Reconstruction of dynamics of aqueous phenols and their products formation in ozonation using differential neural network observers

A differential neural network (DNN) is used to estimate the state dynamics in the phenols-ozone-water system of the model solution of phenol (PH), 4-chlorophenol (4-CPH), and 2,4-dichlorophenol (2,4-DCPH). This new technique, which is based on a differential neural network observer (DNNO), is applied to estimate decomposition dynamics of phenols, byproducts accumulation and decomposition, and final products accumulation. It is considered to be a process with an uncertain model ("black-box") and is affected by internal (variation of experimental conditions) and external perturbations (measurement noises). The monitored ozone concentration in the gas phase in the reactor outlet is used to obtain the summary characteristic curve (ozonogram) in ozonation. The proposed DNNO is trained using the variation of this parameter with the experimental data of the phenols decomposition, obtained using the HPLC technique, at pH 2 and pH 9. The trained DNNO then is applied to reconstruct the dynamics of the phenols decomposition, as well as the byproducts accumulation and the decomposition and the final product accumulation at pH 7 and pH 12. The proposed DNNO technique has been tested to compare estimated results to those experimentally obtained during semibatch ozonation of the model solution of phenols. A good correspondence between the experimental decomposition dynamics and those estimated by DNNO was obtained.