RECURSIVE ROBUST KINETICS ESTIMATION BY USING A MECHANISTIC SHORT-CUTTECHNIQUE AND A PATTERN-RECOGNITION PROCEDURE

In the simulation of complex reactions, the use of mechanistic based k inetic models (KM), although requiring extensive experimental and comp utational effort, presents the advantage of increased prediction relia bility and physically meaningful estimated parameters. Because rapid o ff/on-line process simulation and optimisation usually require reduced KM-s, repeated parameter and model structure adaptations are necessar y. Recently, Maria and Rippin (1995a,b) and Maria (1995) proposed a re liable short-cut technique (MIP, the Modified Integral transformation Procedure) for rapid model identification and approximate parameter es timation. The MIP only implies rapid algebraic manipulations and does not present any convergence problems. Supplementary elements of reacti on path recognition (similarity analysis, problem decomposition, alter native path discrimination, transfer of information rules), and model term-by-term sensitivity and estimate analysis, make the MIP solution more robust and of considerable improved quality compared with the cla ssical direct estimation procedures. The procedure is very suitable fo r non-linear and ill-conditioned cases, being less sensitive to the no ise level, outlier presence, or data and model degeneracy. The MIP, in tegrated in an expert system for kinetic modelling, allows a rapid kin etic data-bank check for suitable KM selection and adaptation to the n ew considered data. The MIP could also be used as a recursive paramete r estimator by transferring previous information about the current pro cess, without use of tuning factors or model linearizations during the identification rule. In this paper some completions to the MIP method are presented in order to improve the initial step when little prior information about the process is available: i) fast identification of a similar KM structure in the data-bank and discrimination among exten ded or reduced reaction path schema; ii) initial use of other direct e stimation techniques and few data from the process to generate rough p rior KM estimates; iii) initial use of non-linear regression (NLS) ste ps and few data from the process to initiate the MIP recursive estimat ion.