Voltage source rectifier and its predictive power control method

Abstract

This application discloses a predictive power control method for a voltage source rectifier, comprising: constructing a causal subspace predictor, which characterizes the mapping relationship between the system state and the control input; constructing a control objective function, configured to balance the system state deviation and the increment of the control input, wherein the system state deviation is the deviation between the system state estimate output by the causal subspace predictor and the system state reference; combining the causal subspace predictor with the control objective function, aiming to minimize the control objective function to obtain the optimal control input, and generating a switching state based on the optimal control input to control the voltage source rectifier. This method is completely independent of physical parameters, enabling the voltage source rectifier to maintain relatively excellent control performance under different operating conditions.

Description

Technical Field

[0001] This application relates to the field of power electronic control, and in particular to a voltage source rectifier and its predictive power control method. Background Technology

[0002] Voltage source rectifiers are widely used in industrial applications, capable of operating at approximately unity power factor under sinusoidal input current conditions. Current mainstream control strategies, such as current vector-oriented control and direct power control, are highly dependent on the accurate mathematical model parameters of the controlled system. When changes in actual operating conditions cause parameter drift, model mismatch occurs, leading to a decline in control performance.

[0003] To reduce dependence on parameters, existing technologies have introduced extended state observers for perturbation estimation. However, linear observers are difficult to accurately describe the nonlinear dynamics of the system, and gain design still partially depends on prior parameters. Summary of the Invention

[0004] This application provides a voltage source rectifier and a predictive power control method. This method is completely independent of physical parameters, enabling the voltage source rectifier to maintain relatively excellent control performance under different operating conditions.

[0005] In a first aspect, this application provides a predictive power control method for a voltage source rectifier, comprising: Construct a causal subspace predictor, which represents the mapping relationship between system state and control input; A control objective function is constructed, which is configured to weigh the system state deviation against the increment of the control input. The system state deviation is the deviation between the system state estimate output by the causal subspace predictor and the system state reference. By combining the causal subspace predictor with the control objective function, the optimal control input is obtained with the goal of minimizing the control objective function. Based on the optimal control input, the switching state is generated to control the voltage source rectifier.

[0006] In one embodiment, the predictive power control method further includes: A power model for a voltage source rectifier is constructed. The power model characterizes how the instantaneous power state of the power grid system in the next sampling period is dynamically related to the current instantaneous power state, the grid voltage, and the rectifier output voltage. The instantaneous power state of the power grid is taken as the system state, and the relevant terms of the grid voltage and rectifier output voltage in the power model are defined as control inputs.

[0007] In one embodiment, the causal subspace predictor comprises three linearly superimposed parts: a first part characterizing the influence of historical system states and historical control inputs on future system states; a second part characterizing the influence of future control inputs on future system states, with the coefficient matrix of the second part configured as a lower triangular matrix; and a third part characterizing the influence between future system states, with the coefficient matrix of the third part configured as a lower triangular matrix with zero on the main diagonal.

[0008] In one embodiment, the predictive power control method further includes: Acquire historical data, including the historical instantaneous power state of the power grid system and the historical composite control inputs of the voltage source rectifier; Based on historical data, a regression vector is constructed for each step of the prediction of the causal subspace predictor, so as to transform the causal subspace predictor into a linear regression form of the dot product of the coefficient vector to be identified and the regression vector. The update law of the coefficient vector is obtained according to the iterative formula of the normalized least mean square algorithm. Based on the update law, the coefficient vector is updated step by step, and the updated coefficient vectors are recombined to update the coefficient matrix of the causal subspace predictor.

[0009] In one embodiment, the update law is expressed by the following formula: ; In the formula, This indicates that the causal subspace predictor is at the ... k In the first cycle i The coefficient vector of step prediction, This indicates that the causal subspace predictor is at the ... k -1 cycle in the first i The coefficient vector of step prediction, This indicates that the causal subspace predictor is at the ... k -1 cycle in the first i The transpose of the predicted coefficient vector. This indicates that the causal subspace predictor is at the ... k In the first cycle i The system state obtained by step prediction, Indicates that for the first i The regression vector constructed by step prediction, Indicates the learning rate. To prevent extremely small constants with a denominator of zero.

