Deadbeat Predictive Control Method for Current Source Inverters Based on Extended State Observer

By combining second-order Taylor expansion and MAESO, the discretization accuracy and parameter drift problems of current source inverters in microgrid off-grid operation mode are solved, realizing high-frequency precise control and steady-state zero-steady-state error tracking, and improving the robustness and dynamic response capability of the system.

CN123121827APending Publication Date: 2026-11-17JINHUIYUAN POWER GRP CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611177807.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-05
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing current source inverters suffer from problems such as insufficient discretization accuracy, sensitivity to parameter drift, computational delay and bandwidth limitations, and an imbalance between computational burden and control effect when operating in off-grid microgrid mode, leading to high-frequency control failure and steady-state error.

Method used

A high-frequency precise discretization prediction model based on second-order Taylor expansion is adopted, combined with a model-assisted extended state observer (MAESO), to construct an extended state equation containing lumped disturbances. An error feedback gain matrix is ​​designed and parameters are tuned. Optimal control commands are achieved through feedforward compensation.

Benefits of technology

It achieves robust mathematical coupling between inverter output pulse current and load inductor current, eliminates high-frequency control blind zone, improves robustness against hardware parameter drift, reduces computational latency, expands bandwidth and improves THD.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123121827A_ABST
    Figure CN123121827A_ABST
Patent Text Reader

Abstract

This invention discloses a deadbeat predictive control method for current source inverters using an extended state observer, belonging to the field of power electronics control. This method targets the high-frequency operating conditions of off-grid microgrids. First, it employs a second-order Taylor expansion for precise discretization, establishing a mathematical coupling between inverter current and inductor current to eliminate high-frequency control blind spots. Then, a Model-Assisted Extended State Observer (MAESO) is designed to deeply integrate prior information from the system's physical model, uniformly estimating and feedforward compensating for parameter drift and load disturbances. The observer gain is tuned based on the Jury criterion to ensure system stability. Finally, the deadbeat optimal control command is calculated using the Moore-Penrose pseudo-inverse. This invention overcomes the shortcomings of traditional methods, such as low discretization accuracy, sensitivity to parameters, and heavy computational burden, achieving high-precision, robust, and fast-response current control, effectively improving the power supply quality of off-grid microgrids.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of power electronic system control technology, specifically relating to a deadbeat predictive control method for a current source inverter based on a model-assisted extended state observer, applicable to off-grid operation mode of a microgrid. Background Technology

[0002] With the rapid development of microgrid technology, inverters, as core interface devices, directly impact power quality and system stability through their control performance. Compared to traditional voltage source inverters (VSI), current source inverters (CSI) offer advantages such as lower output current ripple, natural short-circuit protection, and boost characteristics. In high-performance current control within microgrids, deadbeat predictive current control (DPCC) is widely used due to its fast dynamic response. To further improve system robustness, extended state observers (ESOs) are often introduced to estimate and compensate for parameter fluctuations and load disturbances. Off-grid operation is one of the core operating conditions of microgrids, where the inverter directly supplies power to external, variable resistive electronic loads. Traditional discretization modeling typically employs the first-order forward Euler method, transforming continuous state-space equations into discrete predictive models.

[0003] In applications targeting off-grid power supply for microgrids, existing CSI deadbeat predictive control has the following drawbacks:

[0004] (1) Insufficient discretization accuracy leads to inaccurate prediction (model decoupling): CSI combined with filter components constitutes a second-order system. Microgrid inverters operate at high frequencies (such as 110k-200kHz). Traditional first-order forward Euler discretization will lose higher-order terms, resulting in the current input being unable to affect the next prediction state, causing high-frequency deadbeat control to completely fail.

[0005] (2) Extremely sensitive to parameter drift and complex load disturbances: DPCC is highly dependent on physical parameters (such as filter inductor L and capacitor C). During long-term operation of the microgrid, these hardware parameters will drift significantly (up to 30%-50%) with temperature rise and aging, resulting in steady-state error, waveform distortion and even high-frequency instability of the output current.

[0006] (3) Computational delay and bandwidth limitations: Traditional ESO relies purely on error feedback and does not utilize the known prior physical model of the CSI system. The computational load is large and there is phase lag, which limits the current loop bandwidth and results in slow transient recovery when dealing with sudden changes in microgrid load.

[0007] (4) Imbalance between computational burden and control effect: Existing improvement methods improve accuracy through complex matrix exponents or multi-step prediction, which leads to a sharp increase in computational load and is not conducive to high-frequency real-time control. Summary of the Invention

[0008] The objective of this invention is achieved through the following technical solutions.

