Unscented Kalman filter, model-free predictive control method and device for grid-connected inverters

Abstract

This invention discloses an unscented Kalman filter, a model-free predictive control method and device for grid-connected inverters, and a method for controlling the lumped disturbance F of the system. αβ State observation is performed using an improved unscented Kalman filter, and the proportional gain α in the hyperlocal model is identified using an unweighted adaptive algorithm. The adaptive optimal value of the identified proportional gain α is fed back to the model-free predictive control module, which optimizes the unweighted factor cost function. This allows for prediction of grid-connected current control without any model parameters, thus solving the problem of stable grid-connected inverter operation during system parameter mismatch. This invention introduces an adaptive Sigma method, an enhanced noise model, and a weighted observable. Furthermore, by incorporating voltage fluctuation coupling, dynamic γ optimization, and a nonlinear dynamic model into the model parameter adaptive algorithm, it maintains relatively low computational complexity while improving the control over the lumped disturbance F of the system. αβ Estimation accuracy and the ability to identify the scaling factor α in hyperlocal models.

Description

Technical Field

[0001] This invention belongs to the field of grid-connected inverter control technology, and particularly relates to an unscented Kalman filter, a model-free predictive control method and device for grid-connected inverters. Background Technology

[0002] With the rapid development of renewable energy, distributed generation systems such as solar and wind power are becoming increasingly popular and an important part of modern power systems. Three-level grid-connected inverters, as the core equipment connecting distributed generation systems and the power grid, directly affect the stability and efficiency of the entire system through their control performance. Connecting a large number of power electronic devices can increase grid harmonics, and in more severe cases, lead to power quality degradation and grid-connected equipment failure. Therefore, how to efficiently and stably connect them to the grid has become a key challenge for grid-connected inverters. While traditional grid-connected inverter control methods can meet performance requirements to a certain extent, they still have limitations in terms of system dynamic response and robustness.

[0003] Model-Free Predictive Control (MFPC), as an emerging control strategy, has received widespread attention in recent years because it does not rely on an accurate system model and can predict the future behavior of the system in real time and perform optimized control. However, current MFPC for grid-connected inverters suffers from problems such as response time delay, insufficient control accuracy, sensitivity to disturbances, and poor adaptability under different operating conditions. These problems lead to reduced system stability and efficiency, thereby affecting the reliability and economy of power generation.

[0004] Existing technologies include model-free predictive control based on the Runge-Kutta Algorithm (RKA) and Lagrange difference algorithms. However, this method is computationally complex, and the model parameters of the fourth-order RKA are significantly affected by the time delay of the current response, impacting predictions under poor operating conditions. Model-free predictive control methods based on sliding mode observers also exist, but these methods are complex to design when appropriately selecting the sliding surface and switching gain. Rapid switching near the sliding surface can lead to high-frequency oscillations in the system, and they are poorly sensitive to random noise. Summary of the Invention

[0005] To address the limitations of model-free predictive control in grid-connected inverters and improve system control performance, this invention proposes an unscented Kalman filter, a model-free predictive control method and apparatus for grid-connected inverters. While maintaining relatively low computational complexity, this invention improves the control over the lumped disturbance F of the system. αβ Estimation accuracy and the ability to identify the scaling factor α in hyperlocal models.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention proposes an unscented Kalman filter, the unscented Kalman filter comprising:

[0008] The Sigma point generation module estimates the current state and captures the characteristics of the state distribution; the dynamic adjustment module calculates new Sigma points based on real-time covariance and adjusts the covariance. Dynamic adjustment incorporates historical errors into the covariance update;

[0009] The weighted calculation module is used to generate effective Sigma points that cover the state space by dynamically adjusting the covariance;

[0010] The noise model module is used to model noise using autoregressive models and multiplicative noise theory to obtain noise models applicable to real-world conditions.

[0011] The enhancement module is used to adjust the noise covariance by enhancing the noise model and updating process noise and observation noise;

[0012] A weighted observability analyzer is used to determine the observability of a system.

[0013] In one implementation, the formula for generating Sigma points is:

[0014]

[0015] In the formula,

[0016] The noise model is as follows:

[0017]

[0018] In the formula, a i b i These are model coefficients. p It is the lag order; y(t) and u(t) are the enhanced noise models;

[0019] The process noise update and the observation noise update are as follows:

[0020]

[0021] In the formula, φ and Let Q and R be the process noise and observation noise adjustment coefficients, respectively, and let Q and R be the process noise covariance and measurement noise covariance, respectively. and Let X be the state estimate and the measurement estimate, respectively; pred and Z pred These are the initial values ​​for the state variables and the measured variables, respectively.

[0022] In one implementation, the matrix O of the weighted observability unit w for:

[0023]

[0024] In the formula, w i These are weighting coefficients;

[0025] O can be quickly determined through singular value decomposition. w The rank of is given by:

[0026] O w =UΣV T

[0027] In the formula, O w Let U be the weighted observable matrix, and U and V be the left and right singular vectors in the singular value decomposition method, respectively.

[0028] The number of non-zero singular values ​​of Σ can be used to determine O. w Rank.

