Harmonic space-considered multi-vector model predictive control method for dual three-phase permanent magnet synchronous generator

Abstract

The invention discloses a multi-vector model predictive control method for a dual three-phase permanent magnet synchronous generator considering a harmonic space, which comprises the following steps: selecting three voltage vectors to construct a target voltage vector in each control sector, substituting the target voltage vector into a discrete predictive control model to obtain a predictive current at the next moment, and further constructing a value function to obtain a predictive current at the next moment; and selecting an optimal virtual voltage vector set of the two subspaces and action time of the optimal virtual voltage vector through a minimum value function, combining control of the two subspaces into a control period, recalculating a switching sequence PWM signal, and acting on the rectifier. In the process of selecting the optimal virtual voltage vector set of the two subspaces, a superhelix integral sliding mode observer is used for observing current disturbance, and a composite controller is used for conducting zero static error tracking on harmonic waves in the z1z2 subspace. Under the condition that the dynamic performance is not affected, the fundamental wave space control performance is improved, the harmonic current is suppressed, and the steady-state performance of the DTP-PMSG is improved.

Description

Technical Field

[0001] The present invention relates to the field of model predictive control of dual three-phase permanent magnet synchronous generators, and specifically to a multi-vector model predictive control method for a dual three-phase permanent magnet synchronous generator (DTP-PMSG) considering the harmonic space, which is beneficial to further reduce the harmonic current content, reduce the rectifier switching loss, and improve the steady-state performance of the system. Background Technique

[0002] In recent years, with the progress of science and technology, the all-electricization in fields such as electric vehicles, ship traction, and aerospace has also developed vigorously. The DC power supply with a generator as the core has received extensive attention. The multi-phase permanent magnet synchronous generator (PMSG) has become a key research object due to its characteristics of low voltage and high power output and strong fault tolerance. Among them, the dual three-phase permanent magnet synchronous generator DTP-PMSG is more prominent in terms of static and dynamic performance, torque density, and fault tolerance, and has good application prospects. DTP-PMSG and its control system have become one of the research hotspots at home and abroad.

[0003] The DTP-PMSG control system mainly adopts finite set model predictive current control (MPCC). However, when the optimal voltage vector of the finite set model predictive current control acts throughout the control cycle, it often disrupts the switching frequency, resulting in a larger current ripple. Moreover, since the voltage vector control set used in traditional model predictive current control is the fundamental space voltage vector, and the fundamental space voltage vector has projections of different sizes in the harmonic space, the harmonic current of traditional model predictive control cannot be suppressed, and the stator current distortion degree is relatively high. On the other hand, the control performance of traditional model predictive control depends on accurate motor parameters. However, in the actual operation process, many factors such as the temperature, magnetic saturation, and rectifier nonlinear factors of the motor will cause changes in the motor parameters, resulting in parameter mismatch of the control system and reducing the control performance. Summary of the Invention

[0004] Aiming at the deficiencies in the prior art, the present invention provides a multi-vector model predictive control method for a dual three-phase permanent magnet synchronous generator considering the harmonic space.

[0005] The present invention realizes the above technical objectives through the following technical means.

[0006] Multi-vector model predictive control method for a dual three-phase permanent magnet synchronous generator considering the harmonic space:

[0007] Determine the discrete predictive control model of the dual three-phase permanent magnet synchronous generator in the rotating coordinate system;

[0008] In each control sector, three voltage vectors are selected to construct the target voltage vector, which is then substituted into the discrete predictive control model to obtain the predicted current at the next moment. Further, a cost function is constructed to obtain the optimal virtual voltage vector and its action time.

[0009] The optimal virtual voltage vector sets in the αβ subspace and the z1z2 subspace are selected by minimizing the cost function, and the action time of the optimal virtual voltage vector is obtained. The controls of the two subspaces are combined into one control period, and the switching sequence PWM signal is recalculated and applied to the rectifier.

[0010] Furthermore, the discrete predictive control model is:

[0011]

[0012] where i d (k + 1) is the predicted value of the d-axis current at the next moment, i q (k + 1) is the predicted value of the q-axis current at the next moment, i z1 (k + 1) is the predicted value of the z1-axis current at the next moment, i z2 (k + 1) is the predicted value of the z2-axis current at the next moment, i d (k) is the current value of the d-axis current, i q (k) is the current value of the q-axis current, i z1 (k) is the current value of the z1-axis current, i z2 (k) is the current value of the z2-axis current, T s is the sampling period, L d is the inductance of the dual three-phase permanent magnet synchronous generator on the d-axis, L q is the inductance of the dual three-phase permanent magnet synchronous generator on the q-axis, L z1 is the leakage inductance of the dual three-phase permanent magnet synchronous generator on the z1-axis, L z2 is the leakage inductance of the dual three-phase permanent magnet synchronous generator on the z2-axis, u d (k) is the current value of the d-axis voltage, u q (k) is the current value of the q-axis voltage, u z1 (k) is the current value of the z1-axis voltage, u z2 (k) is the current value of the z2-axis voltage, R s is the stator resistance of the dual three-phase permanent magnet synchronous generator, ω e is the electrical angular velocity of the dual three-phase permanent magnet synchronous generator, ψ f is the permanent magnet flux linkage of the dual three-phase permanent magnet synchronous generator.

