A control method for a model prediction based reactive power compensator

Abstract

This invention discloses a model-based prediction control method for reactive power compensators, belonging to the field of reactive power compensator control technology. The method includes establishing a discretized prediction model for the AC side and constructing a current prediction equation; calculating a reference voltage vector and determining the sector location, then selecting and retaining the switching state vectors within the sector containing the reference voltage vector and adjacent sectors as a candidate set; using a serialized value function to perform hierarchical evaluation of the candidate vectors, selecting a first preferred vector subset with reactive current tracking error as the primary objective, and then determining the optimal switching vector with active current control or midpoint voltage balance as secondary objectives; calculating the optimized duty cycle of the optimal switching vector based on the deadbeat principle, and selecting a second voltage vector and duty cycle from the candidate vector set that further optimizes the value function, applying the synthesized voltage to the converter; this invention improves the steady-state accuracy and dynamic response performance of reactive power compensation while reducing online computation.

Description

Technical Field

[0001] This invention relates to the field of reactive power compensator control technology, and in particular to a control method for reactive power compensators based on model prediction. Background Technology

[0002] With the widespread integration of nonlinear loads, impulsive loads, and distributed renewable energy sources into power systems, reactive power fluctuations in the power grid are becoming increasingly frequent, and voltage stability issues are becoming more prominent. Installing dynamic reactive power compensation devices, such as Static Var Generators (SVG) or Static Synchronous Compensators (STATCOM), at key nodes of the power grid has become an important means to improve power factor, stabilize voltage, and improve power quality.

[0003] Model Predictive Control (MPC), an advanced control strategy based on a system mathematical model for forward prediction and online optimization, has gained widespread attention in the field of power electronic converter control in recent years due to its inherent constraint handling capabilities, multi-objective collaborative optimization capabilities, and rapid dynamic response characteristics. The basic workflow of MPC is as follows: In each sampling period, all available switch state vectors are substituted into the system prediction model one by one to calculate the predicted current value for the next time step for each vector. Then, the control effect of each vector is evaluated through a value function, and finally, the vector that minimizes the value function is selected as the output for the current period.

[0004] However, MPC faces the following technical challenges in its engineering applications with reactive power compensation devices: I. Excessive computational burden of ergonomic optimization in multi-level topologies. When an SVG uses a three-level or five-level converter topology, the number of available switch state vectors increases significantly compared to a two-level topology. Taking a three-level diode midpoint clamping topology as an example, it has as many as 27 available switch state vectors. Traditional MPC performs prediction calculations and value function evaluations on all vectors in each sampling period. When the control frequency increases to above 10kHz, the calculation time may approach or even exceed the sampling period, putting significant pressure on the real-time operation of the digital processor.

[0005] Second, there is a lack of systematic design methods for the weighting coefficients of multiple objectives in the value function. The control objectives of reactive power compensation devices typically include accurate reactive current tracking, active current control, and DC side midpoint voltage balance. Traditional MPC integrates multiple control objectives into a single value function through weighted summation. The weighting coefficients of each objective directly affect the control performance, but their selection lacks theoretical basis and relies heavily on engineering experience and trial and error.

[0006] III. The optimization range of the second vector in traditional dual-vector duty cycle MPC is limited. To improve the current ripple problem of single-vector MPC at lower switching frequencies, dual-vector duty cycle MPC was proposed. This type of method applies two voltage vectors within one control cycle and allocates their respective application times. However, traditional schemes typically limit the search range of the second vector to the zero vector, only adjusting the duration of the effective vector, and cannot change the direction of the synthesized voltage, thus limiting its effectiveness in improving current steady-state error and ripple. Summary of the Invention

