Model-free deadbeat predictive current control method based on adaptive current change component

Abstract

The present invention belongs to the technical field of motor drive and control applications, and proposes a model-free deadbeat predictive current control method based on adaptive current change components. The main solutions are as follows: First, a new model-free deadbeat prediction mechanism is designed, which uses the natural response component of current change and the current change under unit voltage for prediction. This mechanism only involves sampled current and does not require any mechanical parameters, so it has good parameter robustness. Second, in order to obtain the natural response component of current change and the current change under unit voltage required for prediction, the natural response component of current change is used as an adaptive variable, and the measured current change is used as a constraint term. On the premise of pursuing the overall prediction performance, an adaptive law for the natural response component of current change is designed, aiming to obtain the best ratio of current change components when the prediction error converges to zero. Finally, the average filtering strategy is combined with the model-free strategy to suppress the influence of sampling noise on the control performance.

Description

Technical Field

[0001] The present invention relates to the technical field of motor drive and control applications, and particularly to a model-free deadbeat predictive current control method based on an adaptive current change component. Background Art

[0002] Permanent magnet synchronous motors (PMSMs) are widely used in many fields such as transportation, national defense, agricultural production, and industrial manufacturing due to their advantages of simple structure, high reliability, and large power density. In recent years, with the rapid development of digital processors, model predictive control (MPC) has received increasing attention. MPC originated from industrial applications and has been applied to many fields such as traffic network control, power transmission systems, and motor drives. Especially in the field of motor drives, MPC is considered an effective alternative to field-oriented control and direct torque control.

[0003] Deadbeat predictive current control (DPCC) is a common MPC algorithm. Its core idea is to predict the future current behavior based on a model and then select the voltage vector that can make the future current closest to the reference current as the output. When there is a large error between the reference current and the actual current of the PMSM, DPCC will force the stator current to approach the reference current with the maximum driving ability of the system, so it has good dynamic performance. While this control mechanism endows the DPCC algorithm with good dynamic performance, it also makes it extremely sensitive to motor parameters. However, in actual applications, due to the influence of motor temperature and magnetic saturation, the model parameters are always fluctuating, and it is difficult to obtain accurate motor parameters through offline identification. To solve this problem, scholars have conducted a large number of studies.

[0004] Currently, common methods for improving parameter robustness include online parameter identification methods, disturbance compensation methods, and model-free control methods based on super-local models. Since there are many parameters to be identified in the prediction model, the online parameter identification method often has a large computational amount. The disturbance compensation method and the model-free control method based on super-local models can effectively suppress the current tracking static error caused by the mismatch of inductance and flux linkage parameters, but their performance in suppressing the current ripple caused by inductance mismatch is limited. In contrast, the model-free prediction technology based on current change uses the internal relationship between current change and voltage vector to realize current prediction and has stronger parameter robustness. This strategy is often applied to the model predictive current control of a single vector. For DPCC, the system output vector is synthesized by pulse width modulation of the control vector, and it is difficult to obtain the current change corresponding to the zero vector. Therefore, it is impossible to simply apply the model-free strategy based on current change to DPCC. Summary of the Invention

[0005] The object of the present invention is to provide a model-free deadbeat predictive current control method based on an adaptive current change component, aiming to solve the problems of poor robustness of the model predictive control algorithm and large current ripple of the traditional single-vector model-free predictive current control technology. Moreover, it can ensure the prediction accuracy of the strategy, eliminate the influence of sampling noise on the control performance, and improve the performance of MFDPCC control.

[0006] To solve its technical problems, the technical solution adopted by the present invention is as follows:

[0007] The model-free deadbeat predictive current control method based on an adaptive current change component includes the following steps:

[0008] Collect the phase current 、 、 and the bus voltage , and perform a park transformation on the sampled phase current to obtain the dq stator current k at the ( )-th period in the axis rotation coordinate system , read the predicted current k and the voltage vector output by the inverter at the ( k -1)-th period, and apply the optimal reference voltage vector calculated at the ( )-th period to the inverter, and denote it as

[0009] Subtract the sampled current k at the ( )-th period from the sampled current and the sampled current k at the ( 、 -1)-th period to obtain the corresponding current change ;

[0010] Calculate the natural response component and the forced response component of the current change using the current change component adaptability rate;

[0011] Calculate the unit voltage current change considering the sampling error, sum the forced response components of the current change and their corresponding voltage vectors within adjacent M cycles and save them in the arrays and respectively, and use the new forced response component k of the current change and its corresponding voltage vector ​Replace the old value in the array, update the summation result, and then further calculate the unit voltage current change;

[0012] Consider one - beat delay compensation and use the current change component to predict the current value at the future ( k +1) moment;

[0013] Use the reference speed Subtract the feedback speed , and obtain the reference current dq in the d - axis rotating coordinate system after calculation by the PI controller, At the same time, set d the q - axis reference current , and then calculate the reference current change in the ( k +1) cycle;

[0014] According to the basic principle of model - free dead - beat predictive current control, calculate the optimal reference voltage vector k in the ( +1) cycle, and prepare for the inverter output in the next cycle;

[0015] Repeat the above steps in sequence to form a closed - loop control.

