摘要
在开展过程监控的离线建模的工作中,当训练数据集含有离群点时,高斯混合模型(Gaussian Mixture Model,GMM)不能准确刻画多模态数据特征。为解决GMM易受离群点影响的问题,本文提出了Lo OP-GMM的过程监控方法。首先,用局部离群概率(Local Outlier Probability,Lo OP)算法在数据预处理阶段检测并剔除训练数据集中的离群点,并用GMM算法建立离线模型,同时根据后验概率将训练数据集进行聚类。其次,考虑到在线样本的离群概率,构造一个新的全局概率指标作为统计量并用于多模态过程故障监控。最后,通过数值仿真和连续搅拌釜反应器(Continuous Stirred Tank Reactor,CSTR)过程验证了本文所提方法的有效性。
In the work of offline modeling for process monitoring, the process data characteristics of multimode are not described accurately by a Gaussian Mixture Model （GMM） when training data set contain outliers. In order to solve the problem of GMM is extremely susceptible to outlier observations, a novel multimode process monitoring approach based on LoOP-GMM algorithm is proposed. Firstly, Local Outlier Probability （LOOP） algorithm is applied to detect and extract outliers from the training data set that contain outliers in data preprocessing step, then the offline model of multimode process is obtained by GMM, meanwhile, training data set are clustered based on the posterior probability. In addition, taking into account the outlier probability of online sample, a new global probability index is established as a monitoring statistic for multimode failure monitoring. Lastly, the efficiency of the proposed method is demonstrated through a numerical example and Continuous Stirred Tank Reactor （CSTR） process.
作者
许圆圆
宋冰
谭帅
侍洪波
Xu Yuanyuan Song Bing Tan Shuai Shi Hongbo(School of Information Science and Engineering, East China University of Science and Technology, Shanghai 200237, Chin)
出处
《计算机与应用化学》
CAS
2017年第2期128-134,共7页
Computers and Applied Chemistry
基金
国家自然科学基金(61374140)
国家自然科学基金青年基金(61403072)
关键词
过程监控
离群点
高斯混合模型
多模态
全局概率指标
process monitoring
outlier
Gaussian Mixture Model
multimode
global probability index
