Built-in permanent magnet synchronous motor MTPA control method based on improved EKF algorithm

Abstract

The invention discloses a built-in permanent magnet synchronous motor MTPA control method based on an improved EKF algorithm, and belongs to the field of permanent magnet synchronous motor high-performance control. According to the method, a dynamic forgetting factor multi-information extended Kalman filter algorithm is provided for solving the problems of MTPA trajectory offset and operation efficiency reduction caused by IPMSM key parameter time variation, and the algorithm adaptively adjusts historical data weight through a dynamic forgetting factor mechanism to realize rapid tracking when parameters suddenly change; and a multi-information theory is utilized to fuse historical information, so that noise is effectively suppressed, and the steady-state estimation precision is improved. The algorithm serves as an online identification module to be embedded into an MTPA control system, accurate quadrature-axis inductance and permanent magnet flux linkage can be obtained in real time, the current distribution angle is dynamically corrected, and therefore it is ensured that the motor stably operates along the optimal efficiency track under the full working condition. The method does not need extra signal injection, and has high precision, strong robustness and good engineering practicability.

Description

TECHNICAL FIELD

[0001] The application relates to the field of high-performance control of permanent magnet synchronous motors, and particularly relates to an MTPA control method for an interior permanent magnet synchronous motor based on an improved EKF algorithm. BACKGROUND

[0002] Interior permanent magnet synchronous motors (IPMSM) have become the mainstream choice in the fields of electric vehicle driving and industrial servo due to their large reluctance torque, high power density and excellent efficiency. When operating below the base speed, maximum torque per ampere (MTPA) control is a key technology for improving the operating efficiency of the IPMSM. The core of the MTPA control is to optimize the distribution ratio of the direct-axis current (id) and the quadrature-axis current (iq) so that the amplitude of the stator current is minimized when the same electromagnetic torque is output, thereby significantly reducing the copper loss and improving the efficiency. However, the MTPA control performance is highly dependent on the internal parameters of the motor, especially the quadrature-axis inductance (Lq) and the permanent magnet flux linkage (ψf). In actual operation, these parameters will drift due to factors such as temperature changes, magnetic circuit saturation and operating conditions, causing the MTPA trajectory calculated based on fixed parameters to deviate from the actual optimal solution, resulting in decreased efficiency and deteriorated dynamic response, which seriously restricts the full play of the performance of the IPMSM.

[0003] To address the problem of time-varying parameters, online parameter identification combined with MTPA control has become an important research direction. The current mainstream methods mainly include the extended Kalman filter (EKF) and the recursive least squares method (RLS), but they all have fundamental defects and cannot meet the stringent requirements of IPMSM-MTPA control for accuracy, real-time performance and disturbance-free performance: 1. The inherent limitations of the traditional EKF algorithm: Although the EKF can handle the joint estimation of the state and parameters of a nonlinear system, it uses a fixed forgetting factor, which leads to rigid weight distribution of historical data and cannot quickly track parameter mutations. In addition, its single-time innovation update mechanism is sensitive to measurement noise, which can easily lead to slow convergence speed and large estimation fluctuations. In complex dynamic conditions, the identification may lag, the covariance may be mismatched, and even divergence may occur.

[0004] 1. The inherent limitations of the traditional EKF algorithm: Although the EKF can handle the joint estimation of the state and parameters of a nonlinear system, it uses a fixed forgetting factor, which leads to rigid weight distribution of historical data and cannot quickly track parameter mutations. In addition, its single-time innovation update mechanism is sensitive to measurement noise, which can easily lead to slow convergence speed and large estimation fluctuations. In complex dynamic conditions, the identification may lag, the covariance may be mismatched, and even divergence may occur.

[0005] ​​​2. Disadvantages of Classical RLS and its Injection Method: Chinese invention patent CN110890855A discloses a parameter identification method using RLS. Although it introduces a forgetting factor, a low-frequency sinusoidal test signal must be injected into the motor windings to overcome the underrank problem of the identification equation. This active disturbance method has the following negative effects: First, it interferes with normal operation. The injected signal inevitably introduces additional current ripple and torque fluctuation, which destroys the stability of MTPA control and is unacceptable in high-performance servo and precision drive applications. Second, it increases system complexity, requiring the design of additional signal injection, separation, and filtering stages, increasing the computational burden and implementation difficulty of the controller. Third, its applicability is limited. The injection method usually needs to be triggered under specific operating conditions or steady state, and cannot achieve true full-condition, uninterrupted online identification, making it difficult to cope with sudden load or speed changes.

[0006] In summary, among existing technologies, the traditional EKF cannot simultaneously achieve dynamic tracking and steady-state disturbance rejection due to its rigid structure; while the RLS injection method sacrifices control performance and system simplicity for identifiability. Neither of these approaches fundamentally solves the problem of time-varying parameter correction in IPMSM-MTPA control.

[0007] Therefore, designing an online identification algorithm that can adaptively balance fast tracking and high-precision disturbance rejection without the need for external signal injection, and seamlessly embedding it into the MTPA control closed loop, has become a technical bottleneck that urgently needs to be overcome in this field. Summary of the Invention

[0008] The purpose of this invention is to provide a built-in permanent magnet synchronous motor MTPA control method based on an improved EKF algorithm. This method achieves high-precision and fast online identification of IPMSM quadrature axis inductance and permanent magnet flux linkage without injecting any external test signals into the motor. Based on this identification, the maximum torque-to-current ratio (MTPA) control trajectory is dynamically corrected in real time, thereby ensuring that the motor can continuously and stably operate at its optimal efficiency under time-varying parameters and complex operating conditions.

[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The MTPA control method for a built-in permanent magnet synchronous motor based on the improved EKF algorithm includes the following steps: S1. Construct a control system for a built-in permanent magnet synchronous motor. The control system includes a speed outer loop, an MTPA calculation module, a current inner loop, and a DFF-MIEKF identification module based on a dynamic forgetting factor multi-innovation extended Kalman filter. S2. Design a dynamic forgetting factor adaptive mechanism and construct a forgetting factor update formula based on dynamic adjustment of the innovation norm; S3. Construct an extended vector containing information from multiple consecutive historical moments, and based on the extended vector and the dynamic forgetting factor, construct a multi-time gain matrix and a state estimation update equation to form the DFF-MIEKF algorithm. S4. Discretize the voltage equation of the built-in permanent magnet synchronous motor in the dq synchronous rotating coordinate system. Use the direct axis current, quadrature axis current, quadrature axis inductance and permanent magnet flux linkage as state variables to construct identification equation and observation equation. Obtain the state transition function and the corresponding Jacobian matrix through forward Euler discretization, thereby establishing a discrete state space model suitable for the DFF-MIEKF algorithm. S5. During the operation of the control system, the DFF-MIEKF identification module receives motor operation data in real time, executes the DFF-MIEKF algorithm, identifies the real-time estimated values ​​of quadrature axis inductance and permanent magnet flux linkage online, and feeds them back to the MTPA calculation module. S6. The MTPA calculation module dynamically calculates the optimal current distribution angle and the corresponding direct-axis current setpoint and quadrature-axis current setpoint based on the current stator current amplitude command and the real-time estimated values ​​of the quadrature-axis inductance and permanent magnet flux identified in step S5, thereby realizing real-time online correction of the maximum torque-current ratio trajectory.

[0010] Furthermore, in step S2, the forgetting factor update formula is: In the formula, λ is the dynamic forgetting factor. The lower limit of the forgetting factor is set. For adjustment coefficients, For weighted historical new information norm.

[0011] Furthermore, the weighted historical norm Based on historical multi-innovation matrix The historical multi-innovation matrix is ​​calculated to obtain... It consists of the current time and the information from the previous p-1 consecutive time steps. The expression is: In the formula, p is the length of the multiple information window. This is a single-step update.

[0012] Furthermore, in step S3, the expression for the extended vector containing information from multiple consecutive historical moments is: In the formula, p represents the current information at time k, where p is the length of the multiple information window. ,in, Represents the observed value. Indicates the predicted value; The expression for the multi-time gain matrix is: In the formula, For more information gain, This represents the Kalman gain matrix at the i-th step in the past. It is a dynamic forgetting factor; The state estimation update equation is: In the formula, The state estimate is based on time k. For the predicted state based on time k-1, This indicates the total correction amount.

[0013] Further, in step S4, the identification equation is: In the formula, , These are the stator voltages along the d-axis and q-axis, respectively; and These are the stator current components along the d-axis and q-axis of the IPMSM, respectively. and The inductances along the d-axis and q-axis are respectively. For permanent magnet flux linkage; For stator resistance, ω is the rotor rotational angular velocity.

[0014] Furthermore, in step S5, the covariance update process during the execution of the DFF-MIEKF algorithm includes: Prior covariance matrix The update equation is: In the formula, The state transition Jacobian matrix from the previous time step is... The posterior covariance of the previous time step; The process noise covariance matrix; The update equation for the posterior covariance matrix is: In the formula, Let be the posterior covariance matrix at time k. As a dynamic forgetting factor, It is the identity matrix. R is the multi-innovation gain, and R is the observation noise covariance matrix. For the observation matrix, It is the prior covariance matrix. Furthermore, in step S5, the implementation process of executing the DFF-MIEKF algorithm includes: Construct the calculation expressions for multiple innovations and the gain matrix: Then, the state correction equation is constructed: In the formula, Let be the prior state variable at time k. Let K be the Kalman gain matrix at time k. Let k be the innovation vector at time k; Let H be the observation vector at time k, H be the observation matrix, and R be the observation noise covariance matrix. It is the prior covariance matrix.

[0015] Furthermore, in step S6, the expression for calculating the optimal current distribution angle is: In the formula, α is the optimal current distribution angle. L is the real-time estimated value of the permanent magnet flux linkage, and Lq is the real-time estimated value of the quadrature-axis inductance. It is a direct-axis inductor. Provides a signal for the stator current amplitude; Direct-axis current setpoint and quadrature axis current setpoint The calculation expressions are as follows: .

[0016] Furthermore, the lower limit of the forgetting factor The value range is from 0.85 to 0.98, and the adjustment coefficient is... The value ranges from 0.1 to 2.0, and the value range of the multi-information window length p is from 3 to 8.

[0017] Furthermore, the lower limit of the forgetting factor Adjustment coefficient The steps for co-tuning the length p of the multiple information windows are as follows: (1) Initialize parameters; (2) Steady-state operation observation: If the parameter estimation fluctuates too much, increase Or p; if steady-state error still exists after convergence, decrease it. ; (3) Dynamic operating condition observation: If parameter tracking is too slow, reduce or increase If the overshoot is too large or there is oscillation, increase the [adjustment]. or reduce ; (4) Repeat steady-state operation observation and dynamic operating condition observation until a balance is reached between steady-state accuracy and dynamic tracking speed.

[0018] The beneficial effects of the above scheme are as follows: (1) This invention achieves a significant breakthrough in parameter identification performance. By integrating the dynamic forgetting factor and multiple innovation theory, the novel algorithm intelligently balances the contradiction between fast tracking and steady-state noise resistance. It can not only respond quickly when parameters change abruptly, effectively shortening the convergence process, but also greatly suppress noise interference during stable operation, thereby achieving estimation accuracy and robustness far exceeding traditional methods, laying a reliable foundation for high-precision control.

[0019] Simulation and experimental results show that, compared with the traditional EKF algorithm, the DFF-MIEKF algorithm significantly improves performance for parameter identification: In permanent magnet flux linkage identification, the peak identification response decreases from 1.15 Wb to 0.8 Wb, and the follower overshoot is reduced by approximately 28.57%; the identification convergence time is approximately 0.13 seconds, and the steady-state identification error decreases from 3.28% to 0.0035%, essentially achieving zero steady-state error identification. In quadrature-axis inductor identification, the peak identification response decreases from 15.1 mH to 14.4 mH, and the follower overshoot is reduced by approximately 4.64%; the identification convergence time is shortened from 0.5 seconds to 0.3 seconds, increasing the convergence speed by approximately 40%; and the steady-state identification error decreases from 0.1% to 0.028%, also achieving near-zero steady-state error high-precision identification.

[0020] (2) Based on the above-mentioned high-precision real-time parameter identification, the control system of the present invention can ensure that the motor always runs along the optimal efficiency trajectory. The system can dynamically correct the current distribution deviation caused by parameter drift, so that the direct axis and quadrature axis currents can accurately track their optimal set values ​​under various operating conditions, and the current distribution deviation is close to zero. Whether in the dynamic process of sudden load addition and unloading or in steady-state operation, the system can maintain excellent torque response speed, operating efficiency and anti-interference ability, significantly improving the overall control performance.

[0021] (3) In terms of engineering applications, this invention has outstanding practical value and is easy to promote. The entire solution does not require the injection of any test signal, avoiding additional ripple and hardware costs, and can be implemented using only conventional sensors and controllers. The algorithm has moderate computational complexity and is easy to deploy on existing embedded platforms such as DSPs and FPGAs. In addition, this core identification method has good versatility and can be easily ported to other types of motor drive systems, with broad application prospects. Attached Figure Description

[0022] Figure 1 This is a structural diagram of the MTPA control system based on DFF-MIEKF of this invention; Figure 2 This is a schematic diagram of the adaptive adjustment process of the dynamic forgetting factor in an embodiment of the present invention; Figure 3 This is a schematic diagram of the multi-innovation vector construction process in an embodiment of the present invention. Detailed Implementation

[0023] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0024] It should be noted that, unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0025] This invention provides a control method for an embedded permanent magnet synchronous motor (MTPA) based on an improved EKF algorithm. This method creatively integrates the dynamic forgetting factor mechanism with multiple innovation theory to construct a novel dynamic forgetting factor multiple innovation extended Kalman filter (DFF-MIEKF) algorithm. This algorithm is used to identify key time-varying parameters of the IPMSM in real time and with high precision. The identification results are then used to dynamically correct the MTPA control trajectory online, thereby ensuring that the motor operates stably along the optimal efficiency trajectory across the entire operating range. The method includes the following steps: S1. Construct a control system for a built-in permanent magnet synchronous motor. The control system includes a speed outer loop, an MTPA calculation module, a current inner loop, and a DFF-MIEKF identification module based on a dynamic forgetting factor multi-innovation extended Kalman filter. S2. Design a dynamic forgetting factor adaptive mechanism and construct a forgetting factor update formula based on dynamic adjustment of the innovation norm; S3. Construct an extended vector containing information from multiple consecutive historical moments, and based on the extended vector and the dynamic forgetting factor, construct a multi-time gain matrix and a state estimation update equation to form the DFF-MIEKF algorithm. S4. Discretize the voltage equation of the built-in permanent magnet synchronous motor in the dq synchronous rotating coordinate system. Use the direct axis current, quadrature axis current, quadrature axis inductance and permanent magnet flux linkage as state variables to construct identification equation and observation equation. Obtain the state transition function and the corresponding Jacobian matrix through forward Euler discretization, thereby establishing a discrete state space model suitable for the DFF-MIEKF algorithm. S5. During the operation of the control system, the DFF-MIEKF identification module receives motor operation data in real time, executes the DFF-MIEKF algorithm, identifies the real-time estimated values ​​of quadrature axis inductance and permanent magnet flux linkage online, and feeds them back to the MTPA calculation module. S6. The MTPA calculation module dynamically calculates the optimal current distribution angle and the corresponding direct-axis current setpoint and quadrature-axis current setpoint based on the current stator current amplitude command and the real-time estimated values ​​of the quadrature-axis inductance and permanent magnet flux identified in step S5, thereby realizing real-time online correction of the maximum torque-current ratio trajectory.

[0026] The specific implementation process of the present invention will be described in detail below.

[0027] (I) Constructing the overall architecture of the IPMSM control system First, the overall architecture of the control system of this invention is constructed. Figure 1 This is a block diagram of the MTPA control system based on the DFF-MIEKF-based IPMSM. It mainly includes an outer speed loop, an MTPA calculation module, an inner current loop, and a DFF-MIEKF identification module.

[0028] (1) Speed ​​outer loop: The input of the system is the speed command signal ω. The speed outer loop compares the speed command signal ω with the speed feedback signal ω detected by the sensor. The resulting speed error signal is processed by the speed PI controller and outputs the stator current amplitude command signal. This process falls under the category of conventional closed-loop speed control.

[0029] (2) MTPA calculation module: used to receive current amplitude commands from the speed outer loop and quadrature axis inductance Lq and permanent magnet flux linkage from the DFF-MIEKF identification module. The real-time estimated value.

[0030] First, define dq as the synchronous rotating coordinate system of the rotor magnetic field in IPMSM; then, give a signal based on the current stator current amplitude. And the IPMSM permanent magnet flux identified by the DFF-MIEKF module Real-time values ​​of the stator current and quadrature-axis inductance Lq are used to dynamically solve for the optimal current distribution angle α. Based on this, the d-axis component of the stator current command signal required for the optimal torque-to-current ratio (MTPA) control effect is calculated. and q-axis components .

[0031] The dynamic solution expression for the optimal current distribution angle α is as follows: (1) The expressions for the d-axis and q-axis components of the stator current command signal are as follows: (2) (3) Inner current loop: Used to achieve accurate current tracking. The current sensor measures the A-phase and B-phase currents of the motor's three-phase current. and After Clark and Park transformations, the direct-axis and quadrature-axis current feedback signals in the dq coordinate system under synchronous rotation of the rotor magnetic field are obtained. and The feedback current is compared with the given current to obtain the d-axis current deviation and q-axis current deviation, which are then processed by a PI controller to generate the stator voltage command signal in the dq coordinate system. and Subsequently, through inverse Park transformation, space vector pulse width modulation (SVPWM), and a three-phase inverter, a three-phase AC voltage is generated to act on the stator windings of the IPMSM, thereby driving the motor to operate under precise MTPA conditions.

