A permanent magnet synchronous motor parameter identification method based on adaptive Kalman filtering

Abstract

This invention provides an online parameter identification method for permanent magnet synchronous motors based on adaptive Kalman filtering, comprising: establishing parameters of the permanent magnet synchronous motor in... dq The discretized model in the coordinate system is used to obtain the predicted voltage equation through one-step delay compensation. Based on the predicted current deviation between the parameter mismatch model and the ideal model, the expressions for stator inductance and rotor flux linkage are decoupled. State-space models are constructed for the decoupled inductance and flux linkage respectively, and an improved Sage-Husa adaptive Kalman filter algorithm is used to dynamically adjust the covariance matrix of the system and measurement noise. The error covariance matrix is ​​corrected by the optimal weighting coefficient and the sliding window adaptive factor, thereby significantly improving the identification accuracy of parameter mutations and operating condition changes. This achieves high-precision online decoupling identification of inductance and flux linkage, and the identification results are fed back to the control system in real time, effectively improving current tracking performance and control quality.

Description

Technical Field

[0001] This invention relates to the field of motor control technology, specifically a parameter identification method for permanent magnet synchronous motors based on adaptive Kalman filtering. Background Technology

[0002] Permanent magnet synchronous motors (PMSMs) are widely used in industrial drives, new energy vehicles, aerospace, and other fields due to their high efficiency, high power density, and superior control performance. Their high-performance control relies on the accuracy of motor parameters, especially stator inductance and permanent magnet flux linkage. However, during actual operation, factors such as temperature rise, magnetic saturation, and aging can cause slow or abrupt changes in these parameters, leading to parameter mismatch and consequently, decreased control performance, reduced efficiency, and even system instability. Therefore, achieving accurate online identification of key motor parameters is crucial for maintaining high-performance system operation.

[0003] Currently, online parameter identification methods mainly include model reference adaptive methods, extended Kalman filtering, and standard Kalman filtering. However, when performing simultaneous multi-parameter identification, the fixed noise covariance matrix of the standard Kalman filtering algorithm is difficult to adapt to the dynamic characteristic changes caused by changes in operating conditions, sudden load changes, or time-varying parameters during motor operation. This leads to decreased identification accuracy, slower convergence speed, and even filter divergence when parameters are mismatched or operating conditions change. It cannot guarantee continuous, stable, and decoupled accurate identification of inductance and flux linkage across the entire operating range. Specifically, the noise statistical characteristics cannot be adjusted online, and the algorithm's ability to track sudden changes in operating conditions is insufficient, limiting the practicality of parameter identification methods. Summary of the Invention

[0004] The purpose of this invention is to provide a parameter identification method for permanent magnet synchronous motors based on adaptive Kalman filtering, which can dynamically adjust the filter parameters online, achieve multi-parameter decoupling identification, and has strong adaptability to operating conditions and parameter changes.

[0005] The technical solution of this invention is: A method for parameter identification of permanent magnet synchronous motors based on adaptive Kalman filtering includes: Establish permanent magnet synchronous motor in dq The current equation in the coordinate system is discretized and compensated for a one-beat delay to obtain the predicted voltage equation. Based on the deviation of the predicted current between the motor model with parameter mismatch and the ideal motor model without parameter mismatch, the stator inductance and rotor flux are decoupled. For the decoupled stator inductance and rotor flux linkage, corresponding linear discrete system state-space models are established respectively. An improved Sage-Husa adaptive Kalman filter algorithm is used to identify the stator inductance and rotor flux linkage of the motor online. The improved Sage-Husa adaptive Kalman filter algorithm includes: online estimation and dynamic adjustment of the system noise covariance matrix and the measurement noise covariance matrix, and adaptive correction of the error covariance matrix for inductance identification and flux linkage identification respectively.

[0006] Furthermore, the specific process of decoupling the stator inductance and rotor flux linkage includes: Based on surface-mount permanent magnet synchronous motor dq The current equation in the axial coordinate system, after discretization and one-beat delay compensation, yields the predicted voltage equation at time k+1. Calculate the deviation between the predicted current of the motor model with inductance and resistance mismatch and the predicted current of the ideal model without parameter mismatch; Ignoring the influence of current term changes and assuming that the electric angular velocity remains constant between adjacent sampling periods, the process is simplified to consider two adjacent moments as such. q The expression for the difference in shaft current prediction deviation is derived, thereby deriving the calculation expressions for the decoupled stator inductance and rotor flux linkage, respectively.

