A model predictive control algorithm for optimal switching sequence of MMC

Abstract

The present invention provides an optimal switching sequence model predictive control algorithm for a modular multi-level converter, comprising the following steps: converting a current prediction model of a three-phase MMC into a voltage prediction model; converting the predicted voltage vector into a g-h coordinate system, and selecting three nearest voltage base vectors; screening a switch state combination that minimizes the number of switch actions to obtain a symmetrical seven-segment switch sequence; finding an extreme value for a value function to obtain the optimal action time of each vector; allowing the sum of the upper and lower bridge arm voltages to follow an expected value to suppress interphase circulating current; and using a sorting voltage-sharing algorithm to complete the switch control of the MMC converter. The method can be extended to any level, and the complexity of the algorithm remains unchanged. Compared with traditional model prediction, the method achieves a fixed switching frequency, eliminates weight factors, reduces harmonic content, reduces computational burden, and improves output current tracking accuracy.

Description

Technical Field

[0001] The invention relates to the technical field of power electronic converter model predictive control, and in particular to an optimal switching sequence model predictive control algorithm applied to a modular multi-level converter. Background Art

[0002] Modular Multilevel Converter (MMC) has the advantages of high efficiency, low harmonic content and low switching frequency, and is widely used in high-voltage DC transmission systems, high-voltage inverters and active power filters. Due to its highly modular structure, MMC has good scalability and can adapt to various voltage and power level applications.

[0003] The mathematical model of MMC has the characteristics of multiple inputs, multiple outputs and nonlinearity. During its operation, multiple control objectives need to be considered, including output voltage or current control, circulating current suppression, submodule capacitor voltage balance control, etc. Traditional control strategies mostly use closed-loop control, which is complex and difficult to adjust the parameters of proportional integral regulator. In comparison, model predictive control (MPC) is easy to handle multiple system constraints, multiple variables and nonlinear systems, and has the advantages of fast response speed. It has many advantages in the control of MMC. Traditional model predictive control has problems such as large amount of calculation, many times of traversal optimization, and difficulty in adjusting weight factors. Only one optimal switching state is selected in one control cycle, and the required sampling frequency is high. It is impossible to achieve a fixed switching frequency. The harmonic spectrum is widely distributed, making the design of the output filter particularly difficult. In response to this problem, some scholars have proposed a multi-stage model predictive control strategy, which can improve the control accuracy of the output current, reduce the harmonic content, and achieve a fixed switching frequency, but there is still a high computational burden. Therefore, it is very necessary to further study the model predictive control algorithm applied to MMC. Summary of the invention

[0004] The purpose of the present invention is to optimize the control target realization mode on the basis of multi-stage model predictive control, improve the tracking accuracy of output current, reduce harmonic content, and greatly reduce the calculation burden of the controller while being applicable to multi-level series. Combining multi-stage model prediction with voltage vector prediction model, a MMC optimal switch sequence model predictive control algorithm is proposed, which is characterized in that the strategy converts the current prediction model of the three-phase MMC into a voltage prediction model, converts the predicted voltage vector into the gh coordinate system, selects the three nearest voltage base vectors, screens the switch state combination that minimizes the number of switch actions, obtains a symmetrical seven-stage switch sequence, obtains the extreme value of the value function, obtains the optimal action time of each vector, suppresses the interphase circulation by adjusting the number of submodules invested in each bridge arm, and finally uses the sorting voltage balancing algorithm to complete the switch control of the MMC converter, which specifically includes the following steps:

[0005] Step 1: Establish a discrete prediction model for the output voltage vector of the MMC converter and perform overmodulation processing on the predicted voltage vector;

[0006] Step 2: transform the predicted voltage vector into the gh coordinate system and select the three nearest voltage basis vectors to select the switch state combination that minimizes the number of switch actions, and obtain a symmetrical seven-segment switch sequence;

[0007] Step 3: Select the variance function of the predicted current and the reference current as the cost function, and use the method of finding the extreme value to obtain the optimal action time of each voltage basis vector when the cost function takes the minimum value;

[0008] Step 4: Calculate the expected value of the sum of the upper and lower bridge arm voltages of each phase at time k+1 using the circulating current discrete prediction model. Then, based on the discrete prediction model of the submodule capacitor voltage, consider the bridge arm as a whole and calculate the predicted value of the sum of the upper and lower bridge arm voltages of each phase. By adjusting the number of submodules put into each bridge arm, eliminate the error between the expected value and the predicted value, thereby suppressing the circulating current.

[0009] Step 5: Combine the seven-segment switching sequence with the circulating current suppression target to obtain the optimal switching sequence, and then use the sorting method to balance the sub-module capacitor voltage and send out control pulses;

[0010] In the step 1, a discrete prediction model of the output voltage vector is established. The time domain mathematical model of the output current of the MMC converter is obtained according to Kirchhoff's law and the voltage and current constraint relationship of the capacitor and the inductor, and the discrete domain mathematical model is obtained by discretizing it using the forward Euler method. Then, a phase voltage discrete model is obtained based on this model, and the three-phase electrical quantity is synthesized to obtain the output voltage vector prediction model. Due to the sampling time T sVery small, the voltage vector at time k+1 in the formula can be approximated as the voltage vector at time k, which can be obtained through voltage sampling. Similarly, the current vector at time k+1 can be approximated as the reference current vector at time k to control the AC side current;

[0011] In the step 1, the predicted voltage vector is subjected to overmodulation processing. In order to prevent the predicted voltage vector from exceeding the modulation area, the modulus of the maximum vector allowed at the same angle as the predicted voltage is calculated based on the geometric relationship. If the modulus of the predicted vector is greater than the modulus of the maximum vector, the predicted vector is reset to the maximum vector at the same angle; if it is less than the maximum vector, the predicted vector remains unchanged.

[0012] In the step 2, the switch state combination that minimizes the number of switch actions is selected to obtain a symmetrical seven-segment switch sequence. Since one base vector may correspond to multiple different switch state combinations, a value function is defined to perform rolling optimization on the redundant switch states of each voltage vector in the sequence, and the switch state with the smallest change relative to the previous switch state is selected to obtain a symmetrical seven-segment switch sequence.

[0013] In step five, the submodule capacitor voltage is balanced by a sorting method. According to the real-time sorting results of the capacitor voltages of all submodules in the bridge arm, when the bridge arm current has a charging characteristic, a submodule with a lower capacitor voltage is put into operation, and when the bridge arm current has a discharging characteristic, a submodule with a higher capacitor voltage is put into operation.

[0014] Furthermore, the MMC optimal switching sequence model predictive control algorithm is characterized in that the reference output current is converted into the desired output voltage as follows:

[0015] The discrete prediction model of the AC j-phase current of the MMC inverter is:

[0016]

[0017] Where: i j (k+1),u pj (k+1) and u nj (k+1) are the j-phase current, j-phase upper bridge arm voltage and lower bridge arm voltage measured at time k+1; L and R are the reactance and resistance of the AC side respectively; L o and R o are the bridge arm reactance and resistance respectively; T s is the sampling time;

[0018] According to Kirchhoff's voltage law, the following relationship can be obtained:

[0019]

[0020] Where: u j (k+1) is the j-phase output voltage of the MMC at time k+1;

[0021] Combining the above equation with the phase current discrete model and synthesizing the three-phase electrical quantities, we can obtain:

[0022]

[0023] Where: i(k+1), u(k+1) and i(k) are the synthetic vectors of the relevant three-phase electrical quantities;

[0024] In order to track the output current, the current vector at time k+1 is set as the reference current vector i * (k+1), due to the sampling time T s Very small, can be approximated as i * (k); By transforming the above formula, we can get the voltage vector prediction model:

[0025]

[0026] Where: u p (k+1) is the desired output voltage vector.

