Prediction of industrial wastewater quality parameters based on wavelet de-noised ANFIS model

Abstract

Industrial wastewater quality monitoring and prediction have a strong impact on the water resource planning and environmental protection. This study is focused on investigating wastewater quality of Las Vegas Wash. Many artificial intelligence (AI) techniques, including multivariate linear regression (MLR), artificial neural networks (ANNs), adaptive neuro-fuzzy inference system (ANFIS) and ensemble of ANNs and ANFIS, have been proposed to solve the water quality prediction problem. ANFIS has been proven to achieve higher prediction accuracy. Since the water quality data is likely to be polluted by human errors or natural effects, we propose an improved wavelet de-noised technique combined with the ANFIS model to predict the wastewater quality parameters such as total dissolved solids (TDSs) and electrical conductivity (EC). The proposed model is calibrated, validated and tested using the data measured every two weeks since the year 2007 by Southern Nevada Water Authority (SNWA). To evaluate the performance of the proposed model, three statistical indices are used: mean average percentage error (MAPE), root mean square error (RMSE) and the coefficient of determination (R2). Compared with the experimental results of other wavelet de-noised AI models, the proposed wavelet-ANFIS model achieves the best prediction performance, with the RMSE, MAPE and R2 values equal to 16.06, 0.656, 0.987 and 20.95, 0.577, 0.985 for TDSs and EC prediction, respectively.