Application of Artificial Neural Networks - Digital Signal Processing in Power System Protection and Control - page 262

Set of training
cases
small (72 cases)
big (4224 cases)

ANN1

Input signal
preprocessing
none
scalling

ANN2

Activation
function

ANN3

tansig

satlins

ANN4

Fig. 12.18 ANNs and training versions considered

1

s atlin s

0.8

tansig

0.6

0.4

0.2

0

-0.2

-0.4

-0.6

-0.8

-1

-5 -4 -3 -2 -1 0 1 2 3 4 5

Function argument x

Fig. 12.19 Comparison of activation functions ( 12.4 ) and ( 12.9 )

while the other parameters like neuron activation functions, input signal
pre-processing (none or scaling) and size of training set were being changed in
order to identify the best possible configuration and training options. The
parameters that influence the ANN versions are all gathered in Fig. 12.18 . In first
three ANNs from Fig. 12.18 the non-linear tansig activation function ( 12.4 ) was
used in input and hidden layers. The fourth ANN was built of neurons equipped