KERNEL-BASED PREDICTIVE CONTROL OF NONLINEAR MAGNETIC SEPARATION PROCESSES

Анотація

A kernel-based predictive control approach is proposed for nonlinear industrial processes, demonstrated on magnetic separation. The method utilizes data-driven kernel regression to model complex input–output dynamics without requiring physical process equations. A recurrent forecasting scheme is developed to enable multi-step prediction over a defined horizon, even with significant measurement noise and disturbances. Experimental results on real process data show mean absolute forecasting error below 1.6% across all tested scenarios and noise levels. The kernel-based control strategy provides robust, accurate, and computationally efficient performance, allowing for real-time process automation where traditional modeling methods are impractical. Future work will focus on integrating process constraints into the kernel framework and extending applicability to multivariable and large-scale control problems.