[0010] In one embodiment, the predictive power control method further includes: The causal subspace predictor, which consists of three parts including linear superposition, is transformed by merging the third part with the prediction target of the causal subspace predictor. Then, matrix inversion is used to decouple the causal subspace predictor from the future system state to obtain a practical subspace predictor. The future control input sequence in the practical subspace predictor is decomposed into the initial control quantity and the incremental control input sequence to obtain the input increment type practical subspace predictor that characterizes the system state with respect to the control input increment. The input increment type practical subspace predictor is combined with the control objective function, and the optimal control input increment is obtained with the goal of minimizing the control objective function. The optimal control input increment is superimposed with the control input of the previous cycle to obtain the optimal control input.

[0011] In one embodiment, the control objective function is configured as a linear superposition of the norm of the system state deviation and the norm of the control input increment.

[0012] In one embodiment, the switch state is generated using finite set control based on the optimal control input.

[0013] In one embodiment, the predictive power control method further includes: A first cost function, a second cost function, a third cost function, and a fourth cost function are constructed. The first cost function is configured to quantify the error between the candidate output voltage vector and the reference voltage vector. The second cost function is configured to measure the change in switching state. The third cost function is configured to determine the polarity relationship between the neutral point voltage and current. The fourth cost function is configured to introduce tolerance constraints to comprehensively evaluate the neutral point potential balance capability. Based on the optimal control input, a finite set is used to select the optimal voltage vector and its corresponding switching state that minimizes the first cost function: When the obtained optimal voltage vector is a large vector, the unique switching state corresponding to the optimal voltage vector is directly applied. When the obtained optimal voltage vector is zero, among the three corresponding redundant switching states, select the switching state that minimizes the second cost function and does not disrupt the midpoint potential balance. When the obtained optimal voltage vector is a small vector, the switching state that minimizes the second cost function is selected from the two corresponding redundant switching states; if the midpoint potential exceeds the tolerance limit, the switching state that can reduce the fourth cost function is selected based on the polarity relationship between the neutral point voltage and current of the third cost function. When the obtained optimal voltage vector is the midpoint vector, if the fourth cost function is not equal to 2, the corresponding switching state is applied to facilitate the midpoint potential balance; otherwise, the suboptimal switching state of the small vector is selected, and the voltage vector is screened and processed again.

[0014] Secondly, this application also provides a voltage source rectifier, which employs the predictive power control method of the voltage source rectifier in the first aspect.

[0015] The aforementioned predictive power control method for voltage source rectifiers constructs a causal subspace predictor, which characterizes the mapping relationship between system state and control input. It also constructs a control objective function, configured to balance system state deviation with the increment of control input. The system state deviation is the discrepancy between the system state estimate output by the causal subspace predictor and the system state reference. By combining the causal subspace predictor with the control objective function, and minimizing the control objective function, the optimal control input is obtained. Based on this optimal control input, the switching states are generated to control the voltage source rectifier. This method, without relying on the physical parameters of the controlled system, enables the voltage source rectifier to maintain relatively excellent control performance under different operating conditions, exhibiting excellent steady-state and dynamic control performance, and achieving robust power control completely independent of physical parameter dependence. Attached Figure Description

[0016] Figure 1 A flowchart of a predictive power control method for a voltage source rectifier in one embodiment; Figure 2 This is a flowchart for selecting the optimal switching state in one embodiment; Figure 3 This is a simulation waveform diagram of the three-phase current on the AC side in one embodiment; Figure 4 This is a simulation waveform diagram of the actual instantaneous active power and reactive power on the grid side in one embodiment; Figure 5 This is a simulated waveform of the DC-side output voltage in one embodiment; Figure 6 This is a simulated waveform diagram of the grid-side current harmonics in one embodiment. Detailed Implementation

[0017] The present application will be described in detail below with reference to the specific embodiments shown in the accompanying drawings. However, these embodiments do not limit the present application. Any structural, methodological, or functional modifications made by those skilled in the art based on these embodiments are included within the protection scope of the present application.

[0018] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0019] In one embodiment, such as Figure 1 As shown, a predictive power control method for a voltage source rectifier is provided, including the following steps: Step 101: Construct a causal subspace predictor, which represents the mapping relationship between the system state and the control input; Specifically, the system state includes instantaneous active and reactive power, and the control inputs include grid voltage and rectifier output voltage. Based on historical system states and historical control inputs, a historical data vector is constructed.

[0020] Furthermore, the predictions of the causal subspace predictor are represented as a linear superposition of three parts: the influence of historical data, the influence of future inputs, and the influence of the future state itself. To achieve strict causality, the coefficient matrix of the influence of future inputs is configured as a lower triangular matrix, and the coefficient matrix of the influence of the future state itself is configured as a lower triangular matrix with zero on the main diagonal, thus ensuring that the prediction never uses future information.

