Three-phase tmr magnetic field monitoring device and data-driven contactless current reconstruction system

Abstract

The application provides a three-phase TMR magnetic field monitoring device and a data-driven non-contact current reconstruction system, wherein the three-phase TMR magnetic field monitoring device comprises: a closed non-contact measurement body, which is internally provided with a TMR sensing module and a self-adjusting rotating disc for placing the TMR sensing module, and the magnetic sensing direction of the TMR sensing module is adjusted through the self-adjusting rotating disc; an ultrasonic ranging probe is arranged on the upper surface of the non-contact measurement body above the TMR sensing module and is used for detecting the relative position relationship between the three-phase conductor; glass windows are arranged on the opposite two side edges of the non-contact measurement body; monitoring probes are arranged on the side edges of the glass windows; and a support frame is arranged below the non-contact measurement body. The problems existing in the prior art, such as large volume, difficult installation, difficulty in distributed deployment, and limited precision of a traditional analysis model in three-phase system magnetic field inversion, are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of current sensing technology, and in particular to a three-phase TMR magnetic field monitoring device and a data-driven non-contact current reconstruction system. Background Technology

[0002] Currently, power systems are gradually developing towards intelligence, digitalization, and distributed systems, placing higher demands on the immediacy, accuracy, and flexible deployment of current measurement technologies at key nodes. Traditional current measurement technologies typically use current transformers as the mainstream monitoring solution, widely applied in power grids of all levels. However, current transformers suffer from drawbacks such as large size, high power consumption, and inflexible deployment, making it difficult to meet the measurement needs of new distribution networks, distributed energy resources, and end-user electricity scenarios that require multi-point, non-intrusive, and convenient deployment.

[0003] In recent years, with the development of materials science and micro / nano manufacturing technology, high-performance tunneling magnetoresistive (TMR) sensors based on tunneling electrons have been widely used in various fields due to their advantages such as high sensitivity and stability, good linearity, and ease of integration. Magnetoresistive sensors can effectively detect the magnetic field generated by current in a circuit. Using Ampere's circuital law and Biot-Savart's law, a mapping relationship between magnetic induction intensity and current magnitude can be established. However, the mapping relationship between current and magnetic field in three-phase conductors is affected by various factors in practical engineering, such as conductor spatial distribution and interphase coupling, making it impossible for a single linear formula to accurately reconstruct the current. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a three-phase TMR magnetic field monitoring device and a data-driven non-contact current reconstruction system, which effectively solves the problems of large size, difficult installation, and difficulty in distributed deployment of contact measurement methods in the prior art, as well as the limited accuracy of traditional analytical models in the magnetic field inversion of three-phase systems.

[0005] The technical solution provided by this invention is as follows:

[0006] On one hand, the present invention provides a three-phase TMR magnetic field monitoring device, which is disposed around a three-phase conductor and includes:

[0007] The enclosed, non-contact measurement body is internally configured with a TMR sensing module and a self-adjusting rotating disk for placing the TMR sensing module. The magnetic direction of the TMR sensing module is adjusted by the self-adjusting rotating disk.

[0008] An ultrasonic ranging probe is disposed on the upper surface of the non-contact measuring body above the TMR sensing module, and is used to detect the relative positional relationship with the three-phase conductor;

[0009] Glass windows are located on two opposite sides of the non-contact measuring body;

[0010] Monitoring probes positioned on the side of the glass window; and

[0011] A support frame is positioned below the non-contact measuring body.

[0012] On the other hand, the present invention provides a data-driven non-contact current reconstruction system, including the above-mentioned three-phase TMR magnetic field monitoring device, and further including a signal processing device connected to the three-phase TMR magnetic field monitoring device, wherein the signal processing device includes:

[0013] The magnetic field signal processing module is used to stitch together the acquired TMR magnetic field measurement values ​​using a frequency band stitching method.

[0014] The data-driven model is used to establish the influence mapping relationship of various interference sources on the current reconstruction results based on the magnetic field signal processed by the magnetic field signal processing module and the acquired external disturbance signal, to obtain the magnetic field contribution ratio between the three-phase currents, and then reconstruct the magnetic field of the three-phase current.

