to predict the fracture intersection propagation morphology,and the prediction results of Bagging algorithm were compared with those of SVM,DTC,LR and KNN algorithms.The results show that the prediction accuracy of the intersection propagation morphology of hydraulic and natural fractures based on Bagging integrated algorithm reaches 92.58%,and the maximum improvement is 17.95%compared with a single algorithm.The smaller the stress difference,the larger the natural fracture inclination,the larger the fracture size,and the lower the fracturing fluid displacement and viscosity,the more easily shear fractures are generated.Otherwise,the more easily perforated fractures are formed.The path and shape of fracture propagation are intelligently predicted through the established data set of hydraulic fracturing prediction results,which provides a method reference for the optimization design of well completion fracturing by using artificial intelligence and machine learning algorithms and an important reference for the optimization design of actual fracturing process.
万方数据
维普