[0009] Specifically, this invention provides a deadbeat predictive control method for current source inverters with an extended state observer, applied to current source inverters in off-grid operation mode of microgrids, comprising the following steps:

[0010] Step 1, State Sampling and Coordinate Transformation: The current of the filter inductor and the voltage of the filter capacitor output by the inverter are sampled in real time, and the Park transformation is used to transform them from the stationary abc coordinate system to the synchronous rotating dq coordinate system to obtain the state measurement value at the current moment, and the AC bus voltage of the external resistive load is regarded as an external disturbance.

[0011] Step 2: Establish a high-frequency accurate discretized prediction model based on second-order Taylor expansion: Using the Taylor expansion that retains the second-order terms, the continuous state-space equation of the current source inverter in the dq coordinate system is discretized to establish a discrete prediction model. This makes non-zero coupling terms appear in the input matrix of the discrete prediction model, thereby establishing the mathematical coupling relationship between the inverter output pulse current and the load inductor current.

[0012] Step 3: Construct extended state equations that include lumped disturbances: summarize hardware parameter drift, unmodeled line voltage drop and nonlinear load impedance changes into system lumped disturbances, and construct extended state equations that include both system state variables and lumped disturbance variables.

[0013] Step 4: Design and tune the model-assisted extended state observer: Integrate the prior information of the discrete prediction model described in Step 2 to construct a model-assisted extended state observer in the discrete domain. Iteratively update the system state estimate and lumped disturbance estimate through state error feedback and disturbance error feedback. The error feedback gain coefficient of the observer is tuned based on the Jury stability criterion.

[0014] Step 5: Feedforward compensation and calculation of the deadbeat-free optimal command: The lumped disturbance estimate output by the model-assisted extended state observer is fedforward to the deadbeat-free predictive control equation, and the optimal control command in the least squares sense is solved using the Moore-Penrose generalized inverse matrix based on the input matrix of the discrete predictive model.

[0015] The advantages of this invention are:

[0016] (1) Reconstructing high-frequency precise mathematical coupling: Abandoning the first-order approximation, adopting precise discretization modeling that retains the second-order Taylor expansion terms, extracting key coupling parameter terms, establishing a solid mathematical connection between the inverter output pulse current and the load inductor current, and eliminating the high-frequency control blind zone.

[0017] (2) Extremely strong robustness against hardware parameter mismatch: MAESO is used to convert the model error caused by L and C hardware aging or temperature drift into internal lumped disturbance for accurate estimation and feedforward cancellation, completely immune to instability caused by parameter drift.

[0018] (3) Elimination of static error of load disturbance without integrator: Under off-grid conditions, the resistive load parameters are not fixed. The complex nonlinear impedance changes of electronic load and AC bus voltage fluctuations are uniformly regarded as external lumped disturbances and are tracked and compensated by MAESO in real time. Static error-free tracking can be achieved without PI integration, thus improving THD.

[0019] (4) Reduce computational latency and expand bandwidth: MAESO deeply integrates prior information of the second-order CSI discrete physical model (using known models to predict state changes, and the observer only calculates unknown deviations), which greatly reduces the computational burden of the digital controller and shortens the response time. Attached Figure Description

[0020] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0021] Figure 1 This is a schematic diagram of a three-phase two-level off-grid current source inverter (CSI) topology.

[0022] Figure 2 This is a schematic diagram showing that isd and isq cannot track the reference current iw(k) at time (k+1) under the forward Euler approximation discrete model.

[0023] Figure 3 This is a block diagram of the overall structure of the MAESO-CSI-DPCC control method proposed in this invention.

[0024] Figure 4 The following diagram shows the simulation results of the method of the present invention under steady-state conditions, where (a) is the steady-state current waveform and (b) is the THD analysis diagram.

[0025] Figure 5 The following diagram shows the simulation results of the method of the present invention under dynamic working conditions, where (a) is the three-phase current waveform and (b) is the dq-axis current waveform. Detailed Implementation

[0026] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0027] Terminology Explanation:

[0028] Current source inverter (CSI): As the core interface device between distributed power sources and microgrids / loads, it can effectively suppress current ripple due to the structure of parallel filter capacitors on its output side, and has natural short-circuit protection and boost characteristics.

