Fan model identification and authentication method based on start transient multi-channel current fingerprint

Abstract

The application discloses a fan model identification and authentication method based on a starting transient multi-channel current fingerprint in the field of motor and fan equipment state perception and identity authentication, which comprises the following steps: obtaining current signal data of a fan starting process, filtering invalid sections and positioning effective sections to obtain an effective section starting point, and intercepting or retaining the current signal data after the effective section starting point to obtain effective current signal data; detecting a starting point and taking the detected starting point as a timing alignment reference to perform timing alignment; synchronously intercepting a fixed-length starting window, extracting waveform features and envelope features in the starting window, and constructing a fingerprint vector; establishing a model library; identifying and making an authentication decision, and outputting an authentication decision result of passing authentication, rejection or unknown according to an authentication threshold. The application automatically filters invalid data, detects a starting time and aligns, constructs a fingerprint vector with distinguishability, realizes model identification, and outputs through threshold decision.

Description

Technical Field

[0001] This application belongs to the field of state perception and identity authentication of motors and fan equipment, specifically involving a fan model identification and authentication method based on startup transient multi-channel current fingerprint. Background Technology

[0002] Current wind turbine equipment model identification relies heavily on nameplate information, software configuration files, or device IDs reported via communication protocols, which suffers from issues such as susceptibility to replacement / forgery and low efficiency in on-site verification. On the other hand, the current signal during wind turbine startup reflects the coupling characteristics of the motor, inverter, control strategy, and load, possessing the potential to form a "physical fingerprint." However, directly using the startup current for comparison is affected by the following factors: 1) It often contains invalid or junk data before power-on, such as 0 or fixed sentinel values; 2) Inconsistent acquisition trigger times at different times lead to waveform misalignment; 3) Single-channel or simple statistical features are insufficient to balance robustness and discriminative power. Therefore, a multi-channel current fingerprint construction and authentication method for wind turbine startup transients is needed to achieve repeatable and deployable model identification and access determination. Summary of the Invention

[0003] To address the aforementioned technical issues, this application proposes a wind turbine model identification and authentication method based on startup transient multi-channel current fingerprint. This method can automatically filter invalid data, detect and align startup times, construct a fingerprint vector with discriminative power, achieve model identification, and output pass, reject, or unknown results through threshold decision.

[0004] To achieve the above objectives, this application employs the following technical solution:

[0005] This application discloses a method for wind turbine model identification and authentication based on startup transient multi-channel current fingerprinting, which specifically includes the following steps:

[0006] Step 1, Data Acquisition: Obtain current signal data during the wind turbine startup process. The current signal data includes... shaft current and three-phase current Let the sampling frequency be... The number of sampling points recorded per startup is Then time index ,in ;

[0007] Step 2: Perform invalid segment filtering and valid segment location on the raw current signal data obtained in Step 1 to obtain the starting point of the valid segment. And based on the starting point of the valid segment Extract or retain the starting point of the valid segment The subsequent current signal data yields the effective current signal data;

[0008] Step 3: Based on the effective current signal data obtained in Step 2, perform start-up point detection and use the detected start-up point as the timing alignment reference to perform timing alignment on the start-up current signal data of each channel.

[0009] Step 4: Using the starting point detected in Step 3 as the alignment reference, synchronously extract a starting window of fixed length from each current signal data.

[0010] Step 5: Extract waveform features and envelope features within the startup window obtained in Step 4, and fuse the waveform features and envelope features to construct a fingerprint vector;

[0011] Step 6: Based on the fingerprint vector obtained in Step 5, perform standardization and dimensionality reduction embedding to obtain a low-dimensional representation, and calculate the template model according to the low-dimensional representation corresponding to different wind turbine models to establish a model library;

[0012] Step 7, Identification and Authentication Decision: Input a new set of samples to construct fingerprint vectors according to steps 1 to 5, and use the low-dimensional representation and model library trained in step 6 to obtain the embedding representation of the test samples; calculate the Euclidean distance from the embedding representation of the test samples to each template model, output the model identification result according to the principle of minimum Euclidean distance, and output the authentication decision result of pass authentication, rejection or unknown according to the authentication threshold.

