Doubly-fed induction motor parameter identification method based on adaptive extended Kalman filtering

Abstract

The invention discloses a doubly-fed induction motor parameter identification method based on adaptive extended Kalman filtering, and aims to solve the problems of low doubly-fed induction motor parameter identification precision and integrity, sensitive noise and the like in the prior art. According to a dq-axis flux linkage equation and a voltage equation of the doubly-fed induction motor, a state space model with dq-axis current of a stator and a rotor as state variables is built, the dimension of the state space model is expanded, the full-parameter observability of resistance Rs and Rr and inductance Ls, Lr and Lm of the motor is judged by utilizing a self-adaptive extended Kalman filtering theory and combining a rank criterion, and the accuracy and accuracy of the state space model are improved. All-parameter identification of inductance and resistance of the motor is realized, and the robustness of the system is improved through a method of adaptively adjusting noise covariance. The parameter coupling influence can be effectively eliminated, and the identification precision is improved; the self-adaptive noise suppression mechanism enhances the anti-interference capability, and meets the dynamic on-line parameter identification requirements of the doubly-fed motor.

Description

Technical Field

[0001] The present invention relates to the field of on-line parameter identification of motors, and specifically to a parameter identification method for a doubly-fed induction motor based on adaptive extended Kalman filtering. Background Art

[0002] In the field of on-line identification of electrical parameters of doubly-fed induction motors, the doubly-fed induction motor has become the mainstream type of wind turbine generator due to its advantages such as wide speed range operation, high-efficiency energy conversion, and flexible power control. However, its operation is affected by complex working conditions such as wind speed fluctuations and grid disturbances, and real-time dynamic modeling is required to optimize control. The motor parameters will drift with factors such as temperature, magnetic saturation, and aging. The static model identified offline cannot accurately reflect the actual working conditions, and the parameters need to be updated online. The stator and rotor windings of the DFIG are coupled through converters, and the magnetic field is non-linear. Traditional linear models are difficult to accurately describe the dynamic behavior. In actual operation, the measurement noise interference is large, and the traditional Kalman filtering has insufficient robustness.

[0003] The differences compared with the prior art are as follows: Chinese Patent Application Publication No. CN117978016A proposes a parameter identification control method for a permanent magnet synchronous motor, which first obtains the inductance value using the gradient descent method, and then uses the extended Kalman filter to identify the resistance and magnetic flux. The method proposed in this application requires step-by-step identification of the electrical parameters of the motor, and there are problems of high identification delay and large error accumulation.

[0004] Chinese Patent Application Publication No. CN118631112A proposes a parameter identification method using the torque mean square error as the evaluation function, which uses the Kalman filter algorithm to identify the inductance, magnetic flux, and resistance of the permanent magnet synchronous motor, and combines the evaluation function model to predict the current and torque at the next moment. This application has problems of poor anti-interference ability of parameter identification and large torque measurement error.

[0005] Chinese Patent Application Publication No. CN111669093A proposes a parameter identification control method for a brushless DC motor, which establishes a mathematical modeling method based on the αβ axis and establishes a mathematical model of the brushless DC motor. The current in the mathematical modeling of the αβ axis in this application is alternating current, and the cross-coupling effect is strong, resulting in problems of large system disturbance and poor self-disturbance rejection ability.

[0006] Chinese Patent Application Publication No. CN117978016A proposes a parameter identification control method for a permanent magnet synchronous motor. The identification object of this application is a permanent magnet synchronous motor, and the state equation structure is simple. The control method is not suitable for the doubly-fed induction motor with high complexity.

[0007] The Chinese published application CN111293693A proposes a method for identifying the control parameters of a doubly-fed wind turbine converter based on the extended Kalman filter. This application combines the mathematical model of the doubly-fed induction motor to identify the control parameters of the doubly-fed induction motor, but there is a problem that the electrical parameters of the doubly-fed induction motor cannot be identified. Summary of the Invention

[0008] In view of the deficiencies of the prior art, the present invention proposes a parameter identification method for a doubly-fed induction motor based on an adaptive extended Kalman filter. Based on the state space equation of the doubly-fed induction motor with the stator and rotor dq-axis currents as state variables, the adaptive extended Kalman filter algorithm theory is used to achieve high-precision and low-complexity parameter identification. The core of the present invention lies in parameter identification interconnection coupling and adaptive filter optimization.