[0007] Furthermore, the process of establishing the state-space model of the linear discrete system includes: For stator inductance identification, the stator inductance is used as the state variable, and the known quantities in the stator inductance calculation expression are used as the output variables to determine the corresponding state transition matrix and observation matrix. For rotor flux identification, the rotor flux is used as the state variable, and the known quantities in the rotor flux calculation expression are used as the output variables to determine the corresponding state transition matrix and observation matrix.

[0008] Furthermore, the basic framework of the improved Sage-Husa adaptive Kalman filter algorithm includes: A recursive formula with a forgetting factor is used to estimate the system noise covariance matrix and the measurement noise covariance matrix online. A minimum lower bound is set for the noise covariance matrix to prevent it from losing its positive definiteness and causing filter divergence.

[0009] Furthermore, for the identification of stator inductors, methods for adaptively correcting the error covariance matrix include: Calculate the innovation sequence and construct a filter divergence criterion based on the comparison between the sum of squares of the innovation sequence and the theoretical covariance trace; When filtering divergence is determined, an optimal weighting coefficient is calculated by reverse deduction based on the ideal matching condition that the theoretical value of the new information covariance equals the actual value.C k ; Using the optimal weighting coefficients C k Predicted values ​​of the error covariance matrix in the Kalman filter algorithm P ( k / k- 1 The filter is then corrected and updated.

[0010] Furthermore, for rotor flux identification, methods for adaptively correcting the error covariance matrix include: First, filter convergence detection is performed. If the filter is determined to be diverging, the error covariance is corrected. Calculate the innovation sequence, and calculate the theoretical value and the actual estimate based on the sliding window method for the innovation covariance; Based on the relationship between the theoretical value and the actual estimated value, an adaptive factor λ is calculated. Enabled only when the calculated value of the adaptive factor λ is less than 1, and its maximum value is limited; The adaptive factor λ is used to predict the error covariance matrix in the Kalman filter algorithm. Make corrections, and then continue updating the filter.

[0011] Furthermore, the method for solving the adaptive factor λ includes the following steps: First, calculate the theoretical value of the new information covariance: ; In the formula, For the observation matrix, The predicted value of the error covariance matrix. for The transpose of the matrix, To measure the noise covariance matrix; Next, the actual estimate of the new information covariance is calculated using the sliding window method: ; In the formula, m To adjust the sliding window size, k Indicates the current variable, j For the summation variable, For the information sequence at time j, This is the transpose of the new information sequence; Finally, based on the theoretical and actual values ​​of the new information covariance, the adaptive factor λ is solved: ; To avoid unnecessary adjustments, the adaptive factor is only enabled when the calculated value is less than 1; to avoid excessive adjustments, the maximum value of the adaptive factor is set to 0.9.

[0012] Furthermore, it also includes: before starting online identification, setting initial state values ​​for stator inductance and rotor flux linkage based on nominal parameters provided by the motor manufacturer or offline measurement results, and setting initial system noise covariance matrix and measurement noise covariance matrix for the improved Sage-Husa adaptive Kalman filter algorithm.

[0013] Furthermore, the identified stator inductance and rotor flux linkage parameters are input in real time into the current controller of the motor vector control system to update the control parameters of the current loop.

[0014] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for identifying parameters of a permanent magnet synchronous motor based on adaptive Kalman filtering.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This invention effectively overcomes the performance degradation problem of standard Kalman filtering due to fixed noise covariance during permanent magnet synchronous motor operation by introducing an improved Sage-Husa adaptive Kalman filter algorithm and constructing a decoupled parameter identification model. The method first achieves decoupled expression of inductance and flux linkage based on predicted current deviation, and then establishes independent state-space models for synchronous identification. By dynamically estimating and adjusting the system and measurement noise covariance online, and correcting the error covariance matrix using optimal weighting coefficients and sliding window adaptive factors, the algorithm maintains rapid convergence and stable tracking even when motor parameters change abruptly or operating conditions change. Therefore, this invention achieves high-precision, highly adaptable online decoupled identification of stator inductance and rotor flux linkage across the entire operating range, and feeds the identification results back to the control system in real time, significantly improving the tracking accuracy of the current loop and the overall control quality of the system, enhancing the reliability and robustness of the motor drive system under complex operating conditions. Attached Figure Description

[0016] Figure 1 This is a flowchart of the parameter identification method of the present invention.

