2.3. Artificial neural network applied to predict TDS
In this study a feed forward neural network (FFNN) with
back-propagation training algorithm was applied to correlate
the relation between input alkalinity expressed in (pH) and
output salinity expressed in (TDS). The ANN configuration
was identified based on a previous research by Nasr et al.,
2013 and through conducting several trials until reaching the
best regression results with no over-fitting Fig. 2. The network
properties were as follows:
– Network input: pH.
– Network output: TDS concentrations.
– Network type: Feed-forward back-propagation.
– Training function: Levenberg–Marquardt algorithm
(TRAINLM).
– Adaptation learning function: Gradient descent with
momentum weight/bias learning function (LEARNGDM).
– Performance function: Mean square error (MSE).
– Number of layers: 3 (layer-1: five neurons and TANSIG
transfer function; layer-2: three neurons and LOGSIG