[0013] A further improvement in this application is that step 2 specifically includes the following steps:

[0014] Step 2.1: Preprocess the raw current signal data obtained in Step 1, skipping the initial... By sampling points, signal samples are obtained, where... To skip a fixed number of points;

[0015] Step 2.2, Sentinel Value and Invalid Value Removal: Let the set of invalid values ​​be... If the values ​​of consecutive sampling points in the signal samples obtained in step 2.1 belong to the set of invalid values... And the continuous length is not less than the minimum length of the continuous invalid segment. If so, the continuous interval is judged as an invalid segment and removed;

[0016] Step 2.3, Low variance judgment: based on the sliding window length Calculate the standard deviation of signal samples within the sliding window. :

[0017]

[0018] The average value within the sliding window for:

[0019]

[0020] in, Indicates from the first The standard deviation of the signal samples within a sliding window starting from a sampling point is used to characterize the degree of fluctuation of the current signal within that window. This indicates the length of the sliding window, which is the number of sampling points included in each calculation of the standard deviation; This represents the offset of the sampling point within the sliding window, and its value ranges from 0 to... ; Indicates from the first Within the sliding window starting from the sampling point, the first... The signal amplitude corresponding to each sampling point; Indicates from the first Sampling point start, length is The average value of all signal samples within the sliding window;

[0021] Step 2.4: Based on the starting point of the valid segment Extract or retain the starting point of the valid segment The subsequent current signal data yields the valid current signal data.

[0022] A further improvement in this application is that step 3 specifically includes the following steps:

[0023] Step 3.1: Based on the three-phase current in Step 1 Constructing a synthesized three-phase amplitude signal:

[0024]

[0025] Step 3.2: In the three-phase amplitude synthesis signal The root mean square (RMS) energy envelope of the sliding window is calculated. Detect the start point :when Duration not less than When the first satisfaction is determined, the starting point is identified. ;

[0026] Step 3.3, will by Alignment is performed for reference, where, Energy threshold For the minimum duration, , To detect the start point The corresponding sampling points.

[0027] A further improvement in this application is that step 4 specifically involves: from the sampling point... The starting cut length is Startup window :

[0028]

[0029] in, For window duration, This represents the number of corresponding sampling points.

[0030] A further improvement in this application is that step 5 specifically includes the following steps:

[0031] Step 5.1: Perform mean-reduction and normalization processing on the current signal data of each channel within the startup window to obtain the waveform characteristics of the corresponding channel. Simultaneously calculate Hilbert envelope features :

[0032]

[0033] in, This indicates the current signal of any channel within the start window. They represent respectively to The waveform features obtained after mean removal and normalization processing This is the Hilbert transform; Indicates amplitude operation;

[0034] Step 5.2: Extract waveform features The fingerprint vector is formed by concatenating the envelope features in a fixed order:

[0035]

[0036] in, They represent Envelope characteristics of shaft current and three-phase current.

[0037] A further improvement in this application is that step 6 specifically involves: processing the fingerprint vector obtained in step 5... Standardization is performed, and a low-dimensional representation is obtained by dimensionality reduction embedding. For each wind turbine model Low-dimensional representation of sets Computational class center as template model:

[0038]

[0039] in, This represents the low-dimensional representation of the fingerprint vector after standardization and dimensionality reduction embedding. Indicates the first The set of low-dimensional representations corresponding to each wind turbine model; Represents a set The number of low- to mid-dimensional representations; Indicates the first Template models for each wind turbine model, i.e., a set The class centers of each low-dimensional representation.

[0040] A further improvement in this application is that step 7 specifically involves:

[0041] Step 7.1, Identification: For the wind turbine to be identified, the fingerprint vector of the wind turbine is standardized and dimensionality-reduced by embedding in step 6 to obtain a low-dimensional representation. Calculate low-dimensional representation To each template model Euclidean distance:

[0042] ,

[0043] in, Low-dimensional representation of the wind turbine to be identified With the Template model of a fan model The Euclidean distance between them;

[0044] Step 7.2 Authentication: If the fan to be identified declares its model as... Then calculate its low-dimensional representation. To declare the model template distance And the closest distance to other template models. And calculate the authentication interval.

[0045]

[0046] when If authentication is successful, output "authentication passed"; otherwise, output "rejected" or "unknown".

