[codex] Fix load polynomial cost signs

[Kkkakania]

Summary

This fixes the polynomial active-power cost conversion for controllable loads and other elements that are represented with the opposite active-power sign in PYPOWER.

For a controllable load, pandapower cost coefficients describe the load-side power P, while PYPOWER represents it with -P. The transformed polynomial should therefore be:

a * (-P)^2 + b * (-P) + c = a * P^2 - b * P + c

So only the linear active-power coefficient changes sign. The quadratic and constant terms should keep their original signs.

Changes

• Keep cp2_eur_per_mw2 unchanged when filling gencost.
• Keep cp0_eur unchanged for both quadratic and linear polynomial costs.
• Continue applying the existing sign conversion to cp1_eur_per_mw.
• Add a regression test for controllable load polynomial costs, including both quadratic and linear gencost layouts.
• Constrain scipy-stubs below 1.18 for the typing extra, because the current 1.18 stubs use Python 3.12 type statement syntax while mypy.ini intentionally checks with python_version = 3.10.

Fixes #2954.

Validation

• python -m pytest pandapower/test/opf/test_costs_pol.py -q
• python -m pytest pandapower/test/opf/test_costs_mixed.py pandapower/test/opf/test_basic.py::test_storage_opf pandapower/test/opf/test_dcline.py -q
• python -m mypy
• git diff --check

Remote checks after the second push:

• Passed: Codacy, SonarCloud, all Python 3.10-3.14 build splits, numpy1-support, OPF, docs, linting, postgresql, relying jobs, coverage, typing, and tutorial jobs including tutorial_warnings_tests.
• Known baseline failure: the warnings (3.14, *) jobs fail with the same pandapower/test/plotting/test_geo.py / tap_dependency_table DeprecationWarning that is present on the current develop workflow run. I verified this against develop run 27626832854, job 81693439589.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 71.98%. Comparing base (4917c0d) to head (d15df89).

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #3036      +/-   ##
===========================================
- Coverage    71.98%   71.98%   -0.01%     
===========================================
  Files          352      352              
  Lines        38591    38590       -1     
===========================================
- Hits         27779    27778       -1     
  Misses       10812    10812              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[vogt31337]

.values should not be used.

https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.DataFrame.values.html

[Kkkakania]

Fixed in the rebased branch: the new code uses .to_numpy() for the OPF cost coefficient arrays instead of .values.

[Kkkakania]

Follow-up: I expanded the cleanup before pushing again, so _fill_gencost_poly() now uses .to_numpy() for the active and reactive polynomial cost coefficient arrays (cp* and cq*). The focused OPF and converter tests still pass locally.

[vogt31337]

Hi @Kkkakania , thanks for the PR. But since this convention is in pandapower already a long time, I first want to check other OPF tools. pypower is only one of many possible, and I would not prefer to break all the others just for pypower.

[Kkkakania]

Rebased onto the current develop branch, so the PR no longer has merge conflicts. The earlier scipy-stubs pin change is no longer part of this branch because develop already includes the upstream dependency update.\n\nLocal checks after the rebase:\n\n- uv run pytest pandapower/test/opf/test_costs_pol.py::test_controllable_load_polynomial_cost_signs pandapower/test/opf/test_pandamodels_converter.py::test_controllable_load_polynomial_cost_signs_for_powermodels -q\n- uv run pytest pandapower/test/opf/test_costs_pol.py -q\n- uv run pytest pandapower/test/opf/test_pandamodels_converter.py -q\n\nOn the backend-compatibility concern: the remaining patch keeps the sign convention limited to the generated OPF cost coefficients. The added tests cover both the PYPOWER conversion path and the PowerModels converter output. The intended behavior is that controllable load active power uses the opposite sign for the linear term, while the constant and quadratic coefficients stay in the user-provided cost convention. If the preferred project policy is to keep the historical coefficient convention for compatibility, I can narrow this further to a documentation/test-only change instead of changing the generated coefficients.

[Kkkakania]

Added a focused reactive polynomial-cost regression test for the PYPOWER q-cost rows. This covers the cq0/cq1/cq2 coefficient path that Codecov was flagging after the .to_numpy() cleanup.

Fresh local checks:

• uv run pytest pandapower/test/opf/test_costs_pol.py::test_controllable_load_reactive_polynomial_cost_signs -q
• uv run pytest pandapower/test/opf/test_costs_pol.py pandapower/test/opf/test_pandamodels_converter.py -q
• git diff --check

I also checked the failing warnings (3.14, 1) job log. It fails while collecting pandapower/test/plotting/test_geo.py because mv_oberrhein() emits the existing tap_dependency_table is missing in net deprecation warning under -W error; that does not appear related to this OPF cost change.

[Kkkakania]

I trimmed the implementation diff to keep this PR focused on the coefficient sign behavior only. The .to_numpy() cleanup is no longer part of the branch, so the changed production code is now just the active-power coefficient convention plus the duplicate-sign removal.

Fresh local checks after the force-push:

• uv run pytest pandapower/test/opf/test_costs_pol.py::test_controllable_load_polynomial_cost_signs pandapower/test/opf/test_costs_pol.py::test_controllable_load_reactive_polynomial_cost_signs pandapower/test/opf/test_pandamodels_converter.py::test_controllable_load_polynomial_cost_signs_for_powermodels -q
• uv run pytest pandapower/test/opf/test_costs_pol.py pandapower/test/opf/test_pandamodels_converter.py -q
• git diff --check

This should also make the patch easier for Codecov to evaluate.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud