Permanent magnet synchronous motor zero speed starting control method, electronic device and program product

Abstract

The application provides a permanent magnet synchronous motor zero-speed starting control method, electronic equipment and program product, and through three-dimensional direction improvement of position identification, torque control and model fault tolerance, realizes comprehensive starting control scheme with undisturbed position accurate identification, model parameter real-time fusion, control weight adaptive optimization and active fault tolerance capability at zero speed, and has the following advantages: accurate identification of zero-speed rotor position, no swing and no coasting throughout the journey; zero-speed full-torque output, fast dynamic response and small torque ripple; model error active compensation, strong robustness and anti-parameter drift.

Description

Technical Field

[0001] This application relates to the field of motor drive technology, and more specifically, to a zero-speed start control method, electronic equipment, and program product for a permanent magnet synchronous motor. Background Technology

[0002] Permanent magnet synchronous motors (PMSMs) have been widely used in industrial servo drives, new energy vehicles, elevators, cranes, and high-end machine tools due to their high power density, high efficiency, and excellent dynamic performance. The core of achieving high-performance control lies in the precise observation of rotor position and accurate control of closed-loop torque. In the high and medium speed operating range, sensorless technologies based on back-EMF observation (such as sliding mode observers and extended Kalman filters) are relatively mature. However, at zero speed or extremely low speeds, the back-EMF amplitude approaches zero or is very weak, causing back-EMF-based observers to completely fail and unable to obtain effective rotor position information. This has become the core bottleneck for achieving sensorless control of the motor across the entire speed range.

[0003] To address the zero-speed start problem, existing technologies primarily employ high-frequency signal injection. This method injects high-frequency voltage or current signals into the motor stator side, utilizing the motor's salient pole effect (i.e., the unequal inductance between the d-axis and q-axis) to obtain rotor position information. However, the traditional single high-frequency signal injection scheme has the following main drawbacks when applied to zero-speed start of PMSMs:

[0004] The identification process is prone to causing rotor oscillation, making it unsuitable for potential energy loads: In the identification phase of traditional high-frequency injection methods, certain components of the injected signal may generate pulsating torque, causing the rotor to experience unexpected slight oscillations or vibrations during identification. This is fatal for potential energy load applications such as elevators and cranes, potentially leading to serious safety hazards such as load slippage and positioning inaccuracies.

[0005] The model is highly sensitive to parameters and has poor robustness: At zero speed, the stator resistance voltage drop dominates the motor terminal voltage. However, temperature changes, magnetic circuit saturation, and the nonlinear characteristics of the inverter (such as dead-zone effect and tube voltage drop) during motor operation can severely affect the accuracy of key parameters such as stator resistance and inductance, leading to mismatch in the prediction model. Traditional solutions mostly use offline calibration or independent parameter identification modules, which suffer from slow convergence and weak dynamic tracking capabilities. They cannot compensate for significant parameter drift at zero speed in real time, resulting in large deviations in current and torque predictions during startup, easily causing overcurrent and loss of synchronization, and low startup success rate.

[0006] Control strategies struggle to balance multiple objectives and dynamic adaptability: Finite Control Set Model Predictive Torque Control (FCS-MPTC), as an advanced control strategy, exhibits good dynamic performance by selecting the optimal voltage vector online. However, under the special condition of zero-speed start-up, its control objectives are contradictory: on the one hand, it needs to output full torque quickly and stably to overcome the load; on the other hand, it needs to strictly limit current and avoid overcurrent, while also considering switching losses. The value function of traditional FCS-MPTC usually uses fixed weighting coefficients, which cannot adaptively adjust according to the dynamic requirements of different stages during zero-speed start-up (such as zero-speed lock-up and acceleration transition). Fixed weights often compromise on certain aspects, making it difficult to achieve an optimal balance among multiple objectives such as ensuring starting torque, suppressing current surges, and reducing switching losses.

[0007] Lack of proactive fault-tolerance mechanisms for model mismatch: Existing control schemes typically treat model predictions as ideal conditions. When parameter changes or disturbances cause significant deviations between the predicted model and the actual system, there is a lack of effective proactive detection and compensation mechanisms. When the prediction error accumulates to a certain level, hardware protection (such as overcurrent protection) can only be passively triggered, leading to a startup interruption. The system lacks robustness and cannot guarantee reliable startup under harsh operating conditions or parameter perturbations.

[0008] In summary, existing zero-speed sensorless start technology for permanent magnet synchronous motors (PMSMs) cannot simultaneously meet the key requirements of industrial applications such as swing-free identification, fast full-torque response, strong robustness against parameter drift, and adaptive multi-objective optimization, thus restricting the widespread application of high-performance PMSM drive systems in high-end, high-reliability scenarios. Summary of the Invention

[0009] The purpose of this application is to provide a zero-speed start control method, electronic device and program product for permanent magnet synchronous motors, in order to solve the problem that the prior art is unable to simultaneously meet the key requirements of industrial applications such as swing-free identification, fast response at full torque, strong robustness against parameter drift and adaptive multi-objective optimization.

[0010] This application provides a zero-speed start control method for a permanent magnet synchronous motor, comprising:

[0011] Step S1: When the motor is in a zero-speed stationary state, inject a composite voltage signal containing DC bias component and high-frequency pulsation component into the d-axis direction of the rotor position estimation to forcibly enhance the salient pole effect of the motor, and extract the rotor position error in real time based on the high-frequency current response, and simultaneously complete the identification of the rotor initial electrical angle and the determination of magnetic pole polarity; update the identified rotor position electrical angle, saturated inductance and permanent magnet flux linkage observation values ​​to the prediction model in real time in units of control cycle;

[0012] Step S2: Based on the motor speed, the zero-speed start-up process is divided into a zero-speed lock-up interval and a speed-up transition interval. In each MPC control cycle, the weight coefficients of each control objective in the value function are dynamically and adaptively calculated according to the interval and real-time operating conditions. The value function includes at least a torque tracking term, a flux tracking term, a current constraint penalty term, and a switching frequency penalty term. All basic voltage vectors of the inverter are traversed, and the predicted state of each voltage vector is calculated based on the updated prediction model. The cost value of each voltage vector is calculated in combination with the weight coefficients, and the voltage vector with the smallest cost value is selected as the optimal control quantity output for the current cycle.

[0013] Step S3: In each control cycle, calculate the prediction residual between the predicted current value and the actual sampled value at the previous moment, and compensate the prediction residual in real time into the current prediction model through the residual correction factor to correct the model parameter error; at the same time, according to the magnitude of the prediction residual, execute two-level fault-tolerant logic: when the residual magnitude is less than or equal to the set threshold, maintain the normal control mode and enhance the residual penalty term in the value function; when the residual magnitude is greater than the set threshold, switch to the fault-tolerant prediction model and freeze the adaptive weight strategy to maintain the startup process without interruption; Among them, the current prediction model is a specific execution model that embeds the prediction residual into the prediction model; when the residual amplitude is greater than a set threshold, the specific execution model switches to a fault-tolerant prediction model.

[0014] The above technical solution improves upon three dimensions: position recognition, torque control, and model fault tolerance. It achieves a comprehensive start-up control scheme at zero speed, enabling disturbance-free and accurate position recognition, real-time fusion of model parameters, adaptive optimization of control weights, and active fault tolerance. Specifically, it includes:

[0015] Precise zero-speed rotor position identification with no swaying or slippage throughout the entire process: Employing salient pole enhancement high-frequency composite signal injection and saturated salient pole strengthening technology, the difference in inductance between the d and q axes is forcibly increased, enabling stable identification even for surface-mount PMSMs with low salient pole ratios, thus improving position identification accuracy; By estimating d-axis directional injection and q-axis zero-voltage constraint, no torque component is generated during the identification process, and the rotor remains stationary throughout the entire process, completely eliminating the swaying and mechanical shock caused by traditional identification; Utilizing the rate of change of saturated inductance to achieve online determination of magnetic pole polarity without the need for additional pulse injection, ensuring accurate and reliable polarity determination and avoiding start-up synchronization loss caused by 180° polarity errors.

[0016] Zero-speed full-torque output with fast dynamic response and low torque ripple: Employing a dual-interval adaptive weighted value function, the torque weight is significantly increased and the switching loss weight is reduced in the zero-speed interval, achieving stable zero-speed torque output and shortening torque response time. Multiple objectives, including torque, flux linkage, current, and switching losses, are optimized in a unified manner, reducing torque ripple during startup and eliminating impact and jitter.

[0017] Active model error compensation, strong robustness, and resistance to parameter drift: Employing real-time prediction residual correction and active fault-tolerant control techniques, it directly compensates for model errors caused by stator resistance temperature drift, inverter nonlinearity, and core saturation, without requiring separate parameter identification. A two-level fault-tolerant switching mechanism is constructed, automatically switching to fault-tolerant mode when the model error exceeds a threshold, ensuring uninterrupted and shock-free startup throughout the entire process.

[0018] In some optional implementations, step S1 includes:

[0019] The composite voltage signal is injected only into the d-axis of the rotor position estimation, and the q-axis injection voltage is always zero; the high-frequency pulsation component is a standard cosine pulsation waveform; the value range of the DC bias component is 0.5 to 1.5 times the rated voltage of the motor; the angular frequency of the high-frequency pulsation component is 1 / 10 to 1 / 5 of the inverter switching frequency.

[0020] In some optional implementations, step S1, which involves extracting the rotor position error in real time based on the high-frequency current response and simultaneously completing the initial electrical angle identification and magnetic pole polarity determination of the rotor, includes:

[0021] Based on the high-frequency q-axis current component in the injected current response, the rotor position error signal is extracted, and the initial rotor electrical angle is obtained by observation through an adaptive phase-locked loop.

[0022] Based on the DC component of the d-axis current caused by the DC bias component, the d-axis flux change rate is calculated online to determine the magnetic polarity of the permanent magnet.

[0023] If the rate of change of the d-axis flux approaches zero, the current magnetic pole polarity is determined to be incorrect, and the initial rotor electrical angle is corrected by 180 degrees to obtain the rotor position electrical angle.

[0024] In some optional implementations, step S1, which involves updating the identified rotor position electrical angle, saturation inductance, and permanent magnet flux linkage observations to the prediction model in real time, on a control cycle basis, includes:

[0025] Update the rotor position electrical angle to the angle reference of Park transformation and inverse Park transformation;

[0026] Update the d-axis inductance and q-axis inductance in the saturated inductance to the inductance parameters in the current prediction equation;

[0027] The permanent magnet flux linkage observations are then updated to the flux linkage parameters in the torque prediction equation.

[0028] In some optional implementations, in step S2, the zero-speed locking range is the range where the speed is less than or equal to 5% of the rated speed, and the acceleration transition range is the range where the speed is between 5% and 20% of the rated speed; the weighting coefficients of each control objective in the dynamic adaptive calculation value function specifically include:

[0029] In the zero-speed locking range, the weighting coefficients of the torque tracking term and the current constraint penalty term are increased, while the weighting coefficient of the switching frequency penalty term is decreased.

