基于BSMOTE-SVM算法的溢流风险评价  

Overflow Risk Evaluation Based on BSMOTE-SVM Algorithm

作　　者：张禾[1] 池紫欣 ZHANG He;CHI Zixin(School of Mechatronic Engineering,Southwest Petroleum University,Chengdu 610500,China)

机构地区：[1]西南石油大学机电工程学院,四川成都610500

出　　处：《控制工程》2023年第12期2173-2178,共6页Control Engineering of China

基　　金：中石油-西南石油大学创新联合体项目(2020CX040302)。

摘　　要：现今钻井作业中各平台仍然依靠人工坐岗进行溢流预警,溢流风险判断具有主观性导致准确率十分有限。针对目前溢流风险识别能力弱和准确率低的问题,首先,采用了边界样本过采样方法避免了由于溢流发生频率极低导致可获取的样本数量不足的缺陷;其次,引入支持向量机对改善后的样本构造分类器,建立了溢流风险评价模型,并采用已经完钻的油井数据作为测试集进行模型验证。实验结果表明,所提方法将溢流识别准确率提高到了90%,相对于原始不均衡样本训练的分类器准确率,提高了36.67%。同时,此研究成果提高了钻井作业中的溢流识别能力,为安全钻井提供了有力支撑。At present,in drilling operations,each platform still relies on manual sitting for overflow warning,and the overflow risk judgment is subjective,resulting in limited accuracy.In view of the current problems of weak ability and low accuracy of overflow risk identification,firstly,the boundary sample oversampling method is used to avoid the defect of insufficient number of samples that can be obtained due to the extremely low frequency of overflow.Secondly,a support vector machine is introduced to construct a classifier for the improved samples,and an overflow risk evaluation model is established.Using the oil well data that has been drilled as the test set for model verification,experimental results show that the method improves the accuracy of overflow recognition by 90%,and improves the accuracy of the classifier trained by the original unbalanced sample by 36.67%.The research results improve the ability of overflow identification in drilling operations,which provides strong support for safe drilling.

关 键 词：不均衡数据集 支持向量机 边界合成少数类过采样技术 溢流风险评价 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]