[0021] Furthermore, the coefficient matrix is ​​estimated online. For each prediction step i, a regression vector and prediction target are constructed, transforming the problem into a linear regression form. The normalized least mean square algorithm is used to recursively update the parameter vector online, reconstructing the coefficient matrix estimate. Based on the calculated coefficient matrix estimate, the final causal subspace predictor is obtained, completing the data mapping relationship from control input to system state.

[0022] Step 102: Construct the control objective function, which is configured to balance the system state deviation with the increment of the control input. The system state deviation is the deviation between the system state estimate output by the causal subspace predictor and the system state reference. The system state deviation quantifies the error between the future system state estimate output by the causal subspace predictor and the given system state reference value; the control input increment quantifies the magnitude of the change in the future control sequence relative to the previous moment, used to constrain the severity of the control action to smooth the control signal.

[0023] By configuring independent weight matrices for the system state deviation and the increment of the control input, the control objective function can dynamically balance the system state deviation and the increment of the control input, avoiding excessive changes in the control input while keeping the system state deviation small, thus ensuring system stability.

[0024] Step 103: Combine the causal subspace predictor with the control objective function, aim to minimize the control objective function, obtain the optimal control input, and generate the switching state based on the optimal control input to control the voltage source rectifier.

[0025] Specifically, the real-time updated causal subspace predictor is substituted into the control objective function to minimize the control objective function value, thereby obtaining the optimal control input increment sequence that optimizes the overall system performance. The first element of the optimal control input increment sequence is then taken as the optimal control input at the current moment.

[0026] Based on the optimal control input and real-time grid voltage, a reference voltage vector is calculated. Under a finite set control framework, all possible switching states are traversed. By evaluating multiple factors such as voltage tracking error, switching frequency, and midpoint potential balance, the optimal switching state is selected. The power transistors of the voltage source rectifier are then driven based on the optimal switching state, thereby achieving precise and robust control of the system.

[0027] In this embodiment, a data-driven causal subspace predictor is constructed, eliminating dependence on physical parameters and maintaining superior control performance even when system parameters drift or operating conditions change. By constructing a control objective function that balances system state deviation with the increment of control input, high-precision power point tracking can be achieved while effectively constraining abrupt changes in control variables, resulting in smoother control actions and helping to reduce switching losses and current harmonics. Based on the optimal control solution, switching states are generated, enabling fast and direct closed-loop control and achieving robust power control completely independent of physical parameters.

[0028] In one embodiment, the predictive power control method further includes: A power model for a voltage source rectifier is constructed. The power model characterizes how the instantaneous power state of the power grid system in the next sampling period is dynamically related to the current instantaneous power state, the grid voltage, and the rectifier output voltage. The instantaneous power state of the power grid is taken as the system state, and the relevant terms of the grid voltage and rectifier output voltage in the power model are defined as control inputs.

[0029] Specifically, a power model of the active rectifier is established to characterize the instantaneous power state of the power grid system in the next sampling period. αβ The formula in the coordinate system is: ; in, k For the first k Each sampling period, P and Q These represent the instantaneous active and reactive power of the power grid, respectively. T s Indicates the control period. R g and L g These represent the filter resistor and inductor values ​​of the RL filter, respectively. v gα and v gβ These represent voltage source rectifiers in... α β The output voltage value in the coordinate system. e gα and e gβ They represent the power grid at αβ Voltage in coordinate system The 2-norm of the grid voltage. ω g It represents the rotational angular frequency of the grid voltage.

[0030] This equation characterizes the instantaneous power state in the next sampling period (k+1). From the current instantaneous power state Grid voltage and rectifier output voltage The dynamic relationship between them.

[0031] Furthermore, the instantaneous power of the power grid Defined as system state x (k); Integrate the relevant terms of grid voltage and rectifier output voltage in the power model and define them as a composite control input. u ( k ),Right now: .

[0032] In one embodiment, the causal subspace predictor comprises three linearly superimposed parts: the first part characterizes the influence of historical system states and historical control inputs on future system states; the second part characterizes the influence of future control inputs on future system states, and the coefficient matrix of the second part is configured as a lower triangular matrix; the third part characterizes the influence between future system states, and the coefficient matrix of the third part is configured as a lower triangular matrix with zero on the main diagonal.

[0033] Specifically, the equation for the causal subspace predictor is shown below. The causal subspace predictor decomposes the future system state prediction output into a linear superposition of three parts: .

