A method for constructing a frequency response fingerprint database for transformer windings

Abstract

This invention discloses a method for constructing a transformer winding frequency response fingerprint database, comprising the following steps: acquiring winding frequency response data and related information; dividing the 1kHz~1000kHz frequency range into high, medium, and low characteristic frequency bands; performing noise reduction and feature extraction for each frequency band; building a four-level node fingerprint database based on MySQL, consisting of "equipment classification - single equipment - time series - frequency band features"; associating multi-dimensional retrieval indexes to form a "time-frequency band feature" sequence; locating the winding deformation frequency band and inferring its type through single-frequency band trend determination and multi-frequency band association verification; integrating multi-source data into a unified standard to improve data quality; supporting efficient retrieval and dynamic updates; accurately locating problems based on the "frequency band-winding characteristics" mapping to achieve pre-condition warning; covering mainstream voltage levels and insulation media; adapting to various types of FRA detectors; reducing application costs; and being suitable for condition monitoring and fault diagnosis of various types of transformers.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring and fault diagnosis technology, specifically a method for constructing a frequency response fingerprint database for transformer windings. Background Technology

[0002] Transformers are core equipment in power systems, and the mechanical stability of their windings directly determines the safety of power grid operation. Frequency response analysis, as a core technology for diagnosing winding deformation, determines the fault state by comparing the differences in the winding's "frequency response fingerprint" (i.e., impedance / admittance response curves at different frequencies).

[0003] Existing technologies have a certain foundation in transformer frequency response data processing and analysis. Chinese patent application number CN202511072337.8, "Transformer Winding Fault Detection Method, System, and Medium Based on Frequency Response," constructs a multi-dimensional fault fusion model by acquiring offline frequency response test data of the transformer, including at least a smooth frequency response curve, to obtain a value at the resonant frequency reflecting changes in transformer winding performance. Chinese patent application number CN202411953474.8, "Offline Monitoring Method, Device, Equipment, and Storage Medium for Transformer Winding Status," calculates the fluctuation of frequency response data using peak and valley frequencies at neutral point grounding, as well as the equivalent inductance and equivalent capacitance of the transformer winding. The comparison of the test data fluctuation with preset thresholds enables fault detection of the transformer winding.

[0004] However, existing technologies for frequency response fingerprint management and diagnosis of transformer windings still have the following key problems: diverse data sources and inconsistent formats, resulting in high integration costs; traditional fingerprint databases have simple structures and lack multi-dimensional correlation and dynamic update mechanisms; diagnosis relies on human experience and lacks automated and forward-looking trend prediction capabilities; existing machine learning methods have limitations in feature extraction and time series modeling, making it difficult to handle multi-source heterogeneous data.

[0005] With the increasing demand for "condition early warning" in smart grids, there is an urgent need for a method to construct and optimize a frequency response fingerprint database that integrates multi-source data, structured storage, and AI-driven trend diagnosis, in order to solve existing technical bottlenecks and improve the accuracy and foresight of transformer winding fault diagnosis. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method for constructing a frequency response fingerprint database for transformer windings, aiming to solve the problems in the background technology.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for constructing a transformer winding frequency response fingerprint database, comprising the following steps:

[0008] Step S1: Obtain multi-parameter data of the target transformer winding, including frequency response data and related information;

[0009] Step S2: Based on the multi-parameter data of the target transformer winding, according to the correlation between the deformation type of the target transformer winding and the frequency response band, the frequency range of 1kHz~1000kHz is divided into three characteristic frequency bands: high, medium and low. Then, frequency band denoising and feature extraction are performed to obtain core feature information including the main peak frequency, the main peak amplitude, and the correlation coefficient between the denoising curve of each frequency band and the reference curve.

[0010] Step S3: Build a frequency response fingerprint database based on core feature information and associate it with a multi-dimensional search index including "voltage level, detection date, test temperature and humidity" to form a "time-frequency band feature" sequence of the device in the fingerprint database;

[0011] Step S4: Based on the "time-frequency band feature" sequence of the device in the frequency response fingerprint database, the frequency band corresponding to the winding deformation trend is located by a two-step method of single-frequency band trend determination and multi-frequency band correlation verification.

[0012] Furthermore, the frequency response data sources include: on-site FRA testing instrument, which collects frequency domain response data in real time from 1kHz to 1MHz; laboratory winding deformation simulation platform, which simulates axial displacement, radial bulge, and inter-turn short circuit faults, and records frequency response curves for different deformation amounts; historical fault database, containing frequency response data with fault types labeled; and related information including: transformer parameters corresponding to the data; testing environment parameters; and historical maintenance records.

