On-site calibration method and device for transformer vibration monitoring device

Abstract

This invention discloses a field verification method and apparatus for transformer vibration monitoring devices, belonging to the field of transformer sensing technology. The method includes: applying an excitation signal to a vibration table to acquire the vibration signal to be tested and a standard vibration signal; performing Fourier transform and multidimensional anomaly detection to identify abnormal vibration sensors; training an anomaly source determination model using historical anomaly evaluation index values ​​and historical anomaly sources; inputting the anomaly evaluation index values ​​of the abnormal vibration sensors into the model to obtain the anomaly source, and outputting the verification result; the excitation signal includes a fixed-frequency amplitude-sweep excitation signal, a frequency-sweep excitation signal, and a multi-frequency superimposed excitation signal. This invention solves the comprehensive defects of existing verification conditions, single discrimination dimensions, and lack of anomaly source tracing capabilities, leading to low overall verification accuracy and anomaly detection efficiency. It also overcomes the bottleneck of immature current transformer vibration monitoring device verification technology, achieving effective verification of transformer vibration monitoring devices.

Description

Technical Field

[0001] This invention relates to the field of transformer sensing technology, specifically to a field calibration method and apparatus for a transformer vibration monitoring device. Background Technology

[0002] Currently, the industry commonly uses vibration sensors to build transformer vibration monitoring devices. These devices utilize vibration signals to provide early warnings of potential problems such as mechanical faults, core loosening, and winding deformation. The reliability of these devices is crucial for ensuring the validity of the monitoring data. However, current verification technologies for transformer vibration monitoring devices are immature. Existing vibration verification methods suffer from numerous technical shortcomings, making them unsuitable for the actual verification needs of transformer vibration monitoring devices. These shortcomings include: First, traditional vibration verification methods often use a single excitation signal to verify the vibration sensor. This simplistic approach can easily lead to distorted verification results, failing to accurately reflect the actual performance of the transformer vibration monitoring device in the field. Second, current vibration verification methods largely rely on direct comparison of time-domain waveforms, making rough judgments based solely on time-domain amplitude. This ignores the crucial implicit characteristics of the vibration signal in the frequency domain, lacking multi-dimensional quantitative evaluation criteria. This results in insufficient fault identification capabilities for transformer vibration monitoring devices, leading to missed or incorrect diagnoses and compromising verification accuracy. Third, existing vibration verification methods can only qualitatively determine whether a vibration sensor is abnormal, making it difficult to accurately trace the root cause of the abnormality. When anomalies occur on-site, maintenance personnel often struggle to quickly identify the source of the anomaly, relying instead on manual, piecemeal troubleshooting. This process is highly inefficient, time-consuming, and prone to errors, failing to meet the practical maintenance needs of rapid calibration and precise fault location for large numbers of vibration sensors in substations. Therefore, addressing the combined shortcomings of existing technologies—such as limited calibration conditions, singular judgment dimensions, and lack of anomaly tracing capabilities—to improve overall calibration accuracy and anomaly detection efficiency, and ultimately enable effective calibration of transformer vibration monitoring devices, remains a challenging technical problem. Summary of the Invention

[0003] To address the shortcomings of existing technologies, such as limited verification conditions, single discrimination dimensions, and lack of anomaly source tracing capabilities, which result in low overall verification accuracy and anomaly detection efficiency, this invention provides a field verification method and apparatus for transformer vibration monitoring devices. By applying multi-source excitation signals to a vibration table surface to simulate various transformer vibration conditions, signals from both the tested and standard vibration sensors are simultaneously acquired. Feature parameters are extracted using Fourier transform for multi-dimensional anomaly detection. Furthermore, an anomaly source determination model is trained based on historical anomaly evaluation index values ​​and their corresponding historical anomaly sources, enabling the model to automatically identify anomaly sources. This solves the problems of limited verification conditions, single discrimination dimensions, and lack of anomaly source tracing capabilities, thus achieving effective verification of transformer vibration monitoring devices.

[0004] To solve the above-mentioned technical problems, the present invention provides a field verification method for a transformer vibration monitoring device, comprising the following steps: An excitation signal is applied to the vibration table, and the vibration signal to be tested collected by the vibration sensor to be tested and the standard vibration signal collected by the standard vibration sensor are simultaneously acquired. Fourier transform is performed on the vibration signal to be tested to obtain the characteristic parameters to be tested, and Fourier transform is performed on the standard vibration signal to obtain the standard characteristic parameters. Based on the characteristic parameters to be tested and the standard characteristic parameters, multidimensional anomaly detection is performed to identify the abnormal vibration sensor. Historical anomaly evaluation index values ​​are obtained based on historical untested feature parameters and historical standard feature parameters, and an anomaly source determination model is trained based on the historical anomaly evaluation index values ​​and their corresponding historical anomaly sources; wherein the anomaly source determination model is constructed based on random forest. Input the abnormal evaluation index value of the abnormal vibration sensor into the abnormal source determination model to obtain the abnormal source, and output the verification result; The excitation signals include fixed-frequency amplitude sweep excitation signals, frequency sweep excitation signals, and multi-frequency superimposed excitation signals.

[0005] Preferably, the step of performing a Fourier transform on the vibration signal to be tested to obtain the characteristic parameters to be tested, and performing a Fourier transform on the standard vibration signal to obtain the standard characteristic parameters, includes: DC component processing and bandpass filtering are performed on the vibration signal to be tested and the standard vibration signal. The first characteristic parameter to be tested is obtained by performing a Fourier transform on the fixed-frequency sweep vibration signal in the vibration signal to be tested. The swept frequency vibration signal in the vibration signal to be tested is divided into local overlapping segments to obtain the signal to be tested interval. The Fourier transform of the signal to be tested interval is then performed to obtain the second feature parameter to be tested. The multi-frequency superimposed vibration signal in the vibration signal to be tested is iteratively stripped by Fourier transform to obtain the sub-signal to be tested, and the third feature parameter to be tested of the sub-signal to be tested is obtained; wherein, the first feature parameter to be tested, the second feature parameter to be tested, and the third feature parameter to be tested constitute the feature parameter to be tested. The fixed-frequency sweep vibration signal in the standard vibration signal is subjected to Fourier transform to obtain the first standard characteristic parameters; The swept frequency vibration signal in the standard vibration signal is divided into local overlapping segments to obtain the standard interval signal. The standard interval signal is then subjected to Fourier transform to obtain the second standard characteristic parameter. The multi-frequency superimposed vibration signals in the standard vibration signal are iteratively stripped by Fourier transform to obtain the standard sub-signal, and the third standard feature parameter of the standard sub-signal is obtained; wherein, the first standard feature parameter, the second standard feature parameter and the third standard feature parameter constitute the standard feature parameter.

