Comparison of Bayesianneural Network, Artificial Neural Network Gene Expression Programming in River Water Quality (Case Study: Belkhviachay river)

The amount of total dissolved solids (TDS) is an important factor in stream engineering, especially study of river water quality. This study estimates the TDS amount of Belkhviachayriver in Ardabil Province, using bayesian neural network-, gene smart and artificial neural network. Quality variables include hydrogen carbonate, chloride, sulfate, calcium, magnesium, sodium and inflow (Q) in monthly time scale during the period (1976-2009) as input and TDS were chosen as output parameters. The criteria of correlation coefficient, root mean square error and of Nash Sutcliff coefficientwere used to evaluate and performance compare ofmodels. The results showed that however the models could be used to estimate with reasonable accuracy the amount of dissolved solids in water deal, but regarding to accuracy, bayesian neural network model with the highest correlation (0.966), minimum root mean square error (0.094ppm) and the Nash Sutcliff (0.998) were put in the verification phase. The results showed that the bayesian neural network model to estimate high minimum and maximum values ​​of dissolved solids in water.