[0009] The object of the present invention can be achieved through the following technical solutions:

[0010] A parameter identification method for a doubly-fed induction motor based on an adaptive extended Kalman filter includes the following steps:

[0011] 1) Build a state space model of the doubly-fed induction motor with the stator and rotor dq-axis currents as state variables;

[0012] 2) Based on the state space equation of the doubly-fed induction motor with the stator and rotor dq-axis currents as state variables, incorporate the parameters to be identified into the state variables, expand the dimension of the motor state space model, and judge the observability of the parameters;

[0013] 3) Using the adaptive extended Kalman filter theory, adopt the method of parameter iterative identification to achieve the dynamic online identification of the stator resistance R s , rotor resistance R r , stator inductance L s , rotor inductance L r and stator-rotor mutual inductance L m .

[0014] As a further improvement of the present invention, the step 1) is specifically as follows:

[0015] S11. According to the voltage equation and flux linkage equation of the doubly-fed induction motor, convert the flux linkage differential term into a current differential term, and ignore the differential terms of the stator resistance R s , rotor resistance R r , stator inductance L s , rotor inductance L r and stator-rotor mutual inductance L m , that is , where t is time, to obtain the mathematical relationship between the current differential and the flux linkage differential:

[0016] Flux linkage equation:

[0017] , , , ;

[0018] Voltage equation:

[0019] , , , ;

[0020] In the formula , , , , , , , , , , , are the stator d-axis current, stator q-axis current, rotor d-axis current and rotor q-axis current, stator d-axis voltage, stator q-axis voltage, rotor d-axis voltage and rotor q-axis voltage, stator d-axis flux linkage, stator q-axis flux linkage, rotor d-axis flux linkage and rotor q-axis flux linkage, respectively. is the stator electrical angular velocity, is the rotor electrical angular velocity, , , are the stator inductance, rotor inductance and mutual inductance, , are the stator and rotor resistances; Given:

[0021] ;

[0022] Obtain the relationship between the flux linkage differential term and the current differential term,

[0023] ;

[0024] S12. Extract the flux linkage differential term in the voltage equation, let the slip ,

[0025] ;

[0026] S13. Substitute the flux linkage differential term of S12 into S11 to eliminate the flux linkage,

[0027] ;

[0028] S14. Write the state equation with the stator and rotor dq-axis currents of the doubly-fed induction machine as state variables,

[0029] 。

[0030] As a further improvement of the present invention, step 2) is specifically as follows:

[0031] S21. Establish the state equation F:

[0032] ;

[0033] X = [i ds , i qs , i dr , i qr , R s , R r , L s , L r , L m T ; ;

[0034] ;

[0035] The said X is the state variable, the said A is the Jacobian matrix, and the said H is the observation matrix;

[0036] S22. Use the rank criterion to judge:

[0037] ;

[0038] The observability of the system state variable meets the standard.

[0039] As a further improvement of the present invention, step 3) is specifically as follows:

[0040] S31. Initialize the extended Kalman filter algorithm:

[0041] Set the initial state vector and the initial error covariance matrix P0; set the initial values of the process noise covariance matrix Q0 and the observation noise covariance matrix R0; the said are the initial value of the stator d-axis current, the initial value of the stator q-axis current, the initial value of the rotor d-axis current, the initial value of the rotor q-axis current, the initial value of the stator inductance, the initial value of the rotor inductance, the initial value of the mutual inductance, the initial value of the stator resistance, and the initial value of the rotor resistance;

[0042] S32. State prediction:

[0043] Based on the current state estimate = [i ds , i qs , i dr , i qr , R​s ,R r ,L s ,L r ,L m T and the nine-dimensional state equation , predict the state at the next moment : ; the is the current state estimate value, the is the state estimate value at the current moment for the next moment, and the Δt is the sampling time interval; update the prediction error covariance matrix :

[0044] , is the noise covariance matrix at that moment;

[0045] the is the covariance matrix at the current moment for the next moment, the

[0046] ;

[0047] is the Jacobian matrix at the current moment; the is the partial derivative of the state equation F at the current moment with respect to the state variable X;

[0048] S33. Observation update:

[0049] Obtain the actual observation value at the next moment ; calculate the Kalman gain :

[0050]

[0051] the is the actual observation value at the next moment, the is the Kalman gain at the next moment, and the H is the observation matrix; correct the state estimate ;

[0052] ;

[0053] the is the state estimate value at the next moment for the next moment; update the error covariance matrix ;

[0054] ;

[0055] the covariance matrix at the next moment for the next moment, the is the covariance matrix at the current moment for the next moment. ​

[0056] S34. Adaptive adjustment of the noise covariance matrix:

[0057] Based on the residual sequence , dynamically update the observation noise covariance matrix :

[0058] ;

[0059] The described is to dynamically update the observation noise covariance matrix.

[0060] The described α is a smoothing factor, and its value range is ; Adjust the process noise covariance matrix based on the parameter estimation change rate :

[0061] ;

[0062] The described is a regulation coefficient, and the described Q0 is the initial process noise covariance;

[0063] S35. Loop iteration: Repeat steps S2 to S4, and output the identification parameters in real time .

[0064] As a further improvement of the present invention, the calculation of the residual sequence in the adaptive adjustment of the noise covariance matrix in step 3) includes a sliding window mechanism, specifically: calculate its mean value and covariance to suppress the interference of mutation noise; the window size is adaptively adjusted according to the dynamic response speed of the system, and the described window size is the last N residual data.

[0065] As a further improvement of the present invention, the values of the smoothing factor α and the regulation coefficient β in step 3) are calibrated through offline experiments or dynamically optimized online according to the parameter convergence speed.

[0066] Compared with the prior art, the beneficial effects of the present invention are:

[0067] First, by eliminating parameter coupling and optimizing in combination with the adaptive extended Kalman filter algorithm, the identification accuracy of resistance and inductance is significantly improved;

[0068] Second, the adaptive noise covariance adjustment mechanism simplifies the model complexity, combines the sliding window mechanism to optimize the calculation load, and effectively suppresses the measurement noise interference;

[0069] Third, skip the difficulty of flux linkage observation, reduce the observation cost, and lower the identification difficulty;

[0070] Fourth, realize the online identification of motor parameters. Description of the Drawings

[0071] Figure 1 is the flow chart of the state - space equation of the doubly - fed induction motor and its adaptive extended Kalman parameter identification method based on the stator and rotor dq - axis currents of the doubly - fed induction motor as state variables;

[0072] Figure 2 is the flow chart for modeling the state - space equation of the doubly - fed induction motor with the stator and rotor dq - axis currents of the doubly - fed induction motor as state variables;

[0073] Figure 3 is the flow chart for observability judgment;

[0074] Figure 4 is the adaptive extended Kalman algorithm for online identification of the stator resistance R s , rotor resistance R s , stator inductance L s , rotor inductance L r and stator - rotor mutual inductance L m flow chart;

[0075] Figure 5 is the simulation result graph of R s ;

[0076] Figure 6 is the simulation result graph of R s ;

[0077] Figure 7 is the simulation result graph of L s ;

[0078] Figure 8 is the simulation result graph of L r ;

[0079] Figure 9 is the simulation result graph of L m ; Specific implementation manners

[0080] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners:

[0081] Combined with Figure 1 as shown, for the state - space equation of the doubly - fed induction motor and its adaptive extended Kalman parameter identification method based on the stator and rotor dq - axis currents of the doubly - fed induction motor as state variables, by building a state equation with the stator and rotor dq - axis currents of the doubly - fed induction motor as state variables, building with [i ds , i qs , i dr , i qr , R s , R r , L s , Lr , L m , as the state equation F of the state variables, expand the state variables to nine dimensions. According to the observation matrix H and the Jacobian matrix A described above, use the rank criterion to judge the observability of the system, and use the ds , i qs , i dr , i qr , R s , R r , L s , L r , L m , as the adaptive extended Kalman filter algorithm of the state variables to identify the stator resistance R s , rotor resistance R r , stator inductance L s , rotor inductance L r and stator-rotor mutual inductance L m .

[0082] Combined Figure 2 as shown, the modeling of the state space equation of the doubly-fed induction motor with the stator and rotor dq-axis currents of the doubly-fed induction motor as state variables includes the following steps:

[0083] S1. According to the voltage equation and the flux linkage equation of the doubly-fed induction motor, convert the differential term of the flux linkage into the differential term of the current, and ignore the differential terms of the stator resistance R s , rotor resistance R r , stator inductance L s , rotor inductance L r and stator-rotor mutual inductance L m , that is , to obtain the mathematical relationship between the current differential and the flux linkage differential: Flux linkage equation: , , , .

