Three-level parameter-free predictive current control method based on improved super-local model

Abstract

The invention discloses a three-level parameter-free predictive current control method based on an improved hyper-local model, and the method comprises the steps: firstly, collecting the three-phase stator current, the rotor electrical angle and the actual rotating speed of a permanent magnet synchronous motor at the moment k of a current sampling period, and calculating the d-q axis stator current through a vector space decoupling scheme; secondly, a permanent magnet synchronous motor mathematical model is constructed based on a hyper-local model, and estimated values of a system disturbance term and a system gain factor are obtained through a sliding-mode observer and a model reference adaptive system respectively; then, calculating a current prediction value at a next sampling period (k + 1) moment through a forward Euler formula; and finally, performing rolling optimization on the value function to obtain an optimal voltage vector. According to the method, the robustness of a three-level model prediction system to motor parameter changes can be remarkably improved, the problem that motor parameter dependence still exists in traditional model-free prediction current control is effectively solved, and the stability of the system is improved.

Description

Technical Field

[0001] The present invention relates to a motor control method, and in particular to a three-level parameter-free predictive current control method. Background Art

[0002] In modern power electronics applications, diode neutral point clamped (NPC) three-level inverters have become a core component of medium- and high-voltage power conversion systems due to their technical advantages, including high-voltage output capability, high power handling, and stable operation. As control strategies evolve toward higher performance and intelligent control, model predictive current control (MPCC), an advanced control method that combines optimization and responsiveness, has also been widely used in three-level inverters. However, in practical applications, traditional MPCC algorithms rely heavily on the accuracy of motor model parameters, particularly stator inductance and resistance. When the motor operating environment is complex, parameters change, or modeling errors exist, control system performance can significantly degrade, even affecting system stability and exhibiting significant parameter sensitivity. To address MPCC parameter sensitivity and improve parameter robustness, parameter-free predictive current control (PFPCC) has become a research hotspot. This method predicts future motor currents and selects the optimal voltage vector (VV) based on input and output data.

[0003] Traditional model-free predictive current control algorithms primarily utilize hyperlocal mathematical models, which unify all unknown structural and external disturbances into a single total disturbance term. Algebraic identification methods are used to estimate and compensate for the model's total disturbance term. This method, to a certain extent, suppresses the impact of modeling errors and external disturbances on the system's dynamic performance, reduces current ripple, and improves the system's steady-state accuracy and dynamic response speed. Although this type of method avoids the reliance on complete motor parameters in traditional model predictive control, the system gain factors involved in estimating the disturbance compensation term still include inductor parameters, which are not compensated in real time. Under long-term system operation conditions, model mismatch caused by changes in inductor parameters becomes an inevitable problem. Summary of the Invention

[0004] Purpose of the invention: To address the above problems, a three-level parameter-free predictive current control method based on an improved hyperlocal model is proposed to solve the problem of motor parameter dependence in traditional model-free predictive current control, achieve high dynamic response and low current ripple, and improve the stability of the three-level drive system.

[0005] Technical solution: A three-level parameter-free predictive current control method based on an improved hyperlocal model, including:

[0006] Step 1: Obtain the rotor electrical angle θ and three-phase current at time k through the encoder and current sensor respectively, and calculate the dq-axis stator current i using Clark and Park coordinate transformation d (k), i q (k);

[0007] Step 2: Set the value according to the motor speed The actual motor speed N r , the q-axis reference current value at time (k+1) is obtained through the speed PI controller Given d-axis reference current value

[0008] Step 3: Calculate the dq-axis disturbance term F through the sliding film observer d 、F q The estimated value at time k and

[0009] Step 4: Online identification of the dq axis system gain factor to obtain the estimated value at time k and

[0010] Step 5: Get the switching state S of the inverter output at time k a (k), S b (k), S c (k), and the dq-axis stator voltage u at time k is obtained through coordinate transformation d (k),u q (k);

[0011] Step 6: Based on the estimated value of the disturbance term at time k and Estimated value of the gain factor and dq axis stator voltage u d (k),u q (k) and dq axis stator current i d (k), i q (k) Calculate the predicted dq axis current value at time (k+1)

[0012] Step 7: Different Voltage Vectors V j Send it into the value function to obtain the output value g of the value function j ={g1, g2, ···, g 27}, take the basic voltage vector that minimizes the cost function, and output the optimal switching state to drive the inverter through the midpoint potential balancing module.

