摘要
随着数据科学的快速发展,大数据分析法被广泛的应用于非常规油气勘探开发领域,利用机器学习方法构建水力压裂单井产能预测模型,不仅能够对每口井进行准确地压裂效果评价,还可以进行压裂施工参数优化.以苏里格气田东区712口压裂直井为例,应用多元逐步回归、随机森林、支持向量机和BP神经网络算法建立了四种产能预测模型,预测了142口测试井的无阻流量.通过综合评价分析得出随机森林算法建立的产能模型预测精度最高,无阻流量预测值和实际观测值的相关系数达到0.87、均方误差0.204、决定系数0.792,结果表明随机森林预测模型性能最佳,最终选择随机森林算法构建该区单井产能预测模型.
With the rapid development of data science,big data analytics is widely applied to the exploration and development of unconventional oil and gas fields.Application of machine learning methods to predict well productivity can not only accurately evaluate the fracturing effectiveness of a hydraulically fractured well,but also optimize the fracturing treatment parameters.This paper takes 712 vertical fractured wells in the eastern area of Sulige Gas Field as an example,applies multiple stepwise regression,random forest,support vector machine and BP neural network algorithm to establish four well productivity models to estimate the absolute open flow(AOF)potentials of 142 test wells.Through comprehensive evaluation and comparison,it is concluded that the random forest algorithm have the highest prediction performance.The correlation coefficient between the predicted AOF values and the actual observation values reaches 0.87,the mean square error is 0.204,the coefficient of determination is 0.792,indicating that the prediction performance of the random forest model is best.Finally,the random forest algorithm was selected to build the prediction model of the well productivity in this area.
作者
马先林
樊毅龙
MA Xian-lin;FAN Yi-long(College of Petroleum Engineering,Xi’an Shiyou University,Xi’an 710065,China;Shaanxi Key Laboratory of Advanced Stimulation Technology for Oil&Gas Reservoirs,Xi’an 710065,China)
出处
《数学的实践与认识》
2021年第19期186-196,共11页
Mathematics in Practice and Theory
基金
国家自然科学基金(51974253,51934005)
陕西省教育厅重点实验室科研计划项目(18JS085)
陕西省教育厅科研计划项目(20JS117)
陕西省自然科学基础研究计划项目(2020JQ-781)。
关键词
水力压裂
机器学习
预测模型
随机森林
R语言
hydraulic fracturing
machine learning
prediction model
random forest
R language