[0015] The three-phase TMR magnetic field monitoring device and data-driven non-contact current reconstruction system provided by this invention can bring at least the following beneficial effects:

[0016] 1. The non-contact measurement of magnetic field signals using TMR chips avoids the disadvantage of traditional current transformers requiring series wires. The device is small in size and adopts a modular design, which can be quickly deployed in scenarios such as distribution cabinets, substations, and transmission lines. It is low in cost and easy to maintain.

[0017] 2. The system adopts a data-driven model to replace the traditional linear physical formula analysis, and achieves nonlinear mapping relationship fitting from magnetic field to current. It integrates the influence of multiple parameters such as sensor position and interphase interference, effectively distinguishes the main measured current signal from the coupling interference source. The system integrates multiple types of magnetoresistive chips such as high-sensitivity TMR, wide-bandwidth TMR and large-range TMR to adapt to the measurement needs of different current amplitudes, frequency ranges and dynamic characteristics. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the three-phase TMR magnetic field monitoring device in this invention;

[0019] Figure 2 This is a schematic diagram of the TMR sensing module structure in this invention;

[0020] Figure 3 This diagram illustrates the configuration of multiple chip sockets on a PCB board.

[0021] Figure 4 This is a schematic diagram of the interior of a single chip socket in this invention;

[0022] Figure 5 This is a schematic diagram showing the placement relationship between the TMR sensing module and the wires in this invention.

[0023] Figure 6 This is a schematic diagram of the main channel network structure in the data-driven model of this invention;

[0024] Figure 7 This is a schematic diagram of the auxiliary channel network structure in the data-driven model of this invention.

[0025] Figure label:

[0026] 1-Monitoring probe, 2-Glass window, 3-Bottom tray, 4-Ultrasonic ranging probe, 5-TMR sensor module, 6-Shielded wiring conduit, 7-Self-adjusting rotating disk, 8-Support frame, 9-Waterproof box, 10-Triangular bracket, 11-PCB substrate, 12-Signal port, 13-High sensitivity TMR chip, 14-Wideband TMR chip, 15-Large range TMR chip, 16-Three-phase wire, 17-TMR sensor module mounting position, 18-Ground potential, 19-Chip socket, 20-Spring, 21-Electrode. Detailed Implementation

[0027] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the specific implementation methods of the present invention will be described below with reference to the accompanying drawings. Obviously, the accompanying drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings and other implementation methods can be obtained based on these drawings without any creative effort.

[0028] One embodiment of the present invention, such as Figure 1 As shown, a three-phase TMR magnetic field monitoring device includes: a closed non-contact measuring body (including a bottom tray 3), which internally houses a TMR sensing module 5 and a self-adjusting rotating disk 7 for placing the TMR sensing module 5, and adjusts the magnetic direction of the TMR sensing module 5 by means of the self-adjusting rotating disk 7; an ultrasonic ranging probe 4 disposed on the upper surface of the non-contact measuring body above the TMR sensing module 5 for detecting the relative positional relationship with the three-phase conductors; glass windows 2 disposed on two opposite sides of the non-contact measuring body; monitoring probes 1 disposed on the sides of the glass windows 2; and a support frame disposed below the non-contact measuring body.

[0029] In this embodiment, the three-phase TMR magnetic field monitoring device is configured around the three-phase conductors, forming a spatial magnetic field sensing array to sense the power frequency or broadband magnetic induction intensity signal generated by the current conductor during operation in real time. To achieve stable support for the non-contact measuring body, a support frame 8, a waterproof box 9, and a tripod 10 are configured below the non-contact measuring body.

[0030] The core structure of a TMR chip is a magnetic tunnel junction, which relies on the relative magnetization directions between two magnetic layers to sense external magnetic fields. A magnetic tunnel junction typically consists of two layers of ferromagnetic material sandwiching an insulating tunnel layer, such as MgO. The magnetic domains in the free layer can rearrange themselves with changes in the external magnetic field, but the domains in the reference layer remain fixed. Therefore, when the magnetization directions of the free layer and the reference layer are parallel, the probability of electron tunneling is high, leading to increased tunneling conductivity and low overall device resistance. When the magnetization direction of the free layer is affected by an external magnetic field and becomes antiparallel to the reference layer, the spin polarization directions of the two magnetic layers are opposite. This reduces the spin matching of electrons when passing through the tunnel layer, resulting in a lower tunneling probability, decreased conductivity, and an increased overall device resistance.

