Predicting the aluminum sulfate dosage in water treatment

Abstract

The present study shows the strategies used to improve the treatment of clarification of demineralized water in GENSA S. A. E. S. P., Planta Termopaipa, located in Boyacá, Colombia. Experimental data obtained from jar tests were used to build a model based on neuronal nets. The independent variables were pH, turbidity, electrical conductivity, and color of the raw water along with flocculent dosage. The output variable was the Aluminum Sulfate dosage. A three-layer neural network was chosen as a prediction approach. The model was validated to find ten neurons in the hidden layer. Nonlinear optimization was the tool used to train the neural network. The chi- square value was used to test the model and showed that the model is efficient at 90% confidence level.