Kalman filter active damping iterative learning model predictive control method for grid-connected inverter

Abstract

This invention relates to a Kalman filter active damping iterative learning model predictive control method for grid-connected inverters. First, by using only measured signals of inverter-side current and grid-side voltage through a Kalman filter, the state variables of the LCL filter are estimated in real time, replacing physical sensors and reducing hardware costs. Second, based on the estimated capacitor voltage, the active damping current is calculated, and a composite current reference value for the inverter side is generated to actively suppress resonance. Then, an iterative learning control outer loop is introduced to compensate for periodic disturbances such as dead-zone effects, generating a compensation current using historical error data to improve current quality. Finally, finite set model predictive control is used to optimize the switching state, achieving high-precision tracking. This invention reduces the number of sensors, enhances anti-interference capability and robustness, and significantly reduces the total harmonic distortion of the output current. It is suitable for distributed renewable energy generation systems, ensuring the stable and efficient operation of grid-connected inverters.

Description

Technical Field

[0001] This invention relates to the field of model predictive control, and more specifically to a Kalman filter active damping iterative learning model predictive control method for grid-connected inverters. Background Technology

[0002] Today, with the rapid development of smart grids and distributed generation systems based on renewable energy, grid-connected inverters are widely used to connect power sources to the grid. In industrial applications, high performance of grid-connected inverter systems is essential from the perspectives of current quality requirements injected into the grid and reliable operation in the face of various unforeseen circumstances. To meet grid connection standards, LCL filters are typically used to attenuate harmonics generated by high switching frequencies. Compared to other filter types (such as L-type or LC-type), LCL filters offer superior harmonic suppression capabilities. However, controller design requires greater attention to suppress the inherent resonance phenomenon of LCL filters through passive or active methods. Furthermore, due to the inherent uncertainties in the physical system and the constant presence of external interference, designing a controller to ensure compliant output performance is a challenging task. Therefore, optimizing the control strategy of grid-connected inverters has become a core element in ensuring the safe and stable operation of the power grid.

[0003] In recent years, with the continuous development of the computing power of digital controllers, model predictive control (MPC) has gradually become a research hotspot and development direction in the field of grid-connected inverter control. Among them, finite set model predictive control (FCS-MPC) has attracted widespread attention due to its advantages such as strong multi-objective optimization capability and fast dynamic response speed. When FCS-MPC is used in LCL-type grid-connected inverters, due to the existence of resonance spikes in LCL-type filters, measures must be taken to suppress resonance phenomena. Among various resonance suppression methods, active damping strategies are widely used due to their strong robustness and the fact that they do not require additional passive components.

[0004] However, when using active damping, the capacitor voltage in the LCL filter needs to be sampled, while in industrial applications, only the inverter-side inductor current and output-side voltage are typically sampled. Adding capacitor voltage sampling incurs additional cost and increases maintenance costs. Furthermore, the sensor's measurement signal is susceptible to interference from inverter switching noise and grid harmonics, a problem particularly pronounced when the LCL filter resonates. Secondly, while FCS-MPC offers fast dynamic response, its control accuracy depends on the accuracy of the predictive model. When unmodeled dynamics exist (such as switching voltage drop and dead time), a simple FCS-MPC often produces steady-state errors or specific harmonics. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and to provide a Kalman filter active damping iterative learning model predictive control method for grid-connected inverters, which can effectively solve the problems in the background art.

[0006] The Kalman filter active damping iterative learning model predictive control method for grid-connected inverters includes the following steps: S1: Acquire measurement signals from the inverter system, including inverter-side current and grid-side voltage; S2: Based on the measured signal, the state variables of the LCL filter are estimated in real time using a Kalman filter to obtain the estimated state variables, which include at least the estimated value of the capacitor voltage. S3: Based on the estimated capacitor voltage, calculate the active damping current, and superimpose the active damping current with a preset grid-side current reference value to generate an inverter-side composite current reference value; wherein, the specific method for calculating the active damping current is according to the formula Execution, in which This is an estimated value for the capacitor voltage. This is the preset virtual damping resistance value; S4: Introduce an iterative learning control outer loop to compensate for periodic disturbances, generate a compensation current based on historical error data, and superimpose the compensation current with the composite current reference value on the inverter side to obtain the final current reference value. S5: Using the finite set model predictive control FCS-MPC, with the final current reference value as the tracking target, the optimal switching state of the inverter is determined by optimizing the cost function, and the output is used to control the operation of the inverter.

