IDENTIFICATION OF THE TEMPERATURE PRODUCT QUALITY RELATIONSHIP IN A MULTICOMPONENT DISTILLATION COLUMN

In this paper, we show that in finding a mathematical expression to pr edict the relationship between temperatures measured inside a multi-co mponent distillation column and the quality of the produced product at the top of the column, the application of a recently developed system atic procedure to identify Wiener nonlinear systems [20], supports the user in retrieving from the data accurate information about both the structure and initial parameter estimates of the model to be identifie d with iterative parameter optimization methods. This property enables the user to improve his prior knowledge instead of being dependent on it for getting parameter estimates as is the case in most existing pa rametric identification methods. A consequence of this dependency is t hat wrong prior information leads to models with poor prediction capab ility on one hand, and very little information on the other hand on ho w to modify the model structure in order to get improved results. The latter often results in very time consuming ''trial-and-error'' approa ches that furthermore may yield poor results because of the possibilit y of getting stuck in local minima. The outlined approach has the pote ntial to overcome these drawbacks. One common source of the use of wro ng prior knowledge in the identification of multicomponent distillatio n columns is the presence of a static nonlinearity of exponential type that can be removed by taking the logarithm of the measured product q uality. It is shown that this ''trick'' to linearize the system decrea ses the accuracy of the predicted producted quality. The outlined appr oach is also compared to a simple NARX neural network black-box identi fication method that have the potential to approximate general nonline ar input-output behaviours. This comparison shows that the neural netw ork approach easily requires twice as much observations compared to th e Wiener identification approach applied in this paper when the varian ce of the predicted product quality needs to be the same. The real-lif e measurement used in this paper were collected at a refinery of the D utch State Mines (DSM). Finally, in order to use the model obtained wi th one (training) data set under other operational conditions, that is to extrapolate the model a simple observer design is discussed and va lidated with real-life measurements.