[0034] in, p and f This represents the length of the historical time window and the future window, and satisfies... f < p . x =[ PQ ] T For system state variables, W p Represents a historical data vector. U f For future input vectors, X f This will be the output vector for the future. L w , L u and L x These are the historical coefficient matrix, the input coefficient matrix, and the output coefficient matrix, respectively. φ , ρ , θ and γ These are the coefficients for historical output, historical input, future input, and future output, respectively. The coefficient in the upper right corner represents the column number of the matrix it belongs to, and the coefficient in the lower right corner represents the data time it is associated with. L u and L x These are a lower triangular matrix and a lower triangular matrix with 0 on the main diagonal, respectively.

[0035] The first part is L w W p ( k ), L w W p ( kThis characterizes the impact of historical system states and historical control inputs on future system states. Among them, W p (k) is a historical data vector containing data from time (k-k) onwards. k p +1 to k Historical system status x And from time k p arrive k 1's historical control input u Historical coefficient matrix L w Each element ( φ , ρ This is used to quantify the different impact weights of these historical data on the future state of the system.

[0036] Part Two is L u U f ( k ), L u U f ( k This characterizes the impact of future control inputs on the future system state. Among them, U f ( k Let be the future input vector, representing the vector from the current time step. k The Beginning of the Future f Step-control input sequence. Input coefficient matrix. L u It is configured as a lower triangular matrix, and the structure of the lower triangular matrix can guarantee the accuracy of predicting the future... i Step state x ( k + i When, it depends only on the current and future (i.e., the first and last terms). i The control input before the step (i.e.) u ( k )arrive u ( k + i 1), without relying on more future inputs (e.g. u ( k + i This isolates the non-causal effects of future inputs on future outputs.

[0037] The third part is L xX f ( k ), L x X f ( k This represents the interactions between future system states. Output coefficient matrix L x It is configured as a lower triangular matrix with zeros on the main diagonal. Zeros on the main diagonal represent the state at any future time. x ( k + i The influence coefficient of the element on itself at any given moment is zero, thus avoiding logical loops. The lower triangular structure represents the future... i Step state x ( k + i It may be influenced by its previous future state. x ( k +1) to x ( k + i 1) will have an impact, but will not be affected by it. x ( k + i The influence of subsequent states. This structure ensures that the prediction calculations are causal and recursive.

[0038] Based on the above description, the predictive power control method provided in this application constructs an improved causal subspace predictor, making the coefficient matrix of the future control input a lower triangular matrix, and making the coefficient matrix of the future system state a lower triangular matrix with zero on the main diagonal. This allows the improved predictor to effectively isolate future noise, ensure strict causality, and eliminate closed-loop bias.

[0039] In one embodiment, the predictive power control method further includes: Acquire historical data, including the historical instantaneous power state of the power grid system and the historical composite control inputs of the voltage source rectifier; Based on historical data, a regression vector is constructed for each step of the prediction of the causal subspace predictor, so as to transform the causal subspace predictor into a linear regression form of the dot product of the coefficient vector to be identified and the regression vector. The update law of the coefficient vector is obtained according to the iterative formula of the normalized least mean square algorithm. Based on the update law, the coefficient vector is updated step by step, and the updated coefficient vectors are recombined to update the coefficient matrix of the causal subspace predictor.

[0040] Specifically, in each control cycle kTo obtain historical data, the required historical data includes: the historical instantaneous power state of the power grid system, i.e., system state variables. x =[ PQ ] T Historical composite control inputs of voltage source rectifiers, i.e., control input variables Defined as: .

[0041] By concatenating the two types of data vertically in chronological order, a historical data vector is constructed. W p ( k ), W p ( k The structure of ) is:

[0042] in, p Indicates the length of the historical time window.

[0043] To estimate the coefficients of the online predictor, it is necessary to prepare for the future. f Each step in the step prediction (denoted as ) i , i =1, 2, ..., Regression vectors are constructed separately. Further, by shifting the causal subspace predictor forward, the following formula is obtained: .

[0044] Based on this, the normalized least mean square algorithm can be used row by row to... L w , L u and L x Make an estimate for the first i Step prediction: Define local prediction target : .

[0045] Constructing dimensionality-reduced observation vectors : .

[0046] Define the corresponding dimensionality reduction parameter vector : .

[0047] in, Indicates the first i The local prediction target of the step, For the reduced-dimensional observation vector, To and The corresponding dimensionality reduction parameter vector, represent L w The first of the matrix i Entire row. φ , ρ , θ and γ These are the coefficients for historical output, historical input, future input, and future output, respectively. The coefficient in the upper right corner represents the column number of the matrix it belongs to, and the coefficient in the lower right corner represents the data time it is associated with.

