Prediction of TSS and COD using turbidity data from an artificial neural network

The conventional characterization of wastewater relies on laborious and costly analyses, limiting real-time monitoring.This study used turbidity as an indicator to predict suspended solids (SS) and chemical oxygen demand (COD) concentrations at the Koumassi-Digue pretreatment plant (Abidjan).Two hundred (200) samples were collected and analyzed according to AFNOR standards and the methods described by CEAEQ, and then used to train artificial neural networks (ANNs).The 1-6-1 architecture allowed for the modeling of SS with a coefficient of determination R² = 0.94 and a mean relative error MRD = 5.94%.For COD, the 1-4-1 model provided R² = 0.77 and MRD = 11.10%, the discrepancies being related to the dissolved fraction not detectable by turbidimetry.The results confirm the relevance of turbidity for predicting SS and particulate COD.Artificial neural networks (ANNs) thus prove to be a robust, fast, and economically viable tool for the continuous monitoring of urban effluent quality.