[0006] Preferably, the step of iteratively stripping the multi-frequency superimposed vibration signals from the vibration signal to be tested using Fourier transform to obtain the sub-signal to be tested, and obtaining the third characteristic parameter to be tested of the sub-signal to be tested, includes: Perform a Fourier transform on the multi-frequency superimposed vibration signal in the vibration signal to be tested to obtain the first sub-signal to be tested, and obtain the third characteristic parameter to be tested of the first sub-signal to be tested. The signal is reconstructed based on the third characteristic parameter of the first sub-signal to be examined, and the first reconstructed signal is obtained. The first multi-frequency superimposed vibration signal is obtained by performing time-domain difference operation between the multi-frequency superimposed vibration signal in the vibration signal to be tested and the first reconstructed signal. The first multi-frequency superimposed vibration signal is iteratively stripped by Fourier transform to obtain the second sub-signal to be tested, and the third characteristic parameter to be tested of the second sub-signal to be tested is obtained.

[0007] Preferably, the step of iteratively stripping the multi-frequency superimposed vibration signals from the standard vibration signal using Fourier transform to obtain a standard sub-signal, and obtaining the third standard characteristic parameter of the standard sub-signal, includes: Perform Fourier transform on the multi-frequency superimposed vibration signal in the standard vibration signal to obtain the first standard sub-signal, and obtain the third test feature parameter of the first standard sub-signal; The signal is reconstructed based on the third standard characteristic parameters of the first standard sub-signal to obtain the second reconstructed signal; The second multi-frequency superimposed vibration signal is obtained by performing time-domain difference operation between the multi-frequency superimposed vibration signal in the standard vibration signal and the second reconstructed signal. The second multi-frequency superimposed vibration signal is iteratively stripped by Fourier transform to obtain the second standard sub-signal, and the third standard characteristic parameters of the second standard sub-signal are obtained.

[0008] Preferably, the abnormal vibration sensor that performs multidimensional anomaly detection based on the feature parameters to be tested and standard feature parameters includes: Based on the amplitude in the feature parameters to be tested and the amplitude in the standard feature parameters, the mean amplitude error and the spectral error are obtained, and the amplitude error variance is obtained from the mean amplitude error. Based on the amplitude values ​​in the characteristic parameters to be tested, obtain the first average amplitude value within the preset frequency band, and based on the amplitude values ​​in the standard characteristic parameters, obtain the second average amplitude value within the preset frequency band. The spectral correlation coefficient is obtained using the Pearson product-moment correlation coefficient based on the mean of the first amplitude, the mean of the first amplitude within the preset frequency band, the mean of the second amplitude, and the second amplitude within the preset frequency band. The phase error is obtained by comparing the phase in the feature parameters to be verified with the phase in the standard feature parameters. Abnormal vibration sensors are identified by combining the mean amplitude error, variance amplitude error, spectral error, spectral correlation coefficient, and phase error.

[0009] Preferably, the sensor for identifying abnormal vibrations by combining the mean amplitude error, variance of amplitude error, spectral error, spectral correlation coefficient, and phase error includes: If any of the mean amplitude error, variance amplitude error, spectral error, or phase error is greater than the corresponding threshold, or if the spectral correlation coefficient is less than the corresponding threshold, the sensor is identified as an abnormal vibration sensor. If the mean amplitude error, variance amplitude error, spectral error, and phase error are all less than or equal to the corresponding threshold, and the spectral correlation coefficient is greater than or equal to the corresponding threshold, the sensor is identified as a normal vibration sensor. Otherwise, the mean amplitude error, variance amplitude error, spectral error, spectral correlation coefficient, and phase error are normalized and classified, and the average deviation intensity of each category is calculated. If the average deviation intensity of a certain category is greater than the preset mean, and the difference between the average deviation intensity of a certain category and the average deviation intensity of the other categories is greater than the preset difference, the sensor is identified as an abnormal vibration sensor. Otherwise, the sensor is identified as a normal vibration sensor.

[0010] Preferably, obtaining historical anomaly evaluation index values ​​based on historical untested feature parameters and historical standard feature parameters includes: Based on the amplitude values ​​in the historical untested characteristic parameters and the amplitude values ​​in the historical standard characteristic parameters, the mean of historical amplitude error, the variance of historical amplitude error, the historical spectrum error, and the historical spectrum correlation coefficient are obtained. The historical phase error is obtained based on the phase in the historical untested feature parameters and the phase in the historical standard feature parameters. Among them, the historical amplitude error mean, historical amplitude error variance, historical spectrum error, historical spectrum correlation coefficient, and historical phase error constitute the historical anomaly evaluation index values.

[0011] Preferably, training the anomaly source determination model based on the historical anomaly evaluation index values ​​and their corresponding historical anomaly sources includes: A single anomaly source determination model and a comprehensive anomaly source determination model were constructed based on random forest. Input the historical anomaly evaluation index values ​​into the single anomaly source determination model to obtain the predicted single anomaly source probability. Input the predicted single anomaly source probability into the comprehensive anomaly source determination model to obtain the comprehensive anomaly source probability. If the comprehensive anomaly source probability does not match the historical anomaly source probability, adjust the model parameters in the single anomaly source determination model and the comprehensive anomaly source determination model respectively. If the comprehensive anomaly source probability matches the historical anomaly source probability, the training is complete.

[0012] Preferably, if the overall anomaly source probability does not match the historical anomaly source probability, the model parameters in the single anomaly source determination model and the overall anomaly source determination model are adjusted respectively, including: When the overall anomaly source probability does not match the historical anomaly source probability, the matching error is obtained, and the model parameters of the overall anomaly source determination model are adjusted based on the matching error. If the overall anomaly source determination model does not converge within a preset training round during the adjustment process, the single anomaly source probability is matched with the historical anomaly source probability, and the model parameters of the mismatched single anomaly source determination model are adjusted.

