Neuro-predictive process control using on-line controller adaptation

The objective of this paper is to propose a technique of integrating neural networks with conventional controller structures, for the predictive control of complex process systems. In the developed method, a baseline conventional controller, e.g. a Proportional-Integral (PI) controller, is used to control the process. In addition, a recurrent neural network is used in the form of a multi-step-ahead predictor (MSP) to model the process dynamics. Utilizing the MSP capabilities of recurrent neural networks, the parameters of the conventional controller can be tuned by a backpropagation-like approach, to achieve acceptable regulation and stabilization of the controlled process variables. The advantage of such a formulation is the effective online adaptation of the controller parameters while the process is in operation, and the tracking of the different operating regimes and variations in process characteristics. The developed method is applied for the stabilization and transient control of U-Tube Steam Generator (UTSG) water level. Currently utilized constant-gain PI controllers are unable to stabilize the UTSG at low operating powers, resorting in manual control. A significant number of plant shutdowns results from such manual control operation. The proposed predictive controller is able to stabilize the process and improve its performance over its entire operating range.