[0047] in, This indicates the authentication interval, used to measure the credibility of the wind turbine to be identified relative to the claimed model. For authentication threshold

[0048] The beneficial effects of this application are:

[0049] This application is highly robust: it automatically filters invalid or junk data before power-on and performs startup alignment, reducing the impact of manual processing and waveform misalignment.

[0050] This application has high distinguishability: it utilizes direct current. By combining multi-channel information of three-phase current with waveform and envelope features, the separability of different types of wind turbines can be improved.

[0051] This application supports both identification and authentication: it not only outputs model results, but also provides admission decisions based on margin and threshold, which can be used for acceptance, anti-substitution and anti-counterfeiting.

[0052] This application is deployable: it requires minimal computation and can be processed in real time on a host computer or edge device;

[0053] This application is scalable: it can be extended to more models and more channels, such as speed, q-axis current, voltage, etc., to improve generalization ability. Attached Figure Description

[0054] Figure 1 This is a flowchart illustrating the wind turbine model identification and authentication method of this application.

[0055] Figure 2 This is a flowchart illustrating steps 2-4 of this application.

[0056] Figure 3 This is a flowchart illustrating step 5 of this application, which involves constructing the fingerprint vector.

[0057] Figure 4 This is a schematic diagram of step 7, identification and authentication decision, in this application.

[0058] Figure 5 This is a schematic diagram of the confusion matrix of the experimental results of this application.

[0059] Figure 6 This is the certification threshold for this application. A schematic diagram of the curve. Detailed Implementation

[0060] The embodiments of the present invention will be disclosed below with reference to the drawings. For clarity, many practical details will be described in the following description. However, it should be understood that these practical details are not intended to limit the present invention. That is, in some embodiments of the present invention, these practical details are not essential. In addition, for the sake of simplicity, some conventional structures and components will be shown in the drawings in a simple schematic manner.

[0061] like Figure 1 As shown, this application discloses a wind turbine model identification and authentication method based on startup transient multi-channel current fingerprinting, which specifically includes the following steps:

[0062] Step 1, Data Acquisition: Obtain current signal data during the wind turbine startup process. The current signal data includes... shaft current and three-phase current Let the sampling frequency be... The number of sampling points recorded per startup is Then time index ,in ;

[0063] Step 2: Perform invalid segment filtering and valid segment location on the raw current signal data obtained in Step 1 to obtain the starting point of the valid segment. And based on the starting point of the valid segment Extract or retain the starting point of the valid segment The subsequent current signal data yields the valid current signal data.

[0064] like Figure 2 As shown, step 2 specifically includes the following steps:

[0065] Step 2.1: Preprocess the raw current signal data obtained in Step 1, skipping the initial... 1 sampling point is used to quickly remove obvious invalid prefixes and obtain signal samples, where To skip a fixed number of points;

[0066] Step 2.2, Sentinel Value and Invalid Value Removal: Let the set of invalid values ​​be... The set of invalid values The sample consists of pre-set abnormal marker values, empty sample values, and invalid sample values ​​determined in the experiment, including at least one of the following: 0 value, fixed sentinel value, and abnormal values ​​exceeding the normal acquisition range. If the values ​​of consecutive sample points in the signal sample obtained in step 2.1 belong to the invalid value set... And the continuous length is not less than the minimum length of the continuous invalid segment. If so, the continuous interval is judged as an invalid segment and removed;

[0067] Step 2.3, Low variance judgment: based on the sliding window length Calculate the standard deviation of signal samples within the sliding window. :

[0068]

[0069] The average value within the sliding window for:

[0070]

[0071] in, Indicates from the first The standard deviation of the signal samples within a sliding window starting from a sampling point is used to characterize the degree of fluctuation of the current signal within that window. This indicates the length of the sliding window, which is the number of sampling points included in each calculation of the standard deviation; This represents the offset of the sampling point within the sliding window, and its value ranges from 0 to... ; Indicates from the first Within the sliding window starting from the sampling point, the first... The signal amplitude corresponding to each sampling point; Indicates from the first Sampling point start, length is The average value of all signal samples within the sliding window;

[0072] Step 2.4: Based on the starting point of the valid segment Extract or retain the starting point of the valid segment The subsequent current signal data yields the valid current signal data.

