Direct Answer
The PCC (point of common coupling) THD exceeds the limit and the grid operator demands rectification within a deadline. Working through the plant bay by bay with "trial shutdowns" takes weeks and disrupts production. Is there a faster way? The Qianzhi engine embeds a harmonic fingerprint library covering 14 device classes (FP-01 three-phase rectifiers, FP-03 six-pulse VFDs, FP-05 UPS, FP-06 EV chargers, FP-12 PV inverters, and more). It matches each branch current harmonic spectrum against the fingerprints using cosine similarity; a score above 0.85 counts as a hit. A typical source hunt shrinks from weeks to about two hours. The fingerprint turns "who is the polluter" from guesswork into waveform comparison — no shutdowns, no trial disconnections.
A scenario: An industrial park PCC THD reached 8.6% (limit 5%) and the grid operator demanded correction within three months. The traditional approach was to disconnect each bay in turn and watch how the total harmonic content changed — 20 bays meant 20 disconnections, real production loss, and no guarantee of a result. Qianzhi collected one week of harmonic data at the main feeder and each branch CT, matched the spectra against the 14 fingerprint classes, and produced two hits (VFD at 0.91, EV charger at 0.87). The sources were located the same day and responsibility was attributed in about two hours.
1. Why Harmonic Source Hunting Is So Hard
Harmonics are invisible — unlike an oil leak or a hot spot, waveform distortion leaves no on-site trace and cannot be seen without an instrument; Harmonics superpose and mix — several devices emit harmonics at once, and what the bus shows is the summed result, not any single contributor; Harmonics drift — with load changes and operating-mode switches, the same device emits a different harmonic pattern at different times; Traditional hunting relies on trial shutdowns — disconnect and watch, which hurts production, carries operational risk, and does not guarantee a result; Responsibility remains contested — even after a device is found, who bears the responsibility (user side vs grid side) still has to be settled separately.
In one sentence: harmonic source hunting is hard because the distortion is invisible, additive, time-varying, and historically hunted by disconnection — fingerprints beat guesswork.
A trial-shutdown example: A plant with 18 bays exceeded the harmonic limit. Bay-by-bay disconnection took two weeks and 14 interruptions, finally pointing at the VFD bay — while the line lost roughly one shift of output during the hunt. Trial shutdown is destructive hunting — every disconnection is a bet and a loss; fingerprint comparison is non-intrusive.
2. What a Harmonic Fingerprint Is: 14 Device Signatures
Definition — each device class emits harmonics with its own "order combination + phase relation + amplitude ratio": like a human fingerprint, the device has a harmonic signature; Where fingerprints come from — generalization of measured waveforms: a large set of field data plus standardized tests is condensed into 14 typical classes; Dimensions of a fingerprint — more than harmonic order: relative amplitude of each order, phase angle, and how the pattern behaves as load changes; Value of a fingerprint — seen once, recognized again: the same class of device has a basically consistent signature wherever it is installed; Composition of the library — 14 typical classes: rectifiers, VFDs, UPS, EV chargers, PV inverters, arc devices, and others.
In one sentence: a harmonic fingerprint is the "order combination + phase relation + amplitude ratio" a device emits — 14 typical classes each have a signature card.
A fingerprint example: A six-pulse VFD shows typical harmonics at orders 5, 7, 11, 13 (the 6k±1 family); a three-phase rectifier is dominated by orders 5 and 7 with amplitude varying noticeably with load — the two fingerprints differ, so even superimposed they can be separated. Fingerprint comparison presupposes separability — different orders and phases can be told apart.
3. Typical Harmonic Orders of the 14 Device Classes
Three-phase rectifier (FP-01) — dominated by orders 5 and 7: current harmonics concentrate at low orders and the amplitude swings with load; Six-pulse VFD (FP-03) — the 6k±1 family: 5, 7, 11, 13, 17, 19 — a clear signature; Twelve-pulse VFD — the 12k±1 family: 11, 13, 23, 25 — relatively smaller high-order share; UPS (FP-05) — orders 5 and 7 plus a 3rd-order feature: produced by the input rectifier stage, with a large difference between light and full load; EV charger (FP-06) — orders 3, 5, 7, 9: high-power rectification mixed with switch-mode noise, clearly intermittent; PV inverter (FP-12) — high-order harmonics: 19, 23, 25, 31 and so on, concentrated near the switching frequency; Arc-type devices — a broad spectrum: not limited to specific orders, continuous and strongly fluctuating; Other classes — lighting, heating, motors, and more: each has its own features, and the library keeps expanding from field measurements.
| Device class | Fingerprint ID | Typical harmonic orders | Note |
|---|---|---|---|
| Three-phase rectifier | FP-01 | 5, 7 dominant | amplitude swings with load |
| Six-pulse VFD | FP-03 | 5, 7, 11, 13 (6k±1) | clear low-order signature |
| Twelve-pulse VFD | FP-04 | 11, 13, 23, 25 (12k±1) | smaller high-order share |
| UPS | FP-05 | 5, 7 plus 3rd-order feature | large light/full-load gap |
| EV charger | FP-06 | 3, 5, 7, 9 | clearly intermittent |
| PV inverter | FP-12 | 19, 23, 25, 31 etc. | concentrated near switching frequency |
In one sentence: each of the 14 classes has a set of "typical orders + characteristic notes" — the library fixes these features into comparable templates, and matching aligns on order, phase, and ratio.
