A method for predicting current without deadbeat using gradient three parameters in a permanent magnet synchronous motor

Abstract

This invention discloses a method for gradient three-parameter identification and deadbeat-free current prediction of a permanent magnet synchronous motor. The method includes: reconstructing the stator voltage and current equations of the motor based on the initial estimated motor parameters within the controller; and... d A periodic high-frequency excitation signal is injected onto the shaft current reference value to construct parameters that identify the required continuous excitation conditions; based on d shaft and q To address the current prediction error of the motor shaft, a parameter identification model based on inductance, resistance, and flux linkage is established. Based on this model, the gradient descent method is used to update the parameter estimates in the controller. Lyapunov stability analysis is used to determine the range of convergence step sizes, ensuring global convergence and system stability during the gradient iteration process. This invention effectively solves the problem of motor parameter disturbances, significantly improving the stability, dynamic response, and robustness of the motor control system without adding other complex hardware.

Description

Technical Field

[0001] This invention relates to the field of deep-sea motor control technology, and more specifically, to a method for predicting current without deadbeat using gradient three-parameter identification of permanent magnet synchronous motors. Background Technology

[0002] In recent years, with the rapid development of power electronics technology, permanent magnet synchronous motors have been widely used in many cutting-edge fields such as electric vehicles, aerospace, wind power generation, and deep-sea robots due to their high efficiency, high torque density, high reliability, and excellent dynamic performance. However, in actual operation, the accurate estimation of important parameters such as resistance, inductance, and flux linkage is highly dependent. Once a deviation occurs, it will lead to a rapid decline in system stability and dynamic response performance.

[0003] For example, in deep-sea robotic applications, laying submarine cables requires motor-driven robotic arms to withstand the extreme and harsh conditions of the deep-sea environment, such as heavy loads, high voltage, wide temperature differences, and biological impacts. This necessitates that the motor possess the ability to overcome parameter uncertainties. If the controller's preset motor parameters do not match the actual values, the motor must be able to self-correct in real time, ensuring that the current loop and speed loop function properly. Therefore, developing a precise, rapid, and real-time updated motor parameter identification method is crucial for ensuring the high-performance and high-precision operation of permanent magnet synchronous motors under complex working conditions.

[0004] Common online parameter identification methods include particle swarm optimization (PSO), which can identify multiple parameters simultaneously. However, its convergence speed is limited by the iteration speed, and it requires additional conditions to address the underrank problem of the equations. Other methods include multi-step recursive algorithms and other full-parameter identification methods. These methods can obtain more comprehensive parameter information and effectively solve the underrank problem, but they require the injection of high-frequency excitation signals into the system. Since the signal injection time may extend the identification time, it may lead to a "dead zone" phenomenon, thus interfering with the normal operation of the system. Existing parameter identification technologies struggle to achieve an ideal balance between fast matching, high accuracy, simultaneous identification of multiple parameters, and minimal system disturbance, thus limiting the application scenarios of deadbeat predictive current control.

[0005] Therefore, a new solution is needed to address the above problems. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a deadbeat predictive current method for gradient three-parameter identification of permanent magnet synchronous motors. By periodically injecting high-frequency excitation signals, the three parameters of inductance, resistance, and flux linkage in the controller are rapidly converged to near accurate values ​​through gradient step-size iteration. Stability analysis is performed on the iterative process to avoid local convergence problems. This method significantly reduces the impact of parameter mismatch without the need for additional hardware support, while ensuring deadbeat current control and exhibiting strong robustness.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A method for predicting current without deadbeat using gradient three-parameter identification in a permanent magnet synchronous motor includes the following steps:

[0009] Step S1: Reconstruct the motor stator voltage and current equations based on the initial motor parameter estimates in the controller;

[0010] Step S2: Inject a periodic high-frequency excitation signal onto the d-axis current reference value to construct parameters to identify the required continuous excitation conditions.

[0011] Step S3: Based on the current prediction error along the d-axis, establish an identification model for the inductor parameters, and update the estimated inductor value in the controller using the gradient descent method according to the model.

[0012] Step S4: Based on the current prediction error along the d-axis, establish an identification model for the resistance parameters, and update the estimated resistance values ​​in the controller using the gradient descent method according to the model.

[0013] Step S5: Based on the current prediction error of the q-axis, establish an identification model for the flux linkage parameters, and update the flux linkage estimate in the controller using the gradient descent method according to the model.

[0014] Step S6: Determine the range of convergence step size based on Lyapunov stability analysis to ensure global convergence of the gradient iteration process and system stability.

[0015] Furthermore, in step S1, the reconstruction process of the stator voltage and current equations is as follows:

[0016] Step S11, for The dq-axis stator voltage at time t is calculated.

