A broad network of subject integration: Based on slowness feature enhancement

Recently, the Broad Learning System (BLS) has been widely applied to process monitoring, achieved impressive performance. Since its own incremental property, BLS can efficiently train and update model while new data arriving. However, BLS and its variants ignored the time dynamics when monitored the process. To tackle the above problems, various methods have been proposed by researches to obtain the hidden time correlation in data. Slow feature analysis is an unsupervised algorithm, which is used to learn the time correlation representation of process monitoring. In this paper, a method considering the process characteristics and time dynamics of batch process is proposed. For one, the slowly changed features are obtained through the constructed feature extraction model. For other, effectively deal with the non-Gaussian in the data, built the global model to monitor the whole process. The performance superiority of the proposed method is verified on the penicillin simulation platform.