[0013] Furthermore, u d (k) and u q (k) satisfy:

[0014]

[0015] where t i and t j are the action times of two effective voltage vectors in the αβ subspace, u di is the d-axis component of the effective voltage vector u i is the d-axis component of the effective voltage vector u dj is the d-axis component of the effective voltage vector u j is the d-axis component of the effective voltage vector u qi The effective voltage vector u i is the q-axis component of the effective voltage vector u qj The effective voltage vector u j is the q-axis component of the effective voltage vector u i And t d * -i d (k)](s qj -s q0 )+[i q * -i q (k)](s d0 -s dj )+T s (s q0 s dj -s qj s d0 )} / M, t j ={[i d * -i d (k)](s q0 -s qi )+[i q * -i q (k)](s di -s d0 )+T s (s qi s d0 -s q0 s di )} / M, i d * and i q * are the d-axis and q-axis reference currents respectively, s d0 is the d-axis current slope corresponding to the zero voltage vector and s d0 =[-R s i d (k)+ω e L q i q (k)] / L d , sq0 is the q-axis current slope corresponding to the zero voltage vector, and s q0 = [-R s i q (k) - ω e L d i d (k) - ω e ψ f / L q , s di 、s dj 、s qi 、s qj are the current slopes corresponding to u di 、u dj 、u qi 、u qj respectively, and s di = (s d0 + u di ) / L d 、s qi = (s q0 + u qi ) / L q 、s dj = (s d0 + u dj ) / L d 、s qj = (s q0 + u qj ) / L q , M is an intermediate quantity, and M = s q0 s dj + s qi s d0 + s qj s di - s qi s dj - s qj s d0 - s q0 s di ;

[0016] The u z1 (k) and u z2 (k) satisfy:

[0017]

[0018] where t’ i 、t’ j are the action times of two effective voltage vectors in the z1z2 subspace respectively, u z1i is the component of the effective voltage vector u i on the z1 axis, u z2i is the component of the effective voltage vector u iThe component on the z2 axis, u z1j is the effective voltage vector u j The component on the z1 axis, u z2j is the effective voltage vector u j The component on the z2 axis; and t’ i ={[i z1 * -i z1 (k)](s z2j -s z20 )+[i z2 * -i z2 (k)](s z10 -s z1j )+T s (s z20 s z1j -s z2j s z10 )} / M2, t’ j ={[i z1 * -i z1 (k)](s z20 -s z2i )+[i z2 * -i z2 (k)](s z1i -s z10 )+T s (s z2i s z10 -s z20 s z1i )} / M2, i z1 * and i z2 * are the reference currents on the z1 and z2 axes respectively, s z10 is the z1-axis current slope corresponding to the zero voltage vector, and s z10 =[-R s i z1 (k)] / L z1 , s z20 is the z2-axis current slope corresponding to the zero voltage vector, and s z20 =[-R s i z2 (k)] / L z2 , s z1i , s z2i , s z1j , s z2j are the current slopes corresponding to u z1i , u z2i , u z1j , u z2j respectively, and sz1i =(s z10 +u z1i ) / L z1 、s z2i =(s z20 +u z2i ) / L z2 、s z1j =(s z10 +u z1j ) / L z1 、s z2j =(s z20 +u z2j ) / L z2 , M2 is an intermediate quantity, and M2 = s z20 s z1j +s z2i s z10 +s z2j s z1i -s z2i s z1j -s z2j s z10 -s z20 s z1i .

[0019] Furthermore, the constructed value function is:

[0020]

[0021] Furthermore, the optimal virtual voltage vector sets of the αβ subspace and the z1z2 subspace are selected through the minimum value function, which is specifically implemented as follows: when the target vector is located in sector X, only the voltage vectors in the same half-sector as the target vector are selected for optimization by the prediction model.

[0022] Furthermore, the controls of the two subspaces are combined into one control period, specifically: the zero voltage vector in the single subspace control is equivalently replaced by the voltage vector in the other subspace.

[0023] Further, in the process of selecting the optimal virtual voltage vector sets of the αβ subspace and the z1z2 subspace through the minimum value function, when R s 、L d 、L q 、ψ f 、the leakage inductance L of the z1 axis z1 、the leakage inductance L of the z2 axis z2 are perturbed, the hyper-twisting integral sliding mode observer is used to observe the current perturbation and compensate the prediction control model; the parts of each current affected by parameter perturbation are represented by lumped perturbations, and the prediction control model is:

[0024]

[0025] Among them, f d , f q , f z1 , f z2 are the lumped disturbances of each item.

[0026] Furthermore, the structure of the spiral integral sliding mode observer is as follows:

[0027]

[0028] Among them, is the observed value of the d-axis current, is the observed value of the q-axis current, is the observed value of the z1-axis current, is the observed value of the z2-axis current, respectively represent the observed values of the lumped disturbances of the dq-axis and z1z2-axis, s e = [s d s q s z1 s z2 T is the sliding mode surface function matrix, V = [V d V q V z1 V z2 T is the sliding mode control function matrix, and λ and ω are the gains of the sliding mode observer.

