High-frequency transformer litz wire loss prediction method

Abstract

The invention discloses a high-frequency transformer litz wire loss prediction method, and relates to the technical field of circuits, and the method comprises the steps: collecting an electromagnetic physical quantity, a structural physical quantity and a thermal physical quantity of a high-frequency transformer, so as to generate a multi-physical-quantity original feature set; carrying out litz wire equivalent copper foil conversion based on the multi-physical-quantity original feature set to obtain litz wire equivalent structure parameters; based on the litz wire equivalent structure parameters, determining three-dimensional magnetic field distribution of the litz wire, and decomposing the three-dimensional magnetic field distribution into a magnetic field component parallel to the strand and a magnetic field component perpendicular to the strand in combination with the litz wire torsion angle so as to generate a magnetic field component representation result; on the basis of the equivalent structure parameters of the Litz wire and the corresponding current harmonic characteristics, calculating the unit length loss of the Litz wire according to the magnetic field component characterization result so as to determine a basic loss value; and superposing the harmonic loss values to the basic loss value through the mutual impedance equivalent circuit of the litz wire to obtain the predicted loss value of the litz wire, thereby improving the loss prediction precision.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of circuit, in particular to a high-frequency transformer Litz wire loss prediction method. BACKGROUND

[0002] In the field of high-frequency power conversion such as new energy vehicles and photovoltaic power generation, the high-frequency transformer as the core component of energy conversion directly affects the performance of the whole system. Litz wire is widely used in high-frequency transformer winding because it can effectively suppress the skin effect and proximity effect at high frequency. However, as the working frequency increases (usually up to tens of kHz to hundreds of kHz), the loss characteristics of Litz wire become more complex, and accurate prediction of its loss value is of great significance for the optimization design, efficiency improvement and thermal management of the transformer.

[0003] At present, the industry mostly uses simplified models to predict the loss of Litz wire, such as empirical formulas based on classical electromagnetic theory, which estimates the loss by inputting basic parameters such as the number of Litz wire strands, single-strand diameter, working frequency, and combining skin depth and proximity effect coefficient. This method usually assumes that Litz wire strands are uniformly arranged and the effect of twisting structure on loss can be ignored, and the calculation process is relatively simple, suitable for rapid evaluation in the preliminary design stage.

[0004] However, the above simplified model has obvious limitations: as it does not fully consider the spatial arrangement differences of strands in the actual twisting process of Litz wire (such as twisting pitch fluctuations, strand twisting angle changes), microstructure characteristics (such as strand contact resistance, insulation layer distribution), and the coupling effect of multiple harmonic components at high frequency, the prediction error of its loss in a wide frequency range or complex working conditions is large, often unable to meet the high-precision design requirements, especially in the high-frequency band (such as above 100 kHz), the deviation between the predicted value and the actual measured value can be more than 15%. SUMMARY

[0005] To solve the above technical problems, the present application provides a high-frequency transformer Litz wire loss prediction method to at least alleviate the above technical problems.

[0006] The technical scheme provided by the embodiments of the present application is as follows:

[0007] A high-frequency transformer Litz wire loss prediction method, the method comprising:

[0008] Step 1: Collecting electromagnetic physical quantities, structural physical quantities and thermal physical quantities of the high-frequency transformer to generate a multi-physical quantity original feature set;

[0009] Step 2: Litz wire equivalent copper foil conversion based on the multi-physical quantity original feature set to obtain Litz wire equivalent structure parameters, including equivalent thickness, equivalent conductivity and filling factor;

[0010] Step 3: Based on the equivalent structure parameters of the Litz wire, determine the three-dimensional magnetic field distribution of the Litz wire, and combine the twist angle of the Litz wire to decompose the three-dimensional magnetic field distribution into the magnetic field component parallel to the strand and the magnetic field component perpendicular to the strand, to generate the magnetic field component characterization result;

[0011] Step 4: Based on the equivalent structure parameters of the Litz wire and its corresponding current harmonic characteristics, calculate the unit length loss of the Litz wire according to the magnetic field component characterization result to determine the basic loss value;

[0012] Step 5: Superimpose each harmonic loss value on the basic loss value through the mutual impedance equivalent circuit of the Litz wire to obtain the predicted Litz wire loss value.

[0013] In the technical solution of the present application, the multi-physical quantity original feature set is generated by collecting electromagnetic physical quantities, structural physical quantities and thermal physical quantities, which can comprehensively obtain the microstructure characteristics (such as strand contact resistance, insulating layer distribution) and macro operation parameters of the Litz wire, solving the problem of insufficient loss influencing factor capture caused by single input parameter and lack of multi-physical quantity information in the existing simplified model, and laying a data foundation for subsequent accurate analysis. Based on the multi-physical quantity, the equivalent copper foil conversion of the Litz wire is carried out and the equivalent structure parameters (equivalent thickness, equivalent conductivity, filling factor) are obtained, which can convert the complex Litz wire stranded structure into quantifiable equivalent parameters, effectively reflect the influence of strand space arrangement difference (such as stranded pitch fluctuation) on structural characteristics, and solve the problem of inaccurate structural characteristic representation caused by ignoring the stranded structure details in the existing model. The three-dimensional magnetic field distribution is decomposed into the magnetic field component parallel to the strand by combining the twist angle of the Litz wire, which can accurately capture the influence of the change of the strand twist angle on the magnetic field distribution, solve the problem of magnetic field characteristic analysis distortion caused by assuming uniform arrangement of the strand and ignoring the space arrangement difference in the existing model, and make the magnetic field component representation more in line with the actual situation. Based on the equivalent structure parameters, current harmonic characteristics and magnetic field components, the basic loss value is calculated, which can fully take into account the influence of current harmonic characteristics on loss, solve the one-sidedness problem of loss calculation caused by insufficient consideration of high-frequency multi-harmonic components in the existing model, and improve the accuracy of the basic loss value. By superimposing each harmonic loss on the basic loss value through the mutual impedance equivalent circuit, the coupling effect between multi-harmonic components can be effectively reflected, solving the problem of large prediction error (>15%) of high-frequency band loss caused by ignoring harmonic coupling in the existing model, and finally improving the overall precision of Litz wire loss prediction, meeting the high-precision design requirement. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 The figure is a flowchart of the high-frequency transformer Litz wire loss prediction method of the embodiment of the present application. DETAILED DESCRIPTION

[0015] As Figure 1As shown, the embodiment of the present application provides a high-frequency transformer Litz wire loss prediction method, the method comprises:

[0016] Step 1: Collect electromagnetic physical quantities, structural physical quantities and thermal physical quantities of the high-frequency transformer to generate a multi-physical quantity original feature set;

[0017] Step 2: Perform Litz wire equivalent copper foil conversion based on the multi-physical quantity original feature set to obtain Litz wire equivalent structural parameters, including equivalent thickness, equivalent conductivity and filling factor;

[0018] Step 3: Based on the Litz wire equivalent structural parameters, determine the three-dimensional magnetic field distribution of the Litz wire, and combine the Litz wire twist angle to decompose the three-dimensional magnetic field distribution into a magnetic field component parallel to the strand and a magnetic field component perpendicular to the strand to generate a magnetic field component representation result;

[0019] Step 4: Based on the equivalent structural parameters of the Litz wire and its corresponding current harmonic characteristics, calculate the unit length loss of the Litz wire according to the magnetic field component representation result to determine the basic loss value;

[0020] Step 5: Superimpose each harmonic loss value on the basic loss value through the mutual impedance equivalent circuit of the Litz wire to obtain the predicted Litz wire loss value.

[0021] Optionally, the step 1 specifically comprises:

[0022] Step 11: According to the influencing factors of the high-frequency transformer Litz wire loss, determine the specific range of the electromagnetic physical quantities, structural physical quantities and thermal physical quantities to be collected to generate a physical quantity collection list;

[0023] Step 12: According to the physical quantity collection list, use multi-sensor synchronous collection technology to collect data of each physical quantity to generate a multi-dimensional original data set;

[0024] Step 13: Standardize the multi-dimensional original data set to generate a multi-physical quantity original feature set.

[0025] Preferably, for step 11, the data acquisition range is determined by a multi-dimensional physical quantity mapping model, the core of which is to build a knowledge graph of the loss influencing factors of the high-frequency transformer's Litz wire. First, a loss physical model is established based on electromagnetic field theory to quantify the contribution weights of electromagnetic physical quantities (such as winding current density distribution, magnetic field intensity vector), structural physical quantities (such as Litz wire strand diameter, stranding pitch), and thermal physical quantities (such as winding temperature field distribution) to the loss. The relative importance of each physical quantity is calculated using the analytic hierarchy process (AHP), and a weight threshold (such as 0.05) is set to screen the key parameters. Combined with the grey relational analysis algorithm, the correlation (correlation degree ≥ 0.6) between the physical quantity and the loss is verified, and finally a collection list containing 23 core physical quantities is generated. This list not only covers traditional electromagnetic parameters, but also includes microscopic structural parameters such as stranding pitch fluctuation rate and strand contact resistance distribution.

[0026] The above grey relational analysis (GRA) quantifies the dynamic correlation between each physical quantity and the Litz wire loss, which is particularly suitable for small sample, nonlinear data analysis scenarios of high-frequency transformers, and the technical implementation steps are as follows:

[0027] The reference sequence is defined as X0=(χ0(1),χ0(2),…,χ0(n)), where X represents the data sequence, and subscript 0 indicates that the sequence is the "baseline reference"; χ0(k) represents the total loss value of the Litz wire of the high-frequency transformer measured in the kth group of historical data; n represents the total number of historical sampling points (usually ≥ 30 groups, covering different load, frequency, etc. Working conditions).

[0028] Define the comparison sequence X i ={χ i (1),χ i (2),…,χ i (n)}( i=1,2, …,m), where subscript i represents the ith physical quantity (such as current density, stranding pitch, etc.); χ i (k) represents the measured value of the ith physical quantity in the kth group of historical data; m is the total number of physical quantities to be analyzed.

[0029] All sequences are initialized, i.e. χ i (k)'=χ i (k) / χ i (1), where χ i (k)' is the dimensionless result of the ith sequence at the kth point, which eliminates the dimension (such as temperature in ℃, current in A, etc.) by comparing each data point with the first data point of the sequence, so that the historical data of different physical quantities are comparable.

[0030] For each historical sampling point, calculate the absolute difference between the reference sequence and the comparison sequence, Δ i (k)'=|χ0(k)'χ i (k)'|, where Δ i (k) represents the dimensionless difference between the ith physical quantity and the loss value in the kth set of historical data. The smaller the difference, the closer the correlation between the two under the same working condition.

[0031] Calculate the maximum maxmaxΔ i (k) and the minimum minminΔ i (k) among all absolute differences, where "maxmax" represents the maximum absolute difference among all sampling points (k) of all physical quantities (i); "minmin" represents the minimum absolute difference, which is used for subsequent normalization processing.

[0032] Introduce a resolution coefficient (typically 0.5, used to adjust the sensitivity of the correlation degree), and the correlation coefficient formula is: , where ξ i (k) represents the correlation coefficient between the ith physical quantity and the loss value in the kth set of historical data, with a value range of [0, 1]: the closer the value to 1, the more significant the correlation between the physical quantity and the loss under the same working condition.

[0033] Take the average of the correlation coefficients of each comparison sequence to obtain the global correlation degree , where represents the overall correlation degree between the ith physical quantity and the loss value under all historical working conditions, which is a comprehensive evaluation of the single-point correlation degree.

[0034] Arrange all physical quantities in descending order of global correlation degree to form ≥ ≥… , Where the sorting result directly reflects the importance of each physical quantity on the loss: The larger the value, the stronger the overall correlation between the physical quantity and the loss.

[0035] Set the correlation degree threshold value g (0.6 in this step), and keep ≥ the physical quantities with a correlation degree greater than g as the core parameters, where g is a critical value based on engineering experience to ensure that the selected parameters have significant statistical correlation with the loss.

[0036] The consistency of the importance ranking of physical quantities by AHP and GRA is verified by Kendall's coordination coefficient (≥ 0.7) to avoid the deviation of a single method. The mutual information (≥ 0.8) between the reserved parameters is calculated to determine the information overlap, and redundant parameters are removed to simplify the model.

[0037] The influence of parameter fluctuation on loss prediction is verified by Monte Carlo simulation to finally determine 23 core physical quantities, ensuring their sensitivity and independence to loss.

[0038] Therefore, the present application takes the historical loss value as the benchmark. GRA can capture the long-term correlation between physical quantities and loss from accumulated operation data, rather than the accidental correspondence under a single operating condition, improving the reliability of parameter selection. In view of the characteristics of high-frequency transformer loss, such as high cost of actual measurement, limited sample size (usually ≤ 50 groups), GRA can effectively mine the implicit correlation without a large amount of data, and break away from the dependence of traditional regression analysis on large samples. Through trend comparison of historical data sequences, GRA can accurately capture the nonlinear coupling relationship between physical quantities and loss (such as the nonlinear correlation between the fluctuation of twisted pitch and proximity effect loss), making up for the shortcomings of linear analysis.

[0039] Preferably, for step 12, a multi-sensor fusion network is used to realize synchronous acquisition of physical quantities. A distributed sensor array is deployed at the hardware layer: a Rogowski coil sensor (bandwidth 10 MHz) is used to acquire current harmonics, a Hall array sensor (response time < 1 μs) is used to obtain three-dimensional magnetic field distribution, and a fiber Bragg grating sensor (temperature measurement accuracy ± 0.1℃) is used to monitor the temperature field. A distributed data acquisition system based on timestamp synchronization is developed at the software layer, and IEEE 1588 precision clock protocol is used to achieve sub-microsecond synchronization accuracy. Industrial Ethernet (transmission rate 1 Gbps) combined with edge computing technology is used at the data transmission layer to realize real-time data preprocessing (such as digital filtering, sampling rate matching). The final generated multi-dimensional raw data set contains high-density data point cloud with time domain resolution 100 ns and spatial resolution 1 mm, which completely retains the transient physical characteristics of high-frequency transformer during operation.

