Robbybp / surrogate-vs-implicit

Comparing surrogate models and implicit function formulations for chemical process models
Other
1 stars 0 forks source link

initialize ATR optimization problem by solving square system #6

Closed Robbybp closed 7 months ago

Robbybp commented 1 year ago

The autothermal reformer optimization problem's convergence seems to be sensitive to platform. On my linux desktop, it doesn't converge due to higher-than-expected initial infeasibility of the recuperator's hot side enthalpy balances. I'm not sure why this infeasibility is platform-dependent (probably due to recuperator's initialize method?).

Workaround for now is to initialize the optimization problem by solving the square system. This is somewhat time-consuming and depends on https://github.com/Pyomo/pyomo/pull/2882, so this PR shouldn't be merged until that is merged. Maybe a sequential-modular initializer would work as well, and might be faster?

sbugosen commented 1 year ago

Interesting. What about this approach for initialization?

https://idaes.github.io/examples-pse/latest/Tutorials/Basics/HDA_flowsheet_solution_doc.html

They use the SequentialDecomposition tool to find a tear stream and initialize the flowsheet from there.

You are making me think what my laptop could have that everything runs here. What could it be? Should we compare every package version?

sbugosen commented 1 year ago

By the way, I tried running a square problem but didn't work correctly. I'm going to install your branch to run the simulations.

Robbybp commented 7 months ago

Closing this PR as it is outdated.