[0030] During the acceleration transition range, the weighting coefficient of the torque tracking term is reduced as the rotational speed increases, while the weighting coefficient of the switching frequency penalty term is increased as the rotational speed increases.

[0031] In some optional implementations, when the motor speed is in the transition zone between the zero-speed lock-in zone and the acceleration transition zone, the weight coefficients of each control objective in the dynamic adaptive calculation value function are linearly weighted, including:

[0032] Calculate the speed ratio factor based on the current speed;

[0033] Obtain the baseline weights for the zero-speed locking interval and the acceleration transition interval, respectively.

[0034] The two benchmark weights are linearly weighted based on the aforementioned speed ratio factor to calculate the smoothing weight coefficient that is effective at the current moment.

[0035] In some optional implementations, the two-level fault-tolerant logic includes:

[0036] Level 1 fault tolerance: When the predicted residual magnitude When the value is less than or equal to a set threshold, a residual penalty term is enabled. The fault-tolerant value function is used, but the normal prediction model and adaptive weighting strategy are maintained.

[0037] Level 2 fault tolerance: When the predicted residual magnitude When the value exceeds the set threshold, immediately switch to the fault-tolerant prediction model with residual correction factor, pause adaptive weight tuning, freeze the weights to the preset fixed conservative value, and trigger the initial position and parameter re-identification process in step S1.

[0038] Fault-tolerant recovery: When the residual amplitude is continuously less than or equal to the set threshold and the system state is stable, the system smoothly switches back to the normal adaptive control mode without disturbance through linear transition of weights and parallel calculation of the model.

[0039] In some optional implementations, the complete control flow of the method is executed in stages, including:

[0040] During the power-on initialization and initial position identification phase, step S1 is executed to establish an accurate prediction model;

[0041] During the torque closed-loop establishment phase in the zero-speed locking interval, steps S2 and S3 are executed to establish a full torque closed loop at zero speed.

[0042] During the smooth switching phase of the acceleration transition range, steps S2 and S3 are executed to achieve a smooth and uninterrupted acceleration from zero speed to low speed.

[0043] During the steady-state operation phase across the full speed range, steps S1, S2, and S3 are continuously executed in coordination to maintain high-precision control across the full speed range.

[0044] An electronic device provided in this application includes a processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and the machine-readable instructions, when executed by the processor, perform any of the methods described above.

[0045] This application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of any of the methods described above. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 A flowchart illustrating the steps of a zero-speed start control method for a permanent magnet synchronous motor provided in this application embodiment. Detailed Implementation

[0048] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0049] Please refer to Figure 1 , Figure 1 A flowchart of a zero-speed start control method for a permanent magnet synchronous motor provided in this application embodiment includes:

[0050] Step S1: When the motor is in a zero-speed stationary state, inject a composite voltage signal containing DC bias component and high-frequency pulsation component into the d-axis direction of the rotor position estimation to forcibly enhance the salient pole effect of the motor, and extract the rotor position error in real time based on the high-frequency current response, and simultaneously complete the identification of the rotor initial electrical angle and the determination of magnetic pole polarity; update the identified rotor position electrical angle, saturated inductance and permanent magnet flux linkage observation values ​​to the prediction model in real time in units of control cycle;

[0051] Step S2: Based on the motor speed, the zero-speed start-up process is divided into a zero-speed lock-up interval and a speed-up transition interval. In each MPC control cycle, the weight coefficients of each control objective in the value function are dynamically and adaptively calculated according to the interval and real-time operating conditions. The value function includes at least a torque tracking term, a flux tracking term, a current constraint penalty term, and a switching frequency penalty term. All basic voltage vectors of the inverter are traversed, and the predicted state of each voltage vector is calculated based on the updated prediction model. The cost value of each voltage vector is calculated in combination with the weight coefficients, and the voltage vector with the smallest cost value is selected as the optimal control quantity output for the current cycle.

[0052] Step S3: In each control cycle, calculate the prediction residual between the predicted current value and the actual sampled value at the previous moment, and compensate the prediction residual in real time into the current prediction model through the residual correction factor to correct the model parameter error; at the same time, according to the magnitude of the prediction residual, execute two-level fault-tolerant logic: when the residual magnitude is less than or equal to the set threshold, maintain the normal control mode and enhance the residual penalty term in the value function; when the residual magnitude is greater than the set threshold, switch to the fault-tolerant prediction model and freeze the adaptive weight strategy to maintain the startup process without interruption; Among them, the prediction model is equivalent to the basic prediction model; the current prediction model is a specific execution model that embeds the prediction residual into the prediction model; the specific execution model switches to the fault-tolerant prediction model when the residual amplitude is greater than a set threshold.

[0053] The above technical solution improves upon three dimensions: position recognition, torque control, and model fault tolerance. It achieves a comprehensive start-up control scheme at zero speed, enabling disturbance-free and accurate position recognition, real-time fusion of model parameters, adaptive optimization of control weights, and active fault tolerance. Specifically, it includes:

[0054] Precise zero-speed rotor position identification with no swaying or slippage throughout the entire process: Employing salient pole enhancement high-frequency composite signal injection and saturated salient pole strengthening technology, the difference in inductance between the d and q axes is forcibly increased, enabling stable identification even for surface-mount PMSMs with low salient pole ratios, thus improving position identification accuracy; By estimating d-axis directional injection and q-axis zero-voltage constraint, no torque component is generated during the identification process, and the rotor remains stationary throughout the entire process, completely eliminating the swaying and mechanical shock caused by traditional identification; Utilizing the rate of change of saturated inductance to achieve online determination of magnetic pole polarity without the need for additional pulse injection, ensuring accurate and reliable polarity determination and avoiding start-up synchronization loss caused by 180° polarity errors.

[0055] Zero-speed full-torque output with fast dynamic response and low torque ripple: Employing a dual-interval adaptive weighted value function, the torque weight is significantly increased and the switching loss weight is reduced in the zero-speed interval, achieving stable zero-speed torque output and shortening torque response time. Multiple objectives, including torque, flux linkage, current, and switching losses, are optimized in a unified manner, reducing torque ripple during startup and eliminating impact and jitter.

[0056] Active model error compensation, strong robustness, and resistance to parameter drift: Employing real-time prediction residual correction and active fault-tolerant control techniques, it directly compensates for model errors caused by stator resistance temperature drift, inverter nonlinearity, and core saturation, without requiring separate parameter identification. A two-level fault-tolerant switching mechanism is constructed, automatically switching to fault-tolerant mode when the model error exceeds a threshold, ensuring uninterrupted and shock-free startup throughout the entire process.

[0057] In some optional implementations, step S1 includes:

[0058] The composite voltage signal is injected only into the d-axis of the rotor position estimation, and the q-axis injection voltage is always zero; the high-frequency pulsation component is a standard cosine pulsation waveform; the value range of the DC bias component is 0.5 to 1.5 times the rated voltage of the motor; the angular frequency of the high-frequency pulsation component is 1 / 10 to 1 / 5 of the inverter switching frequency.

[0059] In some optional implementations, step S1, which involves extracting the rotor position error in real time based on the high-frequency current response and simultaneously completing the rotor initial electrical angle identification and magnetic pole polarity determination, includes:

[0060] Based on the high-frequency q-axis current component in the injected current response, the rotor position error signal is extracted, and the initial rotor electrical angle is obtained by observation through an adaptive phase-locked loop.

[0061] Based on the DC component of the d-axis current caused by the DC bias component, the d-axis flux change rate is calculated online to determine the magnetic polarity of the permanent magnet.

[0062] If the rate of change of the d-axis flux approaches zero, the current magnetic pole polarity is determined to be incorrect, and the initial rotor electrical angle is corrected by 180 degrees to obtain the rotor position electrical angle.

[0063] In some optional implementations, step S1, which involves updating the identified rotor position electrical angle, saturation inductance, and permanent magnet flux linkage observations to the prediction model in real time, on a control cycle basis, includes:

[0064] Update the rotor position electrical angle to the angle reference of Park transformation and inverse Park transformation;

[0065] Update the d-axis inductance and q-axis inductance in the saturated inductance to the inductance parameters in the current prediction equation;

[0066] The permanent magnet flux linkage observations are then updated to the flux linkage parameters in the torque prediction equation.

[0067] In some optional implementations, in step S2, the zero-speed locking range is the range where the speed is less than or equal to 5% of the rated speed, and the acceleration transition range is the range where the speed is between 5% and 20% of the rated speed; the weighting coefficients of each control objective in the dynamic adaptive calculation value function specifically include:

[0068] In the zero-speed locking range, the weighting coefficients of the torque tracking term and the current constraint penalty term are increased, while the weighting coefficient of the switching frequency penalty term is decreased.

[0069] During the acceleration transition range, the weighting coefficient of the torque tracking term is reduced as the rotational speed increases, while the weighting coefficient of the switching frequency penalty term is increased as the rotational speed increases.

[0070] In some optional implementations, when the motor speed is in the transition zone between the zero-speed lock-in zone and the acceleration transition zone, the weight coefficients of each control objective in the dynamic adaptive calculation value function are linearly weighted, including:

[0071] Calculate the speed ratio factor based on the current speed;

[0072] Obtain the baseline weights for the zero-speed locking interval and the acceleration transition interval, respectively.

[0073] The two benchmark weights are linearly weighted based on the aforementioned speed ratio factor to calculate the smoothing weight coefficient that is effective at the current moment.

[0074] In some optional implementations, the two-level fault-tolerant logic includes:

[0075] Level 1 fault tolerance: When the predicted residual magnitude When the value is less than or equal to a set threshold, a residual penalty term is enabled. The fault-tolerant value function is used, but the normal prediction model and adaptive weighting strategy are maintained.

[0076] Level 2 fault tolerance: When the predicted residual magnitude When the value exceeds the set threshold, immediately switch to the fault-tolerant prediction model with residual correction factor, pause adaptive weight tuning, freeze the weights to the preset fixed conservative value, and trigger the initial position and parameter re-identification process in step S1.

[0077] Fault-tolerant recovery: When the residual amplitude is continuously less than or equal to the set threshold and the system state is stable, the system smoothly switches back to the normal adaptive control mode without disturbance through linear transition of weights and parallel calculation of the model.

[0078] In some optional implementations, the complete control flow of the method is executed in stages, including:

[0079] During the power-on initialization and initial position identification phase, step S1 is executed to establish an accurate prediction model;

[0080] During the torque closed-loop establishment phase in the zero-speed locking interval, steps S2 and S3 are executed to establish a full torque closed loop at zero speed.

[0081] During the smooth switching phase of the acceleration transition range, steps S2 and S3 are executed to achieve a smooth and uninterrupted acceleration from zero speed to low speed.

[0082] During the steady-state operation phase across the full speed range, steps S1, S2, and S3 are continuously executed in coordination to maintain high-precision control across the full speed range.

[0083] The following provides a specific embodiment, including the specific implementation of the prediction model, steps S1, S2, and S3:

[0084] I. Mathematical model derivation of the prediction model MPC:

[0085] The essence of MPC is to predict the system state under different voltage vectors based on the discrete time-domain model of the motor, and select the optimal control quantity through the value function. Therefore, the accurate mathematical model is the core of the scheme and the basis of the subsequent original algorithm.