[0031] In the fabrication of TMR chips, to suppress the influence of the inherent magnetization direction of the reference layer on the consistency of device performance, an external magnetic field is typically applied along the pinning layer direction to pre-cancel or unify the initial magnetic domain alignment. In ferromagnetic materials, the macroscopic manifestation of electron spin magnetic moments is the ordered arrangement of magnetic domains, and their distribution statistics can be described by the Fermi-Dirac distribution function, meaning that the occupancy probability of electrons at different energy levels and spin states is temperature-dependent. From a microstructural perspective, a TMR chip can be approximated as a sandwich-like heterojunction structure composed of a reference magnetic layer, a tunneling barrier layer, and a free magnetic layer. Based on a simplified model and spin polarization theory, its resistance can be derived. R The expression for how magnetization direction changes is as follows:

[0032]

[0033] in, R AP This represents the high configuration resistance value when the free layer is antiparallel to the reference layer. R p This represents the low-resistivity resistance value when the free layer is parallel to the reference layer. H For external magnetic field strength, H s The magnetic field strength when the TMR magnetoresistance reaches saturation; H 0 The magnetic field strength when the TMR reluctance reaches half, i.e. H = H 0 When, R=(R AP + R p ) / 2.

[0034] like Figure 2 As shown, the TMR sensing module 5 includes: a PCB substrate 11; a pluggable magnetoresistive chip slot structure disposed in the PCB substrate 11 for replacing different types of TMR chips 13 / 14 / 15; and a signal port 12 disposed on the side of the PCB substrate, each signal port corresponding to one TMR chip.

[0035] To improve the maintainability and application flexibility of the TMR sensing module 5, this embodiment is designed according to the SOP (Small Outline Package) standard, as follows: Figures 3-4 The pluggable magnetoresistive chip slot structure shown is illustrated. Figure 3 This diagram illustrates the configuration of multiple chip sockets (19) on a PCB board. Figure 4 The diagram illustrates the internal structure of a single chip socket: Utilizing a double-layer PCB, metal contacts are designed on the lower PCB. In chip socket 19, a spring 20 (with an electrode 21 positioned below the spring) works in conjunction with a limiter to secure different standard packaged TMR chips for quick replacement and hot-swapping. This enables chip-level free configuration and modular integration, facilitating switching to the most suitable magnetic field sensing device under different operating conditions.

[0036] The non-contact measurement body is also equipped with a shielded cable tray 6. The signal port of the TMR sensing module 5 is connected to the shielded cable tray 6 through an opto-isolator to output the acquired signal. This can effectively suppress power frequency interference and high-frequency electromagnetic crosstalk, and improve the stability of signal acquisition and anti-interference capability.

[0037] Due to the unique circular distribution of the magnetic field beneath the conductor, such as Figure 5 As shown, the TMR sensor module is preferably installed at position 17 directly below the three-phase conductor 16, with ground potential 18 below the installation position 17. The monitoring probe 1 has openings on both sides, each fitted with a glass window 2. Other parts are made of metal and reliably grounded to reduce external electromagnetic interference. To ensure test sensitivity, the line connecting the two glass windows 2 is perpendicular to the three-phase conductor being measured. The TMR sensing axis is perpendicular to the magnetic field lines, improving measurement sensitivity and directional selectivity. Furthermore, to ensure the magnetic direction of the TMR sensor module 5 is consistent with the magnetic field direction of the target conductor, improving measurement accuracy, a self-adjusting rotating disk 7 is installed below the TMR sensor module 5. This self-adjusting rotating disk 7 includes an angle encoder and a servo drive unit, which automatically adjusts the attitude of the TMR sensor module 5 based on the relative position of the conductor obtained by the ultrasonic ranging probe 4.

[0038] The TMR sensing module 5 is equipped with at least three types of chips: a high-sensitivity TMR chip (small signal TMR chip, frequency band below 10kHz) for monitoring normal operating conditions, a wide-band TMR chip (frequency band greater than 10kHz) for acquiring high-frequency signals, and a large-range TMR chip (magnetic field strength of 200 Oe ~ 1000 Oe) suitable for strong disturbance situations.