[0029] Secondly, the present invention provides a model-free predictive control method for unweighted adaptive grid-connected inverters, wherein the method measures the lumped disturbance F of the system. αβ The method includes: state observation is performed using an unscented Kalman filter, and parameter identification of the scaling factor α in the hyperlocal model is performed using an unweighted adaptive algorithm; the method comprises:

[0030] Obtain the three-phase voltage and current on the output side of the grid-connected inverter, and establish a mathematical model of the three-level grid-connected inverter based on Kirchhoff's laws;

[0031] Clark transformation is performed on the three-phase voltage and current to obtain the mathematical model of the three-level grid-connected inverter in the αβ two-terminal static coordinate system, and the current reference value is obtained by passing the three-phase grid voltage through a phase-locked loop.

[0032] Using the grid-side current and voltage at time k as initial values, we substitute them into the unscented Kalman filter described in any one of claims 1-3 to obtain the estimated identification value.

[0033] Based on the obtained estimated identification values, the three-phase voltage and current on the output side of the grid-connected inverter, and the current reference value, the unweighted adaptive optimal value of the proportional coefficient α is obtained through an unweighted adaptive algorithm.

[0034] The output current of the grid-connected inverter at time k+1 is obtained by model-free prediction using the estimated identification value, the unweighted adaptive optimal value of the proportional coefficient α, and the grid voltage in two stationary coordinate systems.

[0035] By substituting the output current and current reference value of the grid-connected inverter at time k+1 into the unweighted cost function to find the optimal voltage vector control for the switching state of the grid-connected inverter at the next time step, model-free predictive control of the grid-connected inverter is achieved under system parameter mismatch conditions.

[0036] In one implementation, the estimated identification value is:

[0037] In the formula, i αβ (k), i αβ (k+1) represents the output current of the grid-connected inverter at times k and k+1, respectively. , respectively, represent the estimated identification values ​​of the grid-connected inverter at times k and k+1; A is the state transition matrix. T is the sampling period, ω is the grid angular frequency; Z(k) is the observation equation; B is the noise driving matrix. α is the scaling factor, usually set to 1 / L; H is the observation matrix, H =

[10] .

[0038] In one implementation, the step of obtaining the unweighted adaptive optimal value of the proportional coefficient α through an unweighted adaptive algorithm based on the obtained estimated identification value, the three-phase voltage and current on the output side of the grid-connected inverter, and the current reference value includes:

[0039] Define a control error, and select the target signal to be tracked as the reference current i in the stationary coordinate system. αβref (t), the error state equation is defined as:

[0040] e(t) = i αβref (t)-i αβ (t)

[0041] Based on the above formula, the hyperlocal model Substituting the values, we obtain the error dynamic equation as follows:

[0042]

[0043] Using the error dynamic equation as the basis for the adaptive law, the adaptive law is established based on the gradient descent method, and its expression is:

[0044]

[0045] In the formula, γ is the learning rate, which needs to be properly adjusted. Too large a rate may lead to instability, while too small a rate will result in slow convergence. k is the Δu balance gain coefficient.

[0046] After discretizing the proportional coefficient α, a nonlinear dynamic model is introduced. The control parameters are then dynamically adjusted via a nonlinear feedback mechanism along the α and β axes, specifically as follows:

[0047] α α(k+1)=α α (k)-Asgn(e β (k))γ|e β (k)| λ u α (k)T+ksgn(Δu)T

[0048] α β (k+1)=α β (k)-Bsgn(e α (k))γ|e α (k)| λ u β (k)T+ksgn(Δu)T

[0049] In the formula, sgn(e(k)) represents the sign of the error to ensure that the update direction is consistent with the error; |e(k)| λ It is the power of the absolute value of the error (usually λ > 0); A and B are the error coupling coefficients, respectively;

[0050] The Lyapunov function is designed as a polynomial, taking into account the changes in error and control parameters, specifically:

[0051]

[0052] In the formula, the initial value α0 is set to 50; (The following is a separate, unrelated statement: "Introducing...") To improve the capture of transient dynamic behavior, k is a positive weighting coefficient used to adjust the error state. The impact; The derivative of the voltage fluctuation provides further information about the midpoint potential fluctuation;

[0053] According to the Lyapunov stability principle, the error state e(t) and... During the adjustment process, the system converges uniformly to zero, and the control parameter α fluctuates around α0. The system remains stable, ensuring that the error state and parameter α converge to the optimal value α. opt .

[0054] In one implementation, the learning rate γ is dynamically adjusted based on the deviation of the target performance index e(t), and the dynamic adjustment strategy is as follows: if the current error e k Less than the error e from the previous time step k-1 Then increase the learning rate: γ = min(1.05γ, γ max If the current error increases, then reduce the learning rate: γ = max(0.95γ, γ min ).

[0055] In one implementation, the identification value F αβUnweighted adaptive optimal value α opt and u αβ After discretization by the model-free prediction module, i is obtained. αβ Its expression is:

[0056]

[0057] In one implementation, the unweighted cost function is:

[0058] g = (i αref -i α (k+1)) 2 +(i βref -i β (k+1)) 2

[0059] In the formula, i αβref This is the reference current in the stationary coordinate system.