[0048] The coefficient vector is updated recursively using the normalized least mean square algorithm. , The update law is: .

[0049] in, This indicates that the causal subspace predictor is at the ... k In the first cycle i The coefficient vector of step prediction, This indicates that the causal subspace predictor is at the ... k -1 cycle in the first i The coefficient vector of step prediction, This indicates that the causal subspace predictor is at the ... k -1 cycle in the first i The transpose of the predicted coefficient vector. This indicates that the causal subspace predictor is at the ... k In the first cycle i The system state obtained by step prediction, Indicates that for the first i The regression vector constructed by step prediction, Indicates the learning rate. To prevent extremely small constants with a denominator of zero.

[0050] Based on the concatenation order of the dimensionality-reduced parameter vectors, the following steps are performed: f This is the second time I have obtained f The updated coefficient vector corresponding to all coefficient matrices. From each updated coefficient vector... Extracting elements from the matrix and reconstructing them yields estimates of the three coefficient matrices of the causal subspace predictor. , and : ; ; .

[0051] in, , and Represent , and The estimated matrix is ​​in the th i line, number j Column elements, Representative vector The m-th component.

[0052] In this embodiment, the recombination rule ensures that It is a lower triangular matrix. It is a lower triangular matrix with a main diagonal of 0, thus maintaining the causality of the predictor.

[0053] In one embodiment, the predictive power control method further includes: The causal subspace predictor, which consists of three parts including linear superposition, is transformed by merging the third part with the prediction target of the causal subspace predictor. Then, matrix inversion is used to decouple the causal subspace predictor from the future system state to obtain a practical subspace predictor. The future control input sequence in the practical subspace predictor is decomposed into the initial control quantity and the incremental control input sequence to obtain the input increment type practical subspace predictor that characterizes the system state with respect to the control input increment. The input increment type practical subspace predictor is combined with the control objective function, and the optimal control input increment is obtained with the goal of minimizing the control objective function. The optimal control input increment is superimposed with the control input of the previous cycle to obtain the optimal control input.

[0054] Specifically, the corresponding formula for the linearly superimposed three-part causal subspace predictor is:

[0055] Take the third part on the right side of the equation. Move to the left side of the equation, and compare it with the predicted target. By merging, we get: .

[0056] in, 2 f Line 2 f The identity matrix of columns. Because A lower triangular matrix with zero on its main diagonal, guaranteeing... It is reversible. Multiply both sides of the above formula on the left. Using matrix inversion to predict the target Decoupling leads to a practical subspace predictor: ; Define the historical state transition matrix Future control response matrix Then, we can simplify to get: .

[0057] The future control input sequence in the practical subspace predictor The control input is decomposed into an initial control variable and an incremental control input sequence. The incremental control input is defined. ,but It can be decomposed into control quantities based on the previous time step. Initial control quantity and incremental control input sequence sum: .

[0058] l will use the above future control input sequence Substituting the expression into the practical subspace predictor formula, we obtain the input-increment type practical subspace predictor formula characterizing the system state with respect to the control input increment: .

[0059] in, F Indicates free response, H Δ Represents the incremental response matrix. , , I This represents a 2x2 identity matrix.

[0060] The above input incremental practical subspace predictor Combined with the defined control objective function, for example, the control objective function can be expressed by the following formula: .

[0061] in, Indicates the system's reference power. Q and R This represents the custom output error weight matrix and control input weight matrix.

[0062] It should be noted that the reference power include and reactive power reference power Typically, it is 0. Based on the principle of DC bus energy conservation, the optimal active power command is calculated online. (Right now Optimal active power command The calculation is as follows: Based on the principle of energy conservation of instantaneous active power in the system, the energy state variable of the DC bus is defined as follows: ,in, This is the DC bus voltage. Therefore, the continuous-time dynamic physical equation for the output voltage can be expressed as: ; in, This is the optimal active power command; The known DC-side support capacitor parameters; Considered to be caused by an unknown DC load The resulting unmodeled dynamic parameters This indicates the resistance value of the load resistor.

[0063] The system is discretized and reconstructed using a first-order subspace model. In the... k After one sampling period, the system's state update equation is reconstructed as follows: . Define a regression vector containing historical running data. and the parameter vector to be identified. The normalized least mean square algorithm is used to update the model parameters online using the collected system input and output data. The adaptive parameter update rule is as follows: .