[0029] Furthermore, in the process of selecting the optimal virtual voltage vector sets of the αβ subspace and the z1z2 subspace through the minimum value function, a composite controller is used to track the harmonics in the z1z2 subspace without static error, and the composite controller includes a quasi-proportional resonant controller and a repetitive controller.

[0030] Furthermore, the transfer function expression of the quasi-proportional resonant controller in the continuous domain is: The discrete form of the transfer function of the repetitive controller is: Among them, Q(z) is the stability coefficient, z m is the phase lead compensation, S(z) is the low-pass filter, K RC is the internal model gain of the repetitive controller, N is the number of single fundamental period samplings, and m is the number of compensation delay beats.

[0031] The beneficial effects of the present invention are:

[0032] ​​(1) The present invention synthesizes a virtual vector by using the outermost voltage vectors in the αβ subspace and the z1z2 subspace. The integral value of the synthesized virtual vector with respect to time in another space is smaller than that of the traditional synthesis method, which helps to reduce the interference of the αβ subspace vector on the z1z2 subspace. In addition, the multi-virtual vector synthesis method is adopted to achieve full-direction and full-amplitude voltage vector coverage and more accurately track the target voltage vector.

[0033] (2) The present invention uses a super-twisting integral anti-interference sensor to estimate the errors of the motor parameter perturbations and the rectifier non-linear factors, and substitutes the estimated errors into the prediction model for error compensation, so that the optimal vector obtained by the prediction model is more accurate.

[0034] (3) The present invention uses a composite controller to replace the traditional PI controller to more accurately track the harmonic current in the z1z2 subspace and improve the prediction accuracy of the prediction model.

[0035] (4) The present invention simplifies the optimization steps of the prediction model, reduces the number of optimization times required for the prediction model to obtain the optimal vector, from 24 times to 12 times, greatly reducing the computational amount and improving the system operation speed. Description of the Drawings

[0036] Figure 1 is the structural diagram of the dual three-phase permanent magnet synchronous generator of the present invention;

[0037] Figure 2(a) is the voltage vector distribution diagram of the αβ subspace of the six-phase voltage rectifier of the present invention;

[0038] Figure 2(b) is the voltage vector distribution diagram of the z1z2 subspace of the six-phase voltage rectifier of the present invention;

[0039] Figure 3(a) is the αβ subspace distribution diagram of the improved virtual voltage vector of the present invention;

[0040] Figure 3(b) is the z1z2 subspace distribution diagram of the improved virtual voltage vector of the present invention;

[0041] Figure 4 is the structural diagram of the composite controller of the present invention;

[0042] Figure 5 is the structural diagram of the repetitive controller of the present invention;

[0043] Figure 6(a) is the rated speed stator current waveform diagram obtained by the TV-MPCC1 method;

[0044] Figure 6(b) is the rated speed stator current waveform diagram obtained by the TV-MPCC2 method;

[0045] Figure 6(c) is the rated speed stator current waveform diagram obtained by the TV-MPCC3 method;

[0046] Figure 7 This is the strategy flowchart of the present invention. Specific embodiments

[0047] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but the protection scope of the present invention is not limited thereto.

[0048] The present invention considers a multi-vector model predictive control method for a dual three-phase permanent magnet synchronous generator in the harmonic space. First, the method of synthesizing virtual voltage vectors is improved. In each control sector, three voltage vectors are selected to construct the target voltage vector to achieve full-direction and full-amplitude voltage vector coverage. Secondly, considering the errors caused by motor parameter perturbations and rectifier nonlinear factors, a super-twisting integral disturbance rejection observer capable of real-time parameter estimation is proposed to reduce the influence of errors on the motor control performance. Finally, considering that the traditional PI controller cannot accurately track the harmonic currents of each order in the harmonic plane, a composite controller is proposed, which can track and suppress the harmonic currents of each order without static error, further suppressing the harmonic currents. The present invention further proposes a simplified optimization scheme, which significantly reduces the computational amount while not affecting the effective control of the two subspace components.

[0049] As Figure 7 shown, the present invention specifically adopts the following technical solutions:

[0050] Step 1: Use the space vector decoupling method VSD to derive the mathematical model and discrete predictive control model of the dual three-phase permanent magnet synchronous generator ( Figure 1 is the structural schematic diagram of the dual three-phase permanent magnet synchronous generator) in the rotating coordinate system;

[0051] Only the αβ subspace of the dual three-phase permanent magnet synchronous generator participates in energy conversion, and the z1z2 subspace and o1o2 subspace only generate harmonic losses; due to the connection method with isolated neutral points, the variables of each item in the o1o2 subspace are zero, so there is no need to consider the transformation equations of the o1o2 subspace in the natural coordinate system and the rotating coordinate system, and the voltage equation of the DTP-PMSG in the rotating coordinate system is obtained as:

[0052]

[0053] In the formula: u d , u q , u z1 , u z2 , i d , i q , i z1 , i z2 are the voltages and currents of the dual three-phase permanent magnet synchronous generator on the d, q, z1, and z2 axes respectively; L d , Lq are the inductances of the dual three-phase permanent magnet synchronous generator on the d-axis and q-axis respectively; L z1 is the leakage inductance of the dual three-phase permanent magnet synchronous generator on the z1-axis, L z2 is the leakage inductance of the dual three-phase permanent magnet synchronous generator on the z2-axis; R s is the stator resistance of the dual three-phase permanent magnet synchronous generator; ω e , ψ f are the electrical angular velocity and the permanent magnet flux linkage of the dual three-phase permanent magnet synchronous generator respectively.

[0054] After that, the forward Euler method is used to discretize the voltage equation of the motor (i.e., Equation (1)) to obtain the current prediction model at the next moment:

[0055]

[0056] In the formula, i d (k + 1), i q (k + 1), i z1 (k + 1), i z2 (k + 1) are the predicted values of the currents on the d-axis, q-axis, z1-axis, and z2-axis at the next moment respectively; i d (k), i q (k), i z1 (k), i z2 (k) are the current values on the d-axis, q-axis, z1-axis, and z2-axis at the current moment respectively; T s is the sampling period; u d (k), u q (k), u z1 (k), u z2 (k) are the voltage values on the d-axis, q-axis, z1-axis, and z2-axis at the current moment respectively.

[0057] Step 2: Improve the method of synthesizing the virtual voltage vector. In each control sector, select three voltage vectors to construct the target voltage vector to achieve full-direction and full-amplitude voltage vector coverage;

[0058] As shown in Figure 3(a), three voltage vectors on the outermost layer of the αβ subspace in Figure 2(a) are used to synthesize a new virtual voltage vector. Taking vv1 in Figure 3(a) as an example, this virtual voltage vector is composed of the voltage vectors v 45 , v 44 and v 64 in Figure 2(a). The action time of each voltage vector is calculated by the following formula:

[0059]

[0060] In the formula, ε 45 , ε 44 and ε 64 are v45 , v 44 and v 64 The acting time of, and the result is:

[0061]

[0062] The remaining 11 virtual voltage vectors (vv2 - vv12) in Fig. 3(a) are synthesized in the same way. The finally synthesized virtual voltage vector distribution diagram is shown in Fig. 3(a). The amplitude of the synthesized virtual voltage vector is 0.598U dc , U dc is the DC bus voltage.

[0063] The virtual voltage vector in the z1z2 subspace is used in the multi-vector model predictive control to achieve the control of the z1z2 subspace. Similar to the synthesis requirements of the virtual voltage vector in the αβ subspace, it is required that the mapping of the newly synthesized virtual voltage vector in the z1z2 subspace in the αβ subspace is zero. In the present invention, the three voltage vectors on the outermost layer of the z1z2 subspace in Fig. 2(b) are selected to synthesize the virtual voltage vector. Taking the vector vv'1 in Fig. 3(b) as an example, the voltage vectors v 43 , v 42 and v 52 are selected, and the acting time of each voltage vector is calculated by the following formula:

[0064]

[0065] In the formula, ε 43 , ε 42 and ε 52 are the acting times of v 43 , v 42 and v 52 respectively, and the result is:

[0066]

[0067] The remaining 11 virtual voltage vectors (vv'2 - vv'12) in Fig. 3(b) are synthesized in the same way. The finally synthesized virtual voltage vector distribution diagram is shown in Fig. 3(b).

[0068] Compared with the single-vector and double-vector MPCC control strategies, the multi-vector MPCC can cover all plane voltage vectors by using two adjacent virtual voltage vectors and a zero voltage vector in a space, and can realize the output of any amplitude and direction of the voltage vector, and theoretically achieve the optimal control performance in the dq space.

[0069] The multi-vector MPCC simultaneously considers the deadbeat control of the dq-axis current and calculates the slope of the dq-axis current corresponding to the voltage vector. After considering the delay compensation, the slope of the current at the k + 1 moment under the action of the voltage vector is:

[0070]

[0071] Wherein, s d0 and s q0 are the d-axis and q-axis current slopes corresponding to the zero voltage vector, u di and u dj and u qi and u qj correspond to the d-axis and q-axis components of two effective voltage vectors (u i and u j ), respectively. s di and s dj and s qi and s qj are the current slopes corresponding to u di and u dj and u qi and u qj , respectively.

[0072] Furthermore, according to the deadbeat control principle, the calculation formulas for the action times of two adjacent virtual voltage vectors and the zero voltage vector in the αβ subspace are as follows:

[0073]

[0074] Wherein, i d * and i q * are the d-axis and q-axis reference currents, t i and t j and t0 are the action times of two effective voltage vectors and the zero voltage vector in the αβ subspace, respectively, and M is an intermediate quantity.

[0075] Similarly, in the z1z2 subspace, the action times of two adjacent virtual voltage vectors and the zero voltage vector are obtained:

[0076]

[0077] Among them, i z1 * and i z2 * are the reference currents of the z1-axis and z2-axis, respectively, t’ i and t’ j and t’0 are the action times of two effective voltage vectors and the zero voltage vector in the z1z2 subspace, respectively, M2 is an intermediate quantity, s z10 and s z20 are the z1-axis and z2-axis current slopes corresponding to the zero voltage vector, u z1i and u z2i and u z1j and uz2j Corresponding to the components of two effective voltage vectors (u i , u j ) on the z1 and z2 axes, s z1i , s z2i , s z1j , s z2j are respectively the current slopes corresponding to u z1i , u z2i , u z1j , u z2j . L z1 , L z2 are respectively the inductances of the dual three-phase permanent magnet synchronous generator on the z1 and z2 axes.

[0078] Furthermore, the d- and q-axis components of the synthesized voltage vector (i.e., the target voltage vector) after modulation processing are obtained:

[0079]

[0080] Substitute the obtained synthesized voltage vector into the prediction model (Equation (2)) to obtain the predicted current at the next moment, and construct the value function shown below to obtain the optimal virtual voltage vector group and its action time:

[0081]

[0082]

[0083] Step 3: Considering the errors caused by motor parameter perturbations and rectifier nonlinear factors, a super-twisting integral anti-disturbance observer that can estimate parameters in real time is proposed to reduce the influence of errors on the motor control performance;

[0084] Considering the influence brought by parameter perturbations, improve the dq space control performance. According to the MPCC control principle, when the stator resistance R s , d-axis inductance L d , q-axis inductance L q , z1-axis leakage inductance L z1 , z2-axis leakage inductance L z2 and the permanent magnet flux linkage ψ f generate perturbations, they will affect the predicted current, which will lead to inaccurate selection of the optimal virtual voltage vector through the value function, and further affect the performance of the entire system. When the parameters in the model are perturbed, the current prediction model can be expressed as:

[0085]

[0086] In the formula, and They are the d-axis, q-axis, z1-axis, and z2-axis components of the predicted current when there are disturbances in the parameters; ΔR s , ΔL d , ΔL q , ΔL z1 , ΔL z2 , and Δψ f represent the errors between the actual parameters and the parameters in the prediction model.

[0087] To improve the performance of the model prediction against parameter disturbances, based on the model predictive current control, the present invention will use a super-twisting integral sliding mode observer to observe the current disturbances and compensate the prediction model to improve the robustness of the system. The parts of each current affected by parameter disturbances are represented by lumped disturbances. Therefore, the prediction model (Equation (14)) is transformed into:

[0088]

[0089] In the formula, f d , f q , f z1 , f z2 are the lumped disturbances of each item.

[0090] The super-twisting sliding mode observer (STISMO) is a high-order sliding mode observer that combines the super-twisting algorithm (STA) and the sliding mode observer (SMO). By introducing high-order sliding mode control technology, it significantly reduces the chattering phenomenon of the traditional sliding mode observer while maintaining strong robustness against system uncertainties and external disturbances. The present invention designs a super-twisting integral sliding mode observer according to Equation (15), and its structure is as follows:

[0091]

[0092] In the formula, respectively represent the observed values of the dq-axis and z1z2-axis currents, respectively represent the observed values of the lumped disturbances of the dq-axis and z1z2-axis. s e = [s d s q s z1 s z2 T is the sliding mode surface function matrix, V = [V d V q V z1 V z2 T is the sliding mode control function matrix, and λ and ω are the gains of the sliding mode observer.​​

[0093] To solve the problem of slow convergence rate of STISMO and reduce the steady-state error, an integral term of the observation error is added to the traditional linear sliding surface. Therefore, the sliding surface function adopted in the present invention is as follows:

[0094] s e = e + η∫edt (18)

[0095] In the formula, is the current observation error matrix, and the integral gain matrix η = diag([-β d , -β q , -β z1 , -β z2 ).

[0096] Through Lyapunov stability analysis, it can be obtained that the sliding mode gain needs to satisfy the following conditions:

[0097]

[0098] In the formula, δ is a constant greater than 0.

[0099] Step 4: Considering that the traditional PI controller cannot accurately track the harmonic currents of each order in the harmonic plane, a composite controller is proposed, which can track and suppress the harmonic currents of each order without static error;

[0100] Furthermore, considering the improvement of the z1z2 space control performance for harmonic suppression, for the control system of the ideal DTP-PMSG, based on the control method of VSD decoupling, the harmonic current in the z1z2 subspace is zero. However, due to the nonlinear factors in the rectifier, such as dead time, voltage drop of the switching tube, on-off time of the switch, etc., these factors will cause voltage distortion in the rectifier, resulting in the generation of harmonics in the z1z2 subspace. It can be seen from the following formula that the odd-order voltage harmonics caused by the nonlinear factors, mainly the 5th and 7th harmonics, all enter the z1z2 subspace after coordinate transformation.

[0101]

[0102] Among them, V z1n and V z2n are the voltage harmonics in the z1z2 subspace, ΔV dead is the dead-time voltage, V αn and V βn are the voltage harmonics in the αβ subspace.

[0103] Since the traditional PI controller cannot track AC signals without steady-state error, it cannot effectively suppress the 5th and 7th harmonics in the z1z2 subspace. Therefore, the present invention proposes a composite controller to track the harmonics in the z1z2 subspace without steady-state error.