[0073] Step 3: Based on the effective current signal data obtained in Step 2, perform start-up point detection and use the detected start-up point as the timing alignment reference to perform timing alignment on the start-up current signal data of each channel.

[0074] like Figure 1-2 As shown, step 3 specifically includes the following steps:

[0075] Step 3.1: Based on the three-phase current in Step 1 Constructing a synthesized three-phase amplitude signal (to enhance immunity to single-phase noise / phase difference):

[0076]

[0077] Step 3.2: In the three-phase amplitude synthesis signal The root mean square (RMS) energy envelope of the sliding window is calculated. Detect the start point :when Duration not less than When the first satisfaction is determined, the starting point is identified. ;

[0078] Step 3.3, will by Alignment is performed for reference, where, Energy threshold Minimum duration (to avoid false detection of noise spikes). , To detect the start point The corresponding sampling points.

[0079] Step 4: Using the starting point detected in Step 3 as the alignment reference, synchronously extract a fixed-length starting window from each current signal data; specifically: from the sampling point The starting cut length is Startup window :

[0080]

[0081] in, For window duration, This represents the number of corresponding sampling points.

[0082] Step 5: Extract waveform features and envelope features from the startup window obtained in step 4, and fuse the waveform features and envelope features to construct a fingerprint vector.

[0083] like Figure 3 As shown, step 5 specifically includes the following steps:

[0084] Step 5.1: Perform mean-reduction and normalization processing on the current signal data of each channel within the startup window to obtain the waveform characteristics of the corresponding channel. Simultaneously calculate Hilbert envelope features :

[0085]

[0086] in, This represents the current signal of any channel within the startup window, specifically... shaft current or three-phase current ; They represent respectively to The waveform features obtained after mean removal and normalization processing This is the Hilbert transform; Indicates amplitude operation;

[0087] Step 5.2: Extract waveform features The fingerprint vector is formed by concatenating the envelope features in a fixed order:

[0088]

[0089] in, They represent Envelope characteristics of shaft current and three-phase current.

[0090] Step 6: Based on the fingerprint vector obtained in Step 5, perform standardization and dimensionality reduction embedding to obtain a low-dimensional representation, and calculate the template model according to the low-dimensional representation corresponding to different wind turbine models to establish a model library;

[0091] Specifically, the fingerprint vector obtained in step 5 is... Standardization is performed, for example, z-score standardization based on the mean / variance of the training set, and dimensionality reduction embedding is used to obtain a low-dimensional representation. For each wind turbine model Low-dimensional representation of sets Computational class center as template model:

[0092]

[0093] This represents the low-dimensional representation of the fingerprint vector after standardization and dimensionality reduction embedding. Indicates the first The set of low-dimensional representations corresponding to each wind turbine model; Represents a set The number of low- to mid-dimensional representations; Indicates the first Template models for each wind turbine model, i.e., a set The class centers of each low-dimensional representation.

[0094] Step 7, Identification and Authentication Decision: Input a new set of samples to construct fingerprint vectors according to steps 1 to 5, and use the low-dimensional representation and model library trained in step 6 to obtain the embedding representation of the test samples; calculate the Euclidean distance from the embedding representation of the test samples to each template model, output the model identification result according to the principle of minimum Euclidean distance, and output the authentication decision result of pass authentication, rejection or unknown according to the authentication threshold.

[0095] like Figure 4 As shown, step 7 specifically involves:

[0096] Step 7.1, Identification: For the wind turbine to be identified, the fingerprint vector of the wind turbine is standardized and dimensionality-reduced by embedding in step 6 to obtain a low-dimensional representation. Calculate low-dimensional representation To each template model Euclidean distance:

[0097] ,

[0098] in, Low-dimensional representation of the wind turbine to be identified With the Template model of a fan model The Euclidean distance between them;

[0099] Step 7.2 Authentication: If the fan to be identified declares its model as... Then calculate its low-dimensional representation. To declare the model template distance And the closest distance to other template models. And calculate the authentication interval.

[0100]

[0101] when The system outputs "Authentication successful" if authentication is successful, otherwise it outputs "Rejected" or "Unknown". This indicates the authentication interval, used to measure the credibility of the wind turbine to be identified relative to the claimed model. The authentication threshold; the authentication threshold Obtained through development set scanning, or by selecting a more conservative threshold based on the target FAR upper limit.

