Neural modeling and statistical analysis of the degradation process of leachate by the solar photo-Fenton process

Abstract

This study sought to obtain a neural model and statistical analyzes that represented the degradation of leachate in natura by the solar photo-Fenton process, according to the reduction of chemical oxygen demand (COD). The study used leachate from the town of Cachoeira Paulista-SP, which had low biodegradability and required pre-treatment by an oxidative process. Neural networks are presented as an alternative for the modeling of nonlinear processes such as advanced oxidation processes, which involve a large number of control variables and complex reactions. The photo-catalytic process was optimized by a fractional (24-1) factorial design in duplicate and triplicate, with the center point being the input variables at three levels: pH, solar radiation and concentrations of H2O2 and Fe2+. The treatment system used an open reactor in bench scale with a constant volume (3 L) of leachate flow of 13 L min-1 and a 2-h reaction. The optimization process showed that the individual effect of each input variable should operate at its highest level, and that the variable Fe2 + was significant for the combination. The percentage reduction of COD of the best experiment was 88.7%, which is valued at a cost of R$126.67 m-3. The degradation process was modeled via feedforward backpropagation neural networks with linear correlation coefficients for the training sets, validation and test above 0.9, indicating high prediction and generalization of the proposed neural model.