Koopman Operator Learning Control of Dense Medium Separation

For the complex nonlinear characteristics of dense media separation (DMS), this article proposes an efficient data-driven control algorithm based on the Koopman operator. The Koopman framework of global linearized nonlinear dynamics is applied through deep learning technology, and an improved model predictive controller (MPC) based on Koopman is designed to control the DMS process. The controller is composed of an incremental MPC controller and a feedback controller. Moreover, the stability of the closed-loop system is analyzed theoretically. Finally, through the experiment based on operational data from coal washing plant, we establish data-driven model and design the controller for DMS process. More importantly, compared with the nonlinear MPC and Kernel Koopman MPC, our proposed method has higher efficiency.