A model-free predictive control method and device for LC-filtered voltage source inverters

Abstract

This invention discloses a model-free predictive control method and device for an LC-filtered voltage source inverter (VSI). The method includes establishing a mathematical model of the VSI in an αβ stationary coordinate system, obtaining its eight basic vectors; reconstructing the state-space equations of the LC-filtered VSI to achieve real-time updates of voltage and current gradients; predicting future voltage and current based on the updated voltage and current gradients and sampled voltage and current, and calculating voltage and current references; applying the calculated algebraic value functions of the predicted voltage and current, selecting weighting factors to eliminate coupling effects caused by the LC filter; and substituting the predicted voltage and current corresponding to the eight vectors into the value functions for evaluation, selecting the vector with the smallest value function value as the optimal vector for application in the next control cycle. This invention uses measured voltage and current dual gradients to eliminate the dependence of predictive control on system parameters, improving parameter robustness.

Description

Technical Field

[0001] This invention relates to the field of power grid technology, specifically to a model-free predictive control method and device for an LC filter-type voltage source inverter. Background Technology

[0002] In recent years, AC microgrids have played a crucial role in distributed energy systems, configurable in grid-connected or islanded modes. Voltage source inverters (VSIs) are key devices between AC microgrids and independent loads or the grid. When an AC microgrid operates in islanded mode, the VSI requires a high-quality AC voltage from the load to serve as an ideal AC voltage source. Compared to L or LCL-filtered VSIs, LC-filtered VSIs can suppress high-frequency harmonics while generating a sinusoidal output voltage, making them suitable for independently operating systems. For LC-filtered VSIs, this study aims to explore an effective control scheme to achieve low voltage ripple, low total harmonic distortion (THD), and fast dynamic performance under various load conditions.

[0003] To control the output voltage, traditional methods typically employ multi-loop linear control. Although this control method is widely used in practice, its cascaded control loops result in limited dynamic performance and difficulty in parameter adjustment. Compared to linear control methods, Model Predictive Control (MPC) is an effective alternative due to its fast dynamic response and flexible multi-objective handling capabilities. The literature [P. Cortes, G. Ortiz, JIYuz, et al. Model Predictive Control of an Inverter With Output LC Filter for UPS Applications. IEEE Transactions on Industrial Electronics, 2009, 56(6): 1875-1883.] first proposed an MPC method for LC-filtered VSI, which calculates the predicted voltage value at future times using a discrete system model and selects the optimal vector using a value function. Due to the coupling effect between the inductor current and capacitor voltage of the LC filter, it is difficult to obtain satisfactory voltage performance by using a single voltage as the control objective. The literature [T. [Model Predictive Control of Power Converters for Robust and Fast Operation of AC Microgrids. IEEE Transactions on Power Electronics, 2018, 33(7):6304-6317.] By incorporating inductor current tracking evaluation into the value function, a value function evaluation with inductor current and capacitor voltage as dual control objectives is achieved, improving voltage quality. However, the output voltage performance based on MPC depends on an accurate mathematical model, and modeling uncertainties and unpredictable changes in system parameters can affect its steady-state performance.