[0040] Specifically, the Rogowski coil sensor is installed on the lead-out wire of the primary and secondary windings of the high-frequency transformer (near the winding terminal) and is co-linear with the winding axis. This position can directly couple the transient current of the winding, avoid magnetic field interference caused by lead bending, and ensure that the collected current harmonic data can truly reflect the current changes inside the winding, especially for capturing fast current transients at high frequencies (10 kHz-1 MHz). The Hall array sensor is embedded in the magnetic core window in a matrix form (the gap between the winding and the magnetic core column), the plane of the sensor array is parallel to the winding layer, and is uniformly distributed along the winding axis and radial direction (spacing 1 mm). This position can directly detect the three-dimensional magnetic field distribution in the magnetic core window, especially the diffuse magnetic flux near the distributed air gap and the transverse magnetic field component at the end of the winding, solving the problem that traditional single-point measurement cannot reflect the spatial gradient of the magnetic field. The fiber Bragg grating sensor is installed in an embedded manner, with the optical fiber buried in the Litz wire bundle along the winding axis (one sensing point is arranged every 5 turns of the winding), and the optical fiber path is parallel to the strand twisting direction. This position can directly contact the conductor, accurately monitor the working temperature of the winding (not affected by electromagnetic interference), and at the same time avoid introducing additional eddy current loss due to the presence of metal components in the sensor itself, ensuring that the temperature data can reflect the true heating state of the Litz wire.

[0041] Preferably, for step 13, data standardization is realized by adaptive data cleaning and feature enhancement algorithm, solving the consistency problem of multi-source heterogeneous data. First, the Isolation Forest algorithm is used for outlier detection, and the pollution rate parameter is set to 0.15 to automatically identify and exclude error data. Then, the Z-score standardization method is applied to dimensionless processing of each physical quantity, ensuring that the data distribution conforms to the standard normal distribution. For the nonlinear characteristics of some physical quantities, Box-Cox transformation is used for data conversion to improve the linear expression ability of the features. To enhance the generalization ability of the model, the Monte Carlo dropout technology is introduced to randomly disturb the standardized data (the disturbance amplitude is controlled within 5%). The final generated multi-physical quantity original feature set has a unified data scale (mean 0, standard deviation 1) and good linear separability, providing high-quality input for subsequent machine learning modeling.

[0042] Optionally, the step 2 specifically comprises:

[0043] Step 21: Based on the Litz wire structure parameters in the multi-physical quantity original feature set, the geometric equivalence and physical property conservation principles of converting the Litz wire to the equivalent copper foil are determined to generate the equivalent conversion criteria;

[0044] Step 22: According to the equivalent conversion criteria, the thickness, electrical conductivity and filling rate of the Litz wire are calculated to generate the initial values of the equivalent structure parameters;

[0045] Step 23: Perform the twisting effect correction on the initial equivalent structure parameters to generate the Litz wire equivalent structure parameters.

[0046] Optionally, the step 21 specifically includes:

[0047] Step 211: Based on the Litz wire single strand diameter, total strand count and twisting pitch in the multi-physical quantity original feature set, extract the core structure parameters affecting the equivalent conversion to generate the Litz wire structure feature parameter set;

[0048] Step 212: According to the Litz wire structure feature parameter set, analyze the space occupation law when multiple strands are twisted to generate a geometric equivalent sub-criterion that satisfies the principle of equal conductive cross-sectional area of equivalent copper foil and Litz wire;

[0049] Step 213: Based on the geometric equivalent sub-criterion, combined with the physical constraint of constant direct current resistance, generate the equivalent conversion criterion.

[0050] Preferably, in step 211, from the perspective of "multi-scale structure-electromagnetic property mapping", the core parameters that have a significant impact on equivalent conversion are screened:

[0051] Basic geometric parameters: single strand diameter d (need to distinguish the deviation between actual diameter and nominal diameter of the strand, usually take the average of 3 times laser measurement), total strand count N, twisting pitch P (measure the distance between the same position of adjacent two strands along the axial direction, take the average of 5 consecutive pitches to eliminate local distortion error).

[0052] Derived structure parameters: strand space arrangement angle θ (the angle between the axis of a single strand and the overall axis of the Litz wire, calculated from the twisting pitch P and the outer diameter D of the Litz wire, reflecting the degree of inclination of the twisting); single strand diameter deviation rate (statistics of the deviation coefficient of all strand diameters from the average diameter, quantifying the influence of strand manufacturing tolerance on equivalence); twisting tightness coefficient (measure the ratio of radial pressure to theoretical tightness during twisting by pressure sensor, reflecting the degree of extrusion between strands).

[0053] These parameters together constitute the "Litz wire structure feature parameter set", covering multi-dimensional information from macro size to micro tolerance, providing a physical basis for the establishment of subsequent equivalent criteria.

[0054] Optionally, in step 212, considering that the core of geometric equivalence is to ensure the consistency of equivalent copper foil and Litz wire in space occupation and electromagnetic response, rather than simply "area equivalence", based on the structure feature parameter set, three sub-criteria of total conductive cross-sectional area conservation, axial projection area equivalence and radial equivalent diameter matching need to be met to cope with the directional differences of skin effect and proximity effect at high frequency:

[0055] 1. Total conductive area conservation

[0056] The total conductive area of the litz wire needs to be equal to the effective conductive area of the equivalent copper foil. Considering that the strands are arranged in a close-packed hexagonal pattern (the closest to the no-gap state in actual twisting), the total conductive area is the product of the cross-sectional area of a single strand and the effective number of strands, and the formula is as follows:

[0057]

[0058] Wherein:

[0059] S 总 is the total conductive cross-sectional area of the litz wire (unit: mm 2 );

[0060] N 有效 is the actual number of conducting strands after removing broken strands (determined by the conduction test, ≤ total number of strands N);

[0061] d is the diameter of a single strand (unit: mm);

[0062] δ d is the single strand diameter deviation rate (dimensionless, reflecting the unevenness of the strand diameter, δ d ∈ [0, 0.1].

[0063] The conductive area of the equivalent copper foil needs to be equal to this, that is, S 铜箔= W·h = S 总 , where W is the width of the copper foil (unit: mm), and h is the thickness of the copper foil (unit: mm).

[0064] In the above formula, the litz wire strands are usually arranged in a close-packed hexagonal pattern (the highest space utilization arrangement), and this coefficient comes from the geometric area calculation of the regular hexagon - the area of a regular hexagon with side length d is , and the center distance of a single strand is equal to the wire diameter d, so the "effective occupied area" of a single strand in a regular hexagonal arrangement needs to be quantified, which comes from plane geometry derivation. A regular hexagon is composed of 6 equilateral triangles, and the area of each triangle is , and the total area is , which is a basic geometric constant that describes the close packing of regular hexagons, with no empirical error.

[0065] The single strand diameter deviation rate δ d∈ [0, 0.1] reflects the diameter non-uniformity caused by the manufacturing tolerance of the wire, defined as "the maximum deviation of single wire diameter from the average diameter / average diameter", used to correct the deviation of the total conductive area calculation caused by the inconsistency of the wire size. The diameter tolerance of industrial-grade copper wire (Litz wire raw material) usually follows the IEC 60228 or ASTM B355 standard, and the diameter deviation of Class 2 copper wire is allowed to be ±10% (i.e. the maximum deviation rate is 0.1). In actual production, due to factors such as wire drawing die wear and material uniformity, the deviation rate rarely exceeds 0.1, so the value range is limited to [0, 0.1], which not only meets the manufacturing standards, but also covers the actual possible deviation.

[0066] 2. Axial projection area equivalence

[0067] Under high frequency, the current is distributed along the surface of the conductor, and the twisting of the Litz wire will cause its axial projection area to be larger than that in the straight state, which needs to be corrected by the spatial arrangement angle: S 投影= S 总 · (1 / cosθ) · k t， Where: S 投影 is the effective projection area of the Litz wire along the axial direction (unit: mm 2 );

[0068] θ is the spatial arrangement angle of the wire (unit: radian, θ = arctan(P / (πD)), k is the tightness coefficient of twisting (dimensionless, The higher the tightness, the closer the projection area to the theoretical value.

[0069] The axial projection area of the equivalent copper foil needs to meet S 铜箔投影= W·h·(1+αf)-S 投影 , where α is the frequency correction coefficient, f is the working frequency (unit: Hz), to compensate for the deviation of the projection area equivalence caused by the skin effect of the current under high frequency.

[0070] Twisting tightness coefficient k t ∈ [0.85, 1] quantifies the tightness of the wire when twisted, defined as "actual twisting pressure / theoretical maximum tightness pressure", the higher the tightness, the smaller the gap between the wires, the higher the space utilization rate.

[0071] When the ideal twisting (no gap, no elastic deformation), k t=1, but in practice the strands have an insulation layer (thickness 5-20 pm), and the strands will produce a small elastic deformation during the twisting process due to tension, resulting in a decrease in actual tightness; through the actual measurement of 10 different specifications of litz wire (100-500 strands, insulation layer thickness 5-20 pm), the twisting tightness is all above 0.85 (the thinner the insulation layer and the more the number of strands, the higher the tightness), so the value range is [0.85, 1], covering most industrial scenarios.

[0072] Frequency correction coefficient Physical meaning: correct the change of “equivalent projection area” caused by skin effect at high frequency. When the frequency is high, the current is concentrated on the surface of the conductor, which is equivalent to the “effective conductive projection area” decreasing with the increase of frequency, which needs to be compensated by the coefficient. Skin depth is the resistivity, μ0 is the vacuum permeability, μ г is the relative permeability), in the range of 10 kHz-1 MHz, the skin depth of copper decreases from 0.2 mm to 0.02 mm, and the surface current density concentrates with the increase of frequency; through finite element simulation to calculate the ratio of “effective projection area” at different frequencies to that at direct current, the fitting result is , that is, for every 1 kHz increase in frequency, the equivalent projection area needs to be corrected by about 0.1%, which is consistent with the square law characteristic of skin effect.

[0073] 3. Radial equivalent diameter matching

[0074] To ensure the consistency of the radial size of the equivalent copper foil and the litz wire in the magnetic field distribution, the equivalent diameter of the copper foil after winding needs to be matched with the outer diameter of the litz wire:

[0075]

[0076] where D 铜箔等效 is the equivalent diameter of the copper foil after winding (unit: mm), D is the outer diameter of the litz wire (unit: mm), and the coefficient 0.5k t is used to correct the influence of the twisting tightness on the radial size (the higher the tightness, the closer the equivalent diameter to the actual outer diameter).

[0077] Preferably, in step 213, on the basis of geometric equivalence, the structure parameters are associated with electromagnetic characteristics through the physical constraint of “constant direct current resistance” to form a complete equivalent conversion criterion. The direct current resistance of the litz wire is not only related to the total conductive area, but also affected by factors such as the increase in the length of the strands after twisting and the contact resistance between the strands, which needs to be corrected through multi-parameter coupling:

[0078] 1. Litz wire direct current resistance calculation

[0079] Considering the actual length increase of the strands after twisting (from the arrangement angle ​The total DC resistance of the Litz wire is calculated as: R 利兹线 = p · (L0 / cos0 / S 总 ) · (1+k c ·N 有效 ), where p is the resistivity of copper (unit: Ω·mm, taken as 1.72x10 -5 Ω·mm); L0 is the axial length of the Litz wire (unit: mm); 0 is the spatial arrangement angle of the strands (unit: radian, reflecting the length increase multiple caused by stranding 1 / cos0); k c is the contact resistance coefficient between strands (unit: Ω -1 , determined by contact pressure test, k c ∈ [0.01, 0.05], the higher the tightness, the smaller k c ).

[0080] The contact resistance coefficient between strands k c ∈ [0.01, 0.05] Ω -1 quantifies the contribution of the contact resistance between strands to the total resistance, and the contact resistance decreases as the stranding tightness increases (the increase in pressure increases the actual contact area), and the coefficient reflects the coupling relationship between the contact resistance and the effective number of strands. The contact resistance R c is inversely proportional to the contact pressure F (R c ∝1 / F), while the stranding tightness coefficient k t is positively related to the pressure (k t ∝F), so k c decreases as k t increases; experimental measurement of contact resistance under different stranding tension (measurement of strand-to-strand resistance by four-terminal method) found that when k t =0.85 (low tightness), k c ≈0.05 Ω -1 ; when k t =1 (high tightness), k c ≈0.01 Ω -1 , so the value range is [0.01, 0.05], which is consistent with the measured data.

[0081] 2. Equivalent copper foil DC resistance matching

[0082] The DC resistance of the equivalent copper foil needs to be equal to that of the Litz wire, i.e.: R 铝箔 = p · (L0 / W,h,η)=R 利兹线， , where η is the copper foil conductivity efficiency coefficient (dimensionless, , considering the effect of the copper foil surface oxide layer on conductivity).

[0083] The copper foil conductive efficiency coefficient η ∈ [0.95, 1], physical meaning: the influence of the surface oxide layer of the copper foil on the conductive capacity is corrected, the resistivity of the oxide layer (such as Cu or Cu20) is much higher than that of pure copper (about 100-1000 times), which will cause the actual conductive efficiency to be slightly lower than the ideal value. The thickness of the natural oxide layer on the surface of the copper foil is usually 5-20 nm (after being exposed to air for 24 hours), the thickness of the oxide layer is measured by an ellipsometer, and the contribution of the oxide layer to the total resistance is calculated by combining the sheet resistance formula (Rs=ρ oχ / t oχ ) shows that the conductive efficiency loss caused by the oxide layer is about 1%-5%, which is consistent with the statistical distribution of the thickness of the oxide layer (the thickness of the oxide layer of most copper foils is ≤10 nm, and the efficiency loss is ≤3%).

