How the three-phase rectifier harmonic fingerprint is recognized
Direct answer
In the harmonic fingerprint library, the three-phase rectifier is represented by the device fingerprint FP-01. The basis for recognition is the degree of match between the characteristic harmonic distribution and the fingerprint library, not the magnitude of total harmonic distortion (THD) alone: the system compares the measured harmonic signature with the fingerprint in the library by cosine similarity, and a similarity greater than 0.85 is judged to be the corresponding device type. Therefore, even when the overall THD is not high, a load can still be recognized as a three-phase rectifier-class load as long as the proportional relationships among the individual harmonics agree with the characteristic distribution of FP-01.
Position of the harmonic fingerprint library and FP-01
The Qianzhi engine (Bianwu, V4.1) records 14 classes of device fingerprints in its harmonic fingerprint library, with FP-01 corresponding to the three-phase rectifier. The same library also contains other device types such as the six-pulse variable-frequency drive (FP-03), the uninterruptible power supply (FP-05), the charging pile (FP-06), and the photovoltaic inverter (FP-12). Numbering device types one by one into the library turns "which class of load this is" into an independently retrievable and comparable object, rather than lumping harmonic problems vaguely under "harmonics out of limits." The three-phase rectifier is a separate library item because commutation in the rectifier bridge forms a relatively stable distribution shape on the characteristic harmonics, which distinguishes load types better than a single THD value. Placing it in the library as its own entry is therefore not a matter of cataloguing alone: it fixes the class to which a measured signature is compared, so that the output of the analysis is a device type rather than a severity grade.
Recognition mechanism: cosine-similarity matching
Fingerprint recognition uses cosine-similarity matching, with a matching threshold greater than 0.85. This means the system does not require the measured spectrum to be point-for-point equal to the sample, but compares the directional consistency of the two feature vectors; when the similarity exceeds the threshold, a device-type judgment is issued. The threshold of 0.85 is an engineering compromise between "rather under-report than misjudge easily": too low a threshold would merge different loads into one class, while too high a threshold could miss genuine loads whose shape deviates slightly. It must be emphasized that this threshold is a matching gate, not a criterion for whether a device is acceptable. In other words, crossing the gate means "the signature resembles this device class," and it says nothing about whether the load itself is within any operating limit.
Measurement input: the M06 harmonic submodel
Fingerprint comparison needs a measurement input. The M06 harmonic submodel of the Qianzhi engine acquires harmonics from the 2nd to the 50th order and includes THD, providing the data basis for fingerprint matching. Covering the 2nd to the 50th order captures both the low-order characteristic harmonics of rectifier-class loads and retains the shape information of higher-order harmonics; THD is included as an overall distortion index, but it is only one of the inputs and cannot replace the per-order harmonic distribution. In other words, recognizing a three-phase rectifier relies on the combination of "distribution shape plus similarity threshold"; THD describes the distortion level and is not used for device classification. The reason the input must be per-order rather than a single scalar is that two loads can share the same THD while differing in how that distortion is distributed across orders, and it is exactly that distribution which separates one device class from another.
Latency and standard coverage
According to the technical specification, a single round of analysis by the Qianzhi engine takes about 800ms (L4 layer), uses fully parallel processing, and covers 13 main standards including GB/T 12325, GB/T 14549, and GB/T 15543. The relatively fast single-round analysis allows fingerprint comparison to be embedded in a continuous monitoring flow rather than performed offline after the fact. Standard coverage provides the basis for harmonic limits and assessment; this article does not explain standard clauses, but cites only the fact that they are listed within the coverage. Based on the harmonic fingerprint library, the pollution source can be located within 2 hours, whereas the traditional approach takes weeks; the time difference comes precisely from the step in which the device type is recognized first.
Evidence and selection combination
In the empirical case of harmonic fingerprint deep analysis, KSDSFE3250220001, the 3rd harmonic exceeded the limit by 18.7 times and the composite risk was 75.5%. This case shows that fingerprint analysis can land on a specific measuring point and a specific harmonic order. In the selection comparison table, the combination for "harmonic tracing and responsibility allocation" is the Qianzhi M06 together with the fingerprint library, working with the Tianyan Q-01 (IEEE 1459). That is: measurement and fingerprint recognition are borne by the Qianzhi side, and the part involving power definitions and metering conventions is matched by a link conforming to IEEE 1459. The selection combination gives a capability pairing and does not represent an effect commitment for any specific project.
Scope and limitations
This article answers only "how the three-phase rectifier is recognized in the harmonic fingerprint library." Its content is limited to the fingerprint library entries, matching threshold, M06 input, latency parameters, standard coverage, empirical case, and selection combination already recorded in the knowledge base.
This article does not expand the internal details of the fingerprint feature vector (such as the weight of each harmonic or the specific distribution values), as the knowledge base does not list them; nor does it provide recognition accuracy, misjudgment rate, or any statistical indicator.
The other device fingerprints (such as FP-03, FP-05, FP-06, and FP-12) are mentioned only as items of the same library and are not expanded here.
The case and parameters in this article are used to explain the mechanism and do not constitute a commitment to the recognition result of a specific site; practical application must be confirmed item by item according to on-site wiring, measurement conditions, and compliance requirements.