[0017] Step S12, calculate the current by discretizing the motor current equation. The dq-axis stator current at time t;

[0018] Step S13: Based on the analysis of the impact of parameter mismatch, list the equations for the controller's dq-axis output voltage;

[0019] Step S14: Based on the assumption that the controller output voltage is equal to the actual applied voltage, the dq-axis current prediction model is obtained.

[0020] Furthermore, the reconstruction process of the d-axis stator voltage and current equations is as follows:

[0021] Step 1, based on the motor in The dq-axis stator current, rotor electric angular velocity, stator resistance, dq-axis stator inductance, and sampling period at each moment, combined with the corresponding d-axis reference current, are used to... The d-axis stator voltage at time t is calculated.

[0022] Step 2, based on the motor in The dq-axis stator current, rotor electric angular velocity, sampling period, and estimated dq-axis stator inductance and stator resistance at time t are used to deduce the discretized motor current equations. The d-axis stator current at time t;

[0023] Step 3: Given that there may be discrepancies between the actual and estimated values ​​of the stator resistance and dq-axis stator inductance, list the equation for the controller's d-axis output voltage.

[0024] Step 4: Establish the relationship between the controller's d-axis output voltage and d-axis stator current to obtain the d-axis current prediction model.

[0025] Furthermore, the reconstruction process of the q-axis stator voltage and current equations is as follows:

[0026] Step 1, based on the motor in The dq-axis stator current, rotor electric angular velocity, stator resistance, dq-axis stator inductance, flux linkage, and sampling period at each moment, combined with the corresponding q-axis reference current, are used to... The q-axis stator voltage at time t is calculated.

[0027] Step 2, based on the motor in The dq-axis stator current, rotor electric angular velocity, sampling period, and estimated dq-axis stator inductance, stator resistance, and flux linkage at time t are used to deduce the discretized motor current equations. The q-axis stator current at time t;

[0028] Step 3: Given that there may be discrepancies between the actual and estimated values ​​of the stator resistance, dq-axis stator inductance, and flux linkage, list the equation for the controller's q-axis output voltage.

[0029] Step 4: Establish the relationship between the controller's q-axis output voltage and the q-axis stator current to obtain the q-axis current prediction model.

[0030] Furthermore, in step S2, the high-frequency excitation signal is a periodic square wave pulse sequence. The injection frequency band of this signal can be distinguished from the normal operation signal of the motor and can avoid confusion with switching noise. In terms of timing, the actual period is... Intermittent pulse trains, each pulse group containing 8 sampling cycles of effective excitation.

[0031] Furthermore, in step S3, the inductor parameter identification model is implemented in the following way:

[0032] Step S31: Based on the high-frequency excitation signal injected into the d-axis, and using the d-axis as the identification reference system, establish a model of the relationship between the d-axis current and the inductance.

[0033] Step S32: For the surface-mounted permanent magnet synchronous motor, make the stator inductance values ​​of the d-axis and q-axis the same, and simplify the current equation into a form that includes a single inductance parameter.

[0034] Step S33: Construct an inductance error objective function based on the current prediction error. This objective function is based on the d-axis reference current and... The error value between the d-axis stator currents at time t is constructed;

[0035] Step S34: Apply the chain rule to obtain the first-order partial derivative of the identified inductance value, obtain the gradient term and sensitivity term of the corresponding inductance parameter, and confirm the update direction of the inductance estimate.

[0036] Step S35: According to the gradient descent principle, update the inductance parameters based on the current error of the inductance estimate and the iteration step size, so that it gradually converges to the true inductance value in repeated iterations.

[0037] Furthermore, the inductance estimate The update direction is:

[0038] When the inductance estimate Stator inductance value hour, Actual d-axis current at time 1 Less than the d-axis reference current If the gradient is negative, it needs to be increased by increasing the iteration step size. ;

[0039] When the inductance estimate Stator inductance value hour, Actual d-axis current at time 1 Greater than the d-axis reference current If the gradient is positive, it needs to be reduced by adjusting the iteration step size. .

[0040] Furthermore, in step S4, the resistance parameter identification model is implemented in the following way:

[0041] Step S41, based on Once the error identification of the inductor parameters has been completed, the updated inductor is substituted into the resistance identification model. Using the current prediction error in the d-axis direction and the high-frequency excitation signal, a functional relationship between resistance and current is constructed.

[0042] Step S42: Construct a resistance error objective function based on the current prediction error. This objective function is based on the d-axis reference current and... The error value between the d-axis stator currents at time t is constructed;

[0043] Step S43: Apply the chain rule to obtain the first-order partial derivative of the identified resistance value, obtain the gradient term and sensitivity term of the corresponding resistance parameter, and confirm the direction of the resistance estimate update.