[0016] As a further optimization, the mathematical model of the permanent - magnet synchronous motor in the dq d - q axis rotating coordinate system is:

[0017] ,

[0018] In the formula , are respectively dq the stator currents in the d - axis rotating coordinate system, , are respectively dq the stator voltages in the q - axis rotating coordinate system, L d , L q is dq the inductance in the d - q axis rotating coordinate system, , R s and are respectively the magnetic flux linkage of the permanent magnet, the stator resistance, and the electrical angular velocity of the rotor.

[0019] As a further optimization, use the forward Euler method to discretize the mathematical model of the permanent - magnet synchronous motor in the dq d - q axis rotating coordinate system, and deduce:

[0020] ,

[0021] where,T is the control period, k corresponding to the ( k )-th control period, representing the natural response component of the current change, representing the forced response component of the current change to any voltage vector;

[0022] and are expressed in the following form:

[0023] ,

[0024] where represents the input voltage vector, and the matrices A( k ), B( k ), and C( k ) are related to the parameters of the motor in the model.

[0025] As a further optimization, the following reference current change is defined to constrain the permanent magnet synchronous motor:

[0026] ,

[0027] where , are the reference currents. When the actual current change in the next period is equal to the reference current change, the tracking control of the reference current is achieved.

[0028] As a further optimization, the calculation formula for the current change is:

[0029] ,

[0030] In the formula, , are the sampled currents in the ( k )-th control period, , are the sampled currents in the ( k - 1)-th control period.

[0031] As a further optimization, the natural response component of the current change is described in a one-dimensional discrete form as:

[0032] ,

[0033] In the formula, represents the integral of the current prediction error, represents the integral coefficient, represents the unit voltage current change;

[0034] Calculating the unit voltage current change using the current change and its natural response component :

[0035] 。

[0036] As a further optimization, the calculation formula for the unit voltage current change considering the sampling error is

[0037] ,

[0038] where M represents the number of summation periods.

[0039] As a further optimization, the predicted current value at the future ( k +1) moment is:

[0040] ,

[0041] where and represent the voltage vectors applied to the inverter in this period, and the superscript p represents the predicted value.

[0042] As a further optimization, the calculation formula for the reference current change in the ( k +1) period is:

[0043] 。

[0044] As a further optimization, the calculation formula for the optimal reference voltage vector is:

[0045] 。

[0046] The beneficial effects of the present invention are:

[0047] 1. The present invention uses the sampled current to calculate all the information required to predict the optimal voltage vector, completely getting rid of the dependence on the current model of the traditional deadbeat control algorithm. At the same time, this method avoids the cumbersome process of determining the motor parameters, and is easy to debug, improving the reliability of the PMSM drive system.

[0048] 2. The present invention develops a model-free current control technology based on the adaptive current change component, extending the model-free technology based on the sampled current to the deadbeat predictive current control field. Compared with the traditional model-free predictive current control technology, it greatly reduces the current ripple and improves the quality of the three-phase current of the motor.

[0049] 3. The present invention takes into account the influence of sampling noise on the calculation of the forced response of current change, develops an effective noise suppression means, and improves the prediction accuracy of the optimal voltage vector.

[0050] 4. The present invention is significantly different from the DPCC algorithm based on parameter identification. The adaptive rate of the current change component developed by the present invention is designed based on the current prediction error, aiming to improve the overall prediction accuracy, rather than for the purpose of accurately identifying a certain parameter. Benefiting from this, this method also has a certain inhibitory effect on the current sixth harmonic caused by the inverter nonlinearity. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is the block diagram of the model-free deadbeat predictive current control principle in the embodiment of the present invention;

[0052] Figure 2 is the pole distribution of the characteristic equation of the natural response adaptive rate of current change in the embodiment of the present invention under different β conditions;

[0053] Figure 3 is the implementation process of the sampling noise suppression of the unit voltage current change in the embodiment of the present invention;

[0054] Figure 4 is the MFDPCC control flow chart in the embodiment of the present invention;

[0055] Figure 5 is the experimental waveform of DPCC at 0.3 L s parameters in the embodiment of the present invention;

[0056] Figure 6 is the experimental waveform of the model-based DPCC at 3 L s parameters in the embodiment of the present invention;

[0057] Figure 7 is the experimental waveform of the model-based DPCC under nominal parameters in the embodiment of the present invention;

[0058] Figure 8 is the MFDPCC experimental waveform in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention described and illustrated herein can be arranged and designed in various different configurations. Embodiment

[0060] Figure 1 This is the block diagram of the model - free dead - beat predictive current control in this embodiment, which includes a speed outer - loop PI regulator, an adaptive module for the natural response of current variation, adjacent M cycle voltage vector summation modules, adjacent M cycle current variation forced response summation modules corresponding to voltage vectors, a unit voltage - current variation calculation module, a delay compensation and optimal voltage vector prediction module, an inverter, current sensors, voltage sensors, and a permanent - magnet synchronous motor. Therefore, the model - free dead - beat predictive current control method provided in this embodiment is based on the adaptive current variation component and includes the following steps:

[0061] Step 1: Initialize the parameters related to the algorithm execution according to the requirements.

[0062] Specifically, there are two points that need special attention in the parameter initialization process:

[0063] (1) The natural response component of current variation and the forced response component of current variation cannot be initialized to zero simultaneously. and have no such limitation. This is because d the reference current and the actual current on the axis are both zero at startup. When are both zero, the current prediction error is also zero, and it is easy for the adaptive mechanism to fail to start. Although in actual implementation, the sampled current may not always remain at zero, this situation should be avoided to ensure the reliable startup of the system.

[0064] (2) All data in the arrays and must be initialized to zero. The data in the arrays and are updated every M control cycles to ensure that only the data within adjacent M cycles are added when calculating CVPUU. If the initial data in and is not zero, the summation result for the first 2 M cycles will be inaccurate, which will cause the motor startup to fail.

[0065] Step 2: Collect the phase currents , , and the bus voltage through current sensors and voltage sensors, and perform park transformation on the sampled phase currents to obtain dqStator current in the ([[]] k [[]])-th period in the axis rotation coordinate system k , read the predicted current and , as well as the voltage vector output by the inverter in the ([[]] k [[]])-th period , and apply the optimal reference voltage vector calculated in the ([[]] k [[]])-th period to the inverter, and denote it as k k , and

[0066] ; dq Here, the three-phase current collected by the current sensor is transformed into the dq axis rotation coordinate system through the following formula:

[0067] ,

[0068] where θ is the rotor electrical angle, and the voltage vectors are all synthesized by space voltage vector pulse width modulation (SVPWM).

[0069] Step 3: Take the difference between the sampled current k in the ([[]] k [[]])-th period and and the sampled current k in the ([[]] k [[]])-th period and to obtain the corresponding current change , and this process can be written as:

[0070] ,

[0071] In the formula, and are the sampled currents in the ([[]] k [[]])-th control period, k and are the sampled currents in the ([[]] k [[]])-th control period k .

[0072] It should be noted that the current change contains two parts, specifically as follows:

[0073] ,

[0074] In the formula represents the natural response of the current change, mainly caused by the back electromotive force; represents the forced response of the current change to any voltage vector, mainly controlled by the input voltage. By separating the natural response and the forced response in the current change, the future optimal voltage vector can be simply predicted based on the reference current change.

[0075] Step 4: Calculate the natural response component of the current change using the current change component adaptability rate and the forced response component of the current change .

[0076] For the sake of simplicity in description, the one-dimensional discrete form of the adaptability rate of the natural response component of the current change can be expressed as:

[0077] ,

[0078] wherein, represents the integral of the current prediction error, represents the integral coefficient, represents the unit voltage current change; using this adaptability rate to estimate the required k predicted in the ( )th cycle .

[0079] Then calculate the unit voltage current change using the current change and its natural response component:

[0080] ,

[0081] The natural response component of the current change is a free term that changes according to the system operating state, while the measured current change is and the constraint term of the sum. When the motor is in steady-state operation, the of each cycle of DPCC will not fluctuate too much. Therefore, will adjust with the change of and reach a stable state.

