A permanent magnet synchronous motor position sensorless control method based on dynamic weight residual screening adaptive EKF

Abstract

The application discloses a permanent magnet synchronous motor position sensorless control method based on dynamic weight residual screening adaptive EKF, residual energy moving average values of a recent interval and a history interval are represented by working condition characteristic quantities, a working condition perception coefficient r of a permanent magnet synchronous motor at a k moment is calculated k ; r k is converted into a residual weight basic coefficient, a weight coefficient of an i-th historical residual at the moment is calculated, and an effective residual set matched with a current working condition is constructed; the effective residual is weighted and fused to obtain a residual fusion value, an innovation covariance matrix is calculated, a Kalman gain matrix is calculated from an observation noise covariance matrix and a filter error covariance matrix prediction value of a last moment, and observation and process noise covariance matrices are updated; a state vector at the moment is updated from the Kalman gain matrix and a filter residual, a motor rotor state estimation value is obtained, and is fed back to a permanent magnet synchronous motor vector control system. The application enhances the robustness and anti-interference ability of the motor in a high-speed working condition.

Description

Technical Field

[0001] This invention belongs to the field of sensorless control technology for permanent magnet synchronous motors (PMSMs), specifically involving an adaptive covariance matrix adjustment method based on extended Kalman filtering (EKF). It primarily optimizes the observation performance and control robustness of the motor under high-speed, full-condition operation. It can be adapted to industry-standard zero-speed / low-speed start-up strategies, forming a complete sensorless control solution. This solution is widely used in high-performance drive scenarios for permanent magnet synchronous motors, such as industrial servo drives, new energy vehicles, lifting equipment, AGVs, and robot joints, where high precision in motor control, dynamic response, and operational stability are required. Background Technology

[0002] Permanent magnet synchronous motors (PMSMs), with their advantages of high efficiency, high power density, high dynamic response, and low torque ripple, have gradually replaced asynchronous motors and become the core technology in industrial drives, new energy transportation, and other fields. Sensorless control technology eliminates the need for mechanical position sensors such as photoelectric encoders and rotary transformers, significantly reducing system hardware costs and improving the operational reliability of equipment in harsh environments such as humidity, dust, and strong vibration. This is currently an important research direction in the field of high-performance PSM drives.

[0003] Extended Kalman Filter (EKF) can achieve optimal rotor state estimation for nonlinear systems such as permanent magnet synchronous motors in noisy industrial environments. It boasts advantages such as clear algorithm logic, high steady-state accuracy, and strong anti-interference capabilities, making it a widely used sensorless observation algorithm in engineering. In the traditional engineering implementation of EKF, the industry commonly uses a fixed-length sliding window to statistically average the filter residuals to update the covariance matrix parameters. This approach has a low implementation threshold, simple logic, and provides basic observation results under steady-state conditions, making it the currently common EKF engineering solution in the industry.

[0004] However, the traditional fixed-window EKF scheme has significant bottlenecks in high-speed, full-condition operation of permanent magnet synchronous motors, mainly in the following three aspects:

[0005] 1. Fixed windows cannot simultaneously achieve both steady-state smoothness and dynamic response. Furthermore, the length of a fixed window cannot be adjusted once set, while the different operating conditions of a motor have completely opposite requirements for window length: steady-state conditions require a long window to ensure smooth filtering and suppress the influence of sampling noise; dynamic conditions such as sudden load changes, acceleration, and deceleration require a short window to ensure rapid response and avoid observation lag caused by over-filtering. Fixed windows cannot simultaneously meet the needs of both types of operating conditions; regardless of the window length setting, steady-state jitter or dynamic response lag will occur.

[0006] 2. Fixed windows indiscriminately process all historical residuals, making it impossible to distinguish between valid signals and interference signals, which can easily lead to inaccurate updates of the covariance matrix: Fixed windows perform equal-weighted statistical averaging on all historical residuals within the window. Under scenarios of fluctuating operating conditions and parameter perturbations, fixed windows will include a large number of invalid noise residuals in the statistics, resulting in inaccurate updates of the covariance matrix, a significant decrease in observation accuracy, and even causing observation divergence and motor synchronism failure.

[0007] 3. Fixed-window covariance matrix updates have poor adaptability to actual operating conditions, resulting in insufficient algorithm robustness. The residual statistical characteristics of a fixed window are out of sync with the actual operating conditions of the motor. The residual statistical results cannot reflect the dynamic changes in the motor's operating state in real time, causing the covariance matrix update to fail to synchronously adapt to time-varying observation-side noise and model-side errors. This leads to a long-term disconnect between the covariance matrix parameters and the actual operating conditions of the motor, resulting in extremely poor robustness and making it impossible to achieve stable, sensorless control of permanent magnet synchronous motors at medium and high speeds across all operating conditions, from steady state to dynamic scenarios. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of the traditional fixed-window EKF in existing technologies and propose a coupled closed-loop adjustment method based on permanent magnet synchronous motor (PMSM) of operating condition sensing, dynamic weighted residual screening, and dual covariance matrix collaborative adaptation. This method replaces the traditional fixed-window, equal-weighted statistical approach by calculating residuals in real time and constructing operating condition sensing coefficients, and assigning dynamic attenuation weights to historical residuals. It uses a residual validity threshold to screen out effective residuals that match the current operating condition, and then applies these effective residuals to the observation noise covariance matrix. With process noise covariance matrix Adaptive updates are performed to match the EKF filter parameters with the actual operating conditions of the motor.

[0009] The present invention achieves the above-mentioned technical objectives through the following technical means.

[0010] A sensorless control method for permanent magnet synchronous motors based on dynamic weight residual screening adaptive EKF:

[0011] Collect the three-phase stator current of the permanent magnet synchronous motor during operation and convert it into... Stator current sampling values ​​in a coordinate system are used to construct and discretize state equations related to the electromagnetic characteristics of the motor, thus completing the permanent magnet synchronous motor. The prediction of the state at any given time is based on the sampled and predicted values ​​of the stator current, and the extraction is performed. Time-of-flight filter residual ;

[0012] Based on the filter residual ,calculate The time-varying characteristic quantity represents the moving average of residual energy in the recent interval and the moving average of residual energy in the historical interval. It is calculated from the moving average of residual energy in both intervals. Operating condition sensing coefficient of permanent magnet synchronous motor ;

[0013] The working condition perception coefficient Converted to residual weighting base coefficients , and then calculate Time of the first Weighting coefficients of historical residuals Set a residual effective determination threshold that is compatible with the current sampling noise. Historical residuals are selectively filtered to construct an effective residual set that matches the current operating conditions. Meanwhile, engineering constraints are applied to the size of the effective residual set;

[0014] The effective residuals are weighted and fused to obtain the residual fusion value that reflects the current operating conditions. Then, the innovation covariance matrix matching the current working conditions is calculated. Based on the observation noise covariance matrix of the previous time step, combined with the predicted value of the filter error covariance matrix. Calculate the Kalman gain matrix Based on the innovation covariance matrix And the recursive coefficients, to complete the observation noise covariance matrix. Adaptive smooth update; based on the Kalman gain matrix and innovative covariance matrix, the process noise covariance matrix is ​​completed. Adaptive updates;

[0015] Based on the Kalman gain matrix and the filtering residual, complete The state vector is updated at each moment to obtain an estimate of the motor rotor state and feed it back to the permanent magnet synchronous motor vector control system, thereby realizing sensorless control of the permanent magnet synchronous motor.

[0016] Will The state estimate at time step, the filter error covariance matrix, the observation noise covariance matrix, and the process noise covariance matrix are used as... The initial value of the iteration at time step, proceed. The permanent magnet synchronous motor is controlled without position sensors.

[0017] Furthermore, when collecting the stator current during the operation of the permanent magnet synchronous motor, the data from the permanent magnet synchronous motor vector control system is also collected. Shaft-stator voltage, used in permanent magnet synchronous motors Prediction of the state vector at time step and prediction of the filter error covariance matrix.

[0018] Furthermore, the aforementioned Time-of-flight operating condition characteristic quantities The residual energy moving average of the recent interval The residual energy moving average of the historical interval , where M is the sliding step size.

