Neural networks based model predictive control of the drying process

The paper presents the simulation results of using Neural Networks (NN) for building a statistical model, afterwards used for Nonlinear Model Predictive Control (NLMPC) of the drying process. Important incentives of the NN approach are explored, such as modelling the drying process for which detailed governing rules may be difficult to formalize as first principle models and reducing the computation time in nonlinear model based control. Different control structures are tested, using direct or inferred (NN based observer) measured process variables. Incentives and drawbacks of the different control approaches are outlined and the most favourable control scheme is pointed out. Simulation results reveal clear benefits for the NN based NMPC using the NN based observer approach, compared with traditional control methods, and prove incentives for industrial implementation.