[0044] Step S44: According to the gradient descent principle, update the resistance parameters based on the current error of the resistance estimate and the iteration step size, so that they gradually converge to the true resistance value in repeated iterations.

[0045] Furthermore, in step S5, the flux linkage parameter identification model is implemented in the following way:

[0046] Step S51, without considering the influence of the d-axis current, establish a system containing... Error model of q-axis stator current, flux linkage, sampling period, stator inductance, flux linkage estimate, q-axis reference current and rotor electric angular velocity at time t;

[0047] Step S52, based on the q-axis reference current and The objective function is constructed based on the error between the q-axis stator currents at time t.

[0048] Step S53: Apply the chain rule to obtain the first-order partial derivative of the identified magnetic flux value, obtain the gradient term and sensitivity term of the corresponding magnetic flux parameter, and confirm the update direction of the magnetic flux estimate.

[0049] Step S54: According to the gradient descent principle, update the flux linkage parameters based on the current error of the flux linkage estimate and the iteration step size, so that they gradually converge to the true flux linkage value in repeated iterations.

[0050] Furthermore, in step S6, the range of convergence step size is determined based on Lyapunov stability analysis, which is achieved in the following way:

[0051] Step 61: Define the error terms for the three parameters: inductance, resistance, and flux linkage.

[0052] Step S62: Select a Lyapunov function with the error of each parameter as the variable to reflect the magnitude of the overall parameter error;

[0053] Step S63: Analyze the discrete changes of the function to derive sufficient conditions for its monotonically decreasing, thereby determining the range of values ​​that the iteration step size of each of the inductor, resistor and flux linkage should satisfy. By constraining the step size, the gradient update process remains stable.

[0054] The beneficial effects of this invention are:

[0055] 1. The parameter identification method in this invention can realize online identification of three important parameters: resistance, inductance and flux linkage, and perform precise parameter self-compensation to correct parameter mismatches.

[0056] 2. This invention, by introducing a high-frequency square wave injection mechanism and gradient correction features, can maintain the algorithm's sensitivity when disturbances occur. Simultaneously, the gradient module automatically performs an error iteration process based on the error signal direction, giving the identification process self-recovery capabilities, and the current waveform is almost distortion-free.

[0057] 3. This invention employs a step size constraint based on Lyapunov functions. In each sampling period, the error function decreases monotonically, avoiding oscillations and overshoot. At the same control frequency, the convergence time of the three parameters in this method is significantly shorter than that of traditional methods.

[0058] 4. The parameter identification method in this invention maintains the fast dynamic characteristics of deadbeat predictive current control. In current step tests, the response time and overshoot of this method are superior to those of traditional methods.

[0059] 5. The method of the present invention is not only applicable to deep-sea robot electric drive systems, but can also be extended to electric vehicles, industrial servo systems, and wind power generation. Attached Figure Description

[0060] Figure 1 This is a control flowchart of a method for identifying deadbeat-free predictive current of a permanent magnet synchronous motor gradient three-parameter identification in this embodiment;

[0061] Figure 2 This is a schematic diagram illustrating one characteristic of high-frequency excitation signal injection in this embodiment;

[0062] Figure 3 This is a schematic diagram illustrating a step-by-step parameter identification feature in this embodiment;

[0063] Figure 4a This is a waveform diagram of the traditional deadbeat predictive current control method in this embodiment under the q-axis current step response;

[0064] Figure 4b This is a waveform diagram of the q-axis current step response of the extended observer-based deadbeat predictive current control method in this embodiment.

[0065] Figure 4c This is a waveform diagram of the deadbeat prediction current method for gradient three-parameter identification of permanent magnet synchronous motor in this embodiment under the q-axis current step response;

[0066] Figure 5 This is a convergence curve of inductance, resistance and flux linkage for the deadbeat prediction current method of gradient three-parameter identification of permanent magnet synchronous motor in this embodiment when the current jumps.

[0067] Figure 6a This is a waveform diagram of the traditional deadbeat predictive current control method under steady-state conditions in this embodiment;

[0068] Figure 6b This is a waveform diagram of the deadbeat predictive current control method based on the extended observer in this embodiment under steady-state conditions;

[0069] Figure 6c This is a waveform diagram of the deadbeat prediction current method for gradient three-parameter identification of permanent magnet synchronous motor in this embodiment under steady-state conditions;

[0070] Figure 7a This is a phase current FFT diagram of a traditional deadbeat predictive current control method in this embodiment;

[0071] Figure 7b This is a phase current FFT diagram based on the extended observer deadbeat predictive current control method in this embodiment;

[0072] Figure 7c This is a phase current FFT diagram of the deadbeat prediction current method for gradient three-parameter identification of permanent magnet synchronous motor in this embodiment;

[0073] Figure 8a This is a waveform diagram of the traditional deadbeat predictive current control method in this embodiment when the inductance, resistance and flux linkage parameters jump;

[0074] Figure 8b This is a waveform diagram of the deadbeat predictive current control method based on the extended observer in this embodiment when the inductance, resistance and flux linkage parameters are stepped.

[0075] Figure 8c This is a waveform diagram of the deadbeat prediction current method for gradient three-parameter identification of permanent magnet synchronous motor in this embodiment when the inductance, resistance and flux linkage parameters change stepwise. Detailed Implementation

[0076] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0077] Example: A deadbeat-free predictive current system for gradient three-parameter identification of permanent magnet synchronous motors, such as... Figure 1 As shown, the motor outputs three-phase current when running. , , With mechanical angular velocity , The rotor electric angular frequency was obtained through pole logarithmic conversion and differentiation. The three-phase current is first transformed into a stationary coordinate system current through the abc→αβ transformation. , , and then combine The d / q axis feedback current in the rotating coordinate system is obtained through the αβ→dq transformation. , These currents and , The common input is the "gradient-based beat prediction current control module", which calculates and outputs d / q-axis voltage commands. , Subsequently, a voltage command was issued. , The voltage in the stationary coordinate system is obtained by the inverse transformation from dq to αβ. , The input space vector pulse modulation module generates the inverter switching signal. Ultimately, the inverter operates based on the switching signal. DC bus voltage The voltage is inverted to a three-phase AC voltage, which drives the motor to run according to the target, completing the closed-loop control cycle.

[0078] A method for gradient three-parameter identification and deadbeat predictive current for permanent magnet synchronous motors is proposed, addressing the problem of decreased system stability and control accuracy caused by motor parameter mismatch in traditional deadbeat predictive current control. First, a periodic high-frequency excitation signal is injected into the d-axis current to obtain rich parameter identification information. Second, based on the error between the reference current and the actual current, methods are established for the d- and q-axis inductances respectively. , ,resistance Magnetic Link The method first identifies the parameters; then, combining the gradient descent method, iterative steps are used to converge the internal parameter values ​​of the controller to near the actual values ​​of the motor; finally, the real-time updated accurate parameters are used for deadbeat predictive current control to ensure that the control current accurately tracks the reference current value after two beats. This method is used to suppress parameter disturbances, improve the dynamic response capability of the current, and enhance the overall robustness of the motor system. Figure 1 As shown, it includes the following steps:

[0079] Step S1: Reconstruct the motor stator voltage and current equations based on the initial motor parameter estimates within the controller. Specifically:

[0080] Step S11, in At any given time, calculate the control voltage that needs to be applied. and :

[0081]

[0082]

[0083] In the formula, , for Stator voltage along the dq axis at time t; , for The dq-axis stator current at time t; Stator resistance; , For the dq axis stator inductance; The rotor's electric angular velocity; For magnetic linkage; The sampling period; , This is the reference current for the dq axis;

[0084] Step S12, calculate in Predicted current value at time:

[0085]

[0086]

[0087] In the formula, , for The dq-axis stator current at time t;

[0088] Step S13: To analyze the impact of parameter mismatch, the controller control voltage equation is listed:

[0089]

[0090]

[0091] In the formula, and These are the output voltages of the controller's d-axis and q-axis, respectively. This is an estimated value for the controller resistance. This is the estimated flux linkage value for the controller; This is the estimated value of the controller inductance;

[0092] Step S14: To establish the relationship between the controller voltage and the control current, assume that the actual applied voltage is equal to the voltage calculated by the controller. At this point, the d-axis current equation is:

[0093]

[0094] The equation for the q-axis current is:

[0095]

[0096] Step S2: Inject a periodic high-frequency excitation signal onto the d-axis current reference value to construct parameters to identify the required continuous excitation conditions.

[0097] like Figure 2 As shown, the high-frequency excitation signal is a periodic square wave pulse sequence. The injection frequency band of this signal can be distinguished from the normal operation signal of the motor and can avoid confusion with switching noise. In terms of timing, the actual period is... Intermittent pulse trains, each pulse group containing 8 sampling cycles of effective excitation.

[0098] like Figure 3 As shown, when the actual value of any parameter Controller estimate greater than any parameter At that time, The actual current value at any given moment will oscillate significantly, making it impossible to effectively track its corresponding reference current value. ; and when the actual value of any parameter Controller estimate less than any parameter At that time, it was The actual current value at any given moment will exhibit a significant response delay, resulting in slow current tracking; when the actual value of any parameter... equal to the controller estimate with arbitrary parameters At that time, its The actual current value at any given time will be precisely tracked. Reference current value at time This achieves optimal dynamic response performance of the current.