[0013] By adopting the above technical solution, the present invention has the following advantages: (1) By applying multi-source excitation signals to the vibration table, various vibration conditions of the transformer are simulated. The signals of the test and standard vibration sensors are collected simultaneously. The frequency domain information of the vibration signals is obtained through Fourier transform to form multi-dimensional feature parameters, thereby realizing abnormal state detection. The abnormal source judgment model is trained based on the historical abnormal evaluation index values ​​and the historical abnormal sources corresponding to the historical abnormal evaluation index values, so that the abnormal source judgment model can automatically identify the abnormal source. This solves the comprehensive defects of the existing verification conditions, single judgment dimensions, and no ability to trace the abnormal source, resulting in low overall verification accuracy and low efficiency of abnormal investigation, thus realizing effective verification of transformer vibration monitoring devices. (2) Considering that Fourier transform requires the analyzed signal to be stationary in the time domain, while the frequency of the swept-frequency vibration signal changes continuously and is a typical non-stationary signal, directly using Fourier transform will lead to spectral leakage, resulting in large amplitude and phase deviations. Therefore, the swept-frequency vibration signal is divided into multiple locally overlapping short time periods, so that the frequency is approximately constant in each time period. After the division, Fourier transform is performed on the divided interval signals, which improves the accuracy of the obtained feature parameters. Considering that the multi-frequency superimposed vibration signal contains multiple fixed frequency components, it is difficult to achieve full-cycle sampling of all frequencies. The Fourier transform will produce energy crosstalk, and strong frequency leakage may mask weak frequency signals, which will affect the accuracy of amplitude and phase. Therefore, the iterative stripping method is used to extract and subtract the strongest frequency component one by one, thereby avoiding the influence of strong frequency leakage and improving the accuracy of the obtained feature parameters. (3) Considering that the field verification adopts three types of multi-source excitation signals, namely fixed frequency amplitude sweep, frequency sweep, and multi-frequency superposition, there are obvious differences in the distribution of vibration response characteristics and anomaly representation laws corresponding to different excitation methods. The traditional single random forest model is difficult to adapt to the differentiated expression and unified judgment of multi-source heterogeneous features and is easily affected by feature coupling interference, resulting in a decrease in the accuracy of anomaly source identification. Therefore, a two-level architecture of single anomaly source judgment model and comprehensive anomaly source judgment model is constructed: first, the single anomaly source judgment model adapts to the feature laws of various excitation signals and independently outputs the anomaly source prediction probability; then, the comprehensive anomaly source judgment model fuses and decides on multiple single probabilities to achieve accurate identification of vibration sensor anomaly sources under multi-source excitation signals, avoids the discrimination bias caused by the mixing of multi-source features, and improves the accuracy of anomaly source identification.

[0014] This invention also provides a field calibration device for a transformer vibration monitoring device, applicable to the aforementioned field calibration method for a transformer vibration monitoring device, comprising: The signal acquisition module is used to apply an excitation signal to the vibration table and simultaneously acquire the vibration signal to be tested collected by the vibration sensor to be tested installed on the vibration table and the standard vibration signal collected by the standard vibration sensor. The anomaly detection module is used to perform Fourier transform on the vibration signal under test to obtain the characteristic parameters under test, perform Fourier transform on the standard vibration signal to obtain the standard characteristic parameters, and perform multi-dimensional anomaly detection based on the characteristic parameters under test and the standard characteristic parameters to identify the abnormal vibration sensor. The model training module is used to obtain historical anomaly evaluation index values ​​based on historical unexamined feature parameters and historical standard feature parameters, and to train the anomaly source determination model based on the historical anomaly evaluation index values ​​and their corresponding historical anomaly sources. The verification result acquisition module is used to input the abnormal evaluation index value of the abnormal vibration sensor into the abnormal source determination model to obtain the abnormal source, and output the verification result.

[0015] By adopting the above technical solution, the present invention has the following advantages: By applying multi-source excitation signals to the vibration table, various vibration conditions of a transformer are simulated. Signals from both the vibration sensor under test and the standard vibration sensor are simultaneously acquired, and feature parameters are extracted using Fourier transform to achieve multi-dimensional anomaly detection. Furthermore, based on historical anomaly evaluation indicators and their corresponding historical anomaly sources, the anomaly source determination model is trained, enabling the model to automatically identify anomaly sources. This method effectively solves the problems of single verification conditions, limited discrimination dimensions, and lack of anomaly source tracing capabilities in existing technologies, thereby improving the overall verification accuracy and anomaly investigation efficiency, and thus achieving effective verification of transformer vibration monitoring devices. Attached Figure Description

[0016] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings.

[0017] Figure 1 This is a flowchart illustrating the on-site verification method for a transformer vibration monitoring device according to the present invention. Figure 2 This is a schematic diagram of the structure of the field verification device of the transformer vibration monitoring device of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only one preferred embodiment of this invention and are only used to explain this invention. They do not limit the scope of protection of this invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0019] Before discussing the exemplary embodiments in more detail, it should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the operations (or steps) as sequential processes, many of these operations (or steps) can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but it may also have additional steps not included in the figures; the process may correspond to a method, function, procedure, subroutine, subprogram, etc.

[0020] Example 1: like Figure 1 As shown, a field verification method for a transformer vibration monitoring device includes the following steps: An excitation signal is applied to the vibration table, and the vibration signal to be tested collected by the vibration sensor to be tested and the standard vibration signal collected by the standard vibration sensor are simultaneously acquired. Fourier transform is performed on the vibration signal to be tested to obtain the characteristic parameters to be tested, and Fourier transform is performed on the standard vibration signal to obtain the standard characteristic parameters. Based on the characteristic parameters to be tested and the standard characteristic parameters, multidimensional anomaly detection is performed to identify the abnormal vibration sensor. Historical anomaly evaluation index values ​​are obtained based on historical untested feature parameters and historical standard feature parameters, and an anomaly source determination model is trained based on the historical anomaly evaluation index values ​​and their corresponding historical anomaly sources; wherein the anomaly source determination model is constructed based on random forest. Input the abnormal evaluation index value of the abnormal vibration sensor into the abnormal source determination model to obtain the abnormal source, and output the verification result; The excitation signals include fixed-frequency amplitude sweep excitation signals, frequency sweep excitation signals, and multi-frequency superimposed excitation signals.

[0021] In this embodiment, the vibration table is used to provide a controllable, stable, and repeatable excitation signal for the vibration sensor under test and the standard vibration sensor. The technical performance of the vibration table preferably meets the following requirements: The frequency range covers 10Hz to 2000Hz; The acceleration output amplitude can be continuously adjusted, ranging from 0.1 m / s² to 10 m / s². Frequency stability is preferably not less than ±0.1%; Amplitude stability is preferably not less than ±2%; It supports multiple output modes, including fixed-frequency amplitude sweep excitation, frequency sweep excitation, and multi-frequency superposition excitation. It features a platform structure capable of simultaneously mounting several vibration sensors to be tested and one standard vibration sensor, enabling multi-channel synchronous calibration. Understandably, each vibration sensor to be tested corresponds to one channel, while the standard vibration sensor measures the vibration platform output in real time and feeds it back to the controller, ensuring that the frequency and amplitude remain consistent with the vibration platform's set values.

