Online Monitoring Method and System for AC-DC Hybrid Current in Transformer Core Based on Magnetoelectric Coupling Sensor

Abstract

The present invention relates to the technical field of power equipment condition monitoring, and particularly relates to an on-line monitoring method and system for AC-DC hybrid current in a transformer core based on a magnetoelectric coupling sensor. This method uses a magnetoelectric coupling sensor array to real-time monitor the change in magnetic flux generated by the vibration of the transformer core inside, converts the change in magnetic flux into an original electrical signal, processes the original electrical signal, extracts the AC signal component and DC signal component in the hybrid current, and inputs them along with historical operation data and real-time environmental parameters into a deep recurrent neural network to be trained. After iterative training, the deep recurrent neural network outputs the enhanced stray current and the DC current signal after eliminating the environmental impact. The present invention uses a method combining a deep recurrent neural network and adaptive decomposition to achieve high-precision prediction of the core temperature; and uses dynamic data to train the neural network, thereby enhancing the dynamic processing ability of the network, and can effectively improve the operation reliability of the transformer.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment condition monitoring, and particularly relates to an on-line monitoring method and system for AC-DC hybrid current in a transformer core based on a magnetoelectric coupling sensor. Background Technique

[0002] During operation, a transformer core may generate direct current due to reasons such as unbalanced current in a high-voltage direct current transmission system. These direct currents are important causes for the generation of DC bias current. The DC bias current acts in combination with the alternating current, leading to local overheating and increased vibration of the iron core, threatening the safe operation of the equipment. Traditional measurement methods use current transformers and Hall sensors. For example, the utility model patent with the publication number CN205450103U uses a current transformer for acquisition, and the invention patent with the publication number CN114678962B uses a Hall sensor. However, a current transformer can only measure alternating current, and the accuracy of a Hall sensor is easily affected by temperature drift, with limited ability to cope with strong electromagnetic interference and insufficient long-term stability.

[0003] Moreover, most of the sensors currently used for on-line monitoring of transformer cores are active sensors. Active sensors have the following deficiencies when measuring the current in a transformer core:

[0004] (1) Dependence on external power supply. Active sensors require an additional power supply to support the operation of their internal circuits, increasing the system complexity and potential failure points. Especially when working in harsh environments such as high voltage or outdoors for a long time, the reliability is affected;

[0005] (2) Sensitive to temperature. The electronic components inside active sensors are easily affected by temperature fluctuations, resulting in measurement drift;

[0006] (3) Insufficient electromagnetic compatibility. Active sensors will be affected by electromagnetic interference in working conditions with strong electromagnetic fields or complex high-frequency harmonics, resulting in signal distortion or measurement errors;

[0007] (4) High cost. The manufacturing cost of active sensors is significantly higher than that of passive sensors, restricting their large-scale application in low-cost monitoring scenarios, and the accuracy of the monitoring and processing methods used in the existing technology is not guaranteed either.

[0008] For magnetoelectric coupling sensors, for example, the invention patent JP6131601B2 discloses a magnetic sensor, which discloses the specific principle of converting magnetic flux into an electrical signal, and the utility model patent with the publication number CN218766979U discloses a circuit using a magnetoelectric sensor for signal processing. However, for the real-time monitoring and processing method of a magnetoelectric coupling sensor combined with a transformer core, the accuracy and adaptability of the on-line evaluation of the health status of the transformer core are in a blank state, and the specific scheme cannot be anticipated by the existing technology. Summary of the Invention

[0009] Object of the Invention: In order to overcome the deficiencies of the above-mentioned prior art, the present invention provides an on-line monitoring method for AC-DC hybrid current in a transformer core based on a magnetoelectric coupling sensor. This method solves the problems of low decomposition accuracy of the AC-DC hybrid current that occurs during the operation of the transformer core, resulting in low accuracy and poor adaptability in monitoring the health status of the transformer core. The present invention also provides an on-line monitoring system for AC-DC hybrid current in a transformer core based on a magnetoelectric coupling sensor.

[0010] Technical Solution: According to the first aspect of the present invention, there is provided an on-line monitoring method for AC-DC hybrid current in a transformer core based on a magnetoelectric coupling sensor, the method comprising:

[0011] Due to the change of magnetic flux inside the transformer causing the change of vibration of the transformer core, a magnetoelectric coupling sensor array is used to monitor the vibration change in real time, and the DC bias is indirectly obtained; wherein, the magnetoelectric coupling sensor converts mechanical vibration into an electrical signal by detecting the magnetostrictive effect of the core vibration acting on itself. The core vibration mode is directly related to the magnetic flux distribution: when there is no DC bias, the positive and negative cycles of the core magnetic flux are symmetric, and the vibration is mainly at the fundamental frequency; when there is a DC bias, the positive and negative cycles of the core magnetic flux are asymmetric, resulting in changes in the frequency components in the vibration spectrum. By analyzing the changes in the frequency components of the electrical signal output by the sensor, especially the energy levels of the high and low frequency electrical signals, the degree of DC bias can be quantified. The multiple magnetoelectric coupling sensors in the magnetoelectric coupling sensor array are evenly distributed on the surface of the core inside the transformer;

[0012] Perform gain adjustment and digital processing on the original electrical signal, so as to form the output form of the processed standard current, and the standard current includes the hybrid current formed by alternating current and direct current;

[0013] Perform adaptive decomposition on the standard current, extract the AC signal component and DC signal component in the hybrid current, and input them together with historical operation data and real-time environmental parameters into a deep recurrent neural network to be trained. The AC signal component includes stray current, and after iterative training, the deep recurrent neural network outputs the strengthened stray current and the DC current signal after eliminating the environmental influence;

[0014] Apply a reverse harmonic current of a certain intensity through a filter, use the reverse harmonic current to cancel the amplitude and phase of the strengthened stray current in real time, and calculate the effective value of the current of the stray current after cancellation at present.

[0015] Further, the method further includes:

[0016] Based on the relationships between the three current signal data and the core temperature, a dynamic relationship between current and core temperature is constructed, and the predicted core temperature is displayed in real time. The three current signal data include the DC signal after eliminating environmental influence, the AC signal components after adaptive decomposition, and the effective value of the enhanced stray current.

[0017] Further, it includes:

[0018] The construction of the dynamic relationship between current and core temperature is expressed as: Wherein, is the real-time core temperature; is the DC signal after eliminating environmental influence, is the effective value of the AC current, which is obtained by calculating the root mean square of the AC current components obtained through adaptive decomposition ; is the stray current temperature rise coefficient; is the effective value of the enhanced stray current; is the DC temperature rise coefficient; is the AC temperature rise coefficient; is the initial core temperature; is the environmental temperature rise correction term, which is determined by the environmental temperature rise; is the core DC resistance, which is calculated based on the core material characteristics and geometric dimensions; is the heat dissipation coefficient, which is related to the heat dissipation structure; is the material loss coefficient, which is determined by the hysteresis loss and eddy current loss characteristics of the core material; is the frequency; is the saturation magnetic flux density, which is an inherent property of the core material and is determined according to the material magnetization curve.

[0019] Further, the method also includes: setting a dynamic alarm threshold, which is used to determine whether the core temperature or current inside the transformer is in a dangerous state, and dynamically adjusting the intensity of the processing strategy based on the obtained dangerous state;

[0020] The dynamic alarm threshold is expressed as: ; wherein, is the mean value of historical temperature data, which is calculated by rolling based on the historical operation data of the recent several days; is the standard deviation of historical temperature data, which reflects the temperature fluctuation range; is the temperature rise rate weight factor; is the real-time temperature rise rate, which is obtained through the dynamic relationship between current and core temperature.

[0021] Further, it includes:

[0022] After calculating the effective value of the current stray current after current cancellation, it further includes:

[0023] If the effective value of the current stray current after current cancellation is less than a preset threshold, the temperature change of the iron core is monitored in real time;

[0024] Otherwise, if the effective value of the current stray current after current cancellation is still greater than or equal to the preset threshold, the parameter value of the reverse harmonic current is readjusted again to dynamically suppress the stray current until its effective value is less than the preset threshold.

