28.02.2014 Views

The Development of Neural Network Based System Identification ...

The Development of Neural Network Based System Identification ...

The Development of Neural Network Based System Identification ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

4.3 SYSTEM IDENTIFICATION WITH NEURAL NETWORK 107<br />

layer. By performing differentiation as in MLP example, the Jacobian matrix for HMLP<br />

is given as:<br />

⎧<br />

v h (t)<br />

if θ = W 2 ih<br />

ϕ j (t)<br />

if θ = W 3 ij<br />

ψ (t|θ) hmlp<br />

=<br />

∂ŷ (t|θ)<br />

∂θ<br />

⎪⎨<br />

=<br />

1 if θ = B2 i<br />

(<br />

W 2 ih 1 − v<br />

2<br />

h<br />

(t) ) ϕ j (t)<br />

(<br />

W 2 ih 1 − v<br />

2<br />

h<br />

(t) )<br />

if θ = W 1 hj<br />

if θ = B1 h<br />

(4.39)<br />

⎪⎩<br />

0 otherwise<br />

Similarly, the output prediction from the modified Elman network can also be<br />

expressed with the hyperbolic tangent hidden units and linear output unit as follows:<br />

⎛<br />

⎞<br />

m∑<br />

h∑<br />

v h (t) = tanh ⎝ W 1 hj ϕ j (t) + B1 h + W 3 k x k (t) ⎠<br />

j=1<br />

k=1<br />

x k (t) = v h (t − 1) + αx k (t − 1)<br />

H∑<br />

ŷ i (t |θ ) Elman<br />

= W 2 ih v h (t) + B2 i<br />

h=1<br />

with h = 1, 2, 3 · · · H and i = 1, 2, 3 · · · n (4.40)<br />

where W 3 k is the weights vector for connections between context units and hidden<br />

layer. <strong>The</strong> vector x k (t) denotes the output <strong>of</strong> the context units. By performing the<br />

differentiation <strong>of</strong> network output with respect to each weight term, the Jacobian matrix<br />

for Elman network is given as follows:<br />

⎧<br />

v h (t)<br />

if θ = W 2 ih<br />

W 2 ih<br />

(<br />

1 − v<br />

2<br />

h<br />

(t) ) ϕ j (t)<br />

if θ = W 1 hj<br />

ψ (t|θ) Elman<br />

=<br />

∂ŷ (t|θ)<br />

∂θ<br />

⎪⎨ W 2 ih<br />

=<br />

∂v h (t)<br />

∂W 3 k (t−1)<br />

if θ = W 3 k<br />

1 if θ = B2 i<br />

W 2 ih<br />

(<br />

1 − v<br />

2<br />

h<br />

(t) )<br />

if θ = B1 h<br />

(4.41)<br />

⎪⎩<br />

0 otherwise

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!