Permanent magnet wind power generator inductance parameter identification method based on JA model and data fusion

Abstract

The application discloses a permanent magnet wind generator inductance parameter identification method based on JA model and data fusion, comprising the following steps: decomposing the magnetization intensity of ferromagnetic material into reversible component and irreversible component, deducing the nonlinear relationship between inductance and current in combination with Ampere loop law, introducing cross saturation coefficient to correct the inter-axis coupling effect, and establishing an inductance basic model under magnetic saturation state; integrating the influences of current, temperature and frequency to form a composite inductance model; establishing a state equation based on a voltage equation and a flux linkage equation; realizing dynamic prediction and update of state and parameters through extended Kalman filtering, combining a model reference adaptive system to drive parameter correction by using flux linkage error, and finally adjusting weights based on disturbance error of rotational inertia and friction coefficient to realize fusion identification. Through physical mechanism modeling and multi-field coupling correction, the application realizes high-precision tracking of inductance parameters under wide working conditions, and significantly improves the identification precision and generalization ability in the magnetic saturation region.

Description

Technical Field

[0001] This invention relates to the field of wind power, and in particular to a method for identifying inductance parameters of permanent magnet wind turbines based on JA model and data fusion. Background Technology

[0002] Permanent magnet wind generators (PMSGs), with their high efficiency, high power density, zero excitation loss, and excellent low-speed, high-power generation capability, have become the core power generation unit in modern wind power systems, especially in direct-drive and semi-direct-drive wind turbines, and are widely used in onshore and offshore wind farms. During wind power grid-connected operation, the control precision of the power generation system directly determines the overall energy conversion performance and grid adaptability.

[0003] Inductance parameters, as key indicators of the magnetic circuit characteristics of permanent magnet wind turbines, directly affect not only the design and dynamic performance of current regulators in vector control or direct torque control (DTC), but also determine the field weakening capability, grid-connected power regulation accuracy, and the suppression effect of electromagnetic torque and power ripple. Under conditions of high wind speed, wide speed range, and significant magnetic saturation, deviations from actual inductance parameters will lead to power control errors, reduced system stability margins, and even grid-connected current distortion. Therefore, accurate modeling, online identification, and adaptive updating of inductance parameters have become crucial foundations for achieving high-performance, high-reliability wind power generation control systems.

[0004] However, in actual operation, the inductance parameters of permanent magnet generators exhibit significant nonlinear characteristics, mainly affected by the following factors: 1. Magnetic saturation effect: When the motor load increases or enters the weak magnetic region, the stator core magnetic circuit gradually saturates, leading to... shaft and 1. **Cyclical Inductance:** The inductance decreases nonlinearly with increasing current. This nonlinear relationship is particularly pronounced in the deep saturation region and is difficult to accurately describe using traditional linear models or polynomial fitting methods. 2. **Temperature Changes:** During motor operation, the winding resistance and core permeability change significantly with temperature. For example, for every 100°C increase in temperature, the copper winding resistance can increase by approximately 40%, while the remanence and coercivity of the permanent magnet decrease, further affecting inductance characteristics. 3. **Frequency Response:** During high-frequency operation, the skin effect of the stator current and eddy current losses in the core intensify, causing the inductance value to decrease with increasing frequency. This frequency dependence is particularly prominent in high-speed motors or high-frequency modulation systems. 4. **Cross-Saturation Phenomenon:** shaft and The mutual coupling between the axial magnetic circuits makes Changes in shaft current will affect The inductance along the shaft, and vice versa, further complicates the inductance parameters. The coupling effect of these factors makes inductance parameter identification a key challenge in the field of permanent magnet wind turbine control.

[0005] Traditional parameter identification methods can be mainly divided into three categories. The first category is based on physical models: by establishing the motor magnetic circuit equations, the relationship between parameters and current and voltage is derived using finite element analysis or analytical calculation. This type of method has clear physical meaning, but the modeling process is complex and it is difficult to reflect parameter changes under dynamic operating conditions in real time. The second category is based on signal injection: a test signal of a specific frequency is injected into the motor, and parameter information is extracted by analyzing the response. This method is simple to operate, but it is susceptible to noise interference, and the injected signal may affect the normal operation of the motor. The third category is based on data-driven methods: machine learning algorithms (such as neural networks and Kalman filters) are used to learn the parameter change patterns from a large amount of operating data. This type of method is highly adaptable, but it relies on sufficient training samples and lacks physical interpretability, thus limiting its generalization ability.

[0006] In recent years, with the development of industrial and new energy technologies, the operating conditions of motors have become increasingly complex, placing higher demands on parameter identification technology. These demands include: wide operating condition adaptability (covering the entire operating range from no-load to full-load, low temperature to high temperature, and low frequency to high frequency); real-time performance and robustness (maintaining high accuracy even under dynamic load changes and measurement noise); physical interpretability (the model needs to reflect the inherent physical mechanism of parameter changes and support operating condition extrapolation and fault diagnosis); and engineering practicality (low algorithm complexity, easy to implement in embedded systems, and reasonable hardware resource requirements).

[0007] Existing technologies have significant shortcomings in addressing the above challenges, specifically: 1. Insufficient nonlinear modeling: Most methods only consider a single nonlinear factor, such as magnetic saturation or temperature, without establishing a multi-physics coupling model; 2. Weak dynamic tracking capability: Traditional offline calibration or static parameter tables are difficult to adapt to real-time changing operating conditions; 3. Conflict between robustness and accuracy: Data-driven methods may overfit under complex operating conditions, while physical model methods are sensitive to modeling errors; 4. Low computational efficiency: High-precision physical models require a large amount of computation, which cannot meet the needs of real-time control. Summary of the Invention

[0008] To address the aforementioned technical problems, this invention provides a simple and highly accurate method for identifying inductance parameters of permanent magnet wind turbines based on JA model and data fusion.