[0060] Thirdly, the present invention provides a model-free predictive control device for a grid-connected inverter without weights and adaptive control, the device comprising:

[0061] The mathematical model module is used to obtain the three-phase voltage and current on the output side of the grid-connected inverter and to establish a mathematical model of the three-level grid-connected inverter based on Kirchhoff's laws.

[0062] The coordinate system transformation module is used to perform Clark transformation on the three-phase voltage and current to obtain the mathematical model of the three-level grid-connected inverter in the αβ two-term static coordinate system.

[0063] The current reference module is used to obtain the current reference value from the three-phase grid voltage through a phase-locked loop.

[0064] The aforementioned unscented Kalman filter is used to obtain the estimated identification value based on the grid-side current and voltage of the grid-connected inverter at time k.

[0065] The parameter adaptive calculation module is used to obtain the unweighted adaptive optimal value of the proportional coefficient α based on the obtained estimated identification value, the three-phase voltage and current on the output side of the grid-connected inverter, and the current reference value, through an unweighted adaptive algorithm.

[0066] The model-free prediction module is used to obtain the grid-connected inverter output current at time k+1 by using the estimated identification value, the unweighted adaptive optimal value of the proportional coefficient α, and the grid voltage in two static coordinate systems through model-free prediction.

[0067] The optimal switching state selection module is used to substitute the output current and current reference value of the grid inverter at time k+1 into the unweighted cost function to find the optimal voltage vector control for the switching state of the grid-connected inverter at the next time moment, thereby realizing model-free predictive control of the grid-connected inverter under system parameter mismatch conditions.

[0068] The beneficial effects of the solution proposed in this invention are as follows:

[0069] This invention proposes a model-free predictive control strategy for grid-connected inverters using an unweighted adaptive unscented Kalman filter. This strategy primarily addresses the lumped disturbance F of the system. αβ State observation is performed using an improved unscented Kalman filter. While maintaining relatively low computational complexity, an adaptive Sigma method, an enhanced noise model, and a weighted observable are introduced, providing high-accuracy state estimation results in nonlinear models. Furthermore, an adaptive model algorithm for model parameters incorporates voltage fluctuation coupling, dynamic γ optimization, and a nonlinear dynamic model to obtain the unweighted adaptive optimal value α. opt The weighting effect of DC-side capacitor voltage fluctuations in the cost function is optimized, resulting in good performance of predicted current over a wider range. Under system parameter mismatch conditions, it not only greatly reduces computational costs and ensures stable system operation, but also improves current tracking performance and reduces harmonic distortion. Attached Figure Description

[0070] The accompanying drawings, as part of this invention, are provided to further illustrate the invention. The illustrative embodiments and descriptions of the invention are used to explain the invention, but do not constitute an undue limitation thereof. Clearly, the drawings described below are merely some embodiments, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0071] Figure 1 This is a schematic diagram of the structure of an unscented Kalman filter provided in an embodiment of the present invention;

[0072] Figure 2 A flowchart of a model-free predictive control method for an unweighted adaptive grid-connected inverter provided in an embodiment of the present invention;

[0073] Figure 3 It is a three-level grid-connected inverter topology;

[0074] Figure 4 The unweighted adaptive optimal value α provided in this embodiment of the invention. opt Waveform diagram;

[0075] Figure 5This is the three-phase grid-connected current waveform under the conventional MPC steady-state condition provided in this embodiment of the invention, where... Figure 5 (a) is the simulated waveform of the three-phase grid-connected current in steady state under the condition of inductor parameter mismatch, using traditional MPC. Figure 5 (b) is the simulated waveform of grid-connected current error under steady-state conditions of traditional MPC under inductor parameter mismatch. Figure 5 (c) is a schematic diagram of the three-phase grid-connected current THD under the steady state of traditional MPC under inductor parameter mismatch conditions;

[0076] Figure 6 The current simulation waveforms for the model-free predictive control method using existing sliding mode observers are shown below. Figure 6 (a) is the simulated waveform of the three-phase grid-connected current in steady state under the condition of inductance parameter mismatch. Figure 6 (b) is the simulated waveform of the grid-connected current error under steady-state conditions under inductor parameter mismatch. Figure 6 (c) is a schematic diagram of the three-phase grid-connected current THD under steady-state conditions under inductor parameter mismatch;

[0077] Figure 7 The current simulation waveform using the method of the present invention is shown below. Figure 7 (a) is the simulated waveform of the three-phase grid-connected current in steady state under the condition of inductance parameter mismatch using the method of the present invention. Figure 7 (b) is the simulated waveform of the grid-connected current error in steady state under the control strategy of the present invention under the condition of inductor parameter mismatch. Figure 7 (c) is a schematic diagram of the three-phase grid-connected current THD under steady state based on the control strategy described in this invention under the condition of inductor parameter mismatch;

[0078] Figure 8 This is a comparison diagram of grid-connected current simulation between the method of the present invention and existing technologies, wherein... Figure 8 (a) is a simulation comparison diagram of the traditional MPC and the method of the present invention when the grid-connected current is 10A. Figure 8 (b) is a simulation comparison diagram of the model-free predictive control method of the sliding mode observer and the method of the present invention when the grid current is 10A;