A comparison example: A bus carries both a six-pulse VFD and an EV charger: the VFD contributes orders 5 and 7, the charger contributes 3, 5, 7 and 9 — orders 5 and 7 superpose, while orders 3 and 9 belong only to the charger — the "exclusive orders" separate the two devices. Fingerprint matching asks not only "who has what" but "whose pattern looks most like the bus".
4. Cosine Similarity: Sensitive to Shape, Insensitive to Amplitude
Definition — treat the harmonic spectrum as a vector: the cosine of the angle between two vectors measures "how similar the shapes are"; Why it is shape-sensitive — the distribution of orders defines the angle: a spectrum where orders 5 and 7 make up 80% differs clearly in angle from one where they make up 30%; Why it is amplitude-insensitive — cosine looks at direction, not length: if the load rises and the whole spectrum scales up by a factor, the direction stays the same and the cosine does not change; What the 0.85 threshold means — only a highly consistent shape counts as a hit: above 0.85 the order combination and phase relation are closely aligned; Guard against misjudgment — when multiple sources superpose, weighting by "exclusive features" avoids averaging two devices into a mistaken third one.
In one sentence: cosine similarity compares shape, not size — amplitude can change with load, but "what the harmonics look like" is the criterion.
An amplitude example: The same VFD at light load puts 8 A into the 5th order; at full load, 35 A — a fourfold swing, yet the cosine similarity against the fingerprint stays above 0.85 and the hit is still reported. Amplitude follows load; shape does not — that is why cosine similarity is not fooled by load swings.
5. The Four-Step Hunt: From PCC to Responsibility Attribution
Step 1 — measure the PCC: confirm the problem exists: which orders exceed the limit, by how much, and at what times of day — scope the hunt; Step 2 — branch spectra: collect every branch synchronously: main feeder plus branch CTs from the same time base for one week — the harmonic spectrum of each branch; Step 3 — fingerprint matching: run each branch spectrum against the 14 fingerprints with cosine similarity — a score above 0.85 flags a candidate source; Step 4 — responsibility attribution: verify and report: confirm the hit branch against the connected load, then issue the responsibility report.
| Step | Action | Time | Output |
|---|---|---|---|
| 1 | PCC THD and per-order baseline | 30 minutes | quantified over-limit list |
| 2 | synchronous branch spectra | 1 week (online) | per-branch spectrum curves |
| 3 | 14-class cosine matching | real time | candidate hit list |
| 4 | load verification + report | 30 minutes | responsibility report |
In one sentence: the four steps are measure the PCC, collect branch spectra, match fingerprints, and attribute responsibility — the first two gather evidence, the last two make the call — total time falls from weeks to about two hours.
A flow example: An industrial park — day 1 PCC baseline (THD 8.6%); from day 2, one week of online collection; day 8 fingerprint matching: branch 3 hits the six-pulse VFD at 0.91, branch 7 hits the EV charger at 0.87; load verification the same day — the responsibility report is completed in two hours. The week of online collection is evidence-gathering; matching and attribution are the fast actions — the data is slow, the algorithm is fast.
6. Why Weeks Shrink to Two Hours
The bottleneck of trial shutdowns — device by device: 20 branches mean 20 operations plus production impact — weeks is the norm; The speed-up of fingerprints — one collection, batch matching: a week of data is computed at once and every branch is compared simultaneously; The advantage of parallelism — production is not interrupted: collection is monitoring, not operation — the line keeps running during the hunt; The determinism of the algorithm — cosine similarity is reproducible: the same data gives the same result whoever runs it — the conclusion is auditable; The traceability of the report — waveforms, hit scores, and verification records support every link — grid operators accept it.
In one sentence: fingerprint hunting is one collection plus batch matching without stopping production — it turns a "disconnection queue" into a parallel comparison and drops the time from weeks to hours.
A time comparison: In the same park, traditional trial shutdowns took three weeks and four interruptions; fingerprint hunting took two hours for the matching step and zero interruptions. What is saved is not just time — it is production, it is trial disconnection, and the conclusion is traceable.