[0007] The purpose of this invention is to provide a control method for reactive power compensators based on model prediction, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, this invention provides a model-predictive-based control method for a reactive power compensator, applicable to a reactive power compensation device in a multilevel converter topology, comprising the following steps: S1. Establish a discretized prediction model of the AC side of the reactive power compensation device in a three-phase stationary coordinate system, perform forward Euler discretization on the output current with the sampling period, and construct the current prediction equation for the next moment. S2. Calculate the reference voltage vector for the current control cycle, determine the sector position of the reference voltage vector on the space vector plane through coordinate transformation, filter the candidate voltage vector set based on the sector positioning information, and retain only the switch state vectors in the sector where the reference voltage vector is located and its adjacent sectors. S3. The candidate voltage vector set is evaluated hierarchically using a serialized value function: The first-level value function uses the absolute value of the difference between the reactive current reference value and the reactive current prediction value as the evaluation index. Three to four vectors that minimize the evaluation index of the first-level value function are selected from the candidate vector set to form the first preferred vector subset. The second-level value function uses the absolute value of the difference between the active current reference value and the active current prediction value as the evaluation index. The vector that minimizes the evaluation index of the second-level value function is selected from the first preferred vector subset as the optimal switching vector. S4. Calculate the optimal duty cycle of the optimal switching vector within one control cycle based on the deadbeat principle, so that the reactive current prediction value equals the reactive current reference value after the optimized duty cycle. Within the candidate voltage vector set determined in S2, with the optimal switching vector as the center, select the effective vector and zero vector adjacent to the optimal switching vector on the spatial vector plane and with only one bridge arm switching state change, as the candidate range of the second voltage vector. From the candidate range of the second voltage vector, select the vector that further reduces the evaluation index of the first-level value function as the second voltage vector, and calculate the duty cycle of the second voltage vector. Apply the combined voltage obtained by adding the product of the optimal switching vector and the optimized duty cycle to the product of the second voltage vector and the duty cycle of the second voltage vector to the reactive power compensation device converter.

[0009] Preferably, the current prediction equation in S1 is expressed as follows: ; in, for Predicted current value at time 10:00 for Current sampling value at time 10:00 for The converter output voltage vector applied at any given time, for The grid voltage at any given time, The equivalent resistance on the AC side. For the AC side equivalent inductance, The sampling period is , and the sampling period is . satisfy .

[0010] Preferably, the expressions for the first-order value function and the second-order value function in S1 are: ; ; in, For the first-order value function, For the second-order value function, , They are respectively Reference values ​​for reactive and active current at any given time. , They are respectively Predicted values ​​of reactive and active current at any given time.

[0011] Preferably, when the reactive power compensation device is a diode-clamped three-level topology or other multi-level topology with DC-side midpoint voltage balance requirements, the second-order value function... Replace with ,in , These are the upper and lower capacitor voltages on the DC side, respectively.

[0012] Preferably, the selection of the candidate voltage vector set in S2 satisfies: ; in, For the reference voltage vector, Candidate vector Spatial angular deviation from the reference voltage vector, This is a preset angle threshold.

[0013] Preferably, the optimal switching vector in S4 Optimization ratio Calculate using the following formula: ; in, The rate of change of reactive current under the action of the optimal switching vector. The rate of change of reactive current under zero vector action.

[0014] Preferably, the expression for the synthesized voltage in S4 is: ; in, For the synthesized voltage, For the second voltage vector, Let be the duty cycle of the second voltage vector, and satisfy . .

[0015] Preferably, it also includes S5: introducing a one-shot delay compensation mechanism, in Time calculation The predicted current value at time t is based on The predicted current value at time step S3 is evaluated using the value function of step S3 to determine the current prediction. The switching state that should be applied at any given time.

[0016] Preferably, the method further includes an online parameter calibration step: constructing a parameter observer based on the current prediction error, detecting changes in the equivalent resistance and inductance parameters of the AC side of the reactive power compensation device online, and automatically updating the corresponding parameters of the prediction model in step S1 when the parameter changes exceed a preset threshold.