[0007] Preferably, step S2, estimating the state variables using a Kalman filter, includes the following sub-steps: S2.1: Initialize the state estimates and error covariance matrix of the Kalman filter; S2.2: At each sampling time k, perform the prediction step: based on the previous time... k-1 Given the state estimate and input signal, predict the current time step. k The prior estimates of the state and the prior values ​​of the error covariance; Among them, the state prior estimate is obtained through the state equation. Calculation, where The state matrix, For the input matrix, This is the state estimate from the previous time step. For input signals; S2.3: Perform the correction step: Using the measurement value at the current time k, calculate the Kalman gain and correct the prior state estimate to obtain the posterior state estimate at the current time k as the output; wherein, the Kalman gain is calculated using the formula...

[0008] Calculation, where Let the prior value be the error covariance. C For the output matrix, To measure the noise covariance matrix.

[0009] Preferably, in step S2.2, the state matrix A Input matrix B and output matrix C The discrete model is based on an LCL filter, where the state variables include inverter-side current, capacitor voltage, and grid-side current, and the matrix elements are determined by the filter parameters and sampling time.

[0010] Preferably, in step S3, generating the inverter-side composite current reference value further includes capacitor current feedforward compensation: calculating the capacitor current estimate based on the difference between the capacitor voltage estimates at adjacent sampling times, specifically according to the formula... Execution, in which The value of the filter capacitor is given; and the active damping current is added to the grid-side current reference value, and then superimposed with the estimated capacitor current value to obtain the inverter-side composite current reference value.

[0011] Preferably, in step S4, the iterative learning of the control outer loop includes the following sub-steps: S4.1: Define the current tracking error for the current iteration period n and sampling time k as follows: ,in This is the reference value for the composite current on the inverter side. This refers to the inverter-side current obtained from actual sampling. S4.2: Perform low-pass filtering on the current tracking error to obtain the filtered error signal. ,in These are the filter coefficients; S4.3: Based on historical compensation data and the filtered error signal, the current compensation term is generated through iterative learning of the update law, which is: , in For learning gain, For filtering operators; S4.4: Perform phase lead compensation on the current compensation term and output the compensation current. ,in For the number of steps ahead, This represents the number of sampling points for the fundamental frequency period.

[0012] Preferably, in step S4.3, the filtering operator Q(z) The spatial moving average filter is used, specifically as follows: , in These are weighting coefficients, and they satisfy... .

[0013] Preferably, in step S5, the cost function of FCS-MPC includes a squared error term between the inverter-side current reference value and the predicted value, and an absolute value term for the neutral point potential, specifically in the following form: ， in As a weighting factor, This is the predicted value for the neutral point potential.

[0014] Preferably, the method further includes neutral point potential balance control: based on the switching state and inverter-side current, using the formula... Calculate the neutral point current and predict the change in neutral point potential. ,in 、 、 In switch state. This is the DC-side capacitor value.

[0015] Preferably, steps (1) to (5) are executed in real time during each sampling period, and the sampling frequency of the measurement signal is consistent with the control period, ensuring that the Kalman filter, active damping calculation, iterative learning compensation and FCS-MPC optimization form a closed-loop control chain.

[0016] Preferably, the process noise covariance matrix of the Kalman filter... and measurement noise covariance matrix Presets are made based on system model uncertainties and sensor accuracy to optimize the robustness of state estimation.

[0017] Compared with the prior art, the present invention has the following beneficial effects: 1. Reduced number of sensors: LCL filters include grid-side inductors, inverter-side inductors, and intermediate capacitors. Traditional sensor-based solutions require independent sensors (such as Hall sensors) for each inductor current (inverter-side inductor current, grid-side inductor current) and capacitor current. However, Kalman filters can calculate all the signals to be observed by combining a few key measurements (only inverter-side current and grid-side voltage) with the system model, which greatly reduces the cost of sensor procurement and installation.