[0009] The technical solution of this invention to solve the above-mentioned technical problems is: a method for identifying inductance parameters of permanent magnet wind turbines based on JA model and data fusion, comprising the following steps:

[0010] Step 1: Initialize the JA model parameters, decompose the magnetization of the ferromagnetic material into reversible and irreversible components, describe the irreversible magnetization process through the domain motion equation, derive the nonlinear relationship between inductance and current by combining Ampere's circuital law, introduce the cross saturation coefficient to correct the interaxial coupling effect, and establish the basic model of inductance under magnetic saturation.

[0011] Step 2: Construct a dynamic correction model for inductor parameters based on temperature and frequency. The temperature correction model covers changes in stator resistance, core permeability, and JA model parameters. The frequency correction model considers inductor attenuation caused by eddy current and skin effects. A high-frequency inductor correction model is constructed to quantify the skin effect. The effects of current, temperature, and frequency are integrated to form a composite inductor model.

[0012] Step 3: Construct an enhanced state vector containing electrical quantities and parameters to be identified, and establish state equations based on voltage equations and flux linkage equations; achieve dynamic prediction and updating of state and parameters through extended Kalman filtering, combine with model reference adaptive system to drive parameter correction using flux linkage error, and finally adjust weights based on disturbance errors of rotational inertia and friction coefficient to achieve fusion identification.

[0013] The above-mentioned method for identifying inductance parameters of permanent magnet wind turbines based on JA model and data fusion, in step one, decomposes the magnetization intensity of ferromagnetic materials into reversible and irreversible components, and describes the irreversible magnetization process through the magnetic domain motion equation as follows:

[0014] Magnetization of ferromagnetic materials From reversible components With irreversible components composition, The irreversible component describes the pinning and motion of the magnetic domain walls, following the core differential equation:

[0015]

[0016] in, The magnetic field strength, This is the magnetic field direction coefficient. This is the hysteresis loss coefficient. The inter-domain coupling coefficient, The hysteresis-free magnetization is described by the Langevin function:

[0017]

[0018] in, The saturation magnetization is For shape distribution parameters, It is a hyperbolic cotangent function. For effective magnetic field strength, ;

[0019] The relationship between reversible and irreversible components is as follows:

[0020]

[0021] in, is the reversibility coefficient.

[0022] The above-mentioned method for identifying inductor parameters of permanent magnet wind turbines based on JA model and data fusion, specifically the process of establishing the basic inductor model under magnetic saturation state in step one, is as follows:

[0023] In the nonlinear mapping between inductance and current, combined with Ampere's circuital law, Axial magnetic field strength and Axial magnetic field strength They are respectively: ,in, The number of turns in the winding. for shaft current, for shaft current, for Equivalent length of the axial magnetic circuit for Equivalent length of the axial magnetic circuit;

[0024] Inductance is defined as the ratio of magnetic flux linkage to current. shaft flux linkage is , , , is the magnetic flux density , Magnetization Axial components, for Cross-sectional area of ​​the magnetic circuit of the shaft. Given the free permeability, it is derived from this. Shaft inductor , Shaft inductor The expression is:

[0025]

[0026]

[0027] in, for Initial value of shaft inductance, , for Initial value of shaft inductance, , , All are cross-saturation coefficients, and are corrected separately. Axis current pair Shaft inductance, Axis current pair The suppression effect of shaft inductance;

[0028] When the current enters the saturation region, that is , The saturation magnetization of the magnetic material. The shaft inductance is approximately:

[0029]

[0030] Shaft inductance and The principle of axial inductance is the same.

[0031] The above-mentioned method for identifying inductor parameters of permanent magnet wind turbines based on JA model and data fusion, in the temperature correction model of step two, the temperature characteristics of the stator resistance follow the resistance-temperature law of metallic conductors:

[0032]

[0033] in, It is a linear function of temperature and stator resistance. The stator resistance at 25℃ The temperature coefficient of copper, Real-time winding temperature;

[0034] Deducing the temperature rise from the resistance deviation :

[0035]

[0036] in, For real-time measurement of the stator winding resistance;

[0037] The magnetic permeability of the iron core decreases linearly with increasing temperature, and the fitting relationship is as follows:

[0038]

[0039] in, It is a linear function of the temperature change and the magnetic permeability change. The permeability at 25℃ The temperature decay coefficient of the relative permeability of the iron core;

[0040] JA model parameters are modulated by temperature:

[0041]

[0042]

[0043] in, With rising temperature The varying hysteresis loss coefficient This is a linear function of temperature change and the saturation magnetization of the magnetic material. The saturation magnetization of the magnetic material at 25℃. The temperature decay coefficient of saturation magnetization. This represents the temperature increase coefficient of the hysteresis coefficient.

[0044] The above-mentioned method for identifying inductor parameters of permanent magnet wind turbines based on JA model and data fusion, in the frequency correction model of step two, based on Maxwell's equations, the eddy current diffusion equation is:

[0045]

[0046] in, magnetic field strength The Laplace operator describes the spatial second-order rate of change of the magnetic field strength; The permeability of a permanent magnet. For the electrical conductivity of the material, For time; due to sinusoidal excitation , Angular frequency, As the initial value for the excitation, The imaginary unit is used; therefore, the eddy current diffusion equation simplifies to The magnetic field distribution in the planar iron core is as follows:

[0047]

[0048] in, The strength of the spatial magnetic field. As the reference magnetic field strength, It is the distance from the geometric center of the iron core. This is the high-frequency attenuation coefficient. For the thickness of the iron core, , For the depth of the skin effect, ; It is a hyperbolic cosine function;

[0049] Magnetic permeability exists in a complex form with a real part and an imaginary part under a high-frequency alternating magnetic field. Considering the influence of eddy currents, the stored energy component of the real part of the magnetic permeability is defined as:

[0050]

[0051] in, The real part of the permeability is the energy storage component. It is the hyperbolic tangent function.

