An Improved MPC-Based SVPWM Mechanism for NPC Three-Level Z-Source Converters

Model predictive control (MPC) method has been widely used to reduce the computational complexity of the traditional space vector pulse width modulation (SVPWM). However, for a neutral-point clamped three-level Z-source converter, the performance of the normal MPC strategy would highly depend on the computation processing rate because of the multiple times optimization calculation. In this paper, an improved MPC strategy has been developed, with a voltage prediction being designed to replace the current prediction, the calculation of the roll optimization could be effectively simplified significantly, and then the digital execution efficiency would be improved. Besides, in order to obtain a fixed output harmonic frequency, a combination of this improved MPC and SVPWM has been studied and the shoot-through state insertion for the Z-source also has been analyzed in detail. Lastly, comparison experiments have been carried out to make verification of this improved modulation mechanism.


Introduction
NPC three-level converters have significant advantages, e.g., smaller device pressure, lower dv/dt, and better harmonic distortion in the output [1,2]. With a Z-source network, this NPC three-level converters would have a wide range of adjustable voltage, some allowed shootthrough states and better harmonic distortion in the output, and have been widely used in the photovoltaic(PV) grid-connecting system and the ac drive field [3][4][5][6][7][8]. For example, a switched quasi-Z-source DC-DC inverter has been used for PV Systems [9], and a novel Z-source threelevel four-leg inverter has been researched to reduce the leakage current for three-phase PV inverters [10]. Nevertheless, the control difficulty, e.g., modulation method and stability assurance, would also increase because of the existing of Z-source network.
In terms of modulation, many improved algorithms have been proposed, e.g., in [6], the space vectors pulse width modulation along with the shoot-through insertions were introduced to the single Z-source NPC three-level inverters. A pulse width modulation (PWM) based on the stage-space averaging model was researched in [11] for a novel quasi-Z-source inverter topology. In [12], a new modulation strategy was studied in detail to reduce the leakage current for the Z-source four-leg inverters. In [13], a hybrid switching method with the combination of the PWM and pulse amplitude modulation (PAM) was proposed in details, which might result in a complicated computation and a high switching frequency and a switching loss reduction method based on a modified space vector modulation strategy [14].
In the past few years, the finite control set model predictive control (FCS-MPC) has been widely used in the modern power electrical areas for its fast response and being suitable for the multivariable and nonlinear systems [1,15,16]. In [17], the high performance MPC method has been used for the quasi-impedance source inverter and a FCS-MPC method was studied in [18]for the quasi-Z-source fourleg inverters with the consideration of an unbalanced load condition. In [19,20], model predictive control based on the discrete models were also presented. In [21], the MPC strategy with a ripple power compensation structure was presented for the Z-source converters, and the corresponding output harmonic distortions were random which were unsuitable for the filtering.
Unfortunately, when the MPC method was used for the NPC three-level converters, the control performance could be influenced because of the excessive optimization calculations, and for these NPC three-level converters, which combined with the Z-source network, how to make insertions for the shoot-through states is also one of the difficulties. In [22], a modified MPC method by using a novel voltage prediction has been studied for the grid-connected T-type inverter, which could improve the execution efficiency.
In order to solve the above problems and develop an efficient modulation approach for the NPC three-level ZSC to simplify the control complexity, an improved MPC-based SVPWM mechanism has been presented in this paper. In detail, a voltage prediction was designed instead of the normal current prediction, and the calculation of the roll optimization could be simplified significantly although the optimization times are the same. Besides, in order to obtain a relative fixed output harmonic frequency, the optimized voltage vector resulted from the improved MPC method then was generated by the basic vectors of the normal SVPWM and the insertions of the shoot-through states also have been studied in detail.
is study is organized into five sections. Following the introduction, the operation principle of the NPC three-level ZSC is explained in Section 2. In Section 3, this improved MPC-based SVPWM mechanism has been introduced in detail with the comparison to the traditional MPC method. Section 4 illustrates the experimental results with the performance comparisons between the traditional SVPWM strategy. Finally, some conclusions are drawn in Section 5.

Operation Principle of the NPC Three-Level Z-Source Converters
e NPC three-level ZSC researched in this paper is shown in Figure 1, containing two independent DC-power supplies. e connection point O of supplies is also regarded as the zero potential linking with the clamped point.
If the three up switches turn on (e.g., SA 1 , SA 2 , and SA 3 ), it is called upper shoot-through (UST) status being denoted as status "U." On the contrary, if the three down switches turn on, it is called down shoot-through status (DST), which is represented as status "D." e different states of this topology is summarized in Table 1.
Hypothesis C 1 � C 2 and L 1 � L 2 (means V C1 � V C2 � V C and V L1 � V L2 � V L ). All the different states described in Table 1 can be summarized as three kinds of working conditions: nonshoot-through, UST, and DST. ey are shown in Figures 2(a)-2(c), respectively. From Figure 2(a), several equations can be derived as follows: Similarly, the equations derived from Figures 2(b) and 2(c) are as follows: Here, the time duration of UST status is denoted as T sh_U and T sh_D for DST status. Normally, T sh_U should be equal to T sh_D in order to reduce the output voltage harmonic components, which means At the steady state, the average voltages across L 1 and L 2 keep zero over the sampling period T S , and the following equation could be derived from equations (1)-(3): where D � (T sh /T S ) is defined as the duty ratio of the shootthrough states. According to equations (1) and (6), for the nonshootthrough state, the output voltage V i could be obtained as For the Up or Down shoot-through states, the output voltage V i could be obtained according to equations (2), (3), and (6): Now, the three output voltages could also be represented as And the output phase voltage U x could be derived as where M is the modulation coefficient and H � 1/(1-2D) represents the booster ratio.

Traditional MPC.
e mathematical model of the NPC three-level ZSC in αβ coordinates can be given as where R and L are the load resistance and inductance, respectively. By using the Euler approximation for a sampling time T s (shown in equation (12)), equation (10) is discretized as in equation (13): where i α,β is the output current in αβ coordinates, u α,β stands for the output voltage also in αβ coordinates, which is determined by the 27 switching states of the NPC three-level ZSC, and k is the kth sampling period. From this mathematical model, the predictive output current at (k + 1)th could be derived as For the traditional MPC strategy, a roll optimization of objective function could be set, as shown in equation (15), for the current trajectory control: where i * α (k + 1) and i * β (k + 1) are the given current components, respectively, and i P α (k + 1) and i P β (k + 1) are the predictive current components from equation (14).
Obviously, for this traditional method, 27 cycles of optimization should be conducted to obtain the smallest g and the corresponding optimal switching state, in which equation (13) need to be also calculated 27 times in order to obtain the current components with different input voltages, which significantly increases the computation consumption. e implementation flow chart of the traditional MPC strategy is shown in Figure 3.

Improved MPC.
For equation (14), the input voltage can be represented as Assume that the predictive output current in equation (15) equals the given current components in ideal condition, which means Figure 3: Implementation flow chart of the traditional MPC.
With the combination of equations (16) and (17), the predictive voltage components u p α,β (k) could be obtained as where superscript "p" means the expected value and i * α,β (k + 1) is the reference current at (k + 1)th sampling, while the i α,β (k) is the actual current at kth sampling.
For the improved MPC strategy, the roll optimization of objective function was set as In order to make compensation of the time delay, equations (18) and (19) are shifted one step forward. us, the predictive voltage components and the objective function at (k + 1)th could be derived as equations (20) and (21): e implementation flow chart of this improved MPC strategy then could be derived, as shown in Figure 4.
It is obvious that, for the improved MPC, the current calculation within the roll optimization has been omitted, which could improve the execution efficiency to a great degree.
On the other aspect, for the prediction accuracy, the predictive voltage error could be derived with equations (16) and (18): From (22), it can be seen that the voltage tracking performance is consistent with the current, which provides the validation of the proposed improved MPC strategy.

Insertion of the Shoot-rough States.
For both the traditional and improved MPC strategies, their output frequency is a varying variable, which is unsuitable for the filtering design. Here, the optimized voltage u opt (k) (the corresponding switch state is S opt ) has been implemented by the basic space voltage vectors and based on which the shoot-through states of the Z-source are inserted to ensure the boost performance. e MPC-based SVPWM mechanism is shown in Figure 5.

