Data-driven-based model predictive control of reagents addition for tungsten flotation process

In the actual froth flotation process, the complexity of the mechanistic model and the frequent fluctuations of production conditions make the design of control strategies for flotation reagents very difficult. Therefore, this paper proposes a data-driven-based Model Predictive Control (MPC) strategy of reagents addition for tungsten flotation process, which includes a flotation dynamic model based on Gate Recurrent Unit(GRU) network and an optimizer. Firstly, for the flotation process variables with time-series characteristics, a GRU network model is designed to achieve accurate prediction of concentrate grade. Then based on the principles of maximizing the tracking performance of the controller and minimizing the fluctuation of the control signal, a cost function is proposed to form an optimizer combined the constraints of reagents dose. Finally, the output of the optimizer is used to guide the flotation reagents addition process so that the concentrate grade tracks the set value. Experimental results show that the proposed GRU-MPC control strategy has good control performance.