[0025] Furthermore, it includes:

[0026] The standard current is adaptively decomposed to obtain a plurality of intrinsic mode functions, and each of the intrinsic mode functions represents signal components in different frequency bands, so as to extract the corresponding AC signal component and DC signal component, including:

[0027] The signal components in different frequency bands include high-frequency harmonic components and fundamental wave components. The high-frequency harmonic components come from: when there is a DC bias current in the transformer iron core, the magnetic density increases in half a cycle, the saturation increases, the magnetic permeability decreases, and the leakage magnetic flux increases, which leads to an increase in the vibration of the transformer iron core and generates significant high-frequency harmonic components in the vibration signal;

[0028] The intrinsic mode function corresponding to the fundamental wave component is denoted as F1, and the intrinsic mode functions corresponding to the 2nd - 15th harmonics are denoted as F2 - F 15 ; Define the low frequency as below 120Hz and the high frequency as above 120Hz. Therefore, the low-frequency components are the fundamental wave and the second harmonic, and the high-frequency components are the 3rd - 15th harmonics.

[0029] Furthermore, it includes:

[0030] The historical operation data and real-time environmental parameters are input into the deep recurrent neural network to be trained. The AC signal component includes the stray current. After iterative training, the deep recurrent neural network outputs the strengthened stray current and the DC current signal after eliminating the environmental influence, specifically including:

[0031] Weight is assigned to different frequency bands of the signal through the attention mechanism. High weights are assigned to the frequency bands where the fundamental wave, harmonics, and stray current are located in the AC signal component, and low weights are assigned to the noise frequency bands;

[0032] The deep recurrent neural network to be trained is iteratively optimized, and its output is fed back to the input layer for multi-round feature extraction, that is, the key frequency band features corresponding to the high weights are further strengthened through each round of iteration;

[0033] Input the DC temperature rise coefficient in the historical operation data and the dynamically adjusted heat dissipation coefficient into the deep recurrent neural network for iterative training. The deep recurrent neural network analyzes the sensor drift in the historical operation data through the attention mechanism to obtain the correction value of the DC temperature rise coefficient, and finally obtains the DC current signal that eliminates the environmental impact. The DC temperature rise coefficient is obtained by multiplying the DC resistance of the iron core and the heat dissipation coefficient. The dynamically adjusted heat dissipation coefficient is obtained by collecting the environmental temperature rise correction term in real time and adjusting the value of the heat dissipation coefficient accordingly in combination with the current environmental temperature change.

[0034] Further, it includes:

[0035] Applying a certain intensity of reverse harmonic current through the filter. The reverse harmonic current is represented as: ; where n is the harmonic order, and the range is from 2 to 15. That is, the stray current is formed by the AC harmonic components of the 2nd to 15th order. This AC harmonic component is obtained by the trained deep recurrent neural network. is the n th harmonic suppression weight; is the n th harmonic current amplitude; is the n th original phase angle of the harmonic; is the phase compensation amount, which is adjusted in real time according to the operating state of the iron core and environmental interference to ensure that the compensation current is opposite in phase to the corresponding harmonic. , is the power frequency of 50 Hz.

[0036] On the other hand, the present invention also provides an on-line monitoring system for the AC-DC hybrid current of a transformer core based on a magnetoelectric coupling sensor. The system includes:

[0037] An electrical signal acquisition module for using a magnetoelectric coupling sensor array to monitor the vibration change of the transformer core in real time and indirectly obtain the DC bias. Among them, the magnetoelectric coupling sensor converts mechanical vibration into an electrical signal by detecting the magnetostrictive effect of the iron core vibration acting on itself. A plurality of magnetoelectric coupling sensors in the magnetoelectric coupling sensor array are evenly distributed on the surface of the iron core inside the transformer.

[0038] A signal conditioning module for performing gain adjustment and digital processing on the original electrical signal to form an output form of a standard current. The standard current includes a hybrid current formed by AC and DC.

[0039] A data processing module, which is used to adaptively decompose the standard current, extract the AC signal component and DC signal component in the mixed current, and input them, along with historical operation data and real-time environmental parameters, into a deep recurrent neural network to be trained. The AC signal component includes stray current. After iterative training, the deep recurrent neural network outputs the strengthened stray current and the DC current signal with environmental influence eliminated.

[0040] A stray current suppression module, which is used to apply a certain intensity of reverse harmonic current through a filter, use the reverse harmonic current to cancel the amplitude and phase of the strengthened stray current in real time, and calculate the effective value of the current after cancellation currently.

[0041] Furthermore, the system further includes:

[0042] A temperature prediction module, which is used to construct a dynamic relationship between current and core temperature based on the relationship between three current signal data and core temperature, and display the predicted core temperature in real time. The three current signal data include the DC signal with environmental influence eliminated, the AC signal component after adaptive decomposition, and the effective value of the strengthened stray current.

[0043] Furthermore, the system further includes:

[0044] An alarm module, which is used to set a dynamic alarm threshold, judge whether the core temperature or current inside the transformer is in a dangerous state, and dynamically adjust the intensity of the processing strategy according to the obtained dangerous state.

[0045] The dynamic threshold is expressed as: ; where is the mean value of historical temperature data, which is calculated by rolling based on historical operation data of recent several days; is the standard deviation of historical temperature data, which reflects the temperature fluctuation range; is the weight factor of the temperature rise rate; is the real-time temperature rise rate, which is obtained through the dynamic relationship between current and core temperature.

[0046] Furthermore, it includes:

[0047] The upper and lower layers of the magnetoelectric coupling sensor are magnetostrictive materials, which are used to receive magnetic signals in a magnetic field to generate deformations caused by magnetostriction. The middle layer is a piezoelectric material, which is used to receive the strains generated by the upper and lower layers due to magnetostriction, thereby generating a piezoelectric effect and converting the magnetic signal into an electrical signal. The adjacent two layers are bonded with epoxy resin, which is used to fix and transfer the strains between the materials.

[0048] Furthermore, it includes:

[0049] The piezoelectric material has a sheet structure, which includes an upper electrode surface and a lower electrode surface. The upper electrode surface and the lower electrode surface are externally connected to wires for conducting and outputting signals. An excitation coil is wound around the periphery of the magnetoelectric coupling sensing element, and the excitation coil is electrically connected to the signal conditioning module.

[0050] In a third aspect, the present invention also provides an on-line monitoring device for AC-DC hybrid current in a transformer core based on a magnetoelectric coupling sensor. The device includes: a memory, a processor, and an on-line monitoring program for AC-DC hybrid current in a transformer core based on a magnetoelectric coupling sensor stored on the memory and executable on the processor. When the on-line monitoring program for AC-DC hybrid current in a transformer core based on a magnetoelectric coupling sensor is executed by the processor, it realizes the steps of the on-line monitoring method for AC-DC hybrid current in a transformer core based on a magnetoelectric coupling sensor as described above.

[0051] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0052] (1) The present invention uses a composite sensing material combination and sensor array deployment for data acquisition sensors, which can improve the detection accuracy and anti-interference ability;

[0053] (2) When the present invention uses a neural network algorithm to separate DC signals and AC signals, it takes into account the influence of sensor drift and environmental interference on DC signals, combines the DC temperature rise coefficient in historical operation data and real-time environmental parameters, dynamically adjusts the DC temperature rise coefficient, and uses dynamic data to train the neural network, thereby improving the flexibility and processing ability of the neural network, and making the signal output by the trained neural network a DC current signal eliminating environmental influence, thus improving the separation accuracy of AC signals and DC signals. On this basis, the health state of the transformer core is verified through a current-temperature model.

[0054] (3) Aiming at the problem that the stray current characteristics in the original electrical signal may not be obvious, the present invention uses adaptive decomposition and neural network algorithms for feature enhancement, laying a foundation for improving the stray current suppression effect. Then, the present application constructs a reverse harmonic current to suppress the enhanced stray current in real time, and adjusts the main transformer stray current suppression strategy in real time to improve the effect of the suppression strategy and ensure the safe operation of the transformer;

[0055] (4) On the premise that the stray current is suppressed, the present application constructs a current-temperature correlation model according to the dynamic relationship between the current and the core temperature, thereby monitoring the change of the core temperature in real time. On this basis, a dynamic alarm threshold is set, and the dynamic threshold is used to judge whether the core temperature or current parameter is in a dangerous state, and the intensity of the processing strategy is dynamically adjusted accordingly, making the method more accurate in monitoring the health state of the transformer core.

[0056] (5) Compared with the traditional fixed parameter setting method, the dynamic threshold designed by the present invention can dynamically optimize the system operation state according to the real-time temperature rise rate, AC and DC current characteristics and historical data by adopting a dynamic adjustment method, providing more intelligent and flexible temperature prediction and current monitoring. This mechanism effectively avoids the reaction lag caused by fixed parameters in the traditional method, and enhances the adaptability and response speed of the transformer core health prediction under different working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 is a schematic structural diagram of the magnetoelectric coupling sensor described in Embodiment 1 of the present invention;

[0058] Figure 2 is a flowchart of the on-line monitoring method for the AC and DC hybrid current of the transformer core based on the magnetoelectric coupling sensor described in Embodiment 1 of the present invention;

[0059] Figure 3 is a flowchart of the on-line monitoring method for the AC and DC hybrid current of the transformer core based on the magnetoelectric coupling sensor described in Embodiment 2 of the present invention;

[0060] Figure 4 is a flowchart of the on-line monitoring method for the AC and DC hybrid current of the transformer core based on the magnetoelectric coupling sensor described in Embodiment 3 of the present invention.

[0061] Figure 5 is a schematic structural diagram of the on-line monitoring system for the AC and DC hybrid current of the transformer core based on the magnetoelectric coupling sensor described in Embodiment 4 of the present invention

[0062] Figure 6 is a schematic structural diagram of the on-line monitoring system for the AC and DC hybrid current of the transformer core based on the magnetoelectric coupling sensor described in Embodiment 5 of the present invention;

[0063] Figure 7 is a schematic structural diagram of the on-line monitoring system for the AC and DC hybrid current of the transformer core based on the magnetoelectric coupling sensor described in Embodiment 6 of the present invention;

[0064] Figure 8 is a schematic structural diagram of the on-line monitoring system for the AC and DC hybrid current of the transformer core based on the magnetoelectric coupling sensor described in Embodiment 7 of the present invention;

[0065] Figure 9 is a detailed diagram of the on-line monitoring method for the AC and DC hybrid current of the transformer core based on the magnetoelectric coupling sensor described in the embodiments of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0066] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0067] Embodiment 1

[0068] The present invention captures the change of the magnetic field of the iron core through a magnetoelectric coupling sensor, combines a signal processing algorithm to achieve the separation of alternating current and direct current, and realizes the online evaluation of the health state of the transformer iron core. Specifically, an online monitoring method for the alternating current and direct current hybrid current of the transformer iron core based on a magnetoelectric coupling sensor is provided. As Figure 2 shown, the method includes the following steps:

[0069] S1 Use a magnetoelectric coupling sensor array to continuously monitor the change of magnetic flux generated by the vibration of the iron core inside the transformer, and convert the change of the magnetic flux into an original electrical signal. Multiple magnetoelectric coupling sensors in the magnetoelectric coupling sensor array are evenly distributed on the surface of the iron core inside the transformer.

[0070] Since the change of the magnetic flux inside the transformer causes the change of the vibration of the transformer iron core, a magnetoelectric coupling sensor array is used to continuously monitor the change of the vibration, and the DC bias is indirectly obtained. Among them, the magnetoelectric coupling sensor converts mechanical vibration into an electrical signal by detecting the magnetostrictive deformation caused by the vibration of the iron core. The vibration mode of the iron core is directly related to the magnetic flux distribution: when there is no DC bias, the positive and negative cycles of the magnetic flux of the iron core are symmetric, and the vibration is mainly at the fundamental frequency; when there is a DC bias, the positive and negative cycles of the magnetic flux of the iron core are asymmetric, resulting in changes in the frequency components in the vibration spectrum. By analyzing the changes in the frequency components of the electrical signal output by the sensor, especially the energy levels of the high and low frequency electrical signals, the degree of DC bias can be quantified. Multiple magnetoelectric coupling sensors in the magnetoelectric coupling sensor array are evenly distributed on the surface of the iron core inside the transformer.

[0071] Specifically, in this embodiment, in order to measure the separation accuracy of the alternating current and direct current of this method, it is first necessary to inject a certain amount of current into the transformer to generate a corresponding magnetic field in the iron core. Specifically, this embodiment does not limit the injection method. A certain amount of alternating current can be first introduced, and then a certain amount of direct current can be introduced to cause the change of the vibration of the transformer iron core. The magnetoelectric coupling sensor converts mechanical vibration into an electrical signal by detecting the magnetostrictive effect of the iron core vibration acting on itself. It is also possible to introduce a certain amount of mixed alternating current and direct current at the same time to cause the change of the vibration of the transformer iron core. The magnetoelectric coupling sensor converts mechanical vibration into an electrical signal by detecting the magnetostrictive effect of the iron core vibration acting on itself.

[0072] The magnetoelectric coupling sensor used in this embodiment is asFigure 1 As shown, the magnetoelectric coupling sensing element is composed of a sheet material in a multi-layer form. The upper and lower layers are magnetostrictive materials, which are used to receive magnetic signals in a magnetic field and generate deformations caused by magnetostriction. The middle layer is a piezoelectric material, which is used to receive the strains generated by magnetostriction in the upper and lower layers, generate the piezoelectric effect, and convert the magnetic signal into an electrical signal. The layers are bonded with epoxy resin to fix and transfer the strains between the materials. The magnetoelectric coupling sensing element is more sensitive in the length direction, has a more significant magnetostrictive effect than other directions, and can exhibit a better magnetoelectric coupling coefficient.

[0073] The piezoelectric material is a single crystal or polycrystalline ceramic material, specifically one of AlN, quartz, LiNbO3, BaTiO3, ZnO, Pb(Zr,Ti)O3, Pb(Mg,Nb)O3-PbTiO3, Pb(Zn,Nb)O3-PbTiO3, or BiScO3-PbTiO3.

[0074] The magnetostrictive material is an alloy or oxide with magnetostrictive effect or a magnetostrictive composite material formed by their combination with polymers, in the form of long strip multi-layer thin sheets; the alloys or oxides with magnetostrictive effect include Metglas, terbium dysprosium iron alloy [Terfenol-D (Tb0.27-0.30Dy0.73-0.70Fe1.90-1.95)], nickel iron oxide (NiFe2O4), cobalt iron oxide (CoFe2O4), nickel manganese gallium alloy (Ni2MnGa), etc., or magnetostrictive composite materials formed by the combination of the above magnetostrictive materials with polymers. The magnetostrictive material is a single-layer structure or a multi-layer structure.

[0075] In this embodiment, for the method of detecting the dynamic bias current of a transformer based on magnetoelectric coupling sensing, the main steps are as follows:

[0076] (1) Provide a piezoelectric material, process the piezoelectric material into the required size and shape, and clean it thoroughly with ultrapure water by ultrasonic cleaning;

[0077] (2) Electrodes are plated on the upper and lower ends of the surface of the piezoelectric material by annealing, evaporation, or magnetron sputtering;

[0078] (3) The piezoelectric material is polarized along the thickness direction after the electrodes are made;

[0079] (4) Wires are respectively attached to the two electrode end faces of the two piezoelectric materials to conduct electrical signals;

[0080] (5) Provide a magnetostrictive material, process the magnetostrictive material into the required size and shape, and clean it with alcohol;

[0081] (6) Bond the magnetostrictive material to the surface of the piezoelectric material with an adhesive to form a magnetoelectric coupling sensing element;

[0082] (7) Lead out signals from the wires on both sides of the magnetoelectric coupling sensing element to fabricate the magnetoelectric coupling sensing element.

[0083] (8) Wind an excitation coil around the periphery of the magnetoelectric coupling sensing element, and its sensitive axis direction is along the length direction of the magnetoelectric coupling sensing element.

[0084] (9) Connect the two ends of the coil to the signal drive circuit module, and connect the two electrodes of the magnetoelectric coupling sensing element to the filter amplification circuit module for preliminary signal processing.

[0085] In this embodiment, preferably, a magnetoelectric coupling sensor array is formed through the above magnetoelectric coupling sensor. The magnetoelectric coupling sensors are evenly arranged on the surface of the iron core inside the transformer. This sensor is a passive sensor, and it realizes the "magnetic-mechanical-electric" signal conversion through magnetostriction and piezoelectric effect. A single sensor unit consists of a magnetostrictive material layer made of Fe-Ga alloy flakes, a piezoelectric sensing element made of PTZ-5H ceramic chips, and an electromagnetic shielding cover made of permalloy material, with an additional toroidal magnetic yoke structure for directionally capturing stray magnetic fields and reducing magnetic leakage interference.

[0086] The magnetoelectric coupling sensor array is used to monitor the magnetic flux change generated by the vibration of the iron core in real time and convert it into an electrical signal output. Specifically:

[0087] The change in the magnetic flux inside the transformer causes the change in the vibration of the transformer iron core. The vibration mode of the iron core is directly related to the magnetic flux distribution: when there is no DC bias, the positive and negative cycles of the iron core magnetic flux are symmetric, and the vibration is mainly at the fundamental frequency; when there is a DC bias, the positive and negative cycles of the iron core magnetic flux are asymmetric, resulting in changes in the frequency components in the vibration spectrum; the magnetoelectric coupling sensor detects the magnetostrictive effect of the iron core vibration acting on itself, and transfers the deformation of the magnetostrictive material in the sensor to the middle piezoelectric material layer; the piezoelectric material triggers the piezoelectric effect due to mechanical strain, and generates a voltage signal proportional to the strain amplitude between the upper and lower electrode surfaces, which is called the original electrical signal.

[0088] The alternating current in the iron core generates an alternating magnetic field , the direct current generates a static magnetic field , the magnetic field generated by the AC-DC mixed current , causes the iron core to vibrate due to magnetostriction, and the vibration causes the magnetic flux to change. The formula is: In formula (1), is the magnetoelectric coupling coefficient.

[0089] According to Faraday's law of electromagnetic induction, the output voltage V is proportional to the rate of change of magnetic flux, and the formula is: In formula (2), is the sensor sensitivity coefficient, which is an inherent property of the sensor.

[0090] S2 performs gain adjustment and digital processing on the original electrical signal, thereby forming an output form of a standard current, and the standard current includes a mixed current formed by alternating current and direct current;

[0091] In this embodiment, the gain adjustment and digital processing method includes pre-amplification, filtering, and analog-to-digital conversion to obtain an output form of a standard current.

[0092] Specifically, in this embodiment, the electrical signal is subjected to gain adjustment and digital processing, and the magnetic flux change amount is restored from the voltage signal through integral operation , based on the magnetic characteristics of the iron core and the winding parameters, the magnetic flux change is converted into an equivalent alternating current and direct current mixed current , and the formula is: where, is the magnetic permeability, A is the cross-sectional area of the iron core, N is the number of turns of the winding; thereby forming an output form of a standard current, and the standard current is a mixed current including alternating current and direct current. An important reason for the DC bias of the transformer is the existence of a DC current component in the transformer winding. When a DC current flows through the transformer winding, it will cause the magnetic flux of the iron core to be asymmetric in the positive and negative half cycles, exacerbating the vibration of the iron core. Therefore, the existence of the DC current is an important reason for the DC bias phenomenon.

[0093] In this embodiment, the high-frequency / low-frequency ratio can be used for quantitative calculation, and the formula is: where, R reflects the proportion of high-frequency energy relative to low-frequency energy. The greater the DC bias, the more significant the magnetic flux asymmetry, and the higher the proportion of high-frequency energy. That is, this value reflects the possibility of DC bias, and the larger this value, the higher the probability of its being DC bias.

[0094] For the high-frequency / low-frequency component ratio analysis, when there is a DC bias current in the transformer iron core, the magnetic flux density increases in half a cycle, the saturation increases, the magnetic conductivity decreases, and the leakage magnetic flux increases, resulting in an increase in the vibration of the transformer iron core and generating significant high-frequency harmonic components in the vibration signal; through the ratio of the high-frequency harmonic to the fundamental wave component, the amplitude of the DC bias current is indirectly quantified to avoid the interference problem when directly measuring the DC component.

[0095] Among the obtained intrinsic mode functions, the intrinsic mode function corresponding to the fundamental frequency is denoted as F1, and the intrinsic mode functions corresponding to the 2nd to 15th harmonics are denoted as F2 - F 15 , and each intrinsic mode function represents the signal components in different frequency bands.

[0096] The low frequency is defined as below 120 Hz, and the high frequency is defined as above 120 Hz. Therefore, the low-frequency components are the fundamental frequency and the second harmonic, and the high-frequency components are the 3rd to 15th harmonics. Among them, is the root mean square of the low-frequency component, is the root mean square of the high-frequency component, M is the number of sampling points, represents n the i th sampling value of the intrinsic mode function corresponding to the

[0097] S3 adaptively decomposes the standard current, extracts the AC signal component and the DC signal component in the mixed current, and inputs them together with the historical operation data and the real-time environmental parameters into the deep recurrent neural network to be trained. The AC signal component includes the stray current. After iterative training, the deep recurrent neural network outputs the strengthened stray current and the DC current signal after eliminating the environmental influence.

[0098] Specifically, in this embodiment, the specific steps are as follows:

[0099] S31 performs adaptive decomposition on the processed signal through Hilbert-Huang Transform (HHT) to extract the intrinsic mode functions, that is, to preliminarily separate the AC and DC components.

[0100] Specifically, using HHT to process the non-linear and non-stationary signals input to the data processing unit by the signal conditioning module can effectively separate the complex components in the AC-DC mixed current, such as the power frequency fundamental wave, harmonics, DC offset, and high-frequency noise. Its core lies in the local feature adaptability of empirical mode decomposition, without the need to preset basis functions, solving the limitations of traditional Fourier transform for non-stationary signals. Through this step, the system obtains signal components with physical significance, laying a foundation for subsequent component separation and feature enhancement;

[0101] S32 mixes the decomposed AC and DC signal components and inputs them into the pre-trained DRNN model, and performs frequency-domain feature enhancement and dynamic correction of the DC offset in combination with the historical operation database. Specifically, for AC components such as harmonics and stray current, the DRNN model strengthens the features of key frequency bands through the attention mechanism and suppresses non-related noises such as switching noise.

[0102] In this embodiment, the attention mechanism is a mechanism that allows the model to focus on the key parts when processing information, ignore the non-related information, thereby improving the processing efficiency and accuracy. It mimics the characteristics of selective attention when the human vision processes information. It determines the attention Q weight by calculating the similarity between the query vector (Query) and the key vector (Key), and then performs weighted summation on the value vector (Value) to obtain the final output.

[0103] In a preferred embodiment of the present example, the attention mechanism focuses on the input feature processing stage and the hidden layer iteration process in the deep recurrent neural network. Specifically, it mainly performs weight allocation first, enters the DRNN model for cycling, has an attention feedback mechanism during the cycling process, and finally outputs when the requirements are met.

[0104] In a preferred embodiment of the present example, the data input into the DRNN includes signal components, historical operation data, and real-time environmental parameters. Among them, the signal components are the intrinsic mode functions after HHT decomposition, including AC components and DC components; the historical operation data includes pre-stored harmonic characteristics, stray current patterns, DC temperature rise coefficients, and environmental parameters; the real-time environmental parameters are environmental temperature rise data.

[0105] Among them, the DC temperature rise coefficient is expressed as: In Equation (7), is the DC resistance of the iron core, which is calculated based on the material characteristics and geometric dimensions of the iron core; is the heat dissipation coefficient, which is related to the heat dissipation structure.

[0106] The model training process includes:

[0107] S321 Signal input. Input the DC component and AC component separated by HHT into the pre-trained DRNN model;

[0108] S322 Frequency-domain feature enhancement driven by the attention mechanism. In the frequency domain dimension, weight allocation is performed on different frequency bands of the signal through the attention mechanism, high weights are assigned to the frequency bands where the AC fundamental wave, harmonics, and stray current are located, and low weights are assigned to the noise frequency bands;

[0109] S323 Perform multiple rounds of iterative optimization. Feed the output of the DRNN back to the input layer for multiple rounds of feature extraction; in each iteration, further strengthen the key frequency band features through the attention mechanism;

[0110] S324 Dynamic correction and parameter adjustment. Combine the DC temperature rise coefficient in the historical data and the real-time environmental parameters to dynamically adjust the DC temperature rise coefficient to eliminate the influence of sensor drift and environmental interference; use the historical harmonic database to match and calibrate the harmonic amplitude and phase extracted in real time to improve the separation accuracy.

[0111] For step S324, the system real-time collects the environmental temperature rise correction term , combines the current environmental temperature change, and dynamically adjusts the value of, for example, when the environmental temperature rises, the heat dissipation efficiency decreases, needs to be correspondingly reduced; to eliminate the influence of sensor drift, the DRNN analyzes the sensor drift in the historical data through the attention mechanism, so as to correct Value

[0112] In the output result, the AC component includes key frequency band features such as the AC fundamental wave, harmonics, and stray current strengthened by DRNN; the DC component is the DC current signal that eliminates the environmental impact by dynamically adjusting the temperature rise coefficient and the sensor drift correction term.

