k近邻补值方法在工业过程故障诊断中的应用  

k-Nearest Neighbor Imputation Method and Its Application in Fault Diagnosis of Industrial Process

作　　者：李元[1] 吴杰[1] 王国柱[2] 

机构地区：[1]沈阳化工大学信息工程学院,沈阳110142 [2]东北大学信息科学与工程学院,沈阳110819

出　　处：《上海交通大学学报》2015年第6期830-836,共7页Journal of Shanghai Jiaotong University

基　　金：国家自然科学基金重点项目(61034006);国家自然科学基金项目(61174119;60774070)

摘　　要：针对贡献图分析方法在故障分离方面存在拖尾效应以及不能准确定位故障变量的问题,提出一种将k近邻(k-Nearest Neighbor,kNN)补值与传统贡献图相结合的故障定位方法.首先利用主成分分析建模并求取综合控制指标;然后将kNN方法与综合控制指标相结合初步提取故障变量;最终用贡献图从初步提取的故障变量中确定故障根源,该方法有效地避免了正常变量的贡献值对最终诊断结果的影响.本文运用数值算例和TE过程进行仿真,并将该方法与基于重构的贡献方法比较,验证了算法的准确性和有效性.Aimed at the smearing effect in contribution plot method and the falt that fault variables cannot be located, this paper proposed a kNN imputation method for fault diagnosis, combining k-nearest neigh- bor and the contribution plot algorithm. First, PCA was adopted to build an evaluation model and calculate the combined index. Secondly, k-nearest neighbor imputation method and the control index were combined to extract preliminary faulty variables. Finally, the contribution plots were employed to find the funda- mental faulty variables from the preliminary faults. The proposed method can avoid the influence of contri- bution values of normal variables effectively. A numerical example and Tennessee Eastman （TE） process were given to verify the effectiveness and accuracy of the proposed method, compared with the reconstruction-based method.

关 键 词：k最近邻 重构贡献 贡献图 故障诊断 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]