[0013] Furthermore, in step 3, the state equation of the synovial observer is:

[0014]

[0015] Where, and is the observed value of the dq axis current at time k, α d (k), α q (k) are the dq axis system gain factors at time k, u d (k),u q (k) are the components of the stator voltage on the dq axis at time k, They are the dq axis disturbance terms F at time k respectively d 、F q The estimated value of U dsmo , U qsmo is the dq axis sliding mode control quantity;

[0016] Among them, the dq-axis stator current is used as the observation quantity to construct the dq-axis sliding mode surface:

[0017]

[0018] The sliding mode control quantity U is obtained by the exponential approximation law dsmo 、U qsmo :

[0019]

[0020] Where S1 is the sliding mode control function parameter, sign is the sign function, S d 、S q is the dq-axis sliding surface.

[0021] Furthermore, step 4 includes the following specific steps: constructing a reference current model based on the dq-axis stator voltage equation:

[0022]

[0023] Where i d (k-1), i q (k-1) is the dq axis current value at time (k-1), i d (k-2), i q (k-2) is the dq axis current value at the time (k-2), u d (k-1),u q(k-1) is the dq axis voltage value at time (k-1), u d (k-2),u q (k-2) is the dq axis voltage value at time (k-2), T s is the sampling time;

[0024] The adjusted current model is constructed by reference current model:

[0025]

[0026] Where, and Is to adjust the current model to the dq axis current i d (k), i d The estimated value of (k), and is the dq axis system gain factor α d (k), α q (k) identification value;

[0027] The Landau discrete-time recursive parameter identification algorithm is used to design the adaptive law:

[0028]

[0029] Where, and is the dq axis system gain factor α at time k d (k), α q The estimated value of (k), and is the dq axis system gain factor α at time (k-1) d (k-1), α q The estimated value of (k-1), Δu d (k-1), Δu q (k-1) is the dq axis voltage difference at time (k-1), β is the adaptive gain;

[0030] Substitute the reference current model and the adjustment current model into the adaptive law to obtain the dq axis system gain factor α d (k), α q Estimated value of (k) and :

[0031]

[0032] Where, β d , β q is the adaptive gain parameter.

[0033] Furthermore, the step 6 includes the following specific steps: constructing a current model based on a sliding mode observer:

[0034]

[0035] The stator current prediction model is obtained by discretizing the current model through the forward difference algorithm:

[0036]

[0037] Where, and is the predicted value of the dq-axis stator current at time (k+1).

[0038] Beneficial Effects: By constructing a current prediction equation based on a hyperlocal model, the present invention replaces the traditional reliance on precise motor model calculations, effectively improving the robustness of the control system under conditions such as motor parameter changes and modeling errors, and significantly improving the stability and accuracy of the three-level drive system. At the same time, a sliding mode observer is used to estimate the nonlinear disturbance terms in the hyperlocal model, improving the real-time response performance of the system. Furthermore, a model reference adaptive system is introduced to estimate the system gain factor online without manual setting or reliance on static parameter identification, thus achieving true parameter-free predictive control and enhancing the adaptability and practicality of the control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a schematic diagram of the three-level parameter-free predictive current control principle of the permanent magnet synchronous motor of the present invention;

[0040] Figure 2 is the dq axis system gain factor α of the three-level parameter prediction current control method of the present invention d , α q Tracking performance simulation diagram, where (a) is the gain factor α d Tracking performance simulation diagram, (b) is the gain factor α q Tracking performance simulation graph;

[0041] Figure 3 This is a simulation diagram of the q-axis current tracking performance of the three-level parameter prediction current control method proposed in this invention, where (a) is the accurate parameter (L d =3.465mH, L d =3.93mH) under the condition of current tracking simulation waveform, (b) is the dq axis inductance increased by 1.5 times (L d =5.1975mH, L d =5.895mH) under the condition of current tracking simulation waveform;

[0042] Figure 4The three-phase current waveform of the three-level parameter-free predictive current control method proposed by the present invention;

[0043] Figure 5 This is a dynamic performance simulation diagram of the three-level parameter-free predictive current control method proposed in the present invention, where (a) is the simulation waveform under sudden torque conditions, and (b) is the simulation waveform under sudden speed conditions. DETAILED DESCRIPTION

[0044] The present invention will be further described in detail below with reference to the accompanying drawings and through examples. The following examples are intended to explain the present invention but the present invention is not limited to the following examples.

[0045] like Figure 1 As shown in FIG, a three-level parameter-free predictive current control method for a permanent magnet synchronous motor based on an improved sliding mode observer includes the following specific steps:

[0046] Step 1: At the current sampling period k, the rotor electrical angle θ is obtained through the encoder, and the three-phase current i is obtained through the current sensor. s (k), s = a, b, c; the three-phase current is converted into the current component i in the α-β axis stationary coordinate system through Clark transformation of formula (1) α (k), i β (k); then the Park transformation of formula (2) combined with the rotor electrical angle θ is converted into the current component i in the dq axis rotating coordinate system d (k), i q (k);

[0047]

[0048] Step 2: Set the value according to the motor speed The actual motor speed N r The speed difference is obtained by the speed PI controller to obtain the q-axis reference current value at the next sampling period (k+1) Given d-axis reference current value (k+1)=0;

[0049]

[0050]

[0051] Where ΔNr is the speed difference, k p 、k i are the proportional gain and integral gain of the speed loop PI controller respectively, and s is a complex variable.

[0052] Step 3: Calculate the dq-axis disturbance term F through the sliding film observer d 、Fq The estimated value at time k and

[0053] 3.1: Take the dq axis stator current as the observation quantity and construct the dq axis sliding mode surface S d 、S q :

[0054]

[0055] Where, and is the observed value of the dq axis current at time k in the current sampling period;

[0056] 3.2: The sliding mode control function is obtained through the exponential approximation law:

[0057]

[0058] Where U dsmo , U qsmo is the dq axis sliding mode control quantity, S1 is the sliding mode control parameter, S1 is used to adjust the convergence speed of the system along the sliding surface, and sign is the sign function. This step is used to generate the sliding mode control quantity. Through this sliding mode control function, the sliding mode observer can drive the current observation value. Quickly track actual current. Use exponential approximation to ensure the sliding surface quickly converges to 0, improving response speed.

[0059] 3.3: Design the state equation of the dq-axis sliding mode observer, obtain the discrete model by discretization, and obtain the disturbance term F in the discrete domain d 、F q The estimated value at time k in the current sampling period.

[0060] Specifically, the disturbance information is extracted from the sliding mode control quantity, and the dq axis disturbance term F at the current sampling period k is calculated. d 、F q Estimated value of :

[0061]

[0062] Where k1 is the sliding mode design parameter, which is used to adjust the sensitivity and convergence of disturbance estimation. dsmo 、U qsmo In addition to the part of driving current convergence, it also implies the need to compensate for system disturbances. dsmo 、U qsmo Integration can separate the equivalent compensation of disturbance Used for disturbance cancellation of subsequent observers.

[0063] By integrating voltage drive, disturbance compensation, and sliding mode control, the state equation of the sliding mode observer is finally constructed as follows:

[0064]

[0065] Where, α d (k), α q (k) are the dq axis system gain factors at the current sampling period k, u d (k),u q (k) are the dq-axis components of the stator voltage at time k in the current sampling period. The system gain factor replaces the motor parameters in the traditional model and is adaptively identified online using the model reference, achieving parameter-free operation while actively compensating for system disturbances. This provides a precise current observation foundation for subsequent parameter-free predictive current control.

[0066] The state equation of the dq-axis sliding mode observer is discretized using the forward difference algorithm to obtain the discrete model of the sliding mode observer, as shown in formulas (9) and (10):

[0067]

[0068] Where, and is the observed value of the dq-axis stator current at the current sampling period k, and is the predicted value of the dq-axis stator current at the next sampling period (k+1), and They are the dq axis disturbance terms F at the next sampling period (k+1) d 、F q The predicted value, T s is the sampling time.

[0069] Step 4: Online identification of the dq-axis system gain factors, including:

[0070] The reference current model is constructed based on the dq-axis stator voltage equation:

[0071]

[0072] Where i d (k-1), i q (k-1) is the dq axis current value at the previous sampling period (k-1), i d (k-2), i q (k-2) is the dq axis current value at the moment of the last two sampling cycles (k-2), u d (k-1),u q(k-1) is the dq axis voltage value at the previous sampling period (k-1), u d (k-2),u q (k-2) is the dq-axis voltage value at the moment of the last two sampling periods (k-2).