[0064] in, μ This is the adaptive learning step size of the algorithm, used to strike a balance between parameter convergence speed and noise resistance stability; To prevent division by zero anomalies involving tiny normal numbers.

[0065] Setting the system energy state in the future N Smoothly approaching the reference value within each control step. (corresponding to DC voltage reference value) That is, the target energy trajectory is set as follows: .

[0066] The model parameters obtained from this online identification , , and target energy trajectory By substituting the inverse function of the model, the current control period that enables the energy state to track the target can be analytically solved. Optimal active power command : .

[0067] in, Here, N represents the DC voltage reference value, and N is the set step size for the target to reach the reference value.

[0068] To control the objective function Minimize as the objective, for By taking the derivative and setting it to zero, the optimal control input increment sequence can be obtained. : ; Based on the principle of model predictive control, only the optimal control input increment sequence is taken. The first element in As the optimal control input increment. The optimal control input increment is then used as the optimal control input increment. Control input from the previous cycle Superposition yields the optimal control input applied to the system at the current moment. : .

[0069] in, This refers to the optimal control command used to calculate the reference voltage vector and generate the switching state.

[0070] In one embodiment, the switch state is generated using finite set control based on the optimal control input.

[0071] Specifically, based on the optimal control input at the current moment and the sampled grid voltage Calculate the reference control voltage vector of the voltage source rectifier in the αβ coordinate system: .

[0072] Switching state selection is performed within a finite set control framework. The output voltage vector of the three-level converter can be categorized into four types based on amplitude: medium, large, small, and zero, corresponding to a total of 27 switching states. The process iterates through the output voltage vectors corresponding to all candidate switching states, minimizing the fundamental cost function... The optimal voltage vector is initially determined. Based on the type of the optimal voltage vector, a differentiation rule is applied to determine the final applied switching state.

[0073] in, and This is the candidate output voltage vector set of the voltage source rectifier in the αβ coordinate system.

[0074] In one embodiment, selecting the optimal switching state includes the following steps: Step 201: Construct a first cost function, a second cost function, a third cost function, and a fourth cost function. The first cost function is configured to quantify the error between the candidate output voltage vector and the reference voltage vector. The second cost function is configured to measure the change in switching state. The third cost function is configured to determine the polarity relationship between the neutral point voltage and current. The fourth cost function is configured to introduce tolerance constraints to comprehensively evaluate the neutral point potential balance capability. Step 202: Based on the optimal control input, use a finite set to filter the optimal voltage vector and its corresponding switching state that minimizes the first cost function: Step 203: When the obtained optimal voltage vector is a large vector, directly apply the unique switching state corresponding to the optimal voltage vector; Step 204: When the obtained optimal voltage vector is a zero vector, select the switching state that minimizes the second cost function and does not disrupt the midpoint potential balance from the three corresponding redundant switching states. Step 205: When the obtained optimal voltage vector is a small vector, select the switching state that minimizes the second cost function from the two corresponding redundant switching states; if the midpoint potential exceeds the tolerance limit, select the switching state that can reduce the fourth cost function based on the polarity relationship between the neutral point voltage and current of the third cost function. Step 206: When the obtained optimal voltage vector is the midpoint vector, if the fourth cost function is not equal to 2, then apply the corresponding switching state to facilitate the midpoint potential balance; otherwise, select the suboptimal switching state of the small vector and re-screen and process the voltage vector.

[0075] Specifically, in step 201, a first cost function, a second cost function, a third cost function, and a fourth cost function are constructed: First cost function This is used to quantify the tracking error between the candidate output voltage vector and the reference voltage vector; Second cost function It is used to measure changes in switching states in order to optimize switching frequency; Third cost function By determining the neutral point voltage ) and neutral point current The polarity relationship is used to assess the balance trend of the vector pair's midpoint potential; Fourth cost function A voltage tolerance limit is introduced based on the third cost function. To comprehensively evaluate the ability to balance the midpoint potential.

[0076] in, This indicates the switching state of the three bridge arms of the voltage source rectifier. Indicates the neutral point voltage. Represents the neutral point current. This indicates the preset voltage tolerance limit.

[0077] In step 202, the optimal control input is obtained based on the causal subspace predictive control. Calculated reference voltage vector Iterate through all 27 switching states of the three-level rectifier and select the optimal voltage vector and its corresponding set of switching states that minimizes the first cost function.

[0078] In step 203, if the optimal voltage vector is a large vector, the upper and lower capacitors discharge simultaneously, the neutral point current is zero, and the balance is not disrupted. Therefore, the switching state is directly applied.

