Optimal Control of Flotation Industrial Process Using Model-based Reinforcement Learning

In this paper, the optimal control of the flotation industrial process (FIP) is studied. The flotation process uses differences in the physical and chemical properties of mineral surfaces to selectively attach minerals to air bubbles, and separate useful from useless minerals. To optimize control of the process, we use the model-based reinforcement learning (MBRL) method to design the optimal controller for the flotation process. A case study on the flotation mechanism model verifies the efficiency of the proposed method. The results show that the MBRL method can learn the optimal control policy with fewer episodes.