[0104] The parallel composite controller proposed by the present invention consists of the following two parts: (1) The quasi-proportional-resonant controller (QPR) improves the dynamic performance of the system; (2) The repetitive controller (RC) improves the static performance of the system and suppresses high-order harmonics. As Figure 4 shown.

[0105] The transfer function expression of the QPR controller is:

[0106]

[0107] In the formula, K p is the proportional coefficient; K r is the resonant term coefficient; ω c is the cut-off frequency; ω0 is the fundamental angular frequency, and in the present invention, ω0 = 314 rad / s.

[0108] The proportional coefficient K p affects the gain of the QPR controller. The increase of K p can improve the gain of the controller in the entire frequency band except at the resonant point and enhance the anti-interference ability of the system; however, an excessive proportional coefficient may cause the system to oscillate and destroy the system stability. The present invention selects the proportional coefficient K p = 0.122.

[0109] The resonant coefficient K r only affects the gain of the controller at the resonant point. The increase of K r can improve the gain of the controller at the resonant point and eliminate the static error; however, an overly amplified resonant coefficient will cause the frequency band range of the controller to become larger, thereby amplifying other harmonic components and affecting the system stability. The present invention selects the resonant coefficient K r = 50.

[0110] The cut-off frequency ω c affects the bandwidth of the controller. The increase of ω c can increase the bandwidth of the controller at the resonant point without changing the gain at the peak of the system; however, ω c cannot be overly amplified. After the bandwidth of the controller at the resonant point increases, it may couple with other resonant points and affect the system stability. Considering the actual experiment, the present invention selects ω c = 2πΔf = 3.14 rad / s.

[0111] In summary, the transfer function expression of the QPR controller in the continuous domain is:

[0112]

[0113] Considering the digital control of the experiment, the transfer function is discretized by bilinear transformation to obtain the discrete-domain form expression:

[0114]

[0115] Repetitive control is based on the internal model principle. A delay recording link is added to the control path, and the error is recorded and fed back to the error quantity of the next cycle for superposition control in each control cycle. Through periodic accumulation until the tracking error decays to zero, static error-free control of the system is achieved. In actual engineering, it is necessary to introduce a stability coefficient Q(z), a phase lead compensation z m , a low-pass filter S(z), and an RC internal model gain K RC etc. The actual control structure is as shown in Figure 5 . The discrete form of the transfer function of the repetitive controller is:

[0116]

[0117] In the formula, N is the number of single fundamental wave period samplings. In the present invention, N = f s / f0 = 200; m is the number of compensation delay beats. In the present invention, m = 5. In order to ensure better stability and faster dynamic characteristics of the system, K RC is taken as 1.

[0118] Under ideal conditions, the open-loop poles of the internal model of the RC controller are all on the unit circle, which will lead to the system being in a critical state and greatly affecting the system stability. Therefore, generally a digital filter or a constant stability coefficient less than 1 is added to the internal model. In the present invention, a zero-phase shift low-pass filter Q(z) is selected:

[0119]

[0120] The introduction of Q(z) cannot completely achieve anti-interference for high-frequency signals. It is necessary to introduce a low-pass filter S(s) to further attenuate the high-frequency band gain of the system. Generally, S(s) uses a second-order low-pass filter, and the expression in the continuous domain is:

[0121]

[0122] In the formula, ω n is the cut-off frequency. In the present invention, harmonics below the 13th order are considered low-order harmonics. Therefore, ω n = 4084 rad / s, and ζ is the damping coefficient, generally taken as 0.707.

[0123] After discretization processing by the zero-order hold, the expression of the discrete-domain S(z) is obtained:

[0124]

[0125] Step 5: Combine the simplified optimization strategy, select the optimal virtual voltage vector sets of the αβ subspace and the z1z2 subspace through the minimum value function, obtain the action time of the optimal virtual voltage vector, merge the controls of the two subspaces into one control period, recalculate the switching sequence PWM signal, and apply it to the rectifier.

[0126] Considering the voltage vectors in the harmonic plane, the optimal virtual voltage vector and duty cycle are obtained through traversal optimization of the value function. It is necessary to calculate the prediction model 24 times, which is twice the computational amount of the traditional virtual vector-based MPCC. To address the problem of large computational amount, the present invention proposes a simplified optimization algorithm. As shown in Fig. 3(a), when the target vector is in sector I, considering that the vector action time cannot be negative, only the vectors in the same half-sector as the target vector are selected for optimization using the prediction model; for example: when the target vector is in sector I, the vectors in the same half-sector as the target vector are vv3, vv2, vv1, vv12, vv11, and vv 10 , and the number of predictions is reduced to 6 times. The prediction vector control sets for the remaining sectors and the z1z2 subspace can be derived analogously.

