论文部分内容阅读
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 process of off-line modeling of process monitoring, Gaussian Mixture Model (GMM) can not accurately characterize multi-modal data when the training data set contains outliers. In order to solve the problem that GMM is easily affected by outliers, this paper presents a process monitoring method for Lo OP-GMM. Firstly, Local Outlier Probability (Lo OP) algorithm is used to detect and remove outliers from the training dataset during the data preprocessing stage. The GMM algorithm is used to establish the offline model. At the same time, the training data set is clustered according to posterior probability class. Secondly, considering the outlier probability of online samples, a new global probability index is constructed as a statistic and used for multi-modal process fault monitoring. Finally, the effectiveness of the proposed method is verified by numerical simulation and Continuous Stirred Tank Reactor (CSTR) process.