阅读说明：本技术 Lcl型逆变器的无权值多变量顺序模型预测控制方法及装置 (No-weight multivariable sequential model prediction control method and device for LCL inverter ) 是由 张辉 李枝亮 刘业钊 谌金利 史军伟 张晓� 于 2020-06-12 设计创作，主要内容包括：本发明公开了一种LCL型逆变器的无权值多变量顺序模型预测控制方法及装置,该方法通过建立详细的预测模型可以实现直流侧中点电位、逆变器输出电流、滤波电容电压以及并网电流多个变量的控制。针对传统多变量模型预测控制难以选择合理的权重系数的问题,本发明对各个变量进行分步预测评估,每步只对其中的一个变量进行计算,并筛选出部分最优解作为可行域,代入下个变量的预测评估,最终筛选出最优控制目标。该控制方法无需给变量分配权值系数,并在逐步评估过程中缩小寻优空间,减小计算量,提高控制响应速度,该方法控制下的LCL三电平逆变器具有良好的输出效果。(The invention discloses a method and a device for predicative control of a weightless multivariable sequential model of an LCL (lower control limit) inverter. Aiming at the problem that the traditional multivariable model predictive control is difficult to select reasonable weight coefficients, the method carries out step-by-step predictive evaluation on each variable, only one variable in each step is calculated, partial optimal solutions are screened out to be used as feasible domains, the feasible domains are substituted into the predictive evaluation of the next variable, and the optimal control target is screened out finally. According to the control method, a weight coefficient does not need to be distributed to the variable, the optimization space is reduced in the gradual evaluation process, the calculated amount is reduced, and the control response speed is improved.)

1. A multivariable sequential model predictive control method without weight values for an LCL inverter is characterized in that multivariable comprises inverter output current control, filter capacitor voltage control, grid-connected current control and direct-current side midpoint potential control, and the method comprises the following steps:

layering the multivariable, and sequentially predicting and optimizing;

calculating a predicted value of the upper-layer variable in an initial optimization space, and substituting the predicted value into the variable cost function to sort the variables from small to large;

selecting part of the predicted value of the upper-layer variable as an optimization space of the lower-layer variable;

calculating a predicted value and sorting a cost function of the lower-layer variable, and gradually reducing an optimization space in the process;

and selecting the switching sequence which enables the cost function of the lowest layer variable to obtain the minimum value as the optimal control instruction to act on the system.

2. The LCL inverter weightless multivariable sequential model predictive control method according to claim 1, wherein the multivariable layering sequence comprises DC side midpoint potential control, inverter output current control, filter capacitor voltage control and grid-connected current control from top to bottom.

3. The LCL inverter non-weighted multivariable sequential model predictive control method according to claim 1 or 2, wherein the partial predicted value of the upper layer variable is selected as the optimization space of the lower layer variable, and the specific implementation manner is as follows:

sorting cost function values of upper-layer variables from small to large, and screening out the smallest first n values;

and finding n predicted values corresponding to the minimum first n values, and taking the switch states corresponding to the n predicted values as an optimization space of the lower-layer variable.

4. The LCL inverter non-weighted multivariable sequential model predictive control method according to claim 3, wherein the lower layer variables are subjected to predictive value calculation and cost function sequencing, and an optimization space is gradually reduced in the process, and the method is specifically realized as follows:

calculating a cost function value in an optimization space of a lower-layer variable, screening out the minimum first n-s values, and taking the switch states corresponding to the n-s predicted values as the optimization space of the lower-layer variable;

and the switching state of the optimization space of the lower variable gradually reduces the optimization space by sequentially decreasing the switching states of s.

5. The LCL inverter unweighted multivariable sequential model predictive control method of claim 1, wherein the cost function is constructed by,

establishing a multivariable mathematical model in the grid connection process of the LCL type three-level inverter to obtain a multivariable control type;

discretizing the control type of the multiple variables to obtain a prediction model of each variable;

and deducing reference values of other variables according to the given grid-connected current reference value, and constructing independent cost functions of the variables according to the predicted values and the reference values.

