DIRECT AND INDIRECT MODEL BASED CONTROL USING ARTIFICIAL NEURAL NETWORKS

The use of artificial neural networks in model based control, both as process models and as controllers, is investigated: two nonlinear model based control strategies, internal model control (IMC) and multistep predictive control (MPC), are applied to the control of a nonlinear SISO exothermic CSTR. Direct inverse control in the IMC framework was found to require appropriate use of feedback for adequate performance. An IMC-type neural network controller in which the process model was replaced by a nonlinear neural network and inverted on-line to calculate the control action gave very good performance, even when only partial state data were available. An MPC-type neural controller using the same neural network model and extended to include feedback also gave excellent performance. Performance was significantly better for both control techniques when a nonlinear network was used as a process model than when a linear ARMAX model was used. These results indicate that neural networks can learn sufficiently accurate models and give good nonlinear control when model equations are not known or only partial state information is available.