[0079] Figure 9 The waveforms represent the dynamic performance simulation under inductor parameter mismatch conditions. Figure 9 (a) is the dynamic performance simulation waveform of the model-free predictive control method with a sliding mode observer under inductance parameter mismatch conditions. Figure 9 (b) is the dynamic performance simulation waveform of the method of the present invention under the condition of inductor parameter mismatch;

[0080] Figure 10 This is a block diagram of a model-free predictive control system for a grid-connected inverter provided in an embodiment of the present invention;

[0081] It should be noted that these accompanying drawings and textual descriptions are not intended to limit the scope of the invention in any way, but rather to illustrate the concept of the invention to those skilled in the art by referring to specific embodiments. Detailed Implementation

[0082] The model-free predictive control method for grid-connected inverters of the present invention will be further described in detail below with reference to the accompanying drawings. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present invention can be combined with each other. The technical solutions of this application will be described in detail with reference to the accompanying drawings and embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0083] like Figure 1 As shown in the figure, one embodiment of the present invention illustrates an unscented Kalman filter, which includes:

[0084] An adaptive Sigma point generation module is used to estimate the current state and capture the characteristics of the state distribution.

[0085] The prediction module is used to calculate the predicted values ​​of system state variables and observations based on the generated Sigma points;

[0086] The dynamic adjustment module is used to dynamically adjust the mean and covariance based on the predicted values, and incorporates historical errors into the covariance update.

[0087] The enhanced noise model module is used to model noise using autoregressive models and multiplicative noise theory to obtain a noise model suitable for actual conditions. The enhanced noise model is then used to adjust the noise covariance and update the process noise and observation noise.

[0088] A weighted observability is used to determine the observability of a system.

[0089] In a standard unscented Kalman filter, Sigma points are generated around the current state estimate to capture characteristics of the state distribution. The formula for generating Sigma points is:

[0090]

[0091] To dynamically adjust the position of the Sigma point, a new Sigma point can be calculated based on the real-time covariance.

[0092] covariance Dynamic adjustment incorporates historical errors into the covariance update, using a weighted average method:

[0093]

[0094] In the formula, σ is an adjustment factor. When the system changes dynamically, the covariance can be increased to generate more dispersed Sigma points; conversely, the covariance can be decreased to focus on more accurate estimates. This adaptive Sigma point method uses dynamically adjusted covariance to generate more effective Sigma points, ensuring that they better cover the state space.

[0095] Gaussian white noise is typically independent and identically distributed in time, and the process noise w(t) and observation noise follow a normal distribution.

[0096] w(t)~N(0,Q)

[0097] v(t)~N(0,R)

[0098] In practical applications, system noise is often time-dependent. By using autoregressive models and multiplicative noise theory, a noise model applicable to real-world conditions can be obtained:

[0099]

[0100] In the formula, a i b i These are model coefficients. p It is the lag order; y(t) and u(t) are the enhanced noise models.

[0101] By adjusting the noise covariance through an enhanced noise model, the process noise update and the observation noise update are as follows:

[0102]

[0103] In the formula, φ and Let Q and R be the process noise and observation noise adjustment coefficients, respectively, and let Q and R be the process noise covariance and measurement noise covariance, respectively. and Let X be the state estimate and the measurement estimate, respectively; pred and Z pred These are the initial values ​​for the state variables and the measured variables, respectively.

[0104] When a system is dynamically changing, the conventional observability matrix of an unscented Kalman filter may not effectively reflect the observability of the system. Therefore, a weighted observability matrix O is introduced. w for:

[0105]

[0106] In the formula, w i These are weighting coefficients.

[0107] Furthermore, singular value decomposition can be used to quickly determine O.w The rank of is given by:

[0108] O w =UΣV T

[0109] In the formula, O w Let U be the weighted observable matrix, and U and V be the left and right singular vectors in the singular value decomposition method, respectively.

[0110] The number of non-zero singular values ​​of Σ can be used to determine O. w Rank.

[0111] like Figure 2 As shown in the figure, one embodiment of the present invention illustrates a model-free predictive control method for a grid-connected inverter with no weights and adaptive properties. This method uses the lumped disturbance F of the system as a control method. αβ The method involves state observation using an unscented Kalman filter and parameter identification of the scaling factor α in the hyperlocal model using an unweighted adaptive algorithm; the method includes:

[0112] Step S100: Obtain the three-phase voltage and current on the output side of the grid-connected inverter, and establish a mathematical model of the three-level grid-connected inverter based on Kirchhoff's laws.

[0113] Step S200: Perform Clark transformation on the three-phase voltage and current to obtain the mathematical model of the three-level grid-connected inverter in the αβ two-terminal static coordinate system, and obtain the current reference value by passing the three-phase grid voltage through a phase-locked loop.

[0114] Combination Figure 3 In a three-level grid-connected inverter topology, a mathematical model of the grid-connected inverter is established based on Kirchhoff's laws.

[0115] The three-phase current i on the output side of the grid-connected inverter is obtained through sampling calculation. a i b i c Then perform a Clark transformation to obtain i αβ (k);

[0116] Their forms of expression are as follows:

[0117]

[0118] In the formula, i a i b i c For the output current of the grid-connected inverter, u a u b u c Where L is the inverter output voltage, R is the filter inductance, and e is the parasitic resistance. a e be c This is the grid voltage.

[0119] The Clark transform formula is:

[0120]

[0121] By performing Clark transformation on the three-phase voltage and current, the mathematical model of the three-level grid-connected inverter in the αβ two-terminal stationary coordinate system is expressed as follows:

[0122]

[0123] In the formula, i αβ =[i α (k),i β (k] T i α (k) and i β (k) represents the output current of the grid-connected inverter at time (k); u αβ =[u α (k),u β (k)] T ,u α (k) and u β (k) represents the output voltage of the grid-connected inverter at time (k); e αβ =[e α (k),e β (k)] T ,e α (k) and e β (k) represents the grid voltage at time (k).

[0124] The three-phase grid voltage is used to obtain a reference angle θ and a given current value via a phase-locked loop. A coordinate transformation is then used to obtain the current reference value i. αβref .

[0125] Step S300: Use the grid-side current and voltage at time k as initial values ​​and substitute them into the unscented Kalman filter to obtain the estimated identification value.

[0126] i at time (k) αβ (k), u αβ (k) Substitute into the unscented Kalman filter algorithm to update and obtain the estimated identification value. Its expression is:

[0127]

[0128] In the formula, i αβ (k), i αβ (k+1) represents the output current of the grid-connected inverter at times k and k+1, respectively. , respectively, represent the estimated identification values ​​of the grid-connected inverter at times k and k+1; A is the state transition matrix. T is the sampling period, ω is the grid angular frequency; Z(k) is the observation equation; B is the noise driving matrix. α is the scaling factor, usually set to 1 / L; H is the observation matrix, H =

[10] .

[0129] Step S400: Based on the obtained estimated identification value, the three-phase voltage and current on the output side of the grid-connected inverter, and the current reference value, obtain the unweighted adaptive optimal value of the proportional coefficient α through an unweighted adaptive algorithm.

[0130] α is the scaling factor when designing a hyperlocal model. It is usually set to α = 1 / L, but will be affected by i. αβ F αβ u αβ The proportional gain fluctuates with changes in voltage fluctuations, and both excessively large and small parameters α can affect the performance of model-free predictive control in grid-connected inverters. Therefore, this invention designs an adaptive controller based on error feedback and voltage fluctuation coupling to determine the optimal value of the proportional gain α in the hyperlocal model. opt .

[0131] In this embodiment of the application, based on the obtained estimated identification value, the three-phase voltage and current on the output side of the grid-connected inverter, and the current reference value, the unweighted adaptive optimal value of the proportional coefficient α is obtained through an unweighted adaptive algorithm. The specific steps include:

[0132] Define a control error and select the target signal to be tracked as i. αβref (t), then the error state equation can be defined as:

[0133] e(t) = i αβref (t)-i αβ (t)

[0134] Based on the above formula, the hyperlocal model Substituting the values, we obtain the error dynamic equation as follows:

[0135]

[0136] This error dynamic equation can serve as the basis for designing the adaptive law. To dynamically adjust the parameter α to minimize the error dynamic equation, and to achieve weightless optimization of the cost function, an adaptive law can be established based on gradient descent, with the following expression:

[0137]

[0138] In the formula, γ is the learning rate, which needs to be properly adjusted. Too large a rate may lead to instability, while too small a rate will result in slow convergence. k is the Δu balance gain coefficient.

[0139] The learning rate γ can be dynamically adjusted based on the deviation of the target performance index e(t). The dynamic adjustment strategy is as follows: if the current error e k Less than the error e from the previous time step k-1 Then increase the learning rate: γ = min(1.05γ, γ max If the current error increases, then reduce the learning rate: γ = max(0.95γ, γ min ).

[0140] After discretizing the proportional coefficient α, a nonlinear dynamic model is introduced. The control parameters are then dynamically adjusted via a nonlinear feedback mechanism along the α and β axes, resulting in the following adjustment:

[0141] α α (k+1)=α α (k)-Asgn(e β (k))γ|e β (k)| λ u α (k)T+ksgn(Δu)T

[0142] α β (k+1)=α β (k)-Bsgn(e α (k))γ|e α (k)| λ u β (k)T+ksgn(Δu)T

[0143] In the formula, sgn(e(k)) represents the sign of the error to ensure that the update direction is consistent with the error; |e(k)| λ λ is the power of the absolute value of the error (usually λ > 0); A and B are the error coupling coefficients, respectively.

[0144] To ensure system stability, the Lyapunov function is designed as a polynomial, taking into account the changes in error and control parameters:

[0145]

[0146] In the formula, the initial value α0 is set to 50; (The following is a separate, unrelated statement: "Introducing...") To improve the capture of transient dynamic behavior, k is a positive weighting coefficient used to adjust... The impact; The derivative of the voltage fluctuation provides further information about the midpoint potential fluctuation.