[0079] In step 204, if the optimal voltage vector is zero, then in the three redundant switching states, neither the upper nor lower capacitors will discharge. The selection principle is to minimize the switching frequency and choose the switching state that minimizes the second cost function.

[0080] In step 205, if the optimal vector is a small vector, corresponding to two redundant switch states, the state that minimizes the second cost function is selected. If the midpoint potential offset exceeds the tolerance limit at this time... Then, based on the judgment of the third cost function, it is necessary to select the switching state that makes the value of the third cost function smaller (usually negative, indicating that it is beneficial to balance) from the two redundant states, so as to reduce the adverse effect on the midpoint potential. In step 206, if the optimal vector is the median vector, then the fourth cost function is calculated. If the value of the fourth cost function is... This indicates that the vector is beneficial for midpoint balance or offset within tolerance, therefore its corresponding switch state should be applied; otherwise ( This indicates that the midpoint potential offset exceeds the limit and the vector exacerbates the imbalance. In this case, the optimal vector should be abandoned, and a candidate voltage vector (usually a small vector) that makes the first cost function value the second smallest should be selected. Then, the process should jump to the corresponding vector type processing flow in steps 203 to 205 for re-evaluation and screening.

[0081] Through the above steps, the optimal switching state is finally output and applied to the voltage source rectifier.

[0082] Based on the aforementioned predictive power control method for voltage source rectifiers, the proposed predictive power control method was verified through simulation. The simulated voltage source rectifier system parameters are shown in the table below:

[0083] The control method provided in the embodiments of this application is applied to control the system, and the following simulation results are obtained, demonstrating the following advantages: Figure 3 As shown, the grid-side AC three-phase current exhibits a highly symmetrical and smooth sinusoidal waveform, indicating that the system is operating well under steady-state conditions. Figure 4 As shown, the system's actual instantaneous active power P and reactive power Q closely match the reference active power and reference reactive power values, and the power ripple in steady state is extremely small, proving that the control method has extremely high power control accuracy. Figure 5 As shown, a step change is applied to the DC bus voltage setpoint at simulation time t=2.0s. The actual DC output voltage rises rapidly and stabilizes at the new 300V setpoint without overshoot, demonstrating the excellent dynamic response and disturbance rejection capability of this control method. Figure 6 The harmonic spectrum analysis of the grid-side current shows that each harmonic component is strictly suppressed to within 0.5% of the fundamental current amplitude, verifying that the grid-side current has an extremely low total harmonic distortion rate and extremely high power quality under this control method.

[0084] The simulation results of this embodiment show that the predictive power control method proposed in this application embodiment can still enable voltage source rectifiers to achieve better comprehensive control performance than traditional methods that depend on parameters, without relying on the physical parameters of the system. That is, it has excellent steady-state accuracy, fast dynamic response and extremely high grid-side power quality.

[0085] Based on the same concept, this application provides a voltage source rectifier that employs the aforementioned predictive power control method for voltage source rectifiers.

[0086] The voltage source rectifier constructs a lower triangular output coefficient matrix with zero main diagonal. The parameters are estimated row by row using the normalized least mean square algorithm, thereby effectively isolating future noise, ensuring strict causality, and eliminating closed-loop bias. Combined with a finite set control framework, the optimal voltage vector is directly output, achieving robust power control that is completely free from dependence on physical parameters.

[0087] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0088] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. The above are merely preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application are included within the scope of protection of this application.

Claims

1. A predictive power control method for a voltage source rectifier, characterized in that, The predictive power control method includes: Construct a causal subspace predictor, which characterizes the mapping relationship between system state and control input; A control objective function is constructed, which is configured to weigh the system state deviation against the increment of the control input, wherein the system state deviation is the deviation between the system state estimate output by the causal subspace predictor and the system state reference. The causal subspace predictor is combined with the control objective function, and the optimal control input is obtained by minimizing the control objective function. Based on the optimal control input, the switching state is generated to control the voltage source rectifier.

2. The predictive power control method for a voltage source rectifier according to claim 1, characterized in that, The predictive power control method further includes: A power model for a voltage source rectifier is constructed. The power model characterizes the dynamic relationship between the current instantaneous power state, the grid voltage, and the rectifier output voltage in the next sampling period of the power grid system. The instantaneous power state of the power grid is taken as the system state, and the relevant terms of the power grid voltage and the rectifier output voltage in the power model are defined as the control input.

