Online monitoring system and method for grounding current of transformer iron core

Abstract

The invention provides an on-line monitoring system and method for the grounding current of a transformer iron core, and belongs to the field of data processing, and the system comprises a magnetic adsorption type dual-channel current sensor which is used for synchronously collecting a transformer iron core grounding current channel signal and a transformer clamp grounding current channel signal; the self-adaptive signal preprocessing module is used for carrying out self-adaptive differential processing through a self-adaptive differential coefficient and carrying out variable gain adjustment and self-adaptive filtering; the multi-dimensional feature extraction module is used for performing multi-dimensional signal feature extraction on the purified grounding current signal to obtain a multi-dimensional feature parameter set; the feature fusion module is used for performing constraint fusion on the multi-dimensional feature parameter set based on the corrected iron core eddy current loss physical model, and generating a comprehensive health evaluation index through an adaptive weight adjustment mechanism; and the intelligent diagnosis module is used for carrying out transformer iron core fault type identification, health state evaluation and trend prediction according to the comprehensive health evaluation index and the multi-dimensional characteristic parameter set to obtain a transformer iron core state monitoring result.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to an online monitoring system and method for transformer core grounding current. Background Art

[0002] Transformers are core equipment in power systems, and their safe and stable operation is crucial. The iron core is a crucial component of the transformer, and its grounding status is directly related to its safe operation. Under normal circumstances, the transformer core should be reliably grounded at only one point to ensure a stable core potential. Under these conditions, the grounding current is typically several milliamperes. However, when the core experiences faults such as multiple grounding points, insulation aging, or local short circuits, the grounding current can fluctuate abnormally, potentially causing localized overheating, discharge, or even transformer damage.

[0003] Traditional monitoring systems often use single-channel current sensors, which are susceptible to interference from switching transients, harmonics from adjacent equipment, and other factors in the complex electromagnetic environment of substations. This can cause weak ground current signals to be drowned out by noise. Existing signal processing methods often rely on digital filters with fixed parameters, which cannot adaptively adjust to the characteristics of on-site interference and have limited filtering effectiveness in strong interference environments. Traditional fault diagnosis methods rely primarily on simple threshold judgments based on the effective current value. These methods lack in-depth analysis of current waveform characteristics and the physical mechanism of the fault, making it difficult to accurately distinguish between different fault types.

[0004] Chinese invention patent publication number CN118655398B discloses a transformer condition monitoring method and system. This method uses recurrent neural networks and convolutional neural networks to extract global and local features of transformer condition data and then fuses these features to monitor transformer status. This patent uses multiple types of condition data, such as dissolved gas in oil, partial discharge, core grounding current, oil temperature, and oil pressure, as input, performs dimensionality reduction through principal component analysis, and then utilizes a deep learning algorithm for condition monitoring. However, this method only considers core grounding current as one of multiple monitoring parameters and lacks analysis of the core grounding current signal characteristics. Summary of the Invention

[0005] In view of this, the present invention proposes an online monitoring system and method for transformer core grounding current, which solves the problem of low monitoring accuracy caused by using single-dimensional signal analysis in the prior art.

[0006] The technical solution of the present invention is implemented as follows: The present invention provides an online monitoring system for transformer core grounding current, comprising:

[0007] The magnetic adsorption dual-channel current sensor includes a transformer core grounding current channel installed close to the transformer core grounding lead and a transformer clamp grounding current channel installed at a preset distance from the grounding lead, which is used to synchronously collect the transformer core grounding current channel signal and the transformer clamp grounding current channel signal;

[0008] an adaptive signal preprocessing module connected to the magnetic adsorption dual-channel current sensor, configured to perform adaptive differential processing on the transformer core grounding current channel signal and the transformer clamp grounding current channel signal through an adaptive differential coefficient, and perform variable gain adjustment and adaptive filtering on the differential signals to generate a purified grounding current signal;

[0009] a multidimensional feature extraction module, connected to the adaptive signal preprocessing module, for performing multidimensional signal feature extraction on the purified ground current signal to obtain a multidimensional feature parameter set;

[0010] a feature fusion module, connected to the multidimensional feature extraction module, for constraining and fusing the multidimensional feature parameter set based on a modified core eddy current loss physical model, and generating a comprehensive health evaluation index through an adaptive weight adjustment mechanism;

[0011] The intelligent diagnosis module is connected to the feature fusion module and is used to perform fault type identification, health status assessment and trend prediction based on the comprehensive health evaluation index and the multi-dimensional feature parameter set to obtain the transformer core status monitoring result.

[0012] On the basis of the above technical solution, preferably, the adaptive differential processing specifically includes:

[0013] Synchronously collecting signals of the transformer core grounding current channel and the transformer clamp grounding current channel according to a preset sampling frequency to obtain a transformer core grounding current channel signal and a transformer clamp grounding current channel signal;

[0014] The adaptive differential coefficient is calculated according to the transformer core grounding current channel signal and the transformer clamp grounding current channel signal, and the calculation formula is:

[0015]

[0016] Among them, α represents the adaptive differential coefficient, k represents the sampling point index, M represents the sliding window length, I main (k) represents the current value of the transformer core grounding current channel at the kth sampling point, I aux (k) represents the current value of the transformer clamp grounding current channel at the kth sampling point;

[0017] Based on the transformer core grounding current channel signal, the transformer clamp grounding current channel signal and the adaptive differential coefficient, a preliminary purified grounding current signal is obtained, and the calculation formula is:

[0018] I clean (k) = I main (k)-α·I aux (k);

[0019] Among them, I cleak (k) represents the preliminary purified ground current signal at the kth sampling point.

[0020] On the basis of the above technical solution, preferably, the variable gain adjustment and adaptive filtering of the differential signal specifically include:

[0021] Calculating the effective value of the preliminarily purified ground current signal, and determining a gain value of a variable gain amplifier according to the effective value and a reference current threshold;

[0022] Analyzing the frequency spectrum characteristics of the transformer core grounding current channel signal and the transformer clamp grounding current channel signal, identifying the main interference frequency components, determining the interference type based on the main interference frequency components, and performing filtering processing according to the interference type, wherein the filtering processing includes notch filtering, adaptive Wiener filtering and median filtering;

[0023] Adjust the parameters of each filter according to the interference type to generate a purified ground current signal.

[0024] Based on the above technical solutions, preferably, the multidimensional signal feature extraction includes time domain feature extraction, frequency domain feature extraction and time-frequency domain wavelet analysis, and the multidimensional feature parameter set includes time domain features, frequency domain features and time-frequency domain features.

[0025] On the basis of the above technical solution, preferably, the frequency domain feature extraction specifically includes:

[0026] The sliding discrete Fourier transform algorithm is used to calculate the spectrum of the purified ground current signal in real time and extract the amplitudes of the power frequency fundamental wave, second harmonic, third harmonic and fifth harmonic;

[0027] According to the amplitude of the power frequency fundamental wave, second harmonic, third harmonic and fifth harmonic, the temporal variation trend of each harmonic is analyzed, and the transformer core fault characteristic index is calculated to form a frequency domain characteristic vector. The transformer core fault characteristic index includes the transformer core fault total harmonic distortion rate and the multi-point grounding characteristic ratio. The calculation formula is:

[0028]

[0029] Among them, THD coreIndicates the total harmonic distortion rate of transformer core fault, X core [2] represents the amplitude of the second harmonic, X core [3] represents the amplitude of the third harmonic, X core [5] represents the amplitude of the fifth harmonic, X core [1] represents the amplitude of the power frequency fundamental wave, Indicates the multi-point grounding characteristic ratio.

[0030] On the basis of the above technical solution, preferably, the time-frequency domain wavelet analysis specifically includes:

[0031] Perform multi-scale discrete wavelet packet decomposition on the purified ground current signal and calculate the energy distribution of wavelet coefficients in each frequency band;

[0032] According to the energy distribution of wavelet coefficients in each frequency band, the time-varying characteristics of energy distribution are determined to identify transient abnormal components;

[0033] The characteristic parameters of transient abnormal components are extracted to form time-frequency domain features.

[0034] On the basis of the above technical solution, preferably, the constrained fusion of the multi-dimensional feature parameter set based on the modified core eddy current loss physical model specifically includes:

[0035] A modified eddy current loss model is established, and the modified eddy current loss model is:

[0036]

[0037] in, Indicates the corrected eddy current loss power, k loss represents the eddy current loss coefficient, f represents the current frequency, B represents the magnetic induction intensity, t core Indicates the thickness of the transformer silicon steel sheet, β aging Indicates the insulation aging influence coefficient;

[0038] Analyzing the frequency spectrum of the purified ground current signal to obtain a measured value of eddy current loss;

[0039] The eddy current loss deviation is calculated based on the modified eddy current loss model and the actual eddy current loss value, and the deviation is used as a physical constraint condition to constrain the fusion of the multi-dimensional feature parameter set.

[0040] More preferably, the generating of the comprehensive health evaluation index by the adaptive weight adjustment mechanism specifically includes:

[0041] Calculate the initial weight coefficient according to the time domain characteristics, frequency domain characteristics, time-frequency domain characteristics and the modified eddy current loss model;

[0042] The comprehensive health index is calculated based on the initial weight coefficient and the multi-feature parameter set. The calculation formula is:

[0043]

[0044] Among them, Health Index represents the comprehensive health evaluation index, i represents the index of the feature, n represents the total number of features, w i Represents the weight coefficient of the i-th feature, F i represents the normalized eigenvalue of the i-th eigenvalue.

[0045] On the basis of the above technical solution, preferably, the intelligent diagnosis module includes a transformer core fault mode library construction unit, a classification and identification unit and a trend prediction unit, wherein:

[0046] The transformer core fault mode library construction unit includes feature templates for normal state, insulation aging, multi-point grounding and local short circuit, and each transformer core fault mode corresponds to a specific feature parameter range and change trend;

[0047] The classification and identification unit adopts a multi-classifier integration method to identify the transformer core fault type according to the comprehensive health evaluation index and the multi-dimensional feature parameter set to obtain the transformer core fault type;

[0048] The trend prediction unit determines the degradation degree of the core insulation performance based on the change trend of the comprehensive health evaluation index, and uses a time series analysis method to perform trend prediction on the comprehensive health evaluation index, and generates early warning information according to the prediction result.

[0049] The present invention also provides an online monitoring method for transformer core grounding current, which is applied to the online monitoring system as described above, and includes the following steps:

[0050] S1, synchronously collecting the transformer core grounding current channel signal and the transformer clamp grounding current channel signal of the transformer core grounding lead through a magnetic adsorption dual-channel current sensor;

[0051] S2. Calculating an adaptive differential coefficient to perform differential processing on the transformer core grounding current channel signal and the transformer clamp grounding current channel signal, and performing variable gain adjustment and adaptive filtering on the differential signals to obtain a purified grounding current signal;

[0052] S3. Extracting multi-dimensional features of the purified ground current signal to obtain a multi-dimensional feature parameter set;

[0053] S4. Establish a modified eddy current loss physical model, and based on the modified core eddy current loss physical model, perform constraint fusion on the multi-dimensional feature parameter set through an adaptive weight adjustment mechanism to calculate a comprehensive health evaluation index;

[0054] S5. Comprehensive health evaluation index, combined with the preset transformer core fault mode library, identifies the transformer core fault type, evaluates the core health status and predicts the development trend, and generates a monitoring report.

[0055] The online monitoring system for transformer core grounding current of the present invention has the following beneficial effects compared with the prior art:

[0056] (1) The mixed current containing real signals and interference is measured closely to the ground lead through the transformer core grounding current channel. At the same time, the transformer clamp grounding current channel synchronously collects the environmental electromagnetic interference signal at a preset distance. The adaptive differential coefficient calculated based on the two signals is then used to perform differential processing on the signal. This can effectively eliminate the common mode interference in the complex electromagnetic environment on site, significantly improve the signal-to-noise ratio of the core grounding current signal, and ensure the accuracy of the monitoring results.

[0057] (2) By performing constraint fusion based on the modified core eddy current loss physical model, not only can the abnormal state of the transformer core be discovered in a timely manner, but the specific transformer core fault type can also be accurately identified and the development trend can be predicted, and a comprehensive health evaluation index can be generated, which effectively suppresses the interference of noise and irrelevant factors and significantly enhances the robustness of the comprehensive health evaluation index. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0059] Figure 1 Schematic diagram of the structure of an online monitoring system for transformer core grounding current of the present invention;

[0060] Figure 2 The present invention is a flowchart of an online monitoring method for transformer core grounding current. DETAILED DESCRIPTION

[0061] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0062] like Figure 1 As shown, the present invention provides an online monitoring system for transformer core grounding current, comprising:

[0063] The magnetic adsorption dual-channel current sensor includes a transformer core grounding current channel installed close to the transformer core grounding lead and a transformer clamp grounding current channel installed at a preset distance from the grounding lead, which is used to synchronously collect the transformer core grounding current channel signal and the transformer clamp grounding current channel signal;

[0064] an adaptive signal preprocessing module connected to the magnetic adsorption dual-channel current sensor, configured to perform adaptive differential processing on the transformer core grounding current channel signal and the transformer clamp grounding current channel signal through an adaptive differential coefficient, and perform variable gain adjustment and adaptive filtering on the differential signals to generate a purified grounding current signal;

[0065] a multidimensional feature extraction module, connected to the adaptive signal preprocessing module, for performing multidimensional signal feature extraction on the purified ground current signal to obtain a multidimensional feature parameter set;

[0066] a feature fusion module, connected to the multidimensional feature extraction module, for constraining and fusing the multidimensional feature parameter set based on a modified core eddy current loss physical model, and generating a comprehensive health evaluation index through an adaptive weight adjustment mechanism;

[0067] The intelligent diagnosis module is connected to the feature fusion module and is used to identify the transformer core fault type, evaluate the health status and predict the trend based on the comprehensive health evaluation index and the multi-dimensional feature parameter set to obtain the transformer core status monitoring result.

[0068] It is understandable that the current transformers of the transformer core grounding current channel and the transformer clamp grounding current channel have the same specifications. The current transformer of the transformer core grounding current channel is installed close to the transformer core grounding lead by magnetic adsorption to measure the real grounding current signal I main (t). The current transformer of the transformer clamp grounding current channel is installed at an adjustable distance of 3-10 cm from the grounding lead, mainly used for synchronously collecting environmental electromagnetic interference signals I aux (t).

[0069] The present invention measures a mixed current containing real signals and interference through the transformer core grounding current channel close to the grounding lead. At the same time, the transformer clamp grounding current channel synchronously collects the environmental electromagnetic interference signal at a preset distance, and then uses the adaptive differential coefficient calculated based on the two signals to perform differential processing on the signal. It can effectively eliminate common-mode interference in the complex electromagnetic environment on site, significantly improve the signal-to-noise ratio of the weak core grounding current signal, and ensure the accuracy of the monitoring results.

[0070] In one embodiment of the present invention, the transformer core grounding current channel sensor adopts an openable and closable ring structure design, and the outer shell adopts a magnetic adsorption device made of strong magnetic material, which can be firmly adsorbed on the transformer core grounding lead to ensure close contact between the sensor and the grounding lead.

[0071] In one embodiment of the present invention, the adaptive differential processing specifically includes:

[0072] Synchronously collecting signals of the transformer core grounding current channel and the transformer clamp grounding current channel according to a preset sampling frequency to obtain a transformer core grounding current channel signal and a transformer clamp grounding current channel signal;

[0073] The adaptive differential coefficient is calculated according to the transformer core grounding current channel signal and the transformer clamp grounding current channel signal, and the calculation formula is:

[0074]

[0075] Among them, α represents the adaptive differential coefficient, k represents the sampling point index, M represents the sliding window length, I main (k) represents the current value of the transformer core grounding current channel at the kth sampling point, I aux (k) represents the current value of the transformer clamp grounding current channel at the kth sampling point;

[0076] Based on the transformer core grounding current channel signal, the transformer clamp grounding current channel signal and the adaptive differential coefficient, a preliminary purified grounding current signal is obtained, and the calculation formula is:

[0077] I clean (k) = I main (k)-α·I aux( k);

[0078] Among them, I clean (k) represents the preliminary purified ground current signal at the kth sampling point.

[0079] Understandably, The cross-correlation function between the transformer core grounding current channel signal and the transformer clamp grounding current channel signal within the sliding window is calculated, which reflects the degree of correlation between the transformer core grounding current channel signal and the transformer clamp grounding current channel signal. The autocorrelation function of the transformer clamp grounding current channel signal is calculated for normalization. The sliding window length M can be set based on actual usage requirements. Since the auxiliary channel primarily contains interference components, and the transformer core grounding current channel contains a mixed signal of grounding current and interference, the appropriately scaled auxiliary channel signal can effectively estimate the interference components in the main channel. Subtraction effectively eliminates common-mode interference and preserves the true grounding current signal.

[0080] In one embodiment of the present invention, the preset sampling frequency is 10 kHz. The preset sampling frequency can also be set according to actual usage.

[0081] In one embodiment of the present invention, the sliding window is updated point by point.

[0082] Preferably, the sliding window is 20 ms.

[0083] Furthermore, the performing variable gain adjustment and adaptive filtering on the differential signal specifically includes:

[0084] The effective value of the preliminarily purified ground current signal is calculated, and the gain value of the variable gain amplifier is determined according to the effective value and the reference current threshold. The calculation formula is:

[0085]

[0086] Among them, G vga Indicates the gain value of the variable gain amplifier, G max Indicates the maximum gain, I rms Indicates the effective value of the preliminary cleaned ground current signal, I ref represents the reference current threshold, and p represents the index of the steepness of the control curve;

[0087] Analyzing the frequency spectrum characteristics of the transformer core grounding current channel signal and the transformer clamp grounding current channel signal, identifying the main interference frequency components, determining the interference type based on the main interference frequency components, and performing filtering processing according to the interference type, wherein the filtering processing includes notch filtering, adaptive Wiener filtering and median filtering;

[0088] Adjust the parameters of each filter according to the interference type to generate a purified ground current signal.

[0089] The present invention determines the gain value based on the ratio of the effective value to a preset reference threshold. Gain adjustment utilizes a segmented control method, providing high gain to improve the signal-to-noise ratio when the detection current is low, and reducing gain to prevent amplifier saturation when the detection current is high. By applying variable gain control and adaptive filtering to the preliminary purified ground current signal, optimal signal amplification is achieved across a wide dynamic range of current inputs, preventing small signals from being overwhelmed by noise or large signals from causing amplifier saturation, thereby improving signal processing accuracy and system adaptability.

[0090] In one embodiment of the present invention, the maximum gain G max Set to 1000 times, the reference current threshold I ref The current is set to 10 mA, and the exponent p that controls the steepness of the curve is set to 1.2.

[0091] In one embodiment of the present invention, a notch filter is used to precisely suppress power frequency and its harmonic interference. The notch frequency can be dynamically adjusted based on slight deviations from the actual power frequency. For broadband random noise, an adaptive Wiener filtering algorithm is used, which automatically adjusts filter parameters based on the statistical characteristics of the signal and noise. For occasional transient pulse interference, a median filter is used. This method effectively suppresses pulse interference without affecting the waveform characteristics of the useful signal.

[0092] In one embodiment of the present invention, the multidimensional signal feature extraction includes time domain feature extraction, frequency domain feature extraction and time-frequency domain wavelet analysis, and the multidimensional feature parameter set includes time domain features, frequency domain features and time-frequency domain features. The time domain features include mean, standard deviation, skewness and kurtosis. The mean reflects the average level of the grounding current. Under normal circumstances, the mean of the grounding current should be very small and stable. When a DC offset or low-frequency drift occurs, the mean will change. The standard deviation reflects the degree of current fluctuation. In the case of a transformer core fault, the standard deviation usually increases due to the increase in harmonic components or the appearance of irregular waveforms. The skewness reflects the symmetry of the waveform. The skewness of a normal power frequency sine wave is close to zero. When even harmonics or DC components appear, the skewness deviates from zero. The kurtosis reflects the sharpness of the waveform. The kurtosis of a normal sine wave is 1.5. When the signal contains spikes or high-order harmonics, the kurtosis increases significantly.

[0093] Specifically, the frequency domain feature extraction specifically includes:

[0094] The sliding discrete Fourier transform algorithm is used to calculate the spectrum of the purified ground current signal in real time, and the amplitudes of the power frequency fundamental wave, second harmonic, third harmonic and fifth harmonic are extracted. The formula is:

[0095] X core [k,n]=e -j2πk / N·(Xcore [k,n-1]+x[n]-x[nN]);

[0096] Among them, X core [k,n] represents the spectrum value of the kth frequency point at the nth sampling moment, k represents the frequency point index, corresponding to 50Hz fundamental wave, 100Hz second harmonic, 150Hz third harmonic, 250Hz fifth harmonic, N represents the number of DFT points, x[n] represents the nth sampling value, X core [k,n-1] represents the spectrum value of the kth frequency point at the n-1th sampling moment, and x[nN] represents the nNth sampling value;

[0097] According to the amplitude of the power frequency fundamental wave, second harmonic, third harmonic and fifth harmonic, the temporal variation trend of each harmonic is analyzed, and the transformer core fault characteristic index is calculated to form a frequency domain characteristic vector. The transformer core fault characteristic index includes the transformer core fault total harmonic distortion rate and the multi-point grounding characteristic ratio. The calculation formula is:

[0098]

[0099]

[0100] Among them, THD core Indicates the total harmonic distortion rate of transformer core fault, X core [2] represents the amplitude of the second harmonic, X core [3] represents the amplitude of the third harmonic, X core [5] represents the amplitude of the fifth harmonic, X core [1] represents the amplitude of the power frequency fundamental wave, Indicates the multi-point grounding characteristic ratio.

[0101] It is understandable that different types of transformer core faults have different characteristic manifestations in the frequency domain. The total harmonic distortion rate of transformer core fault can quantitatively reflect the degree of distortion of the ground current waveform. Under normal circumstances, THD core The value is very small. When the transformer core fails, the high-order harmonic components increase and THD core When there is multi-point grounding in the transformer core, a circulating current will be formed in the transformer core. These circulating currents are mainly expressed in the form of second harmonics. Therefore, the ratio of the second harmonic to the fundamental wave can well reflect the severity of the multi-point grounding fault. The value of is very small under normal circumstances, usually less than 0.05. It will increase significantly when multiple grounding faults occur. It can be used as an important indicator for transformer core fault diagnosis.

[0102] Furthermore, the time-frequency domain wavelet analysis specifically includes:

[0103] Perform multi-scale discrete wavelet packet decomposition on the purified ground current signal and calculate the energy distribution of wavelet coefficients in each frequency band;

[0104] According to the energy distribution of wavelet coefficients in each frequency band, the time-varying characteristics of energy distribution are determined to identify transient abnormal components;

[0105] The characteristic parameters of transient abnormal components are extracted to form time-frequency domain features.

[0106] As you can understand, time-frequency domain wavelet analysis can simultaneously provide time and frequency information of the signal, making it suitable for analyzing non-stationary signals and transient processes. The wavelet packet decomposition process decomposes the purified ground current signal into sub-signals of different frequency bands step by step. Each level of decomposition divides the purified ground current signal into high-frequency and low-frequency components. Through multi-level decomposition, detailed information on multiple frequency bands can be obtained. The energy distribution reflects the energy concentration of the purified ground current signal in different frequency bands. Under normal circumstances, the energy of the ground current is mainly concentrated in the power frequency and its low-order harmonic frequency bands. When a transient fault or partial discharge occurs, the energy in the high-frequency band increases significantly. When the energy in a certain frequency band suddenly changes, it is identified as a transient abnormal component, which can capture the characteristics of transformer core faults that are difficult to detect with traditional time-domain and frequency-domain analysis.

[0107] In one embodiment of the present invention, the constrained fusion of the multi-dimensional feature parameter set based on the modified core eddy current loss physical model specifically includes:

[0108] A modified eddy current loss model is established, and the modified eddy current loss model is:

[0109]

[0110] in, Indicates the corrected eddy current loss power, k loss represents the eddy current loss coefficient, f represents the current frequency, B represents the magnetic induction intensity, t core Indicates the thickness of the transformer silicon steel sheet, β aging Indicates the insulation aging influence coefficient;

[0111] Analyzing the frequency spectrum of the purified ground current signal to obtain a measured value of eddy current loss;

[0112] The eddy current loss deviation is calculated based on the modified eddy current loss model and the actual eddy current loss value, and the deviation is used as a physical constraint condition to constrain the fusion of the multi-dimensional feature parameter set.

[0113] It can be understood that the eddy current loss coefficient k loss It is a constant related to the material properties of the transformer silicon steel sheet, reflecting the basic properties of the material such as resistivity and permeability. 2This reflects the physical law that eddy current loss is proportional to the square of frequency, because the induced electromotive force is proportional to the frequency, while the eddy current loss is proportional to the square of the induced electromotive force. 2 It is also based on the theoretical relationship that eddy current loss is proportional to the square of magnetic flux density. Reflects the effect of eddy current path length on loss. The greater the thickness, the longer the eddy current path and the greater the loss. Insulation aging influence coefficient β aging This coefficient reflects the impact of aging insulation between the transformer's silicon steel sheets on eddy current losses. When the insulation is intact, the silicon steel sheets are insulated from each other, and eddy currents are confined within a single sheet. When the insulation ages, conductive paths appear between adjacent silicon steel sheets, expanding the eddy current path and increasing losses. This coefficient is usually greater than 1, and the worse the insulation, the greater the coefficient.

[0114] This method estimates the measured eddy current loss by analyzing the spectrum of the purified ground current signal. This is primarily based on the relationship between harmonic components and eddy current loss. Eddy current loss generates specific harmonic components in the ground current. The actual eddy current loss level can be inferred by analyzing the amplitude and phase of these harmonic components.

[0115] By performing constraint fusion based on the modified core eddy current loss physical model, not only can the abnormal state of the core be discovered in a timely manner, but the specific transformer core fault type can also be accurately identified and the development trend predicted, and a comprehensive health evaluation index can be generated, effectively suppressing the interference of noise and irrelevant factors, and significantly enhancing the robustness of the comprehensive health evaluation index.

[0116] In one embodiment of the present invention, the method for obtaining the magnetic induction intensity B is as follows:

[0117] The purified ground current signal is measured by a magnetic adsorption dual-channel current sensor. Faraday's law of electromagnetic induction is used to infer the magnetic flux density based on the current signal. Based on the transformer design parameters and operating conditions, the corresponding relationship between current and magnetic induction intensity is established. The formula is as follows:

[0118]

[0119] Where μ0 represents the vacuum permeability, 4π×10 -7 H / m, I clean Indicates the clean ground current signal, l core Indicates the effective length of the core magnetic circuit.

[0120] In one embodiment of the present invention, the insulation aging influence coefficient β aging The quantification method is as follows:

[0121] The insulation state is determined by measuring the insulation resistance between silicon steel sheets, analyzing the changes in harmonic components, especially the growth of high-order harmonics, and establishing the corresponding relationship between insulation resistance and eddy current loss growth. The calculation formula is:

[0122]

[0123] Among them, α insul Indicates the insulation sensitivity coefficient, ranging from 0.5 to 2.0, R current Indicates the current interlayer insulation resistance, R initial Indicates the initial interlayer insulation resistance.