6. The LCL type inverter non-weight value multivariable sequence model predictive control method according to claim 5, wherein the multivariable mathematical model in the LCL type three-level inverter grid-connected process is established by:

wherein u is_cxIs the grid-connected filter capacitor voltage; u. of_xOutputting a phase voltage for the inverter; i.e. i_1xIs grid-connected side phase current; i.e. i_2xOutputting phase current for the inverter; e.g. of the type_xIs the power grid phase voltage; Δ u_dcThe voltage difference of the upper and lower capacitors at the DC side; i.e. i_c1,i_c2Respectively are direct current side upper and lower capacitance currents; i.e. i_npIs the neutral current; l is₁Is the grid side inductance value; l is₂Is an inverter side inductance value; c₁Is a filter capacitance value; c is the DC side up-down capacitance.

7. The LCL inverter non-weighted value multivariable sequential model predictive control method according to claim 6, wherein the predictive model of each variable is obtained by discretization of a multivariable control type by a forward Eulerian method, and the predictive formula is as follows:

wherein the content of the first and second substances,

8. The method for the predictive control of the unweighted multivariable sequential model of the LCL-type inverter according to claim 7, wherein the derivation of the reference values of the other variables from the given grid-connected current reference value is implemented as:

will the network voltage e_a(k),e_b(k),e_c(k) Clark and park transformation to obtain e_d(k),e_q(k) (ii) a Reference current to be connected to grid

e_dq(k)＝e_d(k)+je_q(k)

the filter capacitor reference voltage vector and the inverter output reference current vector are calculated as:

the calculated component

9. The method for predictive control of an unweighted multivariable sequential model of an LCL-type inverter as claimed in claim 8, wherein the construction of each variable independent cost function from the predicted values and the reference values is:

10. an LCL inverter weightless multivariable sequential model predictive control device characterized in that:

the prediction control device comprises a multivariable layering module, a prediction value calculation module, a cost function calculation module, an optimization space definition module and a control instruction module;

the multivariable layering module is used for independently layering the variables according to the control sequence, so that the multivariable sequential control is easy to realize;

the predicted value calculation module is used for predicting the size of each variable in the next control period according to the discretized mathematical model by using the sampling value of each variable in the control period;

the cost function calculation module is used for calculating the error magnitude between the predicted value and the reference value of the variable in the next control period;

the optimizing space defining module is used for selecting partial optimal values as the optimizing space of the next control variable by sorting the cost function;

the control instruction module is used for applying the optimal control instruction selected by the sequential control to the inverter to drive the switching tube to act;

the number of the predicted value calculation module, the cost function calculation module and the optimization space definition module is equal to the number of variables minus one.

Technical Field

The invention relates to the technical field of multi-level inverter control, in particular to a method and a device for predicatively controlling an LCL inverter through a weightless multivariable sequential model.

Background

With the development of new energy power generation systems such as photovoltaic and wind energy, grid-connected inverters are widely used as key connection equipment. Compared with the traditional inverter, the three-level inverter has the advantages of high voltage resistance and low harmonic content of output voltage and current, and is widely applied to high-voltage and high-power occasions. Meanwhile, compared with an L-type inverter, the inverter with the LCL-type structure not only needs smaller inductance, but also can well inhibit high-frequency harmonic waves caused by switching action, and the distortion rate of output current is lower. Therefore, the LCL three-level neutral-point clamped inverter is an ideal structure and is widely applied to occasions such as new energy grid connection and the like.

As a novel control method, model predictive control is used, a discrete mathematical model and a predictive model of a system are established, and a cost function is evaluated to enable a predicted quantity to track a reference quantity so as to achieve an expected control target. The model prediction control is simple, the response is rapid, the multi-objective optimization can be realized, and the method is widely applied to high-power converters at present. However, the existing multivariable control system needs to set a plurality of weight coefficients, and is difficult to select a proper weight. Therefore, research and improvement on a method for predicting and controlling the unweighted multivariable sequential model of the LCL inverter are urgently needed.

