Largest energy model you've solved using (a recent release of) the HiGHS-solver

Q&A
, , ,
1

Hi openmod’ers!

Just out of interest I’m wondering what’s the largest energy model you’ve been able to solve using HiGHS? Profiling data is quite scarce as far as I am aware. Also, the rapid development of HiGHS might make the few data points I’ve found outdated.

I’m specifically curious to learn about models that work for a whole year, using hourly temporal resolution. I’ve tried to run PyPSA-EUR for a 30 node, H2+electric network. But HiGHS did not seem to be able to pass the presolve step. Might be down to my hardware specs.

Thank you in advance for sharing your insights!

Warm regards,

Seth

2

model.energy (PyPSA server) gives some indication for sector-coupled mode:

193-hourly temporal resolution takes only around 5 minute to solve, but gives reasonable results. This model can only be run at up to 25-hourly resolution for Europe ([one node per country], 25-hourly takes around 16 hours to run) or 3-hourly for single countries (3-hourly takes up to 8 hours to run).

No result for PyPSA-Eur, but I also try to keep track of HiGHS progress here:

https://fneum.github.io/highs-pypsa-progress/

This runs every Tuesday morning with all PyPI versions available, but the problem size is relatively small.

1 Like
3

Please note the forum tag solver-performance.

4

Thanks for the replies! We’ve tried to run a 29 node version of PyPSA-Eur at 4 hour timesteps but haven’t been success in getting past the HiGHS-presolve (left running for 4 hours on a thread ripping AMD cluster). Indicatively, XPRESS solves this problem in about half an hour.

For those interested these are the problem statistics:

Problem Statistics
    35760575 (      0 spare) rows
    15864594 (      0 spare) structural columns
    73167278 (      0 spare) non-zero elements