[0013] Furthermore, based on the correlation between winding deformation type and frequency response band, the frequency range of 1kHz to 1MHz is divided into three characteristic bands, including the low frequency band of 1kHz to 10kHz, the mid frequency band of 10kHz to 100kHz, and the high frequency band of 100kHz to 1MHz.

[0014] Furthermore, frequency band denoising employs a wavelet transform algorithm of "db4 wavelet basis + frequency band layer adjustment" to perform differentiated optimization processing for the noise characteristics of different frequency bands (low, medium, and high).

[0015] For low, medium, and high frequency bands, wavelet decomposition at different levels is performed to reconstruct the low-frequency band noise reduction curves. Mid-frequency noise reduction curve and high-frequency noise reduction curve ;

[0016] Calculate the signal-to-noise ratio for each frequency band:

[0017] ;

[0018] In the formula, Indicates the first Signal-to-noise ratio of each frequency band Indicates frequency band identification. Indicates high frequency, Indicates intermediate frequency, Indicates low frequency; Indicates the first The signal value of each frequency band after noise reduction processing; Indicates the first The original signal values ​​of each frequency band; Represents the natural logarithm operation;

[0019] Verify whether the signal-to-noise ratio of each frequency band after reconstruction meets the preset signal-to-noise ratio threshold. If it does not meet the threshold, adjust the wavelet decomposition level of the corresponding frequency band and perform noise reduction again until the signal-to-noise ratio meets the requirements.

[0020] Furthermore, for each frequency band, three core features are extracted:

[0021] Obtain the main peak frequency of the noise reduction curves for each frequency band. :

[0022] From the low-frequency noise reduction curve Extract the frequency corresponding to the maximum impedance, which is the main peak frequency of the low-frequency noise reduction curve. ;

[0023] From the mid-frequency noise reduction curve Extract the frequency corresponding to the maximum impedance, which is the main peak frequency of the mid-frequency noise reduction curve. ;

[0024] From the high-frequency noise reduction curve Extract the frequency corresponding to the maximum impedance, which is the main peak frequency of the high-frequency noise reduction curve. ;

[0025] Extract the peak amplitude corresponding to the peak frequency of the noise reduction curve for each frequency band. :

[0026] Extract the impedance amplitude corresponding to the main peak frequency of the noise reduction curve for each frequency band, and normalize the extracted impedance amplitude to obtain the main peak amplitude corresponding to the main peak frequency of the noise reduction curve for each frequency band. :

[0027] ;

[0028] In the formula, Indicates the current Impedance amplitude corresponding to the main peak frequency of the frequency band noise reduction curve; express Impedance amplitude corresponding to the main peak frequency of the frequency band reference curve; Represents logarithmic operations to base 10;

[0029] Obtain the correlation coefficients between the noise reduction curves of each frequency band and the baseline curve. :

[0030] ;

[0031] In the formula, Indicates the current The average impedance amplitude corresponding to the main peak frequency of the frequency band noise reduction curve; express Impedance amplitude corresponding to the main peak frequency of the frequency band reference curve; express The average impedance amplitude corresponding to the main peak frequency of the frequency band reference curve;

[0032] Among them, the "first test curve after leaving the factory" or the "first test curve after major overhaul" is used as the reference curve. Extract the main peak frequency of the baseline curve .

[0033] Furthermore, the frequency response fingerprint database is constructed based on a MySQL database, consisting of four levels of nodes: "Device Classification - Single Device - Time Series - Frequency Band Features"; the information for each level of node is as follows:

[0034] The primary node is named "Equipment Classification Node", its storage content is "classified by voltage level", and its index tags are "voltage level, equipment insulation medium".

[0035] The node name of the secondary node is "Single Device Node", the stored content is "Equipment Basic Parameters", and the index tags are "Equipment ID, Model, Substation".

[0036] The third-level node is named "Time Series Node" and stores the timestamp of each device detection. Environmental parameters The index labels are "test date, test season";

[0037] The fourth-level node is named "Frequency Band Feature Node" and stores the main peak frequency of the noise reduction curve for each frequency band. The main peak amplitude corresponding to the main peak frequency of the noise reduction curves in each frequency band Correlation coefficients between noise reduction curves and baseline curves in each frequency band The index labels are "frequency band type, frequency range".

[0038] Furthermore, the baseline curve Disassembled into low-frequency reference curves Mid-frequency reference curve High-frequency band reference curve The data is then stored in the corresponding fourth-level nodes to form the device's "time-frequency band characteristic" sequence.

[0039] Furthermore, the specific process for determining single-band trends is as follows:

[0040] Calculate the average rate of change of the main peak frequency;

[0041] Different frequency band trend change judgment conditions are set, including characteristic average change rate judgment condition, trend stability judgment condition, and benchmark similarity judgment condition. If the characteristic average change rate judgment condition, trend stability judgment condition, and benchmark similarity judgment condition are all met simultaneously, it is determined that there is a winding deformation trend in that frequency band. Otherwise, further verification is performed by referring to the equipment's historical maintenance records. The further verification process is as follows: if a frequency band does not meet the trend change judgment conditions, it is determined by referring to the equipment's historical maintenance records whether there are frequency response changes caused by non-deformation factors.