[0113] S325 uses the DRNN model to perform multiple rounds of feature extraction on the signal components through the attention mechanism, strengthens the time-frequency features of key frequency bands such as harmonics and stray current during each iteration execution, and suppresses irrelevant noise at the same time. In each cycle, the temperature rise coefficient of the DC component and the sensor drift correction term are dynamically updated in combination with the real-time environmental parameters. In this way, the separation accuracy of AC and DC currents can be improved. This process continues until the separation accuracy error of AC and DC currents is less than 0.5%.

[0114] This embodiment illustrates the processed data: After HHT, the DC component, 50Hz fundamental frequency AC, 100Hz harmonics, and high-frequency noise above 2kHz are obtained; the weights of 50Hz fundamental frequency AC and 100Hz harmonics are strengthened through the attention mechanism, and the high-frequency noise above 2kHz is suppressed; the DC temperature rise coefficient of the DC component is corrected according to the environmental temperature ; The DC component and the AC component are output.

[0115] In this embodiment, the separation accuracy error of the DC current is calculated in the following way:

[0116] Before measuring the AC and DC currents, known DC and AC currents are applied first, the output signals of the sensor are recorded, and the corresponding relationship between the sensor output signals and the applied currents is established. The subsequent monitoring results are based on this calibration relationship;

[0117] The current signals output by the sensor when there is only AC and DC are regarded as the true values of the current, the DC , the AC effective value ; During the measurement of AC and DC, the separated DC current is denoted as , and the AC current effective value is denoted as ; Therefore,

[0118] DC error AC error For a clearer illustration, this application gives an example:

[0119] The true value of the DC component is 10mA, and the true value of the AC component effective value is 500mA;

[0120] Separation result: DC 9.97mA; AC effective value 498mA;

[0121] Calculated error:

[0122] DC error AC error ; Therefore, the above calculation error meets the requirements.

[0123] S4 applies a reverse harmonic current of a certain intensity through a filter, uses the reverse harmonic current to cancel the amplitude and phase of the strengthened stray current in real time, and calculates the effective value of the stray current after cancellation.

[0124] In this embodiment, a reverse harmonic current is applied to the power grid connected to the transformer through a filter, and the type of this power grid is not specifically limited in this embodiment. According to the magnitude of the voltage, it may be a transmission grid or a distribution grid. The reverse harmonic current is correspondingly expressed as: Wherein, n is the harmonic order, and the range is from 2 to 15, that is, the stray current is formed by the AC harmonic components of the 2nd to 15th order. This AC harmonic component is obtained by a trained deep recurrent neural network. is the n th harmonic suppression weight; is the n th harmonic current amplitude; is the n th original phase angle of the harmonic; is the phase compensation amount, which is adjusted in real time according to the core operating state and environmental interference to ensure that the compensation current is opposite to the phase of the corresponding harmonic. , is the power frequency of 50 Hz.

[0125] In this embodiment, that is, the 2nd - 15th order harmonic components are separated by the HHT and DRNN combined algorithm, and their amplitudes and phases are extracted. This algorithm calculates the suppression weights and phase compensation amounts of each harmonic dynamically according to the real - time harmonic amplitudes and phases as well as historical operation data; and transfers the above parameters into the reverse harmonic current formula to adjust the stray current suppression strategy in real time to ensure the stable operation of the transformer core.

[0126] In this embodiment, a preferred calculation specific step can be: input real - time harmonic parameters, environmental temperature, historical operation data; through the attention collaboration mechanism of DRNN, strengthen the weights of key frequency bands, suppress noise interference, and combine historical data to generate preliminary and ; adopt the gradient descent method to adjust and according to the remaining current that has not been completely cancelled by the reverse harmonic current; and and An input dynamic processing module generates reverse harmonic current.

[0127] Therefore, reverse harmonic current is generated according to the above formula and injected into the system through an active filter to cancel the stray current. In this embodiment, after calculating the effective value of the current stray current after cancellation, the following steps are further included:

[0128] If the effective value of the current stray current after cancellation is less than the preset threshold, the temperature change of the iron core is monitored in real time;

[0129] Otherwise, if the effective value of the current stray current after cancellation is still greater than or equal to the preset threshold, the parameter value of the reverse harmonic current is readjusted again to dynamically suppress the stray current until its effective value is less than the preset threshold.

[0130] Specifically, a preferred method in this embodiment is: strengthening the weight of the frequency band where the stray current is located through the HHT and DRNN joint algorithm, calculating the root mean square of the enhanced stray current signal to obtain the effective value of the stray current. If the effective value of the stray current is less than 10 mA / m 2 , the temperature change of the iron core is monitored in real time using the current-temperature correlation model; if the effective value of the stray current is greater than or equal to 10 mA / m 2 , then again through the collaborative mechanism between edge computing nodes, a dynamic optimization solution for the stray current suppression parameters is generated to suppress the stray current and ensure the safe operation of the transformer.

[0131] Therefore, a magnetoelectric coupling sensor array is used to monitor the vibration change of the transformer iron core in real time to indirectly obtain the DC bias. Among them, the magnetoelectric coupling sensor converts mechanical vibration into an electrical signal by detecting the magnetostrictive effect of the iron core vibration acting on itself; digitizes the signal; separates the AC and DC current components; uses a method combining a deep recurrent neural network (DRNN) and machine learning to achieve high-precision prediction of the iron core temperature; adopts the main transformer stray current suppression technology to enhance the dynamic processing ability. The present invention realizes high-precision online monitoring of AC-DC hybrid current, has the advantages of strong anti-interference ability, fast dynamic response speed, high intelligence level, etc., and can effectively improve the operation reliability of the transformer.

[0132] Embodiment 2

[0133] As Figure 3 and Figure 9 shown, on the basis of Embodiment 1, the present invention further includes the following steps:

[0134] Based on the relationships between the three current signal data and the core temperature, a dynamic relationship between current and core temperature is constructed, and the predicted core temperature is displayed in real time. The three current signal data include the DC signal after eliminating environmental effects, the AC signal components after adaptive decomposition, and the effective value of the enhanced stray current.

[0135] In this embodiment, the adopted current-temperature correlation model establishes a dynamic relationship between current and core temperature by fusing the output data from the signal conditioning module and the data processing unit. After decoupling the AC and DC current components and synchronously extracting the stray current characteristics through the combined HHT and RDNN algorithms, the system inputs these current signals into the current-temperature correlation model. The model monitors the change of the core temperature in real time according to the relationship between the change of the current signal and the core temperature, such as the relationship between the DC current component, the effective value of the AC current and the core temperature. The output of the model serves as the basis for the dynamic processing module to optimize the stray current suppression strategy, and the predicted core temperature is displayed in real time on the monitoring terminal, further improving the operation safety of the transformer. The expression of the model is as follows: In Equation (9), T is the real-time core temperature; is the DC current component, which is separated from the original signal by the HHT and DRNN algorithms; is the effective value of the AC current, which is obtained by calculating the root mean square of the AC current component obtained after HHT decomposition; is the stray current temperature rise coefficient, which is dynamically optimized by the machine learning model based on historical operation data and real-time environmental parameters; is the effective value of the stray current, which is obtained by calculating the root mean square after extracting the stray current by the combined HHT and DRNN algorithms; is the DC temperature rise coefficient, and its calculation formula is as shown in Equation (7); is the AC temperature rise coefficient, and its calculation formula is as shown in Equation (11); is the initial core temperature, which is the initial value measured by the temperature sensor at the start of the system; is the environmental temperature rise correction term, which is determined by the environmental temperature rise;

[0136] In Equation (10), is the DC resistance of the core, which is calculated based on the material properties and geometric dimensions of the core; is the heat dissipation coefficient, which is related to the heat dissipation structure;

[0137] In Equation (11), is the material loss coefficient, which is determined by the hysteresis loss and eddy current loss characteristics of the core material; is the frequency, 50Hz; is the saturation magnetic flux density, which is an inherent property of the iron core material and is determined according to the material magnetization curve.

