High-precision rogowski coil current measurement method based on intelligent sensor

Abstract

The application belongs to the technical field of current measurement, and particularly relates to a high-precision Rogowski coil current measurement method based on an intelligent sensor, which comprises the following steps: collecting primary current through equal central angle reverse series winding to output high-precision original induced voltage; and building a passive wave trap network to correct wideband signal distortion. Double-layer orthogonal micro-grid is adopted to cooperate with carbon film damping and differential floating front-end to suppress space interference, so that a differential analog signal is obtained. Double-capacitor rotation integration is used to reduce temperature drift, and Hall probe and EKF filtering are combined to fuse AC and DC waveforms. After high-precision ADC sampling, four-dimensional physical parameters are collected and sent into a miniature residual neural network to compensate for multi-physical field errors, a mutual inductance coupling matrix is established, multi-coil channel crosstalk is eliminated through a least square algorithm, and high-precision real-time current values are output. In the application, through the high-precision sensing structure of multi-winding reverse series and I-shaped compensation, wideband, high-precision, strong anti-interference and AC / DC unified current measurement are realized.

Description

Technical Field

[0001] This invention belongs to the field of current measurement technology, specifically a high-precision Rogowski coil current measurement method based on intelligent sensors. Background Technology

[0002] Traditional high-precision Rogowski coil current measurement methods have several core shortcomings: First, they mostly employ a single-winding structure, resulting in weak anti-interference capabilities against spatial magnetic fields. Uneven winding and discontinuous loops easily lead to integration errors, and the single-frame structure exhibits large temperature variations and high residual stress, resulting in severe zero-point drift and poor temperature stability over long-term measurements. Second, they rely on fixed lumped parameter compensation, limiting their ability to correct high-frequency amplitude and phase distortion, and limiting their effective measurement bandwidth, making them unsuitable for wideband and complex harmonic current measurement scenarios. Third, they can only detect AC differential signals, leading to continuous accumulation of temperature drift and leakage current errors in the integrating circuit, and lack of DC observation capabilities, thus failing to achieve integrated AC / DC measurement. Fourth, they lack multi-physics error identification and multi-channel decoupling mechanisms, making them susceptible to temperature, deformation, and crosstalk from adjacent coils, resulting in significant accuracy degradation under complex industrial conditions. Summary of the Invention

[0003] To overcome the shortcomings of existing technologies, this invention proposes a high-precision Rogowski coil current measurement method based on intelligent sensors. This invention primarily addresses issues such as coil temperature drift, crosstalk, poor frequency response, inability to measure DC current, and multi-field coupling errors.

[0004] The high-precision Rogowski coil current measurement method based on intelligent sensors provided by this invention includes: S1: The circuit uses a reverse series structure with independent windings at the center angle to collect the primary current of the power circuit and outputs a native high-precision induced voltage signal.

[0005] S2: Based on the high-stability original induced voltage signal, a finely matched auxiliary compensation winding is added inside the coil to construct a distributed passive notch network, which performs wideband distortion correction on the original induced signal and outputs a wideband pure induced signal.

[0006] S3: Based on the wideband pure induction signal, a double-layer orthogonal microgrid is used to bidirectionally shield stray magnetic fields, and carbon film damping is used to suppress eddy current resonance. It is paired with a differential floating impedance matching front end to output a differential analog voltage signal.

[0007] S4: Based on the differential analog voltage signal, dual capacitor alternating integration is used to suppress leakage current and temperature drift. A miniature Hall probe collects DC zero-point observation values. The differential induction signal and DC reference are fused through EKF extended Kalman filtering, taking into account both wideband AC and DC measurements, and outputting AC and DC analog current waveforms.

[0008] S5: The AC / DC analog current waveform is synchronously sampled by a high-precision ADC to obtain the original digital current. Four-dimensional physical parameters such as temperature and deformation are collected, and the embedded micro residual neural network is input to identify multi-physical field coupling errors. High-precision steady-state and transient current digital data are output.

[0009] S6: Based on high-precision steady-state and transient current digital data, for multi-coil densely arranged measurement scenarios, a multi-loop mutual inductance coupling matrix is ​​constructed. The spatial crosstalk components of adjacent coils are stripped off one channel by one through the least squares iterative algorithm, and high-precision real-time digital current measurement values ​​are output.

[0010] According to the high-precision Rogowski coil current measurement method based on intelligent sensors provided by the present invention, the specific steps in step S1 for outputting the native high-precision induced voltage signal are as follows: S11: Multiple independent windings are evenly arranged along the circumference with equal central angles, and multi-strand twisted Litz wire is wound in the same direction. Adjacent windings are connected in series with opposite ends, so that the induced electromotive force is superimposed in the same direction and the stray magnetic field is canceled in opposite directions, and the initial winding induction signal is output.

[0011] S12: Based on the initial winding induction signal, I-shaped opening compensation magnetic windings are arranged at the gaps of adjacent main windings. The discontinuity integral error of the main winding is compensated by adjusting the number of turns and the opening angle, and the induced voltage signal is output.

[0012] S13: Based on the induced voltage signal after ampere-turn compensation, a gradient low-expansion composite skeleton is used as the support matrix. The thermal expansion coefficients of the three-layer material decrease in a gradient, suppressing temperature deformation interference and outputting a temperature-stable induced voltage signal.

[0013] S14: It takes temperature-stable induced voltage signals as the processing object, adopts micro-stress release winding process, constant tension closed-loop control combined with segmented stress retention and release, low temperature annealing after winding to eliminate residual internal stress, and outputs native high-precision induced voltage signals.

[0014] According to the high-precision Rogowski coil current measurement method based on intelligent sensors provided by the present invention, the specific steps in step S2 for outputting a wideband pure induction signal are as follows: S21: Multiple sets of finely matched auxiliary compensation windings are added along the circumferential direction inside the coil. Each set corresponds to a specific frequency correction band. The number of turns and wire diameter are calculated according to the target resonant frequency to form an inherent resonant circuit and output multi-band auxiliary compensation windings.

[0015] S22: Using the multi-band auxiliary compensation winding as the basic unit, it is cascaded and matched with the main winding distributed parameters. The winding spacing and coupling coefficient are adjusted to control the resonance quality factor, constructing a multi-level RLC notch network and outputting a distributed passive notch network structure.

[0016] S23: The original induction signal is input into the distributed passive notch network structure. The notch unit at each level corrects the amplitude distortion and phase shift segment by segment. The full-band characteristics are optimized by adaptive adjustment of the notch depth, and the induction signal is output.

[0017] S24: Based on the wideband calibrated induction signal, perform harmonic residual detection and signal-to-noise ratio evaluation. If the calibration amount is insufficient, fine-tune the number of turns connected to the auxiliary compensation winding for online correction, and output a wideband pure induction signal.

[0018] According to the high-precision Rogowski coil current measurement method based on intelligent sensors provided by the present invention, the specific steps for outputting the differential analog voltage signal in step S3 are as follows: S31: The coil body is wrapped with an orthogonal staggered conductive microgrid shielding structure. The inner and outer copper microgrids are arranged orthogonally, with the inner layer arranged along the axial direction and the outer layer along the circumferential direction. A polymer insulating layer is set in the middle, and the output has a double-layer orthogonal microgrid shielding structure.

[0019] S32: Based on the double-layer orthogonal microgrid shielding structure, a nano-level carbon film damping layer is set at the grid intersection to suppress eddy current resonance. The two grids are connected to the floating end through high-resistance discharge resistors, and the output damping is optimized by the shielding structure.

[0020] S33: Based on the damping-optimized shielding structure as the shielding carrier, a differentially grounded floating dynamic impedance matching front-end is constructed. The instrumentation amplifier and common-mode choke combination architecture is adopted to dynamically adjust the floating reference point potential and output the dynamic impedance matching front-end circuit.

[0021] S34: The wideband pure induction signal is connected to the dynamic impedance matching front-end circuit, and the damping optimized shielding structure cancels stray magnetic field interference in both radial and axial directions, and outputs a differential analog voltage signal.

[0022] According to the high-precision Rogowski coil current measurement method based on intelligent sensors provided by the present invention, the specific steps for outputting the differential analog voltage signal in step S34 are as follows: Data on the radial and axial components of stray industrial magnetic fields are collected. The induced eddy current intensity of two orthogonal microgrids in two directions is calculated, and radial eddy current cancellation sequences and axial eddy current cancellation sequences are generated.

[0023] The radial eddy current cancellation sequence is matched with the shielding parameters of the outer circumferential grid. The eddy current antiphase depth is dynamically adjusted according to the magnetic field frequency and amplitude to maximize the radial stray magnetic field cancellation efficiency and output the radial shielding optimization parameters.

[0024] Based on the axial eddy current cancellation sequence and radial shielding optimization parameters, a matching calculation is performed with the shielding parameters of the inner axial mesh to adjust the axial shielding depth and output bidirectional shielding collaborative optimization parameters.

[0025] Based on the bidirectional shielding collaborative optimization parameters, the pure inductive signal optimized by radial and axial bidirectional shielding is connected to the differentially grounded floating dynamic impedance matching front end for impedance transformation and common-mode suppression, and outputs a differential analog voltage signal.

[0026] According to the high-precision Rogowski coil current measurement method based on intelligent sensors provided by the present invention, the specific steps in step S4 for outputting AC / DC analog current waveforms are as follows: S41: Construct a dual-capacitor charge cycling integral structure, with fixed-frequency cycling input to the loop. Before cycling, the idle capacitor is cleared and self-calibrated to output a low-drift integral signal.

[0027] S42: A miniature Hall probe is installed at the air gap in the center of the coil based on the low drift integral signal. A high-sensitivity semiconductor material is used to observe the DC zero bias component and convert it into a voltage signal, outputting the DC zero point observation value.

[0028] S43: Based on the DC zero-point observation and the low-drift integral signal, an extended Kalman filter fusion algorithm framework is constructed. The state vector contains multi-dimensional state variables, and the noise covariance is adaptively adjusted to output the EKF filter fusion algorithm model.

[0029] S44: Input the differential induction signal of the coil into the EKF filtering and fusion algorithm model, and achieve optimal fusion of AC and DC signals through prediction-update recursive loop to output AC and DC analog current waveforms.

[0030] According to the high-precision Rogowski coil current measurement method based on intelligent sensors provided by the present invention, the specific steps in step S44 for outputting AC / DC analog current waveforms are as follows: The differential induced voltage signal of the coil is used as the input to the EKF filter as a high-frequency observation channel to provide high-frequency dynamic information on the rate of change of current and output high-frequency observation update quantity.

[0031] The DC zero-point observation value of the miniature Hall probe is used as the input to the EKF filter of the low-frequency observation channel by the high-frequency observation update value, providing DC component and slow-drift steady-state reference information, and outputting the low-frequency observation update value.

[0032] The high-frequency and low-frequency observation updates are used as dual inputs. The noise covariance weights of the two observation channels are adjusted according to the current signal amplitude and signal-to-noise ratio. When the signal amplitude is high, the weight of the high-frequency channel is increased, and when the signal amplitude is low, the weight of the DC channel is increased. The output is an adaptive fusion coefficient.

