Predicting the residual aluminum level in water treatment process

Abstract

In water treatment processes, aluminum salts are widely used as coagulation chemical. High dose of aluminum has been proved to be at least a minor health risk and some evidence points out that aluminum could increase the risk of Alzheimer's disease. Thus it is important to minimize the amount of residual aluminum in drinking water and water used at food industry. In this study, the data of a water treatment plant (WTP) was analyzed and the residual aluminum in drinking water was predicted using Multiple Linear Regression (MLR) and Artificial Neural Network (ANN) models. The purpose was to find out which variables affect the amount of residual aluminum and create simple and reliable prediction models which can be used in an early warning system (EWS). Accuracy of ANN and MLR models were compared. The new nonlinear scaling method based on generalized norms and skewness was used to scale all measurement variables to range [-2⋯ + 2] before data-analysis and modeling. The effect of data pre-processing was studied by comparing prediction results to ones achieved in an earlier study. Results showed that it is possible to predict the baseline level of residual aluminum in drinking water with a simple model. Variables that affected the most the amount of residual aluminum were among others: raw water temperature, raw water KMnO4 and PAC/KMnO4 (Poly-Aluminum Chloride/Potassium permanganate)-ratio. The accuracies of MLR and ANN models were found to be almost the same. Study also showed that data pre-processing affects to the final prediction result.