Application of Artificial Intelligence Techniques for Temperature Prediction in a Polymerization Process

The main feature of the polymerization reaction is its complex nonlinear behaviour, which poses a challenging control system design for the batch reactor. The present work is concerned with the development of intelligent mathematical models to predict the styrene polymerization temperature. In order to improve the final product quality, these models will be used in predictive control schemes. Two techniques from the artificial intelligence field were used: Neuro-fuzzy and artificial neural networks. The pilot plant of styrene production consisted of: a stainless steel jacketed stirred reactor, a storage tank and a variable speed pump for the thermal fluid, temperature sensors (inside reactor, inlet and outlet of the jacket), a densimeter, and a PLC (Programmable Logic Controller). The temperature of the reactor is the process variable to be predicted using the historical data acquired from the pilot plant. Software MatLab 6.0 was used to implement neural and neuro-fuzzy models. The results showed that both models were able to predict online the reactor temperature profile successfully and that they were fast enough to be used in nonlinear predictive control strategies as well.