基于多模型的鲁棒软测量建模方法  

Multiple model based robust soft sensor modeling method

作　　者：闫广实 刘鑫 Yan Guangshi;Liu Xin(School of Management,Heilongjiang University of Science ＆ Technology,Harbin 150022,China;School of Astronautics,Harbin Institute of Technology,Harbin 150001,China)

机构地区：[1]黑龙江科技大学管理学院,哈尔滨150022 [2]哈尔滨工业大学航天学院,哈尔滨150001

出　　处：《黑龙江科技大学学报》2018年第5期598-602,共5页Journal of Heilongjiang University of Science And Technology

基　　金：哈尔滨市科技攻关计划项目(2016AC9QT025)

摘　　要：针对系统输出数据存在异常点和非线性的情况,假设系统噪声服从包含重尾的拉普拉斯分布,提出一种基于多模型的鲁棒软测量建模方法。利用期望最大化算法解决模型参数的估计问题,得到了模型参数的迭代估计算法,并用数值算例验证了该建模方法的有效性。研究表明,鲁棒软测量方法实现了对异常点的自动检测与排除,解决了基于单个模型的传统软测量方法无法得到全局有效模型的问题。其适用于非线性的、具有多个工作模态的实际系统。This paper is an effort to address outliers and nonlinearities in the output data of the system and a proposes a novel multiple model based robust soft sensor modeling method, assuming that the process noise follows the heavy-tailed Laplace distribution instead of the conventional Gaussian distribution where output data contains.The study provides the iterative parameter estimation algorithm by solving the model parameter estimation problems using expectation-maximization algorithm scheme and verifies the effectiveness of the proposed algorithm using he numerical example. The research shows that the robust soft sensor method capable of an automatic detection and elimination of abnormal points could provide an improved alternative to the conventional soft sensor method based on a single model which fails to obtain a global effective model.The method could work better for nonlinear practical systems with multiple working modes.

关 键 词：多模型方法 软测量 异常点 拉普拉斯分布 期望最大化算法 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]