Pandapower converter: fix 3-Ph transformer loading Calculation

[furqan463]

Fixes #319
Fixes #321

Changes proposed in this PR include:

Extended "u1" and "u2" arrays from (n,) dimensions to (n,3) dimensions to remove value error.
multiplied phase loading percent with sqrt(3), (i * v / sqrt(3)) / (sn / 3) --> sqrt(3) * (i * v) / sn
divided total loading percent with sqrt(3) as v_ph = v / sqrt(3).

Electrical Mathematics:

Power per Phase = V_n * I_line_to_line # Basic Definition
Capacity per Phase = Sn / 3
Loading per Phase = Power per Phase / Capacity per Phase

ui = i_line_to_line * V_line_to_line
Loading per Phase = [ui / sqrt(3)] / [Sn / 3]
= [3 / sqrt(3)] * [ui] / [Sn]
= sqrt(3) * ui / Sn

Total Power = sum[Power_Phase_a, Power_Phase_b, Power_Phase_c]
= sum[ ui_a / sqrt(3), ui_b / sqrt(3), ui_c / sqrt(3)]
= [ sum(ui_a, ui_b, ui_c) / sqrt(3) ]
Total Loading = [sum(ui_a, ui_b, ui_c) / sqrt(3)] / Sn

Looking at it from another perspective which is more relevant in case of trafo_loading="current"
loading_per_phase = phase_current(line_to_line) / rated_current
rated_current (I_rated) = Sn / (sqrt(3) x V_line_to_line) #As for 3-Ph power system S = sqrt(3) * V * I
loading_per_phase = I_ph / I_rated
= I_ph / [Sn / (sqrt(3) x V_line_to_line)]
= sqrt(3) * [I_ph * V_line_to_line] / Sn

Checks before merge

• Review the mathematics
• Reproduce; verify that this fixes the issue
• Code review
• Add tests
• Pass CI (Sonar Cloud may fail with HTML error code 403. That is not a problem with this PR)

[mgovers]

Thank you for your contribution @furqan463 ! We greatly appreciate you proposing the fix to the bug you found.

This is just a code review. Still TODO:

• add some tests
• review electrical engineering correctness
• reproduce with a pandapower grid.

[nitbharambe]

Nice spotting the bug!
For reference, as given in pandapower documentation:

As in their code:

https://github.com/e2nIEE/pandapower/blob/7b7ab33a8a9cc1b9802e168dfeea824eeb2ef266/pandapower/results_branch.py#L461-L465

The description and these two definitions are all different I suppose. (Atleast the naming is different :D) Makes it quite confusing.

[furqan463]

@mgovers I've updated the description above elaborating mathematics involved. Moreover, thanks to @nitbharambe proposed changes match with existing code in PandaPower. I'm attaching a small code to reproduce the errors as well as testing the proposed changes. I'm trying to create unit tests, but while working on it, I've found some bugs with existing validation data.
pgm_test.py

[furqan463]

@mgovers I've highlighted in relevant test code, why the unit test was not working for this function.
I'm trying to correct this unit test as well.

[furqan463]

@mgovers I've updated the validation unittest, now the test includes all components and since I loading pp_input to converter, all components are being converted and tested.
Now the for trafo_3ph, loading test will fail on original code for all 4 cases. However with amended code of pandapower_converter under trafo_loading=="current", loading for a & b phase will pass but for c and total will still fail.
Upon investigation I found out that the trafo loading data in validation files is for the case trafo_loading=="power", however, I found that the code for trafo_loading=="power" in pandapower_converter was also wrong, which now I corrected and amended unittest function to use trafo_loading="power", and now all test are passing.
However, for trafo_loading="current", we still need to generate validation data.

