Neural Network Self-Tuning Control Of Hot-Spot Temperature In A Fixed-Bed CatalyticReactor

Abstract

This paper demonstrates the possibility of applying artificial neural networks (ANNs) in the self-tuning PID control of the hot-spot-temperature in a fixed-bed catalytic reactor system. In this reactor system sulfur dioxide is oxidized using vanadium pentoxide catalyst. Unlike the conventional self-tuning PID control algorithm, the ANN applied to the self-tuning PID (NNW-PID) philosophy is an inherent nonlinear estimator and therefore identifies a nonlinear system directly from historical data supplied by the plant. In the majority of control applications, the ANN is employed as a predictor of future outputs within an established predictive control algorithm [1,2]. In this paper we propose a scheme where an ANN is employed on-line with a non predictive PID controller to give adaptive control of the reactor. This is accomplished by applying the values of the ANN weights, that are continually updated by the algorithm, to give the relevant PID controller parameters. Since real plant data is used to train the network, the algorithm can be applied to similar real-world problems.