Neural Network Based Nonlinear Model Predictive Control Vs. Linear Quadratic Gaussian Control (f817bb17-c175-4c62-9b79-c5a20af1c18b)

One problem with the application of neural networks to the multivariable control of mineral and extractive processing is deciding whether and how best to use them. The objective was to compare neural network control to more conventional strategies and find any advantages in terms of set point tracking, rise time, settling time, disturbance rejection, and other criteria. The procedure was to develop neural network controllers using both historic plant data and simulation models. We tried various neural network control patterns including both inverse and direct neural network plant models. These were compared to state space controllers that were, by nature, linear. For grinding and leaching circuits, a nonlinear neural network-based model predictive control strategy was superior to a state space-based linear quadratic Gaussian controller. We also learned the importance of incorporating state space into neural networks by making them recurrent--feeding certain output state variables into input nodes in the neural network. We concluded that neural network controllers can have better disturbance rejection, set point tracking, rise time, settling time, lower set point overshoot, and can be more foolproof and easy to implement in complex, multivariable plants.