[0099] Based on the current prediction errors along the d-axis and q-axis, a relevant cost function is defined, and a parameter identification model based on inductance, resistance, and flux linkage is established. The gradient descent method is then used to update the parameter estimates in the controller based on this model. Finally, Lyapunov stability analysis is used to determine the step-size iterative convergence conditions for the three parameters. The specific method is as follows:

[0100] Step S3: Based on the current prediction error along the d-axis, establish an identification model for the inductance parameters, and update the estimated inductance value in the controller using the gradient descent method according to the model.

[0101] The inductor parameter identification model is implemented in the following way:

[0102] In step S31, when identifying the inductor, since a high-frequency excitation signal is injected into the d-axis, the d-axis is used as the identification reference system. The current equation is:

[0103]

[0104] In the formula, for The d-axis stator current at time t; The sampling period; , For the dq axis stator inductance; The rotor's electric angular velocity; This is the estimated value of the controller inductance; for The q-axis stator current at time t; This is the d-axis reference current;

[0105] Step S32: Select a surface-mounted permanent magnet synchronous motor. Since the inductance values ​​of the d-axis and q-axis are the same, they are unified as L. Furthermore, to highlight the influence of inductance error on current error and for parameter identification, the equation can be simplified as follows:

[0106]

[0107] Step S33, define the objective function for inductor error. :

[0108]

[0109] Step S34: Apply the chain rule to calculate the first-order partial derivative of the identified inductance value to obtain the gradient term of the corresponding inductance:

[0110]

[0111] In the formula, This is the sensitivity term, which directly determines the direction of parameter updates and affects the inductance estimate. renew:

[0112] when At that time, the actual current Less than the reference current If the gradient is negative, it needs to be increased by increasing the iteration step size. ;

[0113] when At that time, the actual current Greater than the reference current If the gradient is positive, it needs to be reduced by adjusting the iteration step size. .

[0114] Step S35, for the inductor, its adaptive law is:

[0115]

[0116] In the formula, This is the new inductance value after one iteration; This is the current inductance value; This represents the inductance iteration step size.

[0117] Step S4: After the inductor parameter identification converges, an identification model for the resistance parameter is established based on the current prediction error along the d-axis, and the estimated resistance value in the controller is updated using the gradient descent method according to the model.

[0118] The resistance parameter identification model is implemented in the following way:

[0119] Step S41, because At this point, the error identification of the inductor parameters has been completed, and the inductance L in the resistance equation is now an accurate estimate. Since the same high-frequency excitation signal is injected into the d-axis, the current equation is forwarded by one cycle, and after simplification, we get:

[0120]

[0121] In the formula, This is an estimated value for the controller resistance. , This is the reference current for the dq axis; for The d-axis stator current at time t; The sampling period; This refers to the stator inductance value;

[0122] Step S42, define the objective function for resistance error. :

[0123]

[0124] Step S43: Apply the chain rule to calculate the first-order partial derivative of the identification resistor value to obtain the gradient term and sensitivity term of the corresponding resistor:

[0125]

[0126] In the formula, This is the sensitivity term; the resistance value is determined based on the sensitivity term. renew:

[0127] when At that time, the actual current Less than the reference current If the gradient is negative, it needs to be increased by increasing the iteration step size. ;

[0128] when At that time, the actual current Greater than the reference current If the gradient is positive, it needs to be reduced by adjusting the iteration step size. ;

[0129] Step S44, for the resistor, its adaptive law is:

[0130]

[0131] In the formula, This is the new resistance value after one iteration; This is the current resistance value; This represents the resistance iteration step size.

[0132] Step S5: Based on the current prediction error of the q-axis, establish an identification model for the flux linkage parameters, and update the flux linkage estimate in the controller using the gradient descent method according to the model.

