Neural networks simulation of photo-Fenton degradation of reactive blue 4

Abstract Multivariate experimental design was applied to the degradation of Reactive Blue 4 dye solutions (RB4) in order to evaluate the use of the Fenton reagent under ultraviolet light irradiation. The efficiency of photocatalytic degradation was determined from the analysis of the following parameters: total organic carbon (TOC), color and sulphates and nitrates content. Factorial experimental design allowed to determine the influence of five parameters (initial concentrations of Fe(II), H2O2 and RB4, pH and temperature) on the value of the decoloration and mineralization kinetic rate constants. Experimental data were fitted using neural networks (NNs). The mathematical model reproduces experimental data within 82–86% of confidence and allows the simulation of the process for any value of parameters in the experimental range studied. Also, a measure of the saliency of the input variables was made based upon the connection weights of the neural networks, allowing the analysis of the relative relevance of each variable with respect to the others. Results showed that intermediate compounds compete to react with OH radicals with the different RB4 species present at each pH. Thus, alkaline pHs were found to be more favourable for a faster decoloration process (higher decoloration kinetic constant) whereas acidic pHs were required to remove total organic carbon. The values of kinetic constants obtained from the NNs fittings were implemented in a computational model together with another 24 reactions to simulate the degradation mechanism and assuming a perfectly mixed reactor. The reaction rate equations were built for each chemical specie and the differential equations were solved in Matlab 6.5.