[0086] 1. Physical model of the motor, a continuous time-domain model in the synchronously rotating dq coordinate system;

[0087] The coupled model of the three-phase stationary coordinate system is decoupled into a dq synchronous rotating coordinate system model through Clark transformation (3s / 2s) and Park transformation (2s / 2r). The derivation process is as follows:

[0088] dq-axis voltage equation:

[0089] ;

[0090] In the formula: : Synchronous rotation of the d-axis stator voltage; : Synchronous rotation of the q-axis stator voltage; : Motor stator phase resistance; : Synchronous rotating d-axis stator current; Synchronous rotating q-axis stator current; The time derivative of the d-axis stator flux linkage; The time derivative of the q-axis stator flux linkage; : Rotor electrical angular velocity; d-axis stator flux linkage; : q-axis stator flux linkage.

[0091] dq-axis flux linkage equation:

[0092] ;

[0093] In the formula: d-axis stator flux linkage; : q-axis stator flux linkage; d-axis stator inductance; : q-axis stator inductance; d-axis stator current; : q-axis stator current; : Permanent magnet flux linkage in motor.

[0094] The current state equation (core of the prediction model) is derived by substituting the flux linkage equation into the voltage equation, eliminating the flux linkage term, and rearranging to obtain the differential state equation of the current, which is used for subsequent discrete prediction and optimization of the original algorithm.

[0095] ;

[0096] In the formula: The d-axis stator current is the derivative with respect to time. The q-axis stator current differential with respect to time; Stator phase resistance; d-axis stator inductance; : q-axis stator inductance; d-axis stator current; : q-axis stator current; : Rotor electrical angular velocity; d-axis stator voltage; : q-axis stator voltage; : Permanent magnet magnetic flux.

[0097] Electromagnetic torque and stator flux linkage equations are used for calculating the control objective of the value function and the state constraints of the original algorithm: In Model Predictive Control (MPC), the current is calculated through step 3. However, the ultimate control objectives are torque (to get the motor running) and flux linkage (to stabilize the motor's magnetic field). Step 4's function is to convert "current" into "torque" and "fluid linkage".

[0098] Electromagnetic torque equation (adapted to SPMSM / IPMSM):

[0099] ;

[0100] Part One ( Permanent magnet torque. Generated by the interaction between the permanent magnet and the q-axis current. This is the primary source of torque; Part Two ( ): Reluctance torque (salient pole effect). Generated by the salient pole effect of the motor core. If The motor can then utilize this portion of torque. Where: : Electromagnetic torque of the motor; Number of pole pairs in a motor; an inherent parameter of the motor. Permanent magnet flux linkage; d-axis stator current. : q-axis stator current; d-axis stator inductance; : q-axis stator inductance;

[0101] The stator flux linkage amplitude equation is used to calculate the magnitude of the magnetic field inside the motor. During zero-speed start-up, it is necessary to ensure a stable magnetic field (neither too weak leading to powerlessness nor too strong leading to magnetic saturation). Stator flux linkage amplitude equation:

[0102] ;

[0103] Substitution and The calculation formula is obtained as follows:

[0104] ;

[0105] In the formula: Stator flux linkage amplitude; d-axis stator flux linkage; : q-axis stator flux linkage; d-axis stator inductance; d-axis stator current; Permanent magnet flux linkage; : q-axis stator inductance; : q-axis stator current.

[0106] 2. Construction and algorithm implementation of prediction model based on discrete time domain (digital control core calculation process);

[0107] The frequency converter uses digital control, which requires discretizing the continuous model. This solution uses the forward Euler method (most commonly used in industrial scenarios due to its low computational cost). The discretization formula is as follows:

[0108] ;

[0109] In the formula: The derivative of the state variable x with respect to time; : The state variable value at the next control moment (the target to be calculated); : The value of the state variable at the current control moment; : Inverter control (sampling) cycle.

[0110] Derivation of the dq-axis current prediction equation:

[0111] Substituting the forward Euler formula into the current state equation and rearranging the terms, we obtain the predicted current value for the next time step, which is also the core foundation of the subsequent original fault-tolerant algorithm:

[0112] d-axis current prediction:

[0113] —Equation (1);

[0114] In the formula: : Predicted d-axis stator current at the next control moment; Stator phase resistance; Control cycle; d-axis stator inductance; : The current d-axis stator current sample value; : q-axis stator inductance; : The current rotor electrical angular velocity; : The current q-axis stator current sample value; : The d-axis stator voltage control quantity at the current moment.

[0115] q-axis current prediction:

[0116] —Equation (2);

[0117] In the formula: : Predicted q-axis stator current at the next control moment; Stator phase resistance; Control cycle; : q-axis stator inductance; : The current q-axis stator current sample value; d-axis stator inductance; : The current rotor electrical angular velocity; : The current d-axis stator current sample value; Permanent magnet flux linkage; : The q-axis stator voltage control quantity at the current moment.

[0118] Zero-speed start With the disappearance of cross-coupling terms, the prediction model is further simplified, requiring less computation and achieving higher accuracy.

[0119] ;

[0120] In the formula: Predicted d-axis current value at the next moment under zero velocity; Predicted q-axis current value at the next moment under zero velocity; Stator phase resistance; Control cycle; d-axis stator inductance; : q-axis stator inductance; : Current d-axis current sampling value at the current moment; : Current q-axis current sample value; : Current d-axis voltage control quantity; : Current q-axis voltage control quantity.

[0121] Torque and flux prediction equations:

[0122] The equations (1) and (2) are... Substituting the torque and flux linkage equations, we obtain the predicted state value for the next time step:

[0123] Torque prediction:

[0124] —Equation (3);

[0125] In the formula: : Predicted electromagnetic torque value at the next control moment; Number of pole pairs in the motor; Permanent magnet flux linkage; : Predicted q-axis current value at the next moment; d-axis stator inductance; : q-axis stator inductance; : Predicted d-axis current value at the next moment.

[0126] Stator flux linkage amplitude prediction:

[0127] —Equation (4);

[0128] In the formula: : Predicted value of stator flux linkage amplitude at the next control moment; d-axis stator inductance; : Predicted d-axis current value at the next moment; Permanent magnet flux linkage; : q-axis stator inductance; : Predicted q-axis current value at the next moment.

[0129] II. Specific implementation of step S1:

[0130] Because the prediction model is at zero velocity (ω) e The original model (≈0) suffers from defects such as unmeasurable position, inaccurate model parameters, and unobservable flux linkage, making it impossible to obtain the rotor position through the fundamental wave model. This embodiment specifically addresses these issues using salient pole enhancement injection, synchronous observation, and polarity determination, making the basic model usable and accurate at zero speed. The identification and observation results from the salient pole enhancement-type oscillation-free initial position identification and zero-speed flux linkage observation fusion algorithm are fed back into the prediction model in the second part in real time, eliminating the need for a separate identification stage. This corrects key parameters such as inductance, flux linkage, and rotor position, forming a closed loop of "model → algorithm → corrected model," thereby improving the overall control accuracy of MPC.

[0131] This algorithm is executed at zero speed throughout, with no rotor oscillation. It consists of four synchronously executed stages: salient pole enhancement injection, synchronous position observation, online polarity determination, and real-time model correction, which are completely synchronized with the MPC control cycle.

[0132] Salient-pole enhancement-mode high-frequency signal injection:

[0133] In a zero-speed stationary state, a composite signal of DC bias voltage and high-frequency pulsating voltage is injected into the d-axis direction for rotor position estimation. The salient pole effect is enhanced by forcing the iron core to saturate through DC bias, and the rotor position error is extracted through the high-frequency pulsating signal.

[0134] Inject composite signals and perform mathematical modeling:

[0135] To strictly limit the injection of composite signals only to the d-axis estimation and keep the estimated q-axis injection voltage at zero throughout the estimation process, thus avoiding rotor oscillation caused by torque components generated by the q-axis signal, a mathematical model for estimating the dq-axis injection voltage is constructed, as shown in the following formula:

[0136] —Equation (5);

[0137] In the formula: : salient pole enhanced DC bias voltage; : Amplitude of high-frequency pulse voltage; : Angular frequency of the high-frequency injected signal; :time; Estimate the composite voltage signal injected along the d-axis; Estimate the q-axis injection voltage.

[0138] DC bias voltage component (salient pole enhancement core design):

[0139] A directional DC bias voltage is applied to the estimated d-axis, causing the core in the d-axis direction to enter a controllable local magnetic saturation state, thus artificially increasing the equivalent inductance of the d-axis after saturation. With q-axis inductance The difference in saliency forcefully enhances the salient pole effect of the motor, solving the problem of ineffective identification when the salient pole ratio is extremely low. The parameter tuning rules are as follows: Value range: The rated voltage of the motor is doubled; the setting basis is: if the voltage is too low, the salient pole enhancement effect will be insufficient, and if the voltage is too high, it will cause overcurrent and irreversible damage due to deep magnetic saturation; zero speed adaptation: there is no back electromotive force at zero speed, the DC bias only generates directional excitation current and does not generate electromagnetic torque, and the rotor remains absolutely stationary.

[0140] High-frequency pulse voltage component (position error detection design):

[0141] A high-frequency pulsating signal is superimposed on a DC bias, utilizing the high-frequency current response generated by the motor's saturated salient poles to carry the error information between the rotor's actual and estimated positions, providing the raw signal for subsequent position error extraction. The parameter tuning rules include: amplitude... :Pick DC bus voltage This ensures that the high-frequency current response can be detected by the ADC without interfering with the main torque control loop; angular frequency :Pick It avoids the fundamental control frequency and the inverter switching frequency to prevent signal aliasing and facilitates subsequent bandpass filtering extraction; waveform characteristics: adopts standard cosine pulse waveform, zero mean, symmetrical distribution, no additional DC component, and does not cause rotor micro-motion.

[0142] Hardware-level composite signal generation:

[0143] Relying on the inverter's hardware PWM generation unit, a dual-carrier hardware superposition technology is employed: the fundamental PWM carrier generates the DC bias component, and a high-frequency independent carrier generates the pulse component. The composite signal is directly synthesized in hardware, eliminating software calculation delays and waveform distortion. The composite signal is synchronized with the MPC control cycle. Hardware-level phase locking is implemented at the ADC sampling time to ensure timing alignment of the entire process of injection, sampling, and calculation.

[0144] Safety constraints and zero-velocity protection during injection:

[0145] Real-time monitoring of the stator three-phase current amplitude; if the current approaches the inverter's maximum allowable current... Automatic linear reduction and This prevents overcurrent damage to components. Zero-speed start-up ensures continuous and stable injection throughout the entire process, without interruptions or pulse jumps, guaranteeing that the rotor remains undulating and free from mechanical vibration, making it suitable for zero-speed identification requirements of potential energy loads such as elevators and cranes.

