Comparative analysis of adaptive neuro-fuzzy inference system (ANFIS) and RSRM models to predict DBP (trihalomethanes) levels in the water treatment plant

Abstract

Since disinfection by-products are a growing concern, it is important to know their quantities in water treatment plants before they are released to the public. As a result, there is a requirement for constant monitoring of disinfection by-products (DBPs), which can have major consequences for human health and productivity. Consequently, in previous studies, several models for predicting disinfection byproduct formation in drinking water have been developed which were either linear/log-linear, hybrid or neural network (radial basis function). In this study, an adaptive neuro-fuzzy inference system (ANFIS) is proposed for predicting trihalomethane levels in real distribution systems. To train and verify the proposed model, 24 sets of data were used, including THMs levels (TCM, BDCM, DBCM and T-THM levels) and five parameters (pH, temperature, UVA254, residual chlorine, and dissolved organic carbon). As compared to response surface modeling (RSRM) coefficient of determination, R2 is between 0.727 < R2 < 0.886, average absolute deviation, AAD = 4.07–10.99 %), MAE = 0.01 – 0.978, and RMSE = 0.017 – 1.449. Further, ANFIS for THMs (T-THMs, TCM, BDCM, and DBCM) prediction consistently show higher regression coefficients between 0.956 < R2 < 0.989, average absolute deviation, AAD = 0.350 – 1.977 %), MAE = 0.002 – 0.133, and RMSE = 0.007 – 0.401, Consequently, based on the statistical indices obtained, ANFIS, on the other hand, proved to be effective for predicting the formation of THMs, and thus allowed improved DBPs monitoring in water treatment systems.

Keywords

• Trihalomethane
• Adaptive neuro-fuzzy inference system
• Disinfection byproduct
• Response surface regression model
• Water treatment plants
• Neural networks