Prediction of Organic Pollution of Waters from the Déganobo Lake System: A Modeling Study

Abstract

This work aimed to study the modeling of the organic pollution of the waters of the Déganobo Lake system by three models: Multiple Linear Regression model (MLR model), Mutilayer Perceptron model (MLP model) and Multiple Linear Regression/ Mutilayer Perceptron hybrid model (MLR/MLP hybrid model). In its implementation, the chemical oxygen demand (COD) of these waters, obtained from August 2021 to July 2022, was used. Two approaches were done in the case of the modeling of their COD by the MLP model and the MLR/MLP hybrid model: static modeling and dynamic modeling. The results have highlighted the low predictions of the COD of these waters by the MLR model (36.2 %) and the MLP models (6-8-1 for the static modeling and 7-3-1 for the dynamic modeling, both predicting less than 35% of the experimental values with high error (RMSE upper than 1.30 and relative error upper than 0.750). However, the MLR/MLP hybrid models (MLR/6-3-1 for the static modeling and MLR/7-3-1 for the dynamic modeling) both well predicted the COD of these waters, around 99% with very low errors (RMSE less than 0.0001 and relative error less than 0.006 in both cases). So, the MLR/MLP hybrid model was the most efficient to predict the COD of these waters. The accuracy of this hybrid model for ecological modeling was again provided during this study.