[0017] Figure 2 This is a comparison of the stator inductance identification values ​​of the standard Kalman filter algorithm and the improved Sage-Husa Kalman filter algorithm under the same Kalman filter parameters.

[0018] Figure 3This is a comparison of rotor flux identification values ​​between the standard Kalman filter algorithm and the improved Sage-Husa Kalman filter algorithm under the same Kalman filter parameters.

[0019] Figure 4 It is the standard Kalman filter algorithm with the same Kalman filter parameters. dq Axis current.

[0020] Figure 5 It is an improved Sage-Husa Kalman filter algorithm with the same Kalman filter parameters. dq Axis current. Detailed Implementation

[0021] The following is combined Figures 1 to 5 The specific embodiments of the present invention will be described in detail below.

[0022] It should be noted that the circuit connections involved in this invention all adopt conventional circuit connection methods and do not involve any innovation.

[0023] Example like Figure 1 As shown, a parameter identification method for permanent magnet synchronous motors based on adaptive Kalman filtering includes the following steps: Establish permanent magnet synchronous motor in dq Voltage equation in rotating coordinate system: (1) in, u d , u q They are respectively dq Voltage of the shaft i d , i q They are respectively dq Stator current of the shaft, R s For stator resistance, L s For stator inductance, ω e Let Φ be the electric angular velocity. f It is a permanent magnet flux linkage.

[0024] By performing forward Euler discretization on the above equations and considering the one-cycle delay of the digital control system, we obtain: k Predicted voltage equation at time +1: (2) , and Represent dq axisk The voltage that should be input at time +1 and They represent the predicted k+ 1 moment d axis, q Shaft predicts current. and They represent d axis, q Shaft reference current, To control the cycle.

[0025] The specific process of decoupling the stator inductance and rotor flux linkage at that moment includes: Based on surface-mount permanent magnet synchronous motor dq The current equation in the axial coordinate system, after discretization and one-beat delay compensation, yields: k The predicted voltage equation at time +1; Calculate the deviation between the predicted current of the motor model with inductance and resistance mismatch and the predicted current of the ideal model without parameter mismatch; Ignoring the influence of current term changes and assuming that the electric angular velocity remains constant between adjacent sampling periods, the process is simplified to consider two adjacent moments as such. q The expression for the difference in shaft current prediction deviation is derived, thereby deriving the calculation expressions for the decoupled stator inductance and rotor flux linkage.

[0026] Specifically, to decouple the motor parameters, the difference between the predicted current of the motor with parameter mismatch and the predicted current of the motor without parameter mismatch is first calculated: (3) In the formula, L s0 For nominal stator inductance, ΔL s For the uncertain component of stator inductance, R s0 The nominal stator resistance, Δ R s Φ is the uncertain component of the stator resistance. f0 For nominal rotor flux, Δ Φ f This represents the uncertain component of the rotor flux linkage.

[0027] As can be seen from equation (3), d The shaft current prediction error includes a current term (first term) and a voltage term (second term); q The shaft current prediction error includes a current term (first term), a voltage term (second term), and a speed term (third term).

[0028] During motor operation, the stator current is relatively small and changes continuously; therefore, the current term can be considered much smaller than the voltage term. Since the sampling period of the motor control is much shorter than the mechanical time constant, the electrical angular velocity between two adjacent sampling periods can be considered almost constant. Based on the above analysis, the electrical angular velocity between two adjacent moments... k+ 1 and k of q The difference in shaft current prediction deviation is:

[0029] (4) According to equation (4), the expression for calculating the decoupled stator inductance is obtained: (5) Substituting (5) into equation (1), we obtain the expression for calculating the decoupled rotor flux linkage: (6) The process of establishing a state-space model for a linear discrete system includes: For stator inductance identification, the stator inductance is used as the state variable, and the known quantities in the stator inductance calculation expression are used as the output variables, such as quantities related to voltage and current, to determine the corresponding state transition matrix and observation matrix. For rotor flux identification, the rotor flux is used as the state variable, and the known quantities in the rotor flux calculation expression are used as the output variables, such as quantities related to voltage, current, and speed, to determine the corresponding state transition matrix and observation matrix.

[0030] Specifically, the mathematical model of a linear discrete system: (7) In the formula, x ( k )and x ( k- 1) respectively k Time and ( k -1) State variables at time t; u ( k-1 )for( k -1) System input at time t; y ( k ) is the first k The system output variable at time t; A ( k () represents the state matrix between adjacent time points; B ( k () represents the system input matrix; H ( k ) represents the observation matrix; w ( k The system noise is denoted as , with a mean of 0 and a covariance of . Q (k Gaussian distribution; v ( k The noise is the measurement noise, with a mean of 0 and a covariance of 0. G ( k The Gaussian distribution of ).