[0018] 2. Stronger anti-interference capability: Sensor measurement signals are easily affected by inverter switching noise, power grid harmonics, etc., while the Kalman filter is essentially an optimal estimator. It can dynamically suppress measurement noise and model error through noise covariance matrix design, and output smoother and more reliable current estimates, which is especially suitable for environments with complex interference.

[0019] 3. Provides redundant observation capability: When a small number of sensors in the system fail, the Kalman filter can continue to output a reliable current estimate based on the remaining measurement values ​​and the system model, achieving a certain degree of fault-tolerant control. In contrast, the failure of a single sensor in a full sensor scheme may directly lead to system instability.

[0020] 4. By introducing an iterative learning outer loop, this invention can effectively identify and compensate for periodic disturbances caused by inverter dead-time effects, switching transistor voltage drops, and nonlinear loads, significantly reducing the total harmonic distortion of the output current without increasing hardware costs. Through a periodic error correction mechanism, it automatically compensates for prediction deviations caused by inductor parameter perturbations or aging, enabling the system to maintain excellent control performance even under varying model parameters. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the Kalman filter active damping iterative learning model predictive control method for grid-connected inverters in this invention.

[0022] Figure 2 This is a three-level LCL grid-connected inverter circuit topology; Figure 3 This is a block diagram of the control strategy in this invention; Figure 4 The figure shows a comparison and simulation analysis of the sensor scheme and the proposed Kalman filter scheme under steady-state conditions in this invention. Figure 5 This is a comparison chart of the capacitor voltage estimated by the Kalman filter and the capacitor voltage measured by the sensor in this invention; Figure 6 This is the grid-connected current diagram when active damping is not used in this invention. Detailed Implementation

[0023] To make the technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention: Step 1: For the LCL-type grid-connected inverter shown in Figure 1, the present invention establishes the following prediction model. When implementing output current control of an LCL-type inverter, it is generally achieved by indirectly controlling the grid-side current by controlling the inverter-side inductor current. Each phase of a midpoint clamped inverter consists of four switching devices and two clamping diodes, which can generate three switching states, namely... Where x represents one of the three phases a, b, and c. The corresponding output voltage of phase x is... ,in This represents the DC side voltage. Variables in the three-phase abc coordinate system can be transformed into variables in the two-phase stationary αβ coordinate system using the Clarke transformation. The specific formula for the Clarke transformation is as follows:

[0024] in Let represent the variables in the three-phase abc coordinate system, respectively. Let represent the variables in the two stationary αβ coordinate systems, respectively. Given... Figure 2 The circuit in the diagram can be modeled using the Clarke transform to establish a dynamic model of the inverter-side current in the continuous time domain within the two-phase stationary αβ coordinate system. (2) in For the inverter side inductor, For inductor parasitic resistance, Let be the current on the inverter side in the two-phase stationary αβ coordinate system. This represents the inverter-side voltage. In the FCS-MPC paradigm, the system state needs to be predicted in the discrete-time domain. The dynamic characteristics of the inverter-side current in the discrete-time domain are as follows: (3) in, It is the sampling time. Indicates the sampling time. This refers to the voltage of the filter capacitor. The DC side of a three-level neutral-point clamped inverter consists of two capacitors with a nominal voltage of (1 / 2) Vdc. Besides load current control, to ensure the inverter operates normally, the voltage values ​​of these two capacitors must be equal. For simplicity, the balance of the two capacitor voltages can be equivalent to the neutral point potential. The control of is defined as follows: (4) in and Two capacitors on the DC side. C1 and C2The voltage. Assuming the switch operates ideally, the current flowing through the neutral point... The relationship between inverter-side current and switch position is as follows: (5) Thus, the prediction equation for the midpoint potential under discrete conditions can be obtained as shown in (6), where DC side capacitor value: (6) in, This represents the midpoint potential at two adjacent moments. The difference.

[0025] Step 2: Design of the FCS-MPC Controller A key feature of the FCS-MPC framework is the design of its cost function and its online rolling optimization mechanism. In this case study, the prediction errors of the inverter-side current and neutral point potential are incorporated into the cost function, as specifically designed below: (7) in For the final reference current value, This is the weighting factor. This indicates that the minimum value is taken. By traversing the 27 candidate voltage vectors in the finite control set, the solution that minimizes equation (7) is found and used as the output of the controller.