[0004] To eliminate the influence of model parameters on predictive control, researchers have studied model-free predictive control (MFPC) methods based on lookup tables (LUTs). This method is simple in principle and easy to implement; it achieves robust current prediction by storing and updating the current gradients under each vector action in a control cycle. However, this method requires updating the current gradients under unapplied vector actions; otherwise, it will lead to stagnation in gradient updates, increasing prediction error. To address this, the literature [Lin Cheng-kai, Yu Jen-te, Lai Yen-shin, et al. Improved Model-Free Predictive Current Control for Synchronous Reluctance Motor Drives. IEEE Transactions on Industrial Electronics, 2016, 63(6):3942-3953.] sets the current gradient update frequency, meaning that if a current gradient is not updated within 50 control cycles, the control system will force the use of its corresponding vector in the next control cycle, thus achieving the current gradient update; the literature [PG Carlet, F. Tinazzi, S. Bolognani, et al. An Effective Model-Free Predictive Current Control for Synchronous Reluctance Motor Drives, IEEE Transactions on Industry Applications, 2019, 55(4):3781-3790.] uses the current gradient sampled in the past three control cycles to update the remaining current gradient; the literature [Ma Chenwei, Li Huayu, Yao Xuliang, et al. An Improved Model-Free Predictive Current Control With Advanced Current] updates the current gradients of the remaining current gradients. [Gradient Updating Mechanism, IEEE Transactions on Industrial Electronics, 2021, 68(12): 11968-11979.] established the current gradient relationship between two control cycles and used this relationship to update the remaining current gradient; [Yu Feng, Zhou Chenhui, Liu Xing, et al.][Model-Free Predictive Current Control for Three-Level Inverter-Fed IPMSM With an Improved Current Difference Updating Technique. IEEE Transactions on Energy Conversion, 2021, 36(4): 3334-3343.] updates the residual gradient by applying the magnitude relationship of vectors; [CAAgustin, Yu Jen-te, Cheng Yu-shan, et al. Model-Free Predictive Current Control for SynRM Drives Based on Optimized Modulation of Triple-Voltage-Vector, IEEE Access, 2021, 9: 130472-130483.] sets three sampling points in one control cycle, improving the update frequency of the current gradient through multi-sampling. Currently, the method for updating the current gradient still needs to be studied, and the above-mentioned LUT-based MFPC methods are all proposed for L-filtered VSI systems. How to design the LUT and how to update the gradient under LC-filtered MFPC has not yet been discussed. Summary of the Invention

[0005] The present invention proposes a model-free predictive control method for an LC filter-type voltage source inverter, which can at least solve one of the above-mentioned technical problems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A model-free predictive control method for an LC-filtered voltage source inverter includes the following steps:

[0008] Step 1: Establish a mathematical model of the LC filter-type voltage source inverter in the αβ stationary coordinate system. Based on the state of its three-phase switching transistors, obtain eight basic voltage vectors, namely the first basic voltage vector v0(0,0,0), the second basic voltage vector v1(1,0,0), the third basic voltage vector v2(1,1,0), the fourth basic voltage vector v3(0,1,0), the fifth basic voltage vector v4(0,1,1), the sixth basic voltage vector v5(0,0,1), the seventh basic voltage vector v6(1,0,1), and the eighth basic voltage vector v7(1,1,1).

[0009] Step 2: Reconstruct the state-space equation of the LC-filtered VSI to achieve real-time updates of voltage and current gradients.

[0010] Step 3: Based on the updated voltage and current gradients from Step 2 and the sampled voltage and current, predict the voltage and current at future times and calculate the voltage and current references.

[0011] Step 4: Substitute the predicted voltage and predicted current calculated in Step 3 into the value function, and select a weighting factor to eliminate the coupling effect caused by LC filtering; substitute the predicted voltage and predicted current corresponding to the eight basic voltage vectors into the value function for evaluation, and select the vector with the smallest value function value as the optimal vector to be applied in the next control cycle.

[0012] Furthermore, step one is described in detail as follows:

[0013] The sampled three-phase capacitor voltage (V) ca v cb v cc ), inverter-side inductor current (i ia i ib i ic ) and three-phase load current (i oa i ob i oc Perform a Clark coordinate transformation to obtain the capacitor voltage (v) in the stationary coordinate system. cα v cβ Inverter-side inductor current (i) in stationary coordinate system iα i iβ ) and load current (i) in stationary coordinate system oα i oβ ):

[0014]

[0015]

[0016]

[0017] The mathematical model of the LC-filtered VSI in the stationary coordinate system is as follows:

[0018]

[0019] Among them, L i C is the inverter-side filter inductor; C is the filter capacitor; v iα v iβ This is the fundamental voltage vector in the stationary coordinate system.