[mgovers]

Hi @furqan463,

Nice find again! I am thinking that maybe we can try to add your validation case as a validation case similar to

import pandapower as pp
import numpy as np
from math import sqrt


def create_net():
    net = pp.create_empty_network('test', f_hz=50)
    pp.create_bus(net, 11, 'source')
    pp.create_bus(net, 11, 'TF_HT')
    pp.create_bus(net, 0.4, 'TF_LT')

    pp.create_ext_grid(net, 0, 1.0, 0, s_sc_max_mva=100, rx_max = 0.1, x0x_max=1, r0x0_max=0.1)
    pp.create_line_from_parameters(net, 0, 1, 2, 0.33, 0.34,
                                0.001, 600, r0_ohm_per_km=0.66, x0_ohm_per_km=0.65, c0_nf_per_km=0.001,
                                g_us_per_km=0, g0_us_per_km=0)
    pp.create_transformer_from_parameters(net, 1,2, 1, 11, 0.415, 0,
                                        4, 0, 0.01, -30, vector_group='Dyn',
                                        vk0_percent=4, vkr0_percent=0, mag0_percent=100, mag0_rx=0, si0_hv_partial=0.9)
    pp.create_asymmetric_load(net, 2, 0.2, 0.19, 0.195,
                            0.05, 0.045, 0.035, 0, type='wye')
    return net


def check_result(net):
    v_hv = net.trafo.vn_hv_kv
    v_lv = net.trafo.vn_lv_kv
    i_max_hv = np.divide(net.trafo.sn_mva, v_hv* np.sqrt(3)) * 1e3
    i_max_lv = np.divide(net.trafo.sn_mva, v_lv* np.sqrt(3)) * 1e3


    i_a_hv = net.res_trafo_3ph.loc[:, 'i_a_hv_ka'] * 1000
    i_b_hv = net.res_trafo_3ph.loc[:, 'i_b_hv_ka'] * 1000
    i_c_hv = net.res_trafo_3ph.loc[:, 'i_c_hv_ka'] * 1000

    i_a_lv = net.res_trafo_3ph.loc[:, 'i_a_lv_ka'] * 1000
    i_b_lv = net.res_trafo_3ph.loc[:, 'i_b_lv_ka'] * 1000
    i_c_lv = net.res_trafo_3ph.loc[:, 'i_c_lv_ka'] * 1000

    np.testing.assert_allclose(np.maximum(i_a_hv / i_max_hv, i_a_lv / i_max_lv)*100, net.res_trafo_3ph.loading_a_percent)
    np.testing.assert_allclose(np.maximum(i_b_hv / i_max_hv, i_b_lv / i_max_lv)*100, net.res_trafo_3ph.loading_b_percent)
    np.testing.assert_allclose(np.maximum(i_c_hv / i_max_hv, i_c_lv / i_max_lv)*100, net.res_trafo_3ph.loading_c_percent)


def compare_result(actual, expected, *, rtol):
    np.testing.assert_allclose(actual.trafo.vn_hv_kv, expected.trafo.vn_hv_kv, rtol=rtol)
    np.testing.assert_allclose(actual.trafo.vn_lv_kv, expected.trafo.vn_lv_kv, rtol=rtol)
    np.testing.assert_allclose(actual.trafo.sn_mva, expected.trafo.sn_mva, rtol=rtol)
    np.testing.assert_allclose(actual.res_trafo_3ph.loc[:, 'i_a_hv_ka'], expected.res_trafo_3ph.loc[:, 'i_a_hv_ka'], rtol=rtol)
    np.testing.assert_allclose(actual.res_trafo_3ph.loc[:, 'i_b_hv_ka'], expected.res_trafo_3ph.loc[:, 'i_b_hv_ka'], rtol=rtol)
    np.testing.assert_allclose(actual.res_trafo_3ph.loc[:, 'i_c_hv_ka'], expected.res_trafo_3ph.loc[:, 'i_c_hv_ka'], rtol=rtol)
    np.testing.assert_allclose(actual.res_trafo_3ph.loc[:, 'i_a_lv_ka'], expected.res_trafo_3ph.loc[:, 'i_a_lv_ka'], rtol=rtol)
    np.testing.assert_allclose(actual.res_trafo_3ph.loc[:, 'i_b_lv_ka'], expected.res_trafo_3ph.loc[:, 'i_b_lv_ka'], rtol=rtol)
    np.testing.assert_allclose(actual.res_trafo_3ph.loc[:, 'i_c_lv_ka'], expected.res_trafo_3ph.loc[:, 'i_c_lv_ka'], rtol=rtol)
    np.testing.assert_allclose(actual.res_trafo_3ph.loading_a_percent, expected.res_trafo_3ph.loading_a_percent, rtol=rtol)
    np.testing.assert_allclose(actual.res_trafo_3ph.loading_b_percent, expected.res_trafo_3ph.loading_b_percent, rtol=rtol)
    np.testing.assert_allclose(actual.res_trafo_3ph.loading_c_percent, expected.res_trafo_3ph.loading_c_percent, rtol=rtol)


