Comparative process optimization of pilot-scale total petroleum hydrocarbon (TPH) degradation by Paspalum scrobiculatum L. Hack using response surface methodology (RSM) and artificial neural networks (ANNs)

• Pilot study of TPH phytoremediation by Paspalum scrobiculatum. L. Hack was employed.
• Optimization process was modeled by RSM and ANN methods.
• Central composite design was used to optimize TPH degradation.
• Artificial neural network was trained and tested using radial basis function networks (RBF) and multilayer feed forward networks (MLP).
• ANN showed a better prediction and fitting ability compared to the RSM.

The aim of this study is to investigate an optimization process for the degradation of total petroleum hydrocarbon (TPH) by a tropical plant, Paspalum scrobiculatum L. Hack, using response surface methodology and artificial neural network. The optimum conditions predicted by RSM were found to be at a diesel concentration of 3%, 72 sampling days and an aeration rate of 1.77 L/min with a 76.8% maximum TPH removal. The coefficients of determination (R2) and adjusted R2 for the RSM model equations were 0.8530 and 0.7208. The optimum conditions predicted by the ANN were found to be at a diesel concentration of 3%, 72 sampling days and an aeration rate of 1.02 L/min with an 85.5% maximum TPH removal. Analysis using the ANN’s prediction data, which showed a higher R2 value of 0.957 and small values of Average Absolute Deviation (AAD) and Root Mean Square Error (RMSE), were 0.33% and 0.302, respectively. Validation analysis showed the predicted values by RSM and ANN were close to the validation values, whereas the ANN showed the lowest deviation, 2.57%, compared to the RSM. This finding suggests that the ANN showed a better prediction and fitting ability compared to the RSM for the non-linear regression analysis.