[0022] In this embodiment, the fixed-frequency sweep excitation signal is defined as: ,in, This represents a fixed-frequency amplitude-sweep excitation signal, where t represents time. This represents the amplitude of the fixed-frequency amplitude-sweep excitation signal. This indicates the fixed frequency of the fixed-frequency amplitude sweep excitation signal. This indicates the initial phase of the fixed-frequency amplitude sweep excitation signal. The optimal setting is 100Hz, which is the common vibration frequency of transformers and avoids 50Hz power frequency interference, thus having good representativeness and stability. The amplitude can be gradually increased from 1 m / s² to 5 m / s², preferably switching every 5 seconds. The amplitude range is set from 1 m / s² to 5 m / s² to cover the vibration intensity levels of the transformer under normal operation and common abnormal conditions, thereby simulating the vibration characteristics during the transition from healthy to abnormal conditions. The step-wise switching every 5 seconds ensures that the output signal from the vibration table and the vibration signal acquisition system are sufficiently stable at each amplitude point, reducing the impact of transient disturbances on the acquisition results and thus improving the accuracy of amplitude error and linearity calculations. Through the above parameter settings, the fixed-frequency sweep amplitude excitation covers both typical and common abnormal conditions, providing a reliable data foundation for subsequent abnormal vibration sensor identification. The sweep excitation signal is defined as: ,in, Indicates the frequency sweep excitation signal. This indicates the amplitude of the sweep excitation signal. This indicates the starting frequency of the sweep excitation signal, which can be 10Hz. Represents the sweep slope, satisfying , The frequency of the sweep excitation signal is the termination frequency, which can be 2000Hz, and T represents the sweep duration. This indicates the initial phase of the frequency sweep excitation signal. By using the frequency sweep method described above, while keeping the vibration amplitude basically constant, the output frequency of the vibration table is gradually changed from low frequency to high frequency, thereby enabling the acquisition of the vibration sensor's response under different frequency conditions. The selectable frequency is 2 m / s², designed to ensure a good signal-to-noise ratio while preventing the vibration table, sensor, or data acquisition card from entering the nonlinear operating region or saturation state. This allows for more realistic response results at different frequencies, improving the accuracy of the frequency sweep. The frequency sweep from 10 Hz to 2000 Hz covers the main frequency range commonly used in transformer vibration monitoring, including low-frequency structural vibration, mid-frequency operating characteristics, and high-frequency local anomaly signals. The sweep duration T can be set to 60 s–120 s. This time range ensures a smooth and continuous sweep process, allowing the sampling system sufficient stabilization time at each frequency point, reducing errors introduced by excessively rapid sweeps, and keeping the calibration time within a reasonable range. The multi-frequency superimposed excitation signal is defined as: ;in, This indicates a multi-frequency superimposed excitation signal. Indicates the index of the frequency component. Indicates the first The frequencies of each frequency component can be 100Hz, 300Hz, or 500Hz. Indicates the first The amplitude of each frequency component, where the amplitude is 2 m / s when the frequency component is 100 Hz. 2 When the frequency component is 300Hz, the corresponding amplitude is 1m / s. 2 When the frequency component is 500Hz, the corresponding amplitude is 0.5m / s. 2 , Indicates the first The initial phase of each frequency component is determined. By superimposing multiple frequency components, the vibration table simultaneously generates vibration outputs of multiple frequency components. 100Hz corresponds to the main frequency component of the iron core's magnetostrictive vibration, while 300Hz and 500Hz correspond to common harmonic vibration components encountered during operation, providing good engineering representativeness. By superimposing multiple frequency components into the same excitation signal, the vibration table simultaneously generates vibration outputs of multiple frequency components, simulating the complex vibration conditions of a real transformer.

[0023] Understandably, characteristic parameters include amplitude and phase. Anomaly evaluation indicators for vibration sensors include mean amplitude error, variance amplitude error, spectral error, spectral correlation coefficient, and phase error. Historical anomaly evaluation indicators include historical mean amplitude error, historical variance amplitude error, historical spectral error, historical spectral correlation coefficient, and historical phase error. Specific sources of anomalies include internal vibration sensor anomalies, abnormal vibration sensor base installation, interface open circuits, broken core wires, poor grounding, damaged shielding layer, amplifier anomalies, filter anomalies, sampling clock anomalies, and reference voltage anomalies.

[0024] In this embodiment, by applying three types of excitation signals to the vibration table, the complex vibration conditions of a transformer are simulated, solving the problems of single excitation form and distorted verification results in traditional verification methods. Vibration signals from both the vibration sensor under test and the standard vibration sensor are simultaneously acquired, and feature parameters are extracted via Fourier transform to achieve multi-dimensional anomaly detection, thereby identifying abnormal vibration sensors. This method overcomes the limitations of traditional methods that rely solely on time-domain amplitude for rough identification. By adding evaluation dimensions and achieving quantitative discrimination, it effectively reduces the probability of missed or false detections of sensor faults. Furthermore, by using historical anomaly evaluation index values ​​and their corresponding historical anomaly sources, the anomaly source determination model is trained. Then, the anomaly evaluation index values ​​of the abnormal sensors are input into the model, automatically and accurately identifying the anomaly source and outputting complete verification results. This method eliminates the traditional method of blindly checking each sensor one by one, achieving intelligent anomaly tracing, significantly improving on-site verification accuracy and fault diagnosis efficiency, and meeting the needs of rapid, accurate, and traceable on-site verification and maintenance of large numbers of vibration sensors in substations.

[0025] In some embodiments, the step of performing a Fourier transform on the vibration signal to be tested to obtain the characteristic parameters to be tested, and performing a Fourier transform on the standard vibration signal to obtain the standard characteristic parameters, includes: DC component processing and bandpass filtering are performed on the vibration signal to be tested and the standard vibration signal. The first characteristic parameter to be tested is obtained by performing a Fourier transform on the fixed-frequency sweep vibration signal in the vibration signal to be tested. The swept frequency vibration signal in the vibration signal to be tested is divided into local overlapping segments to obtain the signal to be tested interval. The Fourier transform of the signal to be tested interval is then performed to obtain the second feature parameter to be tested. The multi-frequency superimposed vibration signal in the vibration signal to be tested is iteratively stripped by Fourier transform to obtain the sub-signal to be tested, and the third feature parameter to be tested of the sub-signal to be tested is obtained; wherein, the first feature parameter to be tested, the second feature parameter to be tested, and the third feature parameter to be tested constitute the feature parameter to be tested. The fixed-frequency sweep vibration signal in the standard vibration signal is subjected to Fourier transform to obtain the first standard characteristic parameters; The swept frequency vibration signal in the standard vibration signal is divided into local overlapping segments to obtain the standard interval signal. The standard interval signal is then subjected to Fourier transform to obtain the second standard characteristic parameter. The multi-frequency superimposed vibration signals in the standard vibration signal are iteratively stripped by Fourier transform to obtain the standard sub-signal, and the third standard feature parameter of the standard sub-signal is obtained; wherein, the first standard feature parameter, the second standard feature parameter and the third standard feature parameter constitute the standard feature parameter.

[0026] It is understandable that the fixed-frequency amplitude-sweeping vibration signal is the acquired signal under fixed-frequency amplitude-sweeping excitation, the frequency-sweeping vibration signal is the acquired signal under frequency-sweeping excitation, and the multi-frequency superimposed vibration signal is the acquired signal under multi-frequency superimposed excitation. As is well known, after obtaining the vibration signal, its amplitude and phase can be obtained through Fourier transform, which will not be elaborated here. Considering that Fourier transform requires the analyzed signal to have a certain degree of stationarity in the time domain, while the frequency of the frequency-sweeping vibration signal changes continuously, making it difficult to meet this condition, direct Fourier transform is prone to spectral leakage, thus causing amplitude and phase deviations. Therefore, the frequency-sweeping vibration signal is divided into multiple locally overlapping short time periods, making the signal frequency approximately constant within each time period. Then, Fourier transforms are performed on the signals of the divided intervals separately, thereby improving the accuracy of the obtained characteristic parameters. Considering that multi-frequency superimposed vibration signals contain multiple fixed frequency components, it is difficult to meet the whole-cycle sampling conditions for all frequencies. Direct Fourier transform can easily generate energy crosstalk. The leakage of strong frequency components may mask weak frequency components, affecting the accuracy of amplitude and phase extraction. Therefore, an iterative stripping method is adopted to extract and subtract the strongest frequency component one by one, thereby reducing the impact of strong frequency leakage and improving the accuracy of feature parameter extraction.

