Passive Identification of Unequal Phase Grid Impedances by Phase-Domain Source Cancellation

arXiv:2610.38131 · eess.SY, cs.SY · Submitted 2026-10-28 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: We're looking at the paper: "Passive Identification of Unequal Phase Grid Impedances by Phase-Domain Source Cancellation".

Dev: Here's the gist of the paper. Passive grid impedance estimators that return a single resistance–inductance pair cannot represent a grid whose phase impedances are unequal The paper identifies a separate pair for each phase from recorded point-of-common-coupling (PCC) voltage and current without injected perturbations.

Rosa: Phase-Domain Regressions The core methodology involves using a periodic difference (PD) or a second-order-generalized-integrator notch difference (SOGID) to cancel the unknown grid-source voltage in each phase.

Dev: Moving-window integration (MWI) is employed to replace the current derivative by an integral and an endpoint feature.

Rosa: The resulting two-parameter regression requires a period or notch-frequency estimate but no phase-locked-loop angle.

Paper summary: Rosa: To recap, this paper tackles the fact that standard passive estimators only give you one resistance and inductance pair, which isn't enough for grids with unequal phase impedances >

Dev: And the authors of "Passive Identification of Unequal Phase Grid Impedances by Phase-Domain Source Cancellation" propose identifying a separate pair for each phase using recorded point-of-common-coupling voltage and current without injecting any perturbations into the system >

Taro: The core methodology they use involves canceling out the unknown grid-source voltage in each phase either with a periodic difference or PD or this second order generalized integrator notch difference SOGID >

Rosa: And to handle the current derivative, they replace it with an integral and an endpoint feature using moving window integration or MWI >

Dev: So the resulting two parameter regression only needs an estimate of the period or notch frequency, not a phase locked loop angle to function >

Taro: They then use voltage-balance analysis to bound the error by comparing that residual source error with how well those two regression features are conditioned >

Rosa: The results show that in a ten second EMTP grid forming record, SOGID moving window integration achieves phase averaged R/L errors of five point zero nine divided by three point three percent during an unequal phase interval from eight point three seconds to nine point eight seconds > <ref:2610.38131#pg2>

Dev: They also noted a shared value error of forty-one point six seven percent for the phase ratio one five two in that same record >

Taro: Furthermore, they tested these estimators on both a grid forming record and a constant impedance grid following record, showing that the estimates can be retained through a synchronization disturbance in the latter case >

Rosa: And finally, when they looked at noise sensitivity, using a noise-referenced update rule reduced the white noise grid following error from fifty five point nine divided by sixty five point zero percent down to twelve point three divided by eleven point five percent >

Conclusion: Dev: Thinking about the title again, "Passive Identification of Unequal Phase Grid Impedances by Phase-Domain Source Cancellation," it points toward a method that doesn't rely on perfect phase matching >

Rosa: It means we can move beyond just getting one lumped resistance and inductance value for the whole grid when we are actually dealing with multiple phases >

Taro: The implication is that for applications where you need high fidelity modeling, especially in power electronics interfaces, this technique offers a path to separate those parameters without needing perfect phase knowledge >

Dev: It suggests that even if the underlying system isn't perfectly synchronized, as long as the frequency mismatch doesn't get too big, we can still get useful R and L information >

Rosa: The authors managed to achieve this by carefully designing their cancellation operators and replacing the derivative with an integral feature without needing a phase locked loop angle in the regression >

Taro: So, it’s about using specific mathematical structures to isolate the unknown source voltage and then integrate the current measurement over a window >

Dev: It's about making sure that these errors are bounded by how much frequency mismatch is present and how sensitive our two regression features are to noise or other things >

Rosa: Overall, this work shows a way to extract multiple R/L pairs passively from PCC data in unequal phase grids using these advanced cancellation methods >

Songyang Zhanga, Xun Gongb, Bo Caob, Reza Iravanic

Department of Electrical and Computer Engineering, University of Alberta · Montreal Research Centre, Huawei Technologies Canada Co., Ltd. · The Edward S. Rogers Sr. Department of Electrical and Computer Engineering, University of Toronto

eess.SY, cs.SY

Submitted: 2026-10-28

Updated: 2026-11-09

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 97/100

The gist: Passive grid impedance estimators that return a single resistance–inductance pair cannot represent a grid whose phase impedances are unequal The paper identifies a separate pair for each phase from…

Key concepts

Passive Grid Impedance Estimators
These are methods that try to determine the resistance and inductance of a power grid without injecting any external signals or perturbations into the system. The paper shows that these simple estimators cannot accurately capture grids where different phases have different impedance characteristics, which is a key problem they solve.
Periodic Difference (PD) / SOGID
These are mathematical tools used to systematically cancel out the unknown voltage source present in each phase of the grid. By using these differences, the researchers can isolate and estimate the true grid parameters without needing a phase-locked loop angle, which simplifies the estimation process.
Moving-Window Integration (MWI)
This technique replaces estimating a current derivative with integrating the current over a moving window and using an endpoint feature. This provides a more robust way to handle noisy measurements, allowing the regression to work effectively even when dealing with complex grid dynamics.