[0147] According to Lyapunov's stability principle, the error state e(t) Δu and During the adjustment process, the system converges uniformly to zero, and the control parameter α fluctuates around α0. The system remains stable, ensuring that the error state and parameter α converge to the optimal value α. opt .

[0148] Figure 4 The parameter α is the unweighted adaptive optimal value α obtained by an adaptive controller based on error feedback and voltage fluctuation coupling. opt The waveform.

[0149] Step S500: Combine the estimated identification value, the unweighted adaptive optimal value of the scaling factor α, and the two grid voltages u in the stationary coordinate system. αβ The output current of the grid-connected inverter at time k+1 is obtained through model-free prediction.

[0150] The estimated identification value, the unweighted adaptive optimal value of the proportional coefficient α, and the grid voltage u are combined. αβ The output current of the grid-connected inverter at time k+1, obtained through model-free prediction, is:

[0151]

[0152] Step S600: Substitute the output current and current reference value of the grid inverter at time k+1 into the unweighted cost function to find the optimal voltage vector control for the switching state of the grid-connected inverter at the next time step, and realize model-free predictive control of the grid-connected inverter under the system parameter mismatch condition.

[0153] Predicted current i of the grid-connected inverter at time (k+1) αβ (k+1) and the current reference value i obtained through the phase-locked loop from the grid-connected voltage. αβref Substituting all values ​​into the unweighted factor cost function for optimization, the optimal voltage vector control is selected to determine the switching state of the grid-connected inverter at the next time step. The unweighted cost function is:

[0154] g = (i αref -i α (k+1)) 2 +(i βref -i β (k+1)) 2

[0155] In the formula, i αβref This is the reference current in the stationary coordinate system.

[0156] To verify the effectiveness of this invention, simulations were performed in the MATLAB / Simulink environment. The system parameters are shown in the table below.

[0157] Table System Parameters

[0158]

[0159] To verify the feasibility and parameter robustness of the proposed control strategy, this invention studies its steady-state performance in simulations and compares it with traditional MPC and the control strategy proposed in Reference 2. Parameter robustness is verified in simulations by setting the mismatch inductance value L. v =0.1H, the reference value of the grid-connected inverter output current is 10A, and the grid current base frequency is 50Hz. Figure 5 , Figure 6 and Figure 7 The diagrams show the steady-state three-phase grid-connected current waveforms, current errors, and THD of the traditional MPC, the control strategy proposed in Reference 2, and the control strategy described in this invention.

[0160] from Figure 5 , Figure 6 and Figure 7 As can be seen, when the reference current is 10A, the grid-connected current obtained by the traditional MPC is relatively turbulent, with a THD of 7.10% and a large current error value, fluctuating within ±3A. The grid-connected current THD of the control strategy proposed in Reference 2 is 3.72%, with a current error value fluctuating within ±2A. Compared with the control strategy proposed in Reference 2, the control strategy described in this invention significantly reduces the grid-connected current THD and error value, with a THD of 1.15% and a current error value fluctuating within ±1A. This verifies the feasibility of the control strategy described in this invention in steady state, avoiding the impact of model parameter mismatch on the grid-connected inverter system.

[0161] Under the same simulation conditions, the steady-state control performance of traditional MPC and the control strategy described in this invention are compared as follows: Figure 8 As shown in Figure a, the steady-state control performance of the control strategy proposed in Reference 2 and the control strategy described in this invention is compared as follows: Figure 8 As shown in b. Specifically, during the period t = 0–100 ms, the control system employs both the traditional MPC and the control strategy proposed in reference 2; starting from t = 100 ms, the control system switches to the control strategy described in this invention.

[0162] When model parameters mismatch, by Figure 8 As can be seen from this, the grid-connected current waveform of the traditional MPC strategy has collapsed, with a THD value reaching 10.17%, while the current THD value under the control strategy described in this invention is only 1.13%; Figure 8 As can be seen from b, the current harmonic distortion (THD) values ​​of the control strategy proposed in Reference 2 and the control strategy described in this invention are 3.73% and 1.12%, respectively. However, the control strategy described in this invention has better steady-state performance, which further verifies its effectiveness.

[0163] To verify the dynamic performance of the proposed control strategy, a comparative analysis was conducted using dynamic simulations of the control strategy described in this invention and the control strategy proposed in Reference 2. Figure 9 It can be seen that when the reference current changes abruptly from 10A to 15A at t=100ms, the dynamic response time of the KF-based MFPC strategy is 1.44ms, while the dynamic response time of the control strategy described in this invention is 0.83ms. This indicates that the control strategy described in this invention has better dynamic response performance.

[0164] The following is an embodiment of a model-free predictive control device for an unweighted adaptive grid-connected inverter according to the present invention, which can be used to execute an embodiment of a model-free predictive control method for an unweighted adaptive grid-connected inverter according to the present invention. For details not disclosed in the embodiment of the model-free predictive control device for an unweighted adaptive grid-connected inverter according to the present invention, please refer to the embodiment of the model-free predictive control method for an unweighted adaptive grid-connected inverter according to the present invention.