[0133] The flux linkage parameter identification model is implemented in the following way:

[0134] Step S51: Since the flux linkage error only affects the q-axis, the influence of the d-axis current is ignored. The flux linkage parameters are identified starting from time k, and the error parameter model is as follows:

[0135]

[0136] In the formula, for The q-axis stator current at time t; The sampling period; The rotor's electric angular velocity; This is the estimated flux linkage value for the controller; For magnetic linkage; This is the q-axis reference current; This refers to the stator inductance value;

[0137] Step S52, define the objective function for flux linkage error. :

[0138]

[0139] Step S53: Apply the chain rule to obtain the first-order partial derivative of the identified flux linkage value to obtain the gradient term and sensitivity term of the corresponding flux linkage:

[0140]

[0141] In the formula, This is the sensitivity term; the flux linkage is estimated based on the sensitivity term. renew:

[0142] when At that time, the actual current Less than the reference current If the gradient is negative, it needs to be increased by increasing the iteration step size. ;

[0143] when At that time, the actual current Greater than the reference current If the gradient is positive, it needs to be reduced by adjusting the iteration step size. ;

[0144] Step S54, for the magnetic flux linkage, its adaptive law is:

[0145]

[0146] In the formula, This represents the new flux linkage value after one iteration. This is the current flux linkage value; This represents the magnetic flux linkage iteration step size.

[0147] Step S6: Determine the range of convergence step size based on Lyapunov stability analysis to ensure global convergence of the gradient iteration process and system stability.

[0148] The range of convergence step size values ​​is determined based on Lyapunov stability analysis, and is achieved in the following way:

[0149] Step S61, define the parameter error terms respectively:

[0150]

[0151]

[0152]

[0153] In the formula, for Estimated inductance value for the timing controller; for Estimated value of controller resistance at any given time; for The estimated flux linkage value for the timing controller; This refers to the stator inductance value; Stator resistance; For magnetic linkage; , , These are the parameter errors for inductance, resistance, and magnetic flux, respectively.

[0154] Step S62, select function :

[0155]

[0156] Step S63, by analyzing the difference function The stability condition is derived as follows:

[0157]

[0158]

[0159]

[0160] In the formula, , , These are the iteration step sizes for inductance, resistance, and flux linkage, respectively. This is the d-axis reference current; The sampling period; The rotor's electric angular velocity; for The q-axis stator current at time t; This refers to the stator inductance value; for The d-axis stator current at time t.

[0161] The effectiveness of the method in this embodiment is verified below:

[0162] Figure 4a , Figure 4b and Figure 4c The figures show a waveform comparison of the traditional deadbeat predictive current control method, the deadbeat predictive current control method based on extended observer, and the deadbeat predictive current control method based on gradient characteristics in this embodiment, under the step response of the q-axis current. The experimental conditions were as follows: the motor was running at a constant speed of n=300rpm, the reference value of the q-axis current jumped from 0A to 0.2A, while the d-axis current remained stable.

[0163] like Figure 4a , Figure 4b and Figure 4c As shown, the first row represents the case where all parameters are matched, and all three methods achieve stable deadbeat current tracking. The second and third rows represent the cases where the controller parameters are less than and greater than the actual parameters, respectively. The traditional deadbeat predictive current control method exhibits a significant DC bias, while the deadbeat predictive current control method based on the extended observer can offset some of the DC bias, but the convergence speed of the current sampling value is still relatively slow. In contrast, the method in this embodiment maintains high dynamic performance of the current throughout.

[0164] Figure 5 This figure shows the convergence curves of the three parameters of the gradient-based deadbeat predictive current control method in this embodiment during a current step. The figure clearly shows that all three parameters converge from an uncertain initial value through repeated step-size iterations, ultimately achieving precise control of the mismatched parameters. Combined with… Figure 4a , Figure 4b and Figure 4c This verifies that it can achieve precise parameter control while still maintaining high-performance deadbeat tracking.

[0165] Figure 6a , Figure 6b and Figure 6c The figures show waveform comparisons under steady-state conditions for the traditional deadbeat predictive current control method, the deadbeat predictive current control method based on extended observer, and the deadbeat predictive current control method based on gradient characteristics in this embodiment. The experimental condition is: dq-axis current waveforms under constant speed of n=100rpm, where the reference value is... =0A, =0.2A.

[0166] like Figure 6a , Figure 6b and Figure 6c As shown, the controller gain underwent two subsequent switching operations. Initially, the controller gain was in a parameter-matched state, at which point the dq-axis currents of all three methods were stable. As the controller gain instantaneously switched to a value less than the actual parameter value, the q-axis current of the traditional deadbeat predictive current control method instantly dropped to 0A, losing normal control. After running for a period of time, the controller gain switched to a value greater than the actual parameter value. Both the traditional deadbeat predictive current control method and the deadbeat predictive current control method based on the extended observer exhibited severe current oscillations. However, the method in this embodiment only experienced a momentary oscillation at the moment of switching, and then maintained a stable and steady dq-axis current operation.