[0084] 3. Final form of the equivalent conversion criterion

[0085] By combining the geometric equivalent sub-criteria (area, projection, diameter) and the conservation of physical properties (direct current resistance), the complete equivalent conversion criterion is obtained:

[0086]

[0087] The criterion associates the structural parameters (d, θ, k t , etc.) of the Litz wire with the size (W, h) of the equivalent copper foil through multi-equation coupling, ensuring both the geometric equivalence and the consistency of electromagnetic properties, and solving the problem of high-frequency loss prediction deviation caused by traditional equivalent methods considering only a single dimension (such as area).

[0088] Optionally, the step 22 specifically comprises:

[0089] Step 221: according to the geometric equivalent sub-criterion in the equivalent conversion criterion, the product of the total number of Litz wires and the diameter of a single wire is calculated and combined with the hexagonal arrangement coefficient to calculate the initial thickness of the equivalent copper foil, thereby generating the initial equivalent thickness;

[0090] Step 222: according to the conservation principle of physical properties in the equivalent conversion criterion, the bulk conductivity of copper is associated with the number of Litz wires to calculate the initial conductivity considering the influence of wire insulation, thereby generating the initial equivalent conductivity;

[0091] Step 223: according to the area ratio principle in the equivalent conversion criterion, the ratio of the total conductive area of the Litz wire to the cross-sectional area of the wire bundle is calculated to calculate the initial filling rate without considering the twisting gap, thereby generating the initial equivalent structural parameter.

[0092] Preferably, steps 221-223 generate the initial equivalent structural parameter (h eq,0, σ eq,η0) not only satisfies the basic constraint of the equivalent conversion criterion, but also retains the key characteristics of the Litz wire twisted structure. Compared with the traditional simplified model, the introduced correction coefficients (such as k p ,θ,k σ ) are all from the actual measurement analysis of the Litz wire structure, which upgrades the parameter calculation from "empirical estimation" to "physical quantification", and lays a high-precision foundation for subsequent twisted effect correction. The specific implementation of each step is as follows:

[0093] Preferably, in step 221: calculation of the initial value of the equivalent thickness (based on the geometric equivalent sub-criterion), the equivalent thickness is the core geometric parameter of the conversion of the Litz wire to the copper foil, and its calculation needs to accurately reflect the spatial arrangement characteristics of the twisted multi-stranded wire, rather than simply area equivalent. This step is different from the traditional "single strand area superposition" idea, and introduces a spatial utilization rate correction of regular hexagonal close packing, and the formula is as follows:

[0094]

[0095] In the above formula, h eq,0 : initial value of equivalent thickness (unit: mm), which is the copper foil thickness without considering the twisted effect; N: total number of Litz wire strands (dimensionless), which needs to exclude broken or virtual connected strands, and take the actual conducting strand number; d: single strand diameter (unit: mm), which is the average value measured by the laser micrometer along the three orthogonal directions of the strand circumference; : regular hexagonal arrangement correction coefficient (dimensionless, value 0.90-0.95), which reflects the spatial utilization rate loss caused by the small gap between the strands in the actual twisting (ideal close packing k p =1); W foil : equivalent copper foil width (unit: mm), which is equal to the actual outer diameter of the Litz wire bundle (obtained by three-dimensional scanning); θ: strand twisting angle (unit: rad), which is calculated from the twisting pitch P and the bundle radius r (θ=arctan(2πr / p)) and is used to correct the change of the axial projection area of the strand caused by the twisting.

[0096] 1. Regular hexagonal area coefficient (Step 221) Physical meaning: This constant is derived from the geometric characteristics of regular hexagonal close packing. The Litz wire strands are usually arranged in a regular hexagonal shape (with the highest space utilization), and the area calculation formula of the regular hexagonal shape with side length d is Therefore, in the calculation of the equivalent thickness, the "effective space occupation area" of the single strand in the regular hexagonal arrangement needs to be quantified by this coefficient.

[0097] From plane geometry derivation - a regular hexagon can be decomposed into 6 equilateral triangles, and the area of each triangle is , and the total area is , is the basic geometric constant describing the regular hexagonal space arrangement, with no experience error, ensuring the theoretical rigor of geometric equivalence.

[0098] 2. Regular hexagonal arrangement correction coefficient k p ∈[0.90, 0.95] Step 221) Physical meaning: Correct the deviation that the strands cannot reach ideal dense packing in actual twisting, reflect the influence of small gaps between strands (caused by insulation layer, uneven twisting tension) on space utilization (k p =1 for ideal no-gap state).

[0099] When ideal dense packing, there is no gap between strands, but the actual litz wire strands have an insulation layer (thickness 5-20 μm), and the strands are subjected to tension during twisting, which causes small misalignment, resulting in reduced space utilization; three-dimensional scanning is performed on 10 typical specifications of litz wire (strand count 50-500 strands, insulation layer thickness 5-20 μm), and it is found that the actual space utilization is 5%-10% lower than the ideal value, so the value range is [0.90, 0.95], which is consistent with the measured data.

[0100] The present application introduces the twisting angle θ and the arrangement correction coefficient k p , the formula not only considers the influence of strand spiral arrangement on axial space occupation, but also corrects the deviation that the actual twisting cannot reach ideal dense packing, so that the initial value of equivalent thickness is more consistent with the real geometric characteristics of litz wire.

[0101] Preferably, step 222: calculation of initial value of equivalent conductivity (based on the principle of conservation of physical properties), the equivalent conductivity needs to reflect the change of the conductive capacity of litz wire due to strand insulation and twisting, rather than simply using the bulk conductivity of copper. The purpose of this step is to quantify the "distributed impedance" effect of the insulation layer between strands on current conduction, and the specific formula used is as follows:

[0102] wherein, : initial value of equivalent conductivity (unit: S / m), reflecting the overall conductive capacity of litz wire; : bulk conductivity of copper (unit: S / m, value at 20°C ); : effective conductive strand count (dimensionless), determined by high-frequency conduction test (some strands may have almost no current due to skin effect); : insulation layer thickness of strand (unit: μm), measured by microscope to measure the average thickness of insulation layer of strand cross section; : relative dielectric constant of insulating material (dimensionless, such as 3.5 for polyimide insulation), reflecting the hindering effect of insulation layer on electric field; : twisting density coefficient (dimensionless, value 1.2-1.5), the more the number of strands, the tighter the twisting, The greater, for the correction of the thickness of the insulating layer changes caused by the extrusion between the strands.

[0103] 3. Bulk conductivity of copper , meaning: the standard value of the electrical conductivity of pure copper at 20℃, which is the reference for the conductive capacity of Litz wire (the material of Litz wire strands is high-purity electrolytic copper, purity ≥ 99.9%).

[0104] In line with the provisions of international standards (such as IEC 60287) for the electrical conductivity of pure copper, the theoretical value at 20℃ is 5.8×10 7 S / m; the actual measurement of the electrical conductivity of industrial-grade high-purity copper wire (Litz wire raw material) has a deviation of ≤2%, so the standard value is taken as the reference to ensure the consistency of the reference of the conservation of physical properties.

[0105] 4. Relative dielectric constant of polyimide =3.5 (step 222, taking common insulating materials as an example) Physical meaning: quantifying the ability of the insulating layer to resist the electric field (the greater the dielectric constant, the stronger the "buffering effect" of the insulating layer on the electric field, and the more difficult it is for the electric current to penetrate), used to correct the attenuation effect of the insulating layer on the overall conductive capacity of the Litz wire.

[0106] Polyimide is a commonly used insulating material for Litz wire (good temperature resistance, high insulating strength), and its relative dielectric constant is measured by parallel plate capacitance method and stabilized at 3.4-3.6 in the high frequency range of 10 kHz-1 MHz, with an average value of 3.5; the relative dielectric constant of other insulating materials (such as polyurethane) can be adjusted according to the actual material (such as polyurethane =2.8-3.2), and polyimide is taken as an example here to illustrate the determination logic of the constant.

[0107] 5. Stranding density coefficient (step 222) Physical meaning: correcting the effect of stranding tightness on the thickness of the insulating layer - the more strands, the tighter the stranding, the greater the pressure between the strands, the thinner the insulating layer is extruded, and the weaker the attenuation effect on the electrical conductivity, so the weight of the single strand diameter d is amplified by the coefficient, and the effect of the insulating layer is reduced.

[0108] The actual thickness of the insulating layer of Litz wire with different number of strands (50-500 strands) and stranding tension (5-20 N) is measured, and it is found that when the number of strands increases or the tension increases, the actual thickness of the insulating layer decreases by 10%-30% compared with the nominal value; through data fitting, when the number of strands ≥300 strands and the tension ≥15 N, =1.5 (insulating layer extrusion is most obvious); when the number of strands ≤100 strands and the tension ≤10 N, =1.2 (extrusion is weaker), so the value range is [1.2, 1.5], covering typical stranding conditions.

[0109] In this application, by / N terms to modify the effect of the strands, quantifies the attenuation effect of the insulating layer thickness and dielectric properties on the conductivity through exponential terms, and breaks away from the limitation of directly using the copper bulk conductivity in traditional models, making the equivalent parameters closer to the actual conductive behavior of the litz wire.

[0110] Preferably, step 223: calculation of the initial filling rate (based on the area ratio principle), the filling rate is a key parameter reflecting the compactness of the litz wire conductive material, and its calculation needs to distinguish the difference between the total conductive area of the strands and the actual cross-sectional area of the bundle. This step introduces a "bundle profile correction" to eliminate the irregularity caused by twisting, and the formula is as follows:

[0111] Wherein, : initial filling rate (dimensionless), i.e. the proportion of conductive material without considering the gap between the strands; : conductive cross-sectional area of a single strand (unit: mm²); : roundness correction coefficient of the strand (dimensionless, value 0.92-0.98), calculated by the roundness error of laser scanning the cross-section of the strand, to correct the area deviation of non-ideal circular strands; : actual cross-sectional area of the litz wire bundle (unit: mm²), calculated by integral method after obtaining the outer profile of the bundle by three-dimensional scanning; : irregularity coefficient of the profile (dimensionless, value 0.03-0.08), reflecting the degree of deviation of the bundle shape from the ideal circle caused by twisting (the more irregular the shape, , the larger the value).

[0112] Circular area coefficient (Step 223) Physical meaning: the formula for calculating the cross-sectional area of a circular strand, when the diameter of a single strand is d, the radius is / 2, and the cross-sectional area is , which is used to quantify the actual conductive area of a single strand and is a basic constant for describing the cross-sectional area of a circular conductor, ensuring the geometric accuracy of the calculation of the conductive area of a single strand.

[0113] Roundness correction coefficient of the strand (Step 223) Physical meaning: correct the roundness error in the manufacturing process of the strand (industrial copper wire cannot be perfectly circular, and may have elliptical or slight protrusions), reflecting the deviation of the actual conductive area from the ideal circular area. Scan the cross-section of the strand with a laser profilometer to measure the deviation of the actual profile from the ideal circle, and calculate the ratio of the actual conductive area to the ideal circular area; the roundness error of industrial-grade copper wire is usually 2%-8% (i.e. the actual area is 2%-8% smaller than the ideal value), so , the value range is [0.92, 0.98], consistent with the manufacturing tolerance standard.

[0114] profile irregularity coefficient (Step 223) Physical meaning: Correct the deviation of the litz wire bundle shape from the ideal circular shape - the spiral arrangement of the strands during the twisting process can cause the outer profile of the bundle to be "wavy", and the actual cross-sectional area is 3%-8% larger than the ideal circular shape (diameter equal to the outer diameter of the bundle). Therefore, this coefficient is used to correct the calculation deviation of the actual cross-sectional area of the bundle. The outer profile of the bundle is obtained by three-dimensional scanning, and the actual cross-sectional area is calculated by integral method, and the ratio of the actual cross-sectional area to the "ideal circular area (D is the outer diameter of the bundle)" is calculated. The actual measurement shows that the smaller the twisting pitch (the more severe the twisting), the more obvious the profile irregularity, and the deviation range is 3%-8%, so the value is [0.03, 0.08], which accurately reflects the influence of twisting on the overall shape.

[0115] It can be seen that by correcting the manufacturing error of the strands and the shape deviation caused by twisting, the initial filling rate not only reflects the "material ratio", but also reflects the actual space compactness of the litz wire, and provides an accurate benchmark for subsequent twisting gap correction.

[0116] Optionally, the step 23 specifically comprises:

[0117] Step 231: Based on the litz wire twisting pitch and the bundle radius, the space twisting coefficient caused by the twisting of the strands is calculated, and the initial equivalent thickness is corrected to generate the corrected equivalent thickness;

[0118] Step 232: According to the coupling relationship between the twisting pitch and the number of strands, a dynamic model of the twisting loss coefficient is established, and the initial equivalent conductivity is corrected to generate the corrected equivalent conductivity;

[0119] Step 233: Combined with the twisting gap data obtained by three-dimensional scanning, the initial filling rate is gap compensated and corrected to generate the corrected filling factor;

[0120] Step 234: Based on the corrected equivalent thickness, the corrected equivalent conductivity and the corrected filling factor, multi-parameter collaborative verification is carried out to generate the equivalent structure parameters of the litz wire.

[0121] Preferably, in step 231, the initial equivalent thickness does not consider the "amplification effect" of the spiral twisting of the strands on the axial space occupation - the twisting causes the projection length of the strands in the axial direction to increase, and the actual equivalent thickness needs to be larger than the initial value. This step introduces "space twisting coefficient" to quantify this effect, and the formula is as follows:

[0122]

[0123] ​​​​: Corrected equivalent thickness (unit: mm), i.e. the final equivalent thickness after considering twisting and torsion; : Initial value of equivalent thickness calculated in step 221 (unit: mm); r: Radius of the Litz wire bundle (unit: mm), which is half of the outer diameter of the bundle obtained by 3D scanning; P: Twisting pitch (unit: mm), which is the length of a single strand of wire measured along the axial direction to complete one spiral; : Strand angle (unit: rad), same as step 221 ( ), reflecting the degree of torsional tilt; : Torsion correction factor (dimensionless, 0.8-1.0), determined by the strand elastic modulus (0.8 for hard copper wire and 1.0 for soft copper wire), corrects the effect of strand deformation on thickness during torsion. : It originates from the geometric characteristics of the helix and reflects the torsion arc within the unit axial length (the smaller the pitch and the larger the radius, the more severe the torsion); : Quantifying the contribution of torsion to axial thickness =0 (no twist), this item is 0 and the correction amount is 0; when When it increases (torsion intensifies), this nonlinear growth is consistent with the amplification effect of torsion on thickness; : The actual torsional amplification effect of hard copper wire (large elastic modulus and small deformation when torsion) is weaker than that of soft copper wire. Actual measurements show that the correction coefficient of hard copper wire is about 0.8, and that of soft copper wire is about 1.0, covering the material characteristics of mainstream Litz wire.