[0020] According to the zero-order hold method, the predicted capacitor voltage and the predicted inverter-side inductor current at time k+1 are:

[0021]

[0022] Where, Φ 11 Φ is the first coefficient in the first row of the first coefficient matrix. 12 Φ is the second coefficient in the first row of the first coefficient matrix. 21 Φ is the first coefficient in the second row of the first coefficient matrix. 22 Γ is the second coefficient in the second row of the first coefficient matrix. 11 Γ is the first coefficient in the first row of the second coefficient matrix. 12 Γ is the second coefficient in the first row of the second coefficient matrix. 21 Γ is the first coefficient in the second row of the second coefficient matrix. 22 For the second coefficient in the second row of the second coefficient matrix, calculate and represent the first coefficient matrix Φ and the second coefficient matrix Γ as follows:

[0023]

[0024] T is the resonant frequency of the LC filter. s To control the cycle.

[0025] To eliminate the influence of model parameters on the predicted current, the sampled capacitor voltage and inductor current can be considered linear in each control cycle. Therefore, the sampled capacitor voltage gradient and inductor current gradient at time k can be expressed as:

[0026]

[0027] Where, Δv cα (k–1), Δv cβ (k–1) represents the capacitor voltage gradient in the stationary coordinate system; Δi iα (k–1), Δi iβ (k–1) represents the inductor current gradient in the stationary coordinate system, which is generated by the fundamental voltage vector acting at time (k–1). Simultaneously, the sampled voltage and current gradients need to correspond to the applied vectors and be stored in the LUT for voltage and current prediction.

[0028] However, voltage and current sampling alone can only update the gradient values ​​under the applied vector in the previous control cycle. For vectors that are not applied, their corresponding gradient values ​​cannot be updated. This phenomenon is called stagnation, which will increase the prediction error and reduce the output voltage performance.

[0029] Furthermore, step two is detailed as follows:

[0030] To eliminate the stagnation phenomenon, equation (5) can be reconstructed as follows:

[0031]

[0032] Because in equation (8): v cα (k)–v cα (k–1)=Δv cα (k–1), v cβ (k)–v cβ (k–1)=Δv cβ (k–1) and i iα (k)–i iα (k–1)=Δi iα (k–1), i iβ (k)–i iβ (k–1)=Δi iβ (k–1); when the basic voltage vector v in equation (8) iα (k–1), v iβ (k–1) becomes the residual vector v jα (k–1), v jβ At (k–1), the voltage gradient Δv corresponding to its residual vector cjα (k–1), Δv cjβ (k–1) and the current gradient Δi corresponding to the residual vector ijα (k–1), Δi ijβ (k–1) is represented as:

[0033]

[0034] Where j∈{0,1,…7} and j≠i. By subtracting formula (8) from formula (9), we can obtain:

[0035]

[0036] To eliminate parameter Γ 21 and Γ 11 The voltage-current gradient relationship at time (k–2) can be obtained through a one-step recursion and expressed as:

[0037]

[0038] Dividing equations (10) and (11) yields the remaining application vector v. jα (k–1), v jβ The capacitor voltage gradient (Δv) under the action of (k–1) cjα (k–1) and Δv cjβ (k–1) and inductor current gradient (Δi) ijα (k–1) and Δi ijβ (k–1) is represented as:

[0039]

[0040] According to equation (12), the calculation of the current gradient is affected by the value of the denominator in the formula. When the denominator is not zero, it can be further simplified to:

[0041]

[0042] When the denominator is 0, according to formula (10), when the coordinate components of a vector are equal, their corresponding voltage and current gradients are also equal. Therefore, the updated formula (12) can be rewritten as:

[0043]

[0044] Therefore, this invention achieves real-time updates of all voltage and current gradients in each control cycle, completely eliminating stagnation.

