浊积岩储层压裂效果数据挖掘研究  

Data Mining Research on Effect Analysis of Fracturing in Shengli Oilfield Turbidite Reservoir Based On Support Vector Machine Algorithm

作　　者：孟庆民[1] 刘恩新[1] 孙辉[1] 刘太昂 李敏杰[3] 

机构地区：[1]中石化胜利油田分公司采油工艺研究院,山东东营257000 [2]上海恒阳数据技术有限公司,上海200444 [3]上海大学理学院化学系,上海200444

出　　处：《石油天然气学报》2014年第1期92-95,7,共4页Journal of Oil and Gas Technology

摘　　要：在压裂设计和效果分析中,主要是依据压裂裂缝模型进行求解,分析结果取决于裂缝模型的选取,导致分析结果差别较大。利用数据挖掘技术,可以从历史的压裂数据中发现规律,建立的模型往往更加准确。以胜利油田浊积岩储层压裂数据为例,利用支持向量机算法建立了是否达到工业油流的定性判别模型和压裂后日产液的定量预测模型。其中定性判别模型的建模准确率为93.18%,留一法准确率为81.82%。定量预测模型建模结果的平均相对误差为10.97%,留一法结果的平均相对误差为20.89%。定性与定量的模型分析结果都和实际效果比较符合,可以为压裂分析提供辅助决策。In hydraulic fracturing design and evaluation of treatment,the analysis was always based on the fracture model.The results often differed greatly because of the selection of the fracture model.Data mining technique was applied to find the rules of analysis from historical fracturing data,by which the model established was more accurate.By taking turbidite reservoirs in Shengli Oilfield for example,support vector machine algorithm was used to establish qualitative model for judging whether the industrial oil being achieved and quantitative model for predicting the daily oil production after fracturing.The accuracy of qualitative model was 93.2% and the accuracy 81.2% was for leave-one-out cross-validation(LOOCV)method.The average relative error of quantitative model was 10.97% and the error was20.9%in LOOCV method.The results of qualitative and quantitative modeling are basically agreed with those of practical application,the data mining method can support the decision for fracturing analysis.

关 键 词：水力压裂 支持向量机 浊积岩 数据挖掘 

分 类 号：TE357[石油与天然气工程—油气田开发工程]