[0032] (4) DFF-MIEKF identification module: As the intelligent observer of this invention, it is embedded in the above-mentioned control loop. The DFF-MIEKF identification module receives measurable physical quantities during the motor operation process in real time, including direct-axis and quadrature-axis voltages. and Direct-axis and quadrature-axis currents and And the rotor electrical angular velocity ω. This module executes the DFF-MIEKF algorithm to estimate the motor parameters online. and The estimated value is then fed back to the MTPA calculation module in real time.

[0033] (II) Design of an adaptive mechanism for dynamic forgetting factor To ensure the algorithm has a fast response speed when parameters change rapidly and maintains high recognition accuracy in steady state, the following dynamic forgetting factor update formula is defined: (3) In the formula, λ is the dynamic forgetting factor. The lower limit of the forgetting factor is set. For adjustment coefficients, For weighted historical new information norm.

[0034] The lower limit of the forgetting factor determines the minimum weight of historical data that the algorithm is willing to retain when dealing with drastic dynamics. It is usually selected between 0.85 and 0.98 depending on the characteristics of the IPMSM system. The adjustment coefficient controls the sensitivity of the dynamic forgetting factor λ to changes in information, with a typical value range of 0.1 to 2.0; This reflects the recent dynamic level of the system.

[0035] The closer the value of λ is to 1, the stronger the "memory" of the identification algorithm in the dynamic process and the better the noise resistance, but the identification and tracking of abrupt parameters is slightly lagging; the smaller the value of λ, the faster the identification and tracking speed, but it may be more sensitive to noise.

[0036] Table 1 shows the selection principles and tuning suggestions for the core parameters of the DFF-MIEKF algorithm. The lower limit parameter of the forgetting factor is also provided for different application scenarios. The setting principle is: (1) For systems like IPMSM with moderate electromechanical time constants, the following is usually selected Between 0.85 and 0.98; (2) If the load or speed changes frequently in the application scenario (such as rapid acceleration / deceleration of electric vehicles), a lower speed can be selected to prioritize ensuring the tracking speed. Values, such as 0.85~0.92; (3) If the application scenario focuses more on high tracking accuracy and low ripple under steady or quasi-steady conditions (such as precision servo, UAV hovering), a higher μ value, such as 0.95~0.98, should be selected in order to make full use of historical data to smooth noise.

[0037] The larger the value, the faster the dynamic forgetting factor decreases as information increases, resulting in a lower threshold and a faster switching speed for the identification algorithm to "fast tracking mode". Parameter The selection of is related to the system sampling period, the response speed of the current / velocity loop, and the expected rate of change of parameters. Its typical value range is between 0.1 and 2.0. For applications with high dynamic response requirements, the value can be taken in the range of 0.5 to 1.5.

[0038] In equation (3) and Let be the historical multiple innovation matrix and its corresponding weighted historical innovation norm, respectively, and their expressions are as follows: (4) (5) In the formula, p is the length of the multiple information window, and its value ranges from 3 to 8; This is a single-step update.

[0039] The choice of the multi-information window length p needs to balance dynamic response speed and noise suppression capability. For motor control systems with sampling frequency of 5-20kHz, the recommended value of p is 3 to 8. Among them, when p=4 or p=5, good noise smoothing effect can be achieved without introducing obvious lag, which is the preferred solution that balances performance and complexity.

[0040] Based on equations (3)-(5), the variation law of the set dynamic forgetting factor can be summarized as follows: (1) When the norm of the new information is relatively large (i.e., the system undergoes drastic dynamic changes), the dynamic forgetting factor... As the value approaches 0, the identification algorithm will be more inclined to forget old data in order to quickly track parameter changes, which can also prevent IPMSM from saturating. (2) When the norm of the new information is relatively small (i.e., the system approaches stability), the dynamic forgetting factor The value will approach 1, and the identification algorithm will be more inclined to retain more historical data in order to suppress the influence of noise.

[0041] like Figure 2 The diagram illustrates the adaptive adjustment process of the dynamic forgetting factor λ. It visually demonstrates how the dynamic forgetting factor λ adapts to the system's innovation norm. When the innovation norm is large, λ approaches the lower limit μ, allowing the algorithm to quickly forget old data and enhance parameter tracking capabilities. When the innovation norm is small, λ approaches 1, allowing the algorithm to retain more historical information to suppress noise and improve estimation accuracy. This mechanism enables DFF-MIEKF to intelligently switch between "fast tracking" and "steady-state noise suppression," solving the tracking lag and noise sensitivity problems caused by the fixed forgetting factor in traditional EKF.

[0042] (III) Constructing multi-innovation extension vectors, multi-time gain matrices, and state estimation update equations To address the issues of traditional EKF single-innovation updates being sensitive to noise and having slow convergence, this invention introduces the multi-innovation theory.

