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.

106 CHAPTER 4 NEURAL NETWORK BASED SYSTEM IDENTIFICATION<br />

<strong>of</strong> the Jacobian matrix ψ(t|θ) would become d × (nN) if multiple input-output variables<br />

were considered in the estimation problem. <strong>The</strong> Jacobian matrix ψ(t|θ) also can be<br />

effectively calculated using an alternative approach such as finite differences method<br />

[Norgaard, 2000, Yu and Wilamowski, 2011, Wilamowski et al., 2008].<br />

Consider a two layer MLP network with hyperbolic tangent hidden units and linear<br />

output units:<br />

ŷ i (t | θ) mlp =<br />

=<br />

⎛ ⎛<br />

⎞ ⎞<br />

H∑<br />

m∑<br />

⎝W 2 ih tanh ⎝ W 1 hj ϕ j (t) + B1 h<br />

⎠ + B2 i<br />

⎠<br />

h=1<br />

j=1<br />

H∑<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 (4.36)<br />

<strong>The</strong> MLP Jacobian matrix ψ(t|θ) is calculated by partial differentiating Equation (4.36)<br />

with respect to parameter vector θ arriving at the following results:<br />

⎧<br />

v h (t)<br />

if θ = W 2 ih<br />

ψ (t|θ) mlp<br />

=<br />

∂ŷ (t|θ)<br />

∂θ<br />

⎪⎨<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 />

0 otherwise<br />

(4.37)<br />

<strong>The</strong> output prediction from the HMLP can also be expressed with the hyperbolic<br />

tangent hidden units and linear output unit as follows:<br />

ŷ i (t |θ ) hmlp<br />

=<br />

=<br />

⎛ ⎛<br />

⎞ ⎞<br />

H∑<br />

m∑<br />

⎝W 2 ih tanh ⎝ W 1 hj ϕ j (t) + B1 h<br />

⎠ + B2 i<br />

⎠ +<br />

h=1<br />

j=1<br />

H∑<br />

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

h=1<br />

m∑<br />

W 3 ij ϕ j (t)<br />

j=1<br />

m∑<br />

W 3 ij ϕ j (t)<br />

with h = 1, 2, 3 · · · H and i = 1, 2, 3 · · · n (4.38)<br />

j=1<br />

where W 3 ij is the weights matrix <strong>of</strong> linear connection between input layer and output

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

Saved successfully!

Ooh no, something went wrong!