Multi-Model MPC for a Complex pH Neutralization Process

Control of feed pH control is an important task in chemical industries, wastewater treatment, polymerization reactions, fatty acid production and biochemical processes. However, this is a challenging problem due to (i) presence of large and varying time delays and (ii) complex time varying nonlinear relation exhibited  by the titration curves. Several advanced control algorithms like multi-model predictive control and neuro-controllers have been used to address the pH control problem. However, the corresponding drawbacks in these approaches are: (i) model-plant mismatch (plant dynamics constantly varies) and (ii) computational complexity involved in designing of a single neuro-controller. In this work, a novel genetic algorithm based neural multi-model predictive controller is developed to control the highly nonlinear and challenging pH control problem. The advantages of the proposed approach are (i) multi-model predictive control approach reduces the computational burden involved in designing a single nonlinear controller and (ii) neural network model can provide better approximation of the process over various regions compared to the linear models and (iii) genetic algorithm, a global optimization approach used to obtain the number of hiden nodes and connecting weights of the multiple neural models helps in reducing the model-plant mismatch that could occur in various regimes. The proposed approach is demonstrated on a wastewater neutralization process and a comparison of results with other commonly used approaches is provided. Simulations studies illustrate the practicality and utility of the proposed scheme to control nonlinear processes.