[0031] The improved Sage-Husa adaptive Kalman filter algorithm has the following basic framework: A recursive formula with a forgetting factor is used to estimate the system noise covariance matrix and the measurement noise covariance matrix online. Set a minimum lower bound for the noise covariance matrix to prevent it from losing its positive definiteness and causing the filter to diverge.

[0032] Specifically, the formula for Kalman filtering is as follows: (8) The first two equations represent the prediction part, and the last four equations represent the update part. In the equations, for k Time-prior state estimation for k Time-prior error covariance For Kalman gain, e ( k ) is the new information sequence.

[0033] For inductor identification, according to equation (5), select L s For state variables, If the output variable is used, then the state transition matrix is... A ( k ) and observation matrix H ( k )for: (9) For magnetic linkage identification, the state variable is selected according to equation (6): (10) Output variable selection: (11) State transition matrix (12) Observation matrix: (13) To implement the adaptive Kalman filter algorithm, this embodiment introduces an improved Sage-Husa adaptive Kalman filter algorithm. The Sage-Husa Kalman filter algorithm achieves adaptive control by online estimation of the noise covariance matrices Q and G.

[0034] The core of the Sage-Husa Kalman filter algorithm is to use the innovation sequence... e ( k The measurement noise covariance G and system noise covariance Q are estimated online. The specific process is as follows: (14) in (15) In the formula, b is the forgetting factor, which generally ranges from 0.95 to 0.99.

[0035] To prevent the covariance matrix from losing its positive definiteness and the filter from diverging, a minimum lower limit needs to be set for the noise covariances Q and G, such as 1e^-30.

[0036] The Sage-Husa Kalman filter algorithm still has some drawbacks, such as the inability to quickly track sudden changes in the system state. Therefore, this embodiment corrects the estimated value of the error covariance matrix.

[0037] For stator inductor identification, adaptive correction methods for the error covariance matrix include: Calculate the innovation sequence and construct a filter divergence criterion based on the comparison between the sum of squares of the innovation sequence and the theoretical covariance trace; When filtering divergence is determined, an optimal weighting coefficient is calculated by reverse deduction based on the ideal matching condition that the theoretical value of the new information covariance equals the actual value. C k ; Using optimal weighting coefficients C k The predicted value of the error covariance matrix in the Kalman filter algorithm is corrected, and then the filtering is updated.

[0038] This embodiment uses the sum of squares of the innovation sequence as the criterion for filtering anomalies, as shown in the following formula: (16) In the formula, γ For the reserve coefficient, γ ≥1; tr This indicates finding the trace of a matrix.

[0039] If equation (16) is satisfied, it means that the actual error will exceed the theoretically predicted value. γ times.

[0040] Assumption G ( k ) =G ( k+ 1), then we have: (17) Substituting equation (17) into equation (16), the filter divergence criterion is: (18) Equation (18) is the filter divergence criterion. When this equation is true, the filter is abnormal.

[0041] The optimal weighting coefficients are inversely derived using the most stringent covariance matching convergence condition. Ideally, the actual covariance of the innovation should equal its theoretical value, i.e.:

[0042] (19) Substituting the corrected formula, which includes the optimal weighting coefficients, into the above ideal conditions, we obtain: (20) In the formula, C k These are the optimal weighting coefficients.

[0043] The optimal weighting coefficients can be obtained by solving equation (20): (twenty one) If equation (18) is satisfied, substituting the optimal weighting coefficients into the second equation of equation (8) yields: (twenty two) Combining equations (8)-(15) and (16)-(22), we can obtain the adaptive Kalman filter identification algorithm for inductors.

[0044] For rotor flux identification, methods for adaptively correcting the error covariance matrix include: Similar to the inductance identification method, a filtering convergence test is first performed, as shown in equation (18). If equation (18) is satisfied, the error covariance is corrected.

[0045] Calculate the innovation sequence, and calculate the theoretical value and the actual estimate based on the sliding window method for the innovation covariance; Based on the relationship between the theoretical value and the actual estimated value, an adaptive factor λ is solved; Enabled only when the calculated value of the adaptive factor λ is less than 1, and its maximum value is limited; Predicting the error covariance matrix in the Kalman filter algorithm using an adaptive factor λ Make corrections, and then continue updating the filter.

