IMPLEMENTING ARTIFICIAL NEURAL NETWORKS MODELLING AFTER THE TREATMENT OF OIL REFINERY EFFLUENTS USING ADVANCED OXIDATION PROCESSES

Abstract

The present study aims at evaluating the treatment of polycyclic aromatic hydrocarbons (PAH) present in oil refinery effluents by advanced oxidation process (AOP), besides analysing the data obtained using artificial neural networks (ANN). The AOP process managed to degrade 10 different PAH initially found in the samples analysed. The efficiency analysis of the process was also evaluated, according to the amounts of total organic carbon (TOC). The ANN Multilayer Perceptron used consisted of 3 layers. Experimental and simulated data used in the training were compared in both trial and validation processes concluding that the amounts were very similar. The network used was able to monitor precisely the tendency of the data and the amounts of TOC, observing the correlation coefficient on both modelling strategies employed. The values of R2 were 0.994 in the first modelling, using the activation function logsig, and 0.996 in the second one, using tansig. Both modelings used the training algorithm Levenberg- Marquardt, corroborating the efficiency of the process employed.