[0052] The above-mentioned method for identifying inductor parameters of permanent magnet wind turbines based on JA model and data fusion, in step two, the high-frequency inductor correction model is as follows:

[0053]

[0054] in, Let frequency be the inductor frequency with respect to frequency. Inductance value at 50Hz For input frequency, For material constants, ;

[0055] Integrating the effects of current, temperature, and frequency coupling including eddy current and skin effects, the final inductance model for:

[0056]

[0057] in, , The magnetic saturation coefficient is temperature-dependent. The reference magnetic saturation coefficient at room temperature. This is the temperature-corrected saturation coefficient; The saturation current threshold; , The temperature-dependent cross-saturation coefficient. The initial cross-saturation coefficient, This is the initial static inductance.

[0058] The above-mentioned method for identifying inductance parameters of permanent magnet wind turbines based on JA model and data fusion, in step three, involves constructing an enhanced state vector containing electrical quantities and parameters to be identified, and establishing the state equation based on the voltage equation and flux linkage equation as follows:

[0059] First, construct the enhanced state vector. The enhanced state vector includes measurable electrical quantities, related environmental parameters, and parameters to be identified:

[0060]

[0061] Then, the state equations are established by combining the voltage equation and the flux linkage equation:

[0062]

[0063]

[0064] in, for Shaft inductor model, The mutual inductance parameter is a function of temperature change. for shaft voltage, for shaft voltage, Electric angular velocity, ; for Axial magnetic flux; For permanent magnet flux linkage;

[0065] Finally, by differentiating the flux linkage equation and substituting it into the voltage equation, we obtain the current dynamic equation:

[0066]

[0067] in, This is the static value of the stator resistance.

[0068] The above-mentioned method for identifying inductance parameters of permanent magnet wind turbines based on JA model and data fusion, in step three, the process of dynamically predicting and updating the state and parameters through extended Kalman filtering includes an extended Kalman filter (EKF) prediction stage and an update stage.

[0069] The prediction stage of the Extended Kalman Filter (EKF) is as follows:

[0070] First, estimate the current state based on the state at the previous time step:

[0071]

[0072] in, for Prior state estimation at time 10:00 for Post-hoc state estimation at time step Sampling time, To control the input vector, This is the state transition function;

[0073] Prediction error covariance matrix:

[0074]

[0075] in, for Prior covariance matrix at time step for The posterior covariance matrix at time step; For Jacobian matrices, for transpose, Key elements yes The first component; For process noise covariance;

[0076] The update phase involves correcting the predicted value using current measurements.

[0077]

[0078]

[0079]

[0080] in, It is the Kalman gain matrix of EKF. This is the actual measured current vector. This is the measured current value; The observation matrix; for Transpose of; To measure the noise covariance, For the state update equation, the state estimate is corrected by measuring the residuals. , so that the state estimate Approaching the true value; The covariance update equation serves to quantify the uncertainty of state estimation and ensure the stability of the identification process.

[0081] The above-mentioned method for identifying inductance parameters of permanent magnet wind turbines based on JA model and data fusion, in step three, involves using flux linkage error to drive parameter correction in conjunction with the model reference adaptive system as follows:

[0082] The Model Reference Adaptive System (MRAS) uses the flux error between the reference model and the adjustable model to design an adaptive law to correct the parameters to be identified, and uses the voltage equation as a transformation to derive the flux calculation formula.

[0083] The reference model is:

[0084]

[0085]

[0086] in, for Axis reference flux linkage for Axis reference flux linkage Indicates time Differentiate;

[0087] The adjustable model is:

[0088]

[0089]

[0090] in, for Axis estimation flux linkage for Axis estimation flux linkage;

[0091] Adaptive law: Definition of flux linkage error:

[0092]

[0093]

[0094] in, for Shaft flux error, for Shaft flux linkage error.

[0095] The above-mentioned method for identifying inductance parameters of permanent magnet wind turbines based on JA model and data fusion, in step three, involves adjusting the weights based on the disturbance error of rotational inertia and friction coefficient to achieve fusion identification.

[0096] First, construct the Lyapunov function:

[0097]

[0098] in, This is a Lyapunov function, which is used to construct a function to determine the stability of a system. The ellipsis represents the estimation error of the cross saturation coefficient, and the ellipsis represents the estimation error of other parameters to be corrected. , All are positive adaptive gains; The inductance estimation error equation is as follows: This is an estimated value for the inductance. ;make That is, the Lyapunov function converges, yielding the adaptive law:

[0099]

[0100] in, for The derivative of yes Time derivative, Estimated value of cross saturation coefficient The time derivative;

[0101] The identification results obtained by the MRAS algorithm are , ;

[0102] Then, a dynamic fusion strategy is adopted: based on rotational inertia. Adjustment weight for disturbance error with friction coefficient B:

[0103]

[0104] , All are nominal mechanical parameters; Moment of inertia The disturbance error; The disturbance error is the friction coefficient B. This is the actual moment of inertia. This is the actual coefficient of friction;

[0105] Finally, the fusion weights are obtained. :

[0106]

[0107] All are sensitivity coefficients;

[0108] Final inductance estimate for:

[0109]

[0110] in, These are the identification values ​​obtained using the EKF algorithm.

[0111] The beneficial effects of this invention are as follows:

[0112] 1. This invention proposes a magnetic saturation parameter identification framework for permanent magnet wind turbines based on the Jiles-Atherton model. Through physical mechanism modeling and multi-field coupling correction, it achieves high-precision tracking of inductance parameters under a wide range of operating conditions, significantly improving the identification accuracy and generalization ability of the magnetic saturation region. It is applicable to direct-drive and semi-direct-drive wind turbines.

[0113] 2. This invention introduces a dynamic temperature and frequency correction mechanism to construct a composite inductance model that includes resistance, permeability, and skin effect, effectively solving the problem of insufficient consideration of environmental factors in traditional methods and enhancing the environmental adaptability of parameter identification.

[0114] 3. This invention designs a fusion algorithm of extended Kalman filter and model reference adaptive system, and combines the disturbance of rotational inertia and friction coefficient to adjust the weight, thereby improving the identification robustness while ensuring dynamic tracking speed and balancing the contradiction between accuracy and stability. Attached Figure Description

[0115] Figure 1 This is the overall flowchart of the present invention.