[0046] Specifically, an adaptive factor based on the sliding window method is used to correct the error covariance matrix.

[0047] (twenty three) In the formula, λ is the adaptive factor, which is solved as follows: First, calculate the theoretical value of the new information covariance: (twenty four) Next, the actual estimate of the new information covariance is calculated using the sliding window method: (25) In the formula, m To adjust the sliding window size, k Indicates the current variable, j For the summation variable.

[0048] Finally, based on the theoretical and actual values ​​of the new information covariance, the adaptive factor λ is solved: (26) To avoid unnecessary adjustments, the adaptive factor is only enabled when the calculated value is less than 1; to avoid excessive adjustments, the maximum value of the adaptive factor is set to 0.9. Combining equations (8)-(15), (18), and (23)-(26), the adaptive Kalman filter algorithm for flux linkage identification can be obtained.

[0049] In one example of using the method provided above, the parameters of the motor are as follows: stator resistance R s0 The Ω is 0.365Ω, and the stator inductance L is... s0 H is 0.001225, and the rotor permanent magnet flux linkage Φ is... f0 The value is 0.1667 Wb. The operating conditions are as follows: the motor model receives a torque step change of (2 Nm to 7 Nm) at 0.1s, and a torque step change of (7 Nm to 4 Nm) at 0.15s; it also receives a speed step change of (300 rpm to 600 rpm) at 0.1s, and a speed step change of (600 rpm to 1200 rpm) at 0.2s. At 0.1s, the inductance changes from L... s0 Set to 2L s0 At 0.15s, the magnetic flux from Φ f0 Set to 0.5Φ f0 .

[0050] The initial values ​​of the noise variance matrix for the standard Kalman filter and the improved Sage-Husa adaptive Kalman filter are as follows: for inductor identification, the system noise variance matrix Q1 is set to 6e-1, and the measurement noise variance matrix G1 is set to 2.8e-8; for flux linkage identification, the system noise variance matrix Q2 is set to 1e-3, and the measurement noise variance matrix G2 is set to 1e+1. For the improved Sage-Husa adaptive Kalman filter, the reserve coefficient for inductor identification is... γ Set to 1 for the magnetic flux identification reserve coefficient. γSet the value to 10, the sliding window size m to 10, and the forgetting factor b to 0.98 and 0.995 in inductance identification and magnetic flux identification, respectively.

[0051] like Figure 2 As shown, under this operating condition and the selected Kalman filter parameters, both methods can accurately identify the inductance. The root mean square error of the inductance of the standard Kalman filter is 2.56e-05, and the root mean square error of the inductance of the adaptive Kalman filter is 1.78e-05.

[0052] like Figure 3 As shown, for flux linkage identification, compared with the standard Kalman filter, the flux linkage identified by the improved Sage-Husa adaptive Kalman filter is less affected by parameter abrupt changes. The root mean square error of flux linkage of the standard Kalman filter is 0.023, while that of the adaptive Kalman filter is 0.01.

[0053] like Figure 4 and Figure 5 As shown, compared to the standard Kalman filter, the adaptive Kalman filter can effectively improve the current tracking performance and is more robust.

[0054] In summary, the parameter identification method for permanent magnet synchronous motors based on adaptive Kalman filtering in this embodiment achieves high-precision and highly adaptable simultaneous online identification of stator inductance and rotor flux linkage through model decoupling, adaptive noise estimation, and targeted error covariance correction strategies. The identification results are effectively applied to motor control, thereby improving the overall system performance.

[0055] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for identifying parameters of a permanent magnet synchronous motor based on adaptive Kalman filtering.

[0056] The above-disclosed embodiments are merely preferred embodiments of the present invention. However, the embodiments of the present invention are not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A permanent magnet synchronous motor parameter identification method based on adaptive Kalman filtering, characterized in that, include: The current equation of permanent magnet synchronous motor in dq Coordinate system is established, and discretization and one-step delay compensation are performed to obtain a prediction voltage equation, the deviation of the predicted current between the motor model with parameter mismatch and the ideal motor model without parameter mismatch is calculated, and the stator inductance and rotor flux linkage are decoupled according to the deviation of the predicted current. For the decoupled stator inductance and rotor flux linkage, corresponding linear discrete system state-space models are established respectively. An improved Sage-Husa adaptive Kalman filter algorithm is used to identify the stator inductance and rotor flux linkage of the motor online. The improved Sage-Husa adaptive Kalman filter algorithm includes: online estimation and dynamic adjustment of the system noise covariance matrix and the measurement noise covariance matrix, and adaptive correction of the error covariance matrix for inductance identification and flux linkage identification respectively.