[0019] Furthermore, the aforementioned Operating condition sensing coefficient of permanent magnet synchronous motor ,in, For numerical protection bias.

[0020] Furthermore, the residual weighting base coefficients Dimensional matching coefficient ,in, The operating condition matching coefficient is... To control the sampling period.

[0021] Furthermore, the aforementioned Time of the first Weighting coefficients of historical residuals ,in, This is the period difference between the current time and the time when the residual was generated.

[0022] Furthermore, the effective residual set ,in, This represents the maximum number of residuals utilized.

[0023] Furthermore, an engineered constraint on the size of the effective residual set is the number of elements in the effective residual subset. Amplitude limiting optimization completed:

[0024] like Take the one with the highest credibility weight. The information is a valid residual set;

[0025] like directly with For an effective residual set;

[0026] like Candidate information is supplemented according to confidence weight from high to low until the sample size reaches a certain level. ;

[0027] in, This represents the minimum number of residuals utilized.

[0028] Furthermore, residual fusion value .

[0029] Furthermore, innovative covariance matrix .

[0030] Compared with the existing traditional fixed-window EKF solution, the present invention has the following outstanding advantages:

[0031] 1. Effectively alleviates the contradiction between steady-state smoothness and dynamic response in traditional fixed-window EKF, achieving both high steady-state smoothness and fast dynamic response under medium- and high-speed full-condition operation: By using a dual-interval moving average design to identify the real-time operating conditions of the motor, a condition perception coefficient is constructed to provide a reliable operating condition benchmark for weight adjustment; then, by using a condition-driven dynamic weight design, the traditional fixed-window equal-weight, fixed-length processing scheme is replaced. Under steady-state conditions, the weight decays slowly to fully utilize historical residuals to suppress noise, while under dynamic conditions, the weight decays quickly to weaken the influence of old residuals and reduce observation lag, thus optimizing the shortcomings of fixed window length in meeting the operating requirements of medium- and high-speed full-condition operation.

[0032] 2. Achieve accurate separation of valid motor operating signals from sampling noise and outdated invalid residuals, improving the accuracy of covariance matrix updates from the data source: Design and current sampling noise 3 The effective judgment threshold of residuals with significance level matching is used to selectively filter the historical residuals after weighting the operating conditions, remove invalid noise residuals with too low weights and old residuals that do not match the current operating conditions, and construct an effective residual set that reflects the true operating conditions of the motor. This improves the noise pollution and inaccurate covariance matrix update problems caused by the indiscriminate inclusion of all residuals in the traditional fixed-window EKF, and significantly improves the rotor state observation accuracy under operating condition fluctuation and parameter perturbation scenarios.

[0033] 3. Significantly improves the adaptability of the covariance matrix to the real-time operating conditions of the motor, enhancing the robustness and anti-interference capability of the motor under medium- and high-speed full-condition operation: The observation noise covariance matrix is ​​completed by selecting effective residuals that match the operating conditions. With process noise covariance matrix The closed-loop collaborative update enables the covariance matrix parameters to be synchronously adapted to the time-varying sampling noise on the observation side and the load disturbance and parameter perturbation on the model side. This solves the problem of the traditional fixed-window EKF covariance matrix parameters being out of sync with the actual operating conditions of the motor, effectively avoids the risk of observation divergence and motor step loss under operating condition fluctuations, and significantly improves the motor's anti-interference capability under parameter perturbation and load change scenarios. Attached Figure Description

[0034] Figure 1 This is a flowchart of the sensorless control process for a permanent magnet synchronous motor based on dynamic weight residual screening adaptive EKF as described in this invention. Detailed Implementation

[0035] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but the scope of protection of the present invention is not limited thereto.

[0036] This embodiment aims to improve upon the three inherent defects of the traditional fixed-window EKF by constructing a coupled closed-loop framework of condition perception, dynamic weighted residual screening, and dual covariance matrix collaborative adaptation. By constructing condition feature quantities and using dynamic weighted residual screening as the core, condition screening of residual data is achieved. This is further combined with the observation noise covariance matrix... Update criteria, process noise covariance matrix A Kalman gain co-update strategy is employed to couple data optimization with covariance matrix updates. The iteration period of this method is synchronized with the sampling period of the permanent magnet synchronous motor vector control (FOC). Strict synchronization is ensured by executing the entire process once per sampling cycle to guarantee synchronization with the motor control timing.

[0037] Step 0: Motor zero-speed / low-speed start and smooth transition between open-loop and closed-loop operation

[0038] This solution primarily optimizes the sensorless control performance of EKF in permanent magnet synchronous motors under all operating conditions at medium and high speeds. Addressing the issue of the EKF having an insignificant back electromotive force in the zero / low speed range, it adapts to the open-loop control strategy of zero-speed rotor initial positioning + low-speed I / F in this field. Only smooth transition and parameter constraint rules are designed for the core algorithm of this solution, enabling complete sensorless control of the motor from standstill to low speed to medium and high speeds.

[0039] 1. General control for zero / low speed range: When the motor is stationary, the initial electrical angle is obtained using a conventional rotor initial positioning method in this field. During the low-speed phase, open-loop I / F control is used for speed increase. During this phase, the EKF algorithm synchronously follows and iterates, with a working condition perception coefficient... Forced to 0, residual weighting base coefficient Take the steady-state minimum value as the effective threshold for residual determination. Setting it to 0.05 allows the EKF algorithm to converge in the low-speed range.

[0040] 2. Open-loop to closed-loop switching rule: When the motor speed rises to the medium-high speed switching threshold, and the deviation between the EKF estimated electrical angle and the open-loop given electrical angle meets the conventional stability conditions in this field, smooth switching is triggered.

[0041] 3. Connection logic of this scheme: After the switch is completed, the parameter constraints of EKF are immediately released, and the medium-to-high speed full-condition adaptive EKF control process of steps one to five is entered (see...). Figure 1 If an electrical angle deviation exceeds the limit during the switching process, the system will switch back to open-loop control according to conventional protection logic in this field, and the switching will be triggered again after the operating conditions stabilize.

[0042] Step 1: PMSM Operation Status Data Acquisition and EKF Basic Prediction Model Construction

[0043] This step constructs an EKF prediction model that is deeply coupled with the electromagnetic characteristics of the PMSM itself, and extracts the filter residuals related to the motor's operating conditions, providing a foundation for subsequent end-to-end optimization.

[0044] 1. PMSM operating status synchronously collected

[0045] The three-phase stator current of the motor is sampled and subjected to Clark transformation to obtain two-phase stationary current. coordinate system Stator current sampling value at time 1 Simultaneously, the output of the FOC system's SVPWM modulation module is acquired. time Shaft stator voltage , as input to the EKF prediction model.

[0046] 2. Definition of EKF model coupled with the electromagnetic properties of PMSM

[0047] Time-state vector Defined as: ( , for Permanent magnet synchronous motor at present Shaft stator current, for The electric angular velocity of the rotor at any given moment. for (electrical angle of the rotor at any moment);

[0048] The time-observation vector is defined as: .

[0049] Observation matrix: Establishes a linear mapping from the state vector to the observation vector, retaining only the components related to current observation to simplify computation. The expression is:

[0050] (1)

[0051] 3. One-step prediction of EKF based on PMSM physical equations

[0052] Based on PMSM The stator voltage state equation and mechanical motion equation in the coordinate system are used to construct state equations related to the electromagnetic characteristics of the motor:

[0053] (2)

[0054] in, , for Stator current of the shaft, , for Stator voltage of the shaft, Stator phase resistance, For stator inductance, It is a permanent magnet flux linkage. For electrical angle, Electric angular velocity, For rotational inertia, The coefficient of friction, This represents the load torque.

[0055] The state equations are discretized using the second-order Runge-Kutta method, balancing numerical stability with the low computational power requirements of the MCU, and adjusting the discretization period and control sampling period. Synchronization, completing the permanent magnet synchronous motor The prediction of the state vector at time step is expressed as:

[0056] (3)

[0057] in, for Predicted value at time, for The optimal state estimate at time t. for Stator voltage input at time t, As an intermediate quantity, This is an intermediate quantity.

[0058] Jacobian matrix based on state equation The prediction of the filter error covariance matrix is ​​completed, and the expression for the Jacobian matrix is:

[0059] (4)

[0060] in, It is a 4th-order identity matrix; the prediction expression for the filter error covariance matrix is:

[0061] (5)

[0062] in, for The process noise covariance matrix at time step 1.

[0063] 4. Filtering Residual Extraction

[0064] Based on the actual current sampling values ​​of PMSM and the predicted values ​​of EKF model, extract Time-based filtering residual (innovation vector):

[0065] (6)

[0066] The filter residual directly reflects the physical deviation between the actual value of the PMSM current and the model prediction value. Its amplitude and variation characteristics are directly related to the motor operating conditions (steady-state / dynamic, load change, parameter perturbation) and are the core physical basis for subsequent operating condition perception.

[0067] Step 2: Constructing PMSM operating condition sensing coefficients based on dual-interval moving average

[0068] This step addresses the shortcomings of traditional fixed-window EKF systems, such as weak condition perception capabilities, difficulty in adapting to changes in operating conditions due to fixed window length, and the need for repeated manual calibration of window parameters. It constructs a condition perception coefficient to provide a core operating condition benchmark for subsequent dynamic weight adjustment and covariance matrix update.

[0069] 1. Calculation of characteristic quantities of PMSM operating condition

[0070] Based on the filter residual extracted in step one, calculate Time-of-flight characteristic quantities:

[0071] (7)

[0072] This physical quantity is a positive real number, which directly reflects the degree of deviation between the predicted and actual values ​​of the PMSM current. The larger the deviation, the more drastic the change in the motor's operating conditions. It is the core physical quantity that characterizes the change in operating conditions.

[0073] 2. Dual-interval moving average processing to match current loop bandwidth

[0074] To accurately capture the dynamic changes in motor operating conditions, a sliding step size M is set to match the bandwidth of the FOC system's current loop. M is related to the bandwidth characteristics of the FOC system's current and speed loops, and its value typically ranges from 1 to 30 control cycles. In this embodiment, 15 is used, but it can be adjusted within this range according to the control cycle and system bandwidth requirements of the actual scenario. The moving average of the residual energy between two adjacent equal-length intervals is calculated to avoid random noise interference from the residual at a single moment, thus accurately capturing the trend of operating condition changes.

[0075] The recent range is The corresponding residual energy moving average is:

[0076] (8)

[0077] Historical interval is The corresponding residual energy moving average is:

[0078] (9)

[0079] This dual-range design, matched with the current loop adjustment cycle, can effectively distinguish between "real changes in operating conditions" and "accidental noise interference," thus improving the insufficient accuracy of operating condition identification in the fixed-window EKF single-window sliding average design.

[0080] 3. Calculation of perception coefficient for low-dependency operating conditions

[0081] Based on the moving average of the residual energy in the two intervals, calculate PMSM operating condition perception coefficient at all times This coefficient is a core indicator characterizing the degree of change in motor operating conditions. To avoid exceeding the limit under extreme operating conditions, [the following is omitted as the text is incomplete and requires further context]. The upper and lower limits are set, and the final expression is:

[0082] (10)

[0083] in, For numerical protection bias (value) ), used to avoid division by zero errors and numerical overflow when the denominator is zero; operating condition perception coefficient The larger the value, the more drastic the physical changes in the motor's operating conditions.

[0084] Step 3: Dynamic Exponential Weight Generation and Effective Residual Screening Driven by Operating Conditions

[0085] This step is key to optimizing the core performance of the fixed-window EKF in this invention. It addresses the shortcomings of the fixed-window EKF, such as processing all residuals with equal weights, difficulty in balancing steady-state smoothness and dynamic response, and the easy contamination of covariance matrix updates by invalid noise. The fixed-window equal-weight and fixed-length processing logic is optimized using operating condition-driven dynamic exponential weights to achieve accurate separation of valid signals and invalid interference.

[0086] 1. Calculation of dynamic weighting base coefficients for working condition matching

[0087] The working condition perception coefficients obtained in step two are converted into residual weight base coefficients. This achieves a direct mapping from physical changes in operating conditions to residual weight adjustments, employing a linear mapping method with minimal computational cost, suitable for MCUs with low computing power requirements. The expression is:

[0088] (11)

[0089] in, Let be the dimension matching coefficient, and:

[0090] (12)

[0091] in, This is the operating condition matching coefficient, which is related to the physical change rate and control cycle characteristics of the PMSM operating condition. Its value range is generally 0.05~0.2, and in this embodiment, it is 0.1. To control the sampling period; if or ,make To avoid abnormal weighting coefficients under steady-state conditions, The value range is generally 50~2000, which is related to the requirements of the sampling period and the smoothness of steady-state filtering in this scheme. In this embodiment, 1000 is used.

[0092] 2. Calculation of the dynamic exponential weight of the dynamic decay coefficient

[0093] Based on the residual weighting base coefficient, calculate Time of the first Weighting coefficients of historical residuals This weight has no fixed attenuation coefficient and is mainly determined by the operating condition perception coefficient. Dynamic control, expressed as:

[0094] (13)

[0095] in, The period difference between the current time and the time when the residual was generated represents the "aging" of the residual.

[0096] Under steady-state conditions, Approaching 0, By keeping the weights near the minimum value, the weight decay rate is slow, which allows for full utilization of historical data to suppress noise and ensures the smoothness of the observation results.

[0097] Under dynamic operating conditions, Rapidly increasing As the weights increase synchronously, the rate of weight decay accelerates. The latest operating condition matching residuals have higher weights and can quickly follow changes in operating conditions, reducing the risk of observation lag.

[0098] 3. Effective residual screening based on weighted thresholds

[0099] Set a residual effective determination threshold that is compatible with current sampling noise. Historical residuals are selectively filtered to construct an effective residual set that matches the current operating conditions. The screening formula is:

[0100] (14)

[0101] 3 from PMSM current sampling noise The significance level was determined by tuning, typically ranging from 0.03 to 0.08; in this embodiment, it was set to 0.05. The maximum number of residuals utilized... The value is related to the upper limit of computing power and the statistical reliability requirements of the covariance matrix. The value range is generally 25~40, and 30 is used in this embodiment.

[0102] Simultaneously, engineering constraints are applied to the size of the effective residual set to balance computational real-time performance with statistical reliability; specifically, this involves limiting the number of elements in the effective residual subset. Amplitude limiting optimization completed:

[0103] like Take the one with the highest credibility weight. This information is the final valid residual set;

[0104] like directly with For an effective residual set;

[0105] like Candidate information is supplemented according to confidence weight from high to low until the sample size reaches a certain level. This ensures the statistical reliability of the covariance matrix update.

[0106] Minimum residual utilization , and covariance matrix ( , The reliability lower limit requirement for statistical calculation is related, and the value range is generally 8~15. In this embodiment, 10 is used.

[0107] During the initial stage of high-speed closed-loop startup of the motor (after EKF closed-loop), when the cumulative sampling period after switching is less than At that time, the working condition perception coefficient is not calculated, and it is set as follows: The residual weighting base coefficients are taken as the steady-state default values. ;when At that time, effective residual set acquisition For all historical residuals within the interval, a sample size is not mandatory. Only through Valid residuals are selected. The cumulative number of sampling periods after entering medium-to-high-speed closed-loop control is not less than [amount missing]. Effective residual sample size And when the system is in a stable closed-loop state, enable and Adaptive updates.

[0108] Step 4: Collaborative Adaptive Update of Q / R Bicovariance Matrix Based on Effective Residues

[0109] The covariance matrix update in this step is based on the effective residual set after the operating conditions are filtered, which solves the problem of noise pollution and disconnection from the operating conditions in the traditional fixed-window covariance matrix from the data source.

