The use of neural networks for fitting complex kinetic data

Abstract

Abstract-In this paper the use of neural networks for fitting complex kinetic data is discussed. To assess the validity of the approach two different neural network architectures are compared with the traditional kinetic identification methods for two cases: the homogeneous esterification reaction between propionic anhydride and 2-butanol. catalysed by sulphuric acid. and tbe heterogeneous Iiquid-liquid toluene mononitration by mixed acid. A large set of experimental data obtained by adiabatic and heat flux calorirnetry and by gas chromatography is used for the training of the neural networks. The results indicate tbat tbe neural network approach can be used to deal with tbe fitting of complex kinetic data to obtain an approximate reaction rale function in a Iimited amount of time. which can be used for design improvement or optimisation when. owing to small production levels or time constraints. it is not possible to develop a detailed kinetic anaIysis. Copyight

Simultaneous influence of gas mixture composition and process temperature on Fe2O3->FeO reduction kinetics: neural network modeling

“…After the learning procedure has been finished the matrix of optimum weight's values is preparedaccumulating hidden, complex information about the process under investigation, thus immediately ready for the practical problems solving. The successful trials of the ANN application in chemical and process engineering problems [e.g., Piotrowski, K. et al (2003), Galván et al (1996), Abilov and Zeybek (2000)] have shown that simple ANNs are capable of complex nonlinear systems modeling, the properly selected and trained Network enables one to interpolate and, in restricted range, extrapolate the data and confirmed that the Networks enable one to perform the complex design calculations quickly and effectively.…”

Kinetic hard-modelling and spectral validation of rank-deficient spectroscopic data: A case study

“…The kinetic mechanism of this reaction has been discussed in previous articles [21,22] and is used here as a test reaction to experimentally validate the theoretical concepts presented previously [17]. Strategies (1) to (4) are applied to this reaction in order to break the rank deficiency in the concentration matrix without distorting the calculated rate constants. As a consequence of Strategy (1), linear dependencies in the concentration profiles translate into the fitted component spectra, and are compared to those theoretically predicted by our method.…”

“…Various chemometric methods capable to analyse time dependent multivariate data measured by spectroscopic techniques have been introduced in recent years [1][2][3][4][5][6][7][8][9][10]. Amongst these chemometric methods, kinetic hard-modelling, based on a hard model (the rate law), can be used to directly determine the kinetic parameters (e.g.…”

“…In a recent article [17], we have introduced a method for the systematic experimental and data analytical design of bi-linear spectroscopic kinetic measurements that allows identifying the species to be incorporated in strategies (1) to (4) to obtain partial or full spectral resolution. When only partial resolution is possible (Strategy 1), we have also presented a method for the calculation of the linear combinations in the fitted component spectra, i.e.…”

Systematic prediction of linear dependencies in the concentration profiles and implications on the kinetic hard-modelling of spectroscopic data

“…as in kinetic and equilibrium investigations) measured by spectroscopy (e.g. mid-IR, UV-Vis, Raman or fluorescence) have considerably progressed in chemistry and chemical engineering [1][2][3][4][5][6][7][8][9][10]. Kinetic hard-modelling is one of these chemometric methods.…”