A Learning-Based Robust Model Prediction Control for Industrial Thickener

This article proposes a deep learning-based robust model predictive control (RMPC) method for regulating the underflow concentration of industrial thickeners. Safely controlling of underflow concentration in thickening processes remains challenging due to complex operational uncertainties. To address this issue, a learning - based tube model is developed to characterize uncertainty propagation mechanisms employing a quantile regression model for probabilistic uncertainty quantification and a classifier-guided adaptive asymmetric tube mechanism to reduce conservatism in conventional approaches. Based on this, an RMPC strategy is designed specifically for thickening processes to effectively track underflow concentration while ensuring system safety. Simulation results demonstrate that the proposed method enhances stability while maintaining operational safety guarantees.