[0042] Among them, the characteristic average rate of change determination criterion for the low-frequency band is "the average rate of change of the main peak frequency in the low-frequency band". The condition for determining the trend stability of the low-frequency band is "the direction of feature change is consistent in 3 consecutive detections"; the condition for determining the benchmark similarity of the low-frequency band is "the correlation coefficient of 2 consecutive detections". >Second preset threshold”;

[0043] Among them, the characteristic average rate of change determination criterion for the mid-frequency band is "the average rate of change of the main peak frequency in the mid-frequency band". The condition for determining the trend stability of the mid-frequency band is "the direction of feature change is consistent in 3 consecutive detections"; the condition for determining the benchmark similarity of the mid-frequency band is "the correlation coefficient of 2 consecutive detections". >Fourth preset threshold”;

[0044] Among them, the characteristic average rate of change determination criterion for the high-frequency band is "the average rate of change of the main peak frequency in the high-frequency band". "≥ Fifth preset threshold", the trend stability judgment condition for the high-frequency band is "the feature change direction is consistent in 3 consecutive detections", and the benchmark similarity judgment condition for the high-frequency band is "the correlation coefficient of 2 consecutive detections". >Sixth preset threshold.

[0045] Furthermore, the specific process of multi-band association verification is as follows:

[0046] By combining the trend determination results of each frequency band in the single-band trend determination, and correlating the correspondence between the winding deformation type and the frequency band, the reliability of the trend is further verified and the deformation type is inferred. Specifically:

[0047] The winding deformation trend only exists in the low-frequency range: if If the value is ≥ the seventh preset threshold and there is no winding deformation trend in the medium / high frequency band, it is verified as a reliable winding deformation trend;

[0048] Only in the mid-frequency range is there a tendency for winding deformation: if If the value is ≥ the eighth preset threshold and there is no winding deformation trend in the low / high frequency band, it is verified as a reliable winding deformation trend;

[0049] The winding deformation trend only exists in the high-frequency band: if If the value is ≥ the ninth preset threshold and there is no winding deformation trend in the low / mid frequency band, it is verified as a reliable winding deformation trend;

[0050] There is a winding deformation trend in the low and mid frequency bands: If the winding deformation trend in the low and mid frequency bands is consistent, it is verified as a reliable winding deformation trend.

[0051] A non-volatile computer storage medium stores computer-executable instructions that execute a method for constructing a transformer winding frequency response fingerprint database.

[0052] Compared with existing technologies, the present invention has the following advantages:

[0053] (1) Improve the quality and standardization of frequency response data: This invention integrates real-time data from the field FRA tester, laboratory winding deformation simulation data and historical fault labeling data, and associates them with equipment parameters, environmental parameters and maintenance records to solve the problem of diverse data sources and inconsistent formats in the existing technology; avoids random errors by using the process of "taking the average value of three consecutive acquisitions", and uses the "first test curve after factory / overhaul" as the benchmark for unified comparison; divides the 1kHz~1MHz frequency band into three characteristic frequency bands: low, medium and high, and combines the db4 wavelet base and the number of differential decomposition layers (7 layers for low / medium frequency and 6 layers for high frequency) to perform frequency band noise reduction, effectively improving data reliability and laying a standardized data foundation for subsequent diagnosis.

[0054] (2) Achieve structured management and efficient retrieval of the frequency response fingerprint database: Based on the MySQL database, a four-level node database is constructed, consisting of "equipment classification - single device - time series - frequency band features". Each level of node is matched with the storage requirements of voltage level / insulation medium, basic equipment parameters, detection time / environment, and core frequency band features, ensuring that the data hierarchy is clear and closely related; multi-dimensional indexes such as "voltage level, equipment ID, detection date, and frequency band type" are designed to support fast combination retrieval of "equipment + time + frequency band", which greatly improves the efficiency of historical data retrieval; after the detection is completed, the feature parameters and timestamp are automatically written into the database to realize dynamic data updates and avoid the defects of traditional fingerprint databases, such as simple structure, slow retrieval, and data solidification.