Disclosure of Invention

The invention provides a method for predicting and controlling an LCL inverter by a non-weighted multivariable sequential model, which realizes the control of each variable of the LCL grid-connected inverter, improves the robustness of a system and does not need to distribute weights to each control variable.

In order to solve the above technical problem, the present invention provides a method for predictive control of an unweighted multivariable sequential model of an LCL inverter, wherein multivariable comprises inverter output current control, filter capacitor voltage control, grid-connected current control and a dc-side midpoint potential, the method comprising:

layering the multivariable, and sequentially predicting and optimizing;

calculating a predicted value of the upper-layer variable in an initial optimization space, and substituting the predicted value into the variable cost function to sort the variables from small to large;

selecting part of the predicted value of the upper-layer variable as an optimization space of the lower-layer variable;

calculating a predicted value and sorting a cost function of the lower-layer variable, and gradually reducing an optimization space in the process;

and selecting the switching sequence which enables the cost function of the lowest layer variable to obtain the minimum value as the optimal control instruction to act on the system.

Further, the sequence of the multi-variable layering is direct-current side midpoint potential control, inverter output current control, filter capacitor voltage control and grid-connected current control from top to bottom in sequence.

Further, the selecting of the partial predicted value of the upper layer variable as the optimization space of the lower layer variable is specifically implemented as follows:

sorting cost function values of upper-layer variables from small to large, and screening out the smallest first n values;

and finding n predicted values corresponding to the minimum first n values, and taking the switch states corresponding to the n predicted values as an optimization space of the lower-layer variable.

Further, the lower layer variables are subjected to predicted value calculation and cost function sorting, and the optimization space is gradually reduced in the process, which is specifically realized as follows:

calculating a cost function value in an optimization space of a lower-layer variable, screening out the minimum first n-s values, and taking the switch states corresponding to the n-s predicted values as the optimization space of the lower-layer variable;

and the switching state of the optimization space of the lower variable gradually reduces the optimization space by sequentially decreasing the switching states of s.

Further, the constructing of the cost function includes,

establishing a multivariable mathematical model in the grid connection process of the LCL type three-level inverter to obtain a multivariable control type;

discretizing the control type of the multiple variables to obtain a prediction model of each variable;

and deducing reference values of other variables according to the given grid-connected current reference value, and constructing independent cost functions of the variables according to the predicted values and the reference values.

Further, the multivariate mathematical model in the grid-connected process of the LCL type three-level inverter is as follows:

wherein u is_cxIs the grid-connected filter capacitor voltage; u. of_xOutputting a phase voltage for the inverter; i.e. i_1xIs grid-connected side phase current; i.e. i_2xOutputting phase current for the inverter; e.g. of the type_xIs the power grid phase voltage; Δ u_dcThe voltage difference of the upper and lower capacitors at the DC side; i.e. i_c1,i_c2Respectively are direct current side upper and lower capacitance currents; i.e. i_npIs the neutral current; l is₁Is the grid side inductance value; l is₂Is an inverter side inductance value; c₁Is a filter capacitance value; c is the DC side up-down capacitance.

Further, the prediction model of each variable is obtained by discretization of a control type of the variables through a forward euler method, and the prediction formula is as follows:

wherein the content of the first and second substances,

the predicted value of the inverter output phase current at the moment of k + 1; i.e. i_2x(k) Outputting phase current for the inverter at the moment k;

the predicted value of the filter capacitor phase voltage at the moment k +1 is obtained; u. of_cx(k) Filtering capacitor phase voltage at the time k;the predicted value of the grid-connected phase current at the moment of k +1 is obtained; i.e. i_1x(k) Grid-connected phase current at the moment k; e.g. of the type_x(k) The voltage of the power grid phase at the moment k; u. of_x(k) Outputting phase voltage for the inverter at the moment k; Δ u_dc ^P(k +1) is a predicted value of the voltage difference between the upper capacitor and the lower capacitor on the direct current side at the moment of k + 1; Δ u_dc(k) The voltage difference between the upper capacitor and the lower capacitor on the direct current side at the moment k; t is_sIs a sampling period; h is_xThe influence factors of the inverter output phase current on the midpoint potential are determined according to different influences of the current on the midpoint potential under various switch states, and x ∈ { a, b, c } is provided above.