[0146] Pre-synchronization of the injected signal with the MPC model:

[0147] Immediately after the injection signal is generated, , , The three core parameters are written into the hardware dual-port RAM, providing real-time parameter benchmarks for the subsequent position error extraction in step 2 and model fusion in step 4, achieving seamless connection of the entire process of injection-observation-control.

[0148] Position error extraction and synchronous observation:

[0149] This step, based on zero-speed operating condition constraints, extracts the rotor position error signal from the current response after high-frequency injection. Through synchronous detection and adaptive phase-locked loop (PLL), the rotor position electrical angle is observed in real time under the premise that the rotor is not oscillating, providing a position reference for MPC control.

[0150] Core constraint premise:

[0151] All calculations in this section are based on the following three zero-speed specific constraints to ensure that the rotor remains stationary and free from torque disturbance throughout the entire process: Zero speed: Back EMF = 0, no velocity coupling term; Injection constraint: only inject composite high-frequency voltage in the estimated d-axis, the estimated q-axis injection voltage is always 0, no torque component is generated; Salient pole saturation: the d-axis inductance is DC biased to... The q-axis remains There is a convex range. .

[0152] Coordinate system rotation transformation:

[0153] Actual rotor synchronous rotation coordinate system: Algorithm for estimating synchronously rotating coordinate systems: Rotor position error angle: (Actual electrical angle - estimated electrical angle).

[0154] Rotation transformation matrix from the estimated coordinate system to the actual coordinate system: (rotation) ) ; Inverse transformation matrix from the actual coordinate system to the estimated coordinate system: (reverse rotation) ) ;

[0155] In the formula: : The actual electrical angle of the rotor; : Estimated electrical angle of rotor; The angle of error between the actual position and the estimated position; : Estimate → actual coordinate system rotation transformation matrix.

[0156] Derivation of high-frequency voltage and current response:

[0157] Estimate the d-axis injected high-frequency voltage by retaining only the high-frequency pulsation component (ignoring DC bias):

[0158] ; ;

[0159] High-frequency voltage in estimated coordinate system → actual coordinate system:

[0160] Substituting the estimated high-frequency voltage in the coordinate system into the transformation matrix, we obtain the actual dq-axis high-frequency voltage:

[0161] ;

[0162] High-frequency current response in actual coordinate system:

[0163] At zero speed with no coupling terms, the actual dq-axis high-frequency current is obtained by integrating the high-frequency voltage:

[0164] ;

[0165] Actual coordinate system → Estimated high-frequency current in the coordinate system:

[0166] The high-frequency current in the actual coordinate system is inversely transformed back to the estimated coordinate system, with a focus on extracting the estimated q-axis high-frequency current. (Core characteristic signal of position error).

[0167] Position error characteristic equation:

[0168] The core relationship between the estimated q-axis high-frequency current and position error is summarized as follows:

[0169] —Equation (6);

[0170] In the formula: Estimate the high-frequency current component extracted along the q-axis; : High-frequency injection voltage amplitude; : Angular frequency of the high-frequency injected signal; d-axis equivalent inductance after DC bias saturation; : q-axis stator inductance;

[0171] Among them: when (Estimated location = actual location) The error signal disappears. When , Include The component directly reflects the magnitude and direction of the position error.

[0172] Synchronous detection and error demodulation:

[0173] Bandpass filtering: for Perform bandpass filtering and retain Frequency components are filtered to remove low frequencies and noise. Synchronous detection: the filtered... With the same frequency reference signal Multiply and demodulate to obtain the DC component of the position error. Error normalization: It is converted into a position error signal that can be input to the PLL, thereby achieving error linearization.

[0174] Adaptive PLL rotor position synchronous observation:

[0175] The demodulated position error is input into an adaptive phase-locked loop (PLL) to form a closed-loop observation: the PLL adjusts the error signal accordingly. Real-time adjustment of estimated electrical angle ;drive This allows the estimated position to converge quickly to the true position; there is no rotor oscillation throughout the entire process at zero speed, and the position observation accuracy can reach [percentage missing]. Electrical angle; the observation results are output to the MPC model in real time for coordinate transformation and torque orientation.

[0176] Online determination of magnetic polarity:

[0177] Traditional methods require a separate pulse voltage injection to determine polarity. This algorithm utilizes the saturation effect of salient pole enhancement injection to simultaneously determine polarity while observing position, eliminating the need for additional signal injection and preventing rotor oscillation.

[0178] Core principle of polarity determination:

[0179] The saturation characteristics of the rotor core in a permanent magnet synchronous motor differ significantly between the N and S pole directions. When the d-axis is estimated to be aligned with the rotor's N pole, the d-axis core is significantly saturated due to the combined effects of the permanent magnet's magnetic field and the injected DC bias, resulting in a rapid decrease in d-axis inductance as the current increases. Conversely, when the d-axis is estimated to be aligned with the rotor's S pole, the d-axis core is in a demagnetized state with extremely weak saturation, and the d-axis inductance remains almost constant with current changes. By calculating the rate of change of the d-axis inductance in real time, the magnetic pole polarity can be distinguished without the need for the separate pulse impact required in traditional methods.

[0180] Detailed execution steps:

[0181] After the PLL converges, the DC component of the d-axis current is extracted in real time. Calculate the rate of change of the d-axis equivalent inductance:

[0182] —Equation (7);

[0183] In the formula: : The rate of change of d-axis inductance with current; d-axis flux linkage observation; : Initial inductance along the d-axis; d-axis current DC component.

[0184] For the calculated Perform numerical sign and amplitude discrimination:

[0185] Case A (negative value and significant amplitude): If Furthermore, the absolute value increases significantly. Conclusion: When the injection direction is consistent with the direction of the permanent magnet's magnetic field (magnetizing), the core magnetic flux density increases, rapidly entering the deep saturation region, and the inductance value decreases sharply. Action: Determine the current estimated angle. The polarity is correct and no correction is needed.

[0186] Case B (approaching zero): If Conclusion: The injection direction is opposite to the direction of the permanent magnet's magnetic field (demagnetization). The externally injected magnetic field is canceled out by the permanent magnet's magnetic field, and the core has not reached the expected saturation depth. Action: Determine that the current polarity is incorrect and trigger a 180° electrical angle correction.

[0187] Execution correction: When "Case B" is determined, the angle correction logic is executed immediately.

[0188] ;

[0189] Closed-loop verification: After correction, the algorithm continues to monitor. The value of the polarity is determined. If it becomes significantly negative, the current polarity is locked as the final result, completing the polarity determination process.

[0190] Result locking and model synchronization: After polarity determination and correction, the final rotor position electrical angle is locked. Lock; synchronize in real time to the coordinate transformation module of the MPC prediction model to ensure complete accuracy of d / q axis decoupling and torque orientation.

[0191] Fusion of identification results with MPC prediction model:

[0192] This step involves the deep integration of the salient-pole enhanced zero-swing initial position identification and zero-speed flux linkage observation fusion algorithm with the MPC control core. The rotor position, saturated inductance, and permanent magnet flux linkage observations obtained from zero-speed synchronous identification are embedded module by module into the MPC discrete prediction model according to the control cycle-level real-time synchronization rules. This completely eliminates the timing discrepancy and parameter mismatch issues between identification and control, and is synchronized with the MPC rolling optimization cycle throughout. Strict synchronization.

[0193] Identification / observation result cache and periodic synchronization lock:

[0194] After the salient-pole enhancement-type oscillation-free initial position identification and zero-speed flux linkage observation fusion algorithm completes salient-pole enhancement injection, position observation, and polarity determination, the core output data is immediately written into the inverter hardware buffer. The data includes: rotor position electrical angle. d-axis equivalent inductance after saturation q-axis inductance Permanent magnet flux linkage observations .

[0195] The hardware clock unit triggers the MPC control cycle synchronization signal to ensure that cached data is synchronized in each cycle. The initial time is read by the MPC prediction model operation unit, achieving zero-delay alignment between the identification output and the model update, and eliminating cross-cycle data deviation.

[0196] Rotor electrical angle correction coordinate transformation reference:

[0197] The identified rotor position electrical angle It replaces the angle reference of Park transformation and inverse Park transformation in the MPC process in real time, replacing the original estimated angle.

[0198] Coordinate transformation correction execution: Park transformation: Current in stationary coordinate system , Convert to dq axis current , At that time, adopt Calculate the transformation matrix to eliminate torque orientation offset caused by angle errors. Inverse Park transformation: This involves converting the dq-axis voltage output from the MPC. , Convert to When using coordinate system voltage, synchronous use This ensures that the voltage vector output is perfectly matched with the actual position of the rotor.

[0199] In zero-speed scenarios, this correction directly guarantees torque orientation accuracy of ±0.5° electrical angle, fundamentally preventing zero-speed step loss and torque jitter.

[0200] Saturated inductor parameter update current prediction model:

[0201] Extracting the saturated d-axis inductance from the fusion algorithm of salient pole enhancement type oscillating initial position identification and zero-velocity flux observation With q-axis inductance Replace the original inductance parameters in the MPC discrete current prediction equations (Equations 1 and 2). , .

[0202] Current prediction model parameter update:

[0203] d-axis current prediction equation updated:

[0204] ;

[0205] q-axis current prediction equation updated:

[0206] ;

[0207] At zero speed ( The simplified current prediction model is simultaneously adopted. , It fully compensates for the inductance drift error caused by core saturation and improves the accuracy of current prediction.

[0208] Torque prediction model corrected by permanent magnet flux linkage observations:

[0209] The permanent magnet flux observation values ​​output from the flux observation stage of the fusion algorithm of salient pole enhancement type oscillating non-oscillating initial position identification and zero velocity flux observation are used. Replace the nominal flux in the MPC torque prediction equation (Equation 3). .

[0210] Torque prediction model update:

[0211] ;

[0212] This correction eliminates the deviation between the nominal and actual values ​​of the magnetic flux linkage, ensuring that the torque prediction error at zero speed is less than 1%, and supporting the accurate calculation of the value function of the dual-interval adaptive weighted model predictive torque control algorithm.

[0213] Closed-loop verification of fused model parameters:

[0214] The MPC processing unit performs threshold verification on the updated inductance, flux linkage, and angle parameters: Inductance parameter: Verification , Whether it is within ±30% of the motor's rated inductance; flux linkage parameters: verification. Whether it is within ±15% of the rated flux linkage; Angle parameter: verification Does the rate of change conform to the zero-velocity stationary characteristic?

[0215] Verification passed: The parameters officially take effect and enter the MPC rolling optimization process; Verification failed: The fusion algorithm of salient pole enhanced non-swaying initial position identification and zero-velocity flux observation is triggered for rapid re-identification, and the parameter correction is completed within 1 control cycle to ensure that the model is always effective.

[0216] The integrated model is integrated into the entire MPC control process: the updated prediction model is directly integrated into the MPC full voltage vector traversal, state prediction, and value function calculation stages, serving as the sole calculation benchmark. The above steps are repeated in each control cycle to achieve a complete closed loop of identification-observation-model correction-control output, completely solving the core defect of traditional solutions that "fixed parameters after identification and inability to dynamically adapt".