[0026] Step 3: Design of the Kalman Filter In this invention, by designing a suitable Kalman filter, all state variables in the LCL filter can be observed by sampling only the inverter-side current and grid-side voltage. The Kalman filter mainly consists of three stages: prediction, update, and correction. The difference equation of the LCL inverter can be set as follows: (8) in A, B, C The matrices represent the system's state matrix, input matrix, and output matrix, respectively. x This represents the state variables, which contain all the key variables describing the dynamic characteristics of the LCL filter; u This represents the input quantity, which includes external input quantities that affect the system state. y This represents the system's observation vector, which embodies the physical quantities actually measured by the sensors. In this design, only the inverter-side current needs to be measured additionally. 。 ， This represents noise. Based on the third-order characteristics of the LCL filter, the state variables and input variables can be defined as follows:

[0027] (9) in, In order to be in k The grid voltage at any given time This represents the grid-side current. The state matrix, input matrix, and output matrix can be set according to the mathematical model of the LCL filter: (10) in, The matrix is ​​the state matrix, which describes the relationships between the state variables within the system, and is derived from the parameters of the LCL filter. , , and the sampling time of the control system A joint decision. The input matrix describes How to affect the state of the system. The output matrix describes the relationship between the state variables and the actual measured values. This matrix shows that the observed values... y(k) Just a state vector x(k) The first element in .

[0028] The Kalman filter is a recursive estimation algorithm whose core idea is "prediction + correction". At each time step, the Kalman filter first predicts the current state based on the state estimate from the previous time step, and then corrects this prediction using the actual measurement value at the current time step, thus obtaining a more accurate estimate. The algorithm mainly consists of the following two stages and five core formulas.

[0029] Initialization: Before the algorithm starts, an initial state estimate needs to be provided. and the initial error covariance matrix .

[0030] Phase 1: Prediction State prediction: based on the previous time step Optimal state estimate and input Predict the current moment The state.

[0031] (11) in, Indicates in The posterior optimal estimate at time t. Indicates in Prior estimates at time 1.

[0032] Error covariance prediction: Update the error covariance matrix, which represents the degree of uncertainty of the predicted state values.

[0033] (12) in for The posterior error covariance matrix at time t. for k The prior error covariance matrix at time t. The process noise covariance matrix reflects the degree of confidence in the system model. The larger the value of the matrix, the less accurate the system model is. Representation matrix The transpose of .

[0034] Phase Two: Correction Calculate the Kalman gain: The Kalman gain is used to determine the weight of the "predicted value" and the "measured value" in the final estimate.

[0035] (13) in, for Kalman gain at time step Representation matrix C The transpose of . The noise covariance matrix is ​​measured to reflect our level of trust in the sensor. The smaller the value, the more accurate the sensor measurement result; Kalman gain. It will get bigger and bigger.

[0036] Status update: using actual measured values To correct the prior estimate To obtain the posterior optimal estimate at the current time. .

[0037] (14) in, It is called the residual, which is the difference between the actual measured value and the predicted measured value. for The posterior optimal estimate at time t is the final result of the filter output at the current time.

[0038] Error covariance update: Update the posterior error covariance matrix to reflect the reduction in uncertainty of the state estimate after measurement correction.

[0039] (15) in, Represents the identity matrix. express k The posterior error covariance matrix at time step 1 will be used as the basis for the next time step 2. The input for the iteration. Derived from the posterior optimal estimate. Estimated output capacitor voltage Smooth and accurate, it is suitable as an input to the active damping element in model predictive control, which can actively suppress the inherent resonance spikes of the LCL filter, thus creating a prerequisite for achieving high-stability grid-connected control.

