Biochemical oxygen demand forecasting through ANN, ANFIS and SVM models

Abstract

Mathematical modelling of the activated sludge process (ASP) enhances the understanding of the process and improves the quality of the effluent released. However, as the process is complex and nonlinear, mathematical modelling of the process has been a challenge. In this study, multi-layer perceptron artificial neural networks (MLP-ANN), artificial neuro fuzzy inference system (ANFIS) and support vector machine (SVR) are investigated and compared to predict biochemical oxygen demand (BOD) for better control of waste water treatment plants employing activated sludge process. The study area selected was in a central district of the southern State of India. The model is evaluated based on statistical parameters of the correlation coefficient R and mean square error (MSE). The software MATLAB 8.6 2015b is used for modelling and simulation study of ANN and ANFIS. Python 3.11 is used for SVR modelling. It has been found that effluent BOD was predicted with maximum correlation coefficient of 0.927 and minimum mean square error of 0.0022 by ANN. It is seen that ANFIS gave a maximum R values of 0.9397 and minimum MSE of 0.0018 for predicting BOD while SVR modelling gave an overall regression R of 0.8661 and MSE of .0021 for BOD prediction.