[0102] To verify this application, for the identification and authentication of two types of fans (1), the starting current data of models A and B fans were collected, each containing With three-phase current (2) Perform invalid segment filtering on each record: skip the first few points, remove consecutive sentinel values, and use a sliding window standard deviation to locate valid segments. (3) Construct And detect the start point (4) After alignment, a startup window of length T=120ms is truncated. (5) The mean of each channel window waveform is removed and normalized, the Hilbert envelope is calculated and normalized, and the fingerprint vector is generated by splicing. (6) Training phase: The fingerprint is standardized and PCA is used for dimensionality reduction, and the class center template of each model is calculated. (7) Testing phase: The fingerprint is generated for the new startup data and projected into the PCA space. The nearest center is output as the model; the margin is calculated and compared with the threshold. Compare the output authentication results (pass / reject / unknown).

[0103] Figure 5 This is a schematic diagram of the confusion matrix of the experimental results of this application. The vertical axis represents the true label, i.e., the actual model of the wind turbine, and the horizontal axis represents the predicted label, i.e., the model the model the wind turbine was identified as. Each number indicates the frequency of the corresponding situation; the darker the color, the larger the sample size. According to... Figure 5 It can be seen that the identification results for the two wind turbine models, 4114 and 17251, are mainly concentrated in the diagonal positions, indicating that most samples were correctly identified. Only a small number of misidentified samples exist in the off-diagonal positions. For example, among the samples whose true model is 4114, 146 were correctly identified as 4114, but 4 were misidentified as 17251. This demonstrates that the startup transient multi-channel current fingerprint constructed in this application has good model differentiation capabilities.

[0104] Figure 6 This is a schematic diagram of the authentication threshold (FAR / FRR) curve for this application, where FAR represents the false acceptance rate and FRR represents the false rejection rate. According to Figure 6 It can be seen that, with the authentication threshold The changes in FAR and FRR show opposite trends. Near the same point, FAR and FRR are quite close, corresponding to an EER of about 5%, indicating that this threshold can achieve a good balance between false acceptance and false rejection, thus realizing the authentication decision after wind turbine model identification.

[0105] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A method for wind turbine model identification and authentication based on start-up transient multi-channel current fingerprinting, characterized in that: The wind turbine model identification and authentication method specifically includes the following steps: Step 1, Data Acquisition: Obtain current signal data during the wind turbine startup process. The current signal data includes... shaft current and three-phase current Let the sampling frequency be... The number of sampling points recorded per startup is Then time index ,in ; Step 2: Perform invalid segment filtering and valid segment location on the raw current signal data obtained in Step 1 to obtain the starting point of the valid segment. And based on the starting point of the valid segment Extract or retain the starting point of the valid segment The subsequent current signal data yields the effective current signal data; Step 3: Based on the effective current signal data obtained in Step 2, perform start-up point detection and use the detected start-up point as the timing alignment reference to perform timing alignment on the start-up current signal data of each channel. Step 4: Using the starting point detected in Step 3 as the alignment reference, synchronously extract a starting window of fixed length from each current signal data. Step 5: Extract waveform features and envelope features within the startup window obtained in Step 4, and fuse the waveform features and envelope features to construct a fingerprint vector; Step 6: Based on the fingerprint vector obtained in Step 5, perform standardization and dimensionality reduction embedding to obtain a low-dimensional representation, and calculate the template model according to the low-dimensional representation corresponding to different wind turbine models to establish a model library; Step 7, Identification and Authentication Decision: Input a new set of samples to construct fingerprint vectors according to steps 1 to 5, and use the low-dimensional representation and model library trained in step 6 to obtain the embedding representation of the test samples; Calculate the Euclidean distance from the embedded representation of the test sample to each template model, output the model identification result according to the principle of minimum Euclidean distance, and output the authentication decision result of pass authentication, rejection or unknown according to the authentication threshold.