[0039] Before use, the TMR chip is calibrated. Then, a high-sensitivity TMR chip and a wideband TMR chip are combined using a bandwidth splicing method. Fourier transform and intensity calibration are employed to combine the signals from the two sensor chips into a single signal. The selection of the TMR chip in TMR sensor module 5 can be configured according to actual conditions. For example, in one instance, a 10kHz (kilohertz) small-signal TMR chip is selected, suitable for monitoring normal operating conditions; a 10MHz wideband TMR chip is suitable for measuring high-frequency signals such as operational overvoltage, power electronic equipment switching interference, and lightning; and a large-range TMR chip with a test value of 1000Oe (Oersted) is used for strong disturbances such as lightning strikes and fault current surges.

[0040] Another embodiment of the present invention provides a data-driven non-contact current reconstruction system, including the aforementioned three-phase TMR magnetic field monitoring device, and a signal processing device connected to the three-phase TMR magnetic field monitoring device. The signal processing device includes: a magnetic field signal processing module, used to stitch together the acquired TMR magnetic field measurements using a frequency band stitching method; and a data-driven model, used to establish a mapping relationship between the influence of various interference sources on the current reconstruction results based on the magnetic field signal processed by the magnetic field signal processing module and the acquired external disturbance signals, to obtain the magnetic field contribution ratio between the three-phase currents, and then reconstruct the three-phase current magnetic field.

[0041] To achieve non-contact, high-precision sensing of three-phase current over a wide frequency range, this embodiment employs dual-channel TMR magnetoresistive sensors to simultaneously acquire magnetic field signals. The two sensor signals are then fused into a single signal using a frequency band splicing and amplitude normalization fusion algorithm. The two channels acquire signals synchronously and amplify them independently, obtaining signal waveforms through front-end ADC conversion. A Fast Fourier Transform is then performed on the two signals, and their frequency domain representation is as follows:

[0042]

[0043] in, B A and B B Two signals from different TMR sensors. Indicates sensor signal B A The test spectrum of the Fourier transform, and These represent the sensor signals respectively. B B The test spectrum of the Fourier transform. In this embodiment, the two corresponding signals are signals acquired by a high-sensitivity TMR chip and a wideband TMR chip.

[0044] The magnetic field signal processing module includes a frequency band splicing unit, which uses Fourier transform to splice the TMR magnetic field measurements to obtain a normalized and fused frequency domain Fourier signal. :

[0045]

[0046] in, and These represent the amplitude normalization coefficients of the test spectrum of the two signals, respectively.

[0047] The inverse Fourier transform unit is used to normalize and fuse the frequency domain Fourier signal from the frequency band splicing unit. Perform an inverse Fourier transform to recover a unified time-frequency domain magnetic field signal, and use this signal for subsequent calculations.

[0048] In a three-phase AC system, the magnetic fields generated by conductors exhibit strong coupling in space. Even if a TMR sensor module is precisely placed directly below the A-phase conductor, the sensed magnetic field strength is not only determined by the A-phase current but also influenced by the superposition of magnetic fields generated by adjacent B / C-phase currents. Especially in scenarios with compact conductor arrangement, suboptimal installation locations, and significant environmental magnetic disturbances, the contribution ratio of the magnetic fields between the three-phase currents is difficult to accurately distinguish using analytical models. Therefore, to achieve high-precision non-contact current reconstruction, a data-driven approach must be introduced to model the three-phase coupling characteristics and clearly define the contribution ratio of different phases to the magnetic field signal. The data-driven model employs a multi-input channel structure, inputting multiple auxiliary variables in addition to the raw TMR magnetic field data, including ultrasonic probe data, surrounding cable layout, and temperature parameters, to establish the mapping relationship between various interference sources and the current reconstruction results.

[0049] In this embodiment, the main channel network in the data-driven model of the components is as follows: Figure 6As shown, the network includes: an input layer (corresponding to the temporal magnetic field sequence in the diagram), a 1D-CNN (for initial extraction of local temporal features), ReLU + BatchNorm + Dropout (corresponding to the activation function, batch normalization function, and random deactivation function to prevent overfitting), multi-layer stacked CNNs (for deep feature extraction), and a Flatten + fully connected layer (the Flatten function is used to compress the multi-dimensional feature maps output by the multi-layer CNNs into one-dimensional vectors, facilitating processing by the fully connected layer). In this network, the 1D-CNN extracts local to complex features of the temporal magnetic field layer by layer. Activation, normalization, and deactivation ensure training stability and feature robustness. Finally, the Flatten + fully connected layer maps the temporal features to the frequency domain. Utilizing the feature extraction capability of CNNs for temporal signals, combined with deep stacking to mine multi-scale patterns, and then achieving cross-domain (temporal → frequency domain) information integration through the fully connected layer, the network extracts the temporal and frequency domain features of the magnetic field signal.

