Feature-based Data Alignment of Multi-stage Batch Processes and Its Application to Optimization

摘要

To solve the unequal length problem of batch data for multi-stage batch processes, a moving window data alignment method based on Warped K-means (WKM) in latent space is proposed in this paper. Firstly, the Warped K-means is used for stage division of batch processes. To capture the data inherent characteristics, principal component analysis (PCA) is utilized to project the original data into low dimensional latent space. Then, in latent space, searching the points closest to reference trajectory in the moving window is carried out to realize the feature-based data alignment. Considering the practical cases that missing data exist in industrial processes, the data complementation is performed by linear or non-linear interpolation according to the trend of the trajectories. Eventually, the aligned batch data are applied to an optimization algorithm of batch processes to verify the validity of the proposed method.