▎ 摘　　要

The present work aimed to compare the predictive power of response surface methodology (RSM), general regression neural network (GRNN), and artificial neural network (ANN) in modelling of an azo dye (AzD) uptake from aqueous solutions by a new developed nanocomposite (NC) based on the graphene sheets and hydroxyapatite/ZnO nanoparticles. The successful fabrication was confirmed by the techniques of Fourier transform infrared (FTIR), Raman, X-ray diffraction (XRD), Brunauer, Emmett and Teller (BET), scanning electron microscope (SEM), Energy-dispersive X-ray spectroscopy (EDX) and mapping. The optimum adsorption conditions based on Genetic algorithm (GA) and desirability function (DF) offered the maximum removal of 94.83% and 96.5%, respectively. RSM, GRNN and ANN, showed the accurate and robustness performance in forecasting AzD adsorption performance. The statistical error terms approved that ANN model showed the best precision and accuracy, owning to RMSE = 0.1257, MAE = 0.0760, AAD = 0.0087, SAE = 3.800, and SSE = 0.7907, followed by the GRNN and RSM models. Following of the obtained data from pseudo-second order kinetic model (R-2 = 0.991), the negative and positive values for Delta G degrees (ranges from -2.33 to -2.758 kJ/mol) and Delta H (6.23 kJ/mol), and Langmuir isotherm model (R-2 = 0.982) supported that the process is chemisorption, endothermic, and homogeneous, respectively. The higher adsorption capacity of the as-prepared NC (similar to 700 mg/g) over a short time (i.e. 4 min) in comparison with those reported until now, along with the potential of application for six consecutive cycles with no significant fail in the performance, make it to a promising option for the treatment of dye-containing aqueous environments. (C) 2022 Elsevier B.V. All rights reserved.