[0033] Using the adaptive fusion coefficients as weight parameters, the optimal state estimation of the two signals is completed through an EKF prediction-update iterative loop, and the AC / DC analog current waveform is output.

[0034] According to the high-precision Rogowski coil current measurement method based on intelligent sensors provided by the present invention, the specific steps in step S5 for outputting high-precision steady-state and transient current digital data are as follows: S51: Employs a high-precision analog-to-digital converter to sample AC / DC analog current waveforms, with an anti-aliasing low-pass filter at the front end, outputting raw digital current data.

[0035] S52: Acquire multi-dimensional physical parameters of the coil based on the original digital current data, including ambient temperature, skeleton deformation, winding displacement and shielding potential difference, and output multi-dimensional physical parameter dataset.

[0036] S53: Using multidimensional physical parameter datasets as feature inputs, and combining them with original digital current data, an embedded micro residual neural network model is constructed. The multi-layer residual structure is activated by leakage rectifier linear units, and the output residual neural network compensation model is generated.

[0037] S54: Input the multidimensional physical parameter dataset and the original digital current data into the residual neural network compensation model, identify the multi-physics coupling residual error, generate dynamic compensation coefficients to correct the amplitude and phase, and output high-precision steady-state and transient current digital data.

[0038] According to the high-precision Rogowski coil current measurement method based on intelligent sensors provided by the present invention, the specific steps in step S54 for outputting high-precision steady-state and transient current digital data are as follows: The multidimensional physical parameter dataset is input into the input layer of the residual neural network, normalized, mapped to the feature space, and the coupling features are extracted to output the multi-physics coupling feature vector.

[0039] The multi-physics coupled feature vector is input into the hidden layer of the residual network. The mapping relationship between the multi-physics and measurement error is fitted through residual connections and multi-layer nonlinear transformation, and the error identification result is output.

[0040] Based on the error identification results, real-time amplitude compensation coefficients and phase compensation coefficients are generated. The compensation coefficients and the original digital current data are then corrected point by point, and the preliminary corrected digital current data is output.

[0041] Based on the preliminary corrected digital current data, residual verification and online incremental learning are performed. When the residual exceeds the threshold, the network weight micro-update is triggered, and high-precision steady-state and transient current digital data are output.

[0042] According to the high-precision Rogowski coil current measurement method based on intelligent sensors provided by the present invention, the specific steps in step S6 for outputting high-precision real-time digital current measurement values ​​are as follows: S61: Based on the spatial coordinates and geometric parameters of each coil, a multi-loop mutual inductance coupling matrix is ​​pre-constructed. The mutual inductance coefficients are obtained through finite element simulation and calibration experiments to obtain initial values, and the initial mutual inductance coupling matrix is ​​output.

[0043] S62: Based on the initial mutual inductance coupling matrix, online correction is performed using known load current data. A sliding window forgetting factor mechanism is used to retain recent valid data, and the actual layout and environmental changes are matched in real time to output a dynamically updated mutual inductance coupling matrix.

[0044] S63: Using the dynamically updated mutual inductance coupling matrix as the decoupling operator, a weighted recursive least squares iterative decoupling algorithm is constructed. The current data collected from each channel is used as the observation vector, and the spatial crosstalk components of adjacent coils are stripped off channel by channel to output the channel current data.

[0045] S64: Performs crosstalk residual detection and accuracy verification based on channel current data, accelerates convergence using the latest sampling data, and outputs high-precision real-time digital current measurement values.

[0046] The high-precision Rogowski coil current measurement method based on intelligent sensors provided by this invention is effective.

[0047] The beneficial effects of this invention are as follows:

[0048] 1. This invention utilizes an integrated structural design combining multi-winding reverse series connection, compensating magnetic winding, gradient composite frame, and micro-stress winding. Combined with orthogonal microgrid shielding and differential floating impedance matching architecture, it comprehensively optimizes the coil's physical structure and signal transmission environment, solving the pain points of traditional Rogowski coils such as weak anti-interference, large temperature drift, high integration error, and poor high-frequency characteristics. This achieves wide-bandwidth, highly stable, and strongly anti-interference raw signal acquisition. The structure employs multi-winding reverse superposition with equal central angles to cancel spatial stray magnetic field interference, and relies on I-shaped compensating windings to compensate for integration errors caused by winding discontinuities. The gradient low-expansion frame suppresses parameter drift caused by temperature deformation, and the micro-stress process eliminates residual winding stress, ensuring the long-term structural stability of the coil. The double-layer orthogonal microgrid shielding bidirectionally cancels radial and axial industrial stray magnetic fields, and the differential floating circuit suppresses common-mode interference and spatial crosstalk. Without sacrificing high-frequency bandwidth, it significantly improves signal purity, providing a stable and high-quality raw signal foundation for high-precision measurements.

[0049] 2. This invention utilizes a multi-layered intelligent algorithm optimization, incorporating a distributed passive notch filter network, dual-capacitor alternating integration, EKF fusion algorithm, micro residual neural network, and least squares decoupling algorithm, to achieve signal distortion correction, dynamic error compensation, and multi-channel crosstalk removal. This solves the problems of poor AC / DC compatibility, error accumulation, insufficient accuracy in multiple scenarios, and severe crosstalk from multiple coils in traditional measurements, achieving high-precision, digital current measurement results across all operating conditions. It achieves wideband signal distortion correction using a passive notch filter network and eliminates capacitor leakage current and operational amplifier temperature drift errors through a dual-capacitor integration structure. Combining a Hall probe with EKF filtering fuses AC and DC signals, balancing wideband response and DC measurement capability. The residual neural network identifies residual errors from multi-physics coupling, dynamically correcting amplitude and phase deviations. A dynamic mutual inductance matrix and least squares iterative algorithm remove channel crosstalk in densely packed multi-coil scenarios, outputting high-precision real-time digital current data. This adapts to complex industrial conditions such as steady-state and transient states, significantly improving measurement accuracy and scenario adaptability. Attached Figure Description

[0050] The invention will now be further described with reference to the accompanying drawings.

[0051] Figure 1 This is a flowchart illustrating the steps of a high-precision Rogowski coil current measurement method based on a smart sensor, provided in an embodiment of the present invention. Detailed Implementation

[0052] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below according to specific embodiments.

[0053] like Figure 1 As shown in the embodiment of the present invention, a high-precision Rogowski coil current measurement method based on a smart sensor is provided. The system includes: S1: It adopts a reverse series structure of 6 to 8 sets of independent windings with equal central angles, combined with I-shaped open compensation magnetic winding, gradient low expansion composite skeleton and micro stress release winding process, to sense and collect the primary current of the power circuit and output native high-precision induced voltage signal.

[0054] S11: Multiple independent windings are evenly arranged along the circumference with equal central angles. Each winding is wound in the same direction using multiple strands of twisted Litz wire. Adjacent windings are connected in series with opposite ends, so that the induced electromotive forces of each group are superimposed in the same direction and the stray magnetic fields in space are canceled in opposite directions. This completes the construction of the multi-winding reverse series structure and outputs the initial winding induced signal.

[0055] To address the issues of weak spatial interference immunity and integration error caused by uneven winding distribution in traditional single-winding Rogowski coils, this embodiment employs a multi-group independent winding architecture with equal central angle distribution. The coil circumference is divided into several sectors, with each sector containing an independently wound group of windings. All windings use multi-strand stranded Litz wire of the same number of turns and diameter, wound in the same direction. Litz wire is composed of multiple strands of fine enameled wire, with each strand insulated from the others, effectively suppressing the skin effect, reducing the AC resistance of the winding at high frequencies, and ensuring stable impedance characteristics of the coil over a wide frequency range.

[0056] Adjacent windings are connected in series with reversed phases. The tail end of the first winding is connected to the head end of the second winding in reverse phase, and the tail end of the second winding is connected to the head end of the third winding in positive phase, and so on, forming an alternating reverse series link. This connection method ensures that the induced electromotive forces generated by each winding in response to the measured primary current are superimposed in the same direction, and the total output voltage is the sum of the induced voltages of each winding, guaranteeing measurement sensitivity. Regarding stray magnetic fields from the external space, because the windings are symmetrically distributed in space, the interference electromotive forces generated by the stray magnetic fields in each winding are similar in magnitude and opposite in direction. After being connected in reverse series, they cancel each other out, thus significantly improving the coil's resistance to spatial magnetic field interference.

[0057] The core advantage of the multi-winding reverse series structure lies in its dual mechanism of spatial symmetrical distribution and electrical reverse connection, which achieves active suppression of stray magnetic fields without sacrificing measurement sensitivity. The equal distribution of the central angles of each winding group ensures uniform ampere-turn density along the circumference of the coil, avoiding local sensitivity differences caused by uneven winding in traditional single-winding systems. The selection of the number of windings comprehensively considers both anti-interference effect and process complexity; a larger number of windings provides stronger anti-interference capability, but also increases the difficulty of the winding process. After all windings are arranged and connected, the initial winding induction signal is formed.

[0058] S12: Using the initial winding induction signal as the correction object, an I-shaped open compensation magnetic winding is arranged at the central angle gap between two sets of adjacent main windings. The integral error caused by the discontinuity of the main winding is compensated by adjusting the number of turns and the opening angle of the compensation winding, and the induced voltage signal after ampere-turn compensation is output.

[0059] Because multiple independent windings have physical gaps in the circumferential direction, the main winding is not a complete and continuous ring structure. This causes the ampere-turn characteristic of the coil to deviate from the ideal state, resulting in integration error. To compensate for this defect, this embodiment arranges an I-shaped open compensating magnetic winding at the central angle gap between adjacent main windings. The compensating winding uses a high-permeability nanocrystalline magnetic core as the magnetic circuit guiding medium. Nanocrystalline materials have advantages such as high initial permeability, large saturation magnetic induction intensity, and low high-frequency loss, which can effectively concentrate the magnetic flux at the gap and enhance the induction capability of the compensating winding.

[0060] The compensating winding is made of fine enameled wire tightly wound on an I-shaped magnetic core. The opening of the I-shaped magnetic core is aligned with the notch in the main winding, so that the two end faces of the core face the main windings on both sides. By precisely adjusting the number of turns and the opening angle of the compensating winding, the magnitude and phase of the induced electromotive force generated by the compensating winding can be controlled, so that it precisely fills the ampere-turn loss caused by the notch in the main winding. The compensating winding and the main winding are connected in series in the same direction to the measurement circuit, and the induced voltage of the compensating winding is superimposed with the induced voltage of the main winding and then output.

[0061] The I-shaped open-ended compensating magnetic winding layout optimizes the overall ampere-turn balance of the coil, effectively eliminating systematic errors caused by winding discontinuities. The number of turns and the opening angle of the compensating winding are determined through a combination of finite element simulation and experimental calibration. First, a three-dimensional model of the coil is established using electromagnetic simulation software to calculate the ampere-turn characteristic curves under different compensation parameters. Then, calibration and verification are performed on actual samples to select the compensation parameter combination that minimizes the ampere-turn error of the coil.