[0073] The adjusted current model is constructed by reference current model:

[0074]

[0075] Where, and Is to adjust the current model to the dq axis current i d (k), i d The estimated value of (k), and is the dq axis system gain factor α d (k), α q (k) is the identification value. and Substitute the real gain α in the reference current model d (k), α q (k), whose output and The deviation from the reference current model reflects the parameter estimation error. The current model is adjusted by continuously adjusting and The adjusted current model is made to approach the reference current model to achieve parameter self-correction.

[0076] The Landau discrete-time recursive parameter identification algorithm is used to design the adaptive law:

[0077]

[0078] Where, and is the dq axis system gain factor α at the current sampling period k d (k), α q The estimated value of (k),

[0079] and is the dq axis system gain factor α at the last sampling period (k-1) d (k-1), α q The estimated value of (k-1), Δu d (k-1), Δu q (k-1) are the dq axis voltage differences at the previous sampling period (k-1); Δi d (k), Δi q(k) are the current deviations between the dq axis reference current model and the adjusted current model at the k moment of the current sampling period, β is the adaptive gain, which determines the parameter update speed. d (k-1)=u d (k-1)-u d (k-2), Δu q (k-1)=u q (k-1)-u q (k-2), the voltage difference is the excitation for the driving current change. When the current deviation between the adjustment model and the reference model is not equal to 0, the voltage difference Δu is proportionally adjusted. Make the deviation gradually approach 0.

[0080] Substitute the reference current model and the adjustment current model into the adaptive law to obtain the dq axis system gain factor α d (k), α q Estimated value of (k) and :

[0081]

[0082] Where, β d , β q Is the adaptive gain parameter, which is dynamically adjusted with the square of the voltage difference. When the voltage changes greatly, β d Approaching βT s , fast update; when the voltage is stable, β d Close to 0 to avoid false updates.

[0083] Step 4 indirectly avoids reliance on a single parameter compared to traditional control by identifying the system gain factor. The identified system gain factor is fed back into the sliding mode observer state equation in step 3 and the stator current prediction model in step 6, replacing the original fixed parameters and achieving parameter-free closed-loop control. This significantly improves the overall control scheme's adaptability to parameter uncertainty and external disturbances.

[0084] Step 5: The dq-axis stator voltage u at the current sampling period k d (k),u q (k) and the three-level inverter output switch state S a (k), S b (k), S c (k) The component u of the stator voltage on the α-β axis at the current sampling period k is obtained by Clark transformation of formula (16): α (k),u β(k), and then after the Park transformation of formula (17), the component u of the stator voltage on the dq axis at the current sampling period k is obtained d (k),u q (k);

[0085]

[0086] Among them, V dc is the DC bus voltage.

[0087] Step 6: Predict the dq-axis stator current value at the next sampling period (k+1).

[0088] The current model constructed based on the sliding mode observer is shown in formula (18); formula (18) is discretized by the forward difference algorithm to obtain the stator current prediction model shown in formula (19);

[0089]

[0090] Where, and is the observed value of the dq-axis stator current at the current sampling period k, and It is the predicted value of the dq-axis stator current at the next sampling period (k+1).

[0091] Step 7: Use the dq axis current prediction value obtained by the stator current prediction model and Input into the value function formula (20), for the 27 voltage vectors that can be output by the three-level inverter, different voltage vectors V j The value function output g under the action j ={g1, g2, ···, g 27}; Then use formula (21) to obtain the minimized value function output g min , g min The corresponding voltage vector U opt That is the optimal voltage vector, U opt As the basic voltage vector of the three-level inverter, the optimal voltage vector is sent to the midpoint potential balance module to obtain the on-off state of the switch tube in the next cycle as the driving signal of the three-level inverter;

[0092]

[0093] g min =min{g1,g2,···,g 27} (twenty one)

[0094] The cost function g is the square sum of the dq axis current tracking errors, reflecting the “applied voltage vector V j After that, the deviation degree between the actual current and the reference current is calculated. The smaller g is, the better the current tracking effect is.