[0124] Furthermore, when the interlayer insulation aging influence coefficient of the silicon steel sheet is (1.0, 1.1), it indicates that the transformer core is in a normal state; when it is (1.1, 1.5), it indicates that the transformer core is slightly aged; when it is (1.5, 2.5), it indicates that the transformer core is moderately aged; and when it is (2.5, 4.0), it indicates that the transformer core is severely aged.

[0125] In one embodiment of the present invention, the deviation is used as a physical constraint condition to perform constraint fusion on the multi-dimensional feature parameter set. The calculation formula for constraint fusion is:

[0126]

[0127]

[0128] in, represents the constrained eigenvalue of the i-th feature, represents the original eigenvalue of the i-th feature, φ(ΔP eddy ) represents the constraint function of eddy current loss deviation, ΔP eddy represents the eddy current loss deviation, represents the corrected eddy current loss power, Indicates the measured value of eddy current loss, δ normal Indicates the normal deviation threshold, with a value of 0.05, α enhance Indicates the enhancement coefficient, the value is 2, δ abnormal It represents the abnormal deviation threshold, which is set to 2, and β represents the attenuation coefficient, which is set to 5.

[0129] It can be understood that when the eddy current loss deviation ΔP eddy Less than 0.5, the transformer core is considered normal. eddy Between 0.05 and 0.2, the transformer core is considered to be slightly abnormal. eddy If it is greater than 0.2, it is considered that the transformer core is seriously abnormal and there is an obvious fault. The constraint function of eddy current loss deviation φ(ΔP eddy) is in the normal range, φ(ΔP eddy )=1, keep the multi-dimensional characteristic parameters unchanged to avoid signal interference, when the constraint function of eddy current loss deviation φ(ΔP eddy ) in the abnormal detection interval, at this time, α enhance = 2, providing a characteristic enhancement effect, when the constraint function of eddy current loss deviation φ(ΔP eddy ) is in the saturation protection interval, at this time, the characteristic value is in δ abnormal Continuous, with the eddy current loss deviation ΔP eddy Increase, eddy current loss deviation constraint function φ(ΔP eddy ) gradually approaches a constant of 1.

[0130] In one embodiment of the present invention, generating a comprehensive health evaluation index through an adaptive weight adjustment mechanism specifically includes:

[0131] The initial weight coefficient is calculated based on the time domain characteristics, frequency domain characteristics, time-frequency domain characteristics and the modified eddy current loss model; the calculation formula is:

[0132]

[0133]

[0134] in, represents the initial weight coefficient of the i-th feature, represents the correlation coefficient between the i-th feature and the eddy current loss model, n represents the total number of features, Cov(·) represents the covariance function, represents the standard deviation of eddy current losses, Represents the normalized eigenvalue F of the i-th i The standard deviation of

[0135] The comprehensive health index is calculated based on the initial weight coefficient and the multi-feature parameter set to reflect the overall health of the core. The calculation formula is:

[0136]

[0137] Among them, Health Index represents the comprehensive health evaluation index, i represents the index of the feature, n represents the total number of features, w i Represents the weight coefficient of the i-th feature, F i represents the normalized eigenvalue of the i-th eigenvalue.

[0138] It can be understood that the correlation coefficient between the characteristics and the eddy current loss model is calculated using the Pearson correlation coefficient to measure the linear correlation between the multidimensional parameter characteristics and the eddy current loss model. At this time, the covariance function reflects the degree to which the multidimensional parameter characteristics and the eddy current loss model deviate from the mean at the same time. When HealthIndex The larger the value, the worse the health of the transformer core.

[0139] In one embodiment of the present invention, the intelligent diagnosis module includes a transformer core fault mode library construction unit, a classification and identification unit, and a trend prediction unit, wherein:

[0140] The transformer core fault mode library construction unit includes feature templates for normal state, insulation aging, multi-point grounding and local short circuit, and each transformer core fault mode corresponds to a specific feature parameter range and change trend;

[0141] The classification and identification unit adopts a multi-classifier integration method to identify the transformer core fault type according to the comprehensive health evaluation index and the multi-dimensional feature parameter set to obtain the transformer core fault type;

[0142] The trend prediction unit determines the degradation degree of the core insulation performance based on the change trend of the comprehensive health evaluation index, and uses a time series analysis method to perform trend prediction on the comprehensive health evaluation index, and generates early warning information according to the prediction result.

[0143] Understandably, the characteristic template for a normal state is a stable, low-level comprehensive health index. Time-domain characteristics are characterized by a low mean, small standard deviation, near-zero skewness, and a kurtosis close to 1.5. Frequency-domain characteristics are characterized by a dominant fundamental component and very low higher-order harmonic content. Time-frequency domain characteristics show energy concentrated near the power frequency. The characteristic template for insulation aging is a slowly rising comprehensive health index, a gradual increase in the total harmonic distortion rate of a transformer core fault, and a slow drift in the time-domain statistical characteristics. Time-frequency domain analysis reveals a gradual increase in energy in the mid- and high-frequency bands. The characteristic template for a multi-point grounding state is a significant increase in the second harmonic component, a significant increase in the multi-point grounding characteristic ratio, a rapid increase in the comprehensive health index, and time-frequency domain analysis shows energy concentration in the second harmonic band. The characteristic template for a local short circuit state is a sudden increase in higher-order harmonic components. Time-frequency domain analysis reveals a sudden energy change in the high-frequency band. The time-domain characteristics show a significant increase in kurtosis, possibly accompanied by a transient impulse signal.