[0027] Furthermore, the MMC optimal switching sequence model predictive control algorithm is characterized by: transforming the predicted voltage vector into the gh coordinate system and selecting three nearest voltage basis vectors, as follows:

[0028] Convert all basic vector coordinates in the gh coordinate system into integers; for any predicted voltage vector, round up and down its coordinates to get four voltage basic vector coordinates, and through logical judgment, select the three nearest voltage basic vectors, and the selected seven-segment switch sequence is composed of the switch states corresponding to these three basic vectors;

[0029] This method can significantly reduce the amount of calculation while ensuring the control effect of the MPC algorithm, and the complexity of the calculation remains unchanged when used at any level.

[0030] Furthermore, the MMC optimal switching sequence model predictive control algorithm is characterized in that: the variance function of the predicted current and the reference current is selected as the cost function, and the action time of the three nearest voltage basis vectors when the cost function takes the minimum value is obtained by using the method of finding the extreme value, which is specifically as follows:

[0031] Since the sampling time T s When predicting the current after each voltage vector acts, the initial current is approximated by the sampled current at time k, as shown in the following formula:

[0032]

[0033] Where: m ranges from 1 to 3; i α,m 、iβ,m is the current prediction value after the selected mth voltage vector acts; t mi is the action time of the mth voltage vector, and satisfies t 1i +t 2i +t 3i =T s ;u α，m 、u β，m 、i α (k), i β (k) is the component of the relevant electric quantity on the α and β axes;

[0034] Considering the current error under the action of each voltage basis vector, the cost function is established as follows:

[0035]

[0036] Where: i * α 、i * β for i * (k) Components on the α and β axes;

[0037] By finding the extreme value of the cost function and using the partial derivative formula, we can get the optimal action time of the three voltage basis vectors.

[0038] Compared with the prior art, the present invention has the following advantages: a fixed switching frequency is achieved, harmonics are concentrated near the switching frequency, and the design difficulty of the output filter is reduced; the weight factor is eliminated, the output current tracking accuracy is improved, the harmonic content is reduced, and the calculation burden is reduced; it can be extended to any level, and the complexity of the algorithm remains unchanged. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is the topology diagram of modular multi-level inverter;

[0040] Figure 2 is a five-level vector sector diagram in one embodiment;

[0041] Figure 3 A five-level space vector diagram in one embodiment;

[0042] Figure 4 This is a carrier comparison chart. DETAILED DESCRIPTION

[0043] In order to illustrate the basic principle, technical solution and performance advantages of the present invention, an embodiment of the present invention will be further described below in conjunction with the accompanying drawings. It should be noted that the present invention can be implemented in a variety of different ways covered by the claims.

[0044] The topology of modular multilevel inverter is as follows Figure 1As shown, it can be divided into three levels: phase unit, bridge arm unit and submodule unit. The submodule can be any structure such as full bridge, half bridge, etc., which is connected to the main circuit after being connected in parallel with the DC capacitor. The DC side of the MMC can be connected to the DC bus of the power grid and output a constant DC voltage U dc The AC side can be connected to an AC power source or a three-phase load to output three-phase sinusoidal AC power.

[0045] Taking a five-level MMC inverter as an example, the implementation of its control method includes the following steps:

[0046] Step 1: Establish a discrete prediction model for the output voltage vector of the MMC inverter. The discrete prediction model for the AC j-phase current of the MMC inverter is as follows:

[0047]

[0048] Where: i j (k+1),u pj (k+1) and u nj (k+1) are the j-phase current, j-phase upper bridge arm voltage and lower bridge arm voltage measured at time k+1; L and R are the reactance and resistance of the AC side respectively; L o and R o are the bridge arm reactance and resistance respectively; T s is the sampling time;

[0049] According to Kirchhoff's voltage law, the following relationship can be obtained:

[0050]

[0051] Where: u j (k+1) is the j-phase output voltage of the MMC at time k+1;

[0052] Combining the above equation with the phase current discrete model and synthesizing the three-phase electrical quantities, we can obtain:

[0053]

[0054] Where: i(k+1), u(k+1) and i(k) are the synthetic vectors of the relevant three-phase electrical quantities;

[0055] In order to track the output current, the current vector at time k+1 is set as the reference current vector i * (k+1), due to the sampling time T s Very small, can be approximated as i * (k); Transform the above formula to obtain the voltage vector prediction model:

[0056]

[0057] Where: up (k+1) is the desired output voltage vector.