[0055] (3) Improve the accuracy and foresight of winding deformation diagnosis: Based on the mapping relationship of "frequency band-winding characteristics" (low frequency band corresponds to overall stiffness, mid frequency band corresponds to inter-pane parameters, and high frequency band corresponds to inter-turn state), extract three types of physically interpretable features: main peak frequency, amplitude, and correlation coefficient with the benchmark. This can accurately locate the winding structure problem corresponding to the abnormal frequency band. Through the two-step method of "single frequency band trend judgment (combining the three conditions of average change rate, trend stability, and benchmark similarity) + multi-frequency band correlation verification (combining the frequency band combination law to screen reliable trends)," the deformation trend can be automatically judged, freeing us from the dependence on manual experience. By tracking the sequence change of "time-frequency band characteristics," deformation signs can be identified before the winding has obvious faults, realizing the transformation from "post-fault diagnosis" to "pre-state warning," and meeting the needs of proactive operation and maintenance of smart grids.

[0056] (4) Enhance the universality and applicability of the method: The primary nodes of the database cover the mainstream voltage levels of 110kV, 220kV, and 500kV, as well as oil-immersed and dry-type transformers with different insulation media. There is no need to build a separate fingerprint database for different equipment. The data acquisition and preprocessing process is standardized and can be adapted to various FRA detectors that output 1kHz~1MHz frequency response data. No customized modification is required, which reduces the threshold and cost of field application and is suitable for long-term condition monitoring and fault diagnosis of multiple types of transformers in the power system. Attached Figure Description

[0057] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0058] like Figure 1 As shown, the present invention provides a technical solution: a method for constructing a frequency response fingerprint database for transformer windings, comprising the following steps:

[0059] Step S1: Obtain multi-parameter data of the target transformer winding, including frequency response data and related information.

[0060] The sources of frequency response data include:

[0061] The on-site FRA testing instrument acquires the frequency response in the 1kHz~1MHz range in real time;

[0062] The laboratory winding deformation simulation platform simulates faults such as axial displacement, radial bulge, and inter-turn short circuit, and records the frequency response curves of different deformation amounts.

[0063] Historical fault database, frequency response data with fault types already labeled.

[0064] The associated information includes:

[0065] The data corresponds to the transformer parameters (model, capacity, voltage level, winding material).

[0066] Detect environmental parameters (temperature, humidity, detection time);

[0067] Historical maintenance records (date of the most recent major overhaul, winding tightness).

[0068] The data acquisition process involves collecting data three times consecutively (with a 5-minute interval between each time), and then taking the average of the three curves at the same frequency to form the "average curve" as the detection curve for this test (denoted as ). To avoid random errors; and simultaneously record the ambient temperature. .

[0069] Among them, the "first test curve after leaving the factory" or the "first test curve after major overhaul" is used as the reference curve. Extract the main peak frequency of the baseline curve .

[0070] Step S2: Based on the multi-parameter data of the target transformer winding, according to the correlation between the deformation type of the target transformer winding and the frequency response band, the frequency range of 1kHz~1000kHz is divided into three characteristic frequency bands: high, medium and low. Then, frequency band denoising and feature extraction are performed to obtain core feature information including the main peak frequency, the main peak amplitude, and the correlation coefficient between the denoising curve of each frequency band and the reference curve.

[0071] Based on the correlation between winding deformation type and frequency response band, the frequency range of 1kHz to 1MHz is divided into three characteristic bands: the low frequency band of 1kHz to 10kHz, the mid frequency band of 10kHz to 100kHz, and the high frequency band of 100kHz to 1MHz.

[0072] Among them, the frequency band noise reduction adopts a wavelet transform algorithm of "db4 wavelet basis + frequency band layer adjustment" to perform differentiated optimization processing for the noise characteristics of low, medium and high frequency bands:

[0073] For the low-frequency band, a 7-level wavelet decomposition is performed to reconstruct the low-frequency band noise reduction curve. ;

[0074] For the mid-frequency band, a 7-level wavelet decomposition is performed to reconstruct the mid-frequency band noise reduction curve. ;

[0075] For the high-frequency band, a 6-level wavelet decomposition is performed to reconstruct the high-frequency band noise reduction curve. .

[0076] Noise characteristics and signal features of different frequency bands: Low frequency band (1kHz~10kHz) has concentrated signal energy and changes slowly; mid frequency band (10kHz~100kHz) reflects the parameters between windings and requires deeper decomposition (7 layers) to effectively extract the overall stiffness features; high frequency band (100kHz~1MHz) has complex noise components and rich signal details. The signal frequency is high but the energy is weak. Excessive decomposition can easily lead to feature loss. 6-layer wavelet decomposition can achieve a balance between noise reduction and signal fidelity.

[0077] Calculate the signal-to-noise ratio for each frequency band:

[0078] ;

[0079] In the formula, Indicates the first Signal-to-noise ratio of each frequency band Indicates frequency band identification. Indicates high frequency, Indicates intermediate frequency, Indicates low frequency; Indicates the first The signal value of each frequency band after noise reduction processing; Indicates the first The original signal values ​​of each frequency band; Represents the natural logarithm operation.