2. The wind turbine model identification and authentication method based on start-up transient multi-channel current fingerprint as described in claim 1, characterized in that: Step 2 specifically includes the following steps: Step 2.1: Preprocess the raw current signal data obtained in Step 1, skipping the initial... By sampling points, signal samples are obtained, where... To skip a fixed number of points; Step 2.2, Sentinel Value and Invalid Value Removal: Let the set of invalid values ​​be... If the values ​​of consecutive sampling points in the signal samples obtained in step 2.1 belong to the set of invalid values... And the continuous length is not less than the minimum length of the continuous invalid segment. If so, the continuous interval is judged as an invalid segment and removed; Step 2.3, Low variance judgment: based on the sliding window length Calculate the standard deviation of signal samples within the sliding window. : The average value within the sliding window for: in, Indicates from the first The standard deviation of the signal samples within a sliding window starting from a sampling point is used to characterize the degree of fluctuation of the current signal within that window. This indicates the length of the sliding window, which is the number of sampling points included in each calculation of the standard deviation; This represents the offset of the sampling point within the sliding window, and its value ranges from 0 to... ; Indicates from the first Within the sliding window starting from the sampling point, the first... The signal amplitude corresponding to each sampling point; Indicates from the first Sampling point start, length is The average value of all signal samples within the sliding window; Step 2.4: Based on the starting point of the valid segment Extract or retain the starting point of the valid segment The subsequent current signal data yields the valid current signal data.

3. The wind turbine model identification and authentication method based on start-up transient multi-channel current fingerprint as described in claim 1, characterized in that: Step 3 specifically includes the following steps: Step 3.1: Based on the three-phase current in Step 1 Constructing a synthesized three-phase amplitude signal: Step 3.2: In the three-phase amplitude synthesis signal The root mean square (RMS) energy envelope of the sliding window is calculated. Detect the start point :when Duration not less than When the first satisfaction is determined, the starting point is identified. ; Step 3.3, will by Alignment is performed for reference, where, Energy threshold For the minimum duration, , To detect the start point The corresponding sampling points.

4. The wind turbine model identification and authentication method based on start-up transient multi-channel current fingerprint as described in claim 3, characterized in that: Step 4 specifically involves: from the sampling point The starting cut length is Startup window : in, For window duration, This represents the number of corresponding sampling points.

5. The wind turbine model identification and authentication method based on start-up transient multi-channel current fingerprint as described in claim 1, characterized in that: Step 5 specifically includes the following steps: Step 5.1: Perform mean-reduction and normalization processing on the current signal data of each channel within the startup window to obtain the waveform characteristics of the corresponding channel. Simultaneously calculate Hilbert envelope features : in, This indicates the current signal of any channel within the start window. They represent respectively to The waveform features obtained after mean removal and normalization processing This is the Hilbert transform; Indicates amplitude operation; Step 5.2: Extract waveform features The fingerprint vector is formed by concatenating the envelope features in a fixed order: in, They represent Envelope characteristics of shaft current and three-phase current.

6. The wind turbine model identification and authentication method based on start-up transient multi-channel current fingerprint as described in claim 1, characterized in that: Step 6 specifically involves: processing the fingerprint vector obtained in step 5. Standardization is performed, and a low-dimensional representation is obtained by dimensionality reduction embedding. For each fan model Low-dimensional representation of sets Computational class center as template model: in, This represents the low-dimensional representation of the fingerprint vector after standardization and dimensionality reduction embedding. Indicates the first The set of low-dimensional representations corresponding to each wind turbine model; Represents a set The number of low- to mid-dimensional representations; Indicates the first Template models for each wind turbine model, i.e., a set The class centers of each low-dimensional representation.

7. The wind turbine model identification and authentication method based on start-up transient multi-channel current fingerprint as described in claim 1, characterized in that: Step 7 specifically includes: Step 7.1, Identification: For the wind turbine to be identified, the fingerprint vector of the wind turbine is standardized and dimensionality-reduced by embedding in step 6 to obtain a low-dimensional representation. Calculate low-dimensional representation To each template model Euclidean distance: ， in, Low-dimensional representation of the wind turbine to be identified With the Template model of a fan model The Euclidean distance between them; Step 7.2 Authentication: If the fan to be identified declares its model as... Then calculate its low-dimensional representation. To declare the model template distance And the closest distance to other template models. And calculate the authentication interval. when If successful, output "Authentication passed"; otherwise, output "Rejected" or "Unknown". in, Indicates the authentication interval, used to measure the credibility of the wind turbine to be identified relative to the claimed model. This is the authentication threshold.

Images