Effectiveness of Different Artificial Neural Network Models in Establishing the Suitable Dosages of Coagulant and Chlorine in Water Treatment Works

Generally, in India, determining the chlorine and coagulant dosage in a WTP depends on the proficiency of operators, which may lead to overdosing or underdosing of coagulants and chlorine. Nevertheless, the determination of both coagulant and chlorine dosages frequently changes as inlet water quality varies which demands extensive laboratory analyses, leading to prolonged experimentation periods in water treatment plants. So objective of the study is to develop the precise relationship between coagulant dose and chlorine dose in a water treatment plant by using an artificial neural network (ANN). As a result, ANN models were developed to predict chlorine dose using coagulant dose by comparing the performance of the number of ANN models. It has been found that radial basis function neural networks (RBFNN) and generalized regression neural networks (GRNN) modeling provide better prediction. In RBFNN and GRNN modeling, the spread factor is varied from 0.1 to 15 to establish a stable and accurate model with high predictive accuracy. It is observed that the RBFNN model showed good prediction (R2 = 0.999). The application of a soft computing model for defining doses of coagulant and chlorine that are inextricably linked at a Water treatment plant (WTP) will be highly beneficial for WTP Managers.