Boosting Weighted Partial Least Squares for Batch Process Quality Prediction

In batch processes, end-product qualities are cumulatively determined by variable dynamic trajectories throughout each batch. Meanwhile, batch processes are inherently time-varying, implying that process variables may have different impacts on end-qualities at different time intervals. To take both the cumulative and the time-varying effects into better consideration for quality prediction, a boosting weighted partial least squares method is proposed. Process variables at each time interval are automatically weighted according to their contributions to quality, while the boosting technique is adopted to further improve the predictions. Application results show the advantages of the proposed method comparing to conventional multivariate statistical models.