Role of modelling on state and parameter estimation

In process industry, plants are generally operated at conditions that differ from the designed ones mainly due to disturbances. Disturbances can enter the system in form of fluctuations in feed flow, temperature and composition, or fluctuation of the utilities quality. These events cause a deterioration of the plant performance that cannot be quantified and online compensated by means of controllers unless online measurements of the quality targets (e.g. concentration, conversion, etc) are available. However the problem of online monitoring cannot be always solved in practice by means of hardware analysers because of unreliable and delayed measurements. An alternative approach is based on estimators that infer the variables of interests by means of secondary measurements and a often nonlinear model of the
process. This type of realization of observers can include the online estimation of model parameters for a more accurate alignment of the model with the process behaviour.
This work addresses the role of the estimation model on estimation performance. Recent studies [1, 2] pointed out that for a defined set of plant measurements the choice of the estimation model and the innovated states play a key role on the performance of the estimator regardless the algorithm employed. Even if in the cited studies some features of the estimation model (such as level of detail, computational complexity) have been taken into account, the effect on the estimation performance of model manipulations such as variables and parameters scaling [3] and transformation have not been investigated yet. For this reason the role of different realizations of the same estimation model needs to be further investigated.