Application of artificial neural networks in the modeling of photocatalytic reduction of Cr(VI) by titanium dioxide nanoparticles: optimization of artificial neural network structure

Chromium is present in two oxidation forms of Cr(III) and Cr(VI). Cr(III) is less toxic than Cr(VI). The aim of this article was to optimize an artificial neural network structure in modeling the photocatalytic reduction of Cr(VI) by TiO2-P25 nanoparticles.

In this work, an artificial neural network (ANN) for the modeling photocatalytic reduction Cr(VI) by TiO2-P25 nanoparticles were used and its structure was optimized. The operating parameters were initial concentration of chromium, amount of photocatalyst, ultraviolet light irradiation time and pH. All the experiments were conducted in a batch photoreactor. The Cr(VI) concentration was measured with a UV/Vis spectrophotometer. ANN calculations were performed using Matlab 7 software and the ANN toolbox.

The results show that the optimization of the ANN structure and the use of an appropriate algorithm and transfer function could significantly improve performance. The proposed neural network in modeling the photoactivity of TiO2-P25 nanoparticles in reducing Cr(VI) was acceptable, based on a good correlation coefficient (0.9886) and a small mean square error (0.00018). All the input variables affected the reduction of Cr(VI), however the effect of pH with an impact factor of 34.15 % was more significant than the others. The results indicated that pH = 2 was the best pH for photocatalytic reduction of Cr(VI). Increasing photocatalyst dosage and irradiation time in the investigated range increased Cr(VI) photocatalytic reduction.

Optimized structure of the ANN includes a three-layer feed-forward back propagation network with 4:10:1 topology and the most appropriate algorithm is a scaled conjugate gradient backpropagation algorithm.