Evolving multilayer perceptron, and factorial design for modelling and optimization of dye decomposition by...

Design of experiment and hybrid genetic algorithm optimized multilayer perceptron (GA-MLP) artificial neural network have been employed to model and predict dye decomposition capacity of the biologically synthesized nano CdS diatomite composite. Impact of independent variables such as, light (UV: on-off), solution pH (5-8), composite weight (CW: 0.5-1 mg), initial dye concentration (DC: 10-20 mg/l) and contact time (0-120 min), mainly in two levels, were examined to evaluate dye removal efficiency of the composite. According to the developed response surface based on the factorial design, all independent variables shown positive interactive effect on dye removal (UV > CW > pH > DC), as well as the pH-CW mutual interaction, while both UV-DC and CW-DC had antagonistic effect. The pH-CW interaction was more influential than pH and DC. Incorporation of the intermediate measurements of dye removal between the start and final contact times in GA-MLP approach, had found to improve the accuracy and predictability of the GA-MLP model. Based on the closeness of the R² (0.98), root mean square error (1.03), variance accounted for (98.23%), mean absolute error (0.61) and model predictive error (9.46%) to their desirable levels, proposed GA-MLP model outperformed the factorial design model. Finally, optimal parameter choice for maximum dye removal using factorial design and GA-MLP were found as: UV (on), pH (9), CW (1 g) and DC (10 mg/l) and UV (on), pH (8.85), CW (0.92 g), DC (12.3 mg/l) and T (117 0.6 min), respectively.