[0082] The value range of the integral coefficient in the adaptability rate described in Step 4 can be determined in the following way. First, the closed-loop transfer function after the z transformation of the adaptability rate of the natural response component of the current change is:

[0083] ,

[0084] The characteristic equation of the transfer function is:

[0085] ,

[0086] Further solving from the system characteristic equation to obtain the poles of :

[0087] ,

[0088] Differentβ at a value z the distribution of the pole positions in the domain is as Figure 2 shown. When β is less than 0, at most only one pole is inside the unit circle; when β is greater than 10000, both poles are outside the unit circle. According to the principle of automatic control, a necessary and sufficient condition for the stability of a discrete system is that all characteristic roots are distributed inside the unit circle. It can be seen from Figure 2 that when β is in the range of 0 to 10000, the system is stable.

[0089] The above adaptive strategy is implemented based on the sampled current. However, in the actual current sampling process, sampling noise is inevitable, and this noise will be introduced into the above adaptive strategy, further deteriorating the prediction of the delay compensation current and the reference voltage vector. Therefore, it is necessary to suppress the sampling noise.

[0090] Step Five: Calculate the unit voltage and current change considering the sampling error. Save the forced response component of the current change and its corresponding voltage vector within adjacent M cycles in arrays and respectively, and sum them separately. When a new control cycle arrives, the new forced response component of the current change and its corresponding voltage vector are used to replace the old values in the arrays, update the summation results, as shown in Figure 3 , and then further calculate the unit voltage and current change considering the sampling error.

[0091] Specifically, based on the principle of mean filtering, the expression of the unit voltage and current change is improved as follows:

[0092] ,

[0093] where M represents the number of cycles for summation.

[0094] The above formula can be further rewritten as:

[0095] ,

[0096] When the summation cycle is large enough, the following equation holds:

[0097] ,

[0098] Substitute the expression of the unit voltage and current change in the( k-2 ) cycle into the expression of the unit voltage and current change in the( k-1 ) cycle, and we get:

[0099] ,

[0100] Among them . It can be seen from this formula that the change of unit voltage and current after improvement is the result of low-pass filtering of the change of unit voltage and current before improvement. Therefore, the influence of sampling noise can be weakened. M The larger the value of λ , the smaller the value of M , and the better the noise suppression performance. The sampling noise suppression process will move forward with time. The and its corresponding voltage vector of the adjacent cycles will be saved in the arrays and M for summation. The selection of

[0101] needs to comprehensively consider the memory usage of the digital signal processor and the noise suppression performance. k Step Six: Considering one-beat delay compensation, use the current change component to predict the current value at the future (

[0102] Specifically, the current value at the ( k +1) moment can be predicted as:

[0103] ,

[0104] Among them, and represent the voltage vectors acting on the inverter in this cycle. The superscript p represents the predicted value.

[0105] Step Seven: Subtract the feedback speed from the reference speed . After calculation by the PI controller, obtain the reference current dq in the axis rotation coordinate system. At the same time, set the d axis reference current , and then calculate the reference current change in the ([[]] k +1) cycle.

[0106] Specifically, the reference current change can be calculated by the following formula:

[0107] ,

[0108] Since the reference current in the ([[]] k-2 ) cycle is approximately equal to the reference current in the ([[]] k ) cycle, and can be used for substitution.

[0109] Step Eight: Calculate the optimal reference voltage vector for the ( k +1)th period according to the basic principle of model-free deadbeat predictive current control , preparing for the inverter output of the next period.

[0110] Specifically, the calculation formula of the optimal reference voltage vector can be expressed as:

[0111] ,

[0112] It can be easily seen from the above formula that the calculation process of the optimal reference voltage vector does not involve any motor mechanical parameters, so it has strong parameter robustness.

[0113] Step Nine: Repeat Steps Two to Eight in sequence to form a closed-loop control. The above MFDPCC control flow is as Figure 4 shown. Only at the startup stage is it necessary to make a non-zero judgment on , because after startup, the value of must be much greater than zero, which is sufficient to meet the calculation requirements of unit voltage and current changes and will not cause the calculation error to be amplified.

[0114] Finally, this embodiment was verified through experiments. The experimental data was collected by an oscilloscope, then the collected data was imported into the MATLAB workspace, and the experimental waveforms were obtained using the From Workspace module in Simulink. Figures 5 - 7 They are respectively the experimental results of model-based DPCC at a speed of 600 r / min under 0.3 L s , 3 L s and nominal parameters. It can be seen that the model-based DPCC has good current control performance under nominal parameters, and the THD of the phase current is 1.76%. Under the condition of 0.3 L s inductance mismatch, the model-based DPCC only has a slight tracking error, and the THD of the phase current slightly increases to 3.37%. However, in the case of three times the rated inductance, dq the ripple content in the shaft current greatly increases, and the THD surges to 18%. The MFDPCC proposed in this embodiment is not affected by motor parameters and always maintains the optimal current tracking performance and ripple control performance, with a THD of only 1.25%, as Figure 8 shown.