[0144] In one embodiment of the present invention, the processing steps of the classification and identification unit specifically include:

[0145] Normalizing the time domain features, frequency domain features and time-frequency domain features to obtain normalized time domain features, normalized frequency domain features and normalized time-frequency domain features;

[0146] Using historical transformer core monitoring data, multiple classifiers are trained separately, and the hyperparameters of each classifier are optimized to obtain trained classifiers. The multiple classifiers include support vector machines, decision trees, K-nearest neighbors, and naive Bayes classifiers.

[0147] Assign a weight to each trained classifier, where the weight is assigned based on the accuracy of each classifier on the validation set;

[0148] The normalized time domain features, normalized frequency domain features, and normalized time-frequency domain features are input into each trained classifier respectively. Each trained classifier will output a prediction result. Weighted voting is performed based on the prediction result of each classifier and the weight of each allocator. The weighted votes obtained by all categories are counted. The category with the highest number of votes is the fault type of the transformer core.

[0149] In one embodiment of the present invention, the degradation degree of the core insulation performance is determined based on the absolute value, change rate, fluctuation of the comprehensive health evaluation index and the total harmonic distortion rate of the transformer core fault.

[0150] In one embodiment of the present invention, the time series analysis method adopts the Autoregressive Integrated Moving Average (ARIMA) model, which can capture characteristics such as autocorrelation, trend and seasonality in time series data.

[0151] The weight distribution of the allocator is described in detail with a specific embodiment:

[0152] Set the validation set data and use the validation set data to calculate the accuracy of each classifier j. The calculation formula is:

[0153]

[0154] Among them, Acc j represents the accuracy of the j-th classifier, Indicates the number of samples correctly classified by the jth classifier, N total Indicates the total number of samples in the validation set data;

[0155] The weights are assigned according to the accuracy of the classifier. The calculation formula is:

[0156]

[0157] Among them, w j represents the weight of the j-th classifier, represents the accuracy of the jth classifier raised to the power of γ, K represents the total number of classifiers, and γ represents the weight adjustment parameter, which ranges from 2.0 to 3.0. When γ > 1, the difference between classifiers with high accuracy and those with low accuracy is magnified.

[0158] like Figure 2 As shown, the present invention also provides an online monitoring method for transformer core grounding current, which is applied to the online monitoring system as described above, comprising the following steps:

[0159] S1, synchronously collecting the transformer core grounding current channel signal and the transformer clamp grounding current channel signal of the transformer core grounding lead through a magnetic adsorption dual-channel current sensor;

[0160] S2. Calculating an adaptive differential coefficient to perform differential processing on the transformer core grounding current channel signal and the transformer clamp grounding current channel signal, and performing variable gain adjustment and adaptive filtering on the differential signals to obtain a purified grounding current signal;

[0161] S3. Extracting multi-dimensional features of the purified ground current signal to obtain a multi-dimensional feature parameter set;

[0162] S4. Establish a modified eddy current loss physical model, and based on the modified core eddy current loss physical model, perform constraint fusion on the multi-dimensional feature parameter set through an adaptive weight adjustment mechanism to calculate a comprehensive health evaluation index;

[0163] S5. Comprehensive health evaluation index, combined with the preset transformer core fault mode library, identifies the transformer core fault type, evaluates the core health status and predicts the development trend, and generates a monitoring report.

[0164] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An online monitoring system for transformer core grounding current, characterized in that: include: The magnetic adsorption dual-channel current sensor includes a transformer core grounding current channel installed close to the transformer core grounding lead and a transformer clamp grounding current channel installed at a preset distance from the grounding lead, which is used to synchronously collect the transformer core grounding current channel signal and the transformer clamp grounding current channel signal; an adaptive signal preprocessing module connected to the magnetic adsorption dual-channel current sensor, configured to perform adaptive differential processing on the transformer core grounding current channel signal and the transformer clamp grounding current channel signal through an adaptive differential coefficient, and perform variable gain adjustment and adaptive filtering on the differential signals to generate a purified grounding current signal; a multidimensional feature extraction module, connected to the adaptive signal preprocessing module, for performing multidimensional signal feature extraction on the purified ground current signal to obtain a multidimensional feature parameter set; a feature fusion module, connected to the multidimensional feature extraction module, for constraining and fusing the multidimensional feature parameter set based on a modified core eddy current loss physical model, and generating a comprehensive health evaluation index through an adaptive weight adjustment mechanism; The intelligent diagnosis module is connected to the feature fusion module and is used to identify the transformer core fault type, evaluate the health status and predict the trend based on the comprehensive health evaluation index and the multi-dimensional feature parameter set to obtain the transformer core status monitoring result.

2. The online monitoring system for transformer core grounding current according to claim 1, characterized in that: The adaptive differential processing specifically includes: Synchronously collecting signals of the transformer core grounding current channel and the transformer clamp grounding current channel according to a preset sampling frequency to obtain a transformer core grounding current channel signal and a transformer clamp grounding current channel signal; The adaptive differential coefficient is calculated according to the transformer core grounding current channel signal and the transformer clamp grounding current channel signal, and the calculation formula is: Among them, α represents the adaptive differential coefficient, k represents the sampling point index, M represents the sliding window length, I main (k) represents the current value of the transformer core grounding current channel at the kth sampling point, L aux (k) represents the current value of the transformer clamp grounding current channel at the kth sampling point; Based on the transformer core grounding current channel signal, the transformer clamp grounding current channel signal and the adaptive differential coefficient, a preliminary purified grounding current signal is obtained, and the calculation formula is: I clean (k)=I main (k)-α·I aux (k); Among them, I clean (k) represents the preliminary purified ground current signal at the kth sampling point.