[0029] Deadbeat Predictive Current Control (DPCC): A type of Model Predictive Control (MPC) that uses a discrete mathematical model of the system to predict the state variables at the next moment, enabling the feedback current to accurately track the reference value and calculate the required inverter output command.

[0030] Model-Aided Extended State Observer (MAESO): An observer that deeply integrates prior information from known physical discrete models of a system. It transforms internal parameter fluctuations (such as hardware aging) and external nonlinear load disturbances into lumped disturbances for accurate estimation and feedforward compensation, and does not suffer from the severe phase lag of traditional observers.

[0031] Off-grid operation mode: The microgrid operates independently from the main grid. In this state, the microgrid system exhibits low inertia characteristics, and the inverter must directly face electronic loads with resistive characteristics, making it highly susceptible to unmodeled voltage drops and external disturbances.

[0032] A. System topology and off-grid mathematical model

[0033] Figure 1 It is a three-phase two-level current source inverter. This topology is configured with 6 switching transistors S1-S6 and a DC-side inductor L. dc The input side has a DC voltage source connected to a DC-DC converter, which uses PI control to achieve independent control of the DC current. The output side is equipped with a voltage sensor and a current sensor to acquire the actual capacitor voltage and inductor current in real time, forming a full-state measurement vector x(k).

[0034] On the AC output side, the pulse width modulation current output by the CSI inverter bridge is defined as i. w The current flowing through the parallel filter capacitor C is i c The output current flowing through the series filter inductor L is i s The AC bus voltage on the resistive load side is v busBased on Kirchhoff's current and voltage laws, the dynamic equations are established as follows:

[0035]

[0036] To achieve zero steady-state error control of AC quantities, the Park transformation is introduced to convert the variables in the stationary abc coordinate system to those in the electric angular velocity system. In the dq coordinate system rotated by e:

[0037]

[0038] In this system, we set the state variable as The system control input variable is The AC bus voltage caused by the electronic load is considered as a known external disturbance moment E of the system. Since the resistance is externally given and its characteristics are variable, we do not fix the resistance parameters in the model matrix. The simplified continuous-time state-space equations are as follows:

[0039]

[0040] B. High-frequency discretization based on second-order Taylor expansion

[0041] The system's control frequency range is 110k-200k, and the sampling period is T. s Extremely short. If discretization using the first-order forward Euler method is employed, the input matrix... The first two lines are still 0, which means that the CSI current command i issued at time k is... w (k) The inductor current i at time k+1 cannot be affected in the calculation. s (k+1), thus causing the beatless predictive control to fail. Therefore, an accurate discretization method that retains the second-order Taylor expansion must be adopted:

[0042]

[0043] The formula for calculating the system discrete matrix is:

[0044]

[0045] We will separately consider the input matrix Expanding on this to verify the effectiveness of the second-order discretization:

[0046]

[0047] Through second-order Taylor expansion, the matrix Non-zero items appeared in the first and second rows. A mathematical coupling between the inverter output pulse current and the load-side filter inductor current was successfully established.

[0048] In actual operation, external electronic loads exhibit nonlinear characteristics, and the L and C parameters drift with temperature and high-frequency conditions. We collectively categorize these factors not accurately represented in the model (including unmodeled line resistance and load impedance voltage drop) as system lumped disturbances. Construct the extended state equations:

[0049]

[0050] C. MAESO Observer Design and Jury Criterion Parameter Tuning

[0051] The design and parameter tuning process of the Model-Aided Extended State Observer (MAESO) and the error feedback gain matrix are as follows: The discrete-domain Model-Aided Extended State Observer is constructed, and its iterative update equation is:

[0052]

[0053]

[0054]

[0055] Where x(k) is the measured value of the actual output state of the system, z1(k) is the observed value of the system state variable, z2(k) is the observed value of the system lumped disturbance, and e(k) is the state observation error. and These are the state error feedback gain matrix and the disturbance error feedback gain matrix, respectively. Further, the... and Configured in diagonal matrix form, specifically represented as follows:

[0056]

[0057] The physical configurations of each gain parameter are as follows: h1 is used to correct the state observation error of the inductor current; h2 is used to correct the state observation error of the filter capacitor voltage; h3 is used to track and compensate for lumped disturbances acting on the current loop (such as inductor parameter drift); h4 is used to track and compensate for lumped disturbances acting on the voltage loop (such as commutator capacitor parameter decay and disturbances caused by sudden changes in external electronic loads of the microgrid).