[0138] Embodiment 3

[0139] Based on Embodiment 2, as Figure 4 shown, the present invention further includes the following steps:

[0140] S6 Set a dynamic alarm threshold, which is used to judge whether the iron core temperature or current inside the transformer is in a dangerous state, and dynamically adjust the intensity of the processing strategy according to the obtained dangerous state;

[0141] The dynamic threshold is expressed as: Wherein, is the mean value of historical temperature data, which is calculated by rolling based on the historical operation data of recent several days; is the standard deviation of historical temperature data, which reflects the temperature fluctuation range; is the weight factor of the temperature rise rate, and generally takes a value of 0.5 - 1.5; is the real-time temperature rise rate, which is obtained through the dynamic relationship between the current and the iron core temperature. Specifically, T here is denoted as the real-time iron core temperature in the above formula (9). Specifically, in this embodiment, the present invention continuously records the iron core temperature data and updates it at fixed time intervals and , generally 1 hour; the data processing unit calculates the temperature rise rate in real time, and combines to generate a dynamic threshold; when the real-time temperature T exceeds , the monitoring terminal triggers an alarm and activates the dynamic processing strategy. Therefore, the present invention adopts a composite sensing material combination and sensor array deployment, which can improve the detection accuracy and anti-interference ability. By introducing a collaborative optimization strategy, dynamic adjustment of system parameters such as stray current suppression is realized. Compared with the traditional fixed parameter setting method, the dynamic threshold designed in this application can dynamically optimize the system operation state according to the real-time temperature rise rate, AC and DC current characteristics and historical data, and provide more intelligent and flexible temperature prediction and current monitoring. This mechanism effectively avoids the reaction lag caused by fixed parameters in the traditional method, and enhances the adaptability and response speed of the system under different working conditions.

[0142] Embodiment 4

[0143] The present invention also provides an on-line monitoring system for AC and DC hybrid current of a transformer core based on a magnetoelectric coupling sensor, as Figure 5 shown, the system includes:

[0144] The electric signal acquisition module is used to monitor the vibration change of the transformer core in real time by using a magnetoelectric coupling sensor array, and indirectly obtain the DC bias magnetization. Among them, the magnetoelectric coupling sensor converts mechanical vibration into an electric signal by detecting the magnetostrictive effect of the core vibration acting on itself. Multiple magnetoelectric coupling sensors in the magnetoelectric coupling sensor array are evenly distributed on the surface of the core inside the transformer.

[0145] The signal conditioning module is used to perform gain adjustment and digital processing on the original electric signal, so as to form an output form of a standard current, and the standard current includes a mixed current formed by alternating current and direct current.

[0146] Specifically, the above-mentioned magnetoelectric coupling sensor is connected to the information conditioning module through a shielded twisted pair cable, and an epoxy resin potting seal kit is used when passing through the transformer oil tank, and the seal interface meets the IP68 protection level.

[0147] The signal conditioning module is connected to the sensor array and includes a preamplifier circuit, a band-pass filter and an analog-to-digital conversion unit, and is used to perform gain adjustment and digital processing on the original signal.

[0148] In this embodiment, the adjustable range of the gain of the preamplifier circuit is 40 - 80 dB, the cut-off frequency of the band-pass filter is set to 50 Hz - 2 kHz. The resolution of the analog-to-digital conversion is not less than 16 bits, and the sampling rate ≥ 100 kSPS.

[0149] The data processing module is used to perform adaptive decomposition on the standard current, extract the AC signal component and DC signal component in the mixed current, and input them together with historical operation data and real-time environmental parameters into the deep recurrent neural network to be trained. The AC signal component includes stray current, and the deep recurrent neural network outputs the strengthened stray current and the DC current signal after eliminating the environmental influence after iterative training.

[0150] The data processing module in this embodiment adopts an FPGA-DSP architecture, and is built-in with an AC-DC current component separation algorithm, a current reconstruction model and a main transformer stray current suppression module. Through the deep recurrent neural network (DRNN) algorithm and machine learning method, it fuses historical operation data and real-time monitoring parameters, and combines the DC temperature rise coefficient in historical data and real-time environmental parameters to dynamically adjust the DC temperature rise coefficient, eliminate the influence of sensor drift and environmental interference, so as to realize the dynamic optimization matching of suppression parameters through the algorithm set inside the data processing unit, perform time-frequency domain joint analysis on the conditioned signal, and calculate the effective values of the AC and DC currents of the core.

[0151] Preferably, the processing process of this module corresponding to this embodiment includes:

[0152] A1: Adaptive decomposition is performed on the processed signal through HHT to extract the intrinsic mode functions, that is, to initially separate the AC and DC components. Specifically, HHT is used to process the non-linear and non-stationary signals in the input data processing unit of the signal conditioning module, which can effectively separate the complex components in the AC-DC mixed current, such as power frequency fundamental waves, harmonics, DC offsets, and high-frequency noises. The core lies in the local feature adaptability of empirical mode decomposition, without the need to preset basis functions, which solves the limitations of traditional Fourier transform for non-stationary signals. Through this step, the system obtains signal components with physical meanings, laying a foundation for subsequent component separation and feature enhancement;

[0153] A2: The decomposed AC and DC signal components are mixed and input into a pre-trained DRNN model, and frequency-domain feature enhancement and dynamic correction of the DC offset are performed in combination with the historical operation database. Specifically, for AC components such as harmonics and stray currents, DRNN strengthens the features of key frequency bands through the attention mechanism and suppresses irrelevant noises such as switching noises.

[0154] The data input into DRNN includes signal components, historical operation data, and real-time environmental parameters. Among them, the signal components are the intrinsic mode functions after HHT decomposition, including AC components and DC components; the historical operation data includes pre-stored harmonic features, stray current patterns, DC temperature rise coefficients, and environmental parameters; the real-time environmental parameter is the environmental temperature rise data.

[0155] The processing steps include:

[0156] B1: Signal input. The DC component and AC component separated by HHT are input into the pre-trained DRNN model;

[0157] B2: Frequency-domain feature enhancement driven by the attention mechanism. In the frequency-domain dimension, weight distribution is performed on different frequency bands of the signal through the attention mechanism, high weights are assigned to the frequency bands where the AC fundamental wave, harmonics, and stray currents are located, and low weights are assigned to the noise frequency bands; that is, the high frequency of 2 kHz is assigned a low weight. In this embodiment, no specific restrictions are made on high and low, as long as the noise frequency band and other frequency bands can be clearly distinguished.

[0158] B3: Perform multiple rounds of iterative optimization. The output of DRNN is fed back to the input layer for multiple rounds of feature extraction; in each iteration, the features of key frequency bands are further strengthened through the attention mechanism;

[0159] B4: Dynamic correction and parameter adjustment. Combining the DC temperature rise coefficient in the historical data and the real-time environmental parameters, the DC temperature rise coefficient is dynamically adjusted to eliminate the influence of sensor drift and environmental interference; using the historical harmonic database, the amplitude and phase of the harmonics extracted in real time are matched and calibrated to improve the separation accuracy.

[0160] For B4, the system collects the environmental temperature rise correction term in real time. Combining with the current environmental temperature change, it dynamically adjusts the value of the heat dissipation coefficient. For example, when the environmental temperature rises, the heat dissipation efficiency decreases, and the heat dissipation coefficient needs to be correspondingly reduced. To eliminate the influence of sensor drift, DRNN analyzes the sensor drift in historical data through the attention mechanism and corrects the value of the heat dissipation coefficient.

[0161] In the output result, the AC component includes key frequency band features such as the AC fundamental wave, harmonics, and stray current strengthened by DRNN, and the noise component is suppressed; the DC component is the DC current signal that eliminates the environmental influence by dynamically adjusting the temperature rise coefficient and the sensor drift correction term.

[0162] The DRNN model is used to perform multiple rounds of feature extraction on the signal components through the attention mechanism. During each iteration, the time-frequency features of key frequency bands such as harmonics and stray current are strengthened, and at the same time, the irrelevant noise is suppressed. In each cycle, the temperature rise coefficient and the sensor drift correction term of the DC component are dynamically updated in combination with the real-time environmental parameters. In this way, the separation accuracy of AC and DC currents can be improved. This process continues until the separation accuracy error of AC and DC currents is less than 0.5%.

[0163] Obviously, after the training process of the neural network described in this embodiment is completed, it is also necessary to test with test data and then put it into use. This is the basic process in engineering and will not be elaborated in this application.

[0164] The stray current suppression module is used to apply a certain intensity of reverse harmonic current through a filter, use the reverse harmonic current to cancel the amplitude and phase of the strengthened stray current in real time, and calculate the effective value of the current after cancellation.