[0080] Verify whether the signal-to-noise ratio of each frequency band after reconstruction meets the requirements. If the requirements are not met, adjust the wavelet decomposition level of the corresponding frequency band to 4-5 levels and perform noise reduction again until the signal-to-noise ratio meets the requirements.

[0081] Specifically, for each frequency band, three physically interpretable core features are extracted to form a "frequency band-feature" mapping relationship:

[0082] 1. Obtain the main peak frequency of the noise reduction curves for each frequency band. :

[0083] From the low-frequency noise reduction curve Extract the frequency corresponding to the maximum impedance, which is the main peak frequency of the low-frequency noise reduction curve. ;

[0084] From the mid-frequency noise reduction curve Extract the frequency corresponding to the maximum impedance, which is the main peak frequency of the mid-frequency noise reduction curve. ;

[0085] From the high-frequency noise reduction curve Extract the frequency corresponding to the maximum impedance, which is the main peak frequency of the high-frequency noise reduction curve. .

[0086] The peak frequency reflects the mechanical characteristics of the winding in the corresponding frequency band: the peak frequency of the low-frequency noise reduction curve. Corresponding inter-turn parameters (inter-turn damage) (Changes); main peak frequency of mid-frequency noise reduction curve Corresponding parameters between pie pieces (if the pie pieces are loose) Offset); main peak frequency of high-frequency noise reduction curve Corresponding to overall stiffness (stiffness changes then) change).

[0087] Among them, the inter-bend parameter specifically refers to the electrical and mechanical correlation parameters between two adjacent winding discs in the transformer winding structure, which is a commonly used industry term in the field of power transformer winding testing; inter-bend loosening refers to the failure of the mechanical fastening state between two adjacent winding discs of a transformer winding, resulting in an increase in the relative displacement or gap between the winding discs.

[0088] 2. Extract the peak amplitude corresponding to the peak frequency of the noise reduction curve for each frequency band. :

[0089] Extract the impedance amplitude corresponding to the main peak frequency of the noise reduction curve for each frequency band, and normalize the extracted impedance amplitude to obtain the main peak amplitude corresponding to the main peak frequency of the noise reduction curve for each frequency band. :

[0090] ;

[0091] In the formula, Indicates the current Impedance amplitude corresponding to the main peak frequency of the frequency band noise reduction curve; express Impedance amplitude corresponding to the main peak frequency of the frequency band reference curve; It represents a logarithmic operation with base 10.

[0092] The amplitude of the main peak reflects the change in the equivalent impedance of the winding in the corresponding frequency band: the amplitude of the main peak corresponding to the main peak frequency of the noise reduction curve in the low frequency band. Offset indicates a change in inter-turn insulation condition; the main peak amplitude corresponding to the main peak frequency of the mid-frequency noise reduction curve. Changes suggest abnormal inter-pane capacitance or inductance; the main peak amplitude corresponding to the main peak frequency of the high-frequency noise reduction curve. Fluctuations indicate overall structural deformation.

[0093] 3. Obtain the correlation coefficient between the noise reduction curves of each frequency band and the baseline curve. (Low-frequency noise reduction curve) Compared with the low-frequency reference curve correlation coefficient Mid-frequency noise reduction curve Compared with the mid-frequency reference curve correlation coefficient High-frequency noise reduction curve High-frequency reference curve correlation coefficient ):

[0094] ;

[0095] In the formula, Indicates the current The average impedance amplitude corresponding to the main peak frequency of the frequency band noise reduction curve; express Impedance amplitude corresponding to the main peak frequency of the frequency band reference curve; express The average impedance amplitude corresponding to the main peak frequency of the frequency band reference curve.

[0096] Step S3: Build a frequency response fingerprint database based on core feature information and associate it with a multi-dimensional search index including "voltage level, detection date, test temperature and humidity" to form a "time-frequency band feature" sequence of the device in the fingerprint database.

[0097] The frequency response fingerprint database is built on a MySQL database, consisting of four levels: "equipment classification - single device - time series - frequency band features," ensuring that historical features can be quickly retrieved by "equipment (transformer) + time + frequency band." The information for each level of the database is as follows:

[0098] The primary node is named "Equipment Classification Node", and its stored content is "Classified by voltage level (110kV, 220kV, 500kV)". The index tags are "Voltage level, equipment insulation medium (oil-immersed, dry)".

[0099] The node name of the secondary node is "Single Device Node", the stored content is "Equipment Basic Parameters (Model, Capacity, Years of Operation)", and the index tags are "Equipment ID, Model, Substation".

