Integrate Power Grid Model as calculation engine in Grid2op

[petersalemink95]

It would be interesting to have Power Grid Model implemented as a calculation engine within Grid2op. Since both projects are within LFEnergy, it would make sense to find synergies and combine the power of both packages.

Currently Grid2op is more focussed towards TSOs, whereas Power Grid Model is focussed towards DSOs. Adding Power Grid Model as a calculation engine could bring DSO functionalities like asymmetric calculations to Grid2op.

There are open issues for Power Grid Model to support more TSO like components, like PV buses. For example: PowerGridModel/power-grid-model#185. If these things are added to Power Grid Model it could even provide both TSO as well as DSO functionalities to Grid2op. Once Power Grid Model can support all functionalities that are currently provided by LightSim2Grid Power Grid Model might even replace LightSim2Grid. Benefits would be that we do not have to support the calculations engine LightSim2Grid anymore, and only one calculation engine needs to be maintained.

[BDonnot]

Hello,

Indeed this would be a super interesting feature and makes total sense.

Grid2op is "backend agnostic". I mean by that that grid2op can work independantly of the backend used.
I tried to document at various places how to implement a new backend in grid2op.

For example you can follow the doc:
https://grid2op.readthedocs.io/en/latest/developer/createbackend.html#main-methods-to-implement

Or if you prefer "test driven programming" you can also follow this "tutorial" : https://grid2op.readthedocs.io/en/latest/developer/createbackend.html#automatic-testing-and-test-driven-programming

And if you prefer "simple basic working step by step examples" you can have a look here: https://github.com/Grid2op/grid2op/tree/master/examples/backend_integration

Let me know if you need any help of course.

NB: to be able ot use already defined environment with a new backend requires at least to have a conversion from the grid format in the environment to the internal backend representation.

At time of writing, most open source grid2op environments rely on pandapower json format to describe the powergrid.

And I noticed here https://power-grid-model-io.readthedocs.io/en/stable/examples/pandapower_example.html#pandapower-conversion that pandapower is able to convert a grid into a powergrid model "format".

So I guess that this is not a huge problem for PGM.

Currently Grid2op is more focussed towards TSOs, whereas Power Grid Model is focussed towards DSOs. Adding Power Grid Model as a calculation engine could bring DSO functionalities like asymmetric calculations to Grid2op.

This is totally correct. Grid2op is mainly think with TSO usecase in mind. But some people have applied it to DSO problematic I think. It would be even better to have DSO specific solver for these I totally agree.

There are open issues for Power Grid Model to support more TSO like components, like PV buses.

This is good to know :-) And it will for sure make PGM better suited for current grid2op usecases.

However, it should be possible to use quite rapidly (with a standard synchronous single phase powerflow) use PGM with grid2op. Indeed, there are lots of environments based on the IEEE case14 and IEEE case118 data. And, if I understood correctly pandapower documentation, PGM can already be used to run powerflow for these grids.
Though I believe the results for these powerflows will not totally represent a real TSO grid before the introduction of PV buses.

If these things are added to Power Grid Model it could even provide both TSO as well as DSO functionalities to Grid2op.

This would be a really nice opportunity indeed :-)

Once Power Grid Model can support all functionalities that are currently provided by LightSim2Grid Power Grid Model might even replace LightSim2Grid

If the computation time (using grid2op, so including the possible data transfer to/from grid2op to PGM) are equivalent (or even faster :-) ) and PGM offer the same functionalities then I totally agree that the need to continue the development of lightsim2grid will be limited.

Benefits would be that we do not have to support the calculations engine LightSim2Grid anymore, and only one calculation engine needs to be maintained.

On that I might not totally agree.

Depending on the usecase:

• fidelity to reality (AC, DC, unbalanced powerflow, transient simulation, etc.)
• data accessibility for on format or another (if an environment has been developped with a backend in mind it might be difficult to change the backend)
• computation speed
• licensing issues (though as PGM is also open source this does not really apply here)

Some people might still want to use lightsim2grid, pandapower, pywpowsybl, powermodels etc.

PGM could definitely be used for some usecases, for TSOs and DSOs but I don't think it would be a good think that grid2op should be limited to only one calculation engine.

But as you said, if lightsim2grid and PGM overlaps in all their usecases then definitely lightsim2grid could be replaced by PGM.