[0165] Reference Figure 10 As shown, in one embodiment, a model-free predictive control device for a grid-connected inverter without weights is provided. The device includes:

[0166] The mathematical model module is used to obtain the three-phase voltage and current on the output side of the grid-connected inverter and to establish a mathematical model of the three-level grid-connected inverter based on Kirchhoff's laws.

[0167] The coordinate system transformation module is used to perform Clark transformation on the three-phase voltage and current to obtain the mathematical model of the three-level grid-connected inverter in the αβ two-term static coordinate system.

[0168] The current reference module is used to obtain the current reference value from the three-phase grid voltage through a phase-locked loop.

[0169] The unscented Kalman filter according to any one of claims 1-3 is used to obtain an estimated identification value based on the grid-side current and voltage of the grid-connected inverter at time k.

[0170] The parameter adaptive calculation module is used to obtain the unweighted adaptive optimal value of the proportional coefficient α based on the obtained estimated identification value, the three-phase voltage and current on the output side of the grid-connected inverter, and the current reference value, through an unweighted adaptive algorithm.

[0171] The model-free prediction module is used to combine the estimated identification value, the unweighted adaptive optimal value of the scaling factor α, and the grid voltage u in two stationary coordinate systems. αβ The output current of the grid-connected inverter at time k+1 obtained through model-free prediction;

[0172] The optimal switching state selection module is used to substitute the output current and current reference value of the grid inverter at time k+1 into the unweighted cost function to find the optimal voltage vector control for the switching state of the grid-connected inverter at the next time moment, thereby realizing model-free predictive control of the grid-connected inverter under system parameter mismatch conditions.

[0173] The functional modules in this embodiment of the invention can be integrated into one processing module, or each unit can exist as a separate physical entity, or two or more units can be integrated into one module. The integrated module can be implemented in hardware or as a software functional module.

[0174] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-described technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A model-free predictive control method for an unweighted adaptive grid-connected inverter, characterized in that, The method measures the lumped disturbance of the system. State observation is performed using an unscented Kalman filter, while the scaling coefficients in the hyperlocal model are adjusted using an unweighted adaptive algorithm. Perform parameter identification; the method includes: Obtain the three-phase voltage and current on the output side of the grid-connected inverter, and establish a mathematical model of the three-level grid-connected inverter based on Kirchhoff's laws; Clark transformation is performed on the three-phase voltage and current to obtain the mathematical model of the three-level grid-connected inverter. The model is in a two-phase stationary coordinate system, and the current reference value is obtained by passing the three-phase grid voltage through a phase-locked loop; The grid-side current and voltage at time k are used as initial values ​​and substituted into the unscented Kalman filter to obtain the estimated identification value. The unscented Kalman filter includes: An adaptive Sigma point generation module is used to estimate the current state and capture the characteristics of the state distribution. The prediction module is used to calculate the predicted values ​​of system state variables and observations based on the generated Sigma points; The dynamic adjustment module is used to dynamically adjust the mean and covariance based on the predicted values, and incorporates historical errors into the covariance update. The enhanced noise model module is used to model noise using autoregressive models and multiplicative noise theory to obtain a noise model suitable for actual conditions. The enhanced noise model is then used to adjust the noise covariance and update the process noise and observation noise. A weighted observability is used to determine the observability of a system. Based on the obtained estimated identification values, the three-phase voltage and current on the output side of the grid-connected inverter, and the current reference value, the proportional coefficient is obtained through an unweighted adaptive algorithm. The unweighted adaptive optimal value; Estimate the identification value and the scaling factor The unweighted adaptive optimal value and the grid voltage in the two-phase stationary coordinate system are used to obtain the output current of the grid-connected inverter at time k+1 through model-free prediction. By substituting the output current and current reference value of the grid-connected inverter at time k+1 into the unweighted cost function to find the optimal voltage vector control for the switching state of the grid-connected inverter at the next time step, model-free predictive control of the grid-connected inverter is achieved under system parameter mismatch conditions.

2. The unweighted adaptive model-free predictive control method for grid-connected inverters according to claim 1, characterized in that, The formula for generating Sigma points is: ， In the formula, , where n is the dimension of the state variables; The noise model is as follows: ， ， In the formula, , These are model coefficients. It is the lag order; , To enhance the noise model; The update process noise and observation noise are as follows: ， In the formula, and Let Q and R be the process noise and observation noise adjustment coefficients, respectively, and let Q and R be the process noise covariance and measurement noise covariance, respectively. and Let them be the state estimate and the measurement estimate, respectively. and These are the initial values ​​for the state variables and the measured variables, respectively.

3. The unweighted adaptive model-free predictive control method for grid-connected inverters according to claim 1, characterized in that, Weighted observable for: ， In the formula, These are weighting coefficients; Singular value decomposition can be used to quickly determine... The rank of is given by: ， In the formula, Let U be the weighted observable matrix, and U and V be the left and right singular vectors in the singular value decomposition method, respectively. The number of non-zero singular values ​​of Σ can be used to determine this. Rank.