[0100] The third-level node is named "Time Series Node" and stores the timestamp of each device detection. Environmental parameters The index tags are "Detection Date (Year / Month / Day), Detection Season (Spring / Summer / Autumn / Winter)";

[0101] The fourth-level node is named "Frequency Band Feature Node" and stores the main peak frequency of the noise reduction curve for each frequency band. The main peak amplitude corresponding to the main peak frequency of the noise reduction curves in each frequency band Correlation coefficients between noise reduction curves and baseline curves in each frequency band The index labels are "frequency band type (low / medium / high) and frequency range".

[0102] Among them, the transformer's "first test curve after leaving the factory" or "first test curve after major overhaul" is used as the benchmark curve. , the baseline curve Disassembled into low-frequency reference curves Mid-frequency reference curve High-frequency band reference curve The data is then stored in the corresponding fourth-level nodes to form the device's "time-frequency band characteristic" sequence.

[0103] Step S4: Based on the "time-frequency band feature" sequence of the device in the frequency response fingerprint database, the frequency band corresponding to the winding deformation trend is located by a two-step method of single-frequency band trend determination and multi-frequency band correlation verification.

[0104] Single-band trend determination:

[0105] 1. Calculate the average rate of change of the main peak frequency:

[0106] calculate The relative rate of change of the main peak frequency of the frequency band is expressed as:

[0107] ;

[0108] In the formula, express frequency band The relative rate of change of the main peak frequency in each detection; express frequency band The main peak frequency of the secondary detection; express frequency band in The main peak frequency of the secondary detection.

[0109] based on The relative rate of change of the main peak frequency in the frequency band is calculated. The average rate of change of the main peak frequency of the frequency band is expressed as:

[0110] ;

[0111] In the formula, express frequency band The average rate of change of the main peak frequency in each detection; This indicates the total number of tests.

[0112] 2. Set trend change judgment conditions for different frequency bands, including characteristic average change rate judgment conditions, trend stability judgment conditions, and benchmark similarity judgment conditions; if the characteristic average change rate judgment conditions, trend stability judgment conditions, and benchmark similarity judgment conditions are all met simultaneously, it is determined that there is a winding deformation trend in that frequency band; otherwise, further verification is performed by referring to the equipment's historical maintenance records. The further verification process is as follows: if a certain frequency band does not meet the trend change judgment conditions, then refer to the equipment's historical maintenance records (such as the time of the most recent major overhaul and the winding tightness) to determine whether there are frequency response changes caused by non-deformation factors, so as to avoid misjudgment.

[0113] Among them, the characteristic average rate of change determination criterion for the low-frequency band is "the average rate of change of the main peak frequency in the low-frequency band". ≥ First preset threshold (the first preset threshold is The trend stability criterion for low-frequency bands is "the direction of feature change is consistent in three consecutive detections," and the benchmark similarity criterion for low-frequency bands is "the correlation coefficient of two consecutive detections." >Second preset threshold (the second preset threshold is) ).

[0114] Among them, the characteristic average rate of change determination criterion for the mid-frequency band is "the average rate of change of the main peak frequency in the mid-frequency band". ≥ Third preset threshold (the third preset threshold is) The trend stability criterion for the mid-frequency band is "the direction of feature change is consistent in three consecutive detections," and the benchmark similarity criterion for the mid-frequency band is "the correlation coefficient of two consecutive detections." > Fourth preset threshold (the fourth preset threshold is )".

[0115] Among them, the characteristic average rate of change determination criterion for the high-frequency band is "the average rate of change of the main peak frequency in the high-frequency band". ≥ Fifth preset threshold (the fifth preset threshold is) The trend stability criterion for high-frequency bands is "the direction of feature change is consistent in three consecutive detections," and the benchmark similarity criterion for high-frequency bands is "the correlation coefficient of two consecutive detections." >Sixth preset threshold (the sixth preset threshold is )".

[0116] Multi-band association verification:

[0117] By combining the trend determination results of each frequency band in the single-band trend determination, and correlating the correspondence between winding deformation type and frequency band, the reliability of the trend is further verified. Specifically:

[0118] The winding deformation trend only exists in the low-frequency range: if ≥Seventh preset threshold (Seventh preset threshold is Furthermore, there is no winding deformation trend in the mid / high frequency bands, which verifies that the winding deformation trend is reliable;

[0119] Only in the mid-frequency range is there a tendency for winding deformation: if ≥Eighth preset threshold (Eighth preset threshold is Furthermore, there is no winding deformation trend in the low / high frequency bands, which verifies that the winding deformation trend is reliable.

[0120] The winding deformation trend only exists in the high-frequency band: if ≥ Ninth preset threshold (the ninth preset threshold is) Furthermore, there is no winding deformation trend in the low / mid frequency range, which verifies that the winding deformation trend is reliable.

[0121] There is a winding deformation trend in the low and mid frequency bands: If the winding deformation trend in the low and mid frequency bands is consistent, it is verified as a reliable winding deformation trend.

