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鲁棒M估计的间歇过程离群点检测方法。该方法首先通过积分方程离散化将模型参数估计问题转化为最小二乘优化问题;分别利用Tikhonov正则化方法及鲁棒M估计消除噪声和离群点对模型参数估计的影响;最后通过分析各个样本点的权值,实现过程数据的离群点检测。将所提出的方法应用于半间歇反应过程,实验结果验证了方法的可行性与有效性。
The batch process modeling and online monitoring with multivariate statistical analysis technique as the core have gradually become the focus of process industry;however, a large number of outliers existing in the process data will directly affect the reliability of these methods, and an outlier detection method for batch processes based on robust M-estimation is proposed to solve this problem. First, integral equation discretization is used to convert the model parameter estimation problem to a least squares optimization problem;Tikhonov regularization method and ro- bust M-estimation are utilized to eliminate the effects of noise and outliers on the model parameter estimation, respectively;finally, the outlier detection for process data is achieved through analyzing the weights of the sample points. The proposed outlier detection method was applied to a semi-batch reaction process;and the experiment results verify the feasibility and effectiveness of the method.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2013年第8期1726-1731,共6页
Chinese Journal of Scientific Instrument
基金
国家高技术研究发展计划(2011AA060204)
国家自然科学基金(61203103)
中央高校基本科研业务费(N110304006)资助项目