[0124] Preferably, in step 232, twisting causes magnetic fields between strands to interfere with each other (proximity effect enhancement), and the actual conductivity is lower than the initial value. This step differs from the traditional "static correction" approach and establishes a dynamic loss model that couples the three parameters of "twisting pitch, number of strands, and frequency." The formula is as follows:

[0125] : Corrected equivalent conductivity (unit: S / m), the final equivalent conductivity after considering the twisting loss; : initial value of equivalent conductivity calculated in step 222 (unit: S / m); N: total number of Litz wire strands (dimensionless); f: operating frequency (unit: Hz). The proximity effect is more significant at high frequencies. : Skin depth (unit: mm), given by calculate( is the resistivity of copper, , =1; : Dynamic coefficient of stranding loss (dimensionless, range 0.02-0.05), quantifies the rate of change of proximity effect with stranding parameters. : The more the number of strands, the more significant the magnetic field superposition between strands, and the adjacent effect loss increases by 0.3 power (derived from experimental data fitting, when the number of strands increases from 10 to 1000, the loss growth trend conforms to the index); : The adjacent effect loss is proportional to the square root of frequency (theoretical derivation result, the higher the frequency, the higher the magnetic field change rate, and the eddy current loss is enhanced); : The measured loss of 5 different specifications of litz wire (at 10 kHz-1 MHz frequency) shows that when P / δ<5 (the pitch is less than 5 times the skin depth), =0.05 (the loss enhancement is significant); when , =0.02 (the loss enhancement is gentle), covering the typical working conditions of high-frequency scenarios.

[0126] Preferably, in step 233, the initial filling rate does not consider the "distributed gap" formed by the twisting of the strands (caused by the spatial misalignment of the spiral arrangement of the strands), and the actual filling rate needs to be reduced by the gap proportion. This step combines the gap data of the three-dimensional scan, and the formula is as follows:

[0127] : The corrected filling factor (dimensionless), the final filling rate after considering the twisting gap; : The initial filling rate calculated in step 223 (dimensionless); : The average gap area around a single strand (unit: mm²), calculated from the gap point cloud data of the three-dimensional scan (taking the average of 100 sampling points per 1 mm axial length); N: the total number of litz wire strands (dimensionless); : The actual cross-sectional area of the litz wire bundle (unit: mm²), as in step 223; : Gap distribution correction coefficient (dimensionless, value 1.1-1.3), reflecting the non-uniformity of the gap in the axial direction (the gap in the edge area is larger than in the center, and needs to be enlarged and corrected). : The three-dimensional scan shows that the twisting gap is "dense in the center and sparse in the edge", and the average gap area of a single strand is about 5%-15% of the cross-sectional area of a single strand (the more the number of strands, the lower the gap proportion); : It is found that the gap area of the edge area of the bundle is 10%-30% larger than that of the center area, so this coefficient is used to correct the influence of the non-uniform distribution of the gap - when the number of strands is small (the edge proportion is high), take 1.3, and when the number of strands is large (the center proportion is high), take 1.1, to ensure the accuracy of the gap compensation.

[0128] Preferably, in step 234, to ensure that the corrected parameters meet the internal consistency of "geometric-physical characteristics", and to avoid contradictions caused by single parameter correction (such as large deviation of direct current resistance after correction of equivalent thickness and electrical conductivity). This step checks by simultaneous equations, and the formula is as follows:

[0129] : DC resistance (unit: Ω) calculated by the corrected parameter, formula is (L is winding length); : measured DC resistance (unit: Ω) measured by four-terminal method; : corrected equivalent copper foil conductive area (unit: mm²), that is ; : actual conductive area of the Litz wire (unit: mm²), that is the product of the wire bundle cross-sectional area and the filling rate. The resistance error threshold 2%: derived from the engineering precision requirement of the high-frequency transformer design (DC resistance is the basis for loss calculation, and the deviation needs to be controlled within 2% to ensure the accuracy of subsequent high-frequency loss prediction); the area error threshold 3%: considering the cumulative error of three-dimensional scanning and parameter calculation, the comprehensive error of the measured geometric parameters is usually ≤3%, and the threshold is strict and consistent with the engineering practice.

[0130] In summary, in the above specific implementation, the space twist correction of step 231 quantifies the nonlinear effect of the spiral structure on the equivalent thickness, which improves the precision by 15% compared to the traditional linear correction; the dynamic loss model of step 232 couples the twisting pitch, the number of strands and the frequency into a continuous function, getting rid of the limitations of fixed coefficient correction; step 233 combines the three-dimensional scanning gap data to realize the upgrade of the filling rate from "estimation" to "actual measurement calibration"; the collaborative verification of step 234 ensures that the corrected parameters are physically self-consistent, avoiding the overall deviation caused by single parameter optimization.

[0131] The finally generated equivalent structure parameters ( ) can accurately reflect the high-frequency loss characteristics of the Litz wire, providing a reliable simplified model for subsequent magnetic field and loss analysis.

[0132] Optionally, the step 3 specifically includes:

[0133] Step 31: based on the equivalent structure parameters of the Litz wire, a three-dimensional electromagnetic simulation model of the high-frequency transformer is constructed and corrected combined with the measured data of the key points to generate three-dimensional magnetic field distribution data around the Litz wire;

[0134] Step 32: according to the three-dimensional magnetic field distribution data and the Litz wire twisting parameters in the multi-physical quantity original feature set, the twist angle of the Litz wire strand relative to the wire bundle is calculated to generate the twist angle parameter;

[0135] Step 33: based on the three-dimensional magnetic field distribution data and the twist angle parameter, the magnetic field is decomposed into components parallel to the strand and perpendicular to the strand to generate the magnetic field component representation result.

[0136] Optionally, the step 31 specifically includes:

[0137] Step 311: Based on the equivalent structure parameters of the Litz wire, combined with the magnetic core material properties and winding arrangement information, a high-frequency transformer three-dimensional electromagnetic simulation model is constructed to restore the winding arrangement and magnetic core air gap distribution, to generate initial three-dimensional magnetic field simulation data;

[0138] Step 312: Magnetic field probes are arranged in the magnetic field distortion significant area including the winding end and the magnetic core air gap edge, and the magnetic field strength measured value of the key point is collected to generate the magnetic field measured data set;

[0139] Step 313: The least square method is used to fit the error of the initial three-dimensional magnetic field simulation data and the magnetic field measured data set, and the magnetic core permeability and winding equivalent conductivity parameters in the simulation model are dynamically adjusted to generate three-dimensional magnetic field distribution data.

[0140] Preferably, in step 311, the three-dimensional electromagnetic simulation model of the high-frequency transformer needs to accurately restore the coupling characteristics of "magnetic core-winding-air gap", especially to reflect the interaction between the Litz wire equivalent structure and the magnetic field of the magnetic core. This step breaks away from the limitations of the traditional "linear magnetic core + simplified winding" model, and introduces the coupling modeling of the nonlinear characteristics of the magnetic core and the equivalent parameters of the Litz wire, the formula is as follows:

[0141]

[0142] : Magnetic flux density vector (unit: T) at spatial position , reflecting the strength and direction of the magnetic field; : Vacuum permeability (constant), basic parameter of electromagnetism; : Relative permeability of the magnetic core (dimensionless), a function (nonlinear characteristic) of magnetic field intensity and saturation magnetic flux density , using the simplified Jiles-Atherton model: ( is the initial permeability, is the coercive force); : Magnetic field intensity vector (unit: A / m) at spatial position ; : Litz wire equivalent conductivity (unit: S / m) after step 232 correction, reflecting the current-carrying capacity of the winding; : Electric field intensity vector (unit: V / m) at the i-th layer winding, derived from the current density ; : Normal unit vector (dimensionless) of the i-th layer winding, distinguishing radial / axial current components; n: total number of windings (dimensionless), determined by the design parameters of the high-frequency transformer. : Physical constant of vacuum permeability, the base reference of electromagnetic calculation; Core nonlinear model parameters ( ): Based on datasheet measured data of core material (e.g. ferrite core , ), obtained by B-H curve fitting, to ensure the model reflects the saturation characteristics of the core at high frequencies; Litz wire equivalent parameters ( ): Directly call the corrected parameters of step 23 to make the winding model consistent with the actual conduction characteristics, avoiding the deviation of traditional simplified models ignoring the high-frequency effect of Litz wire.

[0143] Preferably, in step 312, the measured data of the magnetic field distortion significant area (winding end, air gap edge of the core) is the core basis for correcting the model, the magnetic field gradient of these areas can reach 10-100 T / m (much higher than 1-5 T / m in uniform areas), which needs to be captured by high-density sampling. This step uses the "gradient adaptive sampling" strategy, the formula is as follows:

[0144] : Sampling point spacing (unit: mm), dynamically adjusted with the magnetic field gradient; : Skin depth of Litz wire strand (unit: mm), the formula is ( is the working frequency), to ensure that the sampling accuracy is not lower than the spatial scale of the current skin; : The maximum value of the magnetic field gradient (unit: A / m·mm), determined by pre-scanning (about 50 A / m·mm at the winding end, about 100 A / m·mm at the air gap edge); : The magnetic field gradient at the current position (unit: A / m·mm). Coefficient 0.1: to ensure that the sampling point spacing is less than 1 / 10 of the skin depth, which is enough to capture the rapid change of the magnetic field at high frequencies (the skin depth is about 0.06 mm at 100 kHz, the sampling spacing needs to be ≤0.006 mm); Gradient adaptive logic: in areas with large magnetic field gradients (such as air gap edges), reduce (minimum 0.005 mm); in areas with small magnetic field gradients (such as the center of the core), increase (maximum 0.1 mm), to reduce the amount of data while ensuring accuracy (60% less data than uniform sampling).

[0145] Preferably, in step 313, the error between the initial simulation data and the measured data is mainly due to the nonlinear deviation of the core permeability and the estimation error of the Litz wire equivalent conductivity. This step dynamically adjusts the key parameters by the least squares method to minimize the error, the formula is as follows:

[0146]

[0147]

[0148] : corrected core relative permeability (dimensionless), as a function of spatial position (considering core anisotropy); : corrected winding equivalent conductivity (unit: S / m); : simulated and measured magnetic field strength vector (unit: A / m); M: total number of sampling points (dimensionless), usually ≥1000 (covering all distortion regions); : core initial relative permeability (dimensionless, based on material manual); σeq: initial value of equivalent conductivity corrected in step 232 (unit: S / m). Permeability adjustment range ±20%: the actual permeability of the core is affected by temperature and magnetic flux density, and the deviation is usually ≤20% (the measured permeability of the ferrite core decreases by 15% at 100°C); conductivity adjustment range ±30%: the equivalent conductivity of the litz wire is affected by the tightness of the litz wire, and the measured deviation is about ±25%, which is relaxed to ±30% to ensure convergence; final error ≤2%: based on the engineering precision requirement of high-frequency transformer magnetic field measurement - magnetic field error exceeding 2% will lead to loss prediction deviation >5%, and the threshold ensures the reliability of subsequent loss calculation.

[0149] In the above technical solution, step 311 introduces the coupling of core nonlinearity and litz wire equivalent parameters, which upgrades the model from "static approximation" to "dynamic coupling", which is closer to the electromagnetic characteristics under high frequency; the gradient adaptive sampling of step 312 reduces data redundancy while ensuring the capture of rapid changes in the magnetic field, and balances accuracy and efficiency; the least square method correction of step 313 reduces the magnetic field simulation error from the initial 15%-20% to within 2% through multi-parameter collaborative optimization, which provides a high-precision data basis for subsequent magnetic field component decomposition.

[0150] Optionally, the step 32 specifically comprises:

[0151] Step 321: Extract the litz wire litz pitch and wire bundle outer diameter parameters from the multi-physical quantity original feature set, and perform geometric correlation analysis to generate litz wire space litz feature parameters;

[0152] Step 322: Based on the spiral motion trajectory of the strand wire around the wire bundle axis, combined with the litz wire space litz feature parameters, derive the spatial angle calculation formula of the strand wire center axis and the wire bundle axis to generate a twist angle calculation model;

[0153] Step 323: Substitute the axial position information in the three-dimensional magnetic field distribution data and the litz wire space litz feature parameters into the twist angle calculation model to calculate the twist angle values of the strands at different axial positions to generate a twist angle parameter.

[0154] Preferably, in step 321, the twisting characteristics of the Litz wire depend not only on the basic geometric parameters, but also contain the "micro-unevenness" of the spatial arrangement of the strands (such as pitch fluctuation, layering difference). This step generates a set of characteristic parameters that accurately reflect the twisting rules by quantifying these characteristics, and the formula is as follows:

[0155] Wherein: : Average twisting pitch (unit: mm), averaged by the measured values of 5 consecutive pitches (p ), eliminating accidental errors caused by local distortion; : Pitch fluctuation rate (dimensionless, 0 ), reflecting the stability of the pitch in the twisting process (ideal twisting = 0, actual value 0.02-0.08); r: wire bundle radius (unit: mm), 1 / 2 of the average value of the outer diameter obtained by three-dimensional scanning; : Number of twisting layers (dimensionless, estimated by the total number of strands N: , about 6 strands per layer in a hexagonal arrangement); : Twisting uniformity coefficient (dimensionless, value 0.9-1.0), calculated by the standard deviation of the spatial position of the strands (the smaller the standard deviation, the better the uniformity, closer to 1). Number of pitch measurements 5: the average of 5 consecutive pitches is sufficient to eliminate random errors of a single measurement (error ≤1%), less than 3 is insufficient, more than 7 is redundant; : The pitch control accuracy of industrial twisting equipment is usually ±2%-±8% (the deviation is larger when high-speed twisting), through the actual measurement of 10 kinds of Litz wires, this range covers 95% of the working conditions; the layer estimation formula : In a hexagonal arrangement, the number of strands in the mth layer is about 6m, and the total number of strands , the number of layers is approximately , which is consistent with the actual twisting process (such as 100 strands of Litz wire about 4 layers: 3x4x5=60, close to the actual layering of 100 strands); : The standard deviation of the position of the strands in good twisting (such as precision equipment production) is ≤0.1r, = 0.98; the standard deviation of the strands in poor twisting (such as manual twisting) is ≥0.3r, = 0.9, obtained by fitting the standard deviation of the spatial point cloud.

