Application of Artificial Neural Networks - Digital Signal Processing in Power System Protection and Control - page 246

Fig. 12.1 ANN as a process

model

Output
signals

Input
signals

ANN

Fig. 12.2 McCulloch and
Pitts' neuron model

p 1
p 2

w 1
w 2

o ( p )

p R w R

12.1 Neuron Models and Neural Network Structures

Inspired by the anthropological analogy the ANNs resemble the structure of a
human brain, whose smallest component is a single neuron. The neurons are
connected with neighbouring elements through their axons (sending information),
whereas the incoming signals from other units are received via dendrites synapses.
The human brain consists of hundreds of billions of neurons, while signals in brain
are noisy spike trains of electrical potential. The important feature of the human
neural network is the ability of parallel signal processing as well as distributed
memory, which makes our brain robust to noise and failures.
The first concepts related to artificial neural networks were developed by
McCulloch and Pitts as early as in 1943 [ 4 ]. Their ideas such as threshold and
many simple units combined together to give increased computational power are
still in use today. The McC&P neuron is depicted in Fig. 12.2 , with the corre-
sponding equation ( 12.1 ) for the neuron output given below. It is seen that the
neuron output is equal to 0 or 1, depending on the calculated weighted sum of the
input values p 1 … p R . The inventors assumed further that the inputs p could take
only values 1 or 0 (input activated or dead), and the weighting coefficients were
either +1 or -1. Therefore the application of such neurons could be very limited.

8
<

for P

R

1 w i p i h

i ¼ 1

o ð p Þ¼ ð 12 : 1 Þ

for P

R

:

0 w i p i \h

i ¼ 1

A network of several McC&P neurons can be used for realization of basic logic
operations, like AND, OR, NAND, NOR, some of which are illustrated in
Fig. 12.3 .