PREDICTION OF RESIDUAL CHLORINE IN A WATER TREATMENT PLANT USING GENERALIZED REGRESSION NEURAL NETWORK

Generally, in India in water treatment plant (WTP) chlorine dose is decided by plant operators based on their experience, which may result in under-chlorination or over-chlorination. Thus, there is a need to develop predictive models for residual chlorine in a WTP. This research work focuses on applying artificial neural network (ANN) approach to predict residual chlorine in a WTP. Weekly water quality data spanning 4 years was obtained from plant laboratory for modeling with ANN. Thirty-two ANN models were developed to predict residual chlorine with four combinations of input variables. The ANN models for prediction of residual chlorine were established by several steps of training and testing using feed forward neural network, cascade feed forward neural network, pattern recognition neural network, radial basis neural network and generalized regression neural network. The ANN models were evaluated with error statistics viz coefficient of determination (R2 ), mean square error (MSE), mean absolute error (MAE) and standard statistics viz mean( ), standard deviation (σ), skewness (ɣ1) and kurtosis (ɣ2). Performance of the ANN models increased with the increasing number of input variables such as turbidity, pH and chlorine dose. The best performing model was found to be GRNN with MSE = 0.001 mg/lit, MAE = 0.019 mg/lit and R2= 0.979. Thus, ANN provides a valuable performance assessment tool for plant operators and decision makers to predict residual chlorine.