Naslov
Neural Network Based Feedforward Control Of Two-Stage Heat Exchange Process

Autori
Blažina, Alfred ; Bolf, Nenad

Vrsta, podvrsta i kategorija rada
Radovi u zbornicima skupova, cjeloviti rad (in extenso), znanstveni

Izvornik
1997 IEEE International Conference on Systems, Man And Cybernetics / SMC - : Institute of Electrical and Electronics Engineers (IEEE), 1997, 25-29

Skup
1997 IEEE International Conference on Systems, Man, and Cybernetics

Mjesto i datum
Orlando (FL), Sjedinjene Američke Države, 12.10.1997. - 15.10.1997

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
neural networks; process control; heat excange; feedforward control

Sažetak
An approach to the neural network based feedforward control of a two-stage heat exchange process is presented. A warm liquid with a low boiling point is fed into a distillation column where precise maintenance of the temperature of the fed liquid close to boiling point is required. Therefore, in the first stage, the liquid is heated up to a temperature close to the desired output value, and in the second stage the final value is obtained within a permitted range of only 1°C above the set-point. The first stage of the process takes place in a classical thermostat, while the second stage is conducted in a boiler-shaped tank heat exchanger. When the pressure change occurs in the distillation column, the disturbance causes a variation of material balance. This disturbance is manifested in both stages, producing undesirable material and heat imbalance. The feedforward control is conceived to protect the process against such imbalance. The process model that represents the dependence T(t)= L[p(t)] is applied in the feedforward control and allows us to protect the limits from being exceeded. The task is solved first by developing a mathematical model T(t)= L[p(t)] in a traditional way, and is followed by the implementation of a multilayer feedforward neural network as the basis of a model generator in the feedforward controller. A moving window neural network architecture is used to model the process in a dynamic and unsteady state. Both direct and inverse neural control networks are investigated. The results of the application of the developed neural network based feedforward control are graphically shown and critically examined. Particular attention is given to the application of the developed method in the training of students of chemical engineering.

Izvorni jezik
Engleski

Znanstvena područja
Kemijsko inženjerstvo

POVEZANOST RADA