[0058] Step 2: Over-modulate the predicted voltage vector to prevent it from exceeding the modulation area. Figure 2 As shown, with u p The maximum voltage vector allowed at the same angle is expressed as If u p Model ratio If u p Replace with If less than, remain unchanged. The calculation method is as follows:

[0059]

[0060] Where: is the vector angle of the reference voltage.

[0061] Step 3: Convert the output basic vector of the inverter to the gh coordinate system and convert the coordinates of all basic vectors into integers. The five-level space vector in the gh coordinate system is as follows: Figure 3 As shown, the g-axis coincides with the α-axis in the rectangular coordinate system, and the h-axis is at an angle of 60 degrees to the g-axis.

[0062] Let u p The coordinate in the α-β coordinate system is u α 、u β , its coordinate after transformation to the gh coordinate system is u g 、u h , according to the geometric relationship, the transformation formula of the two coordinate systems is:

[0063]

[0064] Let u p The coordinate in the abc coordinate system is u a 、u b 、u c , according to Clark's formula and the above formula, the transformation formula between the three-phase coordinate system and the gh coordinate system is:

[0065]

[0066] For normalization, the coordinates of all basis vectors are converted to integers.

[0067] Step 4: Select the three nearest voltage basis vectors. Since the coordinates of the basis vectors have been converted to integers, p The four basic voltage vectors closest to the reference vector can be obtained by rounding the coordinates upward and downward, as shown in the following formula:

[0068]

[0069]

[0070] In the formula, the underscore and the upscore represent the rounding up and downs of the variables, respectively. ul 、u lu 、u uu 、u ll Respectively represent and u p The four nearest basis vectors.

[0071] From the geometric relationship, we know that u ul and u lu The basic vectors that are always closest to the reference vector are denoted as u1 and u2; the third closest vector is based on u g +u h -(u ulg +u ulh ) is determined by its sign. If it is positive, then u uu is the third basic vector required, otherwise, u ll is the third basic vector required, denoted as u3.

[0072] Step 5: Screen the switch state combination that minimizes the number of switching actions to obtain a symmetrical seven-segment switch sequence. For the five-level MMC converter, in the gh coordinate system, all the switch states corresponding to any voltage base vector can be determined by the following formula:

[0073] and

[0074] Where: q ranges from 1 to 4; u optg and u opth are the selected voltage basis vectors u opt Projection on the g, h axes.

[0075] It can be seen that the same voltage base vector may correspond to multiple switch states, and it is very important to select an appropriate switch state combination. In order to reduce the number of switching actions, assume that the previous switch state is (S aold ,S bold ,S cold ), the next switch state is (S a ,S b ,S c ), define the switch state change value Y as:

[0076] Y=|S a -S aold |+|S b -S bold|+|S c -S cold |

[0077] First, determine the first switch state, compare the switch state corresponding to u1 with the last switch state of the previous sampling cycle, and select the switch state with the smallest Y value. The second switch state corresponds to u2, the third switch state corresponds to u3, and the fourth switch state has the same basic voltage vector as the first switch state. The second, third, and fourth switches all calculate the Y value based on the selected previous switch state, and determine the switch state with the smallest Y value as the desired switch state.

[0078] In the symmetrical seven-stage switch sequence, the fifth stage is consistent with the third stage, the sixth stage is consistent with the second stage, and the seventh stage is consistent with the first switch state.

[0079] Step 6: Select the variance function of the predicted current and the reference current as the cost function, and use the extreme value method to obtain the optimal action time of the three nearest voltage basis vectors when the cost function takes the minimum value.

[0080] Since the sampling time T s When predicting the current after each vector is applied, the initial current is approximated by the sampled current at time k, as shown in the following formula:

[0081]

[0082] Where: m ranges from 1 to 3; i α,m 、i β,m is the current prediction value after the selected mth voltage vector acts; t mi is the action time of the mth voltage vector, and satisfies t 1i +t 2i +t 3i =T s ;u α，m 、u β，m 、i α (k), i β (k) is the component of the relevant electric quantity on the α and β axes;

[0083] Considering the current error under the action of each voltage basis vector, the cost function is established as follows:

[0084]

[0085] Where: i * α 、i * β for i * (k) Components on the α and β axes;

[0086] Find the extreme value of the cost function and use the partial derivative formula to get the optimal action time of the vector:

[0087]

[0088] Solving the above equation, we can get the optimal action time t of the three voltage base vectors: 1i ,t 2i and t 3i .