[0165] Embodiment 5

[0166] Based on Embodiment 4, as Figure 6 shown, the system further includes:

[0167] The temperature prediction module is used to construct a dynamic relationship between current and core temperature based on the relationship between the three current signal data and the core temperature, and display the predicted core temperature in real time. The three current signal data include the DC signal after eliminating the environmental influence, the AC signal component after adaptive decomposition, and the effective value of the strengthened stray current.

[0168] Embodiment 6

[0169] Based on Embodiment 5, as Figure 7 shown, the system further includes:

[0170] An alarm module, configured to set a dynamic alarm threshold, determine whether the core temperature or current inside the transformer is in a dangerous state, and dynamically adjust the intensity of the processing strategy based on the obtained dangerous state;

[0171] The dynamic threshold is expressed as: Wherein, is the mean value of historical temperature data, calculated by rolling based on historical operation data of recent several days; is the standard deviation of historical temperature data, which reflects the temperature fluctuation range; is the weight factor of the temperature rise rate; is the real-time temperature rise rate, which is obtained through the dynamic relationship between the current and the core temperature.

[0172] Embodiment 7

[0173] Based on any one of Embodiment 4, Embodiment 5 and Embodiment 6, as Figure 8 shown, the system further includes a monitoring terminal, which communicates with the data processing unit through the industrial Internet, integrates a human-machine interaction page, a data storage module and an abnormal alarm unit, and realizes real-time temperature display, historical data backtracking, dynamic processing status feedback and temperature abnormal alarm.

[0174] Other technical features of the on-line monitoring system for AC-DC hybrid current in the transformer core based on the magnetoelectric coupling sensor according to the present invention are similar to the corresponding method of this application, and will not be elaborated here.

[0175] For a better understanding of the implementation manner of the present invention, the present invention will be specifically described in the form of examples. This example is only for further understanding of the invention solution and does not limit the claims.

[0176] Taking the on-line monitoring of the main transformer in a 220 kV substation as an example:

[0177] The system consists of the following parts: a magnetoelectric coupling sensor array, a signal conditioning cabinet, an FPGA-DSP integrated machine, a dynamic processing device, and an industrial Internet platform.

[0178] Implementation steps:

[0179] (1) Uniformly install 12 magnetoelectric coupling sensors on the core surface, apply a 10 A / 50 Hz alternating current to verify the linearity of the sensor output, and then apply a 5 A DC bias magnetic field to test the AC-DC separation accuracy.

[0180] (2) Set the gain of the preamplifier circuit to 60 dB, convert the analog signal into a 16-bit digital signal, and the sampling rate is 200 kSPS.

[0181] Extract the intrinsic mode functions through Hilbert-Huang transform, separate the fundamental wave, DC component and 2-15th harmonics, and calculate the effective values of each component.

[0182] When the DC bias causes the iron core to overheat, perform parameter optimization and active filter control.

[0183] Precisely predict the temperature of the iron core.

[0184] If the temperature exceeds the safety threshold, issue an alarm.

[0185] Compare three processing methods: without DRNN, fixed parameters; with DRNN, fixed parameters; with DRNN, dynamic parameters, that is, the method corresponding to this application. The specific comparison data indicators are shown in Table 1:

[0186] Table 1: Comparison data of three methods

[0187]

[0188] It can be seen from Table 1 that the method adopted in this application has significant advantages in aspects such as AC-DC separation error, temperature prediction error, harmonic suppression efficiency, parameter update time, anti-interference ability, etc., verifying the superiority of this application in terms of system structure and algorithm optimization.

[0189] Finally, the present invention also provides an on-line monitoring device for transformer DC bias based on a magnetoelectric coupling sensor. The device includes: a memory, a processor, and an on-line monitoring program for transformer DC bias based on a magnetoelectric coupling sensor stored on the memory and executable on the processor. When the on-line monitoring program for transformer DC bias based on a magnetoelectric coupling sensor is executed by the processor, it realizes the steps of the on-line monitoring method for transformer DC bias based on a magnetoelectric coupling sensor as described above.

[0190] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0191] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications to these embodiments once they know the basic creative concepts. Therefore, the appended claims are intended to be construed as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0192] Obviously, those skilled in the art can make various changes and modifications to the embodiments of the present invention without departing from the spirit and scope of the embodiments of the present invention. Thus, if these modifications and variations of the embodiments of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these changes and modifications.

Claims

1. A method for online monitoring of AC / DC mixed current in transformer core based on magnetoelectric coupling sensor, characterized in that: The method includes: Since the change of magnetic flux inside the transformer causes the vibration change of the transformer core, a magnetoelectric coupling sensor array is used to monitor the vibration change in real time. The magnetoelectric coupling sensor converts the mechanical vibration into the original electrical signal by detecting the magnetostrictive effect of the core vibration on itself. Performing gain adjustment and digital processing on the original electrical signal to form an output form of a processed current, wherein the processed current includes a mixed current formed by alternating current and direct current; Adaptively decomposing the processed current to obtain a plurality of intrinsic mode functions, each of which represents a signal component in a different frequency band, thereby extracting the corresponding AC signal component and DC signal component, and inputting them into a deep recurrent neural network to be trained along with historical operation data and real-time environmental parameters, wherein the AC signal component includes stray current, and the deep recurrent neural network outputs enhanced stray current and a DC current signal after eliminating environmental influences after iterative training; Applying a reverse harmonic current to the power grid connected to the transformer through a filter, using the reverse harmonic current to offset the amplitude and phase of the enhanced stray current in real time, and calculating the effective value of the stray current after the offset; The processed current is adaptively decomposed to obtain a plurality of intrinsic mode functions, each of which represents a signal component of a different frequency band, thereby extracting the corresponding AC signal component and DC signal component, including: The signal components of different frequency bands include high-frequency harmonic components and fundamental wave components. The high-frequency harmonic components come from: when there is a DC bias current in the transformer core, the half-cycle magnetic density increases, the saturation increases, the magnetic permeability decreases, and the leakage increases, which leads to an increase in the vibration of the transformer core and generates significant high-frequency harmonic components in the vibration signal; The eigenmode function corresponding to the fundamental wave component is recorded as F1, and the eigenmode functions corresponding to the 2nd to 15th harmonics are recorded as F2-F 15 ; And the definition of low frequency is below 120Hz, and high frequency is above 120Hz, so the low frequency component is the fundamental component and the second harmonic, and the high frequency component is the third to fifteenth harmonics; The historical operation data and real-time environmental parameters are input into the deep recurrent neural network to be trained, the AC signal component includes stray current, and the deep recurrent neural network outputs the enhanced stray current and the DC current signal after eliminating the environmental influence after iterative training, specifically including: Assign weights to different frequency bands of the signal, assign high weights to the key frequency bands where the fundamental wave, harmonics and stray current in the AC signal components are located, and assign low weights to the high-frequency components; Iteratively optimize the deep recurrent neural network to be trained, and feed its output back to the input layer for multiple rounds of feature extraction, that is, further strengthen the key frequency band features corresponding to high weights through each round of iteration; The DC temperature rise coefficient in the historical operation data, combined with the dynamically adjusted heat dissipation coefficient, are input into the deep recurrent neural network for iterative training. The deep recurrent neural network analyzes the sensor drift in the historical operation data to obtain a correction value of the DC temperature rise coefficient, and finally obtains a DC current signal that eliminates environmental influences. The DC temperature rise coefficient is obtained by multiplying the DC resistance of the core and the heat dissipation coefficient. The dynamically adjusted heat dissipation coefficient is obtained by collecting the ambient temperature rise correction term in real time, and adjusting the value of the heat dissipation coefficient accordingly in combination with the current ambient temperature changes.

2. The method for online monitoring of AC / DC mixed current of transformer core based on magnetoelectric coupling sensor according to claim 1, characterized in that: The method further includes: Based on the relationship between the three types of current signal data and the core temperature, a dynamic relationship between the current and the core temperature is constructed, and the predicted core temperature is displayed in real time. The three types of current signal data include the DC signal after eliminating environmental influences, the AC signal component after adaptive decomposition, and the effective value of the stray current after enhancement.