[0116] Figure 2 A comparison chart of inductor identification results under different algorithms. Detailed Implementation

[0117] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0118] like Figure 1 As shown, the method for identifying inductance parameters of permanent magnet wind turbines based on JA model and data fusion includes the following steps:

[0119] Step 1: Initialize the JA model parameters, decompose the magnetization of the ferromagnetic material into reversible and irreversible components, describe the irreversible magnetization process through the domain motion equation, derive the nonlinear relationship between inductance and current by combining Ampere's circuital law, introduce the cross saturation coefficient to correct the interaxial coupling effect, and establish the basic model of inductance under magnetic saturation.

[0120] When a permanent magnet wind turbine operates under high load or weak magnetic field conditions, the stator core magnetic circuit may saturate due to increased magnetic flux density, leading to... The inductance of a ferromagnetic material decreases nonlinearly with increasing current. This nonlinearity stems from the domain motion characteristics of ferromagnetic materials—when the external magnetic field intensifies, the domains align themselves along the direction of the magnetic field until saturation magnetization occurs. During this process, the change in inductance is directly related to the magnetization intensity. Therefore, the magnetization intensity of ferromagnetic materials can be decomposed into reversible and irreversible components. The specific process of irreversible magnetization, described by the domain motion equation, is as follows:

[0121] Magnetization of ferromagnetic materials From reversible components With irreversible components composition, The irreversible component describes the pinning and motion of the magnetic domain walls, following the core differential equation:

[0122]

[0123] in, The magnetic field strength, is the magnetic field direction coefficient, which is 1 when the magnetic field strengthens and -1 when the magnetic field weakens; It is the hysteresis loss coefficient, reflecting the energy dissipation of magnetic domain wall motion; is the interdomain coupling coefficient, characterizing the strength of the interaction between magnetic domains; The hysteresis-free magnetization is described by the Langevin function:

[0124]

[0125] in, Saturation magnetization is the upper limit of magnetization for ferromagnetic materials. When the external magnetic field is strong enough, the magnetic domains are completely aligned along the direction of the magnetic field, and the magnetization intensity will not increase after reaching this value. The shape distribution parameter is a coefficient that characterizes the domain size distribution or the effective scale of the magnetic field in a ferromagnetic material and is used to fit the magnetization curve of the actual material. The hyperbolic cotangent function describes the nonlinear growth trend of magnetization with respect to the effective magnetic field strength. For effective magnetic field strength, ;

[0126] The relationship between reversible and irreversible components is as follows:

[0127]

[0128] in, The invertibility coefficient is... The smaller the value, the higher the proportion of irreversible magnetization and the more significant the hysteresis effect.

[0129] The specific process of establishing the basic model of an inductor under magnetic saturation is as follows:

[0130] In the nonlinear mapping between inductance and current, combined with Ampere's circuital law, Axial magnetic field strength and Axial magnetic field strength They are respectively: ,in, The number of turns in the winding. for shaft current, for shaft current, for Equivalent length of the axial magnetic circuit for Equivalent length of the axial magnetic circuit;

[0131] Inductance is defined as the ratio of magnetic flux linkage to current. shaft flux linkage is , , , is the magnetic flux density , Magnetization Axial components, for Cross-sectional area of ​​the magnetic circuit of the shaft. Given the free permeability, it is derived from this. Shaft inductor , Shaft inductor The expression is:

[0132]

[0133]

[0134] in, for Initial value of shaft inductance, , for Initial value of shaft inductance, , , All are cross-saturation coefficients, and are corrected separately. Axis current pair Shaft inductance, Axis current pair The suppression effect of shaft inductance;

[0135] When the current enters the saturation region, that is , The saturation magnetization of the magnetic material. The shaft inductance is approximately:

[0136]

[0137] The above equation clearly reflects the characteristic that magnetic saturation causes the inductance to decrease nonlinearly as the current increases; Shaft inductance and The principle of axial inductance is similar, the only difference is that... Considering the magnetic circuit dominated by armature reaction, the shaft has relatively low magnetic reluctance, leading to... The initial value of the shaft inductance should be greater than axis.

[0138] Step 2: Construct a dynamic correction model for inductor parameters based on temperature and frequency. The temperature correction model covers changes in stator resistance, core permeability, and JA model parameters. The frequency correction model considers inductor attenuation caused by eddy current and skin effects. A high-frequency inductor correction model is constructed to quantify the skin effect. The effects of current, temperature, and frequency are integrated to form a composite inductor model.

[0139] In the temperature correction model, the temperature characteristics of the stator resistance follow the resistance-temperature law for metallic conductors:

[0140]

[0141] in, It is a linear function of temperature and stator resistance. The stator resistance at 25℃ The temperature coefficient of copper, ; Real-time winding temperature;

[0142] Because direct measurement of winding temperature is difficult, the temperature rise is derived from the resistance deviation. :

[0143]

[0144] in, The real-time measured resistance of the stator winding is obtained by measuring the ratio of DC voltage and current across the winding using the ohm method. The calculation error is <2% within the range of -20℃ to 150℃.

[0145] The magnetic permeability of the iron core decreases linearly with increasing temperature, and the fitting relationship is as follows:

[0146]

[0147] in, It is a linear function of the temperature change and the magnetic permeability change. Under normal conditions, the magnetic permeability of the same material has a fixed linear relationship with the temperature change. The permeability at 25℃ The temperature decay coefficient of the relative permeability of the iron core. That is, for every 1°C increase in temperature, the magnetic permeability decreases by 0.12%;

[0148] JA model parameters are modulated by temperature:

[0149]

[0150]

[0151] in, With rising temperature The varying hysteresis loss coefficient This is a linear function of temperature change and the saturation magnetization of the magnetic material. The saturation magnetization of the magnetic material at 25℃. The temperature decay coefficient of saturation magnetization. ; This represents the temperature increase coefficient of the hysteresis coefficient.

