AD Tutorial at the TU Berlin
AD Tutorial at the TU Berlin
AD Tutorial at the TU Berlin
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Optimum Experimental Design in Chemical Engineering (Cont.)<br />
Dynamics: Defined by ODE<br />
Goal: Estim<strong>at</strong>e parameters p = (k 1 , k k<strong>at</strong> , E k<strong>at</strong> )<br />
Problem: Errors in <strong>the</strong> measurements η result in errors in parameters p.<br />
nonlinear regression with additive iid normal errors<br />
η m = h m(t m, x(t m), p, q) + ε m ,<br />
ε m ∼ N (0, σ 2 m )<br />
m = 1, . . . , N M<br />
η m are measurements, h measurement model function (connects model to <strong>the</strong> real world)<br />
Controls q = (n a1 , n a2 , n a4 , c k<strong>at</strong> , θ) influence <strong>the</strong> error propag<strong>at</strong>ion.<br />
Therefore: Find controls q such th<strong>at</strong> <strong>the</strong> “uncertainty” in p is as “small” as possible.<br />
η 2<br />
ˆη(q 1 )<br />
ˆη(q 2 )<br />
J † (q)<br />
p 2<br />
ˆp(q 1 )<br />
ˆp(q 2 )<br />
Sebastian F. Walter, HU <strong>Berlin</strong> () Not So Short <strong>Tutorial</strong> OnAlgorithmic Differenti<strong>at</strong>ion Wednesday, 04.06.2010 19 / 39<br />
η 1<br />
p 1