[0017] Therefore, the present invention employs the above-mentioned control method for a reactive power compensator based on model prediction, which has the following beneficial effects: (1) This invention pre-screens candidate switch state vectors based on the sector position of the reference voltage vector, which effectively reduces the number of vectors that need to be predicted and evaluated in each control cycle. At the same time, it replaces weighted summation with hierarchical evaluation, avoiding the computational overhead of traversing all vectors, making real-time model predictive control more feasible at higher control frequencies.

[0018] (2) The present invention adopts a hierarchical evaluation strategy, which prioritizes reactive current tracking as the first-level screening target, and then takes into account active current control or midpoint voltage balance as the second-level selection target. It does not require combining multiple control targets into a single function through weighting coefficients, thus reducing the complexity of engineering debugging.

[0019] (3) This invention synthesizes two voltage vectors within one control cycle and expands the selection range of the second vector from the zero vector to the effective vector adjacent to the optimal vector, thereby expanding the search space for duty cycle optimization. This helps to reduce current ripple and steady-state tracking error while maintaining dynamic response speed.

[0020] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0021] Figure 1 This is a flowchart of a model-predictive control method for a reactive power compensator according to the present invention. Figure 2 This is a schematic diagram illustrating the spatial vector plane sector division and candidate vector screening of a model prediction-based reactive power compensator control method according to the present invention. Detailed Implementation

[0022] The following detailed description of embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely illustrates selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0023] Example like Figure 1 As shown, this invention provides a model-predictive control method for reactive power compensators, applicable to reactive power compensation devices employing multi-level converter topologies, such as static var generators (SVG) or static synchronous compensators (STATCOM), comprising the following steps: S1. Establish a discretized prediction model for the AC side of the reactive power compensation device in a three-phase stationary coordinate system. Perform forward Euler discretization on the output current with the sampling period to construct the current prediction equation for the next moment. The multilevel converter topology includes, but is not limited to, the three-level diode midpoint clamping topology.

[0024] Specifically, in a three-phase stationary coordinate system (abc coordinate system), the voltage equation for the AC side of the reactive power compensation device is established based on Kirchhoff's voltage law, and then the sampling period is used as the basis for the equation. The differential equation of the output current is discretized using forward Euler discretization. Forward Euler discretization is a first-order numerical integration method. Its core idea is to approximate the rate of change of the derivative over the entire sampling period with the derivative value at the current moment, transforming the continuous domain differential equation into a discrete domain difference equation, thereby obtaining the recursive relationship between the current value at the next moment and the current value at the current moment, the applied voltage vector at the current moment, and the grid voltage at the current moment.

[0025] The current prediction equation is expressed as follows: ; in, for Predicted current value at time 10:00 for Current sampling value at time 10:00 for The converter output voltage vector applied at any given time, for The grid voltage at any given time, The equivalent resistance on the AC side. For the AC side equivalent inductance, The sampling period is , and the sampling period is . satisfy The numerical stability of forward Euler discretization requires that the sampling period be much smaller than the electrical time constant of the system. In practical engineering, the AC side inductance of the reactive power compensation device... Millihenry level, resistance Electrical time constant in the milliohm to ohm range. The timescale is in the range of several milliseconds to tens of milliseconds; when the sampling frequency is selected as 10kHz to 20kHz (corresponding to...). When the time is 50μs to 100μs, the stability condition can be fully satisfied, and the discretized model can accurately reflect the dynamic behavior of the continuous domain system.

[0026] Furthermore, when the sampling frequency is much higher than the fundamental frequency of the power grid (50Hz / 60Hz), the variation in the power grid voltage within adjacent sampling periods is very small, and can be approximated as... This simplifies the process by treating the grid voltage as a constant value across adjacent sampling periods. This simplified assumption avoids the need for additional predictions of the grid voltage in each sampling period. It is obtained by real-time sampling and coordinate transformation of the three-phase voltage at the grid connection point.

