BystrickyK / SINDy

Physically-informed model discovery of systems with nonlinear, rational terms using the SINDy-PI method. Contains functionality for spectral filtering/differentiation.
MIT License
8 stars 0 forks source link

Double pendulum on a cart - Identification results #3

Open BystrickyK opened 3 years ago

BystrickyK commented 3 years ago

Implicit dynamics assuming all parameters are 1 ( to make equations shorter): image

BystrickyK commented 3 years ago

Correlation matrix of the function library: 1615438410_6552799_corr

Matrix of activation distances between identified equations (without labels, since there's too many equations and Matplotlib doesn't let me generate images with sides bigger than 2^16 pixels): 1615438434_1363015_activation_dist

The identified equations, also without labels for the same reason: 1615438439_3838933_implicit_sols These are clustered in the activation space to find models with a consistent (frequently appearing) structure.

If the equations are filtered for equations that have at least one element on the lower triangular part of the activation distance matrix = 0, we get rid of most of them. This is analogous to clustering and keeping clusters with at least two points. Activation distance matrix filtered for equations that appear at least twice: 1615438441_4604166_activation_dist And the solution matrix: 1615438459_5071318_implicit_sols

After clustering and keeping only clusters with 5 or more points, then sorting them so they're together when visualized:

Activation distance matrix: 1615438479_5337195_activation_dist

Solution matrix: 1615438485_8683863_implicit_sols