[0152] In the frequency correction model, during high-frequency operation, the eddy current effect and skin effect cause inductance attenuation. Based on Maxwell's equations, the eddy current diffusion equation is:

[0153]

[0154] in, magnetic field strength The Laplace operator describes the spatial second-order rate of change of the magnetic field strength; The permeability of a permanent magnet. For the electrical conductivity of the material, For time; due to sinusoidal excitation , Angular frequency, As the initial value for the excitation, The imaginary unit is used; therefore, the eddy current diffusion equation simplifies to The magnetic field distribution in the planar iron core is as follows:

[0155]

[0156] in, The strength of the spatial magnetic field. As the reference magnetic field strength, It is the distance from the geometric center of the iron core. For complex wave number, For the thickness of the iron core, , For the depth of the skin effect, That is, the magnetic field is concentrated on the surface, the effective magnetic flux decreases, and the inductance decreases. This explains the mechanism by which the skin effect indirectly affects the inductance by influencing the magnetic flux. It is a hyperbolic cosine function;

[0157] Magnetic permeability exists in a complex form with a real part and an imaginary part under a high-frequency alternating magnetic field. Considering the influence of eddy currents, the stored energy component of the real part of the magnetic permeability is defined as:

[0158]

[0159] in, The real part of the permeability is the energy storage component. It is the hyperbolic tangent function.

[0160] The physical meaning of the above formula lies in calculating the normalized energy storage permeability. At extremely low frequencies, the influence of eddy currents is negligible, and the permeability approaches that of a vacuum. ,Right now Follow The inductance decreases inversely as it increases, directly leading to inductance decay.

[0161] The high-frequency inductor correction model is as follows:

[0162]

[0163] in, Let frequency be the inductor frequency with respect to frequency. Inductance value at 50Hz For input frequency, For material constants, ;

[0164] Integrating the effects of current, temperature, and frequency coupling including eddy current and skin effects, the final inductance model for:

[0165]

[0166] in, , The magnetic saturation coefficient is temperature-dependent. The reference magnetic saturation coefficient at room temperature. This is the temperature-corrected saturation coefficient; The saturation current threshold; , The temperature-dependent cross-saturation coefficient. The initial cross-saturation coefficient, This is the initial static inductance.

[0167] Step 3: Construct an enhanced state vector containing electrical quantities and parameters to be identified, and establish state equations based on voltage equations and flux linkage equations; achieve dynamic prediction and updating of state and parameters through extended Kalman filtering, combine with model reference adaptive system to drive parameter correction using flux linkage error, and finally adjust weights based on disturbance errors of rotational inertia and friction coefficient to achieve fusion identification.

[0168] The process of constructing an enhanced state vector containing electrical quantities and parameters to be identified, and establishing the state equations based on the voltage equation and flux linkage equation, is as follows:

[0169] First, construct the enhanced state vector. The enhanced state vector includes measurable electrical quantities, related environmental parameters, and parameters to be identified:

[0170]

[0171] Then, the state equations are established by combining the voltage equation and the flux linkage equation:

[0172]

[0173]

[0174] in, for Shaft inductor model, The mutual inductance parameter is a function of temperature change. for shaft voltage, for shaft voltage, Electric angular velocity, ; for Axial magnetic flux; For permanent magnet flux linkage;

[0175] Finally, by differentiating the flux linkage equation and substituting it into the voltage equation, we obtain the current dynamic equation:

[0176]

[0177] in, This is the static value of the stator resistance.

[0178] The process of dynamically predicting and updating states and parameters through extended Kalman filtering includes the extended Kalman filter (EKF) prediction stage and the update stage.

[0179] The prediction stage of the Extended Kalman Filter (EKF) is as follows:

[0180] First, estimate the current state based on the state at the previous time step:

[0181]

[0182] in, for Prior state estimation at time 10:00 for Post-hoc state estimation at time step Sampling time, To control the input vector, This is the state transition function;

[0183] Prediction error covariance matrix:

[0184]

[0185] in, for The prior covariance matrix at time step 1, whose subscripts have the following meanings: Time and The prior state after the measurement value is updated at time step, that is, the previous sampled state is used to predict the subsequent state; for The posterior covariance matrix at time step 1, where the subscripts represent: Time and The posterior state after the time-based measurement is updated; For Jacobian matrices, for transpose, Key elements yes The first component; For process noise covariance;

[0186] The update phase involves correcting the predicted value using current measurements.

[0187]

[0188]

[0189]

[0190] in, It is the Kalman gain matrix of EKF. This is the actual measured current vector. This is the measured current value; The observation matrix; for Transpose of; To measure the noise covariance, For the state update equation, the state estimate is corrected by measuring the residuals. , so that the state estimate Approaching the true value; The covariance update equation serves to quantify the uncertainty of state estimation and ensure the stability of the identification process.

[0191] State estimation is achieved through Jacobian matrix linearization, and the process noise covariance is dynamically adjusted with the magnetic saturation depth. When the system passes the measurement residuals... After being converted into state correction quantities, the updated enhancement vector is obtained. :

[0192]

[0193] They are respectively , Posterior estimate of shaft current They are respectively , Posterior estimate of shaft flux linkage They are respectively , Posterior estimate of shaft inductance This is the a posteriori estimate of the stator resistance. This is the a posteriori estimate of the electromagnetic torque.

[0194] For the final output, .

[0195] The process of using flux linkage error to drive parameter correction in a model reference adaptive system is as follows:

[0196] The Model Reference Adaptive System (MRAS) uses the flux error between the reference model and the adjustable model to design an adaptive law to correct the parameters to be identified, and uses the voltage equation as a transformation to derive the flux calculation formula.

[0197] The reference model is:

[0198]

[0199]

[0200] in, for Axis reference flux linkage for Axis reference flux linkage Indicates time Differentiate;

[0201] The adjustable model is:

[0202]

[0203]

[0204] in, for Axis estimation flux linkage for Axis estimation flux linkage;

[0205] Adaptive law: Definition of flux linkage error:

[0206]

[0207]

[0208] in, for Shaft flux error, for Shaft flux linkage error.