[0045] Furthermore, step three is detailed below:

[0046] Based on the voltage and current gradients of each vector action, the predicted voltage and current at time (k+1) can be expressed as:

[0047]

[0048] To compensate for the one-step control delay, the predicted voltage and current at time (k+1) can be expressed as:

[0049]

[0050] Where, Δv cα (k+1), Δv cβ (k+1) and Δi iα (k+1), Δi iβ (k+1) represents the voltage and current gradients under the action of the eight basic voltage vectors;

[0051] Furthermore, voltage reference and current reference can be expressed as:

[0052]

[0053] Among them, v c ref As a reference for capacitor voltage, i i ref V serves as the reference for the inverter-side inductor current. ref ω is the reference amplitude of the capacitor voltage. ref This is the reference angular frequency of the capacitor voltage.

[0054] Furthermore, step four is detailed below:

[0055] To obtain the optimal vector and apply it in the next control cycle, the voltage and current gradients corresponding to the eight basic voltage vectors need to be substituted into equation (18), and the resulting eight voltage and current predictions are substituted into the bi-objective value function for evaluation. The value function can be expressed as:

[0056] g d =g v +λg i (37)

[0057] Among them, g d For a dual-objective value function, g v Let g be the capacitance voltage value function. i λ is the value function of the inductor current on the inverter side, and λ is the weighting factor.

[0058]

[0059] Among them, v cα ref v cβ ref For the capacitor voltage reference in the stationary coordinate system, i iα ref i iβ ref This is the inductor current reference in the stationary coordinate system.

[0060] Finally, the vector that minimizes the value function is selected as the optimal vector and applied in the next control cycle.

[0061] On the other hand, the present invention also discloses a computer-readable storage device storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method described above.

[0062] As can be seen from the above technical solutions, the model-free predictive control method for LC-filtered voltage source inverters of this invention addresses the issue that, when using traditional model predictive control methods for LC-filtered voltage source inverters, the output voltage performance depends on an accurate mathematical model. When the controller parameters do not match the actual parameters, the output voltage prediction error increases significantly. Therefore, this invention proposes a model-free predictive control method for LC-filtered voltage source inverters. It establishes a Level Under Test (LUT) based on voltage and current dual gradients. By calculating and storing the voltage and current gradients under the applied vector action at the previous moment, and combining this with the sampled values ​​at the current moment, it achieves model-free prediction of voltage and current at future moments, improving the robustness of predictive control. It reconstructs the state-space prediction equations, updating the gradient values ​​of unapplied vectors in real time based on the voltage and current gradients under the applied vector action. It uses a dual-objective value function of voltage and current for tracking and evaluation, eliminating the coupling effect of state variables in the second-order LC filter, and thus selecting the optimal vector for the next control cycle.

[0063] In summary, the model-free predictive control method for LC filter voltage source inverters proposed in this invention employs measured voltage and current dual gradients to eliminate the dependence of predictive control on system parameters, thereby improving parameter robustness. Based on the reconstructed state-space equations for different voltage vectors, all voltage and current gradient data are updated in each control cycle, eliminating the adverse effects of gradient update stagnation. The control method employed in this invention effectively solves the problem of strong dependence on model parameters in traditional model predictive control methods, and is particularly suitable for islanded microgrid systems with frequently changing line impedances.

[0064] Specifically, this invention employs a model-free predictive control method with dual voltage and current gradients, which can effectively improve the robustness of predictive control to changes in model parameters. By reconstructing the state-space equations, an improved voltage and current gradient update method is proposed, achieving real-time updates of the voltage and current gradients and effectively avoiding voltage harmonics caused by gradient update stagnation. Attached Figure Description

[0065] Figure 1 This is a topology diagram of an LC-filtered voltage source inverter according to an embodiment of the present invention;

[0066] Figure 2 This is a vector diagram illustrating an embodiment of the present invention;

[0067] Figure 3 This is a voltage-current LUT diagram according to an embodiment of the present invention;

[0068] Figure 4 This is a schematic diagram of vector relationships according to an embodiment of the present invention;

[0069] Figure 5 This is a schematic diagram illustrating the real-time update of the current gradient according to an embodiment of the present invention;

[0070] Figure 6 This is a schematic diagram of the control flow for implementing the present invention;