[0062] S13: Using the induced voltage signal after ampere-turn compensation as the stabilizing object, a gradient low-expansion composite skeleton is used as the winding support matrix. The thermal expansion coefficient of the three-layer material decreases in gradient to suppress temperature deformation interference and output a temperature-stable induced voltage signal.

[0063] Traditional Rogowski coils typically use a single material as the frame support. When the ambient temperature changes, the frame material expands and contracts with temperature changes, causing changes in the winding geometry and resulting in a drift in the coil mutual inductance coefficient, leading to measurement errors. To solve this problem, this embodiment uses a gradient low-expansion composite frame as the winding support base.

[0064] The composite skeleton consists of three layers from the inside out: a low thermal expansion metal core layer, a polymer buffer layer, and a ceramic coating. The inner layer uses a low thermal expansion coefficient metal material (such as Invar) as a rigid support, ensuring the overall mechanical strength and dimensional stability of the skeleton. The middle layer uses a polymer material (such as polyimide) as a buffer transition layer to alleviate the thermal expansion mismatch between the metal core and the outer ceramic layer, preventing coating cracking due to stress concentration during temperature cycling. The outer layer uses a ceramic material (such as microcrystalline glass-ceramic) as the winding base surface. Ceramic materials have high surface hardness, good wear resistance, and an extremely low thermal expansion coefficient, providing a stable and reliable winding reference surface for the winding.

[0065] The thermal expansion coefficients of the three-layer material exhibit a gradient decreasing from the inside out, with the inner metal layer having a slightly higher coefficient, the middle polymer layer next, and the outer ceramic layer having the lowest. This gradient structure allows thermal stress within the framework to be gradually released along the thickness direction when the temperature changes, avoiding stress concentration at the material interfaces. Simultaneously, the extremely low thermal expansion coefficient of the outer ceramic layer ensures the dimensional stability of the winding base surface, keeping the winding's geometric parameters constant over a wide temperature range, thereby suppressing mutual inductance drift caused by temperature variations.

[0066] The gradient low-expansion composite frame, through a multi-layer material gradient matching design, significantly reduces the overall coefficient of thermal expansion while ensuring the mechanical strength of the frame, thereby significantly improving the temperature stability of the coil. The induced voltage signal, after being strengthened for temperature stability, exhibits a significantly reduced temperature drift coefficient, enabling it to adapt to a wider operating temperature range and providing a reliable structural foundation for high-precision current measurement.

[0067] S14: It takes temperature-stable induced voltage signals as the processing object, adopts micro-stress release winding process, constant tension closed-loop control combined with segmented stress retention and release, low temperature annealing after winding to eliminate residual internal stress, and outputs native high-precision induced voltage signals.

[0068] Residual internal stress generated during the winding process can lead to a decrease in the long-term stability of the coil. As the stress gradually releases over time, it causes a slow change in the winding geometry, resulting in zero-point drift and changes in sensitivity. To eliminate the influence of winding stress, this embodiment uses a micro-stress-relieving winding process to complete the winding.

[0069] The winding process employs a constant tension closed-loop control system. A tension sensor monitors the winding tension in real time, and a servo motor dynamically adjusts the unwinding speed to maintain the winding tension consistently near the set value, with tension fluctuations controlled within a minimal range. This constant winding tension ensures uniform tightness in each layer of winding, preventing localized stress concentrations caused by uneven tension.

[0070] A segmented stress release mechanism is introduced during the winding process. After each segment of the winding is completed, the winding is paused and held at the current tension for a period of time, allowing the newly wound winding sufficient time to release internal stress. The interval and duration of the segmented holding are determined comprehensively based on the winding wire diameter, bobbin diameter, and material properties, ensuring that the main stress release is completed before continuing to wind each segment.

[0071] After winding, the coil undergoes low-temperature annealing. The coil is placed in a constant-temperature chamber and held at an appropriate temperature for a certain period, then slowly cooled to room temperature. Low-temperature annealing further eliminates residual internal stress within the winding material, stabilizing the winding's geometry and electrical properties. The annealing temperature and time must be strictly controlled to ensure sufficient stress release while avoiding excessive temperature that could damage the insulation and bobbin materials.

[0072] Coils fabricated using a micro-stress relief winding process exhibit significantly reduced residual internal stress levels and improved long-term stability. The resulting high-precision induced voltage signal possesses characteristics such as high amplitude accuracy, small phase error, low temperature drift, and good long-term stability, laying a solid foundation for subsequent signal processing and error compensation.

[0073] S2: Based on the high-stability original induced voltage signal, a finely matched auxiliary compensation winding is added inside the coil to construct a distributed passive notch network, which performs wideband distortion correction on the original induced signal and outputs a wideband pure induced signal.

[0074] S21: Multiple sets of finely matched auxiliary compensation windings are added along the circumferential direction inside the coil. Each set corresponds to a specific frequency correction band. The number of turns and wire diameter are calculated according to the target resonant frequency to form an inherent resonant circuit and output multi-band auxiliary compensation windings.

[0075] The frequency response of Rogowski coils exhibits amplitude attenuation and phase shift at high frequencies, a result of the combined effects of the distributed inductance, distributed capacitance, and load impedance of the windings. Traditional lumped parameter compensation methods struggle to achieve accurate correction over a wide frequency band. To address this issue, this embodiment employs a distributed multi-band auxiliary compensation winding scheme.

[0076] Multiple sets of auxiliary compensation windings are arranged circumferentially inside the coil, each corresponding to a specific frequency correction band. The number of turns and wire diameter of the auxiliary compensation windings are precisely calculated based on the target resonant frequency. By controlling the distributed inductance and distributed capacitance of the windings, each set of auxiliary compensation windings forms an inherent resonant circuit near its target frequency. The formula for calculating the resonant frequency is:

[0077] In the formula, The resonant frequency, For the distributed inductance of the auxiliary winding, This refers to the distributed capacitance of the auxiliary winding.

[0078] The distributed inductance can be changed by adjusting the number of turns in the winding. The size of the distributed capacitance can be changed by adjusting the winding wire diameter and the inter-turn distance. The size of the value is determined to precisely control the resonant frequency of each auxiliary compensation winding. The resonant frequencies of each auxiliary compensation winding are evenly distributed within the operating frequency band at logarithmic intervals, covering the entire measurement range from low to high frequencies.

[0079] Multiple sets of auxiliary compensation windings are evenly distributed along the circumference. Each set of windings is wound independently and does not overlap, thus avoiding mutual interference between windings. Energy is transferred between the auxiliary compensation windings and the main windings through electromagnetic coupling. No additional active devices are required, making it a passive compensation method with the advantages of high reliability and no introduction of additional noise.

[0080] S22: Based on the multi-band auxiliary compensation winding as the basic unit, it is cascaded and matched with the main winding distributed parameters. The winding spacing and coupling coefficient are adjusted to control the resonance quality factor, construct a multi-level RLC notch network, and output a distributed passive notch network structure.

[0081] Based on the multi-band auxiliary compensation winding, a distributed passive notch network composed of multi-level RLC notch units is constructed by finely adjusting the coupling relationship between each auxiliary compensation winding and the main winding.

[0082] Each auxiliary compensation winding and its corresponding main winding segment together constitute an RLC notch filter unit, where R represents the equivalent loss resistance of the winding, L represents the distributed inductance of the winding, and C represents the distributed capacitance and load capacitance of the winding. The resonant frequency of the notch filter unit is determined by the distributed parameters of the auxiliary winding, while the notch depth and bandwidth are controlled by the coupling coefficient and quality factor Q between the windings.

[0083] The mutual inductance coupling coefficient between the auxiliary compensation winding and the main winding can be changed by adjusting the spacing between them. A smaller spacing results in stronger coupling and a deeper notch depth. A larger spacing results in weaker coupling and a shallower notch depth. Simultaneously, the quality factor (Q) of the resonant circuit can be adjusted by connecting damping resistors of different values ​​in parallel across the auxiliary winding. A higher Q value results in a narrower notch bandwidth and a deeper notch depth. A lower Q value results in a wider notch bandwidth and a shallower notch depth.

[0084] The parameters of each notch filter unit are optimized based on the correction requirements of the overall frequency response of the coil. For frequencies with significant attenuation on the frequency response curve, a deeper notch depth is used for compensation. For frequencies with less attenuation, a shallower notch depth is used. After cascading each notch filter unit, the overall frequency response exhibits a flattened characteristic.

[0085] The advantage of distributed passive notch filters lies in their ability to simultaneously correct amplitude and phase frequency characteristics over a wide frequency band through the synergistic effect of multiple distributed resonant units. Compared with traditional lumped parameter compensation circuits, distributed compensation methods offer higher correction accuracy and a wider operating bandwidth, while being completely passive, consuming no additional power, and introducing no active noise.

[0086] S23: The original induction signal is input into the distributed passive notch network structure. The notch unit at each level corrects the amplitude distortion and phase shift segment by segment. The full-band characteristics are optimized by adaptive adjustment of the notch depth, and the wideband corrected induction signal is output.

[0087] The original induced signal output from the Rogowski coil is fed into a distributed passive notch filter network, where the signal is corrected segment by segment by notch filter units. Near each resonant frequency, the notch filter unit makes specific adjustments to the amplitude and phase of the signal to compensate for frequency distortion caused by the distributed parameters of the coil itself.

[0088] For amplitude-frequency response correction, each stage of the notch filter provides a certain gain boost at its resonant frequency to offset the amplitude attenuation of the coil at that frequency. By rationally designing the gain and frequency distribution of each stage of the notch filter, the flatness of the amplitude-frequency response of the coil is optimized throughout the entire operating frequency band. The flat amplitude-frequency response ensures that current signals of different frequency components can obtain the same gain, avoiding problems such as excessive attenuation of high-frequency components or insufficient gain of low-frequency components.

[0089] For phase frequency characteristic correction, each stage of the notch filter unit provides a certain phase compensation at its resonant frequency to cancel the phase shift of the coil at that frequency point. An ideal current measurement system should have linear phase frequency characteristics, that is, the phase shift of each frequency component is proportional to the frequency, so as to ensure that the signal waveform is not distorted. The distributed notch filter network makes the overall phase frequency characteristic of the coil approach linear through the phase superposition of multiple resonant units, and controls the phase nonlinearity error within a very small range.

[0090] The adaptive notch depth adjustment mechanism dynamically adjusts the compensation depth of each notch unit based on the frequency composition of the input signal. When a strong signal component is detected in a certain frequency band, the notch depth in that band is appropriately increased to improve correction accuracy. When the signal component is weak, the notch depth is appropriately decreased to avoid introducing additional noise. This adaptive adjustment mechanism enables the network to adapt to current signals with different spectral characteristics, always maintaining optimal correction performance.

