Estimation of Contaminants Decomposition in Solid Phase with Ozone by Differential Neural Networks with Discontinuous Learning Law

Abstract

A discontinuous learning law is implemented here to adjust an adaptive non-parametric identifier, based on the differential neural networks (DNNs) approximations. The learning law for DNN uses the vector form of an extended super-twisting algorithm as the output injection term in the DNN structure. The learning laws with discontinuous dynamics have been obtained from the application of a special class of strong lower semi-continuous Lyapunov function. The developed observer was tested on both modelled and experimental input-output information on the specific the ozonation process of a contaminated solid phase. A numerical example illustrates the observer performance when the input-output information is free of the observation noise. The observer has been evaluated using real experimental data, obtained by the direct laboratory analysis. In both cases, modelling and real experiments, the coincidence between the ozonation variables and the estimated states shows a remarkable correspondence.