Neurofuzzy based temperature prediction of an industrial polymerization reactor in real time

The ultimate purpose of this work is the real-time control of a semibatch emulsion polymerization reaction. The main variable under control is the temperature inside the reactor. The keypoint to get accurate control is the early detection of temperature deviations. A neurofuzzy network has been trained to predict the temperature from some of the previously calculated variables. This approach aims to extend the prediction horizon with which the temperature is predicted to an order of magnitude of minutes, based on variables which are not measured online but rather estimated using a reduced calorimetric model. This methodology has been applied to data from a real life emulsion polymerization reactor, and in order to allow its implementation in this factory, it has been ensured that all of the operations can be performed in real time by a common Programmable Logic Controller (PLC). The resulting set of equations predicts the temperature several minutes in advance with good accuracy.