[0115] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A model-free deadbeat predictive current control method based on an adaptive current change component, characterized in that Including the following steps: Collect phase current , , and bus voltage , and perform park transformation on the sampled phase current to obtain dq the stator current of the k cycle in the d-q axis rotating coordinate system , , read the predicted current and the voltage vector output by the inverter in the k -1 cycle , and apply the optimal reference voltage vector calculated in the k -1 cycle to the inverter, and denote it as ; Take the sampling current of the k cycle and the sampling current of the k cycle before the previous cycle, subtract them to obtain the corresponding current change ; Calculating the natural response component of current change using the adaptive rate of current change component and the forced response component of current change ; Calculate the unit voltage and current variation considering the sampling error, and save the forced response components of the current variation and the corresponding voltage vectors within adjacent M cycles into arrays and respectively for summation. In the k cycle, replace the old values in the arrays with the new forced response components of the current variation and the corresponding voltage vectors , update the summation result, and then further calculate the unit voltage and current variation; Consider one - beat delay compensation and use the current change component to predict the future k current value at the +1 moment; Subtract the feedback speed from the reference speed to obtain, after calculation by a PI controller the reference current in the axis rotation coordinate system dq , and at the same time set the axis reference current d , and then calculate the change in the reference current for the +1 cycle; k ​ Calculate the optimal reference voltage vector for the k next cycle according to the basic principle of model-free deadbeat predictive current control to prepare for the inverter output in the next cycle; Repeat the above steps in sequence to form a closed-loop control.

2. The model-free deadbeat predictive current control method based on the adaptive current change component according to claim 1, wherein The mathematical model of the permanent magnet synchronous motor in dq the d-q axis rotating coordinate system is as follows: ， where , are respectively dq the stator currents in the α-axis rotating coordinate system, , are respectively dq the stator voltages in the β-axis rotating coordinate system, L d , L q are dq the inductances in the α-axis rotating coordinate system, , R s and are respectively the magnetic flux linkage of the permanent magnet, the stator resistance, and the electrical angular velocity of the rotor.

3. The model-free deadbeat predictive current control method based on the adaptive current change component according to claim 2, characterized in that Using the forward Euler method to discretize the mathematical model of the permanent magnet synchronous motor in the dq axis rotating coordinate system, it is deduced that: ， Among them, T is the control period, k corresponding to the k th control period, represents the natural response component of the current change, represents the forced response component of the current change to any voltage vector; and is represented in the following form: where represents the input voltage vector, and the matrices A( k ), B( k ) and C( k ) are related to the parameters of the motor in the model.

4. The model-free deadbeat predictive current control method based on the adaptive current change component according to claim 2 or 3, characterized in that Define the following reference current variation for constraining the permanent magnet synchronous motor: , Among them, , are reference currents. When the actual current change in the next cycle is equal to the reference current change, the tracking control of the reference current is achieved.

5. The model-free deadbeat predictive current control method based on the adaptive current change component according to claim 1, wherein The current change The calculation formula is as follows: , In the formula, , are the sampled currents of the k th control period, , are the sampled currents of the k -1th control period.

6. The model-free deadbeat predictive current control method based on the adaptive current change component according to claim 1, characterized in that The natural response component of the current change is described in a one-dimensional discrete form as: , In the formula, represents the integral of the current prediction error, represents the integral coefficient, represents the unit voltage-current change; Calculating the unit voltage current change using the current change and its natural response component : 。 7. The model-free deadbeat predictive current control method based on the adaptive current change component according to claim 1, characterized in that The calculation formula for the unit voltage-current variation considering the sampling error is , Among them, M represents the number of summation periods.

8. The model-free deadbeat predictive current control method based on the adaptive current change component according to claim 1, wherein The future k The predicted current value at the +1 moment is: , Among them, and represent the voltage vectors acting on the inverter in this period, and the superscript p represents the predicted value.

9. The model-free deadbeat predictive current control method based on the adaptive current change component according to claim 1, characterized in that The k calculation formula for the reference current change in the +1 cycle is as follows: 。 10. The model-free deadbeat predictive current control method based on the adaptive current change component according to claim 1, characterized in that, The optimal reference voltage vector The calculation formula is as follows: 。

Images