Neuro-predictive process control using online controller adaptation

A novel architecture for integrating neural networks with industrial controllers is proposed, for use in predictive control of complex process systems. In the proposed method, a conventional PI controller is used to control the process. In addition, a recurrent neural network is used in the form of a multistep-ahead predictor to model the process dynamics. The parameters of the PI controller are tuned by a backpropagation-through-time like approach using "parallel learning" to achieve acceptable regulation and stabilization of the controlled process. 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 process operating regimes and variations. The proposed method is used in the stabilization and transient control of U-tube steam generator water level. The proposed predictive controller stabilizes the process and improves its transient performance over the entire operating range.

I. Introduction

The majority of industrial controllers employ fixed parameter, linear controller structures for controlling complex nonlinear systems [15]. More specifically, an overwhelmingly large fraction of these industrial controllers utilize standard classical structures, such as proportional-integral (PI) control. However, the set of control gains required for desired system performance at different operating conditions might be different. The use of constant-gain controllers is not always desirable because many process parameters drift throughout the system life-time as a result of wear-and-tear. Such behavior suggests the need for on-line adaptation, or self-tuning, of controller parameters to achieve desired process response. Over the past a couple of decades, many methods have been developed for self-tuning of controller parameters, and currently many control vendors offer such solutions [3].