Application of a Multiobjective Artificial Neural Network (ANN) in Industrial Reverse Osmosis Concentrate Treatment with a Fluidized Bed Fenton Process: Performance Prediction and Process Optimization

Abstract

Industrial reverse osmosis concentrate (ROC) treatment with the fluidized bed reactor Fenton (FBR-Fenton) process was modeled with artificial neural network (ANN) and multilinear regression (MLR) techniques. The performance of the models in predicting total organic carbon (TOC) removal and sludge production in ROC treatment was evaluated. The MLR model showed poor prediction accuracy, indicating the linear model is not suitable for describing the FBR-Fenton process. The ANN model achieved an acceptable accuracy for predicting TOC removal (RMSE = 0.0263; R2 = 0.8680) and sludge production (RMSE = 0.0355; R2 = 0.8182) under different operation conditions. ANN model sensitivity analysis revealed the Fe2+ dosage to be a relatively important factor affecting TOC removal (31%) and sludge production (41%) in the FBR-Fenton process. The validated ANN model was implemented for FBR-Fenton process optimization with the aim of achieving a high TOC removal efficiency and minimal sludge production. The optimum operation conditions were determined to be 1.5 mM Fe2+, 10 mM H2O2, 80 min hydraulic retention time, and pH 3–4, and the predicted TOC removal and sludge production were 57% and 0.23 g/L, respectively. Ten more batch studies and a continuous 2 week study with a larger scale FBR-Fenton system were subsequently carried out under the same optimum operation conditions, to further validate the ANN model’s prediction accuracy.