[0027] In this embodiment, the overlap between adjacent segments in the local overlapping division can be 50%. For example, when the division window has 10 sampling points, and the first segment has 1 to 20 points, then the second segment has 11 to 30 points. It can also be flexibly adjusted according to needs.

[0028] Specifically, the iterative stripping of multi-frequency superimposed vibration signals from the vibration signal to be tested using Fourier transform to obtain the sub-signal to be tested, and obtaining the third characteristic parameter to be tested of the sub-signal to be tested, includes: Perform a Fourier transform on the multi-frequency superimposed vibration signal in the vibration signal to be tested to obtain the first sub-signal to be tested, and obtain the third characteristic parameter to be tested of the first sub-signal to be tested. The signal is reconstructed based on the third characteristic parameter of the first sub-signal to be examined, and the first reconstructed signal is obtained. The first multi-frequency superimposed vibration signal is obtained by performing time-domain difference operation between the multi-frequency superimposed vibration signal in the vibration signal to be tested and the first reconstructed signal. The first multi-frequency superimposed vibration signal is iteratively stripped by Fourier transform to obtain the second sub-signal to be tested, and the third characteristic parameter to be tested of the second sub-signal to be tested is obtained.

[0029] Specifically, the step of iteratively stripping the multi-frequency superimposed vibration signals from the standard vibration signal using Fourier transform to obtain a standard sub-signal, and obtaining the third standard characteristic parameter of the standard sub-signal, includes: Perform Fourier transform on the multi-frequency superimposed vibration signal in the standard vibration signal to obtain the first standard sub-signal, and obtain the third test feature parameter of the first standard sub-signal; The signal is reconstructed based on the third standard characteristic parameters of the first standard sub-signal to obtain the second reconstructed signal; The second multi-frequency superimposed vibration signal is obtained by performing time-domain difference operation between the multi-frequency superimposed vibration signal in the standard vibration signal and the second reconstructed signal. The second multi-frequency superimposed vibration signal is iteratively stripped by Fourier transform to obtain the second standard sub-signal, and the third standard characteristic parameters of the second standard sub-signal are obtained.

[0030] As is well known, Fourier transform can be used to achieve frequency domain decomposition and separation of each frequency component in multi-frequency superimposed vibration signals. In this embodiment, during the iterative stripping process, the dominant sub-signals are stripped sequentially in descending order of signal amplitude. The single-frequency sub-signals extracted in each stripping are reconstructed based on their corresponding characteristic parameters. Then, the multi-frequency superimposed vibration signal and the reconstructed signal are time-domain differencing to remove the dominant sub-signals. The remaining residual signal is used as the multi-frequency superimposed vibration signal for the next round of iterative stripping. By iteratively stripping and extracting the characteristic parameters of each order of sub-signals, the mutual coupling interference between multi-frequency components can be effectively reduced, and the errors caused by spectral leakage and component aliasing can be minimized, thereby improving the accuracy and reliability of obtaining the characteristic parameters of each sub-signal in multi-frequency superimposed scenarios.

[0031] As an optional embodiment, the vibration sensor that performs multidimensional anomaly detection and identification based on the characteristic parameters to be tested and standard characteristic parameters includes: Based on the amplitude in the feature parameters to be tested and the amplitude in the standard feature parameters, the mean amplitude error and the spectral error are obtained, and the amplitude error variance is obtained from the mean amplitude error. Based on the amplitude values ​​in the characteristic parameters to be tested, obtain the first average amplitude value within the preset frequency band, and based on the amplitude values ​​in the standard characteristic parameters, obtain the second average amplitude value within the preset frequency band. The spectral correlation coefficient is obtained using the Pearson product-moment correlation coefficient based on the mean of the first amplitude, the mean of the first amplitude within the preset frequency band, the mean of the second amplitude, and the second amplitude within the preset frequency band. The phase error is obtained by comparing the phase in the feature parameters to be verified with the phase in the standard feature parameters. Abnormal vibration sensors are identified by combining the mean amplitude error, variance amplitude error, spectral error, spectral correlation coefficient, and phase error.

[0032] The expression for obtaining the mean amplitude error is: ; This represents the amplitude error of the i-th vibration sensor under test. This represents the mean amplitude error of the i-th vibration sensor under test. Indicates the total number of frequency points. This represents the Kth frequency point. This represents the amplitude of the i-th vibration sensor at the K-th frequency point, obtained from the characteristic parameters to be tested. This represents the amplitude of the standard vibration sensor at the Kth frequency point, obtained from the standard characteristic parameters. The mean amplitude error reflects the degree to which the overall amplitude of the vibration sensor under test deviates from the standard amplitude. The expression for obtaining the variance of the amplitude error is: ; This represents the variance of the amplitude error of the i-th vibration sensor under test. This represents the amplitude error of the i-th vibration sensor under test at the K-th frequency point. The amplitude error variance reflects the fluctuation of the amplitude error of the vibration sensor under test. The expression for obtaining the spectral error is: ; This represents the spectral error of the i-th vibration sensor under test. This indicates that the i-th vibration sensor to be tested, obtained from the characteristic parameters to be tested, is at the frequency point. The amplitude at that point, This indicates the frequency point of the standard vibration sensor obtained from the standard characteristic parameters. The amplitude and spectral error at a given point are used to characterize the overall consistency between the spectrum of the vibration sensor under test and the spectrum of the standard vibration sensor. When the spectral error of the i-th vibration sensor under test is less than 0.05, the spectral consistency of the i-th vibration sensor under test is considered normal. When the spectral error of the i-th vibration sensor under test is greater than or equal to 0.05 and less than 0.10, the spectral consistency of the i-th vibration sensor under test is considered slightly abnormal. When the spectral error of the i-th vibration sensor under test is greater than or equal to 0.10, the spectral consistency of the i-th vibration sensor under test is considered significantly abnormal. The expression for obtaining the spectral correlation coefficient is: ; This represents the spectral correlation coefficient of the i-th vibration sensor under test. and These represent the upper and lower limits of the frequency, respectively, and can be flexibly set according to testing requirements, together forming a preset frequency band. This indicates that the i-th vibration sensor to be tested, obtained from the characteristic parameters to be tested, is at the frequency point. The amplitude at that point, This indicates the frequency point of the standard vibration sensor obtained from the standard characteristic parameters. The amplitude at that point, This represents the average amplitude of the first value within the preset frequency band. This represents the average second amplitude within the preset frequency band. The spectral correlation coefficient reflects the similarity between the spectrum of the vibration sensor under test and the spectrum of the standard vibration sensor. The expression for obtaining the phase error is: ;in, This represents the phase error of the i-th vibration sensor under test. This indicates that the i-th vibration sensor to be tested, obtained from the characteristic parameters to be tested, is at the frequency point. Phase at that point, This indicates the frequency point of the standard vibration sensor obtained from the standard characteristic parameters. The phase at the i-th vibration sensor is determined as follows: when the phase error of the i-th vibration sensor is less than 5°, the phase response of the i-th vibration sensor is considered normal; when the phase error of the i-th vibration sensor is less than or equal to 5° and less than 10°, the phase response of the i-th vibration sensor is considered to have a suspicious deviation; when the phase error of the i-th vibration sensor is greater than or equal to 10°, the response of the i-th vibration sensor is considered to have a significant abnormality.

