Comparisons between classical and neural schemes of process control: applications to a continuous reactor

In this paper we use feedforward neural nets to implement some control schemes on a continuous stirred-tank reactor (CSTR), in order to compare their behaviour with that of a classical PI controller. A first order exothermical reaction takes place in the reactor vessel of the pilot plant, and the thermal condition of the external jacket is used to regulate the temperature of the reaction, The plant is accurately modelled in order to carry out the different neural control experiments simulating the plant and the controllers. White or colored noise is imposed on the process outputs during simulations so as to emulate real industrial situations. This will allow to test the effectiveness and robustness of neural networks for control of these types of reactors in noisy environments.