[0084] Voltage equation: , , , , where , , , , , , , are the stator d-axis voltage, stator q-axis voltage, rotor d-axis voltage and rotor q-axis voltage, stator d-axis flux linkage, stator q-axis flux linkage, rotor d-axis flux linkage and rotor q-axis flux linkage respectively, is the stator electrical angular velocity, is the rotor electrical angular velocity, , , are the stator inductance, rotor inductance and mutual inductance, , are the stator and rotor resistances; Given:

[0085] ; Obtain the relationship between the differential flux linkage and the differential current,

[0086] ;

[0087] S2. Extract the differential flux linkage in the voltage equation and let ,

[0088] ;

[0089] S3. Substitute the differential flux linkage in S2 into S1 to eliminate the flux linkage,

[0090] ;

[0091] S4. Write the state equation with the stator and rotor dq-axis currents of the doubly-fed induction machine as state variables,

[0092] ;

[0093] Combined with Figure 3 as shown, for the state space equation of the doubly-fed induction machine with the stator and rotor dq-axis currents of the doubly-fed induction machine as state variables, incorporating the parameters to be identified into the state variables and expanding the dimension of the machine state space model, the steps for judging the parameter observability include the following:

[0094] S1. Build the state equation F:

[0095] ;

[0096] X = [i ds , i qs , i dr , i qr , R s , R r , L s , L r , L m T ;

[0097] ;

[0098] ;

[0099] S2. Use the rank criterion to judge:​

[0100] ;

[0101] Observability compliance of system state variables;

[0102] Combined with Figure 4 As shown, the dynamic online identification of the stator resistance R s , rotor resistance R r , stator inductance L s , rotor inductance L r and the mutual inductance L m between the stator and rotor of a doubly-fed induction motor is achieved by using the adaptive extended Kalman filter theory and the method of parameter iterative identification, including the following steps:

[0103] S1. Initialize the extended Kalman filter algorithm:

[0104] Set the initial state vector and the initial error covariance matrix P0; set the initial values of the process noise covariance matrix Q0 and the observation noise covariance matrix R0;

[0105] S2. State prediction: Based on the current state estimate = [i ds , i qs , i dr , i qr , R s , R r , L s , L r , L m T and the nine-dimensional state equation , predict the state at the next moment:

[0106] ;

[0107] where Δt is the sampling time interval; update the predicted error covariance matrix :

[0108] ;

[0109] where is the Jacobian matrix;

[0110] S3. Observation update: Obtain the actual observation value ; calculate the Kalman gain :

[0111] ;

[0112] where H is the observation matrix; correct the state estimate​ :

[0113] ;

[0114] Updated error covariance matrix :

[0115] ;

[0116] S4. Adaptive adjustment of noise covariance matrix:

[0117] According to the residual sequence , dynamically update the observation noise covariance matrix :

[0118] ;

[0119] The α mentioned above is a smoothing factor, and its value range is ; Adjust the process noise covariance matrix based on the change rate of parameter estimation :

[0120] ;

[0121] The mentioned above is an adjustment coefficient, and the Q0 mentioned above is the initial process noise covariance;

[0122] S5. Loop iteration: Repeat steps S2 to S4, and output the identification parameters in real time .

[0123] Combined with Figures 5 to 9 shown, use the adaptive extended Kalman algorithm to identify parameters , change the value of L at 0.3 seconds r , change the value of L at 0.5 seconds s , the identification result has high accuracy, fast response speed, and accurate tracking parameters.

[0124] The above is only a preferred embodiment of the present invention, and it is not a limitation of the present invention in any other form. Any modification or equivalent change made according to the technical essence of the present invention still belongs to the scope protected by the present invention.

Claims

1. A parameter identification method for a doubly-fed induction motor based on adaptive extended Kalman filter, characterized in that, Including the following steps: 1) Construct a state - space model of a doubly - fed induction motor with the stator and rotor d - q axis currents of the doubly - fed induction motor as state variables; 2) Based on the state - space equation of the doubly - fed induction motor with the stator and rotor d - q axis currents as state variables, incorporate the parameters to be identified into the state variables, expand the dimension of the motor state - space model, and judge the observability of the parameters; 3) Using the theory of adaptive extended Kalman filter and adopting the method of parameter iterative identification, the dynamic online identification of the stator resistance R s , rotor resistance R r , stator inductance L s , rotor inductance L r and the mutual inductance L m between the stator and rotor is realized.

