How can I optimize the model for the parameter estimation with the greybox modeling tools to compute the estimation faster?

It is sought to perform parameter estimation concerning the non-linear equations for a physical model using the greybox modeling tools from the System Identification Toolbox. However, running this model takes a very long time in Simulink.

How can I optimize the model for the parameter estimation with the greybox modeling tools to compute the estimation faster?

Answer 1

[MathWorks Support Team]

The main speed-up technique is to write a mex version of the physical model for which the parameters need to be estimated. Please refer to the following documentation example for more information,

https://de.mathworks.com/help/ident/ug/two-tank-system-c-mex-file-modeling-of-time-continuous-siso-system.html

Moreover, one could try speed-accuracy trade-off by using different numerical gradient computation options. See for instance the nlgreyestOptions->GradientOptions. Choosing the DifferencingScheme to be forward or backward, rather than central, can reduce the number of function calls, and thus reduce the overall simulation time.

Lastly, one could try to make the objective as smooth and convex as possible, reduce the number of unknowns, place sensible bounds on free parameters etc., in order to speed up the computations.