3. The online monitoring system for transformer core grounding current according to claim 2, characterized in that: The variable gain adjustment and adaptive filtering of the differential signal specifically include: Calculating the effective value of the preliminarily purified ground current signal, and determining a gain value of a variable gain amplifier according to the effective value and a reference current threshold; Analyzing the frequency spectrum characteristics of the transformer core grounding current channel signal and the transformer clamp grounding current channel signal, identifying the main interference frequency components, determining the interference type based on the main interference frequency components, and performing filtering processing according to the interference type, wherein the filtering processing includes notch filtering, adaptive Wiener filtering and median filtering; Adjust the parameters of each filter according to the interference type to generate a purified ground current signal.

4. The online monitoring system for transformer core grounding current according to claim 1, characterized in that: The multi-dimensional signal feature extraction includes time domain feature extraction, frequency domain feature extraction and time-frequency domain wavelet analysis, and the multi-dimensional feature parameter set includes time domain features, frequency domain features and time-frequency domain features.

5. The online monitoring system for transformer core grounding current according to claim 4, characterized in that: The frequency domain feature extraction specifically includes: The sliding discrete Fourier transform algorithm is used to calculate the spectrum of the purified ground current signal in real time and extract the amplitudes of the power frequency fundamental wave, second harmonic, third harmonic and fifth harmonic; According to the amplitude of the power frequency fundamental wave, second harmonic, third harmonic and fifth harmonic, the temporal variation trend of each harmonic is analyzed, and the transformer core fault characteristic index is calculated to form a frequency domain characteristic vector. The transformer core fault characteristic index includes the transformer core fault total harmonic distortion rate and the multi-point grounding characteristic ratio. The calculation formula is: Among them, THD core Indicates the total harmonic distortion rate of the iron core transformer core fault, X core [2] represents the amplitude of the second harmonic, X core [3] represents the amplitude of the third harmonic, X core [5] represents the amplitude of the fifth harmonic, X core [1] represents the amplitude of the power frequency fundamental wave, Indicates the multi-point grounding characteristic ratio.

6. The online monitoring system for transformer core grounding current according to claim 4, characterized in that: The time-frequency domain wavelet analysis specifically includes: Perform multi-scale discrete wavelet packet decomposition on the purified ground current signal and calculate the energy distribution of wavelet coefficients in each frequency band; According to the energy distribution of wavelet coefficients in each frequency band, the time-varying characteristics of energy distribution are determined to identify transient abnormal components; The characteristic parameters of transient abnormal components are extracted to form time-frequency domain features.

7. The online monitoring system for transformer core grounding current according to claim 1, characterized in that: The constrained fusion of the multi-dimensional feature parameter set based on the modified core eddy current loss physical model specifically includes: A modified eddy current loss model is established, and the modified eddy current loss model is: in, Indicates the corrected eddy current loss power, k loss represents the eddy current loss coefficient, f represents the current frequency, B represents the magnetic induction intensity, t core Indicates the thickness of the transformer silicon steel sheet, β aging Indicates the insulation aging influence coefficient; Analyzing the frequency spectrum of the purified ground current signal to obtain a measured value of eddy current loss; The eddy current loss deviation is calculated based on the modified eddy current loss model and the actual eddy current loss value, and the deviation is used as a physical constraint condition to constrain the fusion of the multi-dimensional feature parameter set.

8. The online monitoring system for transformer core grounding current according to claim 4, characterized in that: The generation of the comprehensive health evaluation index through the adaptive weight adjustment mechanism specifically includes: Calculate the initial weight coefficient according to the time domain characteristics, frequency domain characteristics, time-frequency domain characteristics and the modified eddy current loss model; The comprehensive health index is calculated based on the initial weight coefficient and the multi-feature parameter set. The calculation formula is: Among them, Health Index represents the comprehensive health evaluation index, i represents the index of the feature, n represents the total number of features, w i Represents the weight coefficient of the i-th feature, F i represents the normalized eigenvalue of the i-th eigenvalue.

9. The online monitoring system for transformer core grounding current according to claim 1, characterized in that: The intelligent diagnosis module includes a fault mode library construction unit, a classification and identification unit, and a trend prediction unit, wherein: The transformer core fault mode library construction unit includes feature templates for normal state, insulation aging, multi-point grounding and local short circuit, and each transformer core fault mode corresponds to a specific feature parameter range and change trend; The classification and identification unit adopts a multi-classifier integration method to identify the transformer core fault type according to the comprehensive health evaluation index and the multi-dimensional feature parameter set to obtain the transformer core fault type; The trend prediction unit determines the degradation degree of the core insulation performance based on the change trend of the comprehensive health evaluation index, and uses a time series analysis method to perform trend prediction on the comprehensive health evaluation index, and generates early warning information according to the prediction result.

10. An online monitoring method for transformer core grounding current, applied to the online monitoring system according to any one of claims 1 to 9, characterized in that: The following steps are involved: S1, synchronously collecting the transformer core grounding current channel signal and the transformer clamp grounding current channel signal of the transformer core grounding lead through a magnetic adsorption dual-channel current sensor; S2. Calculating an adaptive differential coefficient to perform differential processing on the transformer core grounding current channel signal and the transformer clamp grounding current channel signal, and performing variable gain adjustment and adaptive filtering on the differential signals to obtain a purified grounding current signal; S3. Extracting multi-dimensional features of the purified ground current signal to obtain a multi-dimensional feature parameter set; S4. Establish a modified eddy current loss physical model, and based on the modified core eddy current loss physical model, perform constraint fusion on the multi-dimensional feature parameter set through an adaptive weight adjustment mechanism to calculate a comprehensive health evaluation index; S5. Comprehensive health evaluation index, combined with the preset transformer core fault mode library, identifies the transformer core fault type, evaluates the core health status and predicts the development trend, and generates a monitoring report.

Images