Data Science application for the evaluation of turbidity in the drinking water treatment of the city of Huancayo

Abstract

Turbidity is a physical pollutant with a greater presence in the water to be treated, when its production is intended, above all, for domestic use. A very practical and economical method of reducing this pollutant is by using chemical coagulants or polymers, which, under the principle of bipolar ions, can destabilize colloids and promote the precipitation of total soluble solids. Subsequently, after destabilization, flocs are formed, which become clusters of fine material in suspension which, for their rapid precipitation, are accelerated by the addition of special flocculants. The purpose of this research is related to the application of computational algorithms to create a prediction model of the optimal coagulant dose in the reduction of water turbidity in a treatment plant in the city of Huancayo. For this purpose, data was taken from the plant laboratory, which was nine months old, which were filtered and processed for the application of data science in two major important phases: training, with 70 % of data and; test, with 30 %. The main finding was that the prediction results were 70 % similar to the true results, taking into account the independent variables of initial turbidity, final turbidity, pH, color, among others. It is concluded that the training and validation of the most efficient algorithm is that of Random Forest with 82 % and 72 %, respectively; likewise, the most relevant factors are: turbidity, color of total dissolved solids and conductivity to predict the optimal coagulant dose with the generated model. In that sense, the model could serve purposes of improvement in water treatment.