Estimation of the biological oxygen need (BOD5) of urban wastewater treatment plant with machine learning methods

Treatment of wastewater is important as the clean water resources are getting decreased. In order to operate wastewater treatment plants efficiently, treatment processes should be intervened according to the data obtained from the laboratory and the field. In case of late or incomplete intervention, both operational efficiency and cost losses can occur. Biological Oxygen Demand (BOD5) parameter, which is among the laboratory analyzes in the operation of wastewater treatment plants, results in 5 days. Therefore, it is important to estimate some parameters by utilizing the historical data of the wastewater treatment plant for safer, faster, efficient and low-cost operation and control. For this purpose, various statistical methods and modeling techniques are used. In this study, BOD5 parameter was estimated by using 3 different controlled learning algorithms from machine learning algorithms. Flow (Q), Temperature (T), pH, conductivity, Chemical Oxygen Demand (COD), Suspended Solids (SS) were used as input parameters, Biological Oxygen Demand (BOD5) was used as an output parameter. Since the size of the raw data used in the modeling was different from each other, normalization was performed to bring the data to the same scale. After normalizing, the data was divided into two, using 90% of the data set in education and 10% in the test set. The number of iterations for each model developed is set to 100000 iterations. The regression coefficients (R2) obtained for training and test data in the estimation of BOD5 were found to be for SVM learning algorithm 0.9681; 0.9666, for the SGD Regressor learning algorithm 0.5598; 0.7061, and for the Passive Aggressive Regressor learning algorithm was determined as 0.9840; 0.9808. As can be seen from the developed models, both the SVM learning algorithm and the Passive Aggressive Regressor learning algorithm have given very good results for BOD5 estimation and it was understood that it could be used in treatment plants for estimation purposes.