[0089] Step 7: Obtain the expected value of each phase bridge arm voltage at time k+1 from the circulating current prediction model. The discrete prediction model of the j-phase circulating current of the MMC inverter is shown as follows:

[0090]

[0091] Where: U dc is the DC side voltage; i dj (k+1) is the internal circulation of phase j of the MMC at time k+1;

[0092] It can be seen from the above formula that the circulating current is mainly caused by the inequality between the DC bus voltage and the sum of the upper and lower bridge arm voltages of the phase unit. When the difference between them is greater than zero, the circulating current tends to increase, and when the difference is less than zero, the circulating current decreases.

[0093] In order to eliminate the AC component in the circulating current, the following relationship must be satisfied:

[0094]

[0095] Where: I dc is the DC side current;

[0096] From this, we can get the expected value of the sum of the upper and lower bridge arm voltages of each phase at time k+1:

[0097]

[0098] Step 8: Based on the discrete prediction model of the submodule capacitor voltage, the bridge arm is regarded as a whole to calculate the predicted value of the bridge arm voltage of each phase. Due to the use of the voltage balancing algorithm, the capacitor voltages of each submodule in the same bridge arm are approximately equal, and the bridge arm can be regarded as a whole for voltage prediction, thereby obtaining the average value of the capacitor voltage of the submodule in each bridge arm at time k+1, as shown in the following formula:

[0099]

[0100] Where: u rj,ave (k+1) represents the average value of the capacitor voltage of the j-phase bridge arm unit r (r=p,n) submodule at time k+1; i rj (k) represents the bridge arm current of the bridge arm unit r of phase j; S rjiIndicates the number of conducting submodules of the j-phase r-bridge arm when the i-th switching state in the sequence is in effect; t i represents the action time of the i-th switch state in the sequence;

[0101] Predicted value u of bridge arm voltage of phase j at time k+1 rj (k+1) is:

[0102] u rj (k+1)=S rj7 u rj,ave (k+1)

[0103] Where: S rj7 It indicates the number of conducting submodules of the j-phase r-bridge arm when the seventh switching state in the sequence is in effect.

[0104] Step 9: Appropriately increase or decrease the number of submodules in each bridge arm to suppress the circulation. The number of submodules that should be increased or decreased in each bridge arm within a sampling period is Δn rj,eq It can be obtained by the following formula:

[0105]

[0106] In the above formula, the upper and lower bridge arms of each phase bear half of the error between the expected value and the predicted value of the bridge arm voltage respectively. This method can effectively suppress the circulating current without affecting the AC output.

[0107] Δn rj,eq It is usually a non-integer, and its fractional part is compared with the carrier, such as Figure 4 As shown, an equivalent control effect can be obtained. In the figure, cmp is Δn rj,eq The decimal part of the number of sub-modules that should be inserted into each bridge arm is Δn rj for:

[0108]

[0109] Step 10: Use the sorting method to balance the submodule capacitor voltage and send out control pulses. When the bridge arm current is positive, the submodule with lower voltage is put into operation first. On the contrary, when the bridge arm current is negative, the submodule with higher voltage is put into operation.