[0122] A non-volatile computer storage medium stores computer-executable instructions that execute a method for constructing a transformer winding frequency response fingerprint database.

[0123] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for constructing a frequency response fingerprint database for transformer windings, characterized in that, Includes the following steps: Step S1: Obtain multi-parameter data of the target transformer winding, including frequency response data and related information; Step S2: Based on the multi-parameter data of the target transformer winding, according to the correlation between the deformation type of the target transformer winding and the frequency response band, the frequency range of 1kHz~1000kHz is divided into three characteristic frequency bands: high, medium and low. Then, frequency band denoising and feature extraction are performed to obtain core feature information including the main peak frequency, the main peak amplitude, and the correlation coefficient between the denoising curve of each frequency band and the reference curve. Step S3: Build a frequency response fingerprint database based on core feature information and associate it with a multi-dimensional search index including "voltage level, detection date, test temperature and humidity" to form a "time-frequency band feature" sequence of the device in the fingerprint database; Step S4: Based on the "time-frequency band feature" sequence of the device in the frequency response fingerprint database, the frequency band corresponding to the winding deformation trend is located by a two-step method of single-frequency band trend determination and multi-frequency band correlation verification; The specific process for determining single-band trends is as follows: Calculate the average rate of change of the main peak frequency; Set judgment conditions for trend changes in different frequency bands, including judgment conditions for the average rate of change of features, judgment conditions for trend stability, and judgment conditions for benchmark similarity; If the average rate of change of features, the trend stability, and the benchmark similarity are all met, it is determined that there is a winding deformation trend in the frequency band. Otherwise, it is verified by referring to the equipment's historical maintenance records. The verification process is as follows: if a frequency band does not meet the trend change judgment conditions, it is determined by referring to the equipment's historical maintenance records whether there is a frequency response change caused by non-deformation factors. The specific process of multi-band association verification is as follows: By combining the trend determination results of each frequency band in the single-band trend determination, the corresponding pattern between the winding deformation type and the frequency band is correlated, the reliability of the trend is verified and the deformation type is inferred.

2. The method for constructing a transformer winding frequency response fingerprint database according to claim 1, characterized in that: The frequency response data sources include: on-site FRA testing instrument, which collects frequency domain response data in real time from 1kHz to 1MHz; laboratory winding deformation simulation platform, which simulates axial displacement, radial bulge, and inter-turn short circuit faults, and records frequency response curves for different deformation amounts; historical fault database, which contains frequency response data with fault types labeled; and related information including: transformer parameters corresponding to the data; testing environment parameters; and historical maintenance records.

3. The method for constructing a transformer winding frequency response fingerprint database according to claim 2, characterized in that: Based on the correlation between winding deformation type and frequency response band, the frequency range of 1kHz to 1MHz is divided into three characteristic bands, including the low frequency band of 1kHz to 10kHz, the mid frequency band of 10kHz to 100kHz, and the high frequency band of 100kHz to 1MHz.

4. The method for constructing a transformer winding frequency response fingerprint database according to claim 3, characterized in that: Frequency-band noise reduction employs a wavelet transform algorithm based on a "dB4 wavelet basis + frequency band layer adjustment" to perform differentiated optimization for noise characteristics in low, medium, and high frequency bands. For low, medium, and high frequency bands, wavelet decomposition at different levels is performed to reconstruct the low-frequency band noise reduction curves. Mid-frequency noise reduction curve and high-frequency noise reduction curve ; Calculate the signal-to-noise ratio for each frequency band: ； In the formula, Indicates the first Signal-to-noise ratio of each frequency band Indicates frequency band identification. Indicates high frequency, Indicates intermediate frequency, Indicates low frequency; Indicates the first The signal value of each frequency band after noise reduction processing; Indicates the first The original signal values ​​of each frequency band; Represents the natural logarithm operation; Verify whether the signal-to-noise ratio of each frequency band after reconstruction meets the preset signal-to-noise ratio threshold. If it does not meet the threshold, adjust the wavelet decomposition level of the corresponding frequency band and perform noise reduction again until the signal-to-noise ratio meets the requirements.

