基于Stacking的套损预测方法研究  

Research on Prediction Method of Casing Damage Based on Stacking

作　　者：赵建民[1] 张珺博 崔佳鑫 ZHAO Jianmin;ZHANG Junbo;CUI Jiaxin(School of Computer&Information Technology,Northeast Petroleum University,Daqing 163318)

机构地区：[1]东北石油大学计算机与信息技术学院,大庆163318

出　　处：《计算机与数字工程》2024年第6期1685-1690,共6页Computer & Digital Engineering

基　　金：国家自然科学基金项目“致密油储层孔隙结构跨尺度多源融合及重构”(编号:51774090)资助。

摘　　要：根据油气生产过程中的套管损坏影响因素众多、数据复杂等特点,通过数据预处理、随机森林重要性分析等技术对现场数据进行分析与整合,采用特征工程的方法处理缺失值并选取特征参数。针对传统机器学习模型对套损预测不佳的问题,提出采用双层Stacking模式集成学习预测模型;该模型采用随机森林、支持向量机、梯度提升决策树和K近邻算法为基模型,逻辑回归为元模型,以此构建泛化能力更强的套损预测模型。结果表明,该模型较于单一的机器学习模型准确率与F1值均有提升,该模型最终的准确率达到89.21%。According to the characteristics of many factors affecting casing damage and complex data in the oil and gas produc-tion process,the field data is analyzed and integrated through data preprocessing,random forest importance analysis and other tech-nologies,and the method of feature engineering is used to process missing values and select feature parameters.Aiming at the prob-lem that traditional machine learning models are not good at predicting the set loss,a two-layer stacking mode ensemble learning prediction model is proposed.The model uses random forest,support vector machine,gradient boosting decision tree and K-nearest neighbor algorithm as the base model,and logistic regression as the meta-model to build a more generalized set loss prediction mod-el.The results show that the model has improved accuracy and F1 value compared with a single machine learning model,and the fi-nal accuracy of the model reaches 89.21%.

关 键 词：集成学习 套管损耗 套损预测 Stacking模型融合 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]