[0209] The process of adjusting the weights based on the disturbance error of rotational inertia and friction coefficient to achieve fusion identification is as follows:

[0210] First, construct the Lyapunov function:

[0211]

[0212] in, This is a Lyapunov function, which is used to construct a function to determine the stability of a system. The ellipsis represents the estimation error of the cross saturation coefficient, and the ellipsis represents the estimation error of other parameters to be corrected. , All are positive adaptive gains; The inductance estimation error equation is as follows: This is an estimated value for the inductance. ;make That is, the Lyapunov function converges, yielding the adaptive law:

[0213]

[0214] in, for The derivative of yes Time derivative, Estimated value of cross saturation coefficient The time derivative;

[0215] The identification results obtained by the MRAS algorithm are , ;

[0216] Then, a dynamic fusion strategy is adopted: based on rotational inertia. Adjustment weight for disturbance error with friction coefficient B:

[0217]

[0218] , All are nominal mechanical parameters; Moment of inertia The disturbance error; The disturbance error is the friction coefficient B. This is the actual moment of inertia. This is the actual coefficient of friction;

[0219] Finally, the fusion weights are obtained. :

[0220]

[0221] All are sensitivity coefficients;

[0222] Final inductance estimate for:

[0223]

[0224] in, These are the identification values ​​obtained using the EKF algorithm.

[0225] Convergence criteria: When the inductance change is less than 0.5% and the relative error is less than 5% over 50 consecutive control cycles, the output parameters stabilize. , , , .

[0226] Experimental verification: Tested on a 200kW permanent magnet wind turbine platform with the following parameters: rated voltage 380V, rated current 326A, number of pole pairs 2, core material 35W250 silicon steel sheet, equipped with a 10kHz sampling frequency fast control prototype DS2000, a 10kHz sampling frequency, 0.2-level current / voltage sensor and a PT100 temperature sensor, with a measurement range of -50℃ to 200℃;

[0227] Test conditions: control angle 30°, load coverage 50%-120%, temperature 25℃-120℃, frequency 200Hz-800Hz; adjusted via weak magnetic field control. The shaft current is 0-300A, which is sufficient to cover its saturation state.

[0228] The results show that at a 30° control angle and 100% load: The shaft inductance is identified as 0.3503 mH, and the measured value is 0.3443 mH, with an error of 2.07%. The shaft inductance identification value is 1.0072mH, while the measured value is 1.0116mH, with an error of 3.31%, which is significantly better than the extended Kalman filter (19%) and polynomial fitting (20%).

[0229] Multi-condition verification: The root mean square error of the inductance is 0.0031mH at 200Hz, 0.0065mH at 500Hz, and 0.0044mH at 800Hz; the inductance decays by 8.2% at 120℃, and the model prediction deviation is <1.5%, verifying the effectiveness of temperature and frequency correction.

[0230] Practical application strategy: Identified parameters are used for dynamic adjustment of current loop PI parameters to improve control accuracy across a wide operating range. After establishing the model, parameters are initialized (JA model parameters are taken from the material handbook, and the fusion weight coefficients are calibrated experimentally). The core parameters of the fusion identification are optimally configured by comparing the estimation errors of multiple methods for accurate identification of online inductors.

[0231] To verify the effectiveness of the proposed method, it was compared with other typical parameter identification methods. The comparison method was set as follows:

[0232] Extended Kalman Filter (EKF): A parameter identification method based on extended Kalman filtering, which achieves inductor parameter tracking through nonlinear state estimation and relies on a pre-defined linearized model;

[0233] Model Reference Adaptive System (MRAS): Adjusts parameters based on flux linkage error, suitable for static or slowly changing conditions;

[0234] Polynomial fitting: This method uses high-order polynomials to fit the nonlinear relationship between inductance and current. It is an empirical modeling method and lacks physical mechanism support.

[0235] BP neural network: a black-box modeling method based on backpropagation neural network, which uses a large amount of data to train and fit the parameter mapping relationship;

[0236] JA-Only: Only the basic inductance modeling of the JA model in this invention is retained, and the multiphysics coupling correction and EKF-MRAS fusion links are removed. This is used to verify the necessity of the coupling correction and fusion algorithms. Meanwhile, comparative experiments were conducted on a 200kW permanent magnet wind turbine experimental platform under rated operating conditions (25℃, 50Hz, 100% load) and wide operating conditions (120℃, 800Hz, 120% load). The results are shown in Tables 1 and 2.

[0237] Table 1 Identification results of each method under rated operating conditions

[0238]

[0239] Performance under rated operating conditions (Table 1) results analysis:

[0240] 1. EKF and MRAS, as classic adaptive algorithms, can achieve a certain level of accuracy under rated operating conditions with gradually changing parameters, but they do not consider the physical mechanism of magnetic saturation, and their average relative errors both exceed 15%.

[0241] 2. Polynomial fitting relies on empirical formulas and cannot be extrapolated to an untrained saturated state, with an average relative error of 20.3%. Backpropagation neural networks can approximate nonlinear relationships through data training, reducing the average relative error to 8.7%, but require a large number of samples for training (computation time 50ms) and have poor physical interpretability.

[0242] 3. JA-Only modeling only uses the basic JA model, and the average relative error is reduced to 5.2%, which shows the advantages of physical mechanism modeling. However, since temperature and frequency corrections are not considered, the accuracy is still limited.

[0243] 4. This invention (JA-Fusion) reduces the error to 2.07% and the calculation time to 3.2ms by fusing multiphysics field coupling correction with EKF-MRAS, thus balancing accuracy and real-time performance.

[0244] Table 2 Performance of various methods under wide operating conditions

[0245]

[0246] Performance analysis under a wide range of operating conditions (Table 2):

[0247] Increased temperature (120℃) and high-frequency operation (800Hz) significantly increased the errors of traditional methods: the average relative error of EKF rose to 32.5%, and the average relative error of polynomial fitting reached 41.2% because it did not consider the decay of magnetic permeability and the skin effect; the BP neural network had insufficient generalization ability under wide operating conditions, and the average relative error increased from 8.7% to 22.6%; JA-Only had an average relative error that increased from 5.2% to 9.8% due to the lack of multi-field coupling correction; the present invention, through temperature-frequency dynamic correction and mechanical parameter weighted fusion, only increased the error to 3.31%, verifying the effectiveness of multi-physics modeling and robust fusion algorithm.