[0110] The above is a specific implementation of the present invention, but the protection scope of the present invention is not limited thereto. For those skilled in the art, changes and modifications can be made to the above embodiment without departing from the technical spirit and principle described in the present invention, and these changes and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A MMC optimal switching sequence model predictive control algorithm, characterized in that The current prediction model of the three-phase MMC is converted into a voltage prediction model, the predicted voltage vector is converted into the gh coordinate system, the three nearest voltage base vectors are selected, the switch state combination that minimizes the number of switch actions is screened, and a symmetrical seven-segment switch sequence is obtained. The extreme value of the value function is obtained to obtain the optimal action time of each vector, and the sum of the upper and lower bridge arm voltages is made to follow the expected value to suppress the interphase circulating current. Finally, the sorting voltage balancing algorithm is used to complete the switch control of the MMC converter, which specifically includes the following steps: Step 1: Establish a discrete prediction model for the output voltage vector of the MMC converter and perform overmodulation processing on the predicted voltage vector; Step 2: transform the predicted voltage vector into the gh coordinate system and select the three nearest voltage basis vectors to select the switch state combination that minimizes the number of switch actions, and obtain a symmetrical seven-segment switch sequence; Step 3: Select the variance function of the predicted current and the reference current as the cost function, calculate the partial derivative of the cost function, and obtain the optimal action time of each voltage basis vector; Step 4: Calculate the expected value of the sum of the upper and lower bridge arm voltages of each phase at time k+1 using the circulating current discrete prediction model. Then, based on the discrete prediction model of the submodule capacitor voltage, consider the bridge arm as a whole and calculate the predicted value of the sum of the upper and lower bridge arm voltages of each phase. By adjusting the number of submodules put into each bridge arm, eliminate the error between the expected value and the predicted value of the bridge arm voltage, thereby suppressing the circulating current. Step 5: Combine the seven-segment switching sequence with the circulating current suppression target to obtain the optimal switching sequence, and then use the sorting method to balance the sub-module capacitor voltage and send out control pulses; In the step 1, a discrete prediction model of the output voltage vector is established. The time domain mathematical model of the output current of the MMC converter is obtained according to Kirchhoff's law and the voltage and current constraint relationship of the capacitor and the inductor, and the discrete domain mathematical model is obtained by discretizing it using the forward Euler method. Then, a phase voltage discrete model is obtained based on this model, and the three-phase electrical quantity is synthesized to obtain the output voltage vector prediction model. Due to the sampling time T s Very small, the voltage vector at time k+1 in the formula can be approximated as the voltage vector at time k, which can be obtained through voltage sampling. Similarly, the current vector at time k+1 can be approximated as the reference current vector at time k to control the AC side current; In the step 1, the predicted voltage vector is subjected to overmodulation processing. In order to prevent the predicted voltage vector from exceeding the modulation area, the modulus of the maximum vector allowed at the same angle as the predicted voltage is calculated based on the geometric relationship. If the modulus of the predicted vector is greater than the modulus of the maximum vector, the predicted vector is reset to the maximum vector at the same angle; if it is less than the maximum vector, the predicted vector remains unchanged. In the step 2, the switch state combination that minimizes the number of switch actions is selected to obtain a symmetrical seven-segment switch sequence. Since one base vector may correspond to multiple different switch state combinations, a value function is defined to perform rolling optimization on the redundant switch states of each voltage vector in the sequence, and the switch state with the smallest change relative to the previous switch state is selected to obtain a symmetrical seven-segment switch sequence. In the step 5, the submodule capacitor voltage is balanced by a sorting method, and according to the real-time sorting results of the capacitor voltages of all submodules in the bridge arm, when the bridge arm current is a charging characteristic, a submodule with a lower capacitor voltage is put into operation, and when the bridge arm current is a discharging characteristic, a submodule with a higher capacitor voltage is put into operation; The MMC optimal switching sequence model predictive control algorithm is characterized in that the reference output current is converted into the expected output voltage as follows: The discrete prediction model of the AC j-phase current of the MMC inverter is: Where: i j (k+1),u pj (k+1) and u nj (k+1) are the j-phase current, j-phase upper bridge arm voltage and lower bridge arm voltage measured at time k+1; L and R are the reactance and resistance of the AC side respectively; L o and R o are the bridge arm reactance and resistance respectively; T s is the sampling time; According to Kirchhoff's voltage law, the following relationship can be obtained: Where: u j (k+1) is the j-phase output voltage of the MMC at time k+1; Combining the above equation with the phase current discrete model and synthesizing the three-phase electrical quantities, we can obtain: Where: i(k+1), u(k+1) and i(k) are the synthetic vectors of the relevant three-phase electrical quantities; In order to track the output current, the current vector at time k+1 is set as the reference current vector i * (k+1), due to the sampling time T s Very small, can be approximated as i * (k); By transforming the above formula, we can get the voltage vector prediction model: Where: u p (k+1) is the desired output voltage vector.