[0091] After wideband correction by a distributed passive notch filter network, the amplitude-frequency flatness and phase-frequency linearity of the induced signal are significantly improved, the effective operating frequency band of the coil is greatly widened, and it can accurately measure wideband current signals from power frequency to several megahertz.

[0092] S24: Using the wideband calibrated induction signal as the detection object, perform harmonic residual detection and signal-to-noise ratio evaluation. If the calibration amount is insufficient, fine-tune the number of turns connected to the auxiliary compensation winding for online correction, and output a wideband pure induction signal.

[0093] To ensure the effectiveness of broadband distortion correction, this embodiment introduces a harmonic residual detection and online correction mechanism to monitor and dynamically optimize the quality of the corrected signal in real time.

[0094] First, residual harmonics are detected in the broadband induced signal. The signal is converted from the time domain to the frequency domain using a Fast Fourier Transform (FFT), and the amplitude and phase of each harmonic component are analyzed. The detected frequency response curve is compared with the ideal flat response, and the residual error at each frequency point is calculated. If the residual error in a certain frequency band exceeds the allowable range, it indicates that the correction amount for that frequency band is insufficient, and online correction is required.

[0095] Simultaneously, the signal-to-noise ratio (SNR) of the corrected signal is evaluated, calculating the ratio of effective signal power to noise power. SNR is a crucial indicator of signal quality; a higher SNR indicates a cleaner signal and higher measurement accuracy. If an abnormally low SNR is observed in a particular frequency band, it may be due to excessively deep notch filtering in that band, leading to noise amplification. In such cases, appropriate adjustments to the compensation parameters are necessary.

[0096] When insufficient correction or abnormal signal-to-noise ratio is detected, online correction is performed by fine-tuning the number of turns connected to the auxiliary compensation winding. The auxiliary compensation winding is designed with multiple taps; switching different taps using a high-speed analog switch can change the number of turns in the connected circuit, thereby adjusting the resonant frequency and notch depth. The online correction process does not require power outages or system shutdowns, and parameter adjustments can be completed while the system is operating normally.

[0097] After closed-loop optimization through harmonic residual detection and online correction, the signal quality of the coil remains optimal throughout the entire operating frequency band, and the output wideband pure induction signal has the characteristics of low distortion, high signal-to-noise ratio, and wide bandwidth.

[0098] S3: Based on the wideband pure induction signal, an orthogonal interlaced conductive micro-grid shielding structure is used in conjunction with a differentially floating dynamic impedance matching front end to cancel external industrial stray magnetic field interference in both radial and axial directions. Without sacrificing high-frequency bandwidth, common-mode interference and spatial crosstalk are suppressed, and a differential analog voltage signal is output.

[0099] S31: The coil body is wrapped with an orthogonal staggered conductive microgrid shielding structure. The inner and outer copper microgrids are arranged orthogonally, with the inner layer arranged along the axial direction and the outer layer along the circumferential direction. A polymer insulating layer is set in the middle, and the output has a double-layer orthogonal microgrid shielding structure.

[0100] Industrial environments are filled with stray magnetic field interference, which can enter the Rogowski coil via electromagnetic coupling, causing measurement errors. While traditional metal shielding can block electric field interference, it generates significant eddy currents, severely impacting the coil's high-frequency response characteristics. To address this issue, this embodiment employs an orthogonal staggered conductive microgrid shielding structure.

[0101] The shielding structure consists of two layers of copper microgrids, arranged orthogonally. The inner microgrid is arranged along the axial direction of the coil, with grid lines parallel to the coil axis. The outer microgrid is arranged along the circumferential direction of the coil, with grid lines encircling the coil circumference. A polymer insulating layer is placed between the two microgrids to electrically isolate them and prevent circulating current from forming between the two layers.

[0102] The microgrid structure is composed of finely woven copper wires, with the grid line width and spacing designed based on shielding requirements and high-frequency characteristics. Compared to continuous metal shielding layers, the microgrid structure has the following advantages: First, the gaps in the grid allow magnetic field lines to partially pass through without generating strong eddy current effects, thus having minimal impact on the high-frequency response characteristics of the coil. Second, the conductive lines in the grid can still generate induced eddy currents, which can cancel out interfering magnetic fields, providing a certain degree of magnetic field shielding capability. Third, the two orthogonally arranged grid layers are sensitive to magnetic fields in different directions, enabling directional shielding.

[0103] The inner axial mesh primarily induces eddy currents in magnetic fields varying along the circumference (i.e., axial interference magnetic fields), while the outer circumferential mesh primarily induces eddy currents in magnetic fields varying along the axial direction (i.e., radial interference magnetic fields). The two mesh layers are orthogonally arranged, each performing its specific function, together forming a comprehensive magnetic field shielding system. An insulating layer between the two mesh layers ensures that the induced eddy currents of the two layers do not interfere with each other, allowing each to independently perform its shielding function.

[0104] The double-layer orthogonal microgrid shielding structure effectively shields stray magnetic fields while ensuring good high-frequency characteristics, resolving the contradiction between shielding effectiveness and high-frequency bandwidth in traditional shielding methods. The completed double-layer orthogonal microgrid shielding structure provides a shielding carrier for subsequent damping optimization and impedance matching.

[0105] S32: Based on a double-layer orthogonal microgrid shielding structure, a nano-scale carbon film damping layer is set at the grid intersection to suppress eddy current resonance. The two grids are connected to the floating end through high-resistance discharge resistors, resulting in an output damping optimized shielding structure.

[0106] Microgrid shielding structures may generate eddy current resonance at certain frequencies, leading to a decrease in shielding effectiveness or even introducing additional interference. To suppress eddy current resonance, this embodiment incorporates a nanoscale carbon film damping layer at the intersections of the shielding layer grid.

[0107] A nanometer-thick carbon film with a certain resistivity and sheet resistance is deposited at the wire intersections of the microgrid. When induced eddy currents are generated in the grid, the current experiences energy loss as it passes through the carbon film at the intersections, thereby increasing the equivalent damping of the eddy current loop and suppressing resonance. The carbon film damping layer functions similarly to adding a damping resistor in a resonant circuit, effectively reducing the resonance peak and broadening the effective shielding bandwidth.

[0108] The thickness and sheet resistance of the nano-carbon film damping layer were determined through simulation and experimental optimization. The goal was to ensure sufficient damping to suppress resonance while avoiding excessive damping that would reduce shielding efficiency. The carbon film was prepared using a vacuum deposition process, resulting in uniform thickness, stable performance, and long-term reliable operation.

[0109] Meanwhile, the inner and outer microgrids are each connected to the differential floating ground terminal through independent high-impedance bleeder resistors. The high-impedance bleeder resistors have a large resistance value, which ensures that the electrostatic charge accumulated on the shielding layer is slowly discharged, avoiding the impact of electrostatic high voltage on the measurement circuit. At the same time, because the resistance value is large enough, no low-frequency circulating current will be formed between the shielding layer and ground, ensuring the floating characteristics of the shielding layer.

[0110] The inner and outer mesh layers are independently bleeded, avoiding electrical connection between the two layers through the bleed resistors and ensuring the independence of each layer. High-precision, high-stability high-voltage resistors are used for the bleed resistors to ensure long-term reliability.

[0111] The shielding structure, optimized with nano-carbon film damping, effectively suppresses eddy current resonance, resulting in a wider shielding bandwidth and more stable shielding performance. This optimized damping structure provides a high-quality shielding foundation for subsequent differential floating dynamic impedance matching.

[0112] S33: Using a damped optimized shielding structure as the shielding carrier, a differentially grounded floating dynamic impedance matching front-end is constructed. An instrumentation amplifier and common-mode choke combination architecture is adopted to dynamically adjust the floating reference point potential and output the dynamic impedance matching front-end circuit.

[0113] To further suppress common-mode interference and achieve optimal equipotential matching of the shielding layer, this embodiment constructs a differentially grounded floating dynamic impedance matching front-end circuit. The front-end circuit adopts a combination architecture of an instrumentation amplifier and a common-mode choke to achieve high input impedance, high common-mode rejection ratio, and dynamic impedance matching.

[0114] Instrumentation amplifiers offer advantages such as high input impedance, high common-mode rejection ratio (CMRR), and stable gain, making them ideal for pre-amplifying weak signals. The two input terminals of the instrumentation amplifier are connected to the two output terminals of a Rogowski coil, amplifying the differential signal output from the coil while suppressing common-mode interference signals shared by both terminals. The common-mode rejection ratio of the instrumentation amplifier can reach over 120dB at power frequency, effectively suppressing power frequency common-mode interference.

[0115] A common-mode choke is connected in series in the signal loop, presenting high impedance to common-mode signals and low impedance to differential-mode signals. When used in conjunction with an instrumentation amplifier, a common-mode choke can further enhance the circuit's common-mode rejection capability, particularly providing excellent suppression of high-frequency common-mode interference.

[0116] Differential ground floating technology is the core feature of this front-end circuit. The shielding layer is not directly connected to system ground, but to a floating reference potential point. By detecting the potential difference between the shielding layer and the signal ground in real time, the potential of the floating reference point is dynamically adjusted so that the potential of the shielding layer always follows the changes in the signal common-mode potential, maintaining an equipotential state between the shielding layer and the signal loop.

[0117] When the shielding layer and the signal circuit are at the same potential, the parasitic capacitance between them will not generate displacement current, thus avoiding signal attenuation and phase shift caused by parasitic capacitance. At the same time, equipotential shielding can also effectively suppress external electric field interference from coupling into the signal circuit through parasitic capacitance.

[0118] The potential of the floating reference point is dynamically adjusted by the feedback control circuit. The control circuit detects the potential difference between the shielding layer and the signal ground, and drives the shielding layer potential to approach the target value through an operational amplifier, forming a closed-loop control. The dynamic adjustment has a fast response speed and high accuracy, and can track rapidly changing common-mode potentials.

[0119] The differentially floating dynamic impedance matching front-end circuit achieves optimal impedance matching between the shielding layer and the signal loop, maintaining a high common-mode rejection ratio and low signal distortion over a wide bandwidth, providing a high-quality signal conditioning channel for the output of differential analog voltage signals.

[0120] S34: Connects the wideband pure induction signal to the dynamic impedance matching front-end circuit, and combines the damped optimized shielding structure to cancel stray magnetic field interference in both radial and axial directions, suppress common-mode interference and spatial crosstalk, and output a differential analog voltage signal.

[0121] The clean induction signal, after wideband correction, is connected to the differentially grounded floating dynamic impedance matching front-end circuit. At the same time, the shielding effect of the damped optimized shielding structure is combined to cancel the interference of external industrial stray magnetic fields from both radial and axial directions.

[0122] For radial interference magnetic fields (i.e., magnetic field direction along the coil radius), induced eddy currents are generated in the outer circumferential microgrid. According to Lenz's law, the magnetic field generated by the induced eddy currents is opposite in direction to the original interference magnetic field, thus canceling the radial interference magnetic field. The circumferential grid is arranged around the circumference of the coil and is most sensitive to radially changing magnetic fields, enabling it to efficiently sense and cancel radial interference.