[0043] First, construct an extended vector that integrates multi-time-phase information: (6) In the formula, This is the new information at time k. p represents the length of the multiple information windows, and the setting principles are shown in Table 1.

[0044] Figure 3 The diagram illustrates the process of constructing a multi-inspiration extension vector, demonstrating how to utilize the inspirations from multiple consecutive historical moments to construct an extension vector, replacing the single-moment inspiration in the traditional EKF algorithm. By fusing inspirations from multiple moments, the algorithm can smooth out the influence of random noise, improving the reliability and convergence speed of state estimation. Furthermore, by combining a dynamic forgetting factor λ to assign different weights to inspirations at different moments, the efficiency of utilizing historical information is further optimized, enhancing the algorithm's robustness under complex conditions and overcoming the problems of slow convergence and susceptibility to interference caused by single-inspiration updates in traditional methods.

[0045] Based on the aforementioned extended vector and dynamic forgetting factor, construct the corresponding multi-time gain matrix: (7) In the formula, For more information gain, This represents the Kalman gain matrix at the i-th step in the past. This is a dynamic forgetting factor used to weight the gain at different times.

[0046] Finally, the state estimation update equation that integrates multiple innovation theory and dynamic forgetting factor is as follows: (8) In the formula, The state estimate is based on time k. For the predicted state based on time k-1, This indicates the total correction amount.

[0047] (iv) Constructing the complete DFF-MIEKF algorithm Based on the voltage equation of IPMSM in the dq synchronous rotating coordinate system, the derivative expressions of the quadrature and direct axis currents can be obtained as follows: (9) In the formula, , These are the stator voltages along the d-axis and q-axis, respectively; and These are the stator current components along the d-axis and q-axis, respectively; and Let be the inductances along the d-axis and q-axis, respectively. For permanent magnet flux linkage; For stator resistance, ω is the rotor rotational angular velocity.

[0048] Extend equation (9) to a sixth-order nonlinear coupled model of IPMSM: (10) because This is unrelated to MTPA control and is considered a constant here. Additionally, when the motor operates below its rated speed, constant rotor flux linkage control is generally used, in which case the direct-axis inductance... The change is minimal when operating below the rated speed, and is therefore considered constant here. The current... , and parameters to be identified , As a state variable: (11) Given and Since it is a slowly varying parameter, its derivative term is not considered here. Instead, it is continuously and slowly updated based on state estimation (i.e., calculated from the current AC and DC axis current data). Therefore, the identification equation can be obtained from equation (10): (12) The current moment can be obtained by real-time measurement during motor operation. , The following observation equation is constructed: (13) Based on the state variables defined in equation (11), the observation matrix H can be obtained: (14) (v) Execution of the DFF-MIEKF algorithm By discretizing the identification equation (12) using forward Euler method, the state transition function in the discrete state can be obtained. Based on the current input and the state at the previous time step, predict the state at the current time step: (15) The inputs are the current direct-axis voltage value and rotational angular velocity, i.e.: (16) Based on the state transition function obtained in the discrete state Taking the partial derivative with respect to state x, we can derive the Jacobian matrix F as follows: (17) The dynamic forgetting factor introduced by combining equation (3) It dynamically adjusts based on the current innovation norm to reduce the weight of historical covariance when parameters change abruptly. The covariance update equation is set as follows: (18) In the formula, R is the observation noise covariance matrix. It is the identity matrix. It is the prior covariance matrix, and its expression is: (19) In the formula, The state transition Jacobian matrix of the previous time step reflects the influence of current change and parameter coupling on the propagation of prediction error. The posterior covariance of the previous time step; Let be the process noise covariance matrix.

[0049] Based on the constructed multi-time innovation extension vector (see equation (6)) and the dynamic forgetting factor update mechanism (see equation (3)), the expression for calculating the multi-innovation and gain matrix is ​​constructed as follows: (20) Based on the above, the following state correction equation is constructed: (twenty one) In the formula: For the current moment's news; H is the observation vector at the current moment, containing the current currents along the perpendicular and perpendicular axes at the current moment; H is the observation matrix (see equation (14)). This is the gain matrix.

[0050] This step outputs the updated state estimate, which includes the quadrature-axis inductance Lq and the permanent magnet flux linkage. The real-time estimated value.

[0051] (vi) Real-time correction of MTPA trajectory based on online parameters The high-precision parameter estimates output by the DFF-MIEKF identification module are fed into the MTPA calculation module in real time for dynamic correction of the control trajectory.

[0052] The MTPA calculation module calculates the value of the received stator current based on the received stator current amplitude command. Estimated values ​​of quadrature-axis inductance Lq and permanent magnet flux linkage The optimal current distribution angle α is dynamically solved, and then the optimal direct-axis and quadrature-axis current values ​​are calculated.

