摘要
化工过程长时间处于正常运行状态而积累的故障样本有限,且含有冗余信息,降低了传统故障分类器的准确率。为了提高化工过程故障诊断的准确率,提出了一种基于RS和SVM的化工过程高精度故障诊断方法。首先,在不损失信息的情况下,采用RS约简故障指标体系,去除冗余特征;然后,根据最小约简指标集构建故障数据集,建立优化的SVM故障分类器。将RS-SVM和标准SVM同时应用到预加氢过程的故障分类中,RS不同程度的提高了SVM准确率,当采用RBF核函数且训练样本集容量为60时,准确率提高幅度最大值为11.84%。比较结果表明,剔除数据中的冗余信息有助于提高故障诊断的准确率。
Fault diagnosis plays an important role to prevent accidents in chemical processes that are in the normal operation for a long time. However, the limited number of fault samples with redundant information reduces the accuracy of traditional fault classifiers. In order to improve the accuracy of fault diagnosis in chemical processes, a novel fault diagnosis method for chemical processes with redundant information, which integrates the rough set theory (RS) with a support vector machine (SVM) , named RS-SVM, is proposed in this paper. In the first stage, RS is utilized to eliminate redundant features by reducing fault index system with no information loss. In the second stage, fault dataset based on minimal reduction index set obtained by RS is employed to establish the optimized SVM classifier. The effectiveness of the proposed method is verified by simultaneously applying RS-SVM and standard SVM to a pre-hydrogenation process. Results reveal that based on the simplicity attribute of RS, the accuracy of SVM is increased in various degrees. When the RBF kernel function is used and the fault sample capacity is 60, the accuracy of SVM is increased by the largest margin, 11. 84%. In this aspect, eliminating redundant information is helpful to improve the accuracy of fault diagnosis.
出处
《石油学报(石油加工)》
EI
CAS
CSCD
北大核心
2017年第4期777-784,共8页
Acta Petrolei Sinica(Petroleum Processing Section)
基金
国家自然科学基金项目(51574263)
中国石油大学(北京)青年创新团队C计划(C201602)
中国石油大学(北京)科研基金项目(2462015YQ0403)资助
关键词
化工过程
故障诊断
冗余信息
粗糙集
支持向量机
chemical process
fault diagnosis
redundant information
rough set
support vector machine