[0156] Preferably, in step 322, the motion trajectory of the strands around the wire bundle axis is a spiral line, and the spatial angle between it and the axis (twist angle) depends not only on the basic geometric parameters, but also on the layering twisting (the outer layer strands twist more severely) and the elastic deformation of the strands. This step derives a dynamic angle model considering these factors, and the formula is as follows:

[0157] in: : mth layer, radius The twist angle of the strand at π (unit: rad), is a function of the number of layers and radial position; m: the number of twisted layers (dimensionless, ); : the average radius of the strands in the mth layer (unit: mm), (The outer radius is close to the bundle radius r); : Layer correction factor (dimensionless, ), the outer strands have a greater twisting force, so the actual angle is 10% larger than the geometric value (m-1); z: axial position (unit: mm); : Pitch fluctuation correction term, which simulates the effect of periodic pitch changes on the angle (the fluctuation frequency is consistent with the pitch). : Derived from the geometric nature of the helix - the tangent slope of the helix is ​​equal to "circumference length / pitch", that is, the torsion distance within the unit axial length, which is the basis for calculating the spatial angle; the layer correction coefficient : The actual measurement of 4-layer stranded Litz wire shows that the angle of the second layer is 10% larger than the geometric value, the third layer is 20% larger, and the fourth layer is 30% larger, which is consistent with the "over-twisting" phenomenon caused by the greater twisting force on the outer layer of the wire; the fluctuation period : Pitch fluctuations are usually periodic (consistent with the mechanical vibration frequency of the stranding equipment), with the period equal to the pitch , the sine function can accurately simulate this periodic deviation (the correlation coefficient between the measured fluctuation curve and the sine curve is ≥0.9).

[0158] Preferably, in step 323, the twisting angle of the strand increases linearly with the axial position z (each pitch advances , angle increases ), but due to the influence of pitch fluctuation and layered twisting, the angle change is slightly nonlinear. The formula for this step is as follows:

[0159]

[0160] in, : Axial position z, layer m, radius The twist angle of the strand at (unit: rad); : The initial torsion angle (unit: rad) at step 322 is exist Calculated when : Axial correction factor (dimensionless, value 0.95-1.05), determined by the Poisson's ratio of the strand material (Poisson's ratio of copper is 0.34, = 1.0), to correct the slight influence of axial tension on twist angle; : axial non-linear correction term, simulating the slight angle fluctuation (amplitude 5%, period equal to pitch) caused by the "tight-loose" alternation during stranding. : core geometry coefficient, ensuring that the angle increases (ie 360°) for every pitch, in line with the physical nature of helical lines "advance a pitch, twist a turn"; axial correction coefficient : the strands will be slightly stretched (elongation ≤5%) under axial tension during stranding, and the actual measurement shows that the influence of stretching on angle is ≤5%, so the coefficient range is set to [0.95, 1.05]; non-linear correction term amplitude 0.05: axial angle measurement of 10 types of litz wires shows that the angle deviation (difference from the ideal linear value) is at most 5%, and it is periodic (with the same period as the pitch), and the cosine function can accurately fit this fluctuation (error ≤1%).

[0161] In summary, in step 321, the pitch fluctuation rate and the stranding uniformity coefficient upgrade the stranding characteristics from "macro-geometric description" to "micro-unevenness quantification", laying the foundation for the accuracy of subsequent angle calculation; in step 322, through the layer correction coefficient and the pitch fluctuation correction term, the "over-twisting of the outer layer" and the periodic fluctuation of the pitch in layered stranding are captured, and the accuracy is improved by 20% compared to the traditional geometric model; in step 323, a dynamic function of angle with axial position is established, considering the non-linear influence of axial tension and periodic fluctuation, and the angle calculation error at different positions is reduced from 8%-15% of the traditional method to within 3%.

[0162] Optionally, the step 33 specifically comprises:

[0163] Step 331: Establish a strand local coordinate system according to the twist angle parameter, and perform coordinate system conversion on the three-dimensional magnetic field distribution data to generate local magnetic field data adapted to the strand posture;

[0164] Step 332: Based on the strand local coordinate system, project the magnetic field strength vector in the local magnetic field data onto the longitudinal axis and the radial direction, respectively extract the magnetic field component parallel to the strand and the magnetic field component perpendicular to the strand, to generate magnetic field component decomposition results;

[0165] Step 333, spatial interpolation processing is performed on the magnetic field component decomposition results, so that they are point-by-point matched with the three-dimensional spatial distribution of the litz wire strands, to generate magnetic field component representation results.

[0166] Preferably, in step 331, the helical twisting of the strands causes their spatial posture to change dynamically with their axial position. The traditional global coordinate system (with the bundle axis as the Z axis) cannot directly describe the direction of the magnetic field around the strands. This step uses the twist angle parameter to construct a local coordinate system that adapts to the strand posture in real time, achieving the "posture-following" conversion of the magnetic field data. The formula is as follows:

[0167] Among them, the rotation matrix for:

[0168]

[0169] in,( , , ): Magnetic field intensity component (unit: A / m) in the global coordinate system (X, Y, Z), where the Z axis coincides with the axis of the harness; ( , , ): Magnetic field intensity components in the local coordinate system (x', y', z') of the strand (unit: A / m); z': longitudinal axis of the local coordinate system, coincident with the strand axis (parallel to the strand); x', y': radial axis of the local coordinate system, perpendicular to the strand axis (x' is radially outward, y' is along the circumferential tangent); : The torsion angle at the axial position z calculated in step 323 (unit: rad), that is, the angle between the strand axis and the bundle axis (Z axis); : The rotation matrix that changes dynamically with z realizes the transformation of the global magnetic field to the local coordinate system (the matrix changes with each axial distance of 1 mm). Updated once). Rotation matrix structure: Based on the 2D plane rotation logic (strand twisting is mainly in the radial plane), the angle between the Z axis (harness axis) and the z' axis (strand axis) only affects the component conversion in the XY plane, and the Z direction component remains unchanged ( = ), simplifying the calculation while ensuring accuracy; matrix update frequency (every 1mm): the rate of change of the strand torsion angle with the axial position is approximately (Unit: rad / mm), when =10mm, the angle changes by about 0.6rad per 1mm, and high-frequency updates can ensure that the conversion error is ≤1% (actual measurements show that the conversion accuracy of a 1mm step size is 3 times higher than that of a 5mm step size).

[0170] Preferably, in step 332, the skin effect loss of the Litz wire at high frequency is mainly caused by the magnetic field component perpendicular to the strands ( ) dominates, the proximity effect loss and the parallel component ( ) are related, and the two components need to be accurately separated. This step breaks away from the limitations of traditional "two-dimensional projection" and achieves accurate vector decomposition of the three-dimensional magnetic field. The formula is as follows:

[0171]

[0172] in: : The magnetic field component parallel to the strands (unit: A / m), that is, the magnetic field in the z'-axis direction of the local coordinate system, which directly affects the longitudinal proximity effect between strands; : The magnetic field component perpendicular to the strand (unit: A / m), which is the resultant vector of the x' and y' axis components of the local coordinate system and determines the skin effect intensity on the strand surface; : Vertical component correction coefficient (dimensionless, value 1.02-1.05), compensates for the magnetic field concentration effect caused by the surface curvature of the strand (the smaller the strand diameter, the greater the curvature, and the actual value of the vertical component of the magnetic field is 2%-5% larger than the geometric projection). The parallel component is directly taken : The z'-axis of the local coordinate system strictly coincides with the strand axis, so this component is the magnetic field parallel to the strand, and its physical meaning is clear; the vertical component is the resultant vector: the x' and y' axes are both perpendicular to the strand axis, and their resultant vector fully reflects the magnetic field intensity in the vertical direction, avoiding the error of traditional single radial projection that ignores the circumferential direction component; : Actual measurements of strands with a diameter of 0.1-0.5 mm show that the magnetic field concentration caused by curvature causes the vertical component to be 2%-5% higher than the geometric projection value (the smaller the diameter, the greater the deviation). The correction factor can eliminate this systematic error.

[0173] Preferably, in step 333, the sampling point spacing of the three-dimensional magnetic field distribution data (usually 1-2 mm) is larger than the strand diameter (0.1-0.5 mm). Interpolation is required to match the magnetic field components with the microscopic spatial distribution of the strands (one point every 0.1 mm) point by point. This step uses the "spiral path adaptive interpolation" algorithm, and the formula is as follows:

[0174]

[0175] Among them, the weight coefficient , The interpolation point and The spatial distance between adjacent sampling points (unit: mm).

[0176] in: : Strand position after interpolation The magnetic field component at );z: axial position (unit: mm); : circumferential angle (unit: rad), changes with axial position ( , reflecting the helical trajectory of the strand; : magnetic field component of the i-th neighboring sampling point; : inverse distance square weight, ensuring that the closer the sampling points, the greater the influence on the interpolation result (avoiding the interference of distant points). Number of neighboring sampling points 4: based on the "local linearity" of the strand helical trajectory, 4 points (2 in the axial direction, 2 in the circumferential direction) are sufficient to capture the spatial variation trend of the magnetic field, less than 3 points will cause large interpolation fluctuations, and more than 6 points will cause a sharp increase in calculation amount; the weight function adopts inverse distance square: compared with linear weight, inverse distance square can better highlight the influence of nearby points, so that the deviation of the interpolation result from the actual measured strand magnetic field is ≤3% (linear weight deviation is about 8%); the calculation of the circumferential angle : strictly matches the helical motion of the strand (the circumferential angle increases by with each pitch of the strand, ensuring that the interpolation point corresponds to the actual position of the strand (spatial matching error ≤0.05mm).

[0177] In summary, steps 331-333 realize the synergy of "dynamic coordinate system- vector accurate decomposition-helical adaptive interpolation", which is different from the limitations of traditional fixed coordinate system, so that the magnetic field data can follow the helical posture of the strand in real time, and the conversion error is reduced from 15% to 2%; when distinguishing the parallel / vertical components, the real-time radial and circumferential directions of the strand are considered, which is more suitable for the three-dimensional characteristics of high-frequency magnetic field than two-dimensional decomposition, and the calculation accuracy of proximity effect loss is improved by 20%; the sampling point selection and weight are optimized according to the helical trajectory of the strand, so that the point-by-point matching degree of the magnetic field component and the spatial distribution of the strand is ≥98% (traditional uniform interpolation is only 85%).

[0178] Optionally, the step 4 specifically comprises:

[0179] Step 41: based on the magnetic field component representation result and the equivalent structure parameters of the litz wire, analyzing the loss composition of the skin effect and the proximity effect in the unit length to generate a loss composition model;

[0180] Step 42: according to the loss composition model and the current harmonic characteristics in the multi-physical quantity original feature set, calculating the skin effect loss and the proximity effect loss in the unit length to generate unit length loss data;

[0181] Step 43: performing spatial integration processing on the unit length loss data to generate a basic loss value.

[0182] Optionally, the step 41 specifically comprises:

[0183] Step 411: performing correlation analysis on the magnetic field component representation result and the equivalent structure parameters of the litz wire to generate a magnetic field-structure parameter correlation model;

[0184] Step 412: Based on the magnetic field-structure parameter correlation model, the loss generation mechanism of skin effect and proximity effect is analyzed to generate loss mechanism analysis results;

[0185] Step 413: Based on the loss mechanism analysis results, mathematical modeling is performed to quantify the loss proportion of the two effects in unit length to generate a loss composition model.

[0186] Preferably, in step 411, the coupling relationship between the magnetic field component and the litz wire structure parameter is the core link of the loss analysis. The skin effect loss is more sensitive to the perpendicular magnetic field component and the equivalent thickness, and the proximity effect loss is strongly related to the parallel magnetic field component and the filling rate. This step quantifies the dynamic correlation between "field-structure" to generate a model that can reflect the interaction of parameters, and the formula is as follows:

[0187]

[0188] Wherein, the matrix element is the coupling coefficient (unit: ), the physical meaning is:

[0189] : Coupling coefficient of perpendicular magnetic field component and equivalent thickness , quantifying the influence of on induced loss; : Coupling coefficient of perpendicular magnetic field component and equivalent conductivity , reflecting the amplification effect of conductive ability on skin effect; : Coupling coefficient of perpendicular magnetic field component and filling rate (weak correlation, small value); : Coupling coefficient of parallel magnetic field component and equivalent thickness (weak correlation); : Coupling coefficient of parallel magnetic field component and equivalent conductivity ; : Coupling coefficient of parallel magnetic field component and filling rate , the higher the filling rate, the stronger the magnetic field superposition between strands, and the more significant the coupling. The coupling coefficient is calculated by partial least squares regression (PLSR) based on the measured data of 10 groups of different structure parameters and magnetic field components, and the core value law is: : The closer the equivalent thickness is to the skin depth ( ≈ ), the stronger the coupling (at this time takes 1.2); deviates from The coupling is weakened ( 0.8 when : The higher the equivalent conductivity (such as soft copper wire), the more sensitive the skin effect loss is to the perpendicular magnetic field. The measured coupling coefficient of high conductivity Litz wire is about twice that of low conductivity. : Fill rate When (strands are densely packed), the superposition effect of parallel magnetic fields between strands is significant, and the coupling coefficient is 3.0; When (the gap is large), the superposition is weakened and is taken as 2.0; the weak correlation coefficient ( , ): The value is ≤0.5. Since the filling rate has a weak effect on the skin effect and the equivalent thickness has a weak effect on the proximity effect, they can be simplified to secondary parameters.

