Predicting effluent chemical oxygen demand (cod) in industrial wastewater using artificial neural networks (anns)

Abstract

This Study explores the application of Artificial Neural Networks (ANN) to enhance wastewater treatment processes at Kakira Sugar's Effluent Treatment Plant (ETP). Leveraging three years of historical data combined with daily monitoring of key operational parameters—including FlowRate, pH, Total Suspended Solids (TSS), temperature, Total Dissolved Solids (TDS), and Electrical Conductivity (EC)—the study developed an ANN model capable of accurately predicting effluent Chemical Oxygen Demand (COD) levels. The model was constructed using MATLAB’s Neural Network Toolbox and trained with a 5fold cross-validation approach, ensuring robust generalizability and minimizing overfitting

through iterative weight adjustments via the Levenberg-Marquardt algorithm. Performance

metrics such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and the coefficient of determination (R²) were used to evaluate the model's accuracy, revealing strong

predictive performance on training data and highlighting areas for improvement in validation and testing phases. Sensitivity analysis using Monte Carlo Simulation (MCS) further

identified FlowRate, pH, and temperature as the most influential factors affecting COD

predictions. The outcomes of this project underscore the potential of integrating ANN into wastewater treatment operations, offering a promising tool for real-time process optimization,

regulatory compliance, and enhanced resource management. Future research is recommended

to expand datasets, refine model generalization, and explore hybrid modelling approaches to further elevate the predictive capabilities and operational efficiency of wastewater treatment

systems.