[0027] Based on the prediction model constructed in step S1, an evaluation system for the value function is predefined to meet the requirements of subsequent multi-objective control of the reactive power compensation device. The expressions for the first-level value function and the second-level value function are: ; ; in, For the first-order value function, For the second-order value function, , They are respectively The reference values ​​for reactive and active current at any given time are output by the upper-level reactive power control loop or voltage control loop. , They are respectively The predicted values ​​of reactive and active current at each time step are calculated by combining the current prediction equation from step S1 with the converter output voltage corresponding to the candidate vector after coordinate transformation. It should be noted that the first-level value function uses reactive current tracking error as the evaluation index, and the second-level value function uses active current tracking error as the evaluation index. The two are executed step by step according to priority, rather than being weighted summed by weighting coefficients.

[0028] When the reactive power compensation device is a diode-clamped three-level topology or other multi-level topology with DC-side midpoint voltage balance requirements, the second-order value function... Replace with ,in , These represent the upper and lower capacitor voltages on the DC side, respectively. Maintaining a balance between these voltages is crucial for ensuring the safe and reliable operation of this type of topology converter and the quality of its output waveform. This balance is achieved through real-time sampling by a DC-side voltage sensor. and Whether to use midpoint voltage balance as the second-level evaluation index can be determined in advance based on the topology of the reactive power compensation device before actual implementation, without the need for dynamic switching during operation.

[0029] S2. Calculate the reference voltage vector for the current control cycle, determine the sector position of the reference voltage vector on the space vector plane through coordinate transformation, filter the candidate voltage vector set based on the sector positioning information, and retain only the switch state vectors in the sector where the reference voltage vector is located and its adjacent sectors.

[0030] like Figure 2 The diagram illustrates the spatial vector plane sector division and candidate vector selection in a three-level NPC topology. Figure 2 Reference voltage vector Taking sector I as an example, the 27 switch state vectors are distributed on the αβ plane. Based on the sector positioning information, only the vectors in sector I and the adjacent sectors VI and II are retained. Figure 2 (Solid marker), excluding vectors in sectors III, IV, and V ( Figure 2 Hollow markers), the number of candidate vectors was reduced from 27 to 9. Angle threshold. Set to 120°, indicated by a dashed arc.

[0031] Specifically, the reference voltage vector is obtained as follows: based on the reference values ​​of reactive and active currents, combined with the current grid voltage and current sampling values, the required reference voltage in the dq coordinate system is calculated by the current inner loop controller (such as a proportional-integral controller or a proportional-resonant controller). Then, it is transformed to the αβ stationary coordinate system through the inverse Park transform to obtain the reference voltage vector on the space vector plane. The space vector plane is typically divided into six main sectors, each with an angular range of 60°. Sector boundaries are defined by bisectors of adjacent effective voltage vectors. The sector number of the reference voltage vector can be determined by calculating the magnitudes and arctangent functions of its α-axis and β-axis components.

[0032] The selection of the candidate voltage vector set satisfies: ; in, For the reference voltage vector, Candidate vector Spatial angular deviation from the reference voltage vector, Preset angle threshold, angle threshold The selection is related to the sector partitioning method and topology type: in a typical 6-sector partitioning scheme, The angle can be set from 90° to 120° to ensure that all switch state vectors in the sector where the reference vector is located and in two adjacent sectors are covered. Taking the three-level diode midpoint clamping topology as an example, the total number of switch state vectors is 27. After adopting the above sector positioning and screening rules, the number of candidate vectors is reduced to 7 to 9, which significantly reduces the computational burden for subsequent value function evaluation.

[0033] S3. The candidate voltage vector set is evaluated hierarchically using a serialized value function: The first-level value function uses the absolute value of the difference between the reactive current reference value and the reactive current prediction value as the evaluation index. Three to four vectors that minimize the evaluation index of the first-level value function are selected from the candidate vector set to form the first preferred vector subset. The second-level value function uses the absolute value of the difference between the active current reference value and the active current prediction value as the evaluation index. The vector that minimizes the evaluation index of the second-level value function is selected from the first preferred vector subset as the optimal switching vector.