Further, the reference value of other variables is derived according to the given grid-connected current reference value,

will the network voltage e_a(k),e_b(k),e_c(k) Clark and park transformation to obtain e_d(k),e_q(k) (ii) a Reference current to be connected to gridSubjected to Clark and park transformation to obtainAnd (3) forming a vector:

the filter capacitor reference voltage vector and the inverter output reference current vector are calculated as:

the calculated component

Performing reverse park and Clark conversion to obtain reference values of filter capacitor voltage and inverter output currentAnd the reference value of each variable at the sampling moment of K +1 can be predicted by introducing a Lagrange extrapolation method.

Further, constructing each variable independent cost function according to the predicted value and the reference value:

the invention provides a predictive control device of a weighted-value-free multivariable sequential model of an LCL inverter, which is characterized by comprising a multivariable layering module, a predictive value calculation module, a cost function calculation module, an optimization space defining module and a control instruction module, wherein the multivariable layering module is used for calculating the predicted value of the LCL inverter;

the multivariable layering module is used for independently layering the variables according to the control sequence, so that the multivariable sequential control is easy to realize;

the predicted value calculation module is used for predicting the size of each variable in the next control period according to the discretization mathematical model by using the sampling value of each variable in the control period;

the cost function calculation module is used for calculating the error magnitude between the predicted value and the reference value of the variable in the next control period;

the optimizing space defining module is used for selecting partial optimal values as the optimizing space of the next control variable by sorting the cost function;

the control instruction module is used for applying the optimal control instruction selected by the sequential control to the inverter to drive the switching tube to act;

the number of the predicted value calculation module, the cost function calculation module and the optimization space definition module is equal to the number of variables minus one.

The invention has the beneficial effects that:

the LCL inverter non-weight multivariable sequential model predictive control method comprehensively considers the inverter output current, the filter capacitor voltage, the grid-connected current control and the direct-current side midpoint potential control, ensures that each variable can be directly and effectively controlled, improves the quality of the inverter grid-connected current and greatly improves the robustness of the system; through the step-by-step prediction and evaluation of each variable, only one variable is predicted and evaluated each time, a weight coefficient does not need to be distributed, and the complicated and strong accidental weight distribution work is omitted; in the prediction process of the LCL inverter non-weight multivariable sequential model, the optimization range is gradually reduced, the calculated amount is reduced, and the control efficiency is improved; the LCL inverter non-weight multivariable sequential model predictive control method provided by the invention has universality and can be applied to converters with other topologies.

Drawings

In order to more clearly illustrate the embodiments of the invention, it will be apparent that the drawings needed in the description of the embodiments are briefly presented, and that the drawings in the following description are only some embodiments of the invention.

Fig. 1 is a flowchart of a method for predictive control of an unweighted multivariable sequential model of an LCL inverter according to an embodiment of the present invention;

fig. 2 is a topological structure and a grid-connected schematic diagram of an LCL type three-level grid-connected inverter according to an embodiment of the present invention;

fig. 3 is a multivariable sequential model predictive control block diagram of an LCL type three-level inverter according to an embodiment of the present invention;

FIG. 4 is a simulation diagram comparing the output effects of the multivariable sequential model predictive control and the conventional method provided by the embodiment of the present invention;

FIG. 5 is a graph of a comparative simulation of the robustness of multivariable sequential model predictive control and a conventional method provided by an embodiment of the present invention;

fig. 6 is a schematic structural diagram of an LCL inverter non-weighted multivariable sequential model predictive control apparatus according to an embodiment of the present invention.

