Optimised multivariable nonlinear predictive control for coupled tank applications

This presentation describes the design of a novel nonlinear model predictive control (NMPC) strategy using a stochastic genetic algorithm (GA) to control highly nonlinear, uncertain and complex multivariable process with significant cross-coupling effects between the process input and output variables. Raw multi-input, multi-output (MIMO) data from an experimental setup were collected and analysed. Both a GA and a backpropagation gradient-descent-based approach known as the Levenberg-Marquardt algorithm (LMA) are employed to train an artificial neural network (ANN) nonlinear model. Real-time practical experimental implementation on a MIMO coupled tank system is performed, and the results show the effectiveness of the strategy. The approach can easily be adapted to other industrial processes.