Terminology

Summary

Passive grid impedance estimators that return a single resistance–inductance pair cannot represent a grid whose phase impedances are unequal The paper identifies a separate pair for each phase from recorded point-of-common-coupling (PCC) voltage and current without injected perturbations. p. 0

Phase-Domain Regressions

The core methodology involves using a periodic difference (PD) or a second-order-generalized-integrator notch difference (SOGID) to cancel the unknown grid-source voltage in each phase. Moving-window integration (MWI) is employed to replace the current derivative by an integral and an endpoint feature. p. 0 The resulting two-parameter regression requires a period or notch-frequency estimate but no phase-locked-loop angle. p. 0

Error Analysis

A voltage-balance analysis provides the source residual left by frequency mismatch and bounds the parameter error by the ratio of this residual to the conditioning of the two features. The paper demonstrates that in an unequal-phase interval (8. p. 03–9.8 s) of a 10-s EMTP grid-forming record, SOGID-MWI attains phase-averaged R/L errors of 5. p. 009/0.33%.

Phase Resolution and Robustness

The study evaluates the estimators on two types of records: a GFM record where unequal phase impedances are present, and a constant-impedance GFL record where estimates are retained through a synchronization disturbance. The unequal-phase interval shows the value of separate parameters, with R and L references having ratios 1: 0. p. 45: 2.

Noise Sensitivity and Performance

A controlled noise study showed that a noise-referenced update rule reduces the white-noise GFL error from 55.9/65.0% to 12.3/11.5%. The rank-memory update rule, when applied to the white-noise GFL record, suppresses quiet-window updates and reduces the error from 55. p. 992/65.01% (RLS) to 12. p. 027/11.48%.

Conclusion

The paper concludes that PD and SOGID cancel the periodic grid source, MWI replaces the current derivative by an integral and an endpoint feature, and no PLL angle enters the regression. The voltage-balance analysis relates the parameter error to the source residual left by frequency mismatch and to the conditioning of the two features. p. 9 Over 8.3–9.8 s of the GFM record, SOGID-MWI resolved phase impedances in the ratio 1: 0. p. 95: 2 with 5.09/0.33% R/L error. p. 0 The estimates were retained through a synchronization disturbance in the constant-impedance GFL record. p. 9 The noise-referenced rule reduced the 40-dB white-noise GFL error by 78/82% relative to fixed-forgetting RLS. p. 9 The present evaluation uses a diagonal R–L equivalent at nominal frequency. p. 9 The present phase model supplies three independent R/L pairs. p. 9

--- Page 9 ---

The paper is 2610.38131 EVERY sentence must end with a reference token The single most informative sentence about the paper's subject and finding, with no reference to the paper by its title alone and no reference to your own answer is: Passive grid impedance estimators that return a single resistance–inductance pair cannot represent a grid whose phase impedances are unequal.

--- Page 5 ---

The paper is 2610.38131 EVERY sentence must end with a reference token The single most informative sentence about the paper's subject and finding, with no reference to the paper by its title alone and no reference to your own answer is: Passive grid impedance estimators that return a single resistance–inductance pair cannot represent a grid whose phase impedances are unequal.

--- Page 7 ---

The paper is 2610.38131 EVERY sentence must end with a reference token The single most informative sentence about the paper's subject and finding, with no reference to the paper by its title alone and no reference to your own answer is: Passive grid impedance estimators that return a single resistance–inductance pair cannot represent a grid whose phase impedances are unequal.

--- Page 4 ---

The paper is 2610.38131 EVERY sentence must end with a reference token The single most informative sentence about the paper's subject and finding, with no reference to the paper by its title alone and no reference to your own answer is: Passive grid impedance estimators that return a single resistance–inductance pair cannot represent a grid whose phase impedances are unequal.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Phase-domain regressions can identify unequal phase impedances by using a periodic difference (PD) or a second-order-generalized-integrator notch difference (SOGID) to cancel the unknown grid-source voltage in each phase without needing a phase-locked loop angle.

  2. Moving window integration replaces the current derivative with an integral and an endpoint feature, allowing for identification using only a period or notch-frequency estimate, thus removing the dependence on PLL angle dynamics.

  3. A voltage-balance analysis bounds the parameter error by the ratio of this residual to the conditioning of the two features, providing a rigorous method to assess identifiability based on recorded transients.

  4. The system can retain estimates through synchronization disturbances in constant-impedance grid-following records, demonstrating robustness against frequency mismatches and transient events.

  5. A noise-referenced update rule can reduce white-noise GFL error from 55.9/65.0% to 12.3/11.5%, improving estimation accuracy when measurement noise is present during steady operation.

  6. Excitation-aware adaptive RLS (EAARLS) includes rules that accept an update only when a measure of the regressor exceeds a threshold, allowing the system to decide when a new regression sample is used.

  7. Rank-memory testing can suppress quiet-window updates and reduce error from 55.92/65.01% (RLS) to 12.27/11.48% under white noise conditions by comparing the two regression features against expected noise levels.

Related papers