search

pybop-team / PyBOP

A parameterisation and optimisation package for battery models.
https://pybop-docs.readthedocs.io
BSD 3-Clause "New" or "Revised" License
48 stars 9 forks source link

Parameter grouping and structural identifiability #8

Open davidhowey opened 1 year ago

davidhowey commented 1 year ago

It's clear that sometimes parameters need to be grouped as part of the estimation process. For example, it is not possible to estimate particle radius and diffusivity separately from input-output data, only diffusion time $R^2/D$. In practice, however, one would have to set an (arbitrary) radius or diffusivity and estimate the other parameter, but only report diffusion time. Some packages exist for checking the 'in principle' identifiability of parameters, notably STRIKE-GOLDD, but this is Matlab-based. We probably want to avoid this issue and instead be very careful in hand-holding users through the parameter estimation process to make them aware that not all parameters are 'free' parameters.

martinjrobins commented 1 year ago

I can see this being very easy if we only allow users to use a specific set of models, and very difficult if we allow users to BYOM (bring your own model)