基于加权MMMD的软测量数据过失误差侦测

Gross error detection of soft-sensing data based on WMMMD

在基于聚类分析的软测量过失误差侦测中,针对不同变量对建模影响程度的不同,提出了基于加权MMMD聚类分析的过失误差侦测方法。该方法在聚类的相似度计算时,根据各辅助变量与主导变量的相关性,给每个变量加以不同的权值。在新的相似性度量基础上进行聚类,并侦测出过失误差。用该方法对丙烯浓度的软测量数据进行了过失误差侦测,并用剔除过失误差后的数据做了建模仿真。实验结果表明了新方法的有效性,并能提高软测量数据的整体质量。

Concerning the different influence of variables on the modeling in the gross error detection of soft-sensing based on clustering algorithm, a new method ofgross error detection based on weighted MMMD clustering is proposed. This method uses weighted variables, which based on the correlation between auxiliary variables and leading variable, in the similarity calculation. And then, gross errors are detected by the clustering algorithm based on the new similarity proposed. The method is used to detect the gross errors in modeling data for a propylene concentration soft sensor, and modeling simulations are done. The validity of the new method is proved by the experimental results, and the overall quality of the data is improved.