[0123] For axial interference magnetic fields (i.e., magnetic fields oriented along the coil axis), induced eddy currents are generated in the inner axial microgrid, which, according to Lenz's law, cancel the axial interference magnetic field. The axial grid, arranged along the coil axis, is most sensitive to axially varying magnetic fields and is specifically responsible for canceling axial interference.

[0124] Two orthogonal microgrids correspond to magnetic field interference in two orthogonal directions, respectively, without crossing or interfering with each other, together forming a two-dimensional, omnidirectional magnetic field shielding system. For any stray magnetic field in space, it can be decomposed into radial and axial components, which are then canceled out by the two mesh layers. The bidirectional cancellation efficiency can reach over 85%, significantly reducing the impact of stray magnetic fields on measurements.

[0125] While the shielding structure cancels out the magnetic field, the differentially grounded floating dynamic impedance matching front-end circuit performs common-mode rejection and impedance transformation on the signal. The instrumentation amplifier and common-mode choke work together to suppress common-mode voltage interference on the signal line. The differentially grounded floating technology ensures equipotentiality of the shielding layer, suppressing electric field coupling interference. Dynamic impedance matching ensures optimal impedance for signal transmission, reducing signal reflection and attenuation.

[0126] Through the combined effects of magnetic field cancellation via microgrid shielding and common-mode suppression in the front-end circuitry, the system exhibits exceptional resistance to external industrial interference. Without sacrificing high-frequency bandwidth, it achieves comprehensive suppression of stray magnetic fields, common-mode voltage interference, electric field coupling interference, and spatial crosstalk. The output differential analog voltage signal boasts a high signal-to-noise ratio, low distortion, and strong anti-interference capability.

[0127] S4: Based on the differential analog voltage signal, a dual-capacitor charge cycling integral structure is used to eliminate capacitor leakage current and operational amplifier offset temperature drift. A miniature Hall probe is used to observe DC zero bias in real time. The differential induction signal of the coil and the DC zero point observation value are fused through an EKF Kalman filter to output AC and DC analog current waveforms.

[0128] S41: Construct a dual-capacitor charge cycling integral structure, consisting of two high-precision integrating capacitors and a high-speed analog switch. The capacitors are connected to the circuit at a fixed frequency. Before the cycle, the idle capacitors are cleared and self-calibrated to output a low-drift integral signal.

[0129] The Rogowski coil outputs the differential signal of the current (i.e., di / dt), which needs to be integrated to obtain the actual current waveform. Traditional single-capacitor integrator circuits suffer from capacitor leakage current and operational amplifier offset temperature drift. These errors accumulate over integration time, causing zero-point drift and severely affecting measurement accuracy, especially for DC and slowly varying current measurements. To address this issue, this embodiment employs a dual-capacitor charge-cyclic integrator structure.

[0130] The integrating circuit consists of two high-precision integrating capacitors with identical nominal values ​​and a set of high-speed analog switches. The two capacitors are alternately connected to the integrating loop. When one capacitor is in the working state for integration, the other capacitor is in the idle state for charge clearing and self-calibration. The switching of capacitors is controlled by the high-speed analog switches, which switch at a fixed frequency.

[0131] The integrating capacitors are made of polystyrene or polytetrafluoroethylene (PTFE). These capacitors have advantages such as high insulation resistance, low dielectric loss, low temperature coefficient, and good long-term stability, with extremely low leakage current, making them very suitable for high-precision integrating circuits. The two capacitors are strictly matched with their parameters, and the capacitance deviation is controlled within a very small range to ensure the continuity of the integrating gain during switching.

[0132] Before each rotation, the idle capacitors about to enter the working state are cleared of their charge. The residual charge on the capacitors is completely released through the discharge circuit to ensure that the capacitors integrate from zero. At the same time, self-calibration is performed, measuring the current leakage current of the capacitors and the operational amplifier offset voltage, and the calibration results are stored for subsequent error compensation.

[0133] The core advantage of dual-capacitor alternating integration lies in avoiding the error accumulation caused by long-term integration of a single capacitor through periodic alternation. The integration time of each capacitor is limited to one alternation cycle, and the drift caused by leakage current and offset is controlled within a limited range. The higher the alternation frequency, the shorter the integration time of a single capacitor and the smaller the drift, but the switching noise will also increase accordingly. The alternation frequency needs to strike a balance between drift suppression and switching noise.

[0134] After undergoing dual-capacitor charge cycling integration, the drift of the output integral signal is significantly reduced, and the integration accuracy is significantly improved. The low-drift integral signal provides a high-quality AC component foundation for subsequent AC / DC fusion.

[0135] S42: Using the low-drift integral signal as the drift monitoring benchmark, a miniature Hall probe is installed at the air gap in the center of the coil. High-sensitivity semiconductor materials are used to observe the DC zero-bias component in real time and convert it into a voltage signal, outputting the DC zero-point observation value.

[0136] Rogowski coils operate based on the principle of electromagnetic induction and can only measure changing currents, not direct currents. To achieve direct current measurement, this embodiment places a miniature Hall probe at the air gap in the center of the coil and uses the Hall effect to observe the DC zero-bias component in real time.

[0137] Hall probes are made of highly sensitive semiconductor materials (such as gallium arsenide, GaAs), are small in size, and can be easily installed in the air gap at the center of a coil without affecting the coil's structure. Hall probes operate based on the Hall effect; when current flows through the Hall element and the element is in a magnetic field, a Hall voltage is generated in a direction perpendicular to both the current and the magnetic field. The magnitude of the Hall voltage is proportional to the magnetic field strength.

[0138] The magnetic field strength at the air gap at the center of the coil is proportional to the magnitude of the measured current. Therefore, the magnitude of the measured current, especially the DC component, can be deduced by measuring the Hall voltage. The Hall probe's measurement bandwidth covers the range from DC to several kilohertz, providing accurate DC zero-point reference and low-frequency current information.

[0139] The output of the miniature Hall probe is converted into a DC zero-bias voltage signal after precise amplification and temperature compensation. Since the sensitivity and zero bias of the Hall element change with temperature, temperature compensation is necessary. A temperature sensor is placed near the probe to measure the ambient temperature in real time, and the Hall output is corrected according to a pre-calibrated temperature coefficient to ensure measurement accuracy.

[0140] The low-drift integral signal was used as the drift monitoring benchmark to cross-calibrate the output of the Hall probe. When the drift trend of the integral signal is consistent with the change trend of the Hall signal, it indicates a real current change. When the trends are inconsistent, it indicates a possible drift error, requiring further calibration.

[0141] The miniature Hall probe provides an independent DC observation channel, complementing the AC channel of the Rogowski coil. The DC zero-point observation provides a DC component reference for subsequent EKF fusion, enabling the system to simultaneously measure AC and DC currents.

[0142] S43: Using DC zero-point observations as calibration references, an extended Kalman filter fusion algorithm framework is constructed by combining low-drift integral signals. The state vector contains multi-dimensional state variables, and the noise covariance is adaptively adjusted to output an EKF filter fusion algorithm model.

[0143] To organically fuse the broadband AC signal from the Rogowski coil with the DC signal from the Hall probe, this embodiment constructs an extended Kalman filter (EKF) fusion algorithm framework. The extended Kalman filter is a generalization of the Kalman filter to nonlinear systems, capable of optimally estimating the state of nonlinear systems.

[0144] The algorithm's state vector contains state variables in multiple dimensions: instantaneous current value, rate of change of current, DC bias, and temperature drift coefficient. The instantaneous current value represents the current magnitude being measured and is the primary estimation target. The rate of change of current reflects the trend of current change and is directly related to the differential output of the Rogowski coil. The DC bias represents the DC component of the current and is observed by a Hall probe. The temperature drift coefficient characterizes the system's temperature drift properties and is used to dynamically compensate for errors caused by temperature changes.

[0145] The observation equation incorporates two observation signals simultaneously: one is the differential induced voltage of the Rogowski coil after integration, corresponding to the rate of change of current and the instantaneous value of current in the state vector; the other is the DC observation value of the miniature Hall probe, corresponding to the DC zero bias in the state vector. These two observation signals reflect the state of the measured current from different perspectives, forming redundant observations and improving the reliability of the estimation.

[0146] The process noise covariance and observation noise covariance are adaptively adjusted based on the signal amplitude. When the signal amplitude is large, the coil's signal-to-noise ratio is high, and the observation noise is relatively small, thus increasing the weight of the coil observation. When the signal amplitude is small, the coil's signal-to-noise ratio decreases, while the Hall probe is relatively more accurate in measuring small signals, thus increasing the weight of the Hall observation. This adaptive noise covariance adjustment mechanism enables the fusion algorithm to automatically optimize the fusion ratio of the two observations based on signal conditions, always maintaining the best estimation accuracy.

[0147] The EKF filtering fusion algorithm operates through a predictive-update recursive loop: the prediction step predicts the prior state and prior covariance at the next time step based on the system state equations from the current state. The update step corrects the prior state based on the observations to obtain a posterior state estimate. Each recursive loop completes a state update and outputs the latest current estimate.

[0148] The completed EKF filtering fusion algorithm model provides an algorithmic framework for the optimal fusion of AC and DC signals, which can make full use of the complementary advantages of the two observation signals to obtain high-precision current estimation results.

[0149] S44: Input the differential induction signal of the coil into the EKF filtering and fusion algorithm model, and achieve optimal fusion of AC and DC signals through prediction-update recursive loop. It has both wideband response and DC measurement capability, and outputs AC and DC analog current waveforms.

[0150] The differential induction signal of the Rogowski coil and the DC zero-point observation value of the miniature Hall probe are synchronously input into the EKF filtering fusion algorithm model. The optimal fusion of the AC differential signal and the DC observation signal is achieved through a prediction-update recursive loop.

[0151] At each sampling moment, the algorithm first executes a prediction step: based on the state estimate from the previous moment and the system state equation, it predicts the prior state estimate for the current moment. Simultaneously, based on the process noise covariance, it predicts the prior estimate error covariance for the current moment. This prediction step reflects the dynamic evolution of the system, utilizing historical information to predict the current state.

[0152] Next, the update step is performed: the two observed values ​​are compared with the predicted observed values, and the observation residuals are calculated. Then, the Kalman gain is calculated based on the observation noise covariance and the prior estimation error covariance. Finally, the observation residuals are weighted with the Kalman gain to correct the prior state estimate, resulting in the posterior state estimate. The update step uses the latest observation information to correct the prediction, making the state estimate closer to the true value.

[0153] The two observation signals play different roles in the fusion process: the differential signal from the Rogowski coil provides rich high-frequency dynamic information, ensuring the system's rapid response to transient currents, harmonic currents, and high-frequency interference. The DC signal from the miniature Hall probe provides a stable low-frequency and DC reference, suppressing zero-point offset caused by integral drift and temperature drift. The two signals complement each other, and through information fusion, the resulting current estimation combines wideband response with DC measurement capabilities.