3. The method for online monitoring of AC / DC mixed current of transformer core based on magnetoelectric coupling sensor according to claim 2, characterized in that: The dynamic relationship between the build current and the core temperature is expressed as: in, The real-time temperature of the core; is the DC signal after eliminating the environmental influence, is the effective value of the AC current, which is obtained by adaptive decomposition of the AC current components , and its root mean square is calculated; is the stray current temperature rise coefficient; is the effective value of the stray current after enhancement; is the DC temperature rise coefficient; is the AC temperature rise coefficient; is the initial temperature of the core; It is the correction term for ambient temperature rise, which is determined by the ambient temperature rise; is the DC resistance of the core, which is calculated based on the core material properties and geometric dimensions; is the heat dissipation coefficient, which is related to the heat dissipation structure; is the material loss coefficient, which is determined by the hysteresis loss and eddy current loss characteristics of the core material; is the frequency; is the saturation flux density, which is an inherent property of the core material.

4. The method for online monitoring of AC / DC mixed current of transformer core based on magnetoelectric coupling sensor according to claim 3 is characterized in that: The method further includes: setting a dynamic alarm threshold, the dynamic alarm threshold being used to determine whether the core temperature inside the transformer is in a dangerous state, and dynamically adjusting the intensity of the processing strategy according to the obtained dangerous state; The dynamic alarm threshold is expressed as: ;in, It is the average of historical temperature data, which is calculated based on the historical operating data of the last few days. is the standard deviation of historical temperature data, which reflects the temperature fluctuation range; is the temperature rise rate weight factor; is the real-time temperature rise rate.

5. The method for online monitoring of AC / DC mixed current of transformer core based on magnetoelectric coupling sensor according to claim 1, characterized in that: After calculating the effective value of the stray current after the current offset, the method further includes: If the effective value of the stray current after the current offset is less than the preset threshold, the core temperature change is monitored in real time; Otherwise, if the effective value of the stray current after the current offset is still greater than or equal to the preset threshold, the parameter value of the reverse harmonic current is readjusted again to achieve dynamic suppression of the stray current until its effective value is less than the preset threshold.

6. The method for online monitoring of AC / DC mixed current of transformer core based on magnetoelectric coupling sensor according to any one of claims 1 to 5, characterized in that: The reverse harmonic current is applied to the power grid connected to the transformer through the filter, and the reverse harmonic current is correspondingly expressed as: ;in, n is the harmonic order, ranging from 2 to 15, that is, the stray current is formed by the 2nd to 15th order AC harmonic components, and the AC harmonic components are obtained by the trained deep recurrent neural network. For the n Subharmonic suppression weight; For the n Subharmonic current amplitude; For the n The original phase angle of the subharmonics; is the phase compensation amount, which is adjusted in real time according to the core operating status and environmental interference to ensure that the compensation current is opposite to the phase of the corresponding harmonic. , The power frequency is 50Hz.

7. An online monitoring system for AC / DC mixed current of transformer core based on magnetoelectric coupling sensor, characterized in that: The system includes: An electrical signal acquisition module is used to monitor vibration changes in real time using a magnetoelectric coupling sensor array, wherein the vibration changes are caused by changes in the transformer core vibration caused by changes in the internal magnetic flux of the transformer. The magnetoelectric coupling sensor detects the magnetostrictive effect of the core vibration on itself and converts the mechanical vibration into an original electrical signal; A signal conditioning module, used for performing gain adjustment and digital processing on the original electrical signal, so as to form an output form of a processed current, wherein the processed current includes a mixed current formed by alternating current and direct current; A data processing module, used for adaptively decomposing the processed current, extracting the AC signal component and the DC signal component in the mixed current, and inputting them into a deep recurrent neural network to be trained along with historical operation data and real-time environmental parameters, wherein the AC signal component includes stray current, and the deep recurrent neural network outputs enhanced stray current and a DC current signal after eliminating environmental influence after iterative training; A stray current suppression module, used for applying a reverse harmonic current to the power grid connected to the transformer through a filter, using the reverse harmonic current to offset the amplitude and phase of the enhanced stray current in real time, and calculating the effective value of the stray current after the offset; Wherein, in the data processing module, the processed current is adaptively decomposed to obtain multiple intrinsic mode functions, each of which represents a signal component in a different frequency band, thereby extracting the corresponding AC signal component and DC signal component, including: The signal components of different frequency bands include high-frequency harmonic components and fundamental wave components. The high-frequency harmonic components come from: when there is a DC bias current in the transformer core, the half-cycle magnetic density increases, the saturation increases, the magnetic permeability decreases, and the leakage increases, which leads to an increase in the vibration of the transformer core and generates significant high-frequency harmonic components in the vibration signal; The eigenmode function corresponding to the fundamental wave component is recorded as F1, and the eigenmode functions corresponding to the 2nd to 15th harmonics are recorded as F2-F 15 ; And the definition of low frequency is below 120Hz, and high frequency is above 120Hz, so the low frequency component is the fundamental component and the second harmonic, and the high frequency component is the third to fifteenth harmonics; The historical operation data and real-time environmental parameters are input into the deep recurrent neural network to be trained, the AC signal component includes stray current, and the deep recurrent neural network outputs the enhanced stray current and the DC current signal after eliminating the environmental influence after iterative training, specifically including: Assign weights to different frequency bands of the signal, assign high weights to the key frequency bands where the fundamental wave, harmonics and stray current in the AC signal components are located, and assign low weights to the high-frequency components; Iteratively optimize the deep recurrent neural network to be trained, and feed its output back to the input layer for multiple rounds of feature extraction, that is, further strengthen the key frequency band features corresponding to high weights through each round of iteration; The DC temperature rise coefficient in the historical operation data, combined with the dynamically adjusted heat dissipation coefficient, are input into the deep recurrent neural network for iterative training. The deep recurrent neural network analyzes the sensor drift in the historical operation data to obtain a correction value of the DC temperature rise coefficient, and finally obtains a DC current signal that eliminates environmental influences. The DC temperature rise coefficient is obtained by multiplying the DC resistance of the core and the heat dissipation coefficient. The dynamically adjusted heat dissipation coefficient is obtained by collecting the ambient temperature rise correction term in real time, and adjusting the value of the heat dissipation coefficient accordingly in combination with the current ambient temperature changes.

8. The transformer core AC / DC mixed current online monitoring system based on magnetoelectric coupling sensor according to claim 7 is characterized in that: The system also includes: The temperature prediction module is used to construct a dynamic relationship between current and core temperature based on the relationship between three types of current signal data and core temperature, and to display the predicted core temperature in real time. The three types of current signal data include a DC signal after eliminating environmental influences, an AC signal component after adaptive decomposition, and the effective value of stray current after enhancement.

9. The transformer core AC / DC mixed current online monitoring system based on magnetoelectric coupling sensor according to claim 8 is characterized in that: The system also includes: An alarm module is used to set a dynamic alarm threshold, determine whether the core temperature inside the transformer is in a dangerous state, and dynamically adjust the intensity of the processing strategy according to the obtained dangerous state; The dynamic alarm threshold is expressed as: ; in, It is the average of historical temperature data, which is calculated based on the historical operating data of the last few days. is the standard deviation of historical temperature data, which reflects the temperature fluctuation range; is the temperature rise rate weight factor; is the real-time temperature rise rate.

10. The transformer core AC / DC mixed current online monitoring system based on magnetoelectric coupling sensor according to claim 7, characterized in that: The upper and lower layers of the magnetoelectric coupling sensor are magnetostrictive materials, which are used to receive magnetic signals in a magnetic field to generate deformation caused by magnetostriction. The middle layer is piezoelectric material, which is used to receive the strain generated by the upper and lower layers due to magnetostriction, thereby generating a piezoelectric effect and converting the magnetic signal into an electrical signal. The adjacent layers are bonded with epoxy resin to fix and transfer the strain between the materials.

11. The transformer core AC / DC mixed current online monitoring system based on magnetoelectric coupling sensor according to claim 10, characterized in that: The piezoelectric material is a sheet structure, which includes an upper electrode surface and a lower electrode surface. The upper electrode surface and the lower electrode surface are externally connected to wires for conducting output signals. An excitation coil is wound around the periphery of the magnetoelectric coupling sensor element, and the excitation coil is electrically connected to the signal conditioning module.

12. An online monitoring device for AC / DC mixed current of transformer core based on magnetoelectric coupling sensor, characterized in that: The device comprises: a memory, a processor and an online monitoring program for AC / DC mixed current of a transformer core based on a magnetoelectric coupling sensor, which is stored in the memory and can be run on the processor. When the online monitoring program for AC / DC mixed current of a transformer core based on a magnetoelectric coupling sensor is executed by the processor, the steps of the online monitoring method for AC / DC mixed current of a transformer core based on a magnetoelectric coupling sensor as described in any one of claims 1 to 6 are implemented.

Images