[0034] The hierarchical evaluation mechanism of the serialized value function filters according to the priority level of the control objectives: firstly, reactive current tracking is the primary objective for the first level of selection; then, active current control or midpoint voltage balance is considered for the second level of selection. This mechanism avoids the problems of traditional model predictive control, where the weighting coefficients lack theoretical basis, rely heavily on engineering experience, and involve repeated trial and error when multi-objective weighted summation is performed. In specific implementation, each vector in the candidate voltage vector set needs to be substituted into the prediction equation of step S1 to calculate the corresponding reactive current prediction value and active current prediction value, and then substituted into... and The expression performs a graded evaluation.

[0035] S4. Calculate the optimal duty cycle of the optimal switching vector within one control cycle based on the deadbeat principle, so that the reactive current prediction value equals the reactive current reference value after the optimized duty cycle. Within the candidate voltage vector set determined in S2, with the optimal switching vector as the center, select the effective vector and zero vector adjacent to the optimal switching vector on the spatial vector plane and with only one bridge arm switching state change, as the candidate range of the second voltage vector. From the candidate range of the second voltage vector, select the vector that further reduces the evaluation index of the first-level value function as the second voltage vector, and calculate the duty cycle of the second voltage vector. Apply the combined voltage obtained by adding the product of the optimal switching vector and the optimized duty cycle to the product of the second voltage vector and the duty cycle of the second voltage vector to the reactive power compensation device converter.

[0036] Specifically, the basic idea of ​​deadbeat control is to control the duration of the voltage vector's action, ensuring that the controlled variable accurately tracks the reference value at the end of a sampling period. Applying the deadbeat principle to reactive current control, that is, at the optimal switching vector... Under the effect of optimized duty cycle After the corresponding time period, the predicted reactive current value is exactly equal to the reference reactive current value. This leads to the duty cycle calculation formula. Here, "only one bridge arm switching state changes" means that in the three-phase bridge arms of the multilevel converter, only one phase's switching state changes (e.g., from state "1" to state "0", or from state "0" to state "-1"), while the other two phases remain unchanged. The physical significance of this vector selection strategy is that by selecting the nearest neighboring vector with the closest electrical distance to the optimal vector for synthesis, the coverage of the effective synthesized voltage is expanded while minimizing the number of switching operations, which helps to control switching losses while improving current ripple characteristics.

[0037] Optimal switching vector Optimization ratio Calculate using the following formula: ; in, The rate of change of reactive current under the action of the optimal switching vector. The rate of change of reactive current under zero vector action; the rate of change of reactive current and The calculation method is as follows: Substitute the corresponding voltage vector into the prediction equation in step S1 to obtain the change in reactive current, and then divide by the sampling period. You can get it immediately.

[0038] The expression for the combined voltage is: ; in, For the synthesized voltage, For the second voltage vector, Let be the duty cycle of the second voltage vector, and satisfy . .

[0039] Synthetic voltage The signal is converted into drive pulse signals for the power switching devices of each bridge arm by the space vector pulse width modulation module, driving the converter to output the desired compensation current. By expanding the search range of the second voltage vector from only the zero vector in the traditional scheme to include adjacent effective vectors, a finer voltage vector synthesis can be achieved within one control cycle, expanding the search space for duty cycle optimization, and helping to reduce the ripple of the compensation current and steady-state tracking error.

[0040] S5, introduces a one-shot delay compensation mechanism, Time calculation The predicted current value at time t is based on The predicted current value at time step S3 is evaluated using the value function of step S3 to determine the current prediction. The switching state that should be applied at any given time.