[0217] III. Specific implementation of step S2:

[0218] The three-phase two-level inverter used in this embodiment can only output 8 basic voltage vectors (6 non-zero vectors and 2 zero vectors), and the corresponding switching states (1 for the upper bridge arm and 0 for the lower bridge arm) are shown in the table below:

[0219] Table 1. Voltage Vector Table

[0220] To meet the real-time requirements of industrial frequency converters, this solution adopts predictive time domain. Control Time Domain That is, each control cycle traverses 8 voltage vectors, predicts the system state under their action, selects the optimal vector for output, minimizes the amount of computation, and can complete real-time calculations in ordinary MCUs / DSPs.

[0221] Traditional MPC value functions use fixed weighting coefficients, which have drawbacks: the core control objectives differ between zero-speed start-up, acceleration transition, and steady-state operation, and fixed weights cannot take into account performance under all operating conditions: at zero speed, torque tracking and overcurrent constraints must be prioritized, while at steady state, switching losses and torque ripple must be minimized; the weighting coefficients need to be manually tuned offline, requiring repeated adjustments when adapting to different motors and loads, making engineering implementation difficult; under heavy load and sudden changes, fixed weights cannot quickly strengthen constraint terms, easily leading to overcurrent and loss of synchronization.

[0222] In this embodiment, for the PMSM zero-speed start scenario, a dual-range weighted adaptive architecture is adopted, dividing the system into a zero-speed locking range (5% of rated speed) and a speed-up transition range (5%-20% of rated speed). Different ranges are set with independent weighted adaptive logic to accurately match the control requirements of the entire zero-speed start process. Based on the fuzzy adaptive weight tuning model of operating condition characteristics, no manual weight tuning is required. In each control cycle, the weight coefficients are updated in real time according to torque tracking error, flux deviation, and current margin to achieve optimal control under all operating conditions. The weighted adaptive control is integrated with overcurrent fault-tolerant control. During heavy load sudden changes, the current constraint weight is automatically strengthened to achieve active current limiting, and there is no overcurrent risk during zero-speed heavy load start.

[0223] In Model Predictive Control (MPC) systems, the value function serves as the "decision-making brain" of the controller. Its core function is to address the pain points of traditional MPC during zero-speed startup: difficulty in simultaneously addressing multiple objectives and the need for repeated parameter adjustments. Specifically, its functions are as follows:

[0224] Achieving multi-objective coordinated control: It unifies the core objectives (torque tracking, constant flux) of PMSM control with hardware constraints (current limiting, switching losses) within a single mathematical framework. By calculating the "cost" under different voltage vectors, it selects the vector with the minimum cost, thereby achieving optimal control.

[0225] This function endows the system with "adaptive" capabilities: Unlike traditional fixed weights, this function introduces adaptive weights that change with operating conditions. This means that it can prioritize torque during zero-speed start-up (to prevent runaway) and automatically balance switching losses (to improve efficiency) during acceleration, without the need for manual intervention.

[0226] Ensuring hardware safety (fault tolerance): The function explicitly includes a current penalty term. Once the predicted current exceeds the limit, the penalty term will increase sharply, forcing the controller to abandon the vector, thereby achieving "soft protection" for the frequency converter and preventing the risk of the inverter crashing at zero speed and high torque.

[0227] The value function in this embodiment It is a composite function containing four weighted terms. The following is its complete construction logic and detailed calculation steps:

[0228] The goal of this algorithm is to select the vector that optimizes the system state from the eight basic voltage vectors of the inverter within each control cycle. Value function Defined as:

[0229] ---Equation (8);

[0230] In the formula: : Multi-objective value function value; the smaller the value, the better the control effect. : Adaptive weighting coefficients for the torque tracking term in the current control cycle; Electromagnetic torque setpoint; : Predicted electromagnetic torque value at the next moment; : Adaptive weighting coefficient of the flux tracking term in the current control cycle; Stator flux linkage amplitude setpoint; : Predicted value of stator flux linkage amplitude at the next moment; : Adaptive weighting coefficient of the current constraint penalty term in the current control cycle; Overcurrent constraint penalty function; Predicted values ​​of d-axis and q-axis currents at the next moment; : Adaptive weighting coefficient of the switching frequency penalty term in the current control cycle; Switching frequency penalty function; The current switching status of the three-phase bridge arms of the inverter; : The switching state of the three-phase bridge arm of the inverter at the next moment.

[0231] In order to calculate First, its four components must be calculated.

[0232] Part 1: Torque Tracking Term Calculation )( This parameter ensures that the motor's output torque accurately follows a given value, especially providing sufficient acceleration torque during zero-speed starts. Calculation formula: ;

[0233] Torque prediction value Derivation (based on the MPC model):

[0234] Current prediction: Using the discretized model (Equations 1 and 2), the current voltage vector is... Substitute the values ​​and calculate the current at the next moment:

[0235] ;

[0236] ;

[0237] Torque calculation: the predicted torque Substituting into the torque equation (Equation 3):

[0238] ;

[0239] Error calculation: Calculate the predicted torque and the given torque. The absolute difference.

[0240] Part Two: Calculation of Magnetic Linkage Tracking Term ( This item is used to maintain a constant air gap flux, ensuring that the motor's magnetic circuit is not saturated and that it operates with high efficiency.

[0241] Calculation formula: ;

[0242] Magnetic flux prediction value Calculation:

[0243] Using the predicted Substituting into the flux linkage amplitude equation (Equation 4):

[0244] ;

[0245] Part Three: Calculation of Current Constraint Penalty Term This is the key design feature of the scheme to prevent overcurrent during zero-speed heavy-load startup. It is not a simple linear term, but a piecewise nonlinear function.

[0246] Calculation formula:

[0247] ;

[0248] In the formula: The final score for the current constraint penalty term indicates that the higher the score, the more dangerous and unusable the voltage vector is. : Adaptive penalty intensity coefficient. : Severity of overcurrent.

[0249] penalty function Detailed logic (hard constraint design):

[0250] set up To predict the current amplitude, This is the maximum allowable current for the frequency converter.

[0251] Safety situation ( ): Current safety, no penalty;

[0252] Dangerous situations ( ): The more the current is exceeded, the more the penalty value increases exponentially.

[0253] Part Four: Calculation of Switching Frequency Penalty Term ( This item is used to suppress the switching losses of the inverter and extend the life of the hardware.

[0254] Calculation formula: ;

[0255] Switch state change Calculation:

[0256] Let the current switch state be The switching state of the prediction vector is .

[0257] ;

[0258] If the number of bridge arms switching between the current vector and the vector at the previous time step is greater... The higher the value, the higher the switching loss.

[0259] Dual-interval adaptive weight calculation, weight It's not a constant, but rather automatically adjusted based on the rotational speed. The calculation process is as follows:

[0260] Step 1: Determine the operating range; Range 1 (Zero-speed lock-in zone): Rated speed. Logic: Maximum torque is required at this speed, and overcurrent is not allowed. Calculation: Set to a larger value. Set as the minimum value, Set to the maximum value to prevent starting shock. Range 2 (acceleration transition zone): Rated speed. Logic: At this point, dynamic response and switching losses need to be balanced. Calculation: Decrease as rotational speed increases; It increases with increasing rotational speed.

[0261] Step 2: Real-time weight generation formula (fuzzy adaptive logic); the weights are functions based on error and speed. Taking torque weight as an example (zero speed region):

[0262] ;

[0263] In the formula: : Benchmark weight. Error amplification factor. Current torque error. Derivation: The larger the torque error, The larger the value, the more aggressively the controller will try to correct the torque error in the current cycle.

[0264] In the code or hardware implementation, the calculation process of this function is as follows:

[0265] Step 1: Input the current real-time status of the system. Input: Current current Current speed Current switch status The real-time physical quantities sampled by the frequency converter and given by the system are used as calculation inputs, and all quantities are at the current control moment. Valid values:

[0266] 1. : Current d-axis stator current sampling value;

[0267] 2. : Current q-axis stator current sampling value;

[0268] 3. : The current rotor electrical angular velocity;

[0269] 4. : Current state of the inverter's three-phase bridge arm switches (1 = upper bridge arm on, 0 = lower bridge arm on).

[0270] 5. Electromagnetic torque setpoint;

[0271] 6. Stator flux linkage amplitude setpoint;

[0272] 7. : Maximum allowable output current of the frequency converter (hardware safety limit);

[0273] 8. Fixed parameters of the motor: stator phase resistance d / q axis inductance Permanent magnet flux Extreme logarithms .

[0274] Step 2: Determine the operating range and calculate the adaptive weight coefficient in real time.

[0275] Based on the current rotor electric angular velocity Divide the execution interval and calculate four adaptive weights according to the dual-interval logic (determining the priority of each component of the value function); weight calculation (executed by interval):

[0276] : Adaptive weighting of torque tracking term (priority for controlling torque tracking accuracy);

[0277] Adaptive weighting of flux tracking term (controls flux stability priority);

[0278] : Adaptive weighting of current constraint penalty term (controls the strength of overcurrent protection);

[0279] : Adaptive weighting of switching frequency penalty term (controlling switching loss priority);

[0280] Step 3: Traverse the 8 basic voltage vectors of the inverter, and perform state prediction vector by vector; for the 8 fixed voltage vectors that the inverter can output (6 non-zero vectors) +2 zero vectors Iterate through each step and calculate the next time step using the MPC discrete prediction model. System status:

[0281] Predicting d / q-axis currents: Substitute the zero-velocity optimized discrete current prediction equations and calculate:

[0282] : Predicted value of d-axis stator current at the next moment;

[0283] Predicted q-axis stator current at the next moment;

[0284] Predicting electromagnetic torque and stator flux linkage: Substitute the predicted current into the torque and flux linkage equations to calculate:

[0285] : Predicted electromagnetic torque value at the next moment;

[0286] : Predicted value of stator flux linkage amplitude at the next moment;

[0287] Calculate the predicted current amplitude:

[0288] Predicted total stator current (basis for overcurrent judgment);

[0289] Step 4: Calculate the four core components of the value function vector by vector; for each voltage vector, calculate the four penalty components: torque tracking, flux tracking, current constraint, and switching loss. All quantities represent the cost of the current vector: Torque tracking component. Magnetic Link Tracking Sub-item Current constraint penalty item (Core security items), Switching frequency penalty items .

[0290] Step 5: Calculate the total value function of a single vector. Sum the four components to get the first... The total surcharge of each voltage vector ( (corresponding to 8 vectors)

[0291] formula: ;

[0292] Among them, the overall score of a single vector is as follows: the lower the score, the better and safer the control effect of that vector.

[0293] Step 6: Select the optimal value function and the optimal voltage vector;

[0294] 1. Calculation of the optimal value function:

[0295] formula:

[0296] Optimal value function (minimum total score among 8 vectors)

[0297] 2. Match the optimal vector:

[0298] turn up The corresponding voltage vector number is the optimal control vector for the current cycle.

[0299] Step 7: Output the optimal control value to complete the calculation for this cycle:

[0300] 1. Determine the three-phase switch states corresponding to the optimal voltage vector. Write to the inverter PWM register

[0301] 2. The IGBT power module outputs this vector to control the motor's operation.