[0190] Preferably, in step 412, the loss mechanisms of the skin effect and the proximity effect are essentially different: the former is the energy loss of the current in a single strand due to the vertical magnetic field being concentrated on the surface, while the latter is the additional loss between strands due to the eddy current induced by the parallel magnetic field. This step analyzes the quantitative laws of the two mechanisms through the formula:

[0191] 1. Skin effect loss mechanism

[0192]

[0193] 2. Proximity effect loss mechanism

[0194]

[0195] in: : Skin effect loss per unit length (unit: W / m); : proximity effect loss per unit length (unit: W / m); : Skin effect loss coefficient, derived from electromagnetic field theory (similar to vacuum permeability and the magnetic permeability of copper Related: ; : Proximity effect loss coefficient, taking into account the geometric attenuation of magnetic field coupling between strands (smaller than the skin coefficient due to shorter eddy current paths); : Skin depth (unit: m), reflecting the spatial scale of current concentration ( is the resistivity of copper); N: total number of strands (dimensionless); the greater the number of strands, the more significant the proximity effect (in a 0.8 power relationship, as the edge strands are shielded by the middle strands); : Pitch fluctuation rate (same as step 321). The greater the fluctuation, the more uneven the strand arrangement, and the stronger the proximity effect (correction term quantify this impact). and : Derived from Maxwell's equations, ; Proximity effect is shorter than skin effect in eddy current path, coefficient is about 60% of skin effect ( ), consistent with measured data; : Measured proximity loss of 10-500 strand Litz wire, found that loss increases about 6 times per 10 times of strand count ( ), 0.8 power is more consistent with actual situation than linear relationship ( ); Exponential term : When equivalent thickness is much larger than , skin effect is saturated, loss growth slows down (e.g. when , exponential term ≈ 0.05, loss tends to be stable), consistent with non-linear characteristics of skin effect.

[0196] Preferably, in step 413, the loss constitutive model needs to quantify the proportion of the two effects under different frequencies and structure parameters, rather than simply superimposing them. This step generates a model that can reflect the interaction of parameters by introducing a "dynamic loss proportion coefficient". The formula is as follows:

[0197]

[0198]

[0199]

[0200] Where: : Total loss per unit length (unit: W / m); , : Proportion of skin effect and proximity effect loss (dimensionless, ); : Loss proportion coefficient (dimensionless, value 0.8-1.2), obtained by fitting experimental data (1.2 for Litz wire with thin strands and small pitch, as proximity effect is more significant); : Characteristic frequency (unit: Hz, ), proximity effect proportion rises at high frequency (when , exponential term decays, increases). : Measured loss of different specifications of Litz wire shows that the proximity effect of thin strand wire ( ) is relatively stronger, 1.2; thick strand wire ( ) takes 0.8, covering the main types of Litz wire; : It is found through measurement that 100 kHz is the turning point of loss proportion - below 100 kHz, skin effect proportion is higher ( 50%), above 100 kHz, proximity effect proportion dominates ( 60%, matching the typical working frequency (50kHz-200kHz) of high-frequency transformers; the exponential term : Quantification of the impact of frequency on loss proportion - the term approaches 0 at high frequencies, simplifying the denominator to 1, With increasing, consistent with the rule that the proximity effect dominates in high-frequency scenarios.

[0201] In summary, steps 411-413 form a logical chain of "field-structure coupling-mechanism analysis-proportion quantification", the coupling matrix of step 411 quantifies the nonlinear relationship between magnetic field components and structural parameters, enabling the model to capture "parameter interaction effects" (such as the synergistic effect of equivalent thickness and vertical magnetic field); the loss mechanism formula of step 412 introduces an exponential term and a nonlinear term of the number of strands, accurately reflecting the saturation characteristics of the skin effect and the strand dependence of the proximity effect, with a 30% improvement in accuracy compared to traditional formulas (such as the Proximity Effect Formula and the Skin Effect Formula); the dynamic proportion model of step 413 distinguishes between high and low frequencies , solving the error problem of fixed proportion models in a wide frequency range (loss proportion prediction error ≤5% at frequencies of 10kHz-200kHz).

[0202] Preferably, the step 42 specifically comprises:

[0203] Step 421: Perform spectral analysis on the current harmonic features in the original multi-physical quantity feature set, extract the current amplitude and frequency parameters of each harmonic, and generate a current harmonic parameter set;

[0204] Step 422: Substitute the current harmonic parameter set into the loss composition model to calculate the skin effect loss and proximity effect loss corresponding to each harmonic in the unit length, respectively, to generate sub-harmonic loss data;

[0205] Step 423: Superimpose the sub-harmonic loss data to obtain the total loss in the unit length, and generate unit length loss data.

[0206] Preferably, in step 42, the current of the high-frequency transformer is not an ideal sine wave, but contains a fundamental wave (such as 50Hz) and multiple harmonics (such as 3, 5, 7, or even hundreds of high-frequency harmonics), which significantly increases the high-frequency loss of the Litz wire (as the loss is proportional to the square of the frequency). This step extracts the key parameters of each harmonic through high-precision spectral analysis, and the formula is as follows:

[0207]

[0208] where, ​: Time domain current signal (unit: A), collected by Rogowski coil, with a sampling rate ≥ 10 times the highest harmonic frequency (e.g., a 200kHz signal requires a 2MHz sampling rate); : Hanning window function ( ), used to reduce spectrum leakage (T is the sampling period); FFT: Fast Fourier Transform, which converts the time domain signal into the frequency domain to obtain the amplitude of each frequency component; : The current amplitude of the nth harmonic (unit: A), which is the peak value of the corresponding frequency in the frequency domain; : frequency of the nth harmonic (unit: Hz), ( is the fundamental frequency); : Effective harmonic number (dimensionless, range 20-50), retaining harmonics with amplitudes ≥ 1% of the fundamental (harmonics below this value contribute negligibly to losses). Hanning window coefficient (0.54, 0.46): The weighting coefficient of the classic Hanning window, which can suppress spectrum leakage to -31dB (twice as good as the -13dB of the rectangular window), ensuring that the harmonic amplitude measurement error is ≤2%; Sampling rate (10 times the maximum frequency): Follows the Nyquist sampling theorem while reserving 5 times redundancy (the actual high-frequency harmonics contain 200kHz components, which can be accurately captured by a 2MHz sampling rate); Effective harmonic number : In the measured current harmonics of high-frequency transformers, the amplitude of harmonics above the 30th order is usually less than 1% of the fundamental wave. Therefore, taking the 20th to 50th order can cover 99% of the loss contribution and reduce the amount of calculation.

[0209] Preferably, in step 422, the losses of each harmonic need to be calculated separately and then added together (because the loss is proportional to the square of the frequency and the square of the current, the losses of different harmonics cannot be directly combined). This step is based on the loss composition model of step 41, and the sub-harmonic loss formula is derived:

[0210]

[0211] in: 、 : Skin effect and proximity effect loss of the nth harmonic (unit: W / m); : nth harmonic frequency (unit: Hz), ; 、 : perpendicular and parallel magnetic field components generated by the nth harmonic (unit: A / m); : Current-magnetic field conversion coefficient (dimensionless, value 0.9-1.0), corrects the attenuation caused by the diffusion of the magnetic field at the winding end (based on the derivation of Ampere's loop law, ideally =1); : winding turns (dimensionless), reflecting the amplification of current-generated magnetic field; r: winding average radius (unit: m), affecting the magnetic field strength (the larger the radius, the weaker the magnetic field); : skin depth of the nth harmonic (unit: m), . : the measured magnetic field at the end of the winding is 10% lower than the ideal value (calculated by Ampere's law) due to magnetic flux diffusion, so take 0.9; the magnetic field in the middle of the winding is uniform, take 1.0, and the overall weighted average; : the torsion angle based on step 323 , decompose the magnetic field into perpendicular and parallel components to ensure consistency with the definition of the magnetic field component in step 332; upper limit of harmonic number: when , , it is much smaller than the diameter of the strand, and the skin loss tends to be saturated (exponential term ), which can simplify the calculation.

[0212] Preferably, in step 423, the loss of each harmonic is not simply linearly superimposed, as high-frequency harmonics may produce "cross-coupling loss" between strands (such as the interaction of the magnetic fields of the 3rd and 5th harmonics). This step introduces a coupling correction term, the formula is as follows:

[0213]

[0214] Where: : total loss per unit length (unit: W / m); : harmonic coupling correction coefficient (dimensionless, value 0.01-0.03), quantifying the cross-loss between different harmonics; : phase difference between the nth and mth harmonics (unit: rad), obtained from the frequency spectrum analysis in step 421 (phase difference affects the positive and negative of the coupling loss); double summation term: consider the cross-coupling of all different harmonic pairs ), with the geometric mean of the loss of the two as the weight (as the coupling strength is related to the product of the amplitudes of the two). : the measured data of a 100kHz transformer shows that the harmonic coupling loss accounts for 1%-3% of the total loss (the coupling of the fundamental and 3rd harmonics is the most significant), and this coefficient is obtained by fitting experimental data; phase difference : when the harmonics are in phase ( =0), the coupling loss is enhanced ( ); when they are out of phase, the coupling loss is weakened ( ), which is consistent with the principle of electromagnetic field energy superposition; geometric mean: the coupling loss is related to the loss of both harmonics, and the geometric mean better reflects the nonlinear coupling characteristics than the arithmetic mean (the measurement error is reduced by 2%).

[0215] In summary, steps 421-423 reduce the harmonic amplitude measurement error from 5% to 2% by "precise harmonic extraction-harmonic loss calculation-coupling correction superposition", the Hann window FFT analysis of step 421 provides high-precision input for subsequent loss calculation; the magnetic field component decomposition of step 422 dynamically associates the current harmonics with the magnetic field components, ensuring that the harmonic loss is physically matched with the magnetic field analysis of step 3; the coupling correction term of step 423 first quantifies the cross-loss between different harmonics, making the total loss calculation closer to the measured value than the traditional linear superposition (error reduced from 8% to 3%).

[0216] The final generated unit length loss data not only contains the independent contribution of each harmonic, but also considers the synergistic effect between harmonics, providing a high-precision loss benchmark for the spatial integration of step 43.

[0217] Preferably, the step 43 specifically includes:

[0218] Step 431: Spatial grid discrete processing is performed on the unit length loss data to obtain the loss values of each grid element in the axial and radial directions of the winding, to generate a spatial discrete loss data set;

[0219] Step 432: Based on the three-dimensional geometric model of the winding, the integration boundary and weight coefficient in the axial and radial directions are determined to generate integration parameters;

[0220] Step 433: Weighted integration operation is performed on the spatial discrete loss data set to obtain the reference loss of the entire winding, to generate the basic loss value.

[0221] Preferably, in step 431, the unit length loss data is the average loss of the entire winding, while the actual loss has significant differences in the axial (winding length direction) and radial (winding thickness direction) directions - for example, the loss near the air gap of the magnetic core is 30%-50% higher than that far from the air gap, and the loss at the end of the winding is 20% higher than that in the middle. This step discretizes the continuous loss distribution into calculable units through gridding, and the formula is as follows:

[0222]

[0223] Wherein: : Unit length loss (unit: W / m³) of the i-th axial grid and the j-th radial grid, i.e. volume loss density; : Unit length total loss (unit: W / m) calculated in step 423, as the reference value; : Position (unit: m) of the i-th axial grid, increasing from the end of the winding ( ) to the middle ( ); L: Total length of the winding (unit: m); : The radius of the jth radial grid (unit: m), from the inner diameter of the winding ( ) to the outer diameter ( ) increases; R: winding outer diameter (unit: m); : Axial loss correction factor, quantifying the end effect (the end loss is 30% higher than the middle); : Radial loss correction factor, quantifies the radial magnetic field gradient (loss at the inner diameter is 20% higher than at the outer diameter). Axial correction term : The loss is significant within one-third of the length of the winding end. After exceeding this range, the end effect weakens (exponential decay), which is consistent with the measured data (the loss in the end 1 / 3 area accounts for 40%); radial correction term : The radial magnetic field weakens as the radius increases (proportional to 1 / r), and the loss decreases in a square law, which is consistent with the magnetic field distribution law; the correction coefficient ( .3. .2): By comparing the infrared temperature measurement data at different locations with the average loss, the end loss increment of 30% and the inner diameter loss increment of 20% are fitted to ensure that the grid loss is consistent with the actual temperature distribution.

[0224] Preferably, in step 432, the integral parameter should reflect the difference in contribution of different grid cells to the total loss - cells close to the core (with strong magnetic field) have a greater weight, and cells far from the core have a smaller weight. The formula is as follows:

[0225]

[0226] in: : Weight coefficient of the i-th axial grid (dimensionless), ranging from 1.0 to 1.1, the middle unit ( ) Because the magnetic field is uniform, the weight is the highest; : The weight coefficient of the jth radial grid (dimensionless), proportional to the radius (at the outer diameter , inner diameter ), because the strands are longer at the outer diameter (larger circumference); : Winding inner and outer diameters (unit: m), defining the radial integration range; : Make the axial weight sinusoidal distribution, high in the middle and low at the ends, in line with the law of magnetic field uniformity. Axial weight fluctuation 10% (0.1 coefficient): The measured magnetic field uniformity in the middle of the winding is 10% higher than that at the ends. The loss calculation needs to give a higher weight to avoid the low magnetic field area at the end lowering the total loss; Radial weight benchmark (average radius): ( ) / 2 as the benchmark to ensure that the weight coefficient has a clear physical meaning (for example, if the outer diameter is 1.2 times the average radius, the weight is 1.2); the boundary range: strictly consistent with the three-dimensional geometric model to avoid the integration range exceeding the actual size of the winding (for example, the radial direction does not include the air layer in the non-winding area).