3. The predictive power control method for a voltage source rectifier according to claim 2, characterized in that, The causal subspace predictor comprises three linearly superimposed parts. The first part characterizes the influence of historical system states and historical control inputs on future system states; the second part characterizes the influence of future control inputs on the future system states; and the coefficient matrix of the second part is configured as a lower triangular matrix. The third part characterizes the influence between future system states, and the coefficient matrix of the third part is configured as a lower triangular matrix with zero on the main diagonal.

4. The predictive power control method for a voltage source rectifier according to claim 3, characterized in that, The predictive power control method further includes: Acquire historical data, including the historical instantaneous power state of the power grid system and the historical composite control input of the voltage source rectifier; Based on the historical data, a regression vector is constructed for each step of the prediction of the causal subspace predictor, so as to transform the causal subspace predictor into a linear regression form of the dot product of the coefficient vector to be identified and the regression vector. The update law of the coefficient vector is obtained according to the iterative formula of the normalized least mean square algorithm. Based on the update law, the coefficient vector is updated step by step, and the updated coefficient vectors are recombined to update the coefficient matrix of the causal subspace predictor.

5. The predictive power control method for a voltage source rectifier according to claim 4, characterized in that, The update law is expressed by the following formula: ； In the formula, This indicates that the causal subspace predictor is at the 1st... k In the first cycle i The coefficient vector of step prediction, This indicates that the causal subspace predictor is at the 1st... k -1 cycle in the first i The coefficient vector of step prediction, This indicates that the causal subspace predictor is at the 1st... k -1 cycle in the first i The transpose of the predicted coefficient vector. This indicates that the causal subspace predictor is at the 1st... k In the first cycle i The system state obtained by step prediction, Indicates that for the first i The regression vector constructed by step prediction, Indicates the learning rate. To prevent extremely small constants with a denominator of zero.

6. The predictive power control method for a voltage source rectifier according to claim 3, characterized in that, The predictive power control method further includes: The causal subspace predictor, which consists of three linearly superimposed parts, is modified by merging the third part with the prediction target of the causal subspace predictor. Then, matrix inversion is used to decouple the causal subspace predictor from the future system state to obtain a practical subspace predictor. The future control input sequence in the practical subspace predictor is decomposed into an initial control quantity and an incremental control input sequence to obtain an input-increment type practical subspace predictor that characterizes the system state with respect to the control input increment. The input-increment type practical subspace predictor is combined with the control objective function, and the optimal control input increment is obtained with the goal of minimizing the control objective function. The optimal control input increment is then superimposed with the control input of the previous cycle to obtain the optimal control input.

7. The predictive power control method for a voltage source rectifier according to claim 1, characterized in that, The control objective function is configured as a linear superposition of the norm of the system state deviation and the norm of the control input increment.

8. The predictive power control method for a voltage source rectifier according to claim 1, characterized in that, Based on the optimal control input, the switch state is generated using finite set control.

9. The predictive power control method for a voltage source rectifier according to claim 8, characterized in that, The predictive power control method further includes: A first cost function, a second cost function, a third cost function, and a fourth cost function are constructed. The first cost function is configured to quantify the error between the candidate output voltage vector and the reference voltage vector. The second cost function is configured to measure the change in switching state. The third cost function is configured to determine the polarity relationship between the neutral point voltage and the current. The fourth cost function is configured to introduce tolerance constraints to comprehensively evaluate the neutral point potential balance capability. Based on the optimal control input, a finite set is used to select the optimal voltage vector and its corresponding switching state that minimize the first cost function: When the obtained optimal voltage vector is a large vector, the unique switching state corresponding to the optimal voltage vector is directly applied. When the obtained optimal voltage vector is zero, among the three corresponding redundant switching states, select the switching state that minimizes the second cost function and does not disrupt the midpoint potential balance. When the obtained optimal voltage vector is a small vector, the switching state that minimizes the second cost function is selected from the two corresponding redundant switching states; if the midpoint potential exceeds the tolerance limit, the switching state that can reduce the fourth cost function is selected based on the polarity relationship between the neutral point voltage and current of the third cost function. When the obtained optimal voltage vector is the midpoint vector, if the fourth cost function is not equal to 2, the corresponding switching state is applied to facilitate the midpoint potential balance; otherwise, the suboptimal switching state of the small vector is selected, and the voltage vector is screened and processed again.

10. A voltage source rectifier, characterized in that, The voltage source rectifier employs the predictive power control method for voltage source rectifiers as described in any one of claims 1 to 9.

Images