随机森林的融合模型在压井方法分类中的研究  

Research on Classification of Well Killing Method Based on Random Forest Fusion Model

作　　者：钟原 张泰 李平[1] 杨绪华 ZHONG Yuan;ZHANG Tai;LI Ping;YANG Xuhua(School of Computer Science,Southwest Petroleum University,Chengdu,Sichuan 610500,China)

机构地区：[1]西南石油大学计算机科学学院,四川成都610500

出　　处：《西南石油大学学报（自然科学版）》2022年第1期165-173,共9页Journal of Southwest Petroleum University(Science & Technology Edition)

基　　金：油气藏地质及开发工程国家重点实验室开放基金(PLN201731)。

摘　　要：针对井口压力控制作业中传统方法过度依赖专家经验和数学模型运算精度的问题,提出一种基于随机森林(Random Forest,RF)的多模型融合算法对压井方式进行分类判断。首先,将专家经验结构化、数据化,转化成可被机器学习模型使用的数据形式,同时,结合油气井的基础数据和工况参数,作为智能模型的重要参数来描述压井作业的特征空间;然后,将特征数据通过特征工程进行特征筛选、特征编码和特征选择等处理;最后,构建出基于随机森林的Stacking双层融合模型,实现压井方法的分类预测。通过实验验证,与单模型的机器学习算法相比,本方法具有更高的预测精度。In wellhead pressure control operations,the traditional method relies too much on expert experience and the accuracy of mathematical model calculations.In this paper,we propose a multi-model fusion algorithm based on Random Forest(RF)for judgments of classification of well killing methods.Firstly,the structure and data of expert experience are transformed into a data form that can be used by machine learning models.Meanwhile,the basic data of oil and gas wells and working condition parameters are used as important parameters of intelligent model to describe the feature space of well killing operations.Then,the feature data are processed by feature engineering for feature selection,feature code and feature choose.Finally,a stacking double-layer fusion model based on Random Forest is constructed for the implement of predictions for classification of well killing method.The experimental results show that our method has more high prediction accuracy than other machine learning algorithms that has only single model.

关 键 词：随机森林 融合模型 压井方法分类判断 井口压力控制 

分 类 号：TE21[石油与天然气工程—油气井工程]