[0110] 1. Effective residual fusion and innovative covariance matrix calculation for working condition adaptation

[0111] Based on the effective residuals and corresponding weights obtained in step three, the effective residuals are weighted and fused to obtain a residual fusion value that reflects the current working condition. To reduce the random interference of single residuals, the expression is:

[0112] (15)

[0113] Based on the weighted fusion residuals, the innovation covariance matrix matching the current operating conditions is calculated. This covariance matrix can effectively eliminate the interference of invalid noise and is used to optimize the noise pollution problem of the innovative covariance matrix based on the full residual calculation in fixed-window EKF. The expression is:

[0114] (16)

[0115] in, For numerical protection bias (value) This is used to avoid division by zero errors when the effective residual sample size is insufficient.

[0116] 2. Kalman gain calculation for operating condition matching

[0117] Based on the observation noise covariance matrix of the previous time step Combined with the filter error covariance matrix, one-step prediction value Calculate the Kalman gain matrix This gain can be adapted to the current high-speed real-time operating conditions of the motor, and its expression is:

[0118] (17)

[0119] 3. Update of the filter error covariance matrix

[0120] The filter error covariance matrix is ​​updated to provide the basic data for the next iteration. The expression is:

[0121] (18)

[0122] in, It is a 4th order identity matrix.

[0123] 4. Adaptive update of observation noise covariance

[0124] Define the forgetting factor In this embodiment, the value is 0.95, used to gradually reduce the weight of older data. Calculation Recurrence coefficients at time points:

[0125] (19)

[0126] Based on the innovation covariance matrix and recursion coefficients Complete the observation noise covariance matrix The adaptive smooth update is expressed as:

[0127] (20)

[0128] in, To observe the lower bound of the noise covariance matrix, ensuring the matrix is ​​positive definite and avoiding inversion failures caused by negative definite matrices; the updated Symmetry needs to be achieved: This ensures the numerical stability of the matrix.

[0129] 5. Cooperative adaptive update of process noise covariance matrix

[0130] Based on the Kalman gain matrix and the innovative covariance matrix, the process noise covariance matrix was completed. The adaptive update is expressed as:

[0131] (twenty one)

[0132] (twenty two)

[0133] in, This is the lower bound of the process noise covariance matrix. This is the upper limit for process noise covariance, used to avoid exceeding the limit under extreme operating conditions; the updated Symmetry needs to be achieved: This ensures the numerical stability of the matrix.

[0134] This step, based on the effective residuals after working condition screening, achieves... The dual collaborative adaptive update enables the covariance matrix parameters to match the actual high-speed operating conditions of the motor in real time, significantly improving the defect of inaccurate covariance matrix in fixed window EKF.

[0135] Step 5: EKF State Optimal Estimation and FOC Closed-Loop Feedback

[0136] 1. EKF State Vector Update

[0137] Based on the Kalman gain matrix and the filtering residual, complete The update of the state vector at each time step yields an estimate of the motor rotor state, expressed as:

[0138] (twenty three)

[0139] in, for The optimal state estimate at time step includes The estimated values ​​of stator current, rotor electric angular velocity, and rotor electric angle are all core feedback quantities for FOC closed-loop control.

[0140] 2. Deeply coupled feedback with the FOC closed loop

[0141] The updated rotor state estimate is fed back to the permanent magnet synchronous motor FOC system: the rotor electrical angular velocity estimate is fed back to the speed loop for closed-loop speed regulation; the rotor electrical angle estimate is used for... coordinate system and The Park / Clark transformation of the coordinate system provides a precise angular reference for current loop adjustment; The estimated shaft stator current is used for current loop feedback regulation, realizing deep coupling between FOC closed-loop control and EKF observer, so as to optimize observation performance and control performance in a coordinated manner.

[0142] 3. Full-cycle iterative loop logic

[0143] Will State estimate at time 1 Filtering error covariance matrix Observation noise covariance matrix Process noise covariance matrix As The initial value of the iteration at each time point, after each control sampling cycle. This triggers a complete process from step one to step five, forming a complete iterative closed loop, and achieving real-time matching of the covariance matrix parameters with the motor under all high-speed operating conditions.

[0144] The embodiments described above are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments. Any obvious improvements, substitutions or modifications that can be made by those skilled in the art without departing from the essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A sensorless control method for permanent magnet synchronous motors based on dynamic weight residual screening adaptive EKF, characterized in that: Collect the three-phase stator current of the permanent magnet synchronous motor during operation and convert it into... Stator current sampling values ​​in a coordinate system are used to construct and discretize state equations related to the electromagnetic characteristics of the motor, thus completing the permanent magnet synchronous motor. The prediction of the state at any given time is based on the sampled and predicted values ​​of the stator current, and the extraction is performed. Time-of-flight filtering residual ; Based on the filter residual ,calculate The time-varying characteristic quantity represents the moving average of residual energy in the recent interval and the moving average of residual energy in the historical interval. It is calculated from the moving average of residual energy in both intervals. Operating condition sensing coefficient of permanent magnet synchronous motor ; The working condition perception coefficient Converted to residual weighting base coefficients , and then calculate Time of the first Weighting coefficients of historical residuals ; Set a residual effective determination threshold that is compatible with current sampling noise. Historical residuals are selectively filtered to construct an effective residual set that matches the current operating conditions. Meanwhile, engineering constraints are applied to the size of the effective residual set; The effective residuals are weighted and fused to obtain the residual fusion value that reflects the current operating conditions. Then, the innovation covariance matrix matching the current working conditions is calculated. Based on the observation noise covariance matrix of the previous time step, combined with the predicted value of the filter error covariance matrix. Calculate the Kalman gain matrix Based on the innovation covariance matrix And the recursive coefficients, to complete the observation noise covariance matrix. Adaptive smooth update; based on the Kalman gain matrix and innovative covariance matrix, the process noise covariance matrix is ​​completed. Adaptive updates; Based on the Kalman gain matrix and the filtering residual, complete The state vector is updated at each moment to obtain an estimate of the motor rotor state and feed it back to the permanent magnet synchronous motor vector control system, thereby realizing sensorless control of the permanent magnet synchronous motor. Will The state estimate at time step, the filter error covariance matrix, the observation noise covariance matrix, and the process noise covariance matrix are used as... The initial value of the iteration at time step, proceed. The permanent magnet synchronous motor is controlled without position sensors.

2. The sensorless control method for a permanent magnet synchronous motor according to claim 1, characterized in that, When collecting the stator current during the operation of the permanent magnet synchronous motor, the data from the permanent magnet synchronous motor vector control system is also collected. Shaft-stator voltage, used in permanent magnet synchronous motors Prediction of the state vector at time step and prediction of the filter error covariance matrix.

3. The sensorless control method for a permanent magnet synchronous motor according to claim 1, characterized in that, The Time-of-flight operating condition characteristic quantities The residual energy moving average of the recent interval The residual energy moving average of the historical interval , where M is the sliding step size.

4. The sensorless control method for a permanent magnet synchronous motor according to claim 3, characterized in that, The Operating condition sensing coefficient of permanent magnet synchronous motor ,in, For numerical protection bias.

5. The sensorless control method for a permanent magnet synchronous motor according to claim 4, characterized in that, The residual weighting base coefficient Dimensional matching coefficient ,in, The operating condition matching coefficient is... To control the sampling period.

6. The sensorless control method for a permanent magnet synchronous motor according to claim 5, characterized in that, The Time of the first Weighting coefficients of historical residuals ,in, This is the period difference between the current time and the time when the residual was generated.

7. The sensorless control method for a permanent magnet synchronous motor according to claim 6, characterized in that, Effective residual set ,in, This represents the maximum number of residuals utilized.

8. The sensorless control method for a permanent magnet synchronous motor according to claim 7, characterized in that, An engineering constraint on the size of the effective residual set is the number of elements in the effective residual subset. Amplitude limiting optimization completed: like Take the one with the highest credibility weight. The information is a valid residual set; like directly with For an effective residual set; like Candidate information is supplemented according to confidence weight from high to low until the sample size reaches a certain level. ; in, This represents the minimum number of residuals utilized.

9. The sensorless control method for a permanent magnet synchronous motor according to claim 8, characterized in that, Residual fusion value .

10. The sensorless control method for a permanent magnet synchronous motor according to claim 9, characterized in that, Innovation Covariance Matrix .

Images