[0040] Step 4: Active Damping Control of FCS-MPC Based on Kalman Filter To avoid or mitigate the resonance effect of LCL-type grid-connected inverters, active damping control is required. Traditional passive damping methods, which involve connecting physical resistors in series or parallel within the filter, introduce additional ohmic losses. A superior approach is active damping, which simulates the effect of a virtual resistor in the control algorithm, thereby suppressing resonance without increasing physical losses. Conventional active damping schemes typically rely on direct measurement and feedback of the filter capacitor voltage. However, this approach not only requires additional voltage sensors, increasing hardware costs and system complexity, but also the measurement noise introduced by the sensors can adversely affect the stability of the damping effect. This invention eliminates the need for physical voltage sensors. Instead, it reconstructs the feedback channel using the state observations of a Kalman filter, realizing a virtual sensor function and eliminating the limitation of active damping control bandwidth imposed by physical sensor measurement noise.

[0041] First, define the desired current injected into the grid. It consists of two parts: one part is the grid-side current reference from the power control loop. The other part is the active damping current used to suppress resonance. The active damping current is estimated from the capacitor voltage. and virtual damping resistor The calculation shows that: (16) in, It is a real-time, accurate, and filtered capacitor voltage estimate provided by the Kalman filter described in step three. It is a preset virtual damping resistance value, the size of which can be adjusted according to the desired damping effect.

[0042] The obtained active damping current Reference value of the original power loop network side current to be superimposed on the inverter This forms a new composite network-side current reference value that includes active damping information. : (17) Based on the physical structure of the LCL filter, the inverter-side current... It is the grid-side current. With capacitor current The sum of To ensure the actual grid-side current... It can accurately track the target value calculated in the previous step. The influence of capacitor current must be considered in advance in the current reference on the inverter side, i.e., feedforward compensation must be performed. Using the capacitor voltage estimates provided by the Kalman filter at adjacent time points, the estimated capacitor current required for compensation can be accurately calculated. .

[0043] (18) Reference value of composite network side current Feedforward compensation component of capacitor current By adding them together, we can obtain the complete inverter-side composite current reference value for the FCS-MPC controller. : (19) In this way, the present invention further improves the control accuracy. It makes full use of the rich state information provided by the Kalman filter to achieve precise decoupling and feedforward control of the system's internal dynamics, which is not possible with simple feedback superposition methods, and greatly improves the system's dynamic response speed and steady-state control accuracy.

[0044] Step 5: Iterative learning outer loop design: Although step four obtains the inverter-side composite current reference value through active damping and feedforward compensation. While effectively suppressing resonance, in practical applications, the dead-time effect and the voltage drop of the switching transistor mainly manifest as periodic disturbances synchronized with the fundamental frequency, and FCS-MPC has limited ability to suppress unmodeled periodic disturbances. Therefore, this invention further introduces an iterative learning control loop.

[0045] Introducing step four As the objective, the error is defined as follows: (20) in, For the first n The inverter-side current actually sampled during the next iteration. Considering that the current under FCS-MPC control contains high-frequency switching ripple, the original error is directly used. Learning can lead to control divergence. Therefore, we define the learning error after preprocessing. High-frequency noise is eliminated through a first-order low-pass filter: (twenty one) in, The smaller the value, the stronger the ability to filter out high-frequency noise. It is used to ensure that the error signal input to the learning law mainly contains the fundamental wave and low-order harmonic components.

[0046] Build length is N ( N A ring-shaped storage unit (corresponding to the number of sampling points for the fundamental frequency period) is used to store compensation values. Definition This is the learning compensation term for the nth iteration. To prevent integral drift and improve the robustness of the algorithm across the entire frequency band, a Q-filtering mechanism and learning gain are introduced, with the update law as follows: (twenty two) in, The gain is learned iteratively. Here, to further smooth the waveform, this invention employs a zero-phase-shift spatial moving average filter as... Operator, that is, using the current time of the previous cycle. and its neighboring points The weighted average as historical experience: (twenty three) in, These are weighting coefficients, and they satisfy... The physical significance of this step is to prevent the learning process from diverging due to accidental interference by forgetting outdated error information and smoothing out abrupt changes in adjacent time steps.

