Support for piecewise (convex) costs in ACOPF

[michaelbynum]

Support for piecewise (convex) costs in ACOPF

[bknueven]

I have lots of comments, mostly because I've implemented piecewise cost curves a few times; feel free to consider them completely optional.

The unit commitment model spends quite bit of effort validating and potentially simplifying the data passed in, e.g., ignoring segments outside of [p_min, p_max], combining close segments, etc. Unfortunately some of these boring and tedious things can make a performance difference; it's also nice just to have some of this preprocessing in Egret itself if the data is messy.

I guess there's also the question of if this is the PWL cost formulation for the ACOPF model, see: https://arxiv.org/pdf/2005.14087.pdf. This also discusses a bit of the preprocessing issues.

[michaelbynum]

@bknueven Thanks for all of the feedback. I think this is a greatly improved PR now. Let me know what you think.

[bknueven]

As it is, the validation code does not handle if the p_max <= the first piecewise point or the p_min >= the last piecewise point. We should either address this in this PR or open and issue.