Detailed Description

Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are illustrative only for the purpose of explaining the present invention, and are not to be construed as limiting the present invention.

The LCL inverter non-weight multivariable sequential model predictive control method in the embodiment of the invention is shown in figure 1, and comprises the steps of layering multivariable and sequentially carrying out predictive optimization;

calculating a predicted value of the upper-layer variable in an initial optimization space, and substituting the predicted value into the variable cost function to sort the variables from small to large;

selecting part of the predicted value of the upper-layer variable as an optimization space of the lower-layer variable;

calculating a predicted value and sorting a cost function of the lower-layer variable, and gradually reducing an optimization space in the process;

and selecting the switching sequence which enables the cost function of the lowest layer variable to obtain the minimum value as the optimal control instruction to act on the system.

Further, the sequence of the multi-variable layering is direct-current side midpoint potential control, inverter output current control, filter capacitor voltage control and grid-connected current control from top to bottom in sequence.

Further, the selecting of the partial predicted value of the upper layer variable as the optimization space of the lower layer variable is specifically implemented as follows:

sorting cost function values of upper-layer variables from small to large, and screening out the smallest first n values;

and finding n predicted values corresponding to the minimum first n values, and taking the switch states corresponding to the n predicted values as an optimization space of the lower-layer variable.

Further, the lower layer variables are subjected to predicted value calculation and cost function sorting, and the optimization space is gradually reduced in the process, which is specifically realized as follows:

calculating a cost function value in an optimization space of a lower-layer variable, screening out the minimum first n-s values, and taking the switch states corresponding to the n-s predicted values as the optimization space of the lower-layer variable;

and the switching state of the optimization space of the lower variable gradually reduces the optimization space by sequentially decreasing the switching states of s.

Further, the constructing of the cost function includes,

establishing a multivariable mathematical model in the grid connection process of the LCL type three-level inverter to obtain a multivariable control type;

discretizing the control type of the multiple variables to obtain a prediction model of each variable;

and deducing reference values of other variables according to the given grid-connected current reference value, and constructing independent cost functions of the variables according to the predicted values and the reference values.

Further, a multivariable mathematical model in the LCL type three-level inverter grid connection process is established, and the method is specifically realized as follows: the grid-connected structure of the LCL three-level midpoint clamping type grid-connected inverter is shown in fig. 2, a direct-current voltage source supplies power to two series capacitors, the stable voltage of each capacitor is 1/2 of the direct-current voltage, and the premise that the three-level inverter works normally is that the voltage of a direct-current side capacitor is stabilized at a reference value. Each phase of bridge arm of the inverter is composed of 4 full-control power switching devices and two diodes, each phase of bridge arm has three switching states, three levels-Udc/2, 0 and Udc/2 are correspondingly output, the inverter output phase voltage is specifically shown in formulas (1) and (2), and the three-phase three-level inverter output voltage has 27 combinations in total as each phase has three switching states.

According to fig. 1, a mathematical model of an LCL three-level active midpoint clamp inverter is established:

wherein u is_cxIs the grid-connected filter capacitor voltage; u. of_xIs the inverse ofA converter output phase voltage; i.e. i_1xIs grid-connected side phase current; i.e. i_2xOutputting phase current for the inverter; e.g. of the type_xIs the power grid phase voltage; Δ u_dcThe voltage difference of the upper and lower capacitors at the DC side; i.e. i_c1,i_c2Respectively are direct current side upper and lower capacitance currents; i.e. i_npIs the neutral current; l is₁Is the grid side inductance value; l is₂Is an inverter side inductance value; c₁Is a filter capacitance value; c is the DC side up-down capacitance.