7. Link to Q-01 Harmonic Responsibility Attribution (IEEE 1459)
Locating is only the first step — after the source is found, whether the user side or the grid side bears responsibility must be settled per a standard; The idea of IEEE 1459 — decompose power into fundamental and harmonic components: meter fundamental and harmonic power separately — quantify responsibility; The role of the fingerprint library — fingerprints answer "who emits": waveform comparison identifies the source device; Q-01 answers "how much": harmonic power attribution per IEEE 1459; The linked flow — fingerprint hit → harmonic power decomposition → responsibility ratio → mitigation plan; Supporting standards — GB/T 14549-1993 sets the public-grid harmonic limits: whether the level is exceeded is evidence-based.
In one sentence: fingerprints identify "who is emitting" and IEEE 1459 quantifies "how much" — find the source first, then split responsibility, and the mitigation plan has a basis.
A linked example: In an industrial park, fingerprint matching located branch 3's VFD (0.91); the harmonic power decomposition per IEEE 1459 showed branch 3 contributes 68% of the total harmonic power; compared against the GB/T 14549-1993 limits, a "VFD-side over-limit" responsibility report was issued. Locate, attribute, and benchmark — one closed-loop report, and the grid discussion passes in one round.
8. Common Pitfalls: Fingerprint Matching Is Not Infallible
Pitfall 1 — load swings misread as device change: amplitude changes look like a new device; cosine similarity already handles this — compare shape, not amplitude; Pitfall 2 — multi-source confusion: two devices with similar features; use exclusive-feature weighting and segment matching to avoid averaging into a wrong hit; Pitfall 3 — unsynchronized sampling: branch and main meters on different time bases; synchronous collection and timestamp alignment make spectra comparable; Pitfall 4 — order-only matching: equal orders with opposite phases cancel; the library includes phase relations and compares at the waveform level; Pitfall 5 — a single time slice as the whole story: one moment does not represent the day; a week of data shows the behavioral pattern and a stable hit is the real one.
In one sentence: the five pitfalls are load-swing misjudgment, superposition confusion, unsynchronized sampling, ignored phase, and a one-slice view — the first three are data engineering, the last two are waveform-level comparison.
A pitfall example: One project misjudged once: two VFDs of the same brand had nearly identical features; averaged together the cosine dropped to 0.79 and both were missed — switching to "exclusive features + segment matching" gave 0.90 for each. The fingerprint library is not a static template — field variance is digested by feature engineering.
9. Back to YESLON: How the Harmonic Fingerprint Library Lands
Qianzhi engine M06 — ships with the 14-class harmonic fingerprint library: fingerprint matching is a standard analysis capability; Data source — branch CTs plus the main-feeder CT: online harmonic spectrum collection — no shutdowns, no trial disconnections; Output form — a harmonic-source report: hit device, cosine score, responsibility ratio, and mitigation recommendations; Coordination with Taiyi — the Taiyi central control system aggregates data: Qianzhi diagnoses and Taiyi dispatches — a detect-to-act closed loop; Alignment with standards — GB/T 14549-1993 limit comparison plus IEEE 1459 responsibility attribution: the conclusion holds up.
In one sentence: the Qianzhi harmonic fingerprint library is 14 templates plus cosine matching plus a responsibility report — harmonic hunting moves from weeks of trial shutdowns to a two-hour location.
A landing example: An industrial park PCC exceeded the THD limit; Qianzhi took one week of branch-CT data, matched the fingerprints, and hit a VFD and an EV charger; the responsibility report was issued; after mitigation devices were installed per the recommendations, THD fell to 4.2%. The fingerprint library is not a one-shot check — it keeps growing with field data and gets better with use.
Scope and Limitations
- The conclusions in this article are intended for LV distribution and electrical-safety sites. Standard numbers (GB/T 14549-1993) refer to their currently effective editions, and product parameters follow the latest FEXLINK selection handbook and product knowledge base.
- Quantitative criteria (protection zone, discharge and residual voltage, earth resistance, thresholds and grading) hold under the preconditions stated in this article. Where the site voltage level, earthing arrangement, environmental category or load profile differs, the figures must be recalculated. Specific configurations involving Qianzhi engine M06, harmonic fingerprint library should be verified against actual circuit parameters.
- Specific product models and parameters are subject to the latest selection handbook and product knowledge base; this article does not constitute selection or procurement advice.
- Standard numbers cited here refer to the currently effective editions; when a standard is revised, the related quantitative conclusions must be re-checked.
- Deployment and implementation for specific projects must be carried out by qualified engineers based on site survey and design drawings.
Frequently Asked Questions
The Harmonic Fingerprint Library
The PCC (point of common coupling) THD exceeds the limit and the grid operator demands rectification within a deadline