[0041] In digital control systems, there is an inherent time delay between current sampling, control algorithm calculation, and PWM duty cycle update. Specifically, in the... Sampled in each sampling period Afterwards, a complete model predictive control algorithm operation (including prediction calculation, sector selection, value function evaluation, and duty cycle calculation) is required to determine the voltage vector to be applied. This voltage vector can usually only be determined at the beginning of the next sampling period (i.e., The output can only be updated at a specific time. This means... The switch state that is determined at a specific time actually needs to be determined at... The control effect only applies to the system at a specific moment, causing a one-beat delay. If this delay is not compensated for, the control effect will deviate from the ideal model's predictive control. By using... Time-based predicted value replacement Using the sampled values ​​at each moment as the evaluation benchmark ensures that the optimal control quantity calculated in this cycle accurately corresponds to the system state at the beginning of the next cycle, effectively compensating for the one-step delay introduced by digital implementation.

[0042] Online parameter calibration steps: Construct a parameter observer based on the current prediction error to detect changes in the equivalent resistance and inductance parameters of the AC side of the reactive power compensation device online. When the parameter changes exceed the preset threshold, automatically update the corresponding parameters of the prediction model in step S1.

[0043] During long-term operation of the reactive power compensation device, the core of the AC-side filter inductor may experience saturation drift due to temperature changes or aging, and the winding resistance may also increase with rising temperature, causing the system's equivalent parameters to deviate from the design values. If the prediction model in step S1 continues to use the initial parameters without correction, the model's prediction accuracy will degrade with increasing operating time, thus affecting the accuracy of the value function evaluation and the final control effect. The specific implementation of the parameter observer is as follows: in each control cycle or every few control cycles, the current prediction value calculated by the prediction equation in step S1 is compared with the actual sampled current value to obtain the current prediction error; the prediction error is analyzed using the recursive least squares method or gradient descent method to identify the actual values ​​of the AC-side equivalent resistance R and equivalent inductance L online. To ensure the stability of online parameter calibration, a safety constraint mechanism is set: the magnitude of a single parameter update does not exceed 10% of the current parameter value to avoid sudden changes in the prediction model due to a single identification error; the new identified value is formally written into the prediction model only after passing the consistency check in multiple consecutive sampling cycles. This parameter online calibration step does not require the injection of additional test signals or interruption of normal reactive power compensation function, and can be performed online during normal system operation.

[0044] In terms of control system hardware implementation, this method can be deployed in a dual-core control architecture consisting of a Digital Signal Processor (DSP) and a Field Programmable Gate Array (FPGA). The DSP, as the main control chip, is responsible for executing the main model predictive control algorithm, including predictive model updates, reference voltage calculation and sector location, candidate vector screening, serialized value function evaluation, and dual-vector duty cycle calculation. The FPGA, as a coprocessor, is responsible for high-speed multi-channel synchronous sampling, overcurrent and overvoltage protection logic judgment, and space vector pulse width modulation pulse generation based on the synthesized voltage. Data exchange between the DSP and FPGA occurs via a high-speed parallel bus or serial peripheral interface. At control frequencies of 10kHz to 20kHz, the DSP needs to complete all the above calculations within one sampling cycle. The aforementioned sector screening and hierarchical evaluation strategies can keep the single-cycle computational load within the DSP's real-time processing capabilities.

[0045] Therefore, the present invention adopts the above-mentioned control method of reactive power compensator based on model prediction, which reduces the number of candidate vectors by sector positioning, avoids weight coefficient adjustment by hierarchical evaluation, and optimizes the synthesized voltage by extending the dual duty cycle. This reduces the amount of online calculation while improving the steady-state accuracy and dynamic response performance of reactive power compensation.