[0127] Finally, simplify the optimization algorithm. Consider the action times of the voltage vectors in the two spaces obtained in vector synthesis. Considering that the mutual mapping of the virtual voltage vector control sets used in the two subspaces is zero, the zero voltage vector in the single-subspace control can be equivalently replaced by the voltage vector in the other subspace, and thus the controls in the two subspaces can be merged into one control period. To ensure that the active power output of the system is given priority, the action times of the voltage vectors in the harmonic space are constrained. The action time constraint conditions for the voltage vectors in the two subspaces are:

[0128]

[0129] where, t i、j(dq) represents the action times of the two voltage vectors on the dq axis, and t’ i、j(z1z2) represents the action times of the two voltage vectors on the z1z2 axis.

[0130] The stator current harmonics analysis is carried out by comparing three algorithms: the multi-virtual voltage vector model predictive current control without considering the harmonic space (TV-MPCC1), the multi-virtual voltage vector model predictive current control considering the harmonic space (TV-MPCC2), and the multi-virtual vector model predictive current control based on the observer composite controller and considering the harmonic space (TV-MPCC3) proposed by the present invention, as shown in Figures 6(a), (b), and (c) respectively. It can be seen from the figures that the distortion rate of the stator current using the TV-MPCC1 strategy is 14.53%, the distortion rate of the stator current using the TV-MPCC2 strategy is 7.78%, and the distortion rate of the stator current using the TV-MPCC3 strategy is 5.82%; the amplitude difference of the z1z2-axis current using the TV-MPCC3 strategy is 0.42 A, which is significantly smaller than that using the TV-MPCC1 strategy and the TV-MPCC2 strategy. The steady-state performance of the strategy proposed by the present invention is significantly better than the other two control strategies, verifying the superiority of the control strategy proposed by the present invention.

[0131] The described embodiments are the preferred embodiments of the present invention, but the present invention is not limited to the above embodiments. Without departing from the substance of the present invention, any obvious improvements, substitutions or modifications that can be made by those skilled in the art all fall within the protection scope of the present invention.

Claims

1. A multi-vector model predictive control method for a dual-three-phase permanent magnet synchronous generator considering the harmonic space, characterized in that: Determine the discrete predictive control model of the dual-three-phase permanent magnet synchronous generator in the rotating coordinate system; In each control sector, select three voltage vectors to construct the target voltage vector, substitute it into the discrete predictive control model to obtain the predicted current at the next moment, and further construct a cost function to obtain the optimal virtual voltage vector and its action time; Select the optimal virtual voltage vector sets in the αβ subspace and the z1z2 subspace through the minimum cost function, obtain the action time of the optimal virtual voltage vector, merge the controls of the two subspaces into one control period, recalculate the switching sequence PWM signal, and apply it to the rectifier.

2. The multi-vector model predictive control method for a dual-three-phase permanent magnet synchronous generator according to claim 1, wherein The discrete predictive control model is: where i d (k + 1) is the predicted value of the d-axis current at the next moment, i q (k + 1) is the predicted value of the q-axis current at the next moment, i z1 (k + 1) is the predicted value of the z1-axis current at the next moment, i z2 (k + 1) is the predicted value of the z2-axis current at the next moment, i d (k) is the current value of the d-axis current, i q (k) is the current value of the q-axis current, i z1 (k) is the current value of the z1-axis current, i z2 (k) is the current value of the z2-axis current, T s is the sampling period, L d is the inductance of the dual three-phase permanent magnet synchronous generator on the d-axis, L q is the inductance of the dual three-phase permanent magnet synchronous generator on the q-axis, L z1 is the leakage inductance of the dual three-phase permanent magnet synchronous generator on the z1-axis, L z2 is the leakage inductance of the dual three-phase permanent magnet synchronous generator on the z2-axis, u d (k) is the current value of the d-axis voltage, u q (k) is the current value of the q-axis voltage, u z1 (k) is the current value of the z1-axis voltage, u z2 (k) is the current value of the z2-axis voltage, R s is the stator resistance of the dual three-phase permanent magnet synchronous generator, ω e is the electrical angular velocity of the dual three-phase permanent magnet synchronous generator, ψ f is the permanent magnet flux linkage of the dual three-phase permanent magnet synchronous generator.

