Stacking算法在小样本预测上的适用性研究:以实验室金属挂片的腐蚀速率预测为例  

Applicability of stacking algorithm in small sample prediction:Taking the corrosion rate prediction of metal hanging pieces in laboratory as an example

作　　者：郑鹏飞 杨洋[1] 石鑫[2] 闻小虎 Zheng Pengfei;Yang Yang;Shi Xin;Wen Xiaohu(School of Geoscience and technology,Southwest Petroleum University,Chengdu Sichuan,610500;Key Laboratory for EOR of carbonate fractured vuggy reservoir,Sinopec,Urumqi Xinjiang,830011;Engineering Technology Research Institute of Northwest Oilfield Company,Sinopec,Urumqi Xinjiang,830011)

机构地区：[1]西南石油大学地球科学与技术学院,四川成都610500 [2]中国石油化工集团公司碳酸盐岩缝洞型油藏提高采收率重点实验室,新疆乌鲁木齐830011 [3]中国石油化工股份有限公司西北油田分公司工程技术研究院,新疆乌鲁木齐830011

出　　处：《电子测试》2022年第1期48-50,92,共4页Electronic Test

摘　　要：集成学习在处理小样本问题上具有相当的优势,相较于其他集成模式,Stacking模式对集成单元的类型并没有限制,所以具有相当的研究潜力。在油气领域,如何对实验室条件下对金属腐蚀数据的充分利用,是当前急需解决的问题。为了探究stacking算法在小样本预测上的适用性,本研究以实验室条件下获得的99组金属腐蚀数据为基础,在预处理后,然后选择了11组基础集成模型以stacking的模式进行集成并预测。最终的结果表明stacking模式并不适用于该数据集下的小样本预测。Ensemble learning has considerable advantages in dealing with small sample problems.Compared with other integration modes,stacking mode has no restrictions on the types of integration units,so it has considerable research potential.In the field of oil and gas,how to make full use of metal corrosion data under laboratory conditions is an urgent problem to be solved.In order to explore the applicability of stacking algorithm in small sample prediction,this study is based on 99 groups of metal corrosion data obtained under laboratory conditions.After preprocessing,11 groups of basic integration models are selected to integrate and predict in stacking mode.The final results show that the stacking model is not suitable for small sample prediction under this data set.

关 键 词：小样本 STACKING 油气腐蚀 集成学习 

分 类 号：TE988.2[石油与天然气工程—石油机械设备] TP181[自动化与计算机技术—控制理论与控制工程]