[0167] Figure 7a , Figure 7b and Figure 7cThe image shows a comparison of phase current FFT values ​​for the traditional deadbeat predictive current control method, the deadbeat predictive current control method based on extended observer, and the deadbeat predictive current control method based on gradient characteristics in this embodiment. Figure 7a , Figure 7b and Figure 7c As shown, both the traditional deadbeat predictive current control method and the deadbeat predictive current control method based on extended observer exhibit THD values ​​that are much higher than those when parameters are matched when the controller gain switching causes parameter mismatch, rising to a maximum of 33.2% and 46.96% respectively. However, the steady-state THD value of the phase current in this embodiment remains at around 8%, which can maintain the sensitivity of the algorithm when disturbances occur.

[0168] Figure 8a , Figure 8b and Figure 8c The diagrams show a waveform comparison of the traditional deadbeat predictive current control method, the deadbeat predictive current control method based on extended observer, and the deadbeat predictive current control method based on gradient characteristics in this embodiment, under parameter step conditions. The experiment is conducted under constant speed operating condition n=350rpm. =0.2A is used as a reference value to generate a small square wave signal with an amplitude of 0.2A and a period of about 0.1s. The controller gain switches rapidly at t=0.4. At this time, the controller inductance and flux linkage parameters step to 1 / 3 of their original values, while the controller resistance parameter step to 3 times its original value.

[0169] Both traditional deadbeat predictive current control methods and deadbeat predictive current control methods based on extended observers exhibit current instability and significant oscillations during parameter step changes. However, in the method of this embodiment... The sampled current value exhibits momentary jitter during switching, but as the three parameters rapidly converge to near their actual values, the jitter disappears while the sampled current maintains fast and accurate tracking, demonstrating the robustness of this method.

[0170] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting current without deadbeat using gradient three-parameter identification in a permanent magnet synchronous motor, characterized in that, Includes the following steps: Step S1: Reconstruct the motor stator voltage and current equations based on the initial motor parameter estimates in the controller; Step S2: Inject a periodic high-frequency excitation signal onto the d-axis current reference value to construct parameters to identify the required continuous excitation conditions. Step S3: Based on the current prediction error along the d-axis, establish an identification model for the inductor parameters, and update the estimated inductor value in the controller using the gradient descent method according to the model. Step S4: Based on the current prediction error along the d-axis, establish an identification model for the resistance parameters, and update the estimated resistance values ​​in the controller using the gradient descent method according to the model. Step S5: Based on the current prediction error of the q-axis, establish an identification model for the flux linkage parameters, and update the flux linkage estimate in the controller using the gradient descent method according to the model. Step S6: Determine the range of convergence step size based on Lyapunov stability analysis to ensure global convergence of the gradient iteration process and system stability; In step S3, the inductor parameter identification model is implemented in the following way: Step S31: Based on the high-frequency excitation signal injected into the d-axis, and using the d-axis as the identification reference system, establish a model of the relationship between the d-axis current and the inductance. Step S32: For the surface-mounted permanent magnet synchronous motor, make the stator inductance values ​​of the d-axis and q-axis the same, and simplify the current equation into a form that includes a single inductance parameter. Step S33: Construct an inductance error objective function based on the current prediction error. This objective function is based on the d-axis reference current and... The error value between the d-axis stator currents at time t is constructed; Step S34: Apply the chain rule to obtain the first-order partial derivative of the identified inductance value, obtain the gradient term and sensitivity term of the corresponding inductance parameter, and confirm the update direction of the inductance estimate. Step S35: According to the gradient descent principle, update the inductance parameters based on the current error of the inductance estimate and the iteration step size, so that it gradually converges to the true inductance value in repeated iterations. Inductance estimate The update direction is: When the inductance estimate Stator inductance value hour, Actual d-axis current at time 1 Less than the d-axis reference current If the gradient is negative, it needs to be increased by increasing the iteration step size. ; When the inductance estimate Stator inductance value hour, Actual d-axis current at time 1 Greater than the d-axis reference current If the gradient is positive, it needs to be reduced by adjusting the iteration step size. .

2. The method for predicting current without delay by identifying gradient three parameters of a permanent magnet synchronous motor according to claim 1, characterized in that, In step S1, the reconstruction process of the stator voltage and current equations is as follows: Step S11, for The dq-axis stator voltage at time t is calculated. Step S12, calculate the current by discretizing the motor current equation. The dq-axis stator current at time t; Step S13: Based on the analysis of the impact of parameter mismatch, list the equations for the controller's dq-axis output voltage; Step S14: Based on the assumption that the controller output voltage is equal to the actual applied voltage, the dq-axis current prediction model is obtained.