[0154] The EKF fusion output is a unified AC / DC current waveform, containing both AC and DC components, accurately reflecting the true waveform of the measured current. For pure AC current, the DC component is estimated to be zero, and the waveform is consistent with the measurement results of a traditional Rogowski coil. For currents containing a DC component (such as the output current of a rectifier circuit), the DC component is accurately estimated and superimposed on the AC waveform. For pure DC current, the AC component approaches zero, resulting in a stable DC current output value.

[0155] The AC / DC analog current waveforms, after optimal EKF fusion, offer high measurement accuracy, fast response, wide bandwidth, and excellent DC characteristics, meeting the current measurement requirements under various complex operating conditions. These AC / DC analog current waveforms provide high-quality analog signal input for subsequent analog-to-digital conversion and digital compensation.

[0156] S5: The analog-to-digital conversion of the AC / DC analog current waveform is completed by a high-precision analog-to-digital converter to obtain the original digital current data. The four-dimensional physical parameters of the coil are collected, and the embedded micro residual neural network is used to identify the residual error of multi-physical field coupling. The compensation coefficient is generated to correct the original digital current data and output high-precision steady-state and transient current digital data.

[0157] S51: Uses a high-precision analog-to-digital converter to sample AC / DC analog current waveforms. The front end is equipped with an anti-aliasing low-pass filter. Synchronous sampling triggering ensures that the sampling time is precisely synchronized with the signal phase, and outputs raw digital current data.

[0158] Converting analog-to-digital current waveforms from analog to digital data requires a high-precision analog-to-digital converter (ADC). This embodiment employs a high-speed, high-precision ADC with 16 bits or more and a sampling rate of at least 10 MSPS, which meets the digitization requirements of wideband current signals.

[0159] The ADC input front-end is equipped with an anti-aliasing low-pass filter to filter out frequency components in the signal above the Nyquist frequency, preventing spectral aliasing. The anti-aliasing filter adopts a high-order Butterworth or Chebyshev structure, with a flat passband and fast stopband attenuation, effectively suppressing high-frequency noise and aliasing interference. The filter's cutoff frequency is precisely set according to the ADC sampling rate and signal bandwidth, ensuring that the useful signal passes completely while fully suppressing out-of-band interference.

[0160] The synchronous sampling trigger mechanism ensures precise synchronization between the sampling time and the signal phase. For periodic current signals (such as power frequency sine waves), the fundamental frequency and phase of the signal are extracted using phase-locked loop (PLL) technology to generate a sampling trigger signal synchronized with the signal, enabling the ADC's sampling point to be precisely aligned with a specific phase point of the signal. Synchronous sampling can reduce spectral leakage and improve the accuracy of harmonic analysis, which is particularly important for applications such as power quality monitoring and harmonic analysis.

[0161] The ADC uses a high-precision, low-temperature-drift voltage reference source to ensure the gain accuracy and stability of the analog-to-digital conversion. The ADC's analog and digital grounds are connected at a single point to prevent digital noise from coupling into the analog channel through ground loops. The analog and digital power supplies are powered separately, using ferrite beads and capacitors for decoupling and filtering to ensure power supply purity.

[0162] After high-precision analog-to-digital conversion, the continuous analog current waveform is converted into discrete digital sampling points, forming the raw digital current data. The digitized signal facilitates subsequent digital signal processing, error compensation, and data transmission.

[0163] S52: Using the original digital current data as the timing synchronization reference, it collects multi-dimensional physical parameters of the coil, including ambient temperature, skeleton deformation, winding displacement and shielding potential difference, and outputs a multi-dimensional physical parameter dataset.

[0164] The accuracy of current measurement is affected by various physical factors, such as changes in ambient temperature, mechanical deformation, and electric field interference. To compensate for errors caused by these factors, it is necessary to collect parameters reflecting these physical quantities in real time. This embodiment collects four-dimensional physical parameters of the coil, corresponding to the four dimensions of temperature, mechanical deformation, displacement, and electric field interference.

[0165] The first dimension is the ambient temperature of the coil, acquired by a miniature PT1000 platinum resistance thermometer. The PT1000 platinum resistance thermometer boasts advantages such as high accuracy, good stability, and excellent linearity, enabling it to accurately measure the ambient temperature around the coil. Temperature changes cause thermal expansion and contraction of the coil frame, variations in winding resistance, and changes in the magnetic permeability of the core, all of which affect measurement accuracy. Real-time temperature data acquisition provides a basis for subsequent temperature error compensation.

[0166] The second dimension is the radial deformation of the skeleton, which is acquired by a fiber Bragg grating (FBG) strain sensor. The FBG strain sensor is attached to the skeleton surface and can accurately measure the magnitude of the radial deformation. Skeleton deformation can be caused by factors such as temperature changes, mechanical stress, and vibration. Deformation leads to changes in the winding geometry, which in turn affects the mutual inductance of the coils. Acquiring deformation data provides a basis for mechanical deformation error compensation.

[0167] The third dimension is the axial displacement of the winding, which is acquired by a giant magnetoresistive (GMR) displacement sensor. The GMR displacement sensor measures the axial displacement of the winding relative to the frame in a non-contact manner. The winding may experience axial displacement due to vibration, impact, thermal expansion and contraction, etc. This displacement causes changes in the effective length of the coil, affecting the measurement sensitivity. Acquiring displacement data provides a basis for displacement error compensation.

[0168] The fourth dimension is the shielding layer potential difference, which is acquired by a high-resistance voltage divider detection circuit. The potential difference between the shielding layer and the signal ground reflects the intensity of external electric field interference. An excessively large potential difference will affect the measurement accuracy through parasitic capacitance coupling. Acquiring potential difference data provides a basis for electric field interference compensation.

[0169] Using the sampling clock of the raw digital current data as the timing synchronization reference, the acquisition time of each physical parameter is strictly aligned with the current sampling time, ensuring that each set of current data has corresponding physical parameter data. The multidimensional physical parameter dataset is arranged in chronological order, and each sampling time contains parameter values ​​in four dimensions: temperature, deformation, displacement, and potential difference.

[0170] S53: Using a multidimensional physical parameter dataset as feature input, an embedded micro residual neural network model is constructed by combining the original digital current data. The multi-layer residual structure is activated by the leakage rectifier linear unit, and the output residual neural network compensation model is generated.

[0171] To accurately identify measurement errors caused by multi-physics coupling, this embodiment constructs an embedded micro residual neural network model. Residual neural networks (ResNet) are a type of deep neural network architecture that addresses the training difficulties of deep networks by introducing residual connections, maintaining good training performance and generalization ability even with deep network layers.

[0172] Considering the limited computing resources of embedded platforms, this embodiment employs a miniaturized residual network structure. The network is a 3-layer residual structure, with each layer containing 8-16 neurons. It has a small number of parameters, fast computation speed, and can run in real time on embedded microcontrollers. Although the network size is small, the introduction of the residual structure makes its fitting ability far exceed that of a conventional fully connected network of the same size.

[0173] The network's input layer receives multidimensional physical parameters and raw digital current data, including parameters in four physical dimensions: temperature, deformation, displacement, and potential difference, as well as the current sample value. The input data is first normalized to map parameters of different dimensions and orders of magnitude to a similar numerical range, facilitating network processing.

[0174] The network's hidden layers employ a residual structure. The input to each layer is passed not only to the next layer through weighted connections but also directly to deeper layers through short-circuit connections. Residual connections allow the network to learn residual mappings rather than complete mappings, reducing learning difficulty and improving training efficiency and model accuracy. The activation function of the hidden layers uses Leaky Rectified Linear Units (LeakyReLU). This function remains linear in the positive interval and has a small negative slope in the negative interval, avoiding neuron death and improving the network's convergence speed and stability.

[0175] The network's output layer outputs two compensation coefficients: an amplitude compensation coefficient and a phase compensation coefficient. The amplitude compensation coefficient corrects for amplitude errors in current measurement, while the phase compensation coefficient corrects for phase errors. These two compensation coefficients correspond to the amplitude and phase errors, respectively, enabling precise compensation for both the amplitude and phase of the current simultaneously.

[0176] The network weights were obtained through offline multiphysics simulation data pre-training. First, a multiphysics coupling model of the coil was established using finite element simulation software to simulate current measurement errors under different temperature, deformation, and electric field interference conditions, generating a large amount of simulation training data. Then, this data was used to train the network, and the network weights were optimized using a backpropagation algorithm, enabling the network to accurately predict error compensation coefficients from physical parameters and current data.

[0177] The completed residual neural network compensation model is small in size, fast in computation, and high in accuracy, making it very suitable for deployment and operation on embedded platforms. It provides intelligent algorithm support for the real-time identification and compensation of multi-physics coupling errors.

[0178] S54: Synchronously inputs the multidimensional physical parameter dataset and the original digital current data into the residual neural network compensation model, identifies the multi-physical field coupling residual error in real time, generates dynamic compensation coefficients to correct the amplitude and phase, and outputs high-precision steady-state and transient current digital data.

[0179] The real-time acquired multi-dimensional physical parameter dataset and the original digital current data are synchronously input into the residual neural network compensation model. The network identifies the residual error caused by the coupling of multiple physical fields in real time and generates dynamic compensation coefficients to correct the current data.

[0180] At each sampling time, the current temperature, deformation, displacement, and potential difference (four physical parameters) along with the original current sample value are fed into the neural network input layer. After forward propagation, the network outputs the amplitude compensation coefficient and phase compensation coefficient for the current moment. The forward propagation computation is minimal, involving only a small number of matrix multiplications and activation function operations, which can be completed quickly on a microcontroller, ensuring real-time performance.

[0181] The amplitude compensation factor is a multiplicative coefficient used to correct systematic errors in the current amplitude. Multiplying the original current amplitude by the amplitude compensation factor yields the corrected current value. Amplitude errors can be caused by various factors, such as temperature-induced changes in mutual inductance and deformation-induced changes in winding geometry. Neural networks can comprehensively identify the total amplitude error from multi-dimensional parameters.

[0182] The phase compensation coefficient is an additive phase shift used to correct systematic errors in the current phase. Adding the phase compensation coefficient to the original current phase yields the phase-corrected current value. Phase errors can be caused by various factors such as winding distributed capacitance, shielding eddy currents, and phase shifts in the front-end circuit, and can also be comprehensively identified and compensated for using neural networks.

[0183] The compensation in neural networks is dynamic and real-time, with the compensation coefficients dynamically adjusted as physical parameters change. When the temperature rises, the network automatically increases the temperature-related compensation. When the deformation increases, the network automatically increases the deformation-related compensation. The coupling effects of multiphysics are also implicitly learned and compensated by the network, eliminating the need for explicitly establishing complex coupling mathematical models.

[0184] After real-time compensation of multi-physics field errors by residual neural network, the output high-precision steady-state and transient current digital data have achieved a high level of amplitude and phase accuracy, which can meet the high-precision measurement requirements under various complex environmental conditions.

