Add AlternativeForwardPass

[odow]

Closes #419

Needs

• Docs
• Tests
• An AC-OPF/DC-OPF example

But yeah. We had all the infrastructure for this. It was trivial to implement.

[andrewrosemberg]

Creating a AC-OPF/DC-OPF example. Will post here in a few minutes. Taking advantage of the code used in PowerModelsAnnex.jl.

[andrewrosemberg]

case3.zip

The ACDC test case. I used PowerModels.jl just to create the reference dictionary to make things a bit easier, but I can try to do it manually.

[andrewrosemberg]

I added a few comments and the example based on personal interest, but if it confuses things, please disregard them.

[odow]

Once I merge #612 and rebase, this might take the record for the most complicated feature implemented in the fewest lines of code. It's really only two one-line functions...

[odow]

Provided two models have the same graph structure and state variables, we now have a lot of freedom:

• Simulating with the true distribution, training with the discretized SAA (sampling schemes, Add SimulatorSamplingScheme #594)
• Simulating with the discrete model, training with the continuous relaxation (happens by default)
• Training with a non-convex model, training with a convex approximation (this PR)

[odow]

Preview: https://odow.github.io/SDDP.jl/previews/PR611/tutorial/pglib_opf/

[odow]

@andrewrosemberg this tutorial is so cool: https://odow.github.io/SDDP.jl/previews/PR611/tutorial/pglib_opf/.

It's like a mini-HydroPowerModels.jl.

These new features should make it easy for people to replicate and build upon your paper now.