def test_pgm_pp_converter():
    pgm_net = create_net()
    pp_net = create_net()
    pp.runpp_pgm(pgm_net, symmetric=False)
    pp.runpp_3ph(pp_net)
    check_result(pgm_net)
    check_result(pp_net)
    compare_result(pgm_net, pp_net, rtol=0.04)


if __name__ == "__main__":
    test_pgm_pp_converter()

The current main has huge differences, while your branch passes, so your PR definitely solves the problem.

That said, I do still see some small differences between PGM and PP if I run that script (the compare_result step) of up to 4 percent (see the rtol), particularly in the b phase. However, this does not seem to be caused by the incorrect loading calculation, as the hand calculation seems to match for both calculation backends. Instead, the other result values seem to differ, but that could be caused by different local minima. I did not dig any further into this.

[mgovers]

So far, it seems like you're solving the issues nicely, so I don't think my help is needed as of now. If you do need some input from me, I'm happy to help.

[mgovers]

@furqan463 how is it going with the tests?

Since you found a bug and have already resolved it, I'm also OK with merging in its current state without extensive tests (after a minor clean-up). The tests can then be added as a follow-up. The reasoning behind that is that users can already use the PGM-IO with the bugfix.

What are your thoughts?

[furqan463]

@mgovers agreed. I couldn't find time to finalise the tests, but agreed we can merge without tests, as the changes are broadly conceptual.

[furqan463]

Functions pp_trafo_output and pp_switch_output not being tested for the same reason as was for pp_trafo_output_3ph, the functions are returning from first if because pp_data is not passed to converter. These two functions are also appearing missed in coverage, resulting in coverage of only 89% for this file. Now trying to fix this.

[furqan463]

@mgovers I'm done now. There's one unittest failing that I'm not sure how to resolve. As I understand the test check exactly how we calculate "tan1" in pgm_input_line, but I changed the formula to avoid Nan values.

[mgovers]

@mgovers I'm done now. There's one unittest failing that I'm not sure how to resolve. As I understand the test check exactly how we calculate "tan1" in pgm_input_line, but I changed the formula to avoid Nan values.

Awesome, @furqan463 , I'll have a look when i can 👍

[mgovers]

@nitbharambe looks like this PR is now in a good-to-go state. Since I've also contributed significantly, can you please review?

[mgovers]

OH apparently, CI is still failing; investigating.

[mgovers]

found the reason for the remaining validation tests.

To run the validation tests locally, please either use a test extension or run pytest tests/validation rather than just pytest, as the validation tests are excluded from pytest by default.

[furqan463]

@mgovers please check now, I've updated the data file.
TODOs added for future.

[mgovers]

@mgovers please check now, I've updated the data file. TODOs added for future.

Great! @nitbharambe and I will have a look on Monday.

• concerning the license file for the json data, you can copy-paste one of the other files such that foo.json has a file foo.json.license. The .license files should be the same across the codebase.
• please install the dev dependencies and run pre-commit run --all-files to run the formatter, linter and type checker. It will also run a check that checks whether you added the license file correctly.

[mgovers]

@furqan463 We're doing final review. Please do not add new commits to this branch so that we can merge if everything is OK 👍

[nitbharambe]

LGTM, Thanks for the contribution @furqan463!

[furqan463]

Thanks a lot @mgovers for you kind support and hand-holding, as this was my first experience with open source contribution.
Thanks @nitbharambe for your support as well.

[mgovers]

Anytime! Thank you very much for your contribution. You've been a great contributor. If you like, could you please share your experiences and maybe also any suggestions for improvements?