[0185] S6: Based on high-precision steady-state and transient current digital data, for multi-coil densely arranged measurement scenarios, a multi-loop mutual inductance coupling matrix is ​​constructed in real time. The least squares iterative algorithm is used to remove the spatial crosstalk components of adjacent coils one by one, complete the decoupling optimization of multi-channel data, and output high-precision real-time digital current measurement values.

[0186] S61: For measurement scenarios with densely arranged multi-coil arrays, a multi-loop mutual inductance coupling matrix is ​​pre-constructed based on the spatial coordinates and geometric parameters of each coil. The mutual inductance coefficients are obtained through finite element simulation and calibration experiments to obtain initial values, and the initial mutual inductance coupling matrix is ​​output.

[0187] In measurement scenarios with multiple coils densely arranged, adjacent coils can interfere with each other through spatial electromagnetic fields, a phenomenon known as crosstalk. When multiple Rogowski coils are installed close together, each coil, in addition to inducing the magnetic field generated by its own measured current, also senses the magnetic field generated by the currents of adjacent coils. This results in crosstalk components from adjacent channels being mixed into the measurement results, affecting measurement accuracy. To address this issue, this embodiment constructs a multi-loop mutual inductance coupling matrix for crosstalk modeling and decoupling.

[0188] Based on the spatial coordinates and geometric parameters of each coil, an N×N dimensional multi-loop mutual inductance coupling matrix is ​​pre-constructed, where N is the number of coils operating simultaneously. The diagonal elements of the matrix represent the self-inductance of each coil, reflecting its own electromagnetic characteristics. The off-diagonal elements represent the mutual inductance coefficients between any two coils, reflecting the electromagnetic coupling strength between the two coils, i.e., the magnitude of crosstalk.

[0189] The initial value of the mutual inductance coefficient was obtained through a combination of finite element simulation and calibration experiments. A three-dimensional simulation model of multiple coils was established using electromagnetic simulation software. Based on parameters such as the geometric dimensions, number of turns, and relative positions of each coil, the mutual inductance coefficient between any two coils was numerically calculated to obtain the initial simulation value. Then, calibration experiments were conducted on an actual device, measuring the crosstalk in each channel by applying a known current. The simulation values ​​were then corrected and calibrated to obtain a more accurate initial mutual inductance coefficient.

[0190] The construction of the mutual inductance coupling matrix takes into account the spatial arrangement of the coils, including geometric factors such as axial distance, radial offset, and angular deviation between the coils. For regularly arranged coil arrays, symmetry can be used to reduce the computational load. For irregularly arranged coils, the mutual inductance between each pair of coils is calculated individually.

[0191] The initial mutual inductance coupling matrix reflects the spatial electromagnetic coupling relationship of the multi-coil system in its initial installation state, providing an initial benchmark for subsequent online correction and decoupling calculations. The dimension of the matrix corresponds to the number of coils; the more coils there are, the larger the matrix dimension and the higher the complexity of the decoupling calculation.

[0192] S62: Based on the initial mutual inductance coupling matrix, online correction is performed using known load current data. A sliding window forgetting factor mechanism is used to retain recent valid data, and the actual layout and environmental changes are matched in real time to output a dynamically updated mutual inductance coupling matrix.

[0193] In actual operation, the relative positions of the coils may change slowly due to vibration, thermal expansion and contraction, loose installation, etc. Environmental factors may also affect the mutual inductance coefficient, causing the initial mutual inductance coupling matrix to deviate from the actual situation. To ensure decoupling accuracy, the mutual inductance coupling matrix needs to be corrected online.

[0194] Online correction utilizes current data under known load conditions. Under certain specific conditions, the current in some channels is known or can be accurately obtained through other means. These known current values ​​can be used as calibration references to correct the coefficients of the mutual inductance coupling matrix.

[0195] For example, when the load on a certain channel is disconnected, the actual current of that channel is zero. If the measured value is not zero, then the measured value is entirely caused by crosstalk from other channels. This relationship can be used to inversely deduce the deviation of the mutual inductance coefficient and correct the matrix elements.

[0196] Online correction employs a sliding window forgetting factor mechanism. The system maintains a fixed-length data window, storing current data from the most recent sampling periods. Each time new data enters the window, the oldest data is removed, and the window maintains its fixed length. The forgetting factor determines the weight of historical data; the closer the forgetting factor is to 1, the greater the weight of historical data, resulting in smoother matrix updates. Conversely, a smaller forgetting factor gives greater weight to recent data, leading to more sensitive matrix updates.

[0197] The sliding window mechanism ensures that matrix correction is based only on recent valid data, enabling it to track slow changes in mutual inductance coefficients while avoiding interference from outdated data. A balance must be struck between tracking speed and stability; a window that is too short or a forgetting factor that is too small will result in large matrix fluctuations and poor noise resistance. Conversely, a window that is too long or a forgetting factor that is too large will lead to slow tracking speeds and an inability to reflect changes promptly. Online correction of the mutual inductance coupling matrix allows for real-time matching of the actual spatial arrangement and environmental changes, maintaining consistently high accuracy.

[0198] S63: Using the dynamically updated mutual inductance coupling matrix as the decoupling operator, a weighted recursive least squares iterative decoupling algorithm is constructed. The current data collected from each channel is used as the observation vector, and the spatial crosstalk components of adjacent coils are stripped off channel by channel, and the decoupled current data of each channel is output.

[0199]

[0200] Based on a dynamically updated mutual inductance coupling matrix, this embodiment employs a weighted recursive least squares iterative algorithm to decouple multi-channel current data. Least squares is a classic parameter estimation method that can obtain optimal parameter estimates even in the presence of observation noise.

[0201] The decoupling problem can be modeled as a system of linear equations. Let the actual current of each channel be vector I, the measured current of each channel be vector M, and the mutual inductance coupling matrix be K. Then, the measured value and the actual value satisfy the relationship: M = K·I, meaning that the measured value of each channel equals its own actual current plus the crosstalk component generated by the mutual inductance coupling of the currents of other channels. Decoupling is the process of solving for the actual current vector I given the measured vector M and the coupling matrix K.

[0202] Due to measurement noise, direct inversion may result in unstable results; therefore, the least squares method is used. The goal of the least squares solution is to minimize the sum of squared residuals between the estimated and actual measurements. Weighting coefficients are dynamically allocated based on the signal-to-noise ratio (SNR) of each channel. Channels with higher SNR have more reliable measurement data and are assigned larger weights. Channels with lower SNR have more noisy measurement data and are assigned smaller weights. This dynamic weighting mechanism ensures that the decoupling result relies more on reliable channel data and suppresses the influence of noisy channels on the overall decoupling result.

[0203] The decoupling algorithm is implemented using a recursive iterative approach. Each time new sampled data arrives, the algorithm updates the data based on the previous decoupling result, eliminating the need to recalculate all data. This high computational efficiency makes it suitable for real-time applications. The iteration convergence threshold is set to a minimum value; convergence is considered achieved when the difference between two iterations is less than the threshold, and the iteration stops. The number of iterations is typically kept to a minimum to ensure real-time performance. After stripping the spatial crosstalk components from adjacent coils channel by channel, the true current estimate for each channel is obtained.

[0204] S64: Using the current data of each decoupled channel as the verification object, crosstalk residual detection and accuracy verification are performed. When crosstalk suppression is insufficient, the mutual inductance matrix is ​​quickly reconstructed. The latest sampled data is used to accelerate convergence and output high-precision real-time digital current measurement values.

[0205] Perform residual crosstalk detection on the current data of each decoupled channel. Given that the current of a certain channel is known to be zero or its exact value, check whether the decoupled result for that channel meets expectations. If significant residual crosstalk remains after decoupling, it indicates that the accuracy of the mutual inductance coupling matrix is ​​insufficient, and reconstruction is required.

[0206] Simultaneously, the decoupling accuracy was comprehensively verified, and the crosstalk suppression ratio (RSR) of each channel was calculated. The RRS is an important indicator of decoupling effectiveness, defined as the ratio of the amount of crosstalk before decoupling to the amount of residual crosstalk after decoupling, usually expressed in decibels (dB). A higher RRS indicates better decoupling performance.

[0207] When the crosstalk suppression ratio of a certain channel is detected to be lower than a preset threshold, it indicates that the crosstalk suppression effect between this channel and other channels is insufficient, requiring the triggering of a fast mutual inductance matrix reconstruction process. Fast reconstruction differs from conventional online correction; it is an accelerated convergence mechanism that uses the latest sets of sampled data for centralized matrix updates, enabling the mutual inductance matrix to quickly approximate the true value.

[0208] The triggering conditions for rapid reconstruction include: crosstalk rejection ratio below a threshold, detection of a significant change in the relative position of the coils, and ambient temperature fluctuations exceeding a certain range. When these conditions are met, the system pauses the regular sliding window update and instead uses the latest sampled data for rapid matrix identification, completing a significant correction of the matrix in a short time.

[0209] After rapid reconstruction, the system returns to normal online correction mode and continues fine-tuning of the matrix through a sliding window mechanism. This mechanism, which combines conventional correction with rapid reconstruction, ensures the smooth stability of the matrix under normal operating conditions and enables rapid response when operating conditions change abruptly, allowing the decoupled system to quickly restore steady-state accuracy after changes in operating conditions.

[0210] Through closed-loop optimization involving crosstalk residual verification and rapid reconstruction, the multi-channel decoupling system consistently maintains optimal crosstalk suppression. It outputs high-precision real-time digital current measurements, characterized by high amplitude accuracy, low inter-channel crosstalk, fast response, and strong adaptability, meeting the high-precision current measurement requirements in scenarios with densely arranged multi-coil configurations.

[0211] In summary, this embodiment provides a high-precision Rogowski coil current measurement method based on intelligent sensors. Through the synergistic effect of six core technology modules—a high-precision induction structure with multi-winding reverse series connection and I-type compensation, wideband distortion correction of distributed passive notch filter network, anti-interference design of orthogonal microgrid shielding and floating front end, AC / DC unified measurement with dual-capacitor alternating integration and EKF fusion, multi-physics error compensation of residual neural network, and multi-channel decoupling optimization of least squares iteration—wideband, high-precision, strong anti-interference, and AC / DC unified current measurement is achieved, resulting in significantly improved measurement accuracy and a greatly expanded applicable scenarios.