[0071] Figure 7 This is a diagram of the experimental platform according to an embodiment of the present invention;

[0072] Figure 8a and Figure 8b This is a comparison chart of voltage performance under different update methods; among them... Figure 8a The update method is shown in formula (7). Figure 8b This is a schematic diagram of the method proposed in this invention;

[0073] Figure 9a and Figure 9b This is a comparison chart of steady-state performance under linear load; where, Figure 9a This is a traditional MPC method, and 9b is a schematic diagram of the method proposed in the embodiment of the present invention;

[0074] Figure 10a and Figure 10b This is a comparison chart of steady-state performance under nonlinear loads; among them, Figure 10a For traditional MPC methods, Figure 10b This is a schematic diagram of the method proposed in an embodiment of the present invention;

[0075] Figure 11a , Figure 11b , Figure 11c , Figure 11d , Figure 11e and Figure 11f This is a comparison chart of voltage performance under parameter mismatch; among which, Figure 11a For the traditional MPC method in 0.5L i , Figure 11b The method proposed in the embodiments of the present invention is in 0.5L i ; Figure 11c For the traditional MPC method at 0.5C, Figure 11d The method proposed in the embodiments of the present invention is carried out at 0.5°C; Figure 11e For the traditional MPC method in 0.5L i and 0.5C, Figure 11f The method proposed in the embodiments of the present invention is in 0.5L i And 0.5C. Detailed Implementation

[0076] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0077] To address the shortcomings of the aforementioned background technology, this invention proposes a model-free predictive control method for LC filter-type voltage source inverters, which solves the problem of strong dependence on model parameters in traditional MPC.

[0078] like Figure 6 As shown, this invention provides a model-free predictive control method for an LC-filtered voltage source inverter. This invention eliminates the influence of parameters on predictive control by using measured voltage and current gradients, thus improving the parameter robustness of predictive control. By reconstructing the state-space equations, it achieves real-time updates of all voltage and current gradients in each control cycle, thereby improving output voltage performance. The specific steps are as follows:

[0079] Step 1: Establish a mathematical model of the LC filter-type voltage source inverter in the αβ stationary coordinate system. Based on the state of its three-phase switching transistors, obtain its eight basic voltage vectors, namely the first basic voltage vector v0(0,0,0), the second basic voltage vector v1(1,0,0), the third basic voltage vector v2(1,1,0), the fourth basic voltage vector v3(0,1,0), the fifth basic voltage vector v4(0,1,1), the sixth basic voltage vector v5(0,0,1), the sixth basic voltage vector v6(1,0,1), and the seventh basic voltage vector v7(1,1,1). Use voltage and current gradients to replace the traditional prediction model and save them to the designed LUT.

[0080] The sampled three-phase capacitor voltage (V) ca v cb v cc ), three-phase inverter side inductor current (i ia i ib i ic ) and three-phase load current (i oa i ob i oc Perform a Clark coordinate transformation to obtain the capacitor voltage (v) in the stationary coordinate system. cα v cβ Inverter-side inductor current i in stationary coordinate system iαβ and the load current (i) in the stationary coordinate system oα i oβ ):

[0081]

[0082]

[0083]

[0084] like Figure 1 As shown, the mathematical model of the LC-filtered VSI in the stationary coordinate system is:

[0085]

[0086] Among them, L i C is the inverter-side filter inductor; C is the filter capacitor; v iα v iβ This is the fundamental voltage vector in the stationary coordinate system.

[0087] According to the zero-order hold method, the predicted capacitor voltage and the predicted inverter-side inductor current at time k+1 are:

[0088]

[0089] Where, Φ 11 Φ is the first coefficient in the first row of the first coefficient matrix. 12 Φ is the second coefficient in the first row of the first coefficient matrix. 21 Φ is the first coefficient in the second row of the first coefficient matrix. 22 Γ is the second coefficient in the second row of the first coefficient matrix. 11 Γ is the first coefficient in the first row of the second coefficient matrix. 12 Γ is the second coefficient in the first row of the second coefficient matrix. 21 Γ is the first coefficient in the second row of the second coefficient matrix. 22 For the second coefficient in the second row of the second coefficient matrix, calculate and represent the first coefficient matrix Φ and the second coefficient matrix Γ as follows:

[0090]

[0091] T is the resonant frequency of the LC filter. s To control the cycle.