[0058] To ensure the asymptotic stability of the microgrid off-grid inverter under high-frequency control cycles, the value ranges of the gain parameters h1, h2, h3, and h4 are tuned using an analysis method based on the Jury stability criterion. This tuning method includes the following sub-steps:

[0059] Sub-step 1: Subtract the discrete state-space equation of the actual off-grid inverter from the iterative update equation of the observer to obtain the dynamic equation of the observer's error system, and derive its closed-loop characteristic polynomial P(λ), expressed as:

[0060]

[0061] Where λ is the discrete domain characteristic operator; a4, a3, a2, a1, a0 are the coefficients of the algebraic polynomial composed of the microgrid inverter filter inductor L, filter capacitor C, discrete sampling period Ts, and the gain parameters h1-h4 to be tuned.

[0062] Sub-step 2: Construct a Jury array of characteristic polynomials based on the Jury stability criterion for discrete systems. To ensure that the system error asymptotically converges to zero (i.e., all roots of the characteristic polynomial lie within the unit circle in the Z-plane), establish a set of boundary conditions and determinant constraint inequalities that satisfy stability:

[0063]

[0064] Where bi,ci are the intermediate row coefficients of the Jury array.

[0065] Sub-step 3: Set the system physical reference parameters (such as rated L, C values ​​and control frequency f) s Substituting these values ​​into the system of constraint inequalities, the multivariable nonlinear inequalities are solved to obtain the parameter boundary domain that allows the observer to converge stably. For typical high-frequency control conditions of microgrid inverters (e.g., sampling frequency at 20kHz), after the above Jury criterion tuning calculations, the optimal parameter range of the error feedback gain matrix is ​​set as follows:

[0066]

[0067] By employing the feedback gain matrix tuned based on the Jury stability criterion, optimal tracking of nonlinear disturbances in the electronic load and drift in internal hardware parameters can be achieved while ensuring absolute stability in the discrete control domain of the microgrid off-grid inverter. Appropriately increasing h3 and h4 can effectively accelerate the observer's response to transient disturbances in the microgrid and shorten the dynamic recovery time; simultaneously, limiting them to the upper threshold calculated by the Jury criterion avoids amplifying high-frequency switching noise due to excessive feedback gain, thereby eliminating the potential for high-frequency oscillations in the system.

[0068] D. Feedforward compensation and optimal instruction calculation

[0069] The core idea of ​​deadbeat control is to ensure that the state variable reaches the reference value precisely in the next control cycle. Considering the one-beat delay in digital control systems, we need to predict the state at time k+1 based on the current state at time k. And calculate the effect applied at time k+1 that makes the state at time k+2 reach the reference value. Control commands.

[0070] The reference state vector is set as follows:

[0071]

[0072] The disturbance term predicted by the observer Feedforward compensation is performed by injecting into the deadbeat-free equation:

[0073]

[0074] Thanks to the aforementioned precise discretization that preserves the second-order Taylor expansion, the input matrix... The presence of non-zero current coupling terms makes it a 4×2 column full-rank non-square matrix with control effectiveness. Based on this, to calculate the control variables that minimize the error of the objective function, the Moore-Penrose generalized inverse matrix can be used to obtain the least-squares optimal solution:

[0075]

[0076] Finally, the CSI current reference command used to send to the DSP to generate the SVPWM drive signal is extracted:

[0077]

[0078] F. Control block diagram of the proposed algorithm

[0079] The control structure of MAESO-CSI-DPCC is as follows: Figure 3As shown: Specifically, this invention provides a deadbeat predictive control method for current source inverters with an extended state observer, applied to current source inverters in off-grid operation mode of microgrids, including the following steps:

[0080] Step 1, State Sampling and Coordinate Transformation: The current of the filter inductor and the voltage of the filter capacitor output by the inverter are sampled in real time, and the Park transformation is used to transform them from the stationary abc coordinate system to the synchronous rotating dq coordinate system to obtain the state measurement value at the current moment, and the AC bus voltage of the external resistive load is regarded as an external disturbance.

[0081] Step 2: Establish a high-frequency accurate discretized prediction model based on second-order Taylor expansion: Using the Taylor expansion that retains the second-order terms, the continuous state-space equation of the current source inverter in the dq coordinate system is discretized to establish a discrete prediction model. This makes non-zero coupling terms appear in the input matrix of the discrete prediction model, thereby establishing the mathematical coupling relationship between the inverter output pulse current and the load inductor current.