[0050] Auxiliary channel networks such as Figure 7 As shown, the network comprises: an input layer (corresponding to the auxiliary channel data Z(t) in the diagram), a nonlinear activation MLP layer, and an output layer combining prior knowledge and output (corresponding to the multidimensional perturbation correction factor α(t) in the diagram). Based on the auxiliary channel data, this network leverages the prior-integrated MLP to uncover nonlinear correction patterns and outputs a multidimensional factor for perturbation compensation. This allows domain knowledge and data-driven approaches to work together, improving correction accuracy and model rationality.

[0051] External environment vector This can be summarized as follows:

[0052]

[0053] in, For ultrasonic ranging, temperature, Probe spatial distribution parameters.

[0054] Building upon this, the model introduces a learnable weight extraction module. This module adaptively calculates the contribution of different interference types (such as adjacent phase coupling interference and installation offset) to the reconstruction error based on the data differences in the training set, and injects these contributions as weights into the main reconstruction network for joint optimization. It also outputs the contribution ratio of the three-phase current to the TMR observed magnetic field at the current moment, reflecting which phase dominates the current magnetic field signal and the proportion of each phase current. A softmax-normalized weight allocation submodule is embedded in the network structure. This module learns the influence of different input dimensions on the predicted output from the intermediate features extracted from the main and auxiliary channels and outputs them as three-phase weights.

[0055]

[0056] in, This indicates the normalized contribution weight of the three-phase current to the current TMR magnetic field measurement. This represents the TMR magnetic field measurement value at the current time t. This indicates the amount of external interference input. Represents the weight matrix. This represents the bias weight.

[0057] The final current estimation result is output by the main channel prediction model, with the weight vector reflecting the physical coupling trend. A weight sparsity regularization term L is set for current prediction. Based on engineering experience, TMR chips are typically only dominated by 1-2 phases. Therefore, the weight extraction module's learning process combines the location labels and phase-to-phase distance information of each TMR sensor module, using end-to-end supervised training. The loss function includes a disturbance response regularization term to constrain the sparsity and physical rationality of the weight distribution. Its estimation process relies on only a few sets of features, rather than all dimensions significantly participating. During the trial operation phase, the system undergoes supervised training with real current values ​​to correct estimation biases caused by different site scenarios and installation locations. Since the deployment methods, wire structures, and environmental interference of TMR magnetic field sensors vary significantly in practical applications, the device needs to collect a data sample containing magnetic field signals and real current values ​​after deployment. End-to-end training is used to calibrate the model parameters. The calibration process combines the spatial location information and phase-to-phase distance of the sensors to optimize the coupling weights and regularization term distribution in the magnetic field-current mapping relationship, thereby improving the model's adaptability and estimation accuracy in the current scenario.

[0058] Suppose that the current prediction for a certain phase uses a weighted L1 norm with weight w:

[0059]

[0060] in, Indicates the regularization weight. Describing the L1 norm,

[0061] Adding a regularization term enables the network to automatically suppress redundant or irrelevant feature channels, retaining only the few magnetic field components most representative of the current prediction. By combining the spatial location labels and interphase distance information of each TMR sensor module, the model learns a physically interpretable sparse coupling structure, thereby improving prediction accuracy and model generalization ability. Furthermore, the system has online weight adjustment capabilities; even after model deployment, the interference weights can be dynamically adjusted based on newly acquired data through the local update mechanism of edge nodes to adapt to engineering application requirements.

[0062] Compared with traditional current measurement, the above-mentioned three-phase TMR magnetic field monitoring device and data-driven non-contact current reconstruction system have the advantages of being non-contact and flexible in deployment. Through the data-driven current reconstruction system, the device has a certain current testing accuracy and is suitable for high-voltage, bare conductor and other applications requiring distributed current sensing with simple line structure.