[0095] The simulation results of the method of the present invention are as follows Figure 2 、 Figure 3 、 Figure 4 、 Figure 5 As shown. Figure 2 (a) shows that it can track the q-axis gain factor α well. q ,from Figure 2 (b) shows that it can track the d-axis gain factor α well. d .from Figure 3 (a) shows that under the condition of accurate motor parameters, the actual current can track the reference current very well. Figure 3 As can be seen from (b), the three-level parameter-free predictive current control method for the permanent magnet synchronous motor proposed in the present invention does not require any motor parameters to participate in the calculation, and therefore can always maintain good q-axis current tracking performance. Figure 4 The three-phase current waveform in steady state is shown, and it can be seen that the current can maintain a good sinusoidal degree. Figure 5 It can be seen from (a) and (b) that the three-level parameter-free predictive current control method proposed in the present invention has a faster dynamic response speed in both sudden torque and sudden speed changes.

[0096] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A three-level parameter-free predictive current control method based on an improved hyperlocal model, characterized in that: include: Step 1: Obtain the rotor electrical angle θ and three-phase current at time k through the encoder and current sensor respectively, and calculate the dq-axis stator current i using Clark and Park coordinate transformation d (k), i q (k); Step 2: Set the value according to the motor speed The actual motor speed N r , the q-axis reference current value at time (k+1) is obtained through the speed PI controller Given d-axis reference current value Step 3: Calculate the dq-axis disturbance term F through the sliding film observer d 、F q The estimated value at time k and Step 4: Online identification of the dq axis system gain factor to obtain the estimated value at time k and Step 5: Get the switching state S of the inverter output at time k a (k), S b (k), S c (k), and the dq-axis stator voltage u at time k is obtained through coordinate transformation d (k),u q (k); Step 6: Based on the estimated value of the disturbance term at time k and Estimated value of the gain factor and dq axis stator voltage u d (k),u q (k) and dq axis stator current i d (k), i q (k) Calculate the predicted dq axis current value at time (k+1) Step 7: Different Voltage Vectors V j Send it into the value function to obtain the output value g of the value function j ={g1, g2, ···, g 27 }, take the basic voltage vector that minimizes the cost function, and output the optimal switching state to drive the inverter through the midpoint potential balancing module.

2. The three-level parameter-free predictive current control method based on the improved hyperlocal model according to claim 1, characterized in that: In step 3, the state equation of the synovial observer is: Where, and is the observed value of the dq axis current at time k, α d (k), α q (k) are the dq axis system gain factors at time k, u d (k),u q (k) are the components of the stator voltage on the dq axis at time k, They are the dq axis disturbance terms F at time k respectively d 、F q The estimated value of U dsmo , U qsmo is the dq axis sliding mode control quantity; Among them, the dq-axis stator current is used as the observation quantity to construct the dq-axis sliding mode surface: The sliding mode control quantity U is obtained by the exponential approximation law dsmo 、U qsmo : Where S1 is the sliding mode control function parameter, sign is the sign function, S d 、S q is the dq-axis sliding surface.

3. The three-level parameter-free predictive current control method based on the improved hyperlocal model according to claim 2, characterized in that: The step 4 includes the following specific steps: constructing a reference current model based on the dq-axis stator voltage equation: Where i d (k-1), i q (k-1) is the dq axis current value at time (k-1), i d (k-2), i q (k-2) is the dq axis current value at the time (k-2), u d (k-1),u q (k-1) is the dq axis voltage value at time (k-1), u d (k-2),u q (k-2) is the dq axis voltage value at time (k-2), T s is the sampling time; The adjusted current model is constructed by reference current model: Where, and Is to adjust the current model to the dq axis current i d (k), i d The estimated value of (k), and is the dq axis system gain factor α d (k), α q (k) identification value; The Landau discrete-time recursive parameter identification algorithm is used to design the adaptive law: Where, and is the dq axis system gain factor α at time k d (k), α q The estimated value of (k), and is the dq axis system gain factor α at time (k-1) d (k-1), α q The estimated value of (k-1), Δu d (k-1), Δu q (k-1) is the dq axis voltage difference at time (k-1), β is the adaptive gain; Substitute the reference current model and the adjustment current model into the adaptive law to obtain the dq axis system gain factor α d (k), α q Estimated value of (k) and Where, β d , β q is the adaptive gain parameter.

4. The three-level parameter-free predictive current control method based on the improved hyperlocal model according to claim 3, characterized in that: The step 6 includes the following specific steps: constructing a current model based on a sliding mode observer: The stator current prediction model is obtained by discretizing the current model through the forward difference algorithm: Where, and is the predicted value of the dq-axis stator current at time (k+1).

Images