4. The unweighted adaptive model-free predictive control method for grid-connected inverters according to claim 1, characterized in that, The estimated identification value is: ， In the formula, , These represent the output currents of the grid-connected inverter at times k and k+1, respectively. , These are the estimated identification values ​​of the grid-connected inverter at times k and k+1, respectively. A is the state transition matrix. T is the sampling period. The angular frequency of the power grid; B is the observation equation; B is the noise driving matrix. , It is a proportionality factor, which is usually set. , L H is the filter inductor; H is the observation matrix. .

5. The unweighted adaptive model-free predictive control method for grid-connected inverters according to claim 1, characterized in that, Based on the obtained estimated identification values, the three-phase voltage and current on the output side of the grid-connected inverter, and the current reference value, the proportional coefficient is obtained through a weightless adaptive algorithm. The unweighted adaptive optimal values ​​include: Define a control error, and select the target signal to be tracked as the reference current in the stationary coordinate system. The error state equation is defined as follows: ， Based on the above formula, the hyperlocal model Substituting the values, we obtain the error dynamic equation as follows: ， Using the error dynamic equation as the basis for the adaptive law, the adaptive law is established based on the gradient descent method, and its expression is: ， In the formula, This refers to the learning rate, which needs to be adjusted properly. Too high a rate may lead to instability, while too low a rate will result in slow convergence. k is... Balanced gain coefficient; proportionality coefficient After discretization, a nonlinear dynamic model is introduced, and the control parameters are dynamically adjusted in the α and β axes respectively through a nonlinear feedback mechanism. Specifically, the adjustments are as follows: ， In the formula, The sign indicating the error is used to ensure that the update direction is consistent with the error. It is the power of the absolute value of the error, usually A and B are the error coupling coefficients, respectively. The Lyapunov function is designed as a polynomial, taking into account the changes in error and control parameters, specifically: ， ， In the formula, the initial value Set to 50; introduce This improves the capture of transient dynamic behavior; k is a positive weighting coefficient used to adjust the error state. The impact; The derivative of the voltage fluctuation provides further information about the midpoint potential fluctuation; According to the Lyapunov stability principle, the error state... ,and During the adjustment process, the control parameters converge uniformly to zero. exist Nearby fluctuations, The system remains stable, ensuring error status and parameters. Converging to the optimal value .

6. The unweighted adaptive model-free predictive control method for grid-connected inverters according to claim 5, characterized in that, The learning rate Based on target performance indicators The deviation is dynamically adjusted, and the dynamic adjustment strategy is as follows: if the current error... Less than the error of the previous time step This increases the learning rate. If the current error increases, reduce the learning rate: .

7. The unweighted adaptive model-free predictive control method for grid-connected inverters according to claim 6, characterized in that, The estimated identification value and scaling factor will be... Unweighted adaptive optimal value and grid voltage The output current of the grid-connected inverter at time k+1, obtained through model-free prediction, is: 。 8. The unweighted adaptive model-free predictive control method for grid-connected inverters according to claim 1, characterized in that, The unweighted cost function is: ， In the formula, , The reference current is in the stationary coordinate system. , This represents the output current of the grid-connected inverter at time k+1.

9. A model-free predictive control device for an unweighted adaptive grid-connected inverter, characterized in that, The device includes: The mathematical model module is used to obtain the three-phase voltage and current on the output side of the grid-connected inverter and to establish a mathematical model of the three-level grid-connected inverter based on Kirchhoff's laws. The coordinate transformation module is used to perform Clark transformation on the three-phase voltage and current to obtain the mathematical model of the three-level grid-connected inverter. Model in a two-phase stationary coordinate system; The current reference module is used to obtain the current reference value from the three-phase grid voltage through a phase-locked loop. An unscented Kalman filter is used to obtain an estimated identification value based on the grid-side current and voltage of the grid-connected inverter at time k. The unscented Kalman filter includes: An adaptive Sigma point generation module is used to estimate the current state and capture the characteristics of the state distribution. The prediction module is used to calculate the predicted values ​​of system state variables and observations based on the generated Sigma points; The dynamic adjustment module is used to dynamically adjust the mean and covariance based on the predicted values, and incorporates historical errors into the covariance update. The enhanced noise model module is used to model noise using autoregressive models and multiplicative noise theory to obtain a noise model suitable for actual conditions. The enhanced noise model is then used to adjust the noise covariance and update the process noise and observation noise. A weighted observability is used to determine the observability of a system. The parameter adaptive calculation module is used to obtain the proportional coefficient through an unweighted adaptive algorithm based on the estimated identification values, the three-phase voltage and current on the output side of the grid-connected inverter, and the current reference value. The unweighted adaptive optimal value; The model-free prediction module is used to estimate the identification value and the scaling factor. The unweighted adaptive optimal value and the grid voltage in the two-phase stationary coordinate system are used to obtain the output current of the grid-connected inverter at time k+1 through model-free prediction. The optimal switching state selection module is used to substitute the output current and current reference value of the grid-connected inverter at time k+1 into the unweighted cost function to find the optimal voltage vector control for the switching state of the grid-connected inverter at the next time moment, thereby realizing model-free predictive control of the grid-connected inverter under system parameter mismatch conditions.

Images