[0092] To eliminate the influence of model parameters on the predicted current, the sampled capacitor voltage and inductor current can be considered linear in each control cycle. Therefore, the sampled capacitor voltage gradient and inductor current gradient at time k can be expressed as:

[0093]

[0094] Where, Δv cα (k–1), Δv cβ (k–1) represents the capacitor voltage gradient in the stationary coordinate system; Δi iα (k–1), Δi iβ (k–1) represents the inductor current gradient in the stationary coordinate system, which is generated by the fundamental voltage vector acting at time (k–1). Simultaneously, the sampled voltage and current gradients need to correspond to the applied vectors and be stored... Figure 3 The LUT shown is used for voltage and current prediction.

[0095] However, voltage and current sampling alone can only update the gradient values ​​under the applied vector in the previous control cycle. For vectors that are not applied, their corresponding gradient values ​​cannot be updated. This phenomenon is called stagnation, which will increase the prediction error and reduce the output voltage performance.

[0096] Step 2: Reconstruct the state-space equation of the LC-filtered VSI to achieve real-time updates of voltage and current gradients.

[0097] To eliminate the stagnation phenomenon, equation (5) can be reconstructed as follows:

[0098]

[0099] Because in equation (8): v cα (k)–v cα (k–1)=Δv cα (k–1), v cβ (k)–v cβ (k–1)=Δv cβ (k–1) and i iα (k)–i iα (k–1)=Δi iα (k–1), i iβ (k)–i iβ (k–1)=Δi iβ (k–1); when the basic voltage vector v in equation (8) iα (k–1), v iβ (k–1) becomes the residual vector v jα (k–1), v jβ At (k–1), the voltage gradient Δv corresponding to its residual vector cjα (k–1), Δv cjβ (k–1) and the current gradient Δi corresponding to the residual vector ijα (k–1), Δi ijβ (k–1) is represented as:

[0100]

[0101] Where j∈{0,1,…7} and j≠i. By subtracting formula (8) from formula (9), we can obtain:

[0102]

[0103] To eliminate parameter Γ 21 and Γ 11 The voltage-current gradient relationship at time (k–2) can be obtained through a one-step recursion and expressed as:

[0104]

[0105] Dividing equations (10) and (11) yields the remaining application vector v. jα (k–1), v jβ The capacitor voltage gradient value (Δv) under the action of (k–1) cjα (k–1) and Δv cjβ (k–1) and inductor current gradient (Δi) ijα (k–1) and Δi ijβ (k–1) is represented as:

[0106]

[0107] According to equation (12), the calculation of the current gradient is affected by the value of the denominator in the formula. When the denominator is not zero, it can be further simplified to:

[0108]

[0109] When the denominator is 0, according to formula (10), when the coordinate components of a vector are equal, their corresponding voltage and current gradients are also equal. Figure 4 As shown, the update formula (12) can be rewritten as:

[0110]

[0111] Therefore, this invention achieves real-time updates of all voltage and current gradients in each control cycle, completely eliminating stagnation. Its control block diagram is as follows: Figure 5 As shown.

[0112] Step 3: Based on the updated voltage and current gradients from Step 2 and the sampled voltage and current, predict the voltage and current at future times and calculate the voltage and current references.

[0113] Based on the voltage and current gradients of each vector action, the predicted voltage and current at time (k+1) can be expressed as:

[0114]

[0115] To compensate for the one-step control delay, the predicted voltage and current at time (k+1) can be expressed as:

[0116]

[0117] Where, Δv cα (k+1), Δv cβ (k+1) and Δi iα (k+1), Δi iβ (k+1) represents the voltage and current gradients under the action of the eight basic voltage vectors.