Mathematical Problems in Engineering
Shoot-trough states' insertion not only determines the boost performance but also affects the output harmonic distortion and the switch losses. For the space voltage vectors shown in Figure 6(a), it is being taken as an example that the obtained optimized voltage u opt locates in Zone I and triangle 3, as shown in Figure 6(b).
Assume that the initial switch state is "0 −1 −1," which means SA2, SA3, SB3, SB4, SC3, and SC4 are turned ON: (1) At t � t1, state "0 −1 −1" should be switched to "1 −1 −1" (for phase A, SA is changed from "0" to "1"), and the UST could be inserted in phase A by previously turning on the switch "SA1," as shown in Figure 7 (SA1, SA2, and SA3 are ON while SA4 are OFF). During this insertion, the switch states for phases B and C remain unchanged, which satisfy the "voltsecond" principle. (2) At t � t2, state "1 −1 −1" should be switched to "1 0 −1" (for phase B, SB is changed from "−1" to "0"). If the UST is previously inserted in phase B, the switch state of phase A could be kept as "1," while the switch state of phase C should be clamped from "−1" to "0," which means the "volt-second" state would be destroyed so this insertion should be omitted. (3) At t � t3, state "1 0−1" should be switched to "1 0 0" (for phase C, SC is changed from "−1" to "0"). e DST could be inserted in phase C by persistently turning on the switch "SC4," as shown in Figure 7 (SC2, SC3, and SC4 are ON while SC1 are OFF). During this insertion, the switch states for phases A and B remain unchanged, which satisfy the "volt-second" principle.
Obviously, when the reference vector is located in Zone I and triangle 3, the UST and DST can be inserted in the interaction range of the equivalent zero vector. us, the condition of T sh_U � T sh_D can ensure the balance of the UST status and the DST status.

Experimental Results
In this paper, the effectiveness of the proposed improved MPC-based SVPWM mechanism is verified via an established experimental platform, as shown in Figure 8, along with the deep comparisons with the traditional SVPWM method. e detailed parameters are listed in Table 2.
e control algorithm is implemented on a TMS320F28335 DSP with floating-point arithmetic.
Firstly, when the input voltage V dc changes from 100 V to 200 V (V dc1 � V dc2 change from 50 V to 100 V), the corresponding dynamical line and phase voltages of the improved strategy are shown in Figure 9.
However, because of the voltage prediction error shown in Figure 10, there is some voltage ripple existing in the DClink voltage.
When the duty ratio D changes from 0 to 0.3, which means the output voltage of the Z-source topology V i could be 2.5 times as much as the input voltage V dc (according to equation (7)), the dynamical output line and phase voltages u ab and u a from the traditional SVPWM method and the improved proposed MPC-based Mathematical Problems in Engineering SVPWM mechanism are shown in Figures 11 and 12, respectively.
Correspondingly, the output voltage V i and the capacitors' voltages V c1 and V c2 are shown in Figures 13 and 14. It is obvious that V i changes from 100 V to about 250 V and the capacitors' voltages are basically guaranteed to be in equilibrium.
For the output current, the dynamic comparisons are shown in Figures 15 and 16 along with the harmonic distortion analysis by taking a-phase current as an example.
From Figures 15 and 16, it can be seen that the harmonic distortion changes from 3.69% to 2.11% for the traditional and improved methods without the booster function (D � 0), while from 4.20% to 3.53% with the insertion of the shootthrough states (D � 0.3), which could make verification of the improved mechanism studied in this paper. Besides, with the improved MPC-based SVPWM mechanism, the harmonic frequency is a relative fixed value, which is suitable for the filter designing. Besides, after the dynamic change, the restability time is about half an cycle for this improved MPCbased SVPWM, which is faster than the traditional SVPWM method (about one cycle).
In order to exhibit the superior property of the improved MPC-based SVPWM mechanism. e execution

Parameters
Value   Table 3, where the execution time is calculated by averaging ten-time measured results when D � 0 and D � 0.3 and the sampling frequency is 10 kHz. For the normal MPC-based SVPWM method, it has 10.03% and 13.91% time savings than the traditional SVPWM for the complicated trigonometric calculations has been omitted. Similarly, for the calculation reduction in the roll optimization, the improved scheme has over 26.25% and 19.25% time saving improvements than the traditional SVPWM and the normal MPC-based SVPWM when D � 0, while 28.78% and 17.27% when D � 0.3.

Conclusions
In this paper an improved MPC-based SVPWM mechanism has been proposed for the NPC three-level ZSC. Firstly, in order to improve the execution efficiency, a novel voltage prediction has been derived based on the mathematical models to replace the normal current prediction, by which the optimization could be effectively simplified along with a relative nice performance. Besides, with the combination of the improved MPC and SVPWM, the harmonic frequency of the output current could be kept relatively fixed to be suitable for the filtering design and the shoot-through states insertions have been also studied to realize the voltage boosting. Lastly, comparison experiments have been carried out to make verification of this improved MPC-based SVPWM mechanism. In particular, this improved method has a similar static performance to the traditional SVPWM with a better dynamic performance, while for the execution efficiency, it has over 26.25% and 19.25% time saving improvements than the traditional SVPWM and the normal MPC-based SVPWM when D � 0, while 28.78% and 17.27% when D � 0.3, which is suitable for the digital implementation.
Data Availability e processed data required to reproduce these findings cannot be shared at this time as the data also forms part of an ongoing study.

Conflicts of Interest
e authors declare that they have no conflicts of interest.