Multimodal Industrial Process Monitoring Method Based On High-Dimensional Time Series Analysis

Multimodal industrial process is one of the main production mode of modern biological industry, multimodal dynamic characteristic of the industrial process and differences between batches bring difficulties to monitor, the effective monitoring method can improve the quality of the fermentation process. Existing monitoring method based on multivariate statistical analysis, without considering the non-stationary control input variables, will increase the rate of false positives. This topic in view of the influence of the control input, proposes a method combined Wavelet Transform, Wavelet Transform, WT) method, the Kernel Principal Component Analysis (Kernel Principal Component Analysis, KPCA) method and Fuzzy c-means (Fuzzy C-Means, FCM method, called WT-KPCA, and is applied to the segmentation monitoring of batch process, can monitor the effect of control input, which can be removed to improve monitoring of the effect. This topic proposes a batch process modal clustering FCM method based on WT-KPCA. The method directly looks at the system state through the system response function to determine whether a process is smooth, compared to the batch process modal clustering method based on MKPCA, WT-KPCA method focus on the system dynamic characteristics. This topic proposes a joint weighted SVM modeling method based on the segmentation results of batch process, the method considers the effect of batch process as a whole, can effectively promote the model precision.On-line monitoring method uses CKF to predict data, calculate online monitoring index. Study the method through Pensim simulation platform, the batch process modal clustering FCM method based on WT-KPCA is effective, the joint weighted SVM modeling method based on the segmentation results of batch process works well, on-line monitoring method can effective monitoring of fermentation of failure.