Further, the prediction model of each variable is obtained by discretization processing of a control type with multiple variables by a forward eulerian method, and is specifically realized as follows: applying forward Euler formula discretization to formulas (3) to (6), and utilizing a mathematical model under a discrete domain to deduce predicted values of the output current, the filter capacitor current, the grid-connected current and the direct current side voltage difference of the inverter at the next sampling time as follows:

wherein the content of the first and second substances,the predicted value of the inverter output phase current at the moment of k + 1; i.e. i_2x(k) Outputting phase current for the inverter at the moment k;the predicted value of the filter capacitor phase voltage at the moment k +1 is obtained; u. of_cx(k) Filtering capacitor phase voltage at the time k;the predicted value of the grid-connected phase current at the moment of k +1 is obtained; i.e. i_1x(k) Grid-connected phase current at the moment k; e.g. of the type_x(k) The voltage of the power grid phase at the moment k; u. of_x(k) Outputting phase voltage for the inverter at the moment k; Δ u_dc ^P(k +1) is a predicted value of the voltage difference between the upper capacitor and the lower capacitor on the direct current side at the moment of k + 1; Δ u_dc(k) The voltage difference between the upper capacitor and the lower capacitor on the direct current side at the moment k; t is_sIs a sampling period; l is₁,L₂The inductance values are respectively a grid-connected side inductance and an inverter side inductance; c, C₁Respectively a direct current side capacitor and a filter capacitance value; h is_xThe influence factor of the inverter output phase current on the midpoint potential is determined according to different influences of the current under various switch states on the midpoint potential, wherein x ∈ { a, b, c } is provided, and:

the prediction formula is:

wherein the content of the first and second substances,the predicted value of the inverter output phase current at the moment of k + 1; i.e. i_2x(k) Outputting phase current for the inverter at the moment k;

the predicted value of the filter capacitor phase voltage at the moment k +1 is obtained; u. of_cx(k) Filtering capacitor phase voltage at the time k;the predicted value of the grid-connected phase current at the moment of k +1 is obtained; i.e. i_1x(k) Grid-connected phase current at the moment k; e.g. of the type_x(k) The voltage of the power grid phase at the moment k; u. of_x(k) Outputting phase voltage for the inverter at the moment k; Δ u_dc ^P(k +1) is a predicted value of the voltage difference between the upper capacitor and the lower capacitor on the direct current side at the moment of k + 1; Δ u_dc(k) The voltage difference between the upper capacitor and the lower capacitor on the direct current side at the moment k; t is_sIs a sampling period; h is_xThe influence factors of the inverter output phase current on the midpoint potential are determined according to different influences of the current on the midpoint potential under various switch states, and x ∈ { a, b, c } is provided above.

Further, the derivation of the reference value of each of the other variables according to the given grid-connected current reference value is specifically realized as follows: knowing that the reference value of the given grid-connected current is a sine signal with known amplitude, the reference values of other controlled variables can be obtained by vector operation according to the reference value of the given grid-connected current and by utilizing the components under a two-phase synchronous rotating coordinate system:

will the network voltage e_a(k),e_b(k),e_c(k) Clark and park transformation to obtain e_d(k),e_q(k) (ii) a Reference current to be connected to gridSubjected to Clark and park transformation to obtainAnd (3) forming a vector:

the filter capacitor voltage reference vector and the inverter output current reference vector are calculated as:

the calculated component

Performing reverse park and Clark conversion to obtain reference values of filter capacitor voltage and inverter output currentAnd the reference value of each variable at the sampling moment of K +1 can be predicted by introducing a Lagrange extrapolation method.

The prediction of the reference value adopts a third-order lagrange extrapolation method, for example:

will the network voltage e_a(k),e_b(k),e_c(k) Clark and park transformation to obtain e_d(k),e_q(k) (ii) a Reference current to be connected to grid

Subjected to Clark and park transformation to obtainAnd (3) forming a vector:

the filter capacitor reference voltage vector and the inverter output reference current vector are calculated as:

the calculated component Performing reverse park and Clark conversion to obtain reference values of filter capacitor voltage and inverter output current

And the reference value of each variable at the sampling moment of K +1 can be predicted by introducing a Lagrange extrapolation method.