[0118] Furthermore, voltage reference and current reference can be expressed as:

[0119]

[0120] Among them, v c ref As a reference for capacitor voltage, i i ref V serves as the reference for the inverter-side inductor current. ref ω is the reference amplitude of the capacitor voltage. refThis is the reference angular frequency of the capacitor voltage.

[0121] Step 4: Substitute the predicted voltage and predicted current calculated in Step 3 into the value function, and select a weighting factor to eliminate the coupling effect caused by LC filtering; substitute the predicted voltage and predicted current corresponding to the eight basic voltage vectors into the value function for evaluation, and select the vector with the smallest value function value as the optimal vector to be applied in the next control cycle.

[0122] To obtain the optimal vector and apply it in the next control cycle, the voltage and current gradients corresponding to the eight candidate vectors need to be substituted into equation (18), and the resulting eight voltage and current predictions need to be substituted into the bi-objective value function for evaluation. The value function can be expressed as:

[0123] g d =g v +λg i (18)

[0124] Among them, g d For a dual-objective value function, g v Let g be the capacitance voltage value function. i λ is the value function of the inductor current on the inverter side, and λ is the weighting factor.

[0125]

[0126] Among them, v cα ref v cβ ref For the capacitor voltage reference in the stationary coordinate system, i iα ref i iβ ref This is the inductor current reference in the stationary coordinate system.

[0127] Finally, the vector that minimizes the value function is selected as the optimal vector and applied to the next control cycle. The control block diagram of this invention is as follows: Figure 6 As shown.

[0128] Specific experiment:

[0129] Experimental platform such as Figure 7 As shown in Table 1, the experimental parameters are as follows.

[0130] Table 1. LC-filtered VSI system and control parameters

[0131]

[0132] First, in order to demonstrate the effectiveness of the voltage and current gradient update method proposed in the embodiments of the present invention, experimental tests were conducted on its voltage performance under different update methods.

[0133] like Figure 8a As shown, the update method shown in formula (7) only updates the voltage and current gradients under the applied vector, while the gradient values ​​of the remaining vectors remain the same, exhibiting a significant stagnation phenomenon, leading to spikes in voltage ripple and prediction errors. Figure 8b As shown, when the update method proposed in this invention is implemented, the voltage ripple and prediction error are effectively reduced because the stagnation phenomenon is completely eliminated.

[0134] also, Figure 9a and Figure 9b Experimental comparisons of the traditional MPC method and the method proposed in this invention are presented under linear load and accurate parameters. Figure 9a As shown, when implementing the MPC method, optimal voltage tracking can be achieved, with the lowest THD and steady-state error. Figure 9b As shown, when the method proposed in this invention is implemented, the output voltage THD increases from 3.08% to 3.16%. This is because the voltage performance of the method proposed in this invention depends on the accuracy of voltage and current sampling. However, during the experiment, due to the presence of sampling noise, the output voltage THD and voltage prediction error of the method proposed in this invention are slightly higher than those of the traditional MPC method.

[0135] Figure 10a and Figure 10b Experimental comparisons of the traditional MPC method and the method proposed in this invention are presented under nonlinear loads and with accurate parameters. Both methods can accurately track the reference voltage without significant voltage distortion.

[0136] Finally, to demonstrate the robustness of the parameters of this invention, such as Figure 11a , Figure 11b , Figure 11c , Figure 11d , Figure 11e and Figure 11f Voltage ripple and prediction error were compared between the traditional MPC method and the method proposed in this invention. Figure 11a and 11b It can be seen that when the control parameters of the inductor change from 3.4mH to 1.7mH (i.e., 0.5L), i When the inductance parameter is not affected, the RMSE of the traditional MPC method increases from 3.47V to 6.84V, while the method proposed in this invention maintains an RMSE of 3.66V, which is 3.37V lower than that of the traditional MPC method. Figure 11c and 11dIt can be seen that when the capacitor control parameter changes from 25