Application of Artificial Neural Networks - Digital Signal Processing in Power System Protection and Control - page 261

Fig. 12.16 Basic scheme of CT saturation compensation with ANN

Fig. 12.17 Feed-forward
ANN structure for CT
saturation detection/
correction

i s ( n-a+ 1)

Z -1

z -1

ANNout(n)

Z -1

z -1

Z -1

z -1

i s ( n )

a - number of
ANN inputs

The basic idea of CT saturation compensation with neural networks is depicted
in Fig. 12.16 . The CT secondary current, after low-pass filtering and converting to
digital form, is passed to an ANN, at the output of which the reconstructed primary
CT current should appear. The ANN output is further digitally processed to get
certain relay criterion values, the ones needed for decision making in the particular
relay under consideration.
The structure of ANNs for saturation detection and correction is shown in
Fig. 12.17 . Two- or three-layer feed-forward ANNs were used with non-linear
neurons in input and hidden layers and fully linear output neuron. In case of the
saturation detection a single neural network ANN D (2 layers, structure 13-1) was
used for detection of both saturation beginning and ending points. The ANN D
structure was a result of genetic optimization described in [ 30 ]. The network was
trained to produce output equal 1.0 during the period when the CT was saturated,
not only at the time instants when it goes into and out of saturation. Thus the
border points of saturation intervals are detected when the ANN D output changes
its state from 0 to 1 and opposite.
A number of options were considered for the ANNs intended for CT saturation
correction. The size of the networks ANN C was fixed (three-layer, 10-7-1 neurons)