WATER QUALITY PREDICTION IN DISTRIBUTION SYSTEM USING CASCADE FEED FORWARD NEURAL NETWORK

Abstract

Cascade feed forward ANN models have been developed by using pH, Alkalinity, Hardness, TS and MPN as the input variables to forecast water quality index (WQI) in the various zones of municipal distribution system. Different ANN models were developed using training data set and tested in order to determine optimum number of neurons in the hidden layer and best fitting transfer function. The study reveals that the predictions by logsigmoidal and pure linear transfer function are in good correlation with observed WQI as compared to tansigmoidal transfer function. It is also observed that the model performance changes considerably with change in hidden layer neurons. Hidden layer structure with seven neurons performs better, followed by hidden layer structure with four neurons and one neuron respectively.