基于机器学习的高含水期油井产量预测方法  

Prediction Method of Oil Well Production in High Water Cut Period Based on Machine Learning

作　　者：韩益东 尹洪军 徐国涵[1,2] 桑昌 HAN Yidong;YIN Hongjun;XU Guohan;SANG Chang(College of Petroleum Engineering,Northeast Petroleum University,Daqing 163318,Heilongjiang China;Key Laboratory of Enhanced Oil Recovery of Ministry of Education,Northeast Petroleum University,Daqing 163318,Heilongjiang China)

机构地区：[1]东北石油大学石油工程学院,黑龙江大庆163318 [2]提高油气采收率教育部重点实验室(东北石油大学),黑龙江大庆163318

出　　处：《河南科学》2022年第10期1569-1575,共7页Henan Science

基　　金：国家科技重大专项项目(2017ZX05071);黑龙江省自然科学基金项目(LH2022E023)。

摘　　要：高含水期油井产量预测存在井况资料复杂,数据信息不完整等问题,故提出了结合三种机器学习算法的油井产量预测方法.通过梳理油井特征数据样本,构建数据清洗流程,建立形成高含水油井特征样本库,基于随机森林算法和XGBoost算法确定油田高含水期影响油井产量的主控因素.利用随机森林、XGBoost和BP神经网络算法进行单一模型的拟合和预测,通过选取适当的权值系数结合三种方法得到新的组合预测模型.研究表明:基于随机森林和XGBoost的算法模型提高了特征选择的稳定性和鲁棒性,并且组合预测模型方法比单一模型更精确预测了高含水X区块油井的月产量,与实际产量符合较好.该方法为高含水期油田油井的特征数据集的构建、特征提取及初期产量预测提供了新的思路.Problems such as complicated well condition data and incomplete data information,exist in the prediction of oil well production in high water cut period.In this study,a prediction method of oil well production combining three machine learning algorithms was proposed.By combing the oil well characteristic data samples,the data cleaning process was constructed,and the high water cut oil well characteristic sample database was established and formed.Based on random forest algorithm and XGBoost algorithm,the main controlling factors affecting oil well production in high water cut period were determined.Random forest,XGBoost and BP neural network algorithms were used to fit and predict the single model,and a new combined prediction model was obtained by selecting appropriate weight coefficients combined with the three methods.The results show that the algorithm model based on random forest and XGBoost improves the stability and robustness of feature selection,and the combined prediction model can predict the monthly production of high water cut X block oil well more accurately than the single model,which is in good agreement with the actual production.This method provides a new idea for the construction of characteristic data set,feature extraction and initial production prediction of oil wells in high water cut stage.

关 键 词：随机森林 XGBoost 产量预测 神经网络 组合预测模型 

分 类 号：TE343[石油与天然气工程—油气田开发工程]