[0227] Preferably, in step 433, the losses of all grid units are accumulated by weighted integration to obtain the basic loss value of the whole winding, and the formula is as follows:

[0228]

[0229] Wherein: : basic loss value of the whole winding (unit: W); , : number of axial and radial grids (dimensionless), the more the grids, the higher the accuracy (20 grids can make the error ≤3%); : axial and radial grid step (unit: m), , ; : volume of the first grid (unit: m³), convert the unit length loss (W / m) to volume loss (W / m³) and multiply by the volume; double summation: traverse all grid units to realize the loss accumulation of the whole space. Number of grids (20-50): 20 grids have high calculation efficiency (time consumption <1s), and 50 grids have higher accuracy (error <1%), which balances the efficiency and accuracy of engineering calculation; volume calculation ( ): volume formula based on cylindrical coordinate system, which accurately reflects the volume difference of strands at different radial positions (single circle strand length is longer at outer diameter); weight product ( ): make the units with strong magnetic field and large volume contribute more significantly, and the deviation of the final total loss from the infrared thermal imaging measurement value is ≤5% (the deviation of traditional uniform integration is about 15%).

[0230] In summary, steps 431-433 realize the quantification of the spatial non-uniformity of the loss (end effect, radial gradient) through the grid calculation formula of step 431 “microscopic dispersion-weighted quantization-global integration”, so that the loss of each grid unit corresponds to the actual magnetic field distribution; the weight coefficient of step 432 is determined based on the magnetic field strength and the strand volume, which avoids the error caused by “equal weight integration” (such as overestimating the contribution of low magnetic field area); the volume weighted integration of step 433 converts the unit length loss into three-dimensional volume loss, and the final total loss can be directly compared with the measured power (such as the difference between input and output power) to realize the “calculation-experiment” closed loop verification.

[0231] Optionally, the step 5 specifically comprises:

[0232] Step 51: based on the equivalent structure parameters of the litz wire and the magnetic field component representation results, an equivalent circuit model of mutual impedance containing self-impedance and mutual impedance is constructed to generate a harmonic loss calculation circuit model;

[0233] Step 52: Calculate the loss value of each harmonic according to the harmonic loss calculation circuit model and current harmonic characteristics, and modify the coupling effect through mutual impedance parameters to generate the modified loss value of each harmonic;

[0234] Step 53: Superimpose the modified loss value of each harmonic and the basic loss value to generate the predicted loss value of the Litz wire.

[0235] Optionally, the step 51 specifically comprises:

[0236] Step 511: Correlate the equivalent structure parameters of the Litz wire with the magnetic field component representation results, extract the key parameters affecting the winding impedance characteristics, and generate an impedance characteristic parameter set;

[0237] Step 512: Based on the impedance characteristic parameter set, combine the electromagnetic coupling relationship between the primary and secondary windings of the high-frequency transformer, and establish a matrix equivalent circuit model containing the primary self-impedance, secondary self-impedance and mutual impedance between the primary and secondary windings, to generate a mutual impedance equivalent circuit model;

[0238] Step 513: Modify the impedance parameters in the mutual impedance equivalent circuit model for frequency characteristics to generate a harmonic loss calculation circuit model that can be directly used for harmonic loss calculation.

[0239] Preferably, in step 511, the impedance characteristics (self-impedance and mutual impedance) of the Litz wire are the basis for high-frequency harmonic loss calculation, which is not only related to the geometric structure, but also affected by the magnetic field coupling strength. This step extracts characteristic parameters that can reflect the essence of impedance by analyzing the correlation between equivalent structure parameters and magnetic field components, and the formula is as follows:

[0240]

[0241] Wherein, the physical meaning and extraction logic of each parameter are: : Self-resistance (unit: Ω / m), skin effect resistance varying with frequency, formula is , reflecting the resistance increase of single wire due to current skin effect; : Self-inductance (unit: H / m), generated by the magnetic field of the strand itself, ( radius of single wire), directly related to geometric size; : Mutual resistance (unit: Ω / m), eddy current resistance generated by the magnetic field coupling between strands, , number of strands N and filling rate positively correlated; M: mutual inductance (unit: H / m), magnetic field coupling coefficient between primary and secondary windings, ( primary and secondary self-inductance, coupling coefficient); Impedance coupling coefficient (dimensionless, value 0.7-0.9), characterized by the parallel magnetic field ratio in the results (calculated by ), reflecting the strength of the magnetic field coupling between the strands. Self-resistance frequency term : Theoretical derivation from skin effect - resistance is proportional to the square root of frequency at high frequency ( ), this coefficient ensures that the resistance changes with frequency in accordance with the physical law; self-inductance logarithmic term (2 / ): Based on the geometric definition of inductance (the logarithmic relationship between the size of the wire and the self-inductance), when , this term tends to be stable, avoiding the inductance fluctuation caused by the slight change of geometric size; : The higher the parallel magnetic field ratio (the tighter the strand twisting), the stronger the coupling, closer to 0.9; when the gap is large, the coupling is weak, taking 0.7, consistent with the magnetic field component distribution in step 332.

[0242] Preferably, in step 512, the electromagnetic coupling of the primary and secondary windings of the high-frequency transformer needs to be quantified by the mutual impedance model - the magnetic field generated by the primary current will induce a voltage in the secondary (mutual impedance effect), and vice versa. This step is based on the impedance characteristic parameter set to establish a matrix model to reflect this two-way coupling, and the formula is as follows:

[0243]

[0244] Wherein, the composition and physical meaning of the impedance matrix elements are: : Primary self-impedance (unit: Ω), is the primary self-resistance (the same as step 511 ), is the primary self-inductance; : Secondary self-impedance (unit: Ω), symmetric structure with the primary side; : Mutual impedance (unit: Ω), symmetric due to electromagnetic reciprocity, is the mutual resistance (eddy current loss), M is the mutual inductance; : Primary and secondary voltage phasor (unit: V); , : Primary and secondary current phasor (unit: A); =2 : Angular frequency (unit: rad / s). Mutual impedance symmetry ( = ): Based on the electromagnetic field reciprocity theorem, the coupling strength of the primary to the secondary is equal to the coupling strength of the secondary to the primary, which is the basic characteristic of the electromagnetic coupling of the high-frequency transformer, and no additional correction is needed; reactance term ( , ): reflects the energy storage characteristic of magnetic field, self-inductance is proportional to square of turns number ( ), mutual-inductance is proportional to product of primary and secondary turns number ( ), in line with the law of electromagnetic induction; resistance term ( ): contains skin effect ( ) and proximity effect ( ) loss, ensures the impedance model can reflect both energy loss and magnetic field coupling.

[0245] Preferably, in step 513, traditional impedance model ignores the "frequency dependence" (e.g. self-resistance increases significantly with frequency, mutual-inductance decreases slightly due to eddy current effect) at high frequency, leading to calculation bias of harmonic loss. This step introduces a frequency correction term to adapt the model to each harmonic frequency, the formula is as follows:

[0246]

[0247] The corrected harmonic loss calculation circuit model is:

[0248]

[0249] : reference frequency (unit: Hz, take the rated working frequency, such as 100 kHz), as the reference point for frequency correction; self-resistance correction term ( ): the measured shows that when = 5 , the skin effect causes the self-resistance to be 25% higher than the reference value ( , the total increase of 8.5% after superimposing the reference 1, the cumulative multiple group data fitting exponential 0.3); self-inductance and mutual-inductance correction term (decay coefficients 0.02, 0.03): at high frequency, eddy current between strands generates a counter magnetic field, weakening the main magnetic field, resulting in a slight decrease in inductance - when = 5 , the self-inductance decreases by about 4.5% ( ), the mutual-inductance decreases by about 6.7% ( 7), consistent with the measured law of eddy current loss; exponential term (0.3-0.5): reflects the nonlinearity of loss growth - skin effect loss grows slower than square law (because the current has been highly concentrated on the surface), so the exponential is less than 1, determined by fitting 10 groups of measured data at different frequencies (50 kHz-500 kHz).

[0250] In summary, steps 511-513 build a circuit model that accurately reflects high-frequency harmonic loss through "feature parameter extraction-matrix modeling-frequency dynamic correction". Step 511 directly relates self-resistance, mutual inductance, equivalent structure parameters, and magnetic field components, making the parameter set not only mathematical symbols but also quantitative expressions of electromagnetic properties. Step 512's impedance matrix strictly follows the reciprocity theorem, reflecting both independent losses (self-impedance) and bidirectional coupling losses (mutual impedance), making it more accurate than traditional single-port models. Step 513's correction term uses an exponential function to simulate the nonlinear change of impedance with frequency, addressing the error problem (the impedance calculation error decreases from 15% to 3%) of traditional linear correction at high frequencies. )in the high frequency band ( ).

[0251] The final generated harmonic loss calculation circuit model can directly input each harmonic current and output the corresponding loss value, providing a high-precision circuit basis for step 52's harmonic loss calculation.

[0252] Optionally, the step 52 specifically includes:

[0253] Step 521: Analyze the current harmonic features in the multi-physical quantity original feature set, extract the current amplitude, frequency, and phase information of each harmonic, and generate a harmonic current parameter set;

[0254] Step 522: Substitute the harmonic current parameter set into the harmonic loss calculation circuit model, calculate the initial loss value of each harmonic based on the impedance-power relationship, and generate a harmonic loss initial value set;

[0255] Step 523: Based on the mutual impedance parameters in the mutual impedance equivalent circuit model, quantify the influence of electromagnetic coupling between different harmonics on loss, and cross-correct the harmonic loss initial value set to generate the corrected harmonic loss value.

[0256] Preferably, in step 521, the loss of high-frequency harmonics depends not only on amplitude and frequency but also on phase - for example, in-phase 3rd and 5th harmonics will produce additive loss through mutual impedance, while anti-phase will partially cancel out. This step, based on step 421, focuses on extracting phase information to form a complete complex parameter set, with the formula as follows:

[0257]

[0258] Where: : The current phasor of the nth harmonic (unit: A), which is a complex number containing amplitude and phase; : Harmonic current amplitude (unit: A), consistent with step 421, calculated by FFT; : harmonic frequency (unit: Hz), ( is the fundamental frequency); : Harmonic phase (unit: rad), that is, the current phasor and the reference sine wave ( ) is measured by a lock-in amplifier with an accuracy of ±1°; : Effective harmonic order (dimensionless, range 20-50), retaining harmonics with stable phases and amplitudes ≥ 1% of the fundamental. Phase accuracy ±1°: Phase deviation directly affects the calculation of mutual impedance coupling loss (a 1° phase error can result in a coupling loss error of approximately 1.7%). A 16-bit ADC combined with phase-locked loop technology can achieve this accuracy, meeting engineering requirements. Upper limit of harmonic order: This is consistent with step 421 to ensure the consistency of the parameter set and avoid missing higher-order harmonics (for example, harmonics above the 20th order, while small in amplitude, have high frequencies and significant coupling loss). Complex representation: This provides a mathematical foundation for subsequent mutual impedance coupling loss calculations (involving phase angle differences), avoiding the loss of phase information in traditional real-number amplitude calculations.

[0259] Preferably, in step 522, the harmonic loss calculation circuit model is used to substitute each harmonic current into the impedance parameter to obtain the initial loss value. The loss is essentially the active power absorbed by the impedance, and the real part needs to be extracted through complex operations. The formula is as follows:

[0260]

[0261] in, is the impedance corresponding to the nth harmonic (unit: Ω):

[0262]

[0263] in: : Initial loss value of the nth harmonic (unit: W), without considering the coupling between harmonics; : The voltage phasor of the nth harmonic (unit: V), calculated by multiplying the impedance and the current phasor; : The conjugate complex number of ensures that the power calculation conforms to the definition of active power (the real part is active and the imaginary part is reactive); : The real part of the impedance (unit: Ω), that is, the equivalent resistance, including the self-resistance and mutual resistance (the corrected value in step 513); : Angular frequency of the nth harmonic (unit: rad / s). Real impedance extraction: Active power is only related to the real part of the impedance. The imaginary part (reactance) does not consume energy. This is a basic principle of circuit theory and ensures the physical correctness of loss calculations. Impedance parameter reuse: Directly call the corrected value in step 513. 、 , ensure the consistency with the frequency characteristics of the amendment, avoid the error caused by the dislocation of parameters; calculation accuracy: the measured deviation of the initial loss value and the power analyzer measurement result is less than or equal to 3% for 100 kHz fundamental wave and 3rd harmonic (300 kHz), which verifies the accuracy of the formula.

[0264] Preferably, in step 523, the current of different harmonics produces "cross-coupling loss" through mutual impedance, for example, the magnetic field of the 3rd harmonic will induce additional eddy current in the current of the 5th harmonic, which cannot be captured by single harmonic calculation. The formula of this step is as follows:

[0265]

[0266] Wherein: : the loss value of the nth harmonic after correction (unit: W); : the mutual impedance between the nth and the mth harmonic (unit: Ω), which is obtained by interpolation of the mutual impedance parameter in step 513 When ; : the phase angle of the mutual impedance (unit: rad), ; : phase difference correction term, when the phase is matched (cosine value = 1), the coupling loss is maximum, and when the phase is opposite (cosine value = -1), the loss is minimum; summation term: traverse all other harmonics m, accumulate the additional loss generated by cross-coupling. Mutual impedance interpolation: when the phase difference is large (such as , , ), the mutual impedance is calculated by linear interpolation, and the error is less than or equal to 5% (the measured high-frequency mutual impedance changes slowly with frequency); phase difference weight: the cosine function ensures that the coupling loss is positively related to the degree of phase matching, which is a direct embodiment of the law of electromagnetic induction (the same phase magnetic field produces the same phase eddy current, and the loss is superimposed); coupling strength limitation: when , the mutual impedance amplitude decreases significantly , the cross term can be ignored, and the calculation is simplified (reducing 40% of the operation amount).

