Abstract
Prediction system for normal values of state variables can be classified into two approaches; physical model and black-box model. A physical model is constructed by mass balance and energy balance etc. It is difficult to develop the model and to identify the model parameter because it requires perfect knowledge about the objective system. This is reason why the black-box model is often employed as predictor in chemical processes. It is, however, difficult to obtain complete learning data from historical data during normal operations in order to apply these methods to real chemical processes. So, a simple model with the use of database accumulating a large amount of historical data during normal operations has been proposed and has been called "DateBase (DB) model". DB model predicts present state values using the some past similar data. In previous study, DB model has been examined using a simple continuous process with load fluctuations. And it has been proved that DB model could be used to predict normal values of state variables. But we needed to confirm whether DB model could predict state values of larger scale complex and unsteady-state processes or not. So, we applied DB model to Poly Vinyl Chloride (PVC) batch process as an example of such a process. To get a large amount of historical data during normal operations, the PVC process was constructed on a dynamic process simulator.