[0046] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A control method of a model prediction based reactive power compensator applied to a reactive power compensation device of a multi-level converter topology, characterized in that, Includes the following steps: S1. Establish a discretized prediction model of the AC side of the reactive power compensation device in a three-phase stationary coordinate system, perform forward Euler discretization on the output current with the sampling period, and construct the current prediction equation for the next moment. S2. Calculate the reference voltage vector for the current control cycle, determine the sector position of the reference voltage vector on the space vector plane through coordinate transformation, filter the candidate voltage vector set based on the sector positioning information, and retain only the switch state vectors in the sector where the reference voltage vector is located and its adjacent sectors. S3. The candidate voltage vector set is evaluated hierarchically using a serialized value function: The first-level value function uses the absolute value of the difference between the reactive current reference value and the reactive current prediction value as the evaluation index. Three to four vectors that minimize the evaluation index of the first-level value function are selected from the candidate vector set to form the first preferred vector subset. The second-level value function uses the absolute value of the difference between the active current reference value and the active current prediction value as the evaluation index. The vector that minimizes the evaluation index of the second-level value function is selected from the first preferred vector subset as the optimal switching vector. S4. Calculate the optimal duty cycle of the optimal switching vector within one control cycle based on the deadbeat principle, so that the reactive current prediction value is equal to the reactive current reference value after the effect of the optimized duty cycle. Within the candidate voltage vector set determined by S2, with the optimal switching vector as the center, select the effective vector and the zero vector that are adjacent to the optimal switching vector on the space vector plane and have only one bridge arm switching state change as the candidate range of the second voltage vector; select the vector from the candidate range of the second voltage vector that further reduces the evaluation index of the first-level value function as the second voltage vector, and calculate the duty cycle of the second voltage vector; apply the composite voltage obtained by adding the product of the optimal switching vector and the optimized duty cycle to the product of the second voltage vector and the duty cycle of the second voltage vector to the reactive power compensation device converter.

2. The method of claim 1, wherein, The current prediction equation in S1 is expressed as follows: ； in, for Predicted current value at time [time]. for Current sampling value at time 10:00 for The converter output voltage vector applied at any given time, for The grid voltage at any given time, The equivalent resistance on the AC side. For the AC side equivalent inductance, The sampling period is [value], and the sampling period is [value]. satisfy .

3. The control method for a reactive power compensator based on model prediction according to claim 1, characterized in that, The expressions for the first-order value function and the second-order value function in S1 are: ； ； in, For the first-order value function, For the second-order value function, , They are respectively Reference values ​​for reactive and active current at any given time. , They are respectively Predicted values ​​of reactive and active current at any given time.

4. The control method for a reactive power compensator based on model prediction according to claim 3, characterized in that: When the reactive power compensation device is a diode-clamped three-level topology or other multi-level topology with DC-side midpoint voltage balance requirements, the second-order value function... Replace with ,in , These are the upper and lower capacitor voltages on the DC side, respectively.

5. The control method for a reactive power compensator based on model prediction according to claim 1, characterized in that, The selection of the candidate voltage vector set in S2 satisfies: ； in, For the reference voltage vector, Candidate vector Spatial angular deviation from the reference voltage vector, This is a preset angle threshold.

6. The control method for a reactive power compensator based on model prediction according to claim 1, characterized in that, Optimal switching vector in S4 Optimization ratio Calculate using the following formula: ； in, The rate of change of reactive current under the action of the optimal switching vector. The rate of change of reactive current under zero vector action.

7. The control method for a reactive power compensator based on model prediction according to claim 6, characterized in that, The expression for the combined voltage in S4 is: ； in, For the synthesized voltage, For the second voltage vector, Let be the duty cycle of the second voltage vector, and satisfy . .

8. The control method for a reactive power compensator based on model prediction according to claim 1, characterized in that, It also includes S5: introducing a one-shot delay compensation mechanism, in Time calculation The predicted current value at time t is based on The predicted current value at time step S3 is evaluated using the value function of step S3 to determine the current prediction. The switching state that should be applied at any given time.

9. The control method for a reactive power compensator based on model prediction according to claim 1, characterized in that: It also includes an online parameter calibration step: a parameter observer is constructed based on the current prediction error to detect the changes in the equivalent resistance and inductance parameters of the AC side of the reactive power compensation device online. When the parameter changes exceed the preset threshold, the corresponding parameters of the prediction model in step S1 are automatically updated.

Images