2. The MMC optimal switching sequence model predictive control algorithm according to claim 1, characterized in that: The predicted voltage vector is transformed into the gh coordinate system, and the three nearest voltage basis vectors are selected as follows: Convert all basic vector coordinates in the gh coordinate system into integers; for any predicted voltage vector, round up and down its coordinates to get four voltage basic vector coordinates, and through logical judgment, select the three nearest voltage basic vectors, and the selected seven-segment switch sequence is composed of the switch states corresponding to these three basic vectors; The method of converting the predicted voltage vector into the gh coordinate system and selecting three nearest voltage base vectors can significantly reduce the amount of calculation while ensuring the control effect of the MPC algorithm, and the complexity of the calculation remains unchanged when used for any level.

3. The MMC optimal switching sequence model predictive control algorithm according to claim 1, characterized in that: The variance function of the predicted current and the reference current is selected as the cost function. The method of finding the extreme value is used to obtain the action time of the three nearest voltage basis vectors when the cost function takes the minimum value, as follows: Since the sampling time T s When predicting the current after each voltage vector acts, the initial current is approximated by the sampled current at time k, as shown in the following formula: Where: m ranges from 1 to 3; i α,m 、i β,m is the current prediction value after the selected mth voltage vector acts; t mi is the action time of the mth voltage vector, and satisfies t 1i +t 2i +t 3i =T s ;u α，m 、u β，m 、i α (k), i β (k) is the component of the relevant electric quantity on the α and β axes; Considering the current error under the action of each voltage basis vector, the cost function is established as follows: Where: i * α 、i * β for i * (k) Components on the α and β axes; By finding the extreme value of the cost function and using the partial derivative formula, we can get the optimal action time of the three voltage basis vectors.

4. The MMC optimal switching sequence model predictive control algorithm according to claim 1, characterized in that: The interphase circulation can be suppressed by appropriately increasing or decreasing the number of submodules put into each bridge arm, as follows: The discrete prediction model of the j-phase circulating current of the MMC inverter is: Where: U dc is the DC side voltage; i dj (k+1) is the internal circulation of phase j of the MMC at time k+1; In order to eliminate the AC component in the circulating current, the following relationship must be satisfied: Where: I dc is the DC side current; From this, we can get the expected value of the sum of the upper and lower bridge arm voltages of each phase at time k+1: Due to the use of the sorting and voltage balancing algorithm, the capacitor voltages of each submodule in the same bridge arm are approximately equal; after selecting the optimal switching sequence, the bridge arm is regarded as a whole for voltage prediction, and the average value of the capacitor voltages of the submodules in each bridge arm at time k+1 is obtained, as shown in the following formula: Where: u rj,ave (k+1) represents the average value of the capacitor voltage of the j-phase bridge arm unit r (r=p,n) submodule at time k+1; i rj (k) represents the bridge arm current of the bridge arm unit r of phase j; S rji It indicates the number of conducting submodules of the j-phase r-bridge arm when the i-th switch state in the sequence is in effect; t i represents the action time of the i-th switch state in the sequence; N represents the number of bridge arm submodules; The predicted value u of the sum of the upper and lower bridge arm voltages of phase j at time k+1 rj (k+1) is: you rj (k+1)=S rj7 you rj,ave (k+1) The number of submodules that should be increased or decreased in each bridge arm within a sampling period Δn rj,eq for: Δn rj,eq Usually it is a non-integer. In order to obtain an equivalent control effect, let Δn rj,eq The fractional part of is cmp, and t rj = cmpT s , then the number of submodules that should be inserted into each bridge arm is Δn rj for: Since the upper and lower bridge arms of each phase bear half of the error between the expected value and the predicted value of the bridge arm voltage respectively, this method can effectively suppress the circulating current without affecting the AC output.

Images