[0267] In summary, steps 521-523, through "phase accurate extraction-single harmonic loss calculation-multi harmonic coupling correction", incorporate phase information into harmonic characterization with the complex parameter set of step 521, providing complete mathematical input for subsequent coupling loss calculation (phase error reduced from ±5° to ±1°, making coupling loss calculation accuracy improved by 8%); impedance real part extraction of step 522, based on the active power definition of circuit theory, ensures the physical consistency of the initial loss value with the impedance model, avoiding the approximation error of "direct multiplication of voltage and current amplitude"; the cross correction term of step 523 quantifies the electromagnetic coupling strength between harmonics through the cosine function of phase difference, making the loss calculation cover the full coupling scenario of "fundamental-harmonic" and "harmonic-harmonic". On a 100kHz high-frequency transformer, the total loss calculation error is reduced from 15% of the traditional method to 4.2%.

[0268] Preferably, the step 53 specifically comprises:

[0269] Step 531: Perform frequency characteristic matching verification on the modified harmonic loss values and the basic loss value to generate a loss data set to be superimposed;

[0270] Step 532: Based on the energy proportion of each loss value in the loss data set to be superimposed, use a weighted superimposition algorithm to perform loss aggregation to generate a total loss estimate;

[0271] Step 533: Perform error compensation on the total loss estimate to generate a predicted Litz wire loss value.

[0272] Preferably, in step 531, the modified harmonic loss value (step 523) and the basic loss value (step 433) may have a frequency range mismatch problem - for example, the basic loss may mainly cover the low frequency band (<50kHz), while the loss of high-order harmonics (>100kHz) is only reflected in harmonic correction. This step verifies the consistency of the two through a frequency response function, ensuring that the superimposed loss has a unified physical meaning in the full frequency band, and the formula is as follows:

[0273]

[0274] Where: : Loss matching degree at frequency f (dimensionless), reflecting the deviation of the basic loss and the harmonic loss at that frequency; : Component of the basic loss value at frequency f (unit: W), extracted from the spatial integral result through inverse Fourier transform; : Superimposed value of all harmonic losses at frequency f (unit: W), i.e. ; : Allowable deviation threshold (dimensionless, take 0.1), i.e. deviation ≤10%, to ensure that the loss descriptions of the two at the same frequency are consistent. Constant and parameter value selection basis: deviation threshold =0.1: Based on experimental data, the inherent error of the fundamental loss and the harmonic loss (such as spatial integration approximation, harmonic coupling simplification) is about 5%-8%, setting a threshold of 10% is strict and has room for error, avoiding excessive correction; frequency component extraction: convert the spatial distribution of the fundamental loss into a frequency distribution through inverse Fourier transform, ensuring that it can correspond to the frequency points of the harmonic loss one by one (such as key frequencies such as 50 kHz, 100 kHz must be matched); mismatch processing: if , the high-frequency components of the fundamental loss are corrected by interpolation (reference the frequency trend of the harmonic loss), ensuring the coherence of the data set to be superimposed.

[0275] Preferably, in step 532, the contributions of different frequencies to the total loss are different, and the energy distribution is uneven - for example, the fundamental (50 kHz) loss may account for 60%, while the 3rd harmonic (150 kHz) has a small amplitude, but due to the frequency squared relationship, the loss accounts for up to 20%. This step is based on the energy proportion of each loss to avoid overestimation of low-frequency loss caused by simple linear superposition, and the formula is as follows:

[0276] Wherein, the weight function is:

[0277]

[0278] Wherein: : Total loss estimate (unit: W); F: Full frequency set (unit: Hz), covering the fundamental and all effective harmonics; : Weight coefficient at frequency f (dimensionless), proportional to the frequency squared and magnetic field energy density integral; : Magnetic field strength squared at frequency f and axial position z (unit: (A / m)²), reflecting the magnetic field energy distribution at that frequency; numerator and denominator: normalize the weight by magnetic field energy to ensure that the loss in high-frequency strong magnetic field regions is given higher weight. Constant and parameter value: weight proportional to : Loss is essentially the dissipation of magnetic field energy, and the magnetic field energy is proportional to the square of the frequency ( ), which conforms to the law of electromagnetic energy conversion; magnetic field integration range: integrate along the winding length L to ensure that the weight reflects the magnetic field energy distribution in the whole space (such as the high-frequency magnetic field energy near the air gap of the magnetic core is higher, and the weight of the corresponding frequency is greater); weight normalization: the sum of the weights of all frequencies is 1, ensuring that the total loss estimate has the correct dimension (unit: W) and physical meaning as "the sum of energy dissipation in the full frequency band".

[0279] Preferably, in step 533, the total loss estimate may have systematic errors (such as a low theoretical value due to model simplification or inherent deviation of the measurement equipment), and needs to be corrected by fitting the compensation coefficient of experimental data. The formula is as follows:

[0280]

[0281] in: : Predicted litz wire loss value (unit: W), which is the final output result; : High frequency compensation coefficient (dimensionless, take 0.05), correct the high frequency band ( ) Underestimation of losses due to model simplification; : Maximum harmonic frequency (unit: Hz); : Reference frequency (unit: Hz, 100kHz); : System deviation constant (dimensionless, 0.03), based on the average deviation of historical data statistics (for example, the reading of the measuring device is 3% lower than the actual value). High frequency compensation coefficient :Measurements show that when (5 times the reference frequency), the simplification of the skin effect in the model can reduce the loss; the system deviation =0.03: Calibration of 10 high-frequency transformers of different specifications found that the theoretically calculated values ​​were on average 3% lower than the measured values ​​(due to ignoring the additional loss of the oxide layer on the surface of the strands), so a fixed compensation of 3% was added; Exponential attenuation: The model error in the high-frequency band decreases with increasing frequency (due to the small amplitude of high-order harmonics and limited impact). The exponential function can accurately simulate this attenuation trend and avoid overcompensation.

[0282] In summary, steps 531-533 are implemented through "consistency check-energy weighting-dynamic compensation". In the frequency matching of step 531, the consistency of the basic loss and the harmonic loss is verified through the frequency response function to ensure that the superimposed losses have unified physical meanings in the entire frequency band, avoiding "data disconnection" between low-frequency losses and high-frequency losses; in the energy weighting of step 532, the weight is set based on the magnetic field energy distribution, so that the loss contribution of the high-frequency strong magnetic field area is fully reflected, and the accuracy is improved by 12% compared with the traditional equal-weighted superposition (verified on a 100kHz transformer, the deviation between the total loss calculation and the actual measurement is reduced from 8% to 3%); in the dynamic compensation of step 533, the high-frequency model error and the inherent deviation of the system are distinguished, and "weak compensation at high frequency and strong compensation at low frequency" are achieved through the exponential function, which solves the adaptability problem of fixed coefficient compensation in a wide frequency range.

[0283] To this end, the final predicted loss value can not only reflect the actual loss characteristics of the Litz wire, but also directly serve engineering design (such as heat dissipation solution optimization and efficiency evaluation), realizing a closed loop from theoretical calculation to practical application.

[0284] The above descriptions are only the preferred embodiments of the present application, and are not intended to limit the present application. The present application can have various modifications and changes for those skilled in the art. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method of predicting the loss of a high-frequency transformer's Litz wire, characterized by, The method comprises: Step 1: collecting electromagnetic physical quantities, structural physical quantities and thermal physical quantities of the high-frequency transformer to generate a multi-physical quantity original feature set; Step 2: performing Litz wire equivalent copper foil conversion based on the multi-physical quantity original feature set to obtain Litz wire equivalent structure parameters, including equivalent thickness, equivalent conductivity and filling factor; Step 3: determining the three-dimensional magnetic field distribution of the Litz wire based on the Litz wire equivalent structure parameters, and decomposing the three-dimensional magnetic field distribution into a magnetic field component parallel to the strand and a magnetic field component perpendicular to the strand by combining the Litz wire twist angle to generate a magnetic field component representation result; Step 4: calculating the unit length loss of the Litz wire based on the equivalent structure parameters of the Litz wire and the corresponding current harmonic characteristics according to the magnetic field component representation result to determine the basic loss value; Step 5: superimposing each harmonic loss value on the basic loss value through the mutual impedance equivalent circuit of the Litz wire to obtain the predicted Litz wire loss value.

2. The method of claim 1, wherein, The step 1 specifically comprises: Step 11: determining the specific range of electromagnetic physical quantities, structural physical quantities and thermal physical quantities to be collected according to the influencing factors of the Litz wire loss of the high-frequency transformer to generate a physical quantity collection list; Step 12: collecting data of each physical quantity by using multi-sensor synchronous collection technology according to the physical quantity collection list to generate a multi-dimensional original data set; Step 13: performing standardization processing on the multi-dimensional original data set to generate a multi-physical quantity original feature set.

3. The method of claim 1, wherein, The step 2 specifically comprises: Step 21: determining the geometric equivalence and physical property conservation principle of the Litz wire to equivalent copper foil conversion based on the Litz wire structure parameters in the multi-physical quantity original feature set to generate an equivalent conversion criterion; Step 22: performing equivalent calculation on the thickness, conductivity and filling rate of the Litz wire according to the equivalent conversion criterion to generate equivalent structure parameter initial values; Step 23: performing twisting effect correction on the equivalent structure parameter initial values to generate Litz wire equivalent structure parameters.

4. The method of claim 3, wherein, The step 21 specifically comprises: Step 211: extracting core structure parameters affecting equivalent conversion based on the Litz wire single strand diameter, total strand count and twisting pitch in the multi-physical quantity original feature set to generate a Litz wire structure characteristic parameter set; Step 212: analyzing the space occupation law when multiple strands are twisted according to the Litz wire structure characteristic parameter set to generate a geometric equivalence sub-criterion meeting the principle of equal conductive cross-sectional area of equivalent copper foil and Litz wire; Step 213: generating the equivalent conversion criterion based on the geometric equivalence sub-criterion combined with the physical constraint of constant direct current resistance.

5. The method of claim 1, wherein, The step 3 specifically comprises: Step 31: constructing a three-dimensional electromagnetic simulation model of the high-frequency transformer based on the Litz wire equivalent structure parameters and combining the measured data of key points to generate three-dimensional magnetic field distribution data around the Litz wire; Step 32: calculating the twist angle of the Litz wire strand relative to the wire bundle according to the three-dimensional magnetic field distribution data and the Litz wire twisting parameters in the multi-physical quantity original feature set to generate a twist angle parameter; Step 33: decomposing the magnetic field into components parallel to the strand and perpendicular to the strand based on the three-dimensional magnetic field distribution data and the twist angle parameter to generate a magnetic field component representation result.

6. The method of claim 5, wherein, The step 31 specifically comprises: Step 311: Based on the equivalent structure parameters of the Litz wire, combined with the magnetic core material properties and winding arrangement information, a high-frequency transformer three-dimensional electromagnetic simulation model is constructed to restore the winding arrangement and the magnetic core air gap distribution, so as to generate initial three-dimensional magnetic field simulation data; Step 312: Magnetic field probes are arranged in the magnetic field distortion significant area including the winding end and the magnetic core air gap edge, and the magnetic field strength measured value of the key point is collected to generate the magnetic field measured data set; Step 313: The least square method is used to fit the error of the initial three-dimensional magnetic field simulation data and the magnetic field measured data set, and the magnetic core permeability and winding equivalent conductivity parameters in the simulation model are dynamically adjusted to generate three-dimensional magnetic field distribution data.

7. The method of claim 6, wherein, The step 32 specifically comprises: Step 321: The twisting pitch and wire bundle outer diameter parameters of the Litz wire are extracted from the multi-physical quantity original feature set, and geometric correlation analysis is performed to generate Litz wire space twisting characteristic parameters; Step 322: Based on the spiral motion trajectory of the strand wire winding bundle axis, combined with the Litz wire space twisting characteristic parameters, the spatial angle calculation formula of the strand wire center axis and the wire bundle axis is derived to generate a twist angle calculation model; Step 323: The axial position information in the three-dimensional magnetic field distribution data and the Litz wire space twisting characteristic parameters are substituted into the twist angle calculation model to calculate the twist angle value of the strand wire at different axial positions, so as to generate the twist angle parameter.

8. The method of claim 7, wherein, The step 33 specifically comprises: Step 331: According to the twist angle parameter, a strand wire local coordinate system is established, and the three-dimensional magnetic field distribution data is converted to generate local magnetic field data suitable for the strand wire posture; Step 332: Based on the strand wire local coordinate system, the magnetic field intensity vector in the local magnetic field data is projected onto the longitudinal axis and the radial direction, and the magnetic field components parallel and perpendicular to the strand wire are extracted respectively to generate magnetic field component decomposition results; Step 333, the magnetic field component decomposition results are subjected to spatial interpolation processing to realize point-by-point matching with the three-dimensional space distribution of the Litz wire strand wire, so as to generate magnetic field component representation results.

9. The method of claim 1, wherein, The step 4 specifically comprises: Step 41: Based on the magnetic field component representation results and the equivalent structure parameters of the Litz wire, the loss composition of the skin effect and the proximity effect in the unit length is analyzed to generate a loss composition model; Step 42: According to the loss composition model and the current harmonic characteristics in the multi-physical quantity original feature set, the skin effect loss and the proximity effect loss in the unit length are calculated to generate unit length loss data; Step 43: The unit length loss data is subjected to spatial integral processing to generate a basic loss value.

10. The method of claim 1, wherein, The step 5 specifically comprises: Step 51: Based on the equivalent structure parameters of the Litz wire and the magnetic field component representation results, an equivalent circuit model including self-impedance and mutual-impedance is constructed to generate a harmonic loss calculation circuit model; Step 52: According to the harmonic loss calculation circuit model and the current harmonic characteristics, the loss value of each harmonic is calculated and the coupling effect is corrected through the mutual-impedance parameters to generate the corrected loss value of each harmonic. Step 53: superimposing the modified harmonic loss values and the base loss value to generate a predicted litz wire loss value.

Images