[0248] like Figure 2 As shown, Figure 2 The chart compares the inductor identification results under different algorithms. It compares EKF, MRAS, polynomial fitting, and the method of this invention. Polynomial fitting performs poorly, EKF performs well, and the method of this invention is closest to the actual value.

[0249] Advantages of the method of this invention:

[0250] 1. Supported by physical mechanisms: The JA model characterizes magnetic saturation characteristics from the perspective of magnetic domain motion, and has a stronger ability to extrapolate operating conditions than empirical fitting, i.e., polynomial and data-driven methods, with an error increase of only 0.24% over a wide range of operating conditions;

[0251] 2. Multi-field coupling coverage: Integrating temperature (resistivity, permeability) and frequency (skin effect) corrections, it solves the problem of neglecting environmental factors in traditional methods (EKF, MRAS);

[0252] 3. Robustness of the fusion algorithm: The EKF-MRAS fusion combines mechanical parameter perturbation weighting to dynamically balance tracking speed and stability, resulting in an 89% improvement in accuracy compared to the single algorithm (EKF error 19.0%→2.07%).

[0253] 4. Practical for engineering applications: The calculation time is 3.2ms under rated operating conditions and 3.5ms under wide operating conditions, which meets the real-time requirements of industrial control systems and does not require a large number of samples for training.

[0254] In summary, the present invention is significantly superior to existing methods in terms of accuracy, robustness and engineering applicability in magnetic saturation parameter identification, and is especially suitable for complex working conditions with wide load, variable temperature and wide frequency domain.

Claims

1. A method for identifying inductance parameters of a permanent magnet wind power generator based on a JA model and data fusion, characterized in that, Includes the following steps: Step 1: Initialize the JA model parameters, decompose the magnetization of the ferromagnetic material into reversible and irreversible components, describe the irreversible magnetization process through the domain motion equation, derive the nonlinear relationship between inductance and current by combining Ampere's circuital law, introduce the cross saturation coefficient to correct the interaxial coupling effect, and establish the basic model of inductance under magnetic saturation. In step one, the specific process of establishing the basic inductor model under magnetic saturation is as follows: In the nonlinear mapping of inductance and current, the Ampere loop law is combined, Axial magnetic field strength With Axial magnetic field strength Respectively: Where, The number of turns of the winding, The Axial current, The Axial current, The Equivalent length of axial magnetic circuit, The Equivalent length of axial magnetic circuit; Inductance is defined as the ratio of magnetic flux linkage to current. shaft flux linkage , , , is the magnetic flux density , Magnetization Axial components, for Cross-sectional area of ​​the magnetic circuit of the shaft. Given the vacuum permeability, it is derived from this. Shaft inductor , Shaft inductor The expression is: ； ； in, for Initial value of shaft inductance, , for Initial value of shaft inductance, , , All are cross-saturation coefficients, and are corrected separately. Axis current pair Shaft inductance, Axis current pair The suppression effect of shaft inductance; When the current enters the saturation region, that is , The saturation magnetization of the magnetic material. The magnetization intensity of a ferromagnetic material. The shaft inductance is approximately: ； Shaft inductance and The calculation principle of shaft inductance is the same; Step 2: Construct a dynamic correction model for inductor parameters based on temperature and frequency. The temperature correction model covers changes in stator resistance, core permeability, and JA model parameters. The frequency correction model considers inductor attenuation caused by eddy current and skin effects. A high-frequency inductor correction model is constructed to quantify the skin effect. The effects of current, temperature, and frequency are integrated to form a composite inductor model. Step 3: Construct an enhanced state vector containing electrical quantities and parameters to be identified, and establish state equations based on voltage equations and flux linkage equations; achieve dynamic prediction and updating of state and parameters through extended Kalman filtering, combine with model reference adaptive system to drive parameter correction using flux linkage error, and finally adjust weights based on disturbance errors of rotational inertia and friction coefficient to achieve fusion identification.

2. The method for identifying inductance parameters of permanent magnet wind turbines based on JA model and data fusion as described in claim 1, characterized in that, In step one, the magnetization intensity of the ferromagnetic material is decomposed into reversible and irreversible components. The specific process of describing the irreversible magnetization process through the magnetic domain motion equation is as follows: Magnetization of ferromagnetic materials From reversible components With irreversible components composition, The irreversible component describes the pinning and motion of the magnetic domain walls, following the core differential equation: ； in, The magnetic field strength, This is the magnetic field direction coefficient. This is the hysteresis loss coefficient. The inter-domain coupling coefficient, The hysteresis-free magnetization is described by the Langevin function: ； in, The saturation magnetization is For shape distribution parameters, It is a hyperbolic cotangent function. For effective magnetic field strength, ; The relationship between reversible and irreversible components is as follows: ； in, is the reversibility coefficient.

3. The method for identifying inductance parameters of permanent magnet wind turbines based on JA model and data fusion according to claim 2, characterized in that, In the temperature correction model of step two, the temperature characteristics of the stator resistance follow the resistance-temperature law of metallic conductors: ； in, It is a linear function of temperature and stator resistance. The stator resistance at 25℃ The temperature coefficient of copper, Real-time winding temperature; Deducing the temperature rise from the resistance deviation : ； in, For real-time measurement of the stator winding resistance; The magnetic permeability of the iron core decreases linearly with increasing temperature, and the fitting relationship is as follows: ； in, It is a linear function of the temperature change and the magnetic permeability change. The permeability at 25℃ The temperature decay coefficient of the relative permeability of the iron core; JA model parameters are modulated by temperature: ； ； in, With rising temperature The varying hysteresis loss coefficient This is a linear function of temperature change and the saturation magnetization of the magnetic material. The saturation magnetization of the magnetic material at 25℃. The temperature decay coefficient of saturation magnetization. This represents the hysteresis coefficient and the temperature increase factor.