[0082] Step 3: Construct extended state equations that include lumped disturbances: summarize hardware parameter drift, unmodeled line voltage drop and nonlinear load impedance changes into system lumped disturbances, and construct extended state equations that include both system state variables and lumped disturbance variables.

[0083] Step 4: Design and tune the model-assisted extended state observer: Integrate the prior information of the discrete prediction model described in Step 2 to construct a model-assisted extended state observer in the discrete domain. Iteratively update the system state estimate and lumped disturbance estimate through state error feedback and disturbance error feedback. The error feedback gain coefficient of the observer is tuned based on the Jury stability criterion.

[0084] Step 5: Feedforward compensation and calculation of the deadbeat-free optimal command: The lumped disturbance estimate output by the model-assisted extended state observer is fedforward to the deadbeat-free predictive control equation, and the optimal control command in the least squares sense is solved using the Moore-Penrose generalized inverse matrix based on the input matrix of the discrete predictive model.

[0085] To verify the correctness of the theory, a simulation model was built in the Matlab platform for experimentation.

[0086] A. High steady-state tracking accuracy: Figure 4 The results obtained using the method proposed in this invention in CSI steady-state simulations are presented. Figure 4 As can be seen from (b), under the method proposed in this invention, the steady-state current THD of the three-phase CSI is only 1.56%, and it does not include high-order oscillation frequencies, thus realizing high-quality power supply with zero steady-state error under complex operating conditions.

[0087] B. Extremely fast dynamic response: Figure 5 The transient simulation waveforms obtained using the method proposed in this invention are shown when the DC side current reference is increased from 14A to 18A. Figure 5 Figure (a) shows that after a step change in the DC-side current reference, the system re-enters steady state within half of the fundamental cycle, exhibiting a short transient recovery time. This demonstrates the excellent dynamic performance of the proposed method. Furthermore, the extremely weak dq-axis coupling effect of the grid connection current verifies the effective widening of the control bandwidth.

[0088] C. Extremely strong parameter robustness: To further verify the advantages of the proposed strategy, it is tested under severe hardware aging, magnetic saturation, or temperature drift (e.g., , , When the parameters were mismatched, the isd and isq ripples of FO-CSI-DPCC and MAESO-CSI-DPCC were quantitatively compared. As shown in Table 1, using the MAESO-CSI-DPCC proposed in this invention, under the conditions of R=150%R0, L=150%L0, and C=70%C0, the d-axis and q-axis current ripples under the control of the method of this invention are significantly reduced compared with the traditional FO-CSI-DPCC, eliminating the high-frequency instability risk caused by the deviation of the prediction reference. The above results show that the MAESO-CSI-DPCC strategy has good robustness.

[0089] Table 1 Current Ripple under Parameter Imbalance

[0090] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A deadbeat predictive control method for a current source inverter with an extended state observer, applied to a current source inverter in off-grid operation mode of a microgrid, characterized in that, Includes the following steps: Step 1, State Sampling and Coordinate Transformation: The current of the filter inductor and the voltage of the filter capacitor output by the inverter are sampled in real time, and the Park transformation is used to transform them from the stationary abc coordinate system to the synchronous rotating dq coordinate system to obtain the state measurement value at the current moment, and the AC bus voltage of the external resistive load is regarded as an external disturbance. Step 2: Establish a high-frequency accurate discretized prediction model based on second-order Taylor expansion: Using the Taylor expansion that retains the second-order terms, the continuous state-space equation of the current source inverter in the dq coordinate system is discretized to establish a discrete prediction model. This makes non-zero coupling terms appear in the input matrix of the discrete prediction model, thereby establishing the mathematical coupling relationship between the inverter output pulse current and the load inductor current. Step 3: Construct extended state equations that include lumped disturbances: summarize hardware parameter drift, unmodeled line voltage drop and nonlinear load impedance changes into system lumped disturbances, and construct extended state equations that include both system state variables and lumped disturbance variables. Step 4: Design and tune the model-assisted extended state observer: Integrate the prior information of the discrete prediction model described in Step 2 to construct a model-assisted extended state observer in the discrete domain. Iteratively update the system state estimate and lumped disturbance estimate through state error feedback and disturbance error feedback. The error feedback gain coefficient of the observer is tuned based on the Jury stability criterion. Step 5: Feedforward compensation and calculation of the deadbeat-free optimal command: The lumped disturbance estimate output by the model-assisted extended state observer is fedforward to the deadbeat-free predictive control equation, and the optimal control command in the least squares sense is solved using the Moore-Penrose generalized inverse matrix based on the input matrix of the discrete predictive model.

