Combining artificial neural networks and experimental design to prediction of kinetic rate constants

Abstract

A “soft-modelling” computational approach of artificial neural networks (ANNs) combined with experimental design (ED) has been applied successfully in Chemical Kinetics for the prediction of kinetic rate constants. The system studied comprises two consecutive first-order reactions and the kinetic data were computed determining the values of both rate constants. The kinetic curves were distributed according to an ED, and the central star composite experimental design (CSCED) was chosen as the most appropriate. Computational treatments were performed on synthetic data endowed with noise, after which they were applied to the data measured in an experimental reaction between carbonyl cyanide 3-clorophenylhydrazone with 2-mercaptoethanol, computing the experimental kinetic data of absorbance acquired at 3 wavelengths. The combined ANN and ED approach applied in chemical kinetics proved to be robust and of general applicability and has the advantage of being a “soft-modelling” method such that it was not necessary to solve the system of ordinary differential equations to determine the explicit mathematical function between the data and the kinetic rate constants. Additionally, upon using the CSCED experimental design, it was possible to substantially reduce the number of experiments.