[0033] Specifically, the abnormal vibration sensor that identifies abnormal vibrations by combining the mean amplitude error, variance of amplitude error, spectral error, spectral correlation coefficient, and phase error includes: If any of the mean amplitude error, variance amplitude error, spectral error, or phase error is greater than the corresponding threshold, or if the spectral correlation coefficient is less than the corresponding threshold, the sensor is identified as an abnormal vibration sensor. If the mean amplitude error, variance amplitude error, spectral error, and phase error are all less than or equal to the corresponding threshold, and the spectral correlation coefficient is greater than or equal to the corresponding threshold, the sensor is identified as a normal vibration sensor. Otherwise, the mean amplitude error, variance amplitude error, spectral error, spectral correlation coefficient, and phase error are normalized and classified, and the average deviation intensity of each category is calculated. If the average deviation intensity of a certain category is greater than the preset mean, and the difference between the average deviation intensity of a certain category and the average deviation intensity of the other categories is greater than the preset difference, the sensor is identified as an abnormal vibration sensor. Otherwise, the sensor is identified as a normal vibration sensor.

[0034] Given that the single-threshold discrimination mode only judges each indicator independently, there may be implicit anomalies where all indicators are within limits and no single dimension shows abnormality, but multiple indicators simultaneously show slight deviations, and the overall characteristics of the vibration sensor have deviated from the normal benchmark. Such anomalies are difficult to identify using independent thresholds and are prone to missed detection. This solution relies on indicator clustering and combined with inter-class deviation differences for joint judgment, which can effectively identify potential anomalies where individual indicators are compliant but the overall characteristics are distorted, further improving the accuracy of vibration sensor anomaly judgment. It is understood that the mean amplitude error and variance amplitude error can be grouped into one category, the spectral error and spectral correlation coefficient can be grouped into another, and the phase error can be grouped into another. Summing the indicators within the same category and then dividing by the number of indicators in that category yields the average deviation intensity for that category. In this embodiment, the threshold for spectral error can be 0.10, and the threshold for phase error can be 10°. It is understood that the thresholds, preset mean values, and preset differences for each indicator can also be flexibly adjusted according to actual needs.

[0035] Specifically, obtaining historical anomaly evaluation index values ​​based on historical untested feature parameters and historical standard feature parameters includes: Based on the amplitude values ​​in the historical untested characteristic parameters and the amplitude values ​​in the historical standard characteristic parameters, the mean of historical amplitude error, the variance of historical amplitude error, the historical spectrum error, and the historical spectrum correlation coefficient are obtained. The historical phase error is obtained based on the phase in the historical untested feature parameters and the phase in the historical standard feature parameters. Among them, the historical amplitude error mean, historical amplitude error variance, historical spectrum error, historical spectrum correlation coefficient, and historical phase error constitute the historical anomaly evaluation index values.

[0036] In some embodiments, training the anomaly source determination model based on the historical anomaly evaluation index values ​​and their corresponding historical anomaly sources includes: A single anomaly source determination model and a comprehensive anomaly source determination model were constructed based on random forest. Input the historical anomaly evaluation index values ​​into the single anomaly source determination model to obtain the predicted single anomaly source probability. Input the predicted single anomaly source probability into the comprehensive anomaly source determination model to obtain the comprehensive anomaly source probability. If the comprehensive anomaly source probability does not match the historical anomaly source probability, adjust the model parameters in the single anomaly source determination model and the comprehensive anomaly source determination model respectively. If the comprehensive anomaly source probability matches the historical anomaly source probability, the training is complete.

[0037] In this embodiment, considering that the field verification uses three types of multi-source excitation signals—fixed-frequency amplitude sweep, frequency sweep, and multi-frequency superposition—there are significant differences in the vibration response feature distribution and anomaly characterization patterns corresponding to different excitation signals. Furthermore, the traditional single random forest model is difficult to adapt to the differentiated expression and fusion judgment of multi-source heterogeneous features, resulting in problems such as feature coupling interference and low accuracy in anomaly source identification. Therefore, a two-level architecture—a single anomaly source determination model and a comprehensive anomaly source determination model—is specifically constructed. First, the single anomaly source determination model adapts to the feature patterns of various excitation signals and independently outputs the anomaly source prediction probability. Then, the comprehensive anomaly source determination model fuses and decides on multiple single probabilities, achieving accurate identification of vibration sensor anomaly sources under multi-source excitation signals. This avoids the discrimination bias caused by the mixing of multi-source features and improves the accuracy of anomaly source identification. It is understood that the single anomaly source determination model uses a single historical anomaly evaluation index value as input, mines the mapping relationship between this single historical anomaly evaluation index value and various anomaly sources, and outputs the probability corresponding to each anomaly source, achieving a preliminary independent prediction of the anomaly source. The comprehensive anomaly source determination model receives the probability output by the single anomaly source determination model, fuses, corrects, and correlates the probabilities of multiple paths, avoids the one-sidedness of independent determination by a single indicator, and outputs the comprehensive anomaly source probability after global integration, thus completing the final accurate determination of the anomaly source.

[0038] In this embodiment, if the overall anomaly source probability does not match the historical anomaly source probability, the model parameters in the single anomaly source determination model and the overall anomaly source determination model are adjusted respectively, including: When the overall anomaly source probability does not match the historical anomaly source probability, the matching error is obtained, and the model parameters of the overall anomaly source determination model are adjusted based on the matching error. If the overall anomaly source determination model does not converge within a preset training round during the adjustment process, the single anomaly source probability is matched with the historical anomaly source probability, and the model parameters of the mismatched single anomaly source determination model are adjusted.