[0063] It should be noted that the above embodiments can be freely combined as needed. The above are merely preferred embodiments of the present invention. It should be pointed out that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A data-driven non-contact current reconfiguration system, characterized in that, The device includes a three-phase TMR magnetic field monitoring device, and a signal processing device connected to the three-phase TMR magnetic field monitoring device, wherein the signal processing device includes: The magnetic field signal processing module is used to stitch together the acquired TMR magnetic field measurement values ​​using a frequency band stitching method. The data-driven model is used to establish the influence mapping relationship of various interference sources on the current reconstruction results based on the magnetic field signal processed by the magnetic field signal processing module and the acquired external disturbance signal, to obtain the magnetic field contribution ratio between the three-phase currents, and then reconstruct the magnetic field of the three-phase current. The three-phase TMR magnetic field monitoring device is configured around the three-phase conductors, including: The enclosed, non-contact measurement body is internally configured with a TMR sensing module and a self-adjusting rotating disk for placing the TMR sensing module. The magnetic direction of the TMR sensing module is adjusted by the self-adjusting rotating disk. An ultrasonic ranging probe is disposed on the upper surface of the non-contact measuring body above the TMR sensing module, and is used to detect the relative positional relationship with the three-phase conductor; Glass windows are located on two opposite sides of the non-contact measuring body; Monitoring probes positioned on the side of the glass window; and A support frame is positioned below the non-contact measuring body.

2. The data-driven non-contact current reconfiguration system as described in claim 1, characterized in that, The TMR sensing module includes: PCB substrate; A pluggable magnetoresistive chip slot structure disposed in the PCB substrate is used to replace different types of TMR chips; and Signal ports are configured on the side of the PCB substrate, with each signal port corresponding to a TMR chip.

3. The data-driven non-contact current reconfiguration system as described in claim 1 or 2, characterized in that, The TMR sensing module is equipped with at least three types of chips: a high-sensitivity TMR chip for monitoring normal operating conditions, a wideband TMR chip for acquiring high-frequency signals, and a large-range TMR chip suitable for strong disturbances. The high-sensitivity TMR chip has a frequency band below 10kHz, the wideband TMR chip has a frequency band greater than 10kHz, and the large-range TMR chip has a magnetic field strength of 200Oe to 1000Oe.

4. The data-driven non-contact current reconfiguration system as described in claim 2, characterized in that, The non-contact measurement body is also equipped with a shielded wiring tube. The signal port of the TMR sensing module is connected to the shielded wiring tube through an opto-isolator to output the collected signal.

5. The data-driven non-contact current reconfiguration system as described in claim 1, characterized in that, The self-adjusting rotary disk includes an angle encoder and a servo drive unit, which is used to automatically adjust the attitude of the TMR sensing module according to the relative position of the conductor obtained by the ultrasonic ranging probe, so that its magnetic direction is consistent with the magnetic field direction of the target conductor.

6. The non-contact current reconfiguration system as described in claim 1, characterized in that, The magnetic field signal processing module includes: The frequency band splicing unit is used to splice TMR magnetic field measurements using Fourier transform to obtain a normalized and fused frequency domain Fourier signal. : in, Indicates sensor signal B A The test spectrum of the Fourier transform, Indicates sensor signal B B The test spectrum of the Fourier transform, and These represent the amplitude normalization coefficients of the test spectra of the two signals, respectively. The inverse Fourier transform unit is used to normalize and fuse the frequency domain Fourier signal obtained by the frequency band splicing unit. Perform the inverse Fourier transform.

7. The non-contact current reconfiguration system as described in claim 1, characterized in that, The data-driven model integrates a weight extraction module, which adaptively calculates the contribution of different types of interference to the reconstruction error based on the data differences in the training set, and outputs the contribution ratio of the three-phase current to the TMR magnetic field measurement value at the current time t.

8. The non-contact current reconfiguration system as described in claim 7, characterized in that, The output of the weight extraction module is: in, This indicates the normalized contribution weight of the three-phase current to the current TMR magnetic field measurement. This represents the TMR magnetic field measurement value at the current time t. This indicates the amount of external interference input. Represents the weight matrix. This represents the bias weight.

9. The non-contact current reconfiguration system as described in claim 7, characterized in that, The weight extraction module learning process combines the position labels and inter-phase distance information of each group of TMR sensing modules and is carried out through end-to-end supervised training. The loss function includes a disturbance response regularization term to constrain the sparsity and physical rationality of the weight distribution.

Images