[0053] These current setpoints, calculated based on real-time and accurate parameters, are fed into the inner current loop. Therefore, even when actual motor parameters drift due to factors such as temperature rise and saturation, this invention can ensure that the current vector is always locked on the theoretically optimal MTPA trajectory through a closed loop of "sensing-decision-execution," thereby guaranteeing the continuous and efficient operation of the motor system under various complex time-varying conditions.

[0054] (vii) Parameter tuning and implementation effect To achieve optimal performance, Table 1 provides the selection principles and typical initialization suggestions for key parameters in the DFF-MIEKF algorithm. It summarizes the physical meaning and mechanism of action of three key parameters: the lower limit of the forgetting factor μ, the adjustment coefficient γ, and the length of the multiple innovation window p, along with suggested value ranges and typical initialization values, providing clear parameter tuning guidance for engineering implementation. Specifically, μ determines the minimum value of the dynamic forgetting factor λ, affecting the weight distribution between dynamic and steady-state states; γ controls the sensitivity of λ to innovation changes; and p determines the depth of historical information fusion, affecting the balance between noise smoothing and dynamic response. The table also provides co-tuning steps to guide users in adjusting parameters according to the actual system response to achieve optimal performance of the algorithm in specific application scenarios.

[0055] Table 1. Selection Principles and Tuning Suggestions for the Core Parameters of the DFF-MIEKF Algorithm Note: The above parameters need to be tuned together to achieve optimal performance. It is recommended to follow these steps: (1) Use the “typical values” in the table to set (μ=0.92, γ=1.0, p=4).

[0056] (2) When the motor is running under no-load steady-state conditions, observe the fluctuation (noise) of the parameter estimation. If the fluctuation is too large, μ or p can be increased appropriately; if there is still a steady-state error after convergence, μ can be decreased slightly.

[0057] (3) Apply a step load or speed command and observe the tracking speed and overshoot of the parameter identification. If the tracking is too slow, μ can be appropriately reduced or γ can be increased; if the overshoot is too large or there is oscillation, μ can be appropriately increased or γ can be decreased.

[0058] (4) Repeat steps (2)-(3) until a satisfactory balance is achieved between steady-state accuracy and dynamic tracking speed.

[0059] Through this invention, the parameter identification of IPMSM achieves a synergistic improvement in accuracy and speed. The control system can still maintain accurate tracking of the current to the expected value of MTPA even when the parameters change over time. It exhibits faster recovery capability and stronger robustness when the load or speed changes abruptly, ultimately ensuring the optimal efficiency of the drive system across the entire operating range.

[0060] Finally, it should be noted that any parts of this invention not described in detail are prior art. Those skilled in the art will understand that the above descriptions are merely preferred embodiments of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the invention should be included within the scope of protection of the invention.

Claims

1. A control method for a built-in permanent magnet synchronous motor (MTPA) based on an improved EKF algorithm, characterized in that, Includes the following steps: S1. Construct a control system for the built-in permanent magnet synchronous motor. The control system includes a speed outer loop, an MTPA calculation module, a current inner loop, and a DFF-MIEKF identification module based on a dynamic forgetting factor multi-innovation extended Kalman filter. S2. Design a dynamic forgetting factor adaptive mechanism and construct a forgetting factor update formula based on dynamic adjustment of the innovation norm; S3. Construct an extended vector containing information from multiple consecutive historical moments, and based on the extended vector and the dynamic forgetting factor, construct a multi-time gain matrix and a state estimation update equation to form the DFF-MIEKF algorithm. S4. Discretize the voltage equation of the built-in permanent magnet synchronous motor in the dq synchronous rotating coordinate system. Use the direct axis current, quadrature axis current, quadrature axis inductance and permanent magnet flux linkage as state variables to construct identification equation and observation equation. Obtain the state transition function and the corresponding Jacobian matrix through forward Euler discretization, thereby establishing a discrete state space model suitable for the DFF-MIEKF algorithm. S5. During the operation of the control system, the DFF-MIEKF identification module receives motor operation data in real time, executes the DFF-MIEKF algorithm, identifies the real-time estimated values ​​of quadrature axis inductance and permanent magnet flux linkage online, and feeds them back to the MTPA calculation module. S6. The MTPA calculation module dynamically calculates the optimal current distribution angle and the corresponding direct-axis current setpoint and quadrature-axis current setpoint based on the current stator current amplitude command and the real-time estimated values ​​of the quadrature-axis inductance and permanent magnet flux identified in step S5, thereby realizing real-time online correction of the maximum torque-current ratio trajectory.

2. The MTPA control method for an embedded permanent magnet synchronous motor based on an improved EKF algorithm according to claim 1, characterized in that, In step S2, the forgetting factor update formula is: In the formula, λ is the dynamic forgetting factor. The lower limit of the forgetting factor is set. For adjustment coefficients, For weighted historical new information norm.

3. The MTPA control method for an embedded permanent magnet synchronous motor based on an improved EKF algorithm according to claim 2, characterized in that, The weighted historical norm Based on historical multi-innovation matrix The historical multi-innovation matrix is ​​calculated to obtain... It consists of the current time and the information from the previous p-1 consecutive time steps. The expression is: In the formula, p is the length of the multiple information window. This is a single-step update.

4. The MTPA control method for a built-in permanent magnet synchronous motor based on the improved EKF algorithm according to claim 3, characterized in that, In step S3, the expression for the extended vector containing information from multiple consecutive historical moments is: In the formula, p represents the current information at time k, where p is the length of the multiple information window. ,in, Represents the observed value. Indicates the predicted value; The expression for the multi-time gain matrix is: In the formula, For more information gain, This represents the Kalman gain matrix at the i-th step in the past. It is a dynamic forgetting factor; The state estimation update equation is: In the formula, The state estimate is based on time k. For the predicted state based on time k-1, This indicates the total correction amount.

5. The MTPA control method for an embedded permanent magnet synchronous motor based on an improved EKF algorithm according to claim 1, characterized in that, In step S4, the identification equation is: In the formula, , These are the stator voltages along the d-axis and q-axis, respectively; and These are the stator current components along the d-axis and q-axis of the IPMSM, respectively. and The inductances along the d-axis and q-axis are respectively. For permanent magnet flux linkage; For stator resistance, ω is the angular velocity of the rotor.

6. The MTPA control method for an embedded permanent magnet synchronous motor based on an improved EKF algorithm according to claim 1, characterized in that, In step S5, the covariance update process during the execution of the DFF-MIEKF algorithm includes: Prior covariance matrix The update equation is: In the formula, The state transition Jacobian matrix from the previous time step is... The posterior covariance of the previous time step; The process noise covariance matrix; The update equation for the posterior covariance matrix is: In the formula, Let be the posterior covariance matrix at time k. As a dynamic forgetting factor, It is the identity matrix. R is the multi-innovation gain, and R is the observation noise covariance matrix. For the observation matrix, It is the prior covariance matrix.

7. The MTPA control method for an embedded permanent magnet synchronous motor based on an improved EKF algorithm according to claim 1, characterized in that, In step S5, the implementation process of executing the DFF-MIEKF algorithm includes: Construct the calculation expressions for multiple innovations and the gain matrix: Then, the state correction equation is constructed: In the formula, Let be the prior state variable at time k. Let K be the Kalman gain matrix at time k. Let k be the innovation vector at time k; Let H be the observation vector at time k, H be the observation matrix, and R be the observation noise covariance matrix. It is the prior covariance matrix.

8. The MTPA control method for an embedded permanent magnet synchronous motor based on an improved EKF algorithm according to claim 1, characterized in that, In step S6, the expression for calculating the optimal current distribution angle is: In the formula, α is the optimal current distribution angle. L is the real-time estimated value of the permanent magnet flux linkage, and Lq is the real-time estimated value of the quadrature-axis inductance. It is a direct-axis inductor. Provides a signal for the stator current amplitude; Direct-axis current setpoint and quadrature axis current setpoint The calculation expressions are as follows: 。 9. The MTPA control method for an embedded permanent magnet synchronous motor based on an improved EKF algorithm according to claim 3, characterized in that, The lower limit of the forgetting factor The value range is from 0.85 to 0.98, and the adjustment coefficient is... The value ranges from 0.1 to 2.0, and the value range of the multi-information window length p is from 3 to 8.

10. The MTPA control method for a built-in permanent magnet synchronous motor based on the improved EKF algorithm according to claim 9, characterized in that, Lower limit of forgetting factor Adjustment coefficient The steps for co-tuning the length p of the multiple information windows are as follows: (1) Initialize parameters; (2) Steady-state operation observation: If the parameter estimation fluctuates too much, increase Or p; if steady-state error still exists after convergence, decrease it. ; (3) Dynamic operating condition observation: If parameter tracking is too slow, reduce or increase If the overshoot is too large or there is oscillation, increase the [adjustment]. or reduce ; (4) Repeat steady-state operation observation and dynamic operating condition observation until a balance is reached between steady-state accuracy and dynamic tracking speed.

Images