[0047] Because the FCS-MPC control system has inherent calculation delay and sample-and-hold delay, if the compensation amount calculated at the current moment is directly applied... Superposition can lead to compensation lag and even system oscillation. Therefore, this invention introduces a phase lead compensation mechanism. The final output compensation current is defined. For data from future moments read from storage units: (twenty four) in, The number of phase lead steps. To prevent index out-of-bounds errors during modulo operations, this embodiment sets... This is to precisely compensate for the inherent delay of the digital control system, ensuring that the compensation effect is strictly aligned with the actual disturbance on the time axis. The composite reference value obtained in step four, which includes active damping, is then used... The iterative compensation value obtained in step five By performing linear superposition, the final reference current value used for FCS-MPC cost function evaluation is obtained. : (25) Finally, the result from step five Substitute into the cost function described in step two, replacing . The optimal solution is found by traversing the voltage vector. At this point, the system has completed a full control closed loop, from state observation and active damping suppression to iterative learning error compensation.

[0048] Step Six: Simulation Verification The technical effects of the above-mentioned data-driven three-level neutral-point clamped grid-connected inverter control method based on dynamic linearization are further illustrated below using simulation diagrams: The simulation was performed using Matlab / simulink, and the main parameters in the simulation are shown in Table 1.

[0049]

[0050] in Here, f is the sampling frequency, and f is the power grid frequency. This represents the effective value of the grid current.

[0051] Figure 4 The comparison of grid-connected current waveforms between the solution described in this invention and a traditional sensor solution under the same operating conditions is presented to verify the superiority of this invention. Figure 4 (a) is the grid-connected current waveform obtained by using the control scheme based on the Kalman filter observer described in this invention. Figure 4 (b) is the waveform obtained by using a control scheme with feedback from a traditional capacitor voltage sensor.

[0052] Power quality analysis of the two waveforms shows that the total harmonic distortion (THD) of the proposed solution is 1.37%, significantly lower than the 2.25% of the traditional solution. This result strongly demonstrates that the control strategy of combining a Kalman filter with active damping in this invention can more effectively suppress current harmonics, output higher quality sinusoidal current, and thus achieve superior steady-state control performance.

[0053] Figure 5 Taking phase A as an example, the actual measured value of the filter capacitor voltage is visually compared with the real-time observed value of the Kalman filter in this invention, aiming to verify the accuracy and high-quality characteristics of the observer. It can be clearly seen from the figure that, compared to the measured waveform with obvious glitches and high-frequency noise, the estimated waveform of the Kalman filter is extremely smooth and clean.

[0054] This high-quality estimation signal has a dual advantage: on the one hand, it provides a more accurate initial state for the prediction model of FCS-MPC, improving prediction accuracy; on the other hand, as the core feedback signal of active damping, it avoids introducing measurement noise into the closed-loop control system, thereby greatly enhancing the stability and robustness of the active damping strategy.

[0055] Figure 6 As a key comparative experiment, the system grid-connected current waveform is shown when the active damping function of this invention is disabled, highlighting the necessity of the active damping element. As shown in the figure, without active damping, the grid-connected current exhibits severe distortion and high-frequency oscillations caused by the resonant characteristics of the LCL filter.

[0056] This condition not only severely degrades the grid-connected power quality, failing to meet grid standards, but also risks triggering the system's overcurrent protection and even damaging power semiconductor devices. This result clearly confirms that active damping is a key technology for ensuring the stable and safe operation of LCL-type grid-connected inverters, thus highlighting the significant value of the observer-based high-efficiency active damping scheme proposed in this invention.

[0057] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A Kalman filter active damping iterative learning model predictive control method for grid-connected inverters, characterized in that, Includes the following steps: S1: Acquire measurement signals from the inverter system, including inverter-side current and grid-side voltage; S2: Based on the measured signal, the state variables of the LCL filter are estimated in real time using a Kalman filter to obtain the estimated state variables, which include at least the estimated value of the capacitor voltage. S3: Based on the estimated capacitor voltage, calculate the active damping current, and superimpose the active damping current with a preset grid-side current reference value to generate an inverter-side composite current reference value; wherein, the specific method for calculating the active damping current is according to the formula Execution, in which This is an estimated value for the capacitor voltage. This is the preset virtual damping resistance value; S4: Introduce an iterative learning control outer loop to compensate for periodic disturbances. Generate a compensation current based on historical error data and superimpose it with the composite current reference value on the inverter side to obtain the final current reference value. S5: Using the finite set model predictive control FCS-MPC, with the final current reference value as the tracking target, the optimal switching state of the inverter is determined by optimizing the cost function, and the output is used to control the operation of the inverter.

