Modelling biological oxygen demand in wastewater of Arak City, Iran

Abstract

Due to the fact that the laboratory analysis of biological oxygen demand (BOD) is time-consuming and uncertain because of some interferences during analysis, automatic estimation of BOD by modeling methods would be of great importance to researchers. The main aim of this study was to compare the performance of three models for the prediction of BOD in the wastewater of Arak City, Iran. The methods were artificial neural network (ANN with early stopping and ANN with Bayesian regularization), an Ensemble of ANN (EANN), and partial least squares regression. The models trained and were validated on a data set containing 18 parameters sampled periodically from the wastewater of Arak City. The performance of these models was assessed by mean squared error (MSE) and R2 of the test data set besides correlation coefficient of observed and predicted values of BOD for each model. Concerning the Bayesian regularization algorithm, the perfect fitting of the model to both training and test data with correlation coefficients of 0.999 and 0.945 confirmed that this method outperformed the results of other models. Moreover, the results of sensitivity analysis indicated that chemical oxygen demand (COD), next to sulfate and fecal coliforms are the most important parameters in the prediction of BOD by ANN modeling.