5. The method for constructing a transformer winding frequency response fingerprint database according to claim 4, characterized in that: For each frequency band, three core features are extracted: Obtain the main peak frequency of the noise reduction curves for each frequency band. : From the low-frequency noise reduction curve Extract the frequency corresponding to the maximum impedance, which is the main peak frequency of the low-frequency noise reduction curve. ; From the mid-frequency noise reduction curve Extract the frequency corresponding to the maximum impedance, which is the main peak frequency of the mid-frequency noise reduction curve. ; From the high-frequency noise reduction curve Extract the frequency corresponding to the maximum impedance, which is the main peak frequency of the high-frequency noise reduction curve. ; Extract the peak amplitude corresponding to the peak frequency of the noise reduction curves in each frequency band. : Extract the impedance amplitude corresponding to the main peak frequency of the noise reduction curve for each frequency band, and normalize the extracted impedance amplitude to obtain the main peak amplitude corresponding to the main peak frequency of the noise reduction curve for each frequency band. : ； In the formula, Indicates the current Impedance amplitude corresponding to the main peak frequency of the frequency band noise reduction curve; express Impedance amplitude corresponding to the main peak frequency of the frequency band reference curve; Represents logarithmic operations to base 10; Obtain the correlation coefficients between the noise reduction curves of each frequency band and the baseline curve. : ； In the formula, Indicates the current The average impedance amplitude corresponding to the main peak frequency of the frequency band noise reduction curve; express The average impedance amplitude corresponding to the main peak frequency of the frequency band reference curve; Among them, the "first test curve after leaving the factory" or the "first test curve after major overhaul" is used as the reference curve. Extract the main peak frequency of the baseline curve .

6. The method for constructing a transformer winding frequency response fingerprint database according to claim 5, characterized in that: The frequency response fingerprint database is built on a MySQL database and consists of four levels of nodes: "Device Classification - Single Device - Time Series - Frequency Band Features"; the information for each level of node is as follows: The primary node is named "Equipment Classification Node", its storage content is "classified by voltage level", and its index tags are "voltage level, equipment insulation medium". The node name of the secondary node is "Single Device Node", the stored content is "Equipment Basic Parameters", and the index tags are "Equipment ID, Model, Substation". The third-level node is named "Time Series Node" and stores the timestamp of each device detection. Environmental parameters The index labels are "test date, test season"; The fourth-level node is named "Frequency Band Feature Node" and stores the main peak frequency of the noise reduction curve for each frequency band. The main peak amplitude corresponding to the main peak frequency of the noise reduction curves in each frequency band Correlation coefficients between noise reduction curves and baseline curves in each frequency band The index labels are "frequency band type, frequency range".

7. The method for constructing a transformer winding frequency response fingerprint database according to claim 6, characterized in that: Baseline curve Disassembled into low-frequency reference curves Mid-frequency reference curve High-frequency band reference curve The data is then stored in the corresponding fourth-level nodes to form the device's "time-frequency band characteristics" sequence.

8. The method for constructing a transformer winding frequency response fingerprint database according to claim 7, characterized in that: The criterion for determining the characteristic average rate of change in the low-frequency band is "the average rate of change of the main peak frequency in the low-frequency band". "≥ First preset threshold", the trend stability judgment condition for the low frequency band is "the feature change direction is consistent in 3 consecutive detections", and the benchmark similarity judgment condition for the low frequency band is "the correlation coefficient of 2 consecutive detections". >Second preset threshold”; The criterion for determining the characteristic average rate of change in the mid-frequency band is "the average rate of change of the main peak frequency in the mid-frequency band". "≥Third preset threshold", the trend stability judgment condition for the mid-frequency band is "the feature change direction is consistent in 3 consecutive detections", and the benchmark similarity judgment condition for the mid-frequency band is "the correlation coefficient of 2 consecutive detections". >Fourth preset threshold”; The criterion for determining the characteristic average rate of change in the high-frequency band is "the average rate of change of the main peak frequency in the high-frequency band". "≥ Fifth preset threshold", the trend stability judgment condition for the high-frequency band is "the feature change direction is consistent in 3 consecutive detections", and the benchmark similarity judgment condition for the high-frequency band is "the correlation coefficient of 2 consecutive detections". >Sixth preset threshold.

9. The method for constructing a transformer winding frequency response fingerprint database according to claim 8, characterized in that: Combining the trend determination results of each frequency band in the single-band trend determination, the specific process of verifying the reliability of the trend and inferring the deformation type by associating the correspondence between the winding deformation type and the frequency band is as follows: The winding deformation trend only exists in the low-frequency range: if If the value is ≥ the seventh preset threshold and there is no winding deformation trend in the medium / high frequency band, it is verified as a reliable winding deformation trend; Only in the mid-frequency range is there a tendency for winding deformation: if If the value is ≥ the eighth preset threshold and there is no winding deformation trend in the low / high frequency band, it is verified as a reliable winding deformation trend; The winding deformation trend only exists in the high-frequency band: if If the value is ≥ the ninth preset threshold and there is no winding deformation trend in the low / mid frequency band, it is verified as a reliable winding deformation trend; There is a winding deformation trend in the low and mid frequency bands: If the winding deformation trend in the low and mid frequency bands is consistent, it is verified as a reliable winding deformation trend.

10. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, The computer can execute instructions to perform a method for constructing a transformer winding frequency response fingerprint database as described in any one of claims 1-9.

Images