Implementation of artificial neural network (ann) techniques in some water quality parameters in Karacomak Dam basin

In this study, electrical conductivity (EC), pH, temperature (T), dissolved oxygen (DO), turbidity (TUR), total hardness (Ha), total alkalinity, ammonia nitrogen (NH4-N), nitrite nitrogen (NO2-N), nitrate nitrogen (NO3-N), phosphate (PO4-P), biochemical oxygen demand (BOD), chemical oxygen request (COD) were analysed in accordance to standard methods for the examination of water and wastewater. Between September 2015 and July 2016, the results obtained from the stations have been categorized between excellent to poor water quality for the human use. In this study, the development of the artificial neural network (YAS) for estimating WQI for the Kastamonu Municipality and Karacomak Dam was investigated. The last model structure utilized in this study which is based on a simple feedforward network. The simple feedforward network is applied with two different training algorithms the standard back-propagation algorithm (Levenberg-Marquardt) (train-lm) and Bayesian regulation backpropagation (train-br). In this study, one hidden layer has been selected for modelling and the number of the hidden neuron is set (n+1) and (2n+1) of input nodes. And, many empirical investigations are carried out by using the deferent set of hidden neurons. It used to be determined that If the model has all parameters (input) that are created to predicted the WQI (output) they modified in hidden nodes at (2n+1) however if there is any change in the number of input with 5 or much less than 5 parameters inputs, that it will want to decrease of the hidden node with (n+1). On the other hand, the comparison has been completed between WQI- Calculation and WQI -Predict. The order is as follows model2-abr >model2-alm >model2-5br >the last one is model2-5lm. This discovering can be depicted by using the way that the quantity of hidden neurons straightforwardly affects the execution of the system and we can see that model with Bayesian legislation backpropagation as activation function (train-br) is optimal to the standard back-propagation algorithm (Levenberg-Marquardt) train-lm.