And further, constructing independent cost functions of all variables according to the predicted values and the reference values. The concrete implementation is as follows: the cost function is constructed by using the predicted values and the reference values of all the obtained variables, the existing multivariable model predictive control puts the inverter output current, the filter capacitor voltage, the grid-connected current control and the direct-current side midpoint potential control in the same cost function, and weight coefficients are distributed according to the influence degree on the output result, namely the constructed cost function is as follows:

wherein λ is_iA weighting factor representing the impact of the control quantity on the overall effect. It can be seen that the method needs to determine a plurality of weighting factors, and the selection of the weighting factors has no specific and uniform selection standard, and is often obtained approximately by a trial and error method, so that a large error is caused.

Aiming at the problems, the multivariate sequential model predictive control provided by the invention only calculates one variable at a time by predicting and evaluating each variable step by step without distributing weight coefficients, thereby omitting the complicated and strong accidental weight distribution work, and constructing an independent cost function for each variable according to a predicted value and a reference value:

next, taking the switching state of the three-level inverter 27 as an initial optimization space as an example, a process block diagram of the multivariate sequential model predictive control for selecting the optimal switching state layer by layer is shown in fig. 3 (the right side in fig. 3 is labeled to illustrate that an ellipse in the left-side flowchart represents the optimization space, and a rectangle represents prediction and evaluation calculation), and the following is specifically implemented:

the switching state of the three-level inverter 27 is used as an initial optimization space, substituted into a DC side midpoint potential prediction formula, and a cost function J is selected₁The minimum 9 switch states serve as a new optimization space. Namely, dc-side midpoint potential control: will 27 switch states S₁～S₂₇Substituting the prediction formula (10) to obtain 27 predicted values of the DC side capacitance voltage differenceSequentially substituting an incidence type (17) cost function, screening out the first 9 smallest values from the small value to the large value of the cost function value, and switching states S corresponding to the 9 predicted values_J1min1～S_J1min9And recording as a new optimizing space.

Next, inverter output current control: will switch state S_J1min1～S_J1min9Substituting the corresponding inverter output voltage into the prediction formula (7) to obtain 9 inversionsOutput current prediction valueThe predicted values are sequentially substituted into a cost function (18), and the first 6 switch states S are selected from the cost function values according to the sequence from small to large_J2min1～S_J2min6As a new optimization space.

And then, substituting the inverter output currents corresponding to 6 switch states in the optimization space into a filter capacitor voltage prediction formula, and selecting a cost function J₃The minimum 3 filter capacitor voltages are used as an updated optimization space, namely the filter capacitor voltage control: will select the switch state S_J2min1～S_J2min6Substituting the corresponding inverter output current into a prediction formula (8) to obtain 6 filter capacitor voltage prediction values

Substituting the predicted value into a cost function (19), and selecting the first 3 switch states S from the cost function value according to the sequence from small to large_J3min1～S_J3min3As the final optimization space.

Finally, filter capacitor voltages corresponding to 3 switch states in the optimization space are substituted into a grid-connected current prediction formula, and a cost function J is selected₄And (3) acting the switching sequence corresponding to the minimum grid-connected current as an optimal control instruction on the system, namely grid-connected current control: finally, the selected S is processed_J3min1～S_J3min3Substituting the prediction formula (9) to obtain the predicted values of 3 grid-connected currentsAnd substituting the predicted value into a cost function (20) to find out the grid-connected current predicted value which enables the cost function to be minimum, wherein the corresponding switching sequence is the optimal control instruction.

