基于DHSC的多模态间歇过程测量数据异常检测方法
发布时间：2018-11-24 20:15
【摘要】：多模态间歇过程测量数据异常直接影响数据驱动的多元统计分析过程建模的准确性,导致间歇过程的监控性能降低。针对多模态间歇过程测量数据异常问题,提出了一种基于动态超球结构变化(DHSC)的多模态间歇过程测量数据异常检测方法。该方法通过引入时序约束的模糊C均值聚类(SCFCM),利用隶属度变化划分多模态间歇过程的模态;针对不同模态,采用支持向量数据描述(SVDD)建立基于训练数据的静态超球体和基于待检数据的动态超球体,选择重要的支持向量作为球体结构,进而通过识别超球体发生结构变化实现过程测量数据异常检测。青霉素发酵过程仿真实验表明,所提出的方法能够实现多模态间歇过程的模态划分,减少了模态切换对过程测量数据异常检测精度的影响,并能够根据超球体结构变化检测过程测量数据异常,具有较高的检测精度,降低了误检率。
[Abstract]:The outliers of measurement data in multimodal batch processes directly affect the accuracy of data-driven multivariate statistical analysis process modeling and result in the deterioration of monitoring performance of batch processes. In order to solve the problem of outliers in multimodal batch process measurement, a method of detecting outliers in multimodal batch process measurement based on dynamic hypersphere structure (DHSC) is proposed. In this method, the fuzzy C-means clustering (SCFCM), with time series constraints is introduced to divide the multimodal intermittent processes into modes by using the variation of membership degree. For different modes, support vector data is used to describe (SVDD) to establish static hypersphere based on training data and dynamic hypersphere based on data to be checked. The important support vector is chosen as sphere structure. Then the abnormal detection of process measurement data is realized by recognizing the structural changes of hypersphere. The simulation results of penicillin fermentation process show that the proposed method can realize modal partitioning of multimodal batch process and reduce the influence of modal switching on the accuracy of abnormal detection of process measurement data. It can measure the abnormal data according to the change of the hypersphere structure, and has higher detection accuracy and lower false detection rate.
【作者单位】： 北京化工大学信息科学与技术学院;
【基金】：国家自然科学基金项目(61240047) 北京市自然科学基金项目(4152041)~~
【分类号】：TQ06
[Abstract]:The outliers of measurement data in multimodal batch processes directly affect the accuracy of data-driven multivariate statistical analysis process modeling and result in the deterioration of monitoring performance of batch processes. In order to solve the problem of outliers in multimodal batch process measurement, a method of detecting outliers in multimodal batch process measurement based on dynamic hypersphere structure (DHSC) is proposed. In this method, the fuzzy C-means clustering (SCFCM), with time series constraints is introduced to divide the multimodal intermittent processes into modes by using the variation of membership degree. For different modes, support vector data is used to describe (SVDD) to establish static hypersphere based on training data and dynamic hypersphere based on data to be checked. The important support vector is chosen as sphere structure. Then the abnormal detection of process measurement data is realized by recognizing the structural changes of hypersphere. The simulation results of penicillin fermentation process show that the proposed method can realize modal partitioning of multimodal batch process and reduce the influence of modal switching on the accuracy of abnormal detection of process measurement data. It can measure the abnormal data according to the change of the hypersphere structure, and has higher detection accuracy and lower false detection rate.
【作者单位】： 北京化工大学信息科学与技术学院;
【基金】：国家自然科学基金项目(61240047) 北京市自然科学基金项目(4152041)~~
【分类号】：TQ06
【相似文献】
相关期刊论文 前10条
1 陈嘉祺;间歇过程中釜的安排[J];化工设计;1997年01期
2 李志红,宋子明,袁志敏;间歇过程能量平衡和，
本文编号：2354903
本文链接：https://www.wllwen.com/kejilunwen/huaxuehuagong/2354903.html