2. The method according to claim 1, characterized in that, In step two, the state variables of the continuous state-space equation are the d-axis component of the filter inductor current, the q-axis component of the filter inductor current, the d-axis component of the filter capacitor voltage, and the q-axis component of the filter capacitor voltage; the control input variables are the d-axis and q-axis components of the inverter bridge output pulse width modulation current; the external disturbance variables are the d-axis and q-axis components of the AC bus voltage; the system matrix and the input matrix are determined by the nominal parameters of the filter inductor and the filter capacitor, and the system matrix does not include the load resistance parameter.

3. The method according to claim 2, characterized in that, In step two, the discrete prediction model is characterized by a state transition matrix and an input matrix. The state transition matrix consists of the product of the identity matrix, the system matrix and the sampling period, and half of the product of the square of the system matrix and the square of the sampling period. The input matrix is ​​formed by multiplying the sum of the product of the identity matrix and the sampling period, and half of the product of the system matrix and the square of the sampling period, by left-multiplying by the input matrix of the continuous state space equation.

4. The method according to claim 3, characterized in that, In step three, the extended state equation treats the system's lumped disturbance as a new state variable, and the value of the state variable at the next time step is obtained by summing the value at the current time step with the rate of change of the disturbance. The value of the system state at the previous moment is obtained by adding the product of the state transition matrix and the current state, the product of the input matrix and the control input, and the product of the disturbance matrix and the current lumped disturbance.

5. The method according to claim 4, characterized in that, In step four, the iterative update rule of the model-assisted extended state observer is as follows: the state estimate at the next time step is obtained by adding the product of the state transition matrix and the current state estimate, the product of the input matrix and the current control input, the product of the disturbance matrix and the current lumped disturbance estimate, and the state error feedback quantity; the lumped disturbance estimate at the next time step is obtained by adding the current lumped disturbance estimate and the disturbance error feedback quantity; the state error is the difference between the current state measurement value and the current state estimate; the state error feedback quantity is the product of the first diagonal gain matrix and the state error; the disturbance error feedback quantity is... The product of the second diagonal gain matrix and the state error; the diagonal elements of the first diagonal gain matrix are, in order, the first gain coefficient, the second gain coefficient, the first gain coefficient, and the second gain coefficient; the diagonal elements of the second diagonal gain matrix are, in order, the third gain coefficient, the fourth gain coefficient, the third gain coefficient, and the fourth gain coefficient; the first gain coefficient is used to correct the state observation error of the inductor current, the second gain coefficient is used to correct the state observation error of the capacitor voltage, the third gain coefficient is used to track and compensate for the lumped disturbance acting on the current loop, and the fourth gain coefficient is used to track and compensate for the lumped disturbance acting on the voltage loop.

6. The method according to claim 5, characterized in that, The tuning process for the first to fourth gain coefficients includes: deriving the closed-loop characteristic polynomial of the observer error system, wherein each coefficient of the polynomial is determined by the values ​​of the filter inductor, filter capacitor, sampling period, and the first to fourth gain coefficients; constructing a Jury array of the characteristic polynomial according to the Jury stability criterion, and establishing a set of constraint inequalities that ensure all eigenvalues ​​lie within the unit circle of the Z-plane; substituting the nominal parameter values ​​of the filter inductor and filter capacitor and the control frequency into the set of constraint inequalities to solve for the range of gain coefficient values ​​that enable the observer to converge stably.

7. The method according to claim 6, characterized in that, In step five, the specific process of feedforward compensation and optimal command calculation is as follows: Substitute the state estimate and lumped disturbance estimate of the observer at the next time step into the deadbeat predictive control equation, and set the target state to obtain an equation in which the control command at the next time step is an unknown; use the Moore-Penrose generalized inverse matrix of the input matrix to find the least squares solution of the equation and solve for the optimal control command acting at the next time step; extract the d-axis and q-axis components from the optimal control command and use them as the current reference command for the current source inverter used to generate the SVPWM drive signal.

8. A deadbeat predictive control system for a current source inverter with an extended state observer, characterized in that, include: Voltage sensors and current sensors are used to acquire capacitor voltage and inductor current signals in real time; A digital signal processor for performing the method as described in any one of claims 1 to 7 to generate an SVPWM drive signal; The drive circuit and the three-phase two-level current source inverter are used to convert DC current into desired AC current according to the SVPWM drive signal to power off-grid loads.