2. The method according to claim 1, wherein, The specific process of decoupling stator inductance and rotor flux linkage includes: Based on the current equation of surface-mounted permanent magnet synchronous motor in dq After discretization and one-shot delay compensation, the predicted voltage equation at k+1 time is obtained. Calculate the deviation between the predicted current of the motor model with inductance and resistance mismatch and the predicted current of the ideal model without parameter mismatch; Neglecting the influence of current term change, and according to the condition that the electrical angular velocity is considered as invariable in adjacent sampling period, the adjacent two time q The expression of the deviation of the shaft current prediction is derived, and thus the decoupled calculation expressions of the stator inductance and the rotor flux are derived respectively.

3. The method according to claim 2, wherein, The process of establishing the state-space model of the linear discrete system includes: For stator inductance identification, the stator inductance is used as the state variable, and the known quantities in the stator inductance calculation expression are used as the output variables to determine the corresponding state transition matrix and observation matrix. For rotor flux identification, the rotor flux is used as the state variable, and the known quantities in the rotor flux calculation expression are used as the output variables to determine the corresponding state transition matrix and observation matrix.

4. The method of claim 1, wherein, In the improved Sage-Husa adaptive Kalman filter algorithm, the specific steps for online estimation and dynamic adjustment of the system noise covariance matrix and the measurement noise covariance matrix include: A recursive formula with a forgetting factor is used to estimate the system noise covariance matrix and the measurement noise covariance matrix online. A minimum lower bound is set for the noise covariance matrix to prevent it from losing its positive definiteness and causing filter divergence.

5. The method of claim 1, wherein, Also includes: Before starting online identification, initial state values ​​are set for stator inductance and rotor flux linkage based on nominal parameters provided by the motor manufacturer or offline measurement results, and initial system noise covariance matrix and measurement noise covariance matrix are set for the improved Sage-Husa adaptive Kalman filter algorithm.

6. The method of claim 4, wherein, For stator inductor identification, adaptive correction methods for the error covariance matrix include: Calculate the innovation sequence and construct a filter divergence criterion based on the comparison between the sum of squares of the innovation sequence and the theoretical covariance trace; When the filter divergence is determined, a best weighting coefficient is calculated according to the ideal matching condition that the innovation covariance theoretical value is equal to the actual value C k ; Utilizing the optimal weighting coefficients C k The error covariance matrix prediction value in the Kalman filtering algorithm is corrected, and then the filtering update is continued. P k / k-1 ​​ 7. The method of claim 4, wherein the method is based on an adaptive Kalman filter. For rotor flux identification, methods for adaptively correcting the error covariance matrix include: Calculate the innovation sequence, and calculate the theoretical value and the actual estimate based on the sliding window method for the innovation covariance; Based on the relationship between the theoretical value and the actual estimated value, an adaptive factor λ is calculated. Enabled only when the calculated value of the adaptive factor λ is less than 1, and its maximum value is limited; The predicted value of the error covariance matrix in the Kalman filter algorithm is corrected using the adaptive factor λ, and then the filtering is updated.

8. The method according to claim 7, wherein, The method for solving the adaptive factor λ includes the following steps: Perform filter convergence detection, and when the filter is determined to be diverging, correct the error covariance. Calculate the theoretical value of the new information covariance: ； In the formula, For the observation matrix, The predicted value of the error covariance matrix. for The transpose of the matrix, To measure the noise covariance matrix; Next, the actual estimate of the new information covariance is calculated using the sliding window method: ； In the formula, m To adjust the sliding window size, k Indicates the current variable, j For the summation variable, For the information sequence at time j, This is the transpose of the new information sequence; Finally, based on the theoretical and actual values ​​of the new information covariance, the adaptive factor λ is solved: ； Enabled only when λ < 1, with a maximum limit of 0.

9.

9. The method of claim 1, wherein, The identified stator inductance and rotor flux linkage parameters are input in real time into the current controller of the motor vector control system to update the control parameters of the current loop.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by the processor, it implements the parameter identification method for permanent magnet synchronous motors based on adaptive Kalman filtering as described in any one of claims 1-9.

Images