2. The Kalman filter active damping iterative learning model predictive control method for grid-connected inverters according to claim 1, characterized in that, In step S2, estimating the state variables using a Kalman filter includes the following sub-steps: S2.1: Initialize the state estimates and error covariance matrix of the Kalman filter; S2.2: At each sampling time k, perform the prediction step: based on the previous time... k-1 Given the state estimate and input signal, predict the current time step. k The prior estimates of the state and the prior values ​​of the error covariance; Among them, the state prior estimate is obtained through the state equation. Calculation, where The state matrix, For the input matrix, This is the state estimate from the previous time step. For input signals; S2.3: Perform the correction step: Using the measurement value at the current time k, calculate the Kalman gain and correct the prior state estimate to obtain the posterior state estimate at the current time k as the output; wherein, the Kalman gain is calculated using the formula... Calculation, where Let the prior value be the error covariance. C For the output matrix, To measure the noise covariance matrix.

3. The Kalman filter active damping iterative learning model predictive control method for grid-connected inverters according to claim 2, characterized in that, In step S2.2, the state matrix A Input matrix B and output matrix C The discrete model is based on an LCL filter, where the state variables include inverter-side current, capacitor voltage, and grid-side current, and the matrix elements are determined by the filter parameters and sampling time.

4. The Kalman filter active damping iterative learning model predictive control method for grid-connected inverters according to claim 1, characterized in that, In step S3, generating the inverter-side composite current reference value also includes capacitor current feedforward compensation: calculating the capacitor current estimate based on the difference between the capacitor voltage estimates at adjacent sampling times, specifically according to the formula... Execution, in which This is the value of the filter capacitor; and the active damping current Compared with grid-side current reference value After adding them together, they are then superimposed with the estimated capacitor current to obtain the reference value of the composite current on the inverter side.

5. The Kalman filter active damping iterative learning model predictive control method for grid-connected inverters according to claim 1, characterized in that, In step S4, the iterative learning of the control outer loop includes the following sub-steps: S4.1: Define the current tracking error for the current iteration period n and sampling time k as follows: ,in This is the reference value for the composite current on the inverter side. This is the actual sampling current; S4.2: Perform low-pass filtering on the current tracking error to obtain the filtered error signal. ,in These are the filter coefficients; S4.3: Based on historical compensation data and the filtered error signal, the current compensation term is generated through iterative learning of the update law, which is: ， in For learning gain, For filtering operators; S4.4: Perform phase lead compensation on the current compensation term and output the compensation current. ,in For the number of steps ahead, This represents the number of sampling points for the fundamental frequency period.

6. The Kalman filter active damping iterative learning model predictive control method for grid-connected inverters according to claim 5, characterized in that, In step S4.3, the filtering operator Q(z) A spatial moving average filter designed to enhance robustness is employed, specifically as follows: ， in These are weighting coefficients, and they satisfy... .

7. The Kalman filter active damping iterative learning model predictive control method for grid-connected inverters according to claim 1, characterized in that, In step S5, the cost function of FCS-MPC includes a squared error term between the inverter-side current reference value and the predicted value, and an absolute value term for the neutral point potential, specifically in the form of: ， in As a weighting factor, This is the predicted value for the neutral point potential.

8. The Kalman filter active damping iterative learning model predictive control method for grid-connected inverters according to claim 1, characterized in that, The method also includes neutral point potential balance control: based on the switching state and inverter-side current, using a formula... Calculate the neutral point current and predict the change in neutral point potential. ,in 、 、 In switch state. This is the DC-side capacitor value.

9. The Kalman filter active damping iterative learning model predictive control method for grid-connected inverters according to claim 1, characterized in that, Steps (1) to (5) are executed in real time during each sampling period, and the sampling frequency of the measured signal is consistent with the control period, ensuring that the Kalman filter, active damping calculation, iterative learning compensation and FCS-MPC optimization form a closed-loop control chain.

10. The Kalman filter active damping iterative learning model predictive control method for grid-connected inverters according to claim 1, characterized in that, The process noise covariance matrix of the Kalman filter and measurement noise covariance matrix Presets are made based on system model uncertainties and sensor accuracy to optimize the robustness of state estimation.

Images