3. The method for predicting current without delay by identifying gradient three parameters of a permanent magnet synchronous motor according to claim 2, characterized in that, The reconstruction process of the d-axis stator voltage and current equations is as follows: Step 1, based on the motor in The dq-axis stator current, rotor electric angular velocity, stator resistance, dq-axis stator inductance, and sampling period at each moment, combined with the corresponding d-axis reference current, are used to... The d-axis stator voltage at time t is calculated. Step 2, based on the motor in The dq-axis stator current, rotor electric angular velocity, sampling period, and estimated dq-axis stator inductance and stator resistance at time t are used to deduce the discretized motor current equations. The d-axis stator current at time t; Step 3: Given that there may be discrepancies between the actual and estimated values ​​of the stator resistance and dq-axis stator inductance, list the equation for the controller's d-axis output voltage. Step 4: Establish the relationship between the controller's d-axis output voltage and d-axis stator current to obtain the d-axis current prediction model.

4. The method for predicting current without delay by identifying gradient three parameters of a permanent magnet synchronous motor according to claim 2, characterized in that, The reconstruction process of the q-axis stator voltage and current equations is as follows: Step 1, based on the motor in The dq-axis stator current, rotor electric angular velocity, stator resistance, dq-axis stator inductance, flux linkage, and sampling period at each moment, combined with the corresponding q-axis reference current, are used to... The q-axis stator voltage at time t is calculated. Step 2, based on the motor in The dq-axis stator current, rotor electric angular velocity, sampling period, and estimated dq-axis stator inductance, stator resistance, and flux linkage at time t are used to deduce the discretized motor current equations. The q-axis stator current at time t; Step 3: Given that there may be discrepancies between the actual and estimated values ​​of the stator resistance, dq-axis stator inductance, and flux linkage, list the equation for the controller's q-axis output voltage. Step 4: Establish the relationship between the controller's q-axis output voltage and the q-axis stator current to obtain the q-axis current prediction model.

5. The method for predicting current without delay by identifying gradient three parameters of a permanent magnet synchronous motor according to claim 1, characterized in that, In step S2, the high-frequency excitation signal is a periodic square wave pulse sequence. The injection frequency band of this signal can be distinguished from the normal operation signal of the motor and can avoid confusion with switching noise. The actual period formed in the timing sequence is... Intermittent pulse trains, each pulse group containing 8 sampling cycles of effective excitation.

6. The method for predicting current without delay by identifying gradient three parameters of a permanent magnet synchronous motor according to claim 1, characterized in that, In step S4, the resistance parameter identification model is implemented in the following way: Step S41, based on Once the error identification of the inductor parameters has been completed, the updated inductor is substituted into the resistance identification model. Using the current prediction error in the d-axis direction and the high-frequency excitation signal, a functional relationship between resistance and current is constructed. Step S42: Construct a resistance error objective function based on the current prediction error. This objective function is based on the d-axis reference current and... The error value between the d-axis stator currents at time t is constructed; Step S43: Apply the chain rule to obtain the first-order partial derivative of the identified resistance value, obtain the gradient term and sensitivity term of the corresponding resistance parameter, and confirm the direction of the resistance estimate update. Step S44: According to the gradient descent principle, update the resistance parameters based on the current error of the resistance estimate and the iteration step size, so that they gradually converge to the true resistance value in repeated iterations.

7. The method for predicting current without delay by identifying gradient three parameters of a permanent magnet synchronous motor according to claim 1, characterized in that, In step S5, the flux linkage parameter identification model is implemented in the following way: Step S51, without considering the influence of the d-axis current, establish a system containing... Error model of q-axis stator current, flux linkage, sampling period, stator inductance, flux linkage estimate, q-axis reference current and rotor electric angular velocity at time t; Step S52, based on the q-axis reference current and The objective function is constructed based on the error between the q-axis stator currents at time t. Step S53: Apply the chain rule to obtain the first-order partial derivative of the identified magnetic flux value, obtain the gradient term and sensitivity term of the corresponding magnetic flux parameter, and confirm the update direction of the magnetic flux estimate. Step S54: According to the gradient descent principle, update the flux linkage parameters based on the current error of the flux linkage estimate and the iteration step size, so that they gradually converge to the true flux linkage value in repeated iterations.

8. The method for predicting current without delay by identifying gradient three parameters of a permanent magnet synchronous motor according to claim 1, characterized in that, In step S6, the range of convergence step size is determined based on Lyapunov stability analysis, and this is achieved in the following way: Step S61: Define the error terms for the three parameters: inductance, resistance, and flux linkage. Step S62: Select a Lyapunov function with the error of each parameter as the variable to reflect the magnitude of the overall parameter error; Step S63: Analyze the discrete changes of the function to derive sufficient conditions for its monotonically decreasing, thereby determining the range of values ​​that the iteration step size of each of the inductor, resistor and flux linkage should satisfy. By constraining the step size, the gradient update process remains stable.

Images