[0302] 3. Update system status variables and wait for the next control cycle. Triggered, the entire process described above is repeated;

[0303] Dual-interval adaptive weight tuning logic:

[0304] This section addresses the entire PMSM zero-speed start-up process, dividing the operating conditions into a zero-speed lock-up interval and a speed-up transition interval. It uses an adaptive weighting tuning rule to achieve optimal control with dual objectives: full torque at zero speed and stable operation at speed-up. The weighting coefficients are updated in real time according to the operating conditions, eliminating the need for manual offline tuning and making it compatible with all models of PMSMs in the same power range.

[0305] Interval 1: Zero-speed lockout interval (0%~5% of rated speed). The core control objectives for this interval are zero-speed full torque output, no overcurrent, and no step loss. The weighted adaptive logic is as follows:

[0306] 1. Torque tracking weights:

[0307] ;

[0308] In the formula: Adaptive weights for the torque tracking term in the current control cycle; Torque baseline weight; Torque error amplification factor; Electromagnetic torque setpoint; : The actual value of the electromagnetic torque at the current moment;

[0309] Tuning logic: The larger the torque tracking error, the higher the weight, forcing the controller to prioritize correcting the torque deviation.

[0310] 2. Current constraint weights:

[0311] ;

[0312] In the formula: : Adaptive weighting of the current constraint penalty term in the current control cycle; Current reference weight; : Current margin amplification factor; : Current d / q-axis stator current sampling value; : Maximum allowable output current of the frequency converter;

[0313] Setting logic: The closer the predicted current is to the current limit value, the greater the weight increases exponentially, actively avoiding the risk of overcurrent.

[0314] 3. Magnetic Linkage Tracking Weights:

[0315] ;

[0316] In the formula: : Weight of the flux tracking term in the current control cycle; : Magnetic flux reference weight;

[0317] Tuning logic: Fixed weights ensure constant stator flux amplitude at zero speed, avoiding magnetic saturation / undermagnetization.

[0318] 4. Switching loss weight:

[0319] ;

[0320] In the formula: Weight of the switching frequency penalty term in the current control cycle; : Switching loss baseline weight;

[0321] Tuning logic: extremely low fixed weight, prioritizes torque performance at zero speed, and does not optimize switching losses for the time being.

[0322] Interval 2: Acceleration transition range (5%~20% of rated speed); The core control objectives for this interval are smooth acceleration, torque ripple suppression, and switching loss balancing. The weighted adaptive logic is as follows:

[0323] 1. Torque tracking weights:

[0324] ;

[0325] In the formula: Adaptive weights for the torque tracking term in the current control cycle; : Torque benchmark weight during acceleration phase; Rotational speed attenuation coefficient; Current rotor electrical angular velocity; Rated electrical angular velocity of the motor;

[0326] Tuning logic: The higher the speed, the torque weight gradually decreases to balance dynamic response and stability.

[0327] 2. Switching loss weight:

[0328] ;

[0329] In the formula: : Adaptive weighting of the switching frequency penalty term in the current control cycle; : Baseline weight for switching losses in the acceleration phase; Speed ​​amplification factor; Current rotor electrical angular velocity; Rated electrical angular velocity of the motor;

[0330] Tuning logic: The higher the rotational speed, the higher the weight of switching loss, which suppresses the increase of switching frequency and reduces losses.

[0331] 3. Magnetic Linkage Tracking Weights:

[0332] ;

[0333] In the formula: : Adaptive weights for the flux tracking term in the current control cycle; : Flux flux error amplification factor; Stator flux linkage amplitude setpoint; : The actual value of the stator flux linkage amplitude at the current moment;

[0334] Tuning logic: The greater the magnetic flux deviation, the higher the weight, ensuring the smooth operation of the magnetic flux during the acceleration process.

[0335] 4. Current constraint weight: Fixed weight retains basic overcurrent protection capability.

[0336] ;

[0337] Definition of physical quantity: Weight of the current cycle current constraint penalty term; The current reference weight for the speed-up transition range is set according to the inverter's power rating, overcurrent protection threshold, and load characteristics.

[0338] Interval switching logic:

[0339] When the motor speed transitions from a low-speed range to a high-speed range (especially in the critical range of 4% to 6% of rated speed), a weighted smooth transition strategy is required to avoid abrupt changes in control parameters (such as step changes in weighting coefficients) caused by the range switching. This strategy uses linear weighting of the speed ratios to ensure a continuous and smooth transition of the weighting coefficients between the two ranges, thereby suppressing torque fluctuations and improving dynamic performance. The specific implementation is as follows:

[0340] The operating conditions are divided into three zones based on the PMSM zero-speed start characteristics, clearly defining the speed boundaries and core control objectives:

[0341] 1. Range 1 (Zero-speed lock-up zone): Speed ​​range ;

[0342] Control objective: Suppress zero-speed jitter and enhance torque output capability.

[0343] Weighting characteristics: Emphasis on torque tracking (weighting coefficients) Larger) and current suppression (weighting coefficient) high).

[0344] 2. Range 2 (Acceleration Transition Zone): Speed ​​Range ;

[0345] Control objective: Rapidly increase rotational speed while maintaining both efficiency and stability.

[0346] Weighting characteristics: Balance torque and flux tracking (weighting coefficient adjustment, such as reducing) ,Increase ).

[0347] 3. Transition zone (smooth switching zone): Speed ​​range ;

[0348] Function: To establish a smooth transition zone between interval 1 and interval 2 to avoid abrupt changes in weights.

[0349] The core mechanism of the weighted smooth transition strategy is as follows: Weight coefficients are not directly switched within the transition zone. Instead, a linear weighting method based on rotational speed ratio is used, allowing the weights to change continuously and gradually with rotational speed. The core formulas and definitions are as follows:

[0350] 1. Calculation of speed ratio factor:

[0351] ;

[0352] 2. Smoothing weight calculation:

[0353] ;

[0354] In the formula: : Real-time electrical angular velocity of the rotor; Rated electrical angular velocity of the motor; Rotational speed ratio factor, dimensionless, range of values. ; Weighting coefficients for the zero-speed locking region (interval 1), including... , , , ; Weighting coefficients for the acceleration transition zone (interval 2), including... , , , ; : Smoothing weights, the weight coefficients that take effect in the current period of the smooth transition zone.

[0355] Switching logic:

[0356] when hour, The weights are entirely based on interval 1. When hour, It uses the weights of interval 2 entirely. Within the interval, Linear growth.

[0357] Implementation steps and logical flow:

[0358] This process and Model Predictive Control (MPC) control cycle Strict synchronization, executed once per control cycle:

[0359] 1. Rotor speed detection and range determination: The current rotor electrical angular velocity is acquired in each control cycle. . judge Has it fallen into the transition zone? ).

[0360] 2. Calculate the speed ratio factor Substitute the values ​​into the speed ratio formula to calculate the current... The value determines the position of the rotational speed within the transition zone.

[0361] 3. Linear weighting of weighting coefficients: The weighting coefficients for each control objective (such as torque, flux linkage, current, and switching losses) are calculated using weighting.

[0362] ;

[0363] In the formula: Real-time weight of torque tracking term; Torque tracking weight in the zero-speed lock-up zone; Torque tracking weight in the acceleration transition zone;

[0364] ;

[0365] In the formula: Real-time weights of magnet link tracking items; Zero-velocity locking region flux tracking weights; : Flux tracking weight in the acceleration transition zone;

[0366] ;

[0367] In the formula: Real-time weights of current constraint terms; : Current constraint weight in the zero-velocity locking region; : Current constraint weight in the acceleration transition region;

[0368] ;

[0369] In the formula: Real-time weight of switching loss term; Zero-speed lockout region switching loss weight; Weight of switching losses in the speed-up transition zone.

[0370] 4. Value function calculation and vector selection:

[0371] The value function of each voltage vector is calculated using weighted coefficients. Select The corresponding voltage vector output.

[0372] 5. Dynamic updates:

[0373] Each cycle is recalculated based on the latest rotational speed. Weights are used to achieve a dynamic and smooth transition.

[0374] Detailed design of each constraint:

[0375] 1. Overcurrent constraint penalty term: Implements hard overcurrent constraint on the frequency converter to prevent overcurrent damage to components during zero-speed heavy-load startup. The penalty term is designed as follows:

[0376] ;

[0377] In the formula: Output value of the overcurrent constraint penalty function; : Predicted d-axis current value at the next moment; : Predicted q-axis current value at the next moment; : Maximum allowable output current of the frequency converter.

[0378] 2. Switching frequency penalty term: To suppress the drawback of the FCS-MPC's non-fixed switching frequency and reduce inverter losses, the penalty term is the number of switching state changes.

[0379] ;

[0380] In the formula: Output value of the switching frequency penalty function; : The switching status of the three-phase bridge arms A, B, and C at the next moment (1 indicates the upper bridge arm is on, 0 indicates the lower bridge arm is on). : The current switching status of the three-phase bridge arms A, B, and C.

[0381] IV. Specific implementation of step S3:

[0382] The predictive fault-tolerant control algorithm with active compensation for zero-speed model error solves the prediction error problem caused by inaccurate model parameters (such as changes in resistance and inductance) and inverter nonlinearity (such as dead zone effect and tube voltage drop) when the motor is operating at zero speed or extremely low speed. The algorithm improves the robustness and control accuracy of MPC through active compensation mechanism.

[0383] At zero speed, the stator resistance voltage drop accounts for a very high proportion. Resistance temperature drift, inverter nonlinearity, and core saturation can lead to inaccurate prediction models. Traditional solutions use passive compensation based on parameter identification, which has three main drawbacks:

[0384] 1. Parameter identification at zero speed is slow to converge and prone to divergence, making it impossible to track parameter changes in real time;

[0385] 2. The compensation mechanism is disconnected from MPC rolling optimization, making it impossible to proactively avoid the risks of overcurrent and loss of synchronization caused by model errors;

[0386] 3. When the model error is too large, there is no fault-tolerant switching mechanism, which can easily lead to startup failure.

[0387] Advantages of the algorithm in this embodiment:

[0388] 1. By adopting an active correction factor for the prediction residual and directly embedding it into the current prediction model, the prediction model coefficients are corrected in real time according to the residual between the actual current and the predicted current in each control cycle. No separate parameter identification is required, which improves the response speed at zero speed.

[0389] 2. Integrate model residuals into the value function to achieve proactive fault tolerance control of model errors. When the model becomes inaccurate, the constraint terms are automatically strengthened to avoid overcurrent and loss of synchronization.

[0390] 3. Two-level fault-tolerant switching mechanism: When the model error exceeds the threshold, it automatically switches to the fault-tolerant prediction model to ensure that the zero-speed startup process is not interrupted.

[0391] Algorithm principle and calculation process:

[0392] Prediction residual calculation: For each control cycle, calculate the residual between the predicted current value and the actual sampled value at the previous moment, which serves as a quantitative indicator of model error.

[0393] —Equation (9);

[0394] In the formula: : The predicted residuals of the d-axis and q-axis currents at the current moment; : The actual sampled values ​​of the d-axis and q-axis currents at the current moment; : The d-axis and q-axis current values ​​predicted in the previous cycle at the current moment.

