基于偏鲁棒M-回归的间歇过程离群点检测  

Outlier Detection for Batch Processes Based on Partial Robust M-Regression

作　　者：贾润达[1] 毛志忠[1] 

机构地区：[1]东北大学信息科学与工程学院,辽宁沈阳110819

出　　处：《东北大学学报（自然科学版）》2013年第11期1558-1561,共4页Journal of Northeastern University(Natural Science)

基　　金：国家高技术研究发展计划项目(2011AA060204);国家自然科学基金资助项目(61203103)

摘　　要：以多元统计分析技术为核心的间歇过程建模、在线监测逐渐成为过程工业的关注焦点,然而过程数据中存在的大量离群点将直接影响上述方法的可靠性,为此提出了一种基于偏鲁棒M-回归的间歇过程离群点检测方法.首先基于极大相关熵估计建立鲁棒预测模型;然后利用偏鲁棒M-回归算法计算模型的回归系数;最后采用Hampel识别器分析最终的权值,从而实现离群点的检测.将所提方法应用于某间歇反应过程,实验结果验证了方法的有效性.Batch processes modeling, online monitoring with multivariate statistical analysis at the core have gradually became the research focus of process industry, however, the reliability of such methods would be affected by a large number of outliers in process data. Thus, an outlier detection method for batch processes based on partial robust M-regression was proposed to solve this issue. First, the robust prediction model was established based on maximum correntropy estimator. Then partial robust M-regression algorithm was utilized to calculate the model regression coefficients. Finally, Hampel identifier was used to analyze the final weights, and the outlier detection was fulfilled. The proposed outlier detection method was applied to a batch reaction process, and the experimental results indicated that effectiveness of the r^ror^osed method_

关 键 词：间歇过程 离群点 偏鲁棒M-回归 相关熵 Hampel识别器 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]