Neural network based predictor for control of cascaded thermal process

Process control has been by far the most popular area of neural network application in chemical engineering. The use of neural networks in this area offers a potentially effective means of handling three difficult problems: complexity, nonlinearity and uncertainty. In comparison with other empirical models, neural networks are relatively less sensitive to noise and incomplete information, and thus deal with higher levels of uncertainty when applied in process control problems. The artificial neural network (ANN) technique is applied to the simulation of the time-dependent behaviour of a cascaded thermal process (heat exchangers) and is used to control the temperature of hot fluid flowing in the inner tube of the second heat exchanger. Two neuro predictor control schemes are presented. In the first scheme, a methodology is proposed for training and prediction of dynamic behavior of heat exchanger using feed forward neural network with external recurrent connections. Then a non-linear predictive control strategy based on identified model is proposed for heat exchanger control. In the second scheme, neural network based predictor for predicting outputs of first and second stages is presented. The performances of neuro schemes are evaluated by simulation for both servo and regulatory problems and the results are compared with a PID controller. An approach to the neural network-based predictor for control of cascaded thermal process is presented for predicting the effect of major load disturbance on the controlled variable in advance. First a mathematical model is developed in a traditional way, followed by the implementation of a neural network multilayer predictor trained based on mathematical model. The dynamic and steadystate behaviors are investigated.