Modeling trihalomethanes concentrations in water treatment plants using machine learning techniques.

Water disinfection process in a water treatment process results in the formation of disinfection by-products (DBPs), including total trihalomethanes (TTHMs). It takes a relatively long time to estimate TTHMs concentration level in the water treatment plants; thereby it is impossible to timely control operation parameters to reduce the TTHMs concentration. Here, we developed a predictive model to quantify TTHMs concentration using conventional water quality parameters from six water treatment plants in Han River. Before the developing the model, self-organizing map (SOM) and artificial neural network (ANN) restored missing values in input and output parameters. Then, an ANN model was trained to predict TTHMs by using relevant water quality parameters investigated from Pearson correlation. Pearson Correlation test selected six significant input parameters such as temperature, algae, pre-middle chlorine, post chlorine, total chlorine, and total organic carbon. Based on five-fold jackknife cross-validation, the ANN models built using different types of input data showed different performance in training (range of R² from 0.62 to 0.92) and validation (range of R² from 0.62 and 0.80) steps. This study can be a useful tool for predicting TTHMs concentrations using conventional water quality data in drinking water treatment plants. Machine learning models can be readily developed and utilized by managers working with drinking waters. [ABSTRACT FROM AUTHOR]