[0212] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0213] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A high-precision Rogowski coil current measurement method based on intelligent sensors, characterized in that, include: S1: The circuit uses a reverse series structure with independent windings at the center angle to collect the primary current of the power circuit and outputs a native high-precision induced voltage signal. S2: Based on the high-stability original induced voltage signal, a finely matched auxiliary compensation winding is added inside the coil to construct a distributed passive notch network, which performs broadband distortion correction on the original induced signal and outputs a broadband pure induced signal. S3: Based on the wideband pure induction signal, a double-layer orthogonal microgrid is used to bidirectionally shield stray magnetic fields, carbon film damping is used to suppress eddy current resonance, and a differential floating impedance matching front end is used to output a differential analog voltage signal. S4: Based on the differential analog voltage signal, a dual-capacitor alternating integration is used to suppress leakage current and temperature drift. A miniature Hall probe collects DC zero-point observation values. The differential induction signal and DC reference are fused through an EKF extended Kalman filter to take into account both wideband AC and DC measurements, and output AC and DC analog current waveforms. S5: The AC / DC analog current waveform is synchronously sampled by a high-precision ADC to obtain the original digital current, four-dimensional physical parameters are collected, the embedded micro residual neural network is input to identify multi-physical field coupling errors, and high-precision steady-state and transient current digital data are output. S6: Based on the high-precision steady-state and transient current digital data, for the multi-coil densely arranged measurement scenario, a multi-loop mutual inductance coupling matrix is ​​constructed. The spatial crosstalk components of adjacent coils are stripped off one channel by one through the least squares iterative algorithm, and high-precision real-time digital current measurement values ​​are output.

2. The high-precision Rogowski coil current measurement method based on intelligent sensors according to claim 1, characterized in that: In step S1, the specific steps for outputting the native high-precision induced voltage signal are as follows: S11: Multiple independent windings are evenly arranged along the circumference with equal central angles, and multi-strand twisted Litz wire is wound in the same direction. Adjacent windings are connected in series with opposite ends, so that the induced electromotive force is superimposed in the same direction and the stray magnetic field is canceled in opposite directions, and the initial winding induced signal is output. S12: Based on the initial winding induction signal, I-shaped opening compensation magnetic windings are arranged at the gaps of adjacent main windings. The discontinuity integral error of the main windings is compensated by adjusting the number of turns and the opening angle, and an induced voltage signal is output. S13: According to the induced voltage signal, a gradient low-expansion composite skeleton is used as the support matrix. The thermal expansion coefficient of the three-layer material decreases in a gradient, suppressing temperature deformation interference and outputting a temperature-stable induced voltage signal. S14: Taking the temperature-stable induced voltage signal as the processing object, a micro-stress release winding process is adopted, constant tension closed-loop control is combined with segmented stress retention and release, and low-temperature annealing after winding is used to eliminate residual internal stress, and a native high-precision induced voltage signal is output.

3. The high-precision Rogowski coil current measurement method based on intelligent sensors according to claim 1, characterized in that: In step S2, the specific steps for outputting the wideband pure induction signal are as follows: S21: Multiple sets of finely matched auxiliary compensation windings are added along the circumferential direction inside the coil. Each set corresponds to a specific frequency correction band. The number of turns and wire diameter are calculated according to the target resonant frequency to form an inherent resonant circuit and output multi-band auxiliary compensation windings. S22: Using the multi-band auxiliary compensation winding as the basic unit, it is cascaded and matched with the main winding distribution parameters. The winding spacing and coupling coefficient are adjusted to control the resonance quality factor, construct a multi-level RLC notch network, and output a distributed passive notch network structure. S23: Input the original induction signal into the distributed passive notch network structure, and the notch unit at each level corrects the amplitude distortion and phase shift segment by segment. The full-band characteristics are optimized by adaptive adjustment of the notch depth, and the induction signal is output. S24: Based on the wideband corrected induction signal, perform harmonic residual detection and signal-to-noise ratio evaluation. If the correction amount is insufficient, fine-tune the number of turns connected to the auxiliary compensation winding for online correction, and output a wideband pure induction signal.

4. The high-precision Rogowski coil current measurement method based on intelligent sensors according to claim 1, characterized in that: In step S3, the specific steps for outputting the differential analog voltage signal are as follows: S31: The coil body is wrapped with an orthogonal staggered conductive microgrid shielding structure. The inner and outer copper microgrids are orthogonally arranged, with the inner layer arranged along the axial direction and the outer layer arranged along the circumferential direction. A polymer insulating layer is set in the middle, and the output double-layer orthogonal microgrid shielding structure is provided. S32: According to the double-layer orthogonal microgrid shielding structure, a nano-level carbon film damping layer is set at the grid intersection to suppress eddy current resonance. The two grids are respectively connected to the floating end through a high-resistance discharge resistor, and the output damping optimized shielding structure is achieved. S33: Based on the damping optimized shielding structure, a differentially grounded floating dynamic impedance matching front-end is constructed as the shielding carrier. An instrumentation amplifier and common-mode choke combination architecture is adopted to dynamically adjust the floating reference point potential and output the dynamic impedance matching front-end circuit. S34: The wideband pure induction signal is connected to the dynamic impedance matching front-end circuit, and the damping optimized shielding structure cancels out stray magnetic field interference in both radial and axial directions, and outputs a differential analog voltage signal.

5. The high-precision Rogowski coil current measurement method based on intelligent sensors according to claim 4, characterized in that: In step S34, the specific steps for outputting the differential analog voltage signal are as follows: Data on the radial and axial components of external industrial stray magnetic fields are collected. The induced eddy current intensity of two orthogonal microgrids in two directions is calculated, and radial eddy current cancellation sequence and axial eddy current cancellation sequence are generated. The radial eddy current cancellation sequence is matched with the shielding parameters of the outer circumferential grid. The eddy current antiphase depth is dynamically adjusted according to the magnetic field frequency and amplitude to maximize the radial stray magnetic field cancellation efficiency and output the radial shielding optimization parameters. Based on the axial eddy current cancellation sequence and the radial shielding optimization parameters, a matching calculation is performed with the shielding parameters of the inner axial mesh to adjust the axial shielding depth and output the bidirectional shielding collaborative optimization parameters. Based on the bidirectional shielding collaborative optimization parameters, the pure inductive signal optimized by radial and axial bidirectional shielding is connected to the differentially grounded floating dynamic impedance matching front end for impedance transformation and common-mode suppression, and a differential analog voltage signal is output.

6. The high-precision Rogowski coil current measurement method based on intelligent sensors according to claim 1, characterized in that: In step S4, the specific steps for outputting the AC / DC analog current waveform are as follows: S41: Construct a dual-capacitor charge cycling integral structure, with fixed frequency cycling input to the loop. Before cycling, the idle capacitor is cleared and self-calibrated to output a low-drift integral signal. S42: Based on the low-drift integral signal, a miniature Hall probe is installed at the air gap in the center of the coil. A high-sensitivity semiconductor material is used to observe the DC zero-bias component and convert it into a voltage signal, and the DC zero-point observation value is output. S43: Based on the DC zero-point observation and low-drift integral signal, construct an extended Kalman filter fusion algorithm framework. The state vector contains multi-dimensional state variables, and the noise covariance is adaptively adjusted to output an EKF filter fusion algorithm model. S44: Input the differential induction signal of the coil into the EKF filtering and fusion algorithm model, and achieve optimal fusion of AC and DC signals through prediction-update recursive loop to output AC and DC analog current waveforms.

7. The high-precision Rogowski coil current measurement method based on intelligent sensors according to claim 6, characterized in that: In step S44, the specific steps for outputting the AC / DC analog current waveform are as follows: The differential induced voltage signal of the coil is used as the input to the EKF filter as a high-frequency observation channel to provide high-frequency dynamic information on the rate of change of current and output high-frequency observation update quantity. The DC zero-point observation value of the miniature Hall probe is used as the input to the EKF filter of the low-frequency observation channel using the high-frequency observation update amount to provide DC component and slow-drift steady-state reference information, and output the low-frequency observation update amount. The high-frequency observation update and the low-frequency observation update are dual inputs. The noise covariance weights of the two observation channels are adjusted according to the current signal amplitude and signal-to-noise ratio. When the signal amplitude is high, the weight of the high-frequency channel is increased, and when the signal amplitude is low, the weight of the DC channel is increased. The adaptive fusion coefficient is output. Using the adaptive fusion coefficients as weighting parameters, the optimal state estimation of the two signals is completed through an EKF prediction-update iterative loop, and the AC / DC analog current waveform is output.

8. The high-precision Rogowski coil current measurement method based on intelligent sensors according to claim 1, characterized in that: In step S5, the specific steps for outputting high-precision steady-state and transient current digital data are as follows: S51: Uses a high-precision analog-to-digital converter to sample AC / DC analog current waveforms, with an anti-aliasing low-pass filter at the front end to output raw digital current data; S52: Acquire multi-dimensional physical parameters of the coil based on the original digital current data, including ambient temperature, skeleton deformation, winding displacement and shielding potential difference, and output multi-dimensional physical parameter dataset; S53: The multidimensional physical parameter dataset is used as feature input, and an embedded micro residual neural network model is constructed by combining the original digital current data. The multi-layer residual structure is activated with the leakage rectifier linear unit to output the residual neural network compensation model. S54: Input the multidimensional physical parameter dataset and the original digital current data into the residual neural network compensation model, identify the multi-physical field coupling residual error, generate dynamic compensation coefficients to correct the amplitude and phase, and output high-precision steady-state and transient current digital data.

9. The high-precision Rogowski coil current measurement method based on intelligent sensors according to claim 8, characterized in that: In step S54, the specific steps for outputting high-precision steady-state and transient current digital data are as follows: The multi-dimensional physical parameter dataset is input into the input layer of the residual neural network, normalized, mapped to the feature space, and the coupling features are extracted to output the multi-physics coupling feature vector. The multi-physics coupling feature vector is input into the hidden layer of the residual network. The mapping relationship between the multi-physics and measurement error is fitted through residual connections and multi-layer nonlinear transformation, and the error identification result is output. Based on the error identification results, real-time amplitude compensation coefficients and phase compensation coefficients are generated. The compensation coefficients and the original digital current data are then corrected point by point, and preliminary corrected digital current data is output. Based on the preliminary corrected digital current data, residual verification and online incremental learning are performed. When the residual exceeds the threshold, the network weight micro-update is triggered, and high-precision steady-state and transient current digital data are output.

10. The high-precision Rogowski coil current measurement method based on intelligent sensors according to claim 1, characterized in that: In step S6, the specific steps for outputting high-precision real-time digital current measurement values ​​are as follows: S61: Based on the spatial coordinates and geometric parameters of each coil, a multi-loop mutual inductance coupling matrix is ​​pre-constructed. The mutual inductance coefficients are obtained through finite element simulation and calibration experiments to obtain initial values, and the initial mutual inductance coupling matrix is ​​output. S62: Based on the initial mutual inductance coupling matrix, online correction is performed using known load current data. A sliding window forgetting factor mechanism is used to retain recent valid data, and the actual layout and environmental changes are matched in real time to output a dynamically updated mutual inductance coupling matrix. S63: Using the dynamically updated mutual inductance coupling matrix as the decoupling operator, construct a weighted recursive least squares iterative decoupling algorithm, use the current data collected from each channel as the observation vector, strip away the spatial crosstalk components of adjacent coils channel by channel, and output the channel current data. S64: Based on the channel current data, perform crosstalk residual detection and accuracy verification, accelerate convergence using the latest sampling data, and output high-precision real-time digital current measurement values.

Images