2. The parameter identification method of the doubly-fed induction motor based on the adaptive extended Kalman filter according to claim 1, wherein The specific content of step 1) is as follows: S11. According to the voltage equation and flux linkage equation of the doubly-fed induction motor, convert the differential term of the flux linkage into the differential term of the current, and ignore the stator resistance R s , the rotor resistance R r , the stator inductance L s , the rotor inductance L r and the differential terms of the mutual inductance L m between the stator and the rotor, that is , where t is time, and obtain the mathematical relationship between the current differential and the flux linkage differential: Flux linkage equation: ， ， ， ； Voltage equation: ， ， ， ; where , , , , , , , , , , , are the stator d-axis current, stator q-axis current, rotor d-axis current and rotor q-axis current, stator d-axis voltage, stator q-axis voltage, rotor d-axis voltage and rotor q-axis voltage, stator d-axis flux linkage, stator q-axis flux linkage, rotor d-axis flux linkage and rotor q-axis flux linkage, respectively, is the electrical angular velocity of the stator, is the electrical angular velocity of the rotor, , , are the stator inductance, rotor inductance and mutual inductance, , are the stator and rotor resistances; Given: ； Obtain the relationship between the differential terms of the flux linkage and the differential terms of the current, ； S12. Extract the differential term of magnetic flux linkage in the voltage equation and set the slip , ； S13. Substitute the differential term of the flux linkage in S12 into S11 to eliminate the flux linkage, ； S14. Write the state equation with the stator and rotor d - q axis currents of the doubly - fed induction motor as state variables, 。 3. The parameter identification method for a doubly-fed induction motor based on adaptive extended Kalman filter according to claim 2, characterized in that, The specific content of step 2) is as follows: S21. Construct the state equation F: ； X = [i ds , i qs , i dr , i qr , R s , R r , L s , L r , L m T ; ;​ ； Where X is the state variable, A is the Jacobian matrix, and H is the observation matrix; S22. Use the rank criterion to judge: ； The observability of the system state variables meets the standard.

4. The parameter identification method of the doubly-fed induction motor based on the adaptive extended Kalman filter according to claim 3, characterized in that, The specific content of step 3) is as follows: S31. Initialize the extended Kalman filter algorithm: Set the initial state vector and the initial error covariance matrix P0; Set the initial values of the process noise covariance matrix Q0 and the observation noise covariance matrix R0; the are the initial value of the stator d-axis current, the initial value of the stator q-axis current, the initial value of the rotor d-axis current, the initial value of the rotor q-axis current, the initial value of the stator inductance, the initial value of the rotor inductance, and the initial value of the mutual inductance, the initial value of the stator resistance, and the initial value of the rotor resistance; S32. State prediction: Based on the current state estimate =[i ds ,i qs ,i dr ,i qr ,R s ,R r ,L s ,L r ,L m T and the nine-dimensional state equation , predict the state at the next moment : ; the is the current state estimate value, the is the state estimate value at the current moment for the next moment, and the Δt is the sampling time interval; update the prediction error covariance matrix :​ , is the noise covariance matrix at that time; The is the covariance matrix of the current moment with respect to the next moment. The ； is the Jacobian matrix at the current moment; the is the partial derivative of the state equation F with respect to the state variable X at the current moment; S33. Observation update: Obtain the actual observed value at the next moment ; Calculate the Kalman gain : ； The is the actual observed value at the next moment, and the is the Kalman gain at the next moment, where H is the observation matrix; the corrected state estimate : ； The described is the state estimate value for the next moment; update the error covariance matrix : ； The covariance matrix for the next moment with respect to the next moment, the is the covariance matrix for the next moment with respect to the current moment; S34. Adaptively adjust the noise covariance matrix: According to the residual sequence , dynamically update the observation noise covariance matrix : ； The described is to dynamically update the observation noise covariance matrix; The α is a smoothing factor, and its value range is ; Adjusting the process noise covariance matrix based on the change rate of parameter estimation : ； The described is the adjustment coefficient, and the described Q0 is the initial process noise covariance; S35. Loop iteration: Repeat steps S2 to S4 and output the identification parameters in real time. .

5. The parameter identification method of a doubly-fed induction motor based on adaptive extended Kalman filtering according to claim 4, characterized in that, In step 3), the calculation of the residual sequence in the adaptive adjustment of the noise covariance matrix includes a sliding window mechanism, specifically: calculating its mean and covariance to suppress the interference of mutation noise; the window size is adaptively adjusted according to the dynamic response speed of the system, and the window size is the last N residual data.

6. The parameter identification method of a doubly-fed induction motor based on adaptive extended Kalman filtering according to claim 4, characterized in that In step 3), the values of the smoothing factor α and the adjustment coefficient β are calibrated through offline experiments or dynamically optimized online according to the parameter convergence rate.

Images