Enhancement of Advanced Oxidation Processes in Oil Refinery Wastewater Treatment Using Deep Neural Networks

The complex composition of persistent and resistant contaminants in oil refinery wastewater presents a significant environmental challenge that conventional treatment methods frequently fail to effectively address. Advanced Oxidation Processes (AOPs), specifically the photo-Fenton method, and a deep learning framework known as the Infallible Deep Neural Network (InfDNN) with a novel activation function known as Infallible Linear Units (InfLU) are the focus of this study's integrated approach. The investigation focused on the removal of polycyclic aromatic hydrocarbons (PAHs) and other pollutants from refinery wastewater.  Nine PAHs were found in the GC-MS analysis, including benzo(a)pyrene, phenanthrene, and naphthalene. Before treatment, benzo(a)pyrene and benzo(k)fluoranthene exceeded the CPCB limit of 0.06 g/L. The photo-Fenton process demonstrated high efficiency, with naphthalene levels reduced from 373.47 µg/L to 5.08 µg/L at the Inlet—a 98.6% degradation rate—and phenol completely eliminated at most sampling points (100% removal).  Overall, PAH degradation efficiencies ranged from 84.5% to 100%, and partial mineralization was confirmed via Total Organic Carbon (TOC) analysis.  Iron (Fe) levels at the effluent discharge point reached 30.00 mg/L, exceeding the IRSGP limit and indicating the need for further treatment.  The InfDNN model was trained with the Levenberg–Marquardt algorithm and the Multilayer Perceptron (MLP) architecture. It was validated with normalization in the range [0.2–0.8] using 80% of the data for training and 10% each for testing and validation. This study demonstrates the effectiveness of combining photo-Fenton AOPs with AI-based modeling for scalable, cost-efficient, and regulation-compliant treatment of oil refinery wastewater.