Full control set model prediction control method for three-level converter

Abstract

The invention relates to the technical field of three-phase three-level NPC converter control, in particular to a three-level converter full control set model prediction control method, which comprises the following steps: constructing a first prediction model of a three-level converter based on an electrical physical quantity; equally dividing each control period into n identical state intervals to form a full control set; converting the full control set after the redundant vector is removed into an alpha-beta coordinate system; obtaining a second prediction model under the alpha-beta coordinate system under the action of the full control set; constructing a midpoint potential difference prediction model; establishing a cost function by using the two prediction models; and carrying out hierarchical optimization on the cost function, removing redundant vectors from the full control set, carrying out cost function calculation, and selecting a state vector which enables the cost function to be minimum from the full control set as an optimal vector of a corresponding control period. According to the three-level converter neutral-point potential balancing method, a hierarchical optimization method is utilized, the neutral-point potential balancing problem of the three-level converter is solved, meanwhile, the tedious weight factor design step is eliminated, and the calculation burden is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of three-phase three-level NPC converter control, and in particular to a three-level converter full control set model predictive control method. Background Art

[0002] Three-phase three-level NPC converters are widely used in various power conversion fields. The control performance of the converter is directly related to the quality of the transmitted power. For many years, the control strategy research of three-level converters has been the focus of scientific researchers. A large number of studies and experiments are aimed at finding high-performance methods to improve the control of three-level converters.

[0003] Among many control strategies, model predictive control is widely used in power electronic converters due to its advantages such as simple design and fast dynamic response speed. However, the traditional finite set model predictive control (FCS-MPC) has problems such as poor steady-state performance and unstable switching frequency due to its incomplete control set definition. Although its steady-state performance can be improved by increasing the sampling frequency, the sampling time is limited by the amount of calculation in the algorithm, and it cannot solve the root problem of incomplete control set.

[0004] In addition, the midpoint voltage imbalance is a critical point for three-level converters. Although the multi-objective optimization capability of FCS-MPC can alleviate this problem to a certain extent, the cost function has to be reconstructed in a way that adds the midpoint voltage constraint to achieve the goal of high-quality output current and midpoint voltage balance. However, it is difficult for FCS-MPC to design the weighting coefficients in the cost function when considering both the output current quality and the midpoint voltage constraints, especially when the parameters or control conditions change.

[0005] Therefore, improving the control set and introducing new optimization methods are of great significance to solving the above-mentioned problems. Summary of the invention

[0006] In view of the above problems, the purpose of the present invention is to provide a model predictive control method for the full control set of a three-level converter. By dividing a control cycle into n identical state intervals, the traditional finite control set is expanded to achieve better control performance; at the same time, a hierarchical optimization method is introduced to achieve midpoint potential balance and reduce computational complexity.

[0007] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is specifically as follows:

[0008] A three-level converter full control set model predictive control method comprises the following steps:

[0009] Step 1, using sensors to collect electrical physical quantities, constructing a first prediction model of a three-level converter based on the electrical physical quantities, converting the first prediction model to an αβ coordinate system, and performing delay compensation; wherein the electrical physical quantities include a grid-side current, a grid voltage, a DC-side voltage, and upper and lower capacitor voltages on the DC-side at the current moment;

[0010] Step 2, divide each control cycle into n equal state intervals, define the switching function according to the n state intervals, and form a full control set;

[0011] Step 3, convert the full control set after removing redundant vectors into the αβ coordinate system;

[0012] Step 4, bringing the converted control set into the first prediction model to obtain a second prediction model in the αβ coordinate system under the action of the full control set;

[0013] Step 5: There is a midpoint potential imbalance in the three-level converter. Based on the relationship between capacitance and switch state, a prediction model for the midpoint potential difference is constructed.

[0014] Step 6: Establish a cost function using the second prediction model and the prediction model of the midpoint potential difference;

[0015] Step 7, using the hierarchical optimization method, convert the cost function into two parts (current term and midpoint potential difference term), remove redundant vectors from the full control set, calculate the cost function, select the state vector that minimizes the cost function from the full control set as the optimal vector for the corresponding control cycle, and perform coordinate transformation, and use the transformed optimal vector to control the switching state of the three-level converter.

[0016] Furthermore, in step 1, the first electrical prediction model is transformed into the αβ coordinate system by Clark transformation, and the model expression is obtained as follows:

[0017]

[0018] Where, L is the filter inductance, i αβ is the grid-side current in the stationary αβ coordinate system, V gαβ is the grid voltage in the stationary αβ coordinate system, R is the line load, V dc is the DC side voltage, S αβ , S α , S β represents the switching function in the stationary αβ coordinate system, S a , S b , S c represents the switching function in the three-phase coordinate system;

[0019] The load current derivative di / dt is approximated by the forward Euler method, and di / dt is brought into the model for delay compensation to obtain:

[0020]

[0021] Among them, i αβ (k+1) represents the grid-side current at time k+1 in the stationary αβ coordinate system, V gαβ (k+1) represents the grid voltage at time k+1 in the stationary αβ coordinate system, V dc (k+1) represents the DC side voltage at time k+1, S αβ (k+1) represents the switching function at time k+1 in the stationary αβ coordinate system.

[0022] Furthermore, the implementation process of step 2 includes:

[0023] At time k, the prediction interval at time k+1 is divided into n parts, and the proportion of high level is set to m, 0≤m≤n. It is defined that the switch tube is turned on during some continuous intervals in each control cycle and remains off during the remaining intervals. The full control set after division is Contains a combination of n+1 switch states; S a , S b , S c is the switching function in the three-phase coordinate system.

[0024] Further, step 3 includes the following steps:

[0025] Combining each parameter in the full control set, for a three-phase three-level converter, we can get (n+1) 3 There are redundant state vectors in these state vectors. After removing the redundant state vectors, we can get 3n 2 +3n+1 kinds of state vectors; Clark transform the non-redundant state vector to the αβ coordinate system to obtain the switching function S in the αβ coordinate system α , S β :

[0026]

[0027] Furthermore, the expression of the second prediction model obtained in step 4 is:

[0028]

[0029] Furthermore, constructing a prediction model for the midpoint potential difference in step 5 comprises the following steps:

[0030] First, write the expression of the capacitor voltage dynamic process, and approximate the capacitor voltage derivative du / dt in the expression by the forward Euler method to obtain:

[0031]

[0032] Among them, u c1 (k),u c2 (k) represents the voltage of the upper and lower capacitors on the DC side at time k, C1 and C2 are the capacitances of the DC side capacitors, and i c1 (k), i c2 (k) represents the current of the upper and lower capacitors on the DC side at time k;

[0033] Then, based on the switch state and converter output current and Kirchhoff's current law, the current flowing out of the midpoint is:

[0034]

[0035] Among them, i a 、i b 、i c is the grid-side current in the three-phase coordinate system;

[0036] The prediction model of midpoint potential difference is obtained as follows:

[0037]

[0038] Among them, Δu(k) represents the midpoint potential difference at time k, i a (k), i b (k), i c (k) represents the grid-side current at time k.

[0039] Furthermore, in step 6, in order to achieve the goal of using the reference current to track the current and balance the midpoint potential, a weighting coefficient of λ is used. dc The MPC cost function is expressed as:

[0040]

[0041] in, Represents the current reference value at time k+2.

[0042] Different redundant vectors have the same effect on the grid-side current, but different effects on the midpoint potential. The iterative method that ignores the redundant vector can only achieve the purpose of tracking the reference current, but cannot meet the requirement of balancing the midpoint potential. Therefore, the hierarchical optimization method is adopted, which can not only reduce the calculation burden, but also eliminate the cumbersome weight factor design steps.

[0043] Furthermore, in step 7, the cost function is optimized hierarchically to obtain:

[0044]

[0045] Substitute the control set after removing the redundant vector into the cost function (1) for calculation and evaluation, and select the optimal switching function S α , S β ; Then, use S α , S β The corresponding redundant vector is evaluated in part (2) of the cost function, and finally the optimal switching sequence S that can track the reference current and balance the midpoint potential is found. a , S b , S c ; Since one switch vector corresponds to at most n+1 redundant vectors, the number of iterations increases by at most n+1 times.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] (1) The present invention applies full control set model predictive control to a three-level converter, which not only achieves better control performance, but also retains the advantages of simple structure and fast dynamic response speed of traditional model predictive control. It also has the advantages of fixed switching frequency, small output current ripple, and good steady-state output current waveform quality, thereby improving the control performance of the power electronic converter.

[0048] (2) The present invention utilizes a hierarchical optimization method to solve the problem of midpoint potential balance in a three-level converter, while eliminating cumbersome weight factor design steps and reducing the computational burden. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0050] Figure 1 It is a flow chart of a control method for solving the midpoint potential imbalance of a three-level converter using a full control set model prediction in an embodiment;

[0051] Figure 2 This is a schematic diagram of the control set interval division in the embodiment;

[0052] Figure 3 Schematic diagram of the switch state when the division interval n is 4 in the embodiment;

[0053] Figure 4 Schematic diagram of non-redundant vectors in an embodiment;

[0054] Figure 5 Schematic diagram of the DC side circuit of the converter in the embodiment;

[0055] Figure 6 It is a comparison diagram of steady-state performance simulation results in the embodiment;

[0056] Figure 7 A comparison diagram of switch function values ​​in the embodiment;

[0057] Figure 8 It is a comparison diagram of dynamic response simulation results in the embodiment;

[0058] Fig. 9 Graph showing the grid-side current spectrum in the embodiment. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. Of course, the specific embodiments described here are only used to explain the present invention and are not used to limit the present invention.

[0060] Example 1

[0061] like Figure 1 As shown, a three-level converter full control set model predictive control method includes the following steps:

[0062] Step 1, using sensors to collect electrical physical quantities, constructing a first prediction model of a three-level converter based on the electrical physical quantities, converting the first prediction model to an αβ coordinate system, and performing delay compensation; wherein the electrical physical quantities include a grid-side current, a grid voltage, a DC-side voltage, and upper and lower capacitor voltages on the DC-side at the current moment;

[0063] After the first prediction model is transformed into the αβ coordinate system by Clark transformation in step 1, the model expression obtained is:

[0064]

[0065] Where, L is the filter inductance, i αβ is the grid-side current in the stationary αβ coordinate system, V gαβ is the grid voltage in the stationary αβ coordinate system, R is the line load, V dc is the DC side voltage, S αβ , S α , S β represents the switching function in the stationary αβ coordinate system, S a , S b , S c represents the switching function in a stationary three-phase coordinate system;

[0066] Then, the load current derivative di / dt is approximated by the forward Euler method, and we get:

[0067]

[0068] Where k represents the sampling time, T s Indicates the sampling time;

[0069] Substituting the load current derivative di / dt into the model, we get:

[0070]

[0071] Among them, i αβ (k) represents the grid-side current at time k in the stationary αβ coordinate system, V gαβ (k) represents the grid voltage at time k in the stationary αβ coordinate system, V dc (k) represents the DC side voltage at time k, S αβ (k) represents the switching function at time k in the stationary αβ coordinate system.

[0072] After delay compensation, the first prediction model becomes:

[0073]

[0074] Step 2: Divide each control cycle into n equal state intervals, define the switching function according to the n state intervals, form a full control set, and use the full control set to implement the full control set model predictive control of the three-phase three-level NPC converter.

[0075] Specifically, the implementation process of step 2 includes:

[0076] At time k, the prediction interval at time k+1 is divided into n parts, and the proportion of high level is set to m, 0≤m≤n. It is defined that the switch tube is turned on during some continuous intervals in each control cycle and remains off during the remaining intervals. The full control set after division is Contains a combination of n+1 switch states; S a , S b , S c is the switching function in the three-phase coordinate system.

[0077] The traditional finite control set method only contains the combination of three switch states {-1, 0, 1}, while the full control set after dividing the prediction interval contains The combination of switch states improves the control performance of the power electronic converter by redefining the control set and adding optional vector states. Figure 2 As shown in the upper part, the traditional finite control set model predicts the switching state S xi (x=a,μb,c;i=1,2,3,4) can only be 0 or 1 in one control cycle. In this embodiment, one control cycle of the full control set model predictive control is divided into n parts, Sxi It can be m1 / n or m2 / n. In an example, when n is 4, the five possible switch states of the full control set model predictive control (CFCS-MPC) correspond to S x =-1, S x =-1 / 2, S x =0, S x =1 / 2, S x =1, such as Figure 3 As shown, S a = -1 / 2 and S a =1 / 2, the switching state of the a-phase bridge arm.

[0078] Step 3: Convert the full control set after removing redundant vectors to the αβ coordinate system.

[0079] Specifically, the implementation process of step 3 includes:

[0080] Combining each parameter in the full control set, for a three-phase three-level converter, we can get (n+1) 3 There are redundant state vectors in these state vectors. After removing the redundant state vectors, we can get 3n 2 +3n+1 kinds of state vectors; Clark transform the non-redundant state vector to the αβ coordinate system to obtain the switching function S in the αβ coordinate system α , S β :

[0081]

[0082] For a three-phase three-level converter, such as Figure 4 As shown in (a), the traditional FCS-MPC has a total of 27 switch state combinations, including 19 non-redundant vectors and 8 redundant vectors; while for the CFCS-MPC described in the present invention, as the value of n increases, the selectable states show a polynomial growth, Figure 4 (b) and show the non-redundant state vectors when n is 4 and n is 8, respectively. When n is 4, the possible switching states are reduced to 61, and when n is 8, the possible switching states are reduced to 217.

[0083] Step 4: convert the full control set to the αβ coordinate system and bring it into the first prediction model after delay compensation to obtain the second prediction model in the αβ coordinate system under the action of the full control set.

[0084] Specifically, in step 4, the whole control set is transformed into the αβ coordinate system by Clark transformation, and is introduced into the first prediction model. The expression of the obtained second prediction model is:

[0085]

[0086] Program directly in the three-phase coordinate system,

[0087] Through three for loops, the states of the three bridge arms are traversed Need to calculate (n+1) 3 times; converted to the stationary αβ coordinate system, the amount of calculation is greatly reduced to 3n due to the removal of redundant vectors 2 +3n+1 times, four boundary conditions are needed to find all the correct states. The boundary conditions are as follows:

[0088]

[0089] Step 5: construct a prediction model for the midpoint potential difference.

[0090] The DC side circuit of the three-phase three-level NPC converter is as follows Figure 5 As shown, C1 and C2 are capacitor capacitances, u c1 、u c2 is the upper and lower capacitor voltage, i c1 、i c2 is the current of the upper and lower capacitors, i o is the current flowing out of the midpoint O. Specifically, the implementation process of step 5 includes:

[0091] Write down the expression of capacitor voltage dynamic process:

[0092]

[0093] Approximating the capacitor voltage derivative du / dt by the forward Euler method, we get:

[0094]

[0095] Substituting the capacitor voltage derivative du / dt into the model, we get:

[0096]

[0097] Among them, u c1 (k),u c2 (k) represents the voltage of the upper and lower capacitors on the DC side at time k, i c1 (k), i c2 (k) represents the current of the upper and lower capacitors on the DC side at time k

[0098] Then, based on the switch state and converter output current and Kirchhoff's current law, the current flowing out of the midpoint can be obtained as:

[0099]

[0100] Among them, ia 、i b 、i c is the grid-side current in the three-phase coordinate system;

[0101] The expression of midpoint potential difference is:

[0102] Δu(k+1)=u c1 (k+1)-u c2 (k+1)

[0103] Therefore, the prediction model of midpoint potential difference can be obtained as follows:

[0104]

[0105] Among them, Δu(k) represents the midpoint potential difference at time k, i a (k), i b (k), i c (k) represents the grid-side current at time k.

[0106] Step 6: Establish a cost function using the second prediction model and the prediction model of the midpoint potential difference.

[0107] Specifically, in step 6, in order to achieve the goal of using the reference current to track the current and balance the midpoint potential, a weighting coefficient of λ is used. dc The MPC cost function is expressed as:

[0108]

[0109] in, Represents the current reference value at time k+2.

[0110] Different redundant vectors have the same effect on the grid-side current, but different effects on the midpoint potential. The iterative method that ignores the redundant vector can only achieve the purpose of tracking the reference current, but cannot meet the requirement of balancing the midpoint potential. Therefore, the hierarchical optimization method is adopted, which can not only reduce the calculation burden, but also eliminate the cumbersome weight factor design steps.

[0111] Step 7, using the hierarchical optimization method, convert the cost function into two parts (current term and midpoint potential difference term), remove redundant vectors from the full control set, calculate the cost function, select the state vector that minimizes the cost function from the full control set as the optimal vector for the corresponding control cycle, and perform coordinate transformation, and use the transformed optimal vector to control the switching state of the three-level converter.

[0112] Specifically, in step 5, the cost function is optimized hierarchically to obtain:

[0113]

[0114] Substitute the control set after removing the redundant vector into the cost function (1) for calculation and evaluation, and select the optimal switching function S α , S β ; Then, use S α , S β The corresponding redundant vector is evaluated in part (2) of the cost function, and finally the optimal switching sequence S that can track the reference current and balance the midpoint potential is found. a , S b , S c Since one switch vector corresponds to at most n+1 redundant vectors, the number of iterations increases by at most n+1 times.

[0115] Since various digital processors today (such as DSP, FPGA, etc.) all use timer modules to control the size of the duty cycle when outputting PWM waves with different duty cycles, the continuous variable of duty cycle will be discretized into an integer variable by the timer clock of the PWM generation module.

[0116] For example, when a 100MHz controller is used to generate a 10kHz center-symmetrical PWM signal, the peak value of the triangular carrier waveform will be 5000 (100MHz / 10kHz / 2), which means that only 5000 duty cycles can be output, that is, the maximum value of n is 5000.

[0117] Similarly, considering that the clock frequency of the digital controller is limited, the number of divided state intervals is also limited. Therefore, n has a maximum value. When n reaches the maximum, the corresponding control set is the full control set.

[0118] In one example, simulation verification is performed according to the parameters listed in Table 1, and the following results are obtained:

[0119] Figure 6 For traditional FCS-MPC ( Figure 6 (a)), PI control ( Figure 6 (b)) and CFCS-MPC with n=10 ( Figure 6 (c)) and CFCS-MPC with n equal to 50 ( Figure 6 (d)) shows the steady-state performance simulation diagram. It can be seen that the steady-state current ripple of the CFCS-MPC with n = 10 is significantly lower than that of the traditional FCS-MPC, but higher than that of the PI control strategy. The DC side capacitor voltage shows that the midpoint potential difference of the CFCS-MPC with n = 10 is well suppressed compared with the traditional FCS-MPC and PI, and the midpoint potential fluctuation is less than 2V. For the CFCS-MPC with n = 50, not only the steady-state current ripple becomes smaller, but also the midpoint potential difference is well suppressed. The steady-state current ripple is only slightly larger than that of the PI control strategy, and the midpoint potential fluctuation is less than 1V.

[0120] Figure 7 CFCS-MPC with n set to 10 ( Figure 7 (a)) and CFCS-MPC with n equal to 50 ( Figure 7 (b)) The switching function value shows that with the increase of the division interval n, the switching state waveform gradually becomes smoother and closer to a saddle wave, which is very similar to the SVPWM modulation waveform.

[0121] Figure 8 For traditional FCS-MPC ( Figure 8 (a)), PI control ( Figure 8 (b)) and CFCS-MPC with n=10 ( Figure 8 (c)) and CFCS-MPC with n equal to 50 ( Figure 8 (d)) Dynamic response results comparison diagram, FCS-MPC and CFCS-MPC dynamic response speed is basically the same, CFCS-MPC and traditional FCS-MPC current tracking speed is significantly faster than PI control strategy;

[0122] Fig. 9 For traditional FCS-MPC ( Fig. 9 (a)), PI control ( Fig. 9 (b)) and CFCS-MPC with n=10 ( Fig. 9 (c)) and CFCS-MPC with n equal to 50 ( Fig. 9 From the grid-side current spectrum diagram (d), it can be seen that the current spectrum distribution of the traditional FCS-MPC is relatively dispersed, the switching frequency is uncertain, and most of them are concentrated below 10 kHz. The current spectrum distribution of the CFCS-MPC and PI proposed in this embodiment is concentrated near the integer multiples of the switching frequency 10 kHz.

[0123] Table 1 Electrical physical quantity parameter values

[0124]

[0125] The above experimental results show that the full control set model predictive control used in the present invention has better control performance than the traditional finite control set model predictive control; the problem of midpoint potential balance of the three-level converter is achieved by using a hierarchical optimization method.

[0126] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A three-level converter full control set model predictive control method, characterized in that: The following steps are involved: Step 1: Use sensors to collect electrical physical quantities, build a first prediction model of the three-level converter based on the electrical physical quantities, convert the first prediction model to the αβ coordinate system, and perform delay compensation; the electrical physical quantities include the grid side current, grid voltage, DC side voltage, and DC side upper and lower capacitor voltages at the current moment; Step 2: Divide each control cycle into n equal state intervals, define the switching function according to the n state intervals, and form a full control set; Step 3: Convert the full control set after removing redundant vectors to the αβ coordinate system; Step 4: Bring the converted full control set into the first prediction model to obtain the second prediction model in the αβ coordinate system under the action of the full control set; Step 5: There is a midpoint potential imbalance in the three-level converter. Based on the relationship between capacitance and switch state, a prediction model for the midpoint potential difference is constructed. Step 6: Establish a cost function using the second prediction model and the prediction model of the midpoint potential difference; Step 7: Using the hierarchical optimization method, the cost function is converted into two parts: the current term and the midpoint potential difference term. After removing the redundant vectors from the full control set, the cost function is calculated. The state vector that minimizes the cost function is selected from the full control set as the optimal vector for the corresponding control period, and the coordinate transformation is performed. The transformed optimal vector is used to control the switching state of the three-level converter.