3. The multi-vector model predictive control method for a dual-three-phase permanent magnet synchronous generator according to claim 2, characterized in that The said u d (k) and u q (k) satisfy: where t i and t j are the action times of two effective voltage vectors in the αβ subspace, u di is the d-axis component of the effective voltage vector u i , u dj is the d-axis component of the effective voltage vector u j , u qi is the q-axis component of the effective voltage vector u i , u qj is the q-axis component of the effective voltage vector u j ; and t i = {[i d * - i d (k)](s qj - s q0 ) + [i q * - i q (k)](s d0 - s dj ) + T s (s q0 s dj - s qj s d0 )} / M, t j = {[i d * - i d (k)](s q0 - s qi ) + [i q * - i q (k)](s di - s d0 ) + T s (s qi s d0 - s q0 s di )} / M, i d * and i q * are the d-axis and q-axis reference currents respectively, s d0 is the d-axis current slope corresponding to the zero voltage vector and s d0 = [-R s i d (k) + ω e L q i q (k)] / L d , s q0 is the q-axis current slope corresponding to the zero voltage vector, and s q0 = [-R s i q (k)-ω e L d i d (k)-ω e ψ f / L q ,s di 、s dj 、s qi 、s qj are respectively the current slopes corresponding to u di 、u dj 、u qi 、u qj , and s di =(s d0 +u di ) / L d 、s qi =(s q0 +u qi ) / L q 、s dj =(s d0 +u dj ) / L d 、s qj =(s q0 +u qj ) / L q , M is an intermediate quantity, and M = s q0 s dj +s qi s d0 +s qj s di -s qi s dj -s qj s d0 -s q0 s di ; The said u z1 (k) and u z2 (k) satisfy: where t’ i and t’ j are the action times of two effective voltage vectors in the z1z2 subspace, u z1i is the component of the effective voltage vector u i on the z1 axis, u z2i is the component of the effective voltage vector u i on the z2 axis, u z1j is the component of the effective voltage vector u j on the z1 axis, u z2j is the component of the effective voltage vector u j on the z2 axis; and t’ i = {[i z1 * - i z1 (k)](s z2j - s z20 ) + [i z2 * - i z2 (k)](s z10 - s z1j ) + T s (s z20 s z1j - s z2j s z10 )} / M2, t’ j = {[i z1 * - i z1 (k)](s z20 - s z2i ) + [i z2 * - i z2 (k)](s z1i - s z10 ) + T s (s z2i s z10 - s z20 s z1i )} / M2, i z1 * and i z2 * are the reference currents of the z1 and z2 axes respectively, s z10 is the z1-axis current slope corresponding to the zero voltage vector, and s z10 = [-R s i z1 (k)] / L z1 , s z20 is the z2-axis current slope corresponding to the zero voltage vector, and s z20 = [-R s i z2 (k)] / L z2 , s z1i , s z2i , s z1j , s z2j are respectively the current slopes corresponding to u z1i , u z2i , u z1j , u z2j , and s z1i = (s z10 + u z1i ) / L z1 , s z2i = (s z20 + u z2i ) / L z2 , s z1j = (s z10 + u z1j ) / L z1 , s z2j = (s z20 + u z2j ) / L z2 , M2 is an intermediate quantity, and M2 = s z20 s z1j + s z2i s z10 + s z2j s z1i - s z2i s z1j - s z2j s z10 - s z20 s z1i .

4. The multi-vector model predictive control method for a dual-three-phase permanent magnet synchronous generator according to claim 3, wherein The constructed cost function is:

5. The multi-vector model predictive control method for a dual-three-phase permanent magnet synchronous generator according to claim 4, wherein Select the optimal virtual voltage vector sets in the αβ subspace and the z1z2 subspace through the minimum cost function, which is specifically realized by the following method: when the target vector is in sector X, only select the voltage vectors in the same half-sector as the target vector for optimization using the prediction model.

6. The multi-vector model predictive control method for a dual-three-phase permanent magnet synchronous generator according to claim 5, wherein Merge the controls of the two subspaces into one control period, specifically: the zero voltage vector in the single-subspace control is equivalently replaced by the voltage vector in the other subspace.

7. The multi-vector model predictive control method for a dual-three-phase permanent magnet synchronous generator according to claim 2, wherein In the process of selecting the optimal virtual voltage vector sets of the αβ subspace and the z1z2 subspace through the minimum value function, when R s 、L d 、L q 、ψ f 、the leakage inductance L of the z1 axis z1 、the leakage inductance L of the z2 axis z2 are disturbed, a super-twisting integral sliding mode observer is used to observe the current disturbance and compensate the predictive control model; the parts of each current affected by parameter disturbances are represented by lumped disturbances, and the predictive control model is: Among them, f d , f q , f z1 , f z2 are the lumped disturbances of each item.

8. The multi-vector model predictive control method for a dual-three-phase permanent magnet synchronous generator according to claim 7, wherein The structure of the spiral integral sliding mode observer is: Among them, is the observed value of the d-axis current, is the observed value of the q-axis current, is the observed value of the z1-axis current, is the observed value of the z2-axis current, respectively represent the observed values of the lumped disturbances on the dq-axis and z1z2-axis, s e = [s d s q s z1 s z2 T is the sliding mode surface function matrix, V = [V d V q V z1 V z2 T is the sliding mode control function matrix, and λ, ω are the gains of the sliding mode observer.​​ 9. The multi-vector model predictive control method for a dual-three-phase permanent magnet synchronous generator according to claim 2, characterized in that In the process of selecting the optimal virtual voltage vector sets in the αβ subspace and the z1z2 subspace through the minimum cost function, use a composite controller to perform static-error-free tracking on the harmonics in the z1z2 subspace, and the composite controller includes a quasi-proportional-resonant controller and a repetitive controller.

10. The multi-vector model predictive control method for a dual-three-phase permanent magnet synchronous generator according to claim 9, characterized in that, The transfer function expression of the quasi-proportional resonant controller in the continuous domain is as follows: The discrete form of the transfer function of the repetitive controller is as follows: Among them, Q(z) is the stability coefficient, z m is the phase lead compensation, S(z) is the low-pass filter, K RC is the internal model gain of the repetitive controller, N is the number of samples per fundamental wave period, and m is the number of compensation delay beats.

Images