[0039] Understandably, when the anomaly source corresponding to the maximum comprehensive anomaly source probability in the comprehensive anomaly source probability is consistent with a historical anomaly source, it indicates a match between the comprehensive anomaly source probability and the historical anomaly source; otherwise, it indicates a mismatch. In the case of a mismatch, the probability corresponding to the historical anomaly source is set to 100% to obtain the matching error. During the adjustment of the model parameters in the comprehensive anomaly source determination model, if the comprehensive anomaly source determination model fails to converge within the preset training epochs, the convergence anomaly is not due to the parameter configuration of the comprehensive anomaly source determination model itself, but rather to the model parameter adaptation bias of the single anomaly source determination model. Therefore, the model parameters in the mismatched single anomaly source determination model are adjusted. Understandably, when the anomaly source corresponding to the maximum value in the single anomaly source probability is consistent with a historical anomaly source, it indicates a match; otherwise, it indicates a mismatch. The preset training epochs can be flexibly adjusted according to actual usage needs.

[0040] Example 2: This embodiment also provides a field calibration device for a transformer vibration monitoring device, applicable to the aforementioned field calibration method for a transformer vibration monitoring device, including: The signal acquisition module is used to apply an excitation signal to the vibration table and simultaneously acquire the vibration signal to be tested collected by the vibration sensor to be tested installed on the vibration table and the standard vibration signal collected by the standard vibration sensor. The anomaly detection module is used to perform Fourier transform on the vibration signal under test to obtain the characteristic parameters under test, perform Fourier transform on the standard vibration signal to obtain the standard characteristic parameters, and perform multi-dimensional anomaly detection based on the characteristic parameters under test and the standard characteristic parameters to identify the abnormal vibration sensor. The model training module is used to obtain historical anomaly evaluation index values ​​based on historical unexamined feature parameters and historical standard feature parameters, and to train the anomaly source determination model based on the historical anomaly evaluation index values ​​and their corresponding historical anomaly sources. The verification result acquisition module is used to input the abnormal evaluation index value of the abnormal vibration sensor into the abnormal source determination model to obtain the abnormal source, and output the verification result.

[0041] In this embodiment, as Figure 2 As shown, the field verification device consists of a control and drive module, a physical experimental platform, and a signal acquisition and analysis module. The control and drive module includes a vibration controller and a power amplifier. The system generates an excitation signal based on set vibration parameters (such as frequency, amplitude, and waveform) and receives standard feedback signals from a standard vibration sensor, forming a closed-loop control to ensure the accuracy and stability of the excitation signal output by the vibration table (set on the physical experimental platform). A power amplifier amplifies the weak electrical signal output from the vibration controller into a sufficiently powerful drive signal to drive the vibration table. Under the action of the power drive signal, the vibration table generates high-precision, controllable mechanical vibration, providing a unified excitation signal for all sensors. A standard vibration sensor is installed on the vibration table, providing real-time feedback on the actual vibration state of the table; the signal is sent back to the vibration controller for closed-loop control. The test vibration sensor and the standard vibration sensor are installed on the same vibration table, receiving the exact same excitation signal. The signals acquired by the standard vibration sensor and the test vibration sensor are output to the signal acquisition and analysis module. The signal acquisition module synchronously acquires the test vibration signals output by eight channels of the test vibration sensor and the standard vibration signals output by the standard vibration sensor. The signal acquisition module is specifically a multi-channel data acquisition card. The anomaly detection module, model training module, and verification result acquisition module together constitute the host computer, which performs anomaly detection, model training, and verification result acquisition. To make the verification results more intuitive, the output interface displays time-domain waveforms and frequency-domain spectrograms. Normal channels are displayed in the default color, while abnormal channels are marked in red, thus enabling intuitive differentiation of abnormal channels. It can be understood that one channel corresponds to one vibration sensor under test. Finally, the diagnostic results for each abnormal channel are output in report form.

[0042] The specific embodiments described above are preferred embodiments of the field verification method and device for a transformer vibration monitoring device of the present invention, and are not intended to limit the specific scope of the present invention. The scope of the present invention includes but is not limited to the specific embodiments described above. All equivalent changes made in accordance with the shape and structure of the present invention are within the protection scope of the present invention.

Claims

1. A field verification method for a transformer vibration monitoring device, characterized in that, Includes the following steps: An excitation signal is applied to the vibration table, and the vibration signal to be tested collected by the vibration sensor to be tested and the standard vibration signal collected by the standard vibration sensor are simultaneously acquired. Fourier transform is performed on the vibration signal to be tested to obtain the characteristic parameters to be tested, and Fourier transform is performed on the standard vibration signal to obtain the standard characteristic parameters. Based on the characteristic parameters to be tested and the standard characteristic parameters, multidimensional anomaly detection is performed to identify the abnormal vibration sensor. Historical anomaly evaluation index values ​​are obtained based on historical untested feature parameters and historical standard feature parameters, and an anomaly source determination model is trained based on the historical anomaly evaluation index values ​​and their corresponding historical anomaly sources; wherein the anomaly source determination model is constructed based on random forest. Input the abnormal evaluation index value of the abnormal vibration sensor into the abnormal source determination model to obtain the abnormal source, and output the verification result; The excitation signals include fixed-frequency amplitude sweep excitation signals, frequency sweep excitation signals, and multi-frequency superimposed excitation signals.

2. The on-site verification method for a transformer vibration monitoring device according to claim 1, characterized in that, The process of obtaining the characteristic parameters by performing a Fourier transform on the vibration signal to be tested and obtaining the standard characteristic parameters by performing a Fourier transform on the standard vibration signal includes: DC component processing and bandpass filtering are performed on the vibration signal to be tested and the standard vibration signal. The first characteristic parameter to be tested is obtained by performing a Fourier transform on the fixed-frequency sweep vibration signal in the vibration signal to be tested. The swept frequency vibration signal in the vibration signal to be tested is divided into local overlapping segments to obtain the signal to be tested interval. The Fourier transform of the signal to be tested interval is then performed to obtain the second feature parameter to be tested. The multi-frequency superimposed vibration signal in the vibration signal to be tested is iteratively stripped by Fourier transform to obtain the sub-signal to be tested, and the third feature parameter to be tested of the sub-signal to be tested is obtained; wherein, the first feature parameter to be tested, the second feature parameter to be tested, and the third feature parameter to be tested constitute the feature parameter to be tested. The fixed-frequency sweep vibration signal in the standard vibration signal is subjected to Fourier transform to obtain the first standard characteristic parameters; The swept frequency vibration signal in the standard vibration signal is divided into local overlapping segments to obtain the standard interval signal. The standard interval signal is then subjected to Fourier transform to obtain the second standard characteristic parameter. The multi-frequency superimposed vibration signals in the standard vibration signal are iteratively stripped by Fourier transform to obtain the standard sub-signal, and the third standard feature parameter of the standard sub-signal is obtained; wherein, the first standard feature parameter, the second standard feature parameter and the third standard feature parameter constitute the standard feature parameter.

