You are on page 1of 1

Multilayer Artificial Neural Network 25

For a system of first-order ODEs (Equation 3.36), we have the derivatives of


the error function with respect to wj and vj as follows:
2
æ é ùö
ç ê ì dyt1 ( xi , p1 ) ü ú÷
ç êí - f1 ( xi ,yt1 ( xi ,p1 ), ¼ ,yt ( xi ,p ) ) ý ú÷
ç êî dx þ ú÷
¶ ç ú÷
h
¶E( x ,p) 1ê
¶w j
= ç
¶w j ç å i =1
ê +í

ì dyt2 ( xi , p2 )
î dx
ü
- f 2 ( xi ,yt1 ( xi ,p1 ), ¼ ,yt ( xi ,p ) ) ý
þ
ú÷
ú÷
ç üú ÷
ç
ê ì ( )
ê +  + ï dyt x i , p - f ( x ,y ( x ,p ), ¼ ,y ( x ,p ) ) ïú ÷
ç ê í l i t1 i 1 t i  ý ÷
ú
ç ïî dx ïþû ÷ø
è ë
(3.37)
2
æ é ì dyt1 ( xi , p1 ) ü ùö
ç êí - f1 ( xi ,yt1 ( xi ,p1 ), ¼ ,yt ( xi ,p ) ) ý ú÷
ç êî dx þ ú÷
¶ ç ú÷
h
¶E( x ,p) 1ê
¶v j
= ç
¶v j ç å
i =1
ê +í

ì dyt2 ( xi , p2 )
î dx
ü
- f 2 ( xi ,yt1 ( xi ,p1 ), ¼ ,yt ( xi ,p ) ) ý
þ
ú÷
ú÷
ç ê ì dyt ( xi , p ) üú ÷
çç ê + + í - f l ( xi ,yt1 ( xi ,p1 ), ¼ ,yt ( xi ,p ) ) ýú ÷÷
è ë î dx þû ø
(3.38)
It may be noted that the detailed procedure of handling IVPs and BVPs using
a single-layer ANN model is discussed in subsequent chapters.
Next, we address the computation of the gradient of ODEs using the tradi-
tional multilayer ANN model.

3.2.5 Computation of Gradient of ODEs for Multilayer ANN Model


Error computation not only involves the output but also the derivative of
the network output with respect to its input [6]. So it requires finding the
gradient of the network derivative with respect to its input. Let us now con-
sider a multilayer ANN with one input node x that has h number of data
(x = (x1, x2,  … , xh)T), a hidden layer with m nodes, and one output unit. The
network output N(x, p) is formulated as
m

N ( x , p) = åv s(z )
j =1
j j (3.39)

The derivative of N(x, p) with respect to input x is


m
dk N
dx k
= åv w s( )
j =1
j
k k
j j (3.40)

You might also like