FlexMeasures / flexmeasures / 36885988224

86%

push

github

Check each commitment for a convex cost curve, rather than their pooled prices (#2631)

* Check each commitment for a convex cost curve, rather than their pooled prices

A commitment carries its own pair of deviation variables, priced by its own pair of prices, so a convex cost curve is a property of one commitment at a time.
Summing every commitment's prices per time step let one commitment's prices mask another's non-convexity, which left out the commitment-sign variables that keep such a commitment bounded.
The solver then returned the problem as infeasible or unbounded, with nothing to point at.

The check now reads a commitment's price through the same helper that the Pyomo and the HiGHS path use to price their deviation variables, so that the check and the model cannot disagree about a commitment's price.

Fixes #2534

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015He1XE7G7HN1svS7Nxgth4
Signed-off-by: F.N. Claessen <claessen@seita.nl>

* Changelog: convexity is checked per commitment

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015He1XE7G7HN1svS7Nxgth4
Signed-off-by: F.N. Claessen <claessen@seita.nl>

* Cover the split halves of an any-type commitment

Copilot asked for the interaction with split multi-slot commitments to be verified, so this holds it.
An "any"-type commitment groups every time slot together, and such a group is split into an upwards and a downwards half, each keeping one price column.
The test passes both before and after the per-commitment check, which is the point: this change leaves the split path alone.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015He1XE7G7HN1svS7Nxgth4
Signed-off-by: F.N. Claessen <claessen@seita.nl>

---------

Signed-off-by: F.N. Claessen <claessen@seita.nl>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

12 of 12 new or added lines in 3 files covered. (100.0%)

20772 of 24086 relevant lines covered (86.24%)

0.86 hits per line