Based on the above embodiment, the present invention provides a prediction control apparatus for an unweighted multivariable sequential model of an LCL inverter, as shown in fig. 6, the apparatus includes a multivariable hierarchical module, a predicted value calculation module, a cost function calculation module, an optimization space definition module, and a control instruction module; the multivariable layering module is used for independently layering the variables according to the control sequence, so that the multivariable sequential control is easy to realize; the predicted value calculation module is used for predicting the size of each variable in the next control period according to the discretization mathematical model by using the sampling value of each variable in the control period; the cost function calculation module is used for calculating the error magnitude between the predicted value and the reference value of the variable in the next control period;

the optimizing space defining module is used for selecting partial optimal values as the optimizing space of the next control variable by sorting the cost function; the control instruction module is used for applying the optimal control instruction selected by the sequential control to the inverter to drive the switching tube to act;

and the number of the predicted value calculation module, the cost function calculation module and the optimization space definition module is equal to the number of variables minus one.

Furthermore, the multi-variable layering module is used for independently layering all control variables, completing the acquisition of the variables and other required signals in each layer and preparing data; sending the data collected in the layer module to a predicted value calculation module, and performing prediction calculation in an optimization space to obtain a plurality of corresponding predicted values; sending all the obtained predicted values to a cost function calculation module, and calculating the deviation of the predicted values and the reference values; and sending the deviation corresponding to each predicted value into an optimization space defining module to sort and select a part of control signals corresponding to the predicted values from small to large as an optimization space of a lower layer variable. And then, controlling the variables of the lower layer, and repeating the steps until the optimal control signal comprehensively considering all the variables is selected to act on the system.

The weighted-value-free multivariable sequential model prediction control is characterized in that model prediction is sequentially carried out on all variables, only one variable is calculated each time, on the premise of realizing multivariable control, setting of weight coefficients of all variables is omitted, workload is reduced, control accuracy is improved, an optimization searching space is gradually reduced in sequential model prediction, calculated amount is reduced, the optimization searching calculated amount in each actual control period is only 27+9+6+3 times which is 45 times, and is far less than the traditional 27+ 4 times which is 108 times, and calculation and control efficiency is greatly improved. Through the step-by-step prediction and evaluation of each variable, a weight coefficient does not need to be distributed, the optimizing space is gradually reduced in the process, and the calculated amount is reduced.

A multivariable sequential model predictive control simulation of the three-level active midpoint clamp type inverter is built by utilizing an MATLAB/Simulink module, the feasibility of the proposed control method is verified, and the parameters of a simulation system are shown in a table 1.

TABLE 1 simulation parameters

Parameter(s)	Size and breadth
		Voltage on the direct current side	600V
Network phase voltage	220V
		Frequency of the grid	50Hz
Sampling frequency	20kHz
		DC side capacitor	1500uF
Filter capacitor	50uF
		Electric network side inductor	1.5mH
Inverter side inductor	2.2mH

FIG. 4 is a comparison of the output effects of the multivariate sequential model predictive control proposed by the present invention and the conventional method, wherein FIG. 4(b) is an effect diagram of the predictive control using the conventional multivariate model under the same parameters, and the waveform distortion rate THD is 1.08%; fig. 4(a) is the multivariate sequential model predictive control provided by the invention, and it can be seen that the grid-connected current is regular, the THD is only 0.39%, and the effectiveness of the control algorithm provided by the invention is verified.

In order to verify the robustness of the multivariate sequential model predictive control, 5-order and 7-order harmonics are injected into the power grid voltage when t is 0.05s, the output performance of the two algorithms is compared, as shown in fig. 4, the output effect of the traditional control method after the power grid voltage is distorted is shown in fig. 5(b), the grid-connected current THD is increased from 1.08% to 4.43%, and the distortion is serious; fig. 5(a) shows the grid-connected current under the multivariate sequential model prediction control, and it can be seen that the grid-connected current THD increases from 0.39% to 1.29% after the grid distortion, and still has a good output effect. The conclusion can be obtained by analyzing the above results, the multivariable sequential model predictive control provided by the invention not only improves the grid-connected current output effect of the inverter, but also greatly improves the robustness of the system.

The above examples are preferred embodiments of the present invention, but the present invention is not limited to the above examples, and any other changes, modifications, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and they are included in the scope of the present invention.