Distributed nonlinear model predictive control for cobalt removal process in zinc hydrometallurgy considering error compensation modelling

Abstract

AbstractTo address the strong nonlinearity, uncertainty, and mutual coupling in the cobalt removal process of zinc hydrometallurgy, an algorithm based on an improved genetic algorithm (GA) backpropagation (BP) neural network combined with distributed nonlinear model predictive control (NMPC) is proposed. This method was applied to improve the quality of the purification solution and reduce the consumption of zinc powder, overcoming the challenges faced by the current cobalt removal process. First, a synergistic continuously stirred tank reactor (SCSTR) model was constructed for the dynamic cobalt removal process. Second, aiming at the problem that a single SCSTR model has difficulty describing the process accurately, based on the highly nonlinear mapping ability of data‐driven models, a method that organically integrates the SCSTR model and an error compensation model based on the GA‐BP neural network was proposed (GA‐BP‐SCSTR) to provide a more accurate online prediction of the process indicators. Then, a distributed NMPC architecture was developed using the GA‐BP‐SCSTR model, control vector parameterization (CVP) technique, and sequential quadratic programming (SQP) algorithm to achieve the coordinated control of the cobalt removal process. Finally, simulation results of an actual site showed that the prediction accuracy of the GA‐BP‐SCSTR model was higher than those of other models. The proposed predictive control method can maintain the outlet cobalt ion concentrations at the set values while achieving accurate control of the zinc powder addition. This approach can provide guidance for on‐site production and eliminate the blindness of manual experience control.