Optimal Control of Flotation Reagents Via Sos and Reinforcement Learning

Flotation reagent optimal control is critical for enhancing metal recovery while minimizing reagent consumption. However, model uncertainties complicate the control of flotation reagents. This paper presents a novel data-driven control strategy based on the sum of squares and reinforcement learning (SOSRL), which relies solely on measurable data, bypassing the need for precise system dynamics. The flotation reagent control problem is reformulated as a polynomial optimization task, with the value and Hamiltonian functions approximated by polynomial vectors. Through iterative optimization, the control policy is evaluated and improved. The proposed method ensures global stability while reducing computational complexity. Numerical simulations validate the effectiveness of the SOS-RL based control strategy, highlighting its potential to optimize flotation performance in practical applications.