2. A three-level converter full control set model predictive control method according to claim 1, characterized in that: In step 1, the first prediction model is transformed into the αβ coordinate system by Clark transformation, and the model expression is obtained as follows: Where, L is the filter inductance, i αβ is the grid-side current in the stationary αβ coordinate system, V gαβ is the grid voltage in the stationary αβ coordinate system, R is the line load, V dc is the DC side voltage, S αβ , S α , S β represents the switching function in the stationary αβ coordinate system, S a , S b , S c represents the switching function in the three-phase coordinate system; The load current derivative di / dt is approximated by the forward Euler method, and di / dt is brought into the model for delay compensation to obtain: Among them, i αβ (k+1) represents the grid-side current at time k+1 in the stationary αβ coordinate system, V gαβ (k+1) represents the grid voltage at time k+1 in the stationary α coordinate system, V dc (k+1) represents the DC side voltage at time k+1, S αβ (k+1) represents the switching function at time k+1 in the stationary αβ coordinate system.

3. The three-level converter full control set model predictive control method according to claim 2, characterized in that: The step 2 comprises the following steps: At time k, the prediction interval at time k+1 is divided into n parts, and the proportion of high level is set to m, 0≤m≤n. It is defined that the switch tube is turned on during some continuous intervals in each control cycle and remains off during the remaining intervals. The full control set after division is Contains a combination of n+1 switch states; S a , S b , S c is the switching function in the three-phase coordinate system.

4. The three-level converter full control set model predictive control method according to claim 3, characterized in that: The step 3 comprises the following steps: Combining each parameter in the full control set, for a three-phase three-level converter, we can get (n+1) 3 There are redundant state vectors in these state vectors. After removing the redundant state vectors, we can get 3n 2 +3n+1 kinds of state vectors; Clark transform the non-redundant state vector to the αβ coordinate system to obtain the switching function S in the αβ coordinate system α , S β :

5. The three-level converter full control set model predictive control method according to claim 4, characterized in that: The expression of the second prediction model obtained in step 4 is:

6. A three-level converter full control set model predictive control method according to claim 5, characterized in that: The prediction model of the midpoint potential difference is constructed in step 5, comprising the following steps: First, write the expression of the capacitor voltage dynamic process, and approximate the capacitor voltage derivative du / dt in the expression by the forward Euler method to obtain: Among them, u c1 (k) and u c2 (k) represents the voltage of the upper and lower capacitors on the DC side at time k, c1 and c2 are the capacitances of the DC side capacitors, and i c1 (k), i c2 (k) represents the current of the upper and lower capacitors on the DC side at time k; Then, based on the switch state and converter output current and Kirchhoff's current law, the current flowing out of the midpoint is: Among them, i a 、i b 、i c is the grid-side current in the three-phase coordinate system; The prediction model of midpoint potential difference is obtained as follows: Among them, Δu(k) represents the midpoint potential difference at time k, i a (k), i b (k), i c (k) represents the grid-side current at time k.

7. The three-level converter full control set model predictive control method according to claim 1, characterized in that: In step 6, in order to achieve the goal of tracking the current and balancing the midpoint potential with the reference current, a weighting coefficient of λ is used. dc The MPC cost function is expressed as: in, Represents the current reference value at time k+2.

8. The three-level converter full control set model predictive control method according to claim 1, characterized in that: In step 7, the cost function is optimized hierarchically to obtain: Substitute the control set after removing the redundant vector into the cost function (1) for calculation and evaluation, and select the optimal switching function S α , S β ; Then, use S α , S β The corresponding redundant vector is evaluated in part (2) of the cost function, and finally the optimal switching sequence S that can track the reference current and balance the midpoint potential is found. a , S b , S c ; Since one switch vector corresponds to at most n+1 redundant vectors, the number of iterations increases by at most n+1 times.

Images