Design of ANFIS Structures and GMDH Type-Neural Network for Prediction of Optimum Coagulant Dosage in Water Treatment Process Case Study: Great Water Treatment Plant in Guilan Province

Given the increasing importance of surface water bodies as supply sources of drinking water and regarding the requirement for using different chemicals at various stages of water treatment processes, it is essential to investigate coagulant consumption in water treatment plants. Determination of the required dosage of coagulants used in the coagulation and flocculation unit is one of the most important decisions in water treatment operations. For this purpose, the jar test is generally used to determine the type and concentration of suitable coagulants in a water treatment plant. However, the test is rather time-consuming and unreliable due to the inaccurate results it yields. Instead, intelligent methods can be employed to overcome this shortcoming of the jar test. In this study, experimental data were collected over the period from 2011 to 2012 and further refined for study. Two non-linear models based on adaptive neuro-fuzzy inference system (ANFIS) and GMDH-type neural networks were then developed and experimental results were used to determine the optimum poly-aluminium chloride dosage for use at Guilan water treatment plant. The effects of input parameters including temperature, pH, turbidity, suspended solids, electrical conductivity, and color were investigated on coagulant dosage. The ANFIS model was found to outperform the GMDH model in predicting the required poly-aluminium chloride dosage.