3. The on-site verification method for a transformer vibration monitoring device according to claim 2, characterized in that, The process involves iteratively stripping the multi-frequency superimposed vibration signals from the vibration signal under test using Fourier transform to obtain the sub-signal under test, and then obtaining the third characteristic parameter of the sub-signal under test, including: Perform a Fourier transform on the multi-frequency superimposed vibration signal in the vibration signal to be tested to obtain the first sub-signal to be tested, and obtain the third characteristic parameter to be tested of the first sub-signal to be tested. The signal is reconstructed based on the third characteristic parameter of the first sub-signal to be examined, and the first reconstructed signal is obtained. The first multi-frequency superimposed vibration signal is obtained by performing time-domain difference operation between the multi-frequency superimposed vibration signal in the vibration signal to be tested and the first reconstructed signal. The first multi-frequency superimposed vibration signal is iteratively stripped by Fourier transform to obtain the second sub-signal to be tested, and the third characteristic parameter to be tested of the second sub-signal to be tested is obtained.

4. The on-site verification method for a transformer vibration monitoring device according to claim 2, characterized in that, The process involves iteratively stripping the multi-frequency superimposed vibration signals from the standard vibration signal using Fourier transform to obtain a standard sub-signal, and then obtaining the third standard characteristic parameter of the standard sub-signal, including: Perform Fourier transform on the multi-frequency superimposed vibration signal in the standard vibration signal to obtain the first standard sub-signal, and obtain the third test feature parameter of the first standard sub-signal; The signal is reconstructed based on the third standard characteristic parameters of the first standard sub-signal to obtain the second reconstructed signal; The second multi-frequency superimposed vibration signal is obtained by performing time-domain difference operation between the multi-frequency superimposed vibration signal in the standard vibration signal and the second reconstructed signal. The second multi-frequency superimposed vibration signal is iteratively stripped by Fourier transform to obtain the second standard sub-signal, and the third standard characteristic parameters of the second standard sub-signal are obtained.

5. The on-site verification method for a transformer vibration monitoring device according to claim 1, characterized in that, The abnormal vibration sensor that performs multidimensional anomaly detection and identification based on the test feature parameters and standard feature parameters includes: Based on the amplitude in the feature parameters to be tested and the amplitude in the standard feature parameters, the mean amplitude error and the spectral error are obtained, and the amplitude error variance is obtained from the mean amplitude error. Based on the amplitude values ​​in the characteristic parameters to be tested, obtain the first average amplitude value within the preset frequency band, and based on the amplitude values ​​in the standard characteristic parameters, obtain the second average amplitude value within the preset frequency band. The spectral correlation coefficient is obtained using the Pearson product-moment correlation coefficient based on the mean of the first amplitude, the mean of the first amplitude within the preset frequency band, the mean of the second amplitude, and the second amplitude within the preset frequency band. The phase error is obtained by comparing the phase in the feature parameters to be verified with the phase in the standard feature parameters. Abnormal vibration sensors are identified by combining the mean amplitude error, variance amplitude error, spectral error, spectral correlation coefficient, and phase error.

6. The on-site verification method for a transformer vibration monitoring device according to claim 5, characterized in that, The comprehensive amplitude error mean, amplitude error variance, spectral error, spectral correlation coefficient, and phase error are used to identify abnormal vibration sensors, including: If any of the mean amplitude error, variance amplitude error, spectral error, or phase error is greater than the corresponding threshold, or if the spectral correlation coefficient is less than the corresponding threshold, the sensor is identified as an abnormal vibration sensor. If the mean amplitude error, variance amplitude error, spectral error, and phase error are all less than or equal to the corresponding threshold, and the spectral correlation coefficient is greater than or equal to the corresponding threshold, the sensor is identified as a normal vibration sensor. Otherwise, the mean amplitude error, variance amplitude error, spectral error, spectral correlation coefficient, and phase error are normalized and classified, and the average deviation intensity of each category is calculated. If the average deviation intensity of a certain category is greater than the preset mean, and the difference between the average deviation intensity of a certain category and the average deviation intensity of the other categories is greater than the preset difference, the sensor is identified as an abnormal vibration sensor. Otherwise, the sensor is identified as a normal vibration sensor.

7. The on-site verification method for a transformer vibration monitoring device according to claim 1, characterized in that, The process of obtaining historical anomaly evaluation index values ​​based on historical unverified feature parameters and historical standard feature parameters includes: Based on the amplitude values ​​in the historical untested characteristic parameters and the amplitude values ​​in the historical standard characteristic parameters, the mean of historical amplitude error, the variance of historical amplitude error, the historical spectrum error, and the historical spectrum correlation coefficient are obtained. The historical phase error is obtained based on the phase in the historical untested feature parameters and the phase in the historical standard feature parameters. Among them, the historical amplitude error mean, historical amplitude error variance, historical spectrum error, historical spectrum correlation coefficient, and historical phase error constitute the historical anomaly evaluation index values.

8. The on-site verification method for a transformer vibration monitoring device according to claim 1, characterized in that, The step of training the anomaly source determination model based on the historical anomaly evaluation index values ​​and their corresponding historical anomaly sources includes: A single anomaly source determination model and a comprehensive anomaly source determination model were constructed based on random forest. Input the historical anomaly evaluation index values ​​into the single anomaly source determination model to obtain the predicted single anomaly source probability. Input the predicted single anomaly source probability into the comprehensive anomaly source determination model to obtain the comprehensive anomaly source probability. If the comprehensive anomaly source probability does not match the historical anomaly source probability, adjust the model parameters in the single anomaly source determination model and the comprehensive anomaly source determination model respectively. If the comprehensive anomaly source probability matches the historical anomaly source probability, the training is complete.

9. The on-site verification method for a transformer vibration monitoring device according to claim 8, characterized in that, If the overall anomaly source probability does not match the historical anomaly source probability, the model parameters in the single anomaly source determination model and the overall anomaly source determination model are adjusted respectively, including: When the overall anomaly source probability does not match the historical anomaly source probability, the matching error is obtained, and the model parameters of the overall anomaly source determination model are adjusted based on the matching error. If the overall anomaly source determination model does not converge within a preset training round during the adjustment process, the single anomaly source probability is matched with the historical anomaly source probability, and the model parameters of the mismatched single anomaly source determination model are adjusted.

10. A field calibration device for a transformer vibration monitoring device, applicable to the field calibration method for a transformer vibration monitoring device according to any one of claims 1-9, characterized in that, include: The signal acquisition module is used to apply an excitation signal to the vibration table and simultaneously acquire the vibration signal to be tested collected by the vibration sensor to be tested installed on the vibration table and the standard vibration signal collected by the standard vibration sensor. The anomaly detection module is used to perform Fourier transform on the vibration signal under test to obtain the characteristic parameters under test, perform Fourier transform on the standard vibration signal to obtain the standard characteristic parameters, and perform multi-dimensional anomaly detection based on the characteristic parameters under test and the standard characteristic parameters to identify the abnormal vibration sensor. The model training module is used to obtain historical anomaly evaluation index values ​​based on historical unexamined feature parameters and historical standard feature parameters, and to train the anomaly source determination model based on the historical anomaly evaluation index values ​​and their corresponding historical anomaly sources. The verification result acquisition module is used to input the abnormal evaluation index value of the abnormal vibration sensor into the abnormal source determination model to obtain the abnormal source, and output the verification result.

Images