Chaotic surrogate models

xn+1 = .671 - .416xn - 1.014xn2 + 1.738xnxn-1 +.836xn-1 -.814xn-12

Data

Model

Auto-correlation function (1/f noise)

Previous slide

Next slide

Back to first slide

View graphic version

Can use arbitrary nonlinear functions

Brute-force method (variant of simulated annealing)

Easy to fit particular characteristics (i.e., power spectrum)

Not good for prediction (doesn't preserve dynamics)

Model is (usually) NOT unique, but (usually) IS robust to small variations of parameters