Comparative Analysis of Machine Learning Algorithms for Forecasting Effluent Chemical Oxygen Demand in Wastewater Treatment Plants

Abstract

Accurate prediction of wastewater effluent parameters is crucial for evaluating the performance of wastewater treatment plants, as it significantly contributes to reducing time, energy, and costs. This study employed three machine learning algorithms such as Artificial Neural Network (ANN), Support Vector Machine (SVM), and Gaussian Process Regression (GPR) in order to forecast the output COD values of Wastewater Treatment Plant No. 1 in Parkand Abad, Mashhad, Iran. The input data for the models included BOD5, COD, TSS, Temprature, and pH of influent sewage, recorded daily from March 2018 to June 2019. The findings indicated that the SVM model surpassed the ANN and GPR models in predicting effluent COD parameters across all three phases, with GPR also performing better compared to ANN throughout the training, validation, and testing stages. The SVM model achieved values of r = 0.82, R2 = 0.67, RMSE = 19.02, MAPE = 0.069, and MAE = 13.26 during the training phase, and the model exhibits values of r= 0.74, R2= 0.45, RMSE=28.02, MAPE=0.080, and MAE=18.46 in the testing phase.

Keywords

• chemical oxygen demand
• coefficient of determination
• machine learning algorithms
• wastewater treatment plant