Data-driven predictive control for the industrial processes with multiphase and transition

Processes with multiphase are commonly found in process industries. Process dynamics and correlations among variables tend to change with the transitions across such phases. In this paper, we propose a new data-driven predictive control strategy with the consideration of the important multiphase feature. The method aims to feature the multiphase data and use the data to design the controller. First, the data is divided and weighted based on the multiple phases and transitions. Through the minimal image representation, the data-driven prediction of future trajectory can be obtained and thus the computation of dynamic optimization. In the proposed controller, data Hankel matrices is direct incorporated in the predictive control laws, without a model or an intermediate step to meet the given performance specifications. Finally, the proposed predictive controller is demonstrated on a multiphase process.