4. The method for identifying inductance parameters of permanent magnet wind turbines based on JA model and data fusion as described in claim 3, characterized in that, In the frequency correction model of step two, based on Maxwell's equations, the eddy current diffusion equation is: ； in, magnetic field strength The Laplace operator describes the spatial second-order rate of change of the magnetic field strength; The permeability of a permanent magnet. For the electrical conductivity of the material, For time; due to sinusoidal excitation , Angular frequency, As the initial value for the excitation, The imaginary unit is used; therefore, the eddy current diffusion equation simplifies to The magnetic field distribution in the planar iron core is as follows: ； in, The strength of the spatial magnetic field. As the reference magnetic field strength, It is the distance from the geometric center of the iron core. For high-frequency attenuation coefficient, For the thickness of the iron core, , For the depth of the skin effect, ; It is a hyperbolic cosine function; Magnetic permeability exists in a complex form with a real part and an imaginary part under a high-frequency alternating magnetic field. Considering the influence of eddy currents, the stored energy component of the real part of the magnetic permeability is defined as: ； in, The real part of the permeability is the energy storage component. It is the hyperbolic tangent function.

5. The method for identifying inductance parameters of permanent magnet wind turbines based on JA model and data fusion according to claim 4, characterized in that, In step two, the high-frequency inductor correction model is as follows: ； in, Let frequency be the inductor frequency with respect to frequency. Inductance value at 50Hz For input frequency, For material constants, ; Integrating the effects of current, temperature, and frequency coupling including eddy current and skin effect, ultimately... Axial inductance model for: ； in, , The magnetic saturation coefficient is temperature-dependent. The reference magnetic saturation coefficient at room temperature. This is the temperature-corrected saturation coefficient; The saturation current threshold; , The temperature-dependent cross-saturation coefficient. The initial cross-saturation coefficient, This is the initial static inductance.

6. The method for identifying inductance parameters of permanent magnet wind turbines based on JA model and data fusion according to claim 5, characterized in that, In step three, the process of constructing an enhanced state vector containing electrical quantities and parameters to be identified, and establishing the state equation based on the voltage equation and flux linkage equation, is as follows: First, construct the enhanced state vector. The enhanced state vector includes measurable electrical quantities, related environmental parameters, and parameters to be identified: ； Then, the state equations are established by combining the voltage equation and the flux linkage equation: ； ； in, for Shaft inductor model, The mutual inductance parameter is a function of temperature change. for shaft voltage, for shaft voltage, Electric angular velocity, ; for Axial magnetic flux; For permanent magnet flux linkage; Finally, by differentiating the flux linkage equation and substituting it into the voltage equation, we obtain the current dynamic equation: ； in, This is the static value of the stator resistance.

7. The method for identifying inductance parameters of permanent magnet wind turbines based on JA model and data fusion according to claim 6, characterized in that, In step three, the process of dynamically predicting and updating the state and parameters through extended Kalman filtering includes an extended Kalman filter (EKF) prediction stage and an update stage. The prediction stage of the Extended Kalman Filter (EKF) is as follows: First, estimate the current state based on the state at the previous time step: ； in, for Prior state estimation at time 10:00 for Post-hoc state estimation at time step Sampling time, To control the input vector, This is the state transition function; Prediction error covariance matrix: ； in, for Prior covariance matrix at time step for The posterior covariance matrix at time step 1. For Jacobian matrices, for transpose, Key elements yes The first component; For process noise covariance; The update phase involves correcting the predicted value using current measurements. ； ； ； in, It is the Kalman gain matrix of EKF. This is the actual measured current vector. This is the measured current value; The observation matrix; for Transpose of; To measure the noise covariance, For the state update equation, the state estimate is corrected by measuring the residuals. , so that the state estimate Approaching the true value; The covariance update equation serves to quantify the uncertainty of state estimation and ensure the stability of the identification process.

8. The method for identifying inductance parameters of permanent magnet wind turbines based on JA model and data fusion according to claim 7, characterized in that, In step three, the process of using the magnetic flux error to drive parameter correction in conjunction with the model reference adaptive system is as follows: The Model Reference Adaptive System (MRAS) uses the flux linkage error between the reference model and the adjustable model to design an adaptive law to correct the parameters to be identified, and uses the voltage equation as a transformation to derive the flux linkage calculation formula. The reference model is: ； ； in, for Axis reference flux linkage for Axis reference flux linkage Indicates time Differentiate; The adjustable model is: ； ； in, for Axis estimation flux linkage for Axis estimation flux linkage; Adaptive law: Definition of flux linkage error: ； ； in, for Shaft flux error, for Shaft flux linkage error.

9. The method for identifying inductance parameters of permanent magnet wind turbines based on JA model and data fusion according to claim 8, characterized in that, In step three, the process of adjusting the weights based on the disturbance error of the moment of inertia and friction coefficient to achieve fusion identification is as follows: First, construct the Lyapunov function: ； in, This is a Lyapunov function, which is used to construct a function to determine the stability of a system. The ellipsis represents the estimation error of the cross saturation coefficient, and the ellipsis represents the estimation error of other parameters to be corrected. , All are positive adaptive gains; The inductance estimation error equation is as follows: This is an estimated value for the inductance. ;make That is, the Lyapunov function converges, yielding the adaptive law: ； in, for The derivative of yes Time derivative, Estimated value of cross saturation coefficient The time derivative; The identification results obtained by the MRAS algorithm are , ; Then, a dynamic fusion strategy is adopted: based on rotational inertia. Adjustment weight for disturbance error with friction coefficient B: ； , All are nominal mechanical parameters; Moment of inertia The disturbance error; The disturbance error is the friction coefficient B. This is the actual moment of inertia. This is the actual coefficient of friction; Finally, the fusion weights are obtained. : ； All are sensitivity coefficients; Final inductance estimate for: ； in, These are the identification values ​​obtained using the EKF algorithm.

Images