[0395] Active correction factor for residuals and reconstruction of prediction model: Based on the prediction residuals, an active correction factor for a first-order low-pass filter is designed and directly embedded into the current prediction model to compensate for model errors in real time. The reconstructed prediction model is as follows:

[0396] —Equation (10);

[0397] In the formula: The d-axis and q-axis residual correction coefficients are tuned according to the control cycle. : The predicted residuals of the d-axis and q-axis currents at the current moment; the meanings of the other physical quantities are consistent with those of equations (1) and (2).

[0398] This reconstructed model does not require identification of specific parameter drift amounts. It directly corrects the prediction results through residuals and can simultaneously compensate for all model errors caused by stator resistance temperature drift, inverter nonlinearity, and core saturation. It exhibits extremely strong robustness at zero speed.

[0399] The residual fault-tolerant value function is fused, which incorporates the magnitude of the model residuals into the value function to achieve active fault-tolerant control. A residual penalty term is added to Equation (8):

[0400] —Equation (11);

[0401] In the formula: : Weighting coefficient of residual penalty term; : Current d-axis and q-axis current prediction residuals at the current moment. When the model error increases, the residual penalty term weights automatically take effect, prioritizing voltage vectors with smaller model errors to avoid loss of synchronization and overcurrent caused by error accumulation.

[0402] The two-level fault-tolerant switching mechanism is the core safety protection logic of the algorithm. It is based on the real-time amplitude of the d / q axis current prediction residual to make a graded judgment, dividing it into two levels of protection: Level 1 fault tolerance (active compensation for minor errors) and Level 2 fault tolerance (model switching for major errors). It is executed synchronously with the MPC control cycle throughout the process to ensure that the model error can respond quickly and be fault-tolerant without impact, regardless of its size. It guarantees that the zero-speed start-up is uninterrupted, does not exceed current limits, and does not lose synchronization.

[0403] Fault tolerance criterion: The residual magnitude of the d / q-axis current prediction is used as the sole criterion. The formula for calculating the residual magnitude is as follows:

[0404] ;

[0405] In the formula: , Predict the residual for the d / q axis current in the current period; This is a quantitative indicator for model error.

[0406] Fault tolerance grading thresholds: Level 1 fault tolerance threshold: ( (Rated motor current); Secondary fault tolerance threshold: ( (This refers to the motor's rated current).

[0407] Level 1 fault tolerance, active error compensation mode:

[0408] Level 1 fault tolerance triggering conditions: The model has small errors (such as slight stator resistance temperature drift, small inverter nonlinearity error, and mild core saturation), and the current prediction residual amplitude is ≤5% of the rated current.

[0409] The specific actions to be taken for Level 1 fault tolerance:

[0410] 1. Prediction model maintenance: Without switching the basic prediction model, the residual active correction factor (Equation 10) is continuously enabled to correct the current prediction results in real time and passively offset small model errors.

[0411] 2. Enhanced Fault Tolerance of Value Function: Enable the fault-tolerant value function (Equation 11) that incorporates residual penalty terms, automatically increase the model error penalty weight, and prioritize the selection of voltage vectors with smaller prediction errors by the control logic to avoid error accumulation.

[0412] 3. Control logic remains unchanged: the dual-range adaptive weights are normally tuned, and the weights of current constraint, torque tracking, and switching loss are dynamically adjusted according to the original range logic to maintain zero-speed full torque output and basic current limiting capability.

[0413] 4. No extra actions: No position re-identification is triggered, and control parameters are not frozen, ensuring a continuous and smooth control process with no torque fluctuations.

[0414] Level 2 fault tolerance, forced switching mode for heavy error model:

[0415] The triggering conditions for Level 2 fault tolerance are: the model has a large error (such as severe resistance temperature drift, core deep saturation, inverter nonlinear distortion, or severe parameter drift), and the current prediction residual amplitude is greater than 5% of the rated current.

[0416] The specific actions to be taken for Level 2 fault tolerance:

[0417] 1. Forced switching of prediction model: Immediately stop the conventional adaptive prediction model and automatically switch to the fault-tolerant prediction model (i.e., the reconstructed prediction model with residual correction in the prediction fault-tolerant control algorithm with active compensation for zero-speed model error, Equation 10), shield all non-core adaptive terms, and retain only the core logic of residual correction to maximize the improvement of model robustness.

[0418] 2. Adaptive weight strategy freeze: Pause the dual-interval dynamic weight tuning of the dual-interval adaptive weight model predictive torque control algorithm, and switch the torque tracking weight and current constraint weight to fixed, highly robust, conservative values: significantly increase the current constraint weight to strengthen overcurrent hard protection; lock the torque tracking weight to ensure stable zero-speed torque output and prevent overshoot / loss of synchronization.

[0419] 3. Linkage parameter recalibration: Hardware triggers and software synchronously starts the fusion algorithm of salient pole enhanced non-swaying initial position identification and zero-speed flux linkage observation, and re-executes rotor initial position identification, saturated inductance correction and permanent magnet flux linkage observation to quickly repair the basic parameter errors of the prediction model.

[0420] 4. Output constraint hardware enhancement: The software tightens the current limit, and the hardware synchronously locks the minimum switching frequency of the PWM to prevent abnormal switching frequency, current surges, and torque ripples caused by model inaccuracies.

[0421] 5. Uninterrupted startup process: The entire process maintains a zero-speed torque closed loop, with no slippage of the potential energy load and no rotor oscillation, ensuring continuous execution of the startup process without interruption or restart.

[0422] Fault-tolerant recovery and uninterrupted smooth switchback:

[0423] This stage is the final stage and closed loop of the two-level fault-tolerant mechanism. The core objective is to smoothly switch back from the two-level fault-tolerant forced mode to the normal adaptive control mode of the predictive fault-tolerant control algorithm with dual-interval adaptive weighted model predictive torque control algorithm + zero-speed model error active compensation after the model error is eliminated and the system state is stable, without torque shock, speed fluctuation, or switching disturbance, ensuring continuous controllability throughout the zero-speed start-up process.

[0424] The fault-tolerant recovery judgment condition is that the recovery and switchback logic will only be triggered if all four of the following conditions are met, thus preventing frequent mode switching caused by repeated errors:

[0425] 1. Residual amplitude meets the standard: d / q-axis current prediction residual amplitude ( (Rated current of the motor);

[0426] 2. Stability meets the standard: The residual amplitude remains stable below the threshold for three consecutive control cycles, without rebound or sudden change;

[0427] 3. Parameter calibration completed: The fusion algorithm of salient pole enhanced non-swaying initial position identification and zero-speed flux linkage observation has re-completed the accurate calibration of rotor position, d / q axis saturated inductance, and permanent magnet flux linkage, and the basic parameters of the model have returned to normal;

[0428] 4. Normal operating status: The motor has no overcurrent, overload, or hardware failure, and is still in the zero-speed lock-in range or speed-up transition range. The startup process is not interrupted.

[0429] Uninterrupted, smooth, step-by-step execution process:

[0430] Pre-recovery preparation (1 control cycle): The system marks the recovery ready state and suspends the mandatory constraints of the secondary fault tolerance; the fault tolerance prediction model (Equation 10) is kept running, the residual correction factor is retained, and the prediction accuracy is maintained; the current motor torque, current, and speed are locked as the switchback reference values ​​to avoid sudden changes in the reference.

[0431] Adaptive weight smooth unfreezing (core disturbance-free process): Instead of directly restoring the dynamic weight, a linear weighted transition is adopted to prevent torque fluctuations caused by sudden weight changes: according to the current speed range (zero speed lock / acceleration transition), the adaptive weight target value in normal mode is read; every 1 control cycle, the frozen fixed conservative weight is linearly approximated to the target value, and the transition time is set to 5~10 control cycles;

[0432] During the transition period, the torque tracking weight, current constraint weight, and switching loss weight change synchronously and smoothly without any abrupt changes.

[0433] Unperturbed switching back of the prediction model: Gradually reduce the residual correction coefficient of the fault-tolerant prediction model. The fault-tolerant high value is smoothly reduced to the normal operating value; the normal mode discrete prediction model (Equation 1 and Equation 2) is activated simultaneously and calculated in parallel with the fault-tolerant model for 1 to 2 cycles; after confirming that the output error of the two models is <1% of the rated current, the fault-tolerant model is completely stopped and the normal prediction model is switched to.

[0434] Full recovery of control logic: The complete tuning logic of the dual-interval adaptive weight model predictive torque control algorithm is restored, and the weights are updated in real time according to torque error, current margin, and speed.

[0435] The prediction fault-tolerant control algorithm for restoring zero-speed model error active compensation is first-level fault-tolerant active compensation, retaining the value function of the residual penalty term (Equation 11), forming a long-term error protection;

[0436] Release the secondary fault-tolerant constraints such as PWM switching frequency lock and current limit tightening, and return to normal output rules.

[0437] Recovery complete confirmation: After monitoring for 5 consecutive control cycles, it is confirmed that the residual, torque, current, and speed are all stable within the normal range; the system exits the fault-tolerant mode, is marked as normal control state, and continues to execute the zero-speed start and smooth acceleration process; the residual peak value and duration of this fault-tolerant event are recorded for subsequent algorithm optimization and fault diagnosis.

[0438] Smooth Switchback Core Guarantee (No Shock Key): Linear Weight Transition: Rejects abrupt changes in weights, transitioning gradually and uniformly from conservative to target values, eliminating torque overshoot / drop; Parallel Model Transition: Dual-model parallel computation ensures continuous predicted output during switching, with no sudden state changes; Timing Synchronization Control: The switchback process is strictly synchronized with the PWM control cycle, with all parameter updates completed at the midpoint of the PWM cycle, avoiding switching noise; Gradual Constraint Removal: The mandatory protection of secondary fault tolerance is gradually released, not abruptly removed, ensuring a smooth system transition.

[0439] Long-term protection after recovery: After switching back to normal mode, the first-level fault tolerance active compensation continues to be effective: real-time monitoring of current prediction residuals, small errors are automatically compensated by residual correction factors; the fault tolerance value function is retained, and slight fluctuations in model error are suppressed in advance to avoid triggering the second-level fault tolerance again.

[0440] This embodiment achieves zero-speed start-up without any impact, swaying, or slippage. It relies on the collaborative efforts of three key technologies: a salient-pole enhancement-mode sway-free initial position identification algorithm and a zero-speed flux linkage observation fusion algorithm for initial position identification; a dual-interval adaptive weighted model predictive torque control algorithm; and a zero-speed model error active compensation predictive fault-tolerant control algorithm. The system executes seamlessly in four stages: power-on initialization → zero-speed torque closed-loop control → smooth acceleration → steady-state operation. The complete control logic is as follows:

[0441] During the power-on initialization and initial position identification phase:

[0442] Core actions: The frequency converter completes hardware initialization, current / voltage sampling calibration, and fault self-check; and synchronously starts the salient pole enhancement-type oscillation-free initial position identification and zero-speed flux linkage observation fusion algorithm.

[0443] Algorithm execution: Simultaneously completes accurate rotor position identification, disturbance-free magnetic pole polarity determination, and real-time correction of stator inductance / permanent magnet flux linkage parameters at zero speed. Identification and control are deeply integrated, with no separate identification stage and the rotor remains stationary without oscillation throughout the entire process.

[0444] Control objective: To establish an accurate MPC prediction mathematical model to provide a reliable rotor position and parameter basis for subsequent torque closed-loop control, and to eliminate the risk of slippage under potential energy load.

[0445] During the torque closed-loop establishment phase in the zero-speed locking range:

[0446] Core actions: The dual-interval adaptive weighted model predictive torque control algorithm is activated, and the system automatically enters the zero-speed lock-in interval (0~5% of rated speed); at the same time, the predictive fault-tolerant control algorithm with active compensation for zero-speed model error is put into operation throughout the process.

[0447] Algorithm execution: The torque is given by a ramp function that linearly increases from 0 to the target value to avoid starting shock; the dual-interval adaptive weight model predictive torque control algorithm adaptively tunes the weights according to the zero-speed interval logic to prioritize torque tracking and overcurrent constraints; the zero-speed model error active compensation predictive fault-tolerant control algorithm calculates the current prediction residual in real time and actively compensates for model errors such as stator resistance temperature drift and inverter nonlinearity to enhance fault-tolerant protection.

[0448] Control objective: To quickly establish rated full torque at zero speed, the potential energy load first outputs torque to balance the load torque, and then executes the speed-up command to achieve full torque at zero speed, without overcurrent or loss of synchronism.

[0449] During the smooth transition phase of the acceleration transition range:

[0450] Core action: When the motor speed increases to 5% of the rated speed, the dual-range adaptive weighted model predictive torque control algorithm automatically switches to the speed-up transition range (5%~20% of the rated speed) through a weighted smooth transition strategy.

[0451] Algorithm execution: Adaptively adjust the weighting coefficients of torque, flux linkage, and switching loss to balance dynamic response and operational stability; The predictive fault-tolerant control algorithm with zero-speed model error active compensation continuously performs model error compensation and two-level fault-tolerant monitoring to avoid torque pulsation and runaway risk during acceleration; Maintain MPC rolling optimization and traverse the optimal voltage vector output to achieve smooth acceleration.

[0452] Control objective: To achieve a smooth transition from zero speed to low speed without fluctuations, suppress torque pulsation, reduce switching losses, and ensure a smooth and stable acceleration process.

[0453] During steady-state operation across the full speed range:

[0454] Core action: Once the motor speed reaches the target speed, the system automatically switches to full-speed-range steady-state control logic.

[0455] Algorithm execution: Three original algorithms work together continuously. The dual-interval adaptive weight model predictive torque control algorithm maintains weight adaptation across all operating conditions. The salient pole enhanced non-swaying initial position identification and zero-speed flux observation fusion algorithm corrects position and parameters in real time. The zero-speed model error active compensation predictive fault-tolerant control algorithm ensures model accuracy and operational fault tolerance.

[0456] Control objective: To maintain high-precision torque and speed dual closed-loop control, ensure long-term stable operation of the motor, and balance control performance and operating efficiency.

[0457] One possible structure of the electronic device provided in this application includes: a processor, a memory, and a communication interface, which are interconnected and communicate with each other via a communication bus and / or other forms of connection mechanisms.

[0458] The memory includes one or more, which may be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The processor and other possible components can access the memory, reading and / or writing data therein.

[0459] A processor may be one or more, and can be an integrated circuit chip with signal processing capabilities. The aforementioned processors can be general-purpose processors, including Central Processing Units (CPUs), Microcontroller Units (MCUs), Network Processors (NPs), or other conventional processors; they can also be special-purpose processors, including Neural-network Processing Units (NPUs), Graphics Processing Units (GPUs), Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. Furthermore, when there are multiple processors, some may be general-purpose processors, while others may be special-purpose processors.

[0460] A communication interface includes one or more devices that can be used to communicate directly or indirectly with other devices to exchange data. The communication interface may include interfaces for wired and / or wireless communication.

[0461] One or more computer program instructions may be stored in the memory, and the processor may read and execute these computer program instructions to implement the methods provided in the embodiments of this application.

[0462] Understandably, electronic devices can include more or fewer components, or different structures, and these components can be implemented using hardware, software, or a combination thereof. Electronic devices can be physical devices, such as PCs, laptops, tablets, mobile phones, servers, and embedded devices, or they can be virtual devices, such as virtual machines and virtualized containers. Furthermore, electronic devices are not limited to single devices; they can also be combinations of multiple devices or clusters of numerous devices.

[0463] This application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of any of the methods described above.

[0464] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0465] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0466] Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0467] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0468] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A zero-speed start control method for a permanent magnet synchronous motor, characterized in that, include: Step S1: When the motor is in a zero-speed stationary state, inject a composite voltage signal containing DC bias component and high-frequency pulsation component into the d-axis direction of the rotor position estimation to forcibly enhance the salient pole effect of the motor, and extract the rotor position error in real time based on the high-frequency current response, and simultaneously complete the rotor initial electrical angle identification and magnetic pole polarity judgment. The identified rotor position electrical angle, saturation inductance, and permanent magnet flux linkage observations are updated to the prediction model in real time, in units of control cycles. Step S2: Based on the motor speed, divide the zero-speed start-up process into a zero-speed lock-up interval and a speed-up transition interval; In each MPC control cycle, the weight coefficients of each control objective in the value function are dynamically and adaptively calculated according to the interval and real-time operating conditions; the value function includes at least torque tracking term, flux tracking term, current constraint penalty term and switching frequency penalty term; Traverse all the basic voltage vectors of the inverter, calculate the predicted state of each voltage vector based on the updated prediction model, and calculate the cost value of each voltage vector in combination with the weight coefficients. Select the voltage vector with the smallest cost value as the optimal control output for the current cycle. Step S3: In each control cycle, calculate the prediction residual between the predicted current value and the actual sampled value at the previous moment, and compensate the prediction residual in real time into the current prediction model through the residual correction factor to correct the model parameter error; at the same time, according to the magnitude of the prediction residual, execute two-level fault-tolerant logic: when the residual magnitude is less than or equal to the set threshold, maintain the normal control mode and enhance the residual penalty term in the value function; when the residual magnitude is greater than the set threshold, switch to the fault-tolerant prediction model and freeze the adaptive weight strategy to maintain the startup process without interruption; Among them, the current prediction model is a specific execution model that embeds the prediction residual into the prediction model; when the residual amplitude is greater than a set threshold, the specific execution model switches to a fault-tolerant prediction model.

2. The method according to claim 1, characterized in that, Step S1 includes: The composite voltage signal is injected only into the d-axis of the rotor position estimation, and the q-axis injection voltage is always zero; the high-frequency pulsation component is a standard cosine pulsation waveform; the value range of the DC bias component is 0.5 to 1.5 times the rated voltage of the motor; the angular frequency of the high-frequency pulsation component is 1 / 10 to 1 / 5 of the inverter switching frequency.

3. The method according to claim 2, characterized in that, In step S1, the real-time extraction of rotor position error based on high-frequency current response, and the simultaneous completion of rotor initial electrical angle identification and magnetic pole polarity determination, include: Based on the high-frequency q-axis current component in the injected current response, the rotor position error signal is extracted, and the initial rotor electrical angle is obtained by observation through an adaptive phase-locked loop. Based on the DC component of the d-axis current caused by the DC bias component, the d-axis flux change rate is calculated online to determine the magnetic polarity of the permanent magnet. If the rate of change of the d-axis flux approaches zero, the current magnetic pole polarity is determined to be incorrect, and the initial rotor electrical angle is corrected by 180 degrees to obtain the rotor position electrical angle.

4. The method according to claim 3, characterized in that, In step S1, the process of updating the identified rotor position electrical angle, saturation inductance, and permanent magnet flux linkage observations to the prediction model in real time, using control cycles as the unit, includes: Update the rotor position electrical angle to the angle reference of Park transformation and inverse Park transformation; Update the d-axis inductance and q-axis inductance in the saturated inductance to the inductance parameters in the current prediction equation; The permanent magnet flux linkage observations are then updated to the flux linkage parameters in the torque prediction equation.

5. The method according to claim 4, characterized in that, In step S2, the zero-speed locking range is the range where the speed is less than or equal to 5% of the rated speed, and the speed-up transition range is the range where the speed is between 5% and 20% of the rated speed. The weight coefficients of each control objective in the dynamically adaptive calculation value function specifically include: In the zero-speed locking range, the weighting coefficients of the torque tracking term and the current constraint penalty term are increased, while the weighting coefficient of the switching frequency penalty term is decreased. During the acceleration transition range, the weighting coefficient of the torque tracking term is reduced as the rotational speed increases, while the weighting coefficient of the switching frequency penalty term is increased as the rotational speed increases.

6. The method according to claim 5, characterized in that, When the motor speed is in the transition zone between the zero-speed lock-in zone and the acceleration transition zone, the weight coefficients of each control objective in the dynamic adaptive calculation value function are linearly weighted, including: Calculate the speed ratio factor based on the current speed; Obtain the baseline weights for the zero-speed locking interval and the acceleration transition interval, respectively. The two benchmark weights are linearly weighted based on the aforementioned speed ratio factor to calculate the smoothing weight coefficient that is effective at the current moment.

7. The method according to claim 6, characterized in that, The two-level fault-tolerant logic includes: Level 1 fault tolerance: When the predicted residual magnitude When the value is less than or equal to a set threshold, a residual penalty term is enabled. The fault-tolerant value function is used, but the normal prediction model and adaptive weighting strategy are maintained; among them, The weighting coefficient for the residual penalty term; Level 2 fault tolerance: When the predicted residual magnitude When the value exceeds the set threshold, immediately switch to the fault-tolerant prediction model with residual correction factor, pause adaptive weight tuning, freeze the weights to the preset fixed conservative value, and trigger the initial position and parameter re-identification process in step S1. Fault-tolerant recovery: When the residual amplitude is continuously less than or equal to the set threshold and the system state is stable, the system smoothly switches back to the normal adaptive control mode without disturbance through linear transition of weights and parallel calculation of the model.

8. The method according to any one of claims 1-7, characterized in that, The complete control flow of the method is executed in stages, including: During the power-on initialization and initial position identification phase, step S1 is executed to establish an accurate prediction model; During the torque closed-loop establishment phase in the zero-speed locking interval, steps S2 and S3 are executed to establish a full torque closed loop at zero speed. During the smooth switching phase of the acceleration transition range, steps S2 and S3 are executed to achieve a smooth and uninterrupted acceleration from zero speed to low speed. During the steady-state operation phase in the full-speed domain, steps S1, S2, and S3 are continuously executed in coordination to maintain full-speed domain control.

9. An electronic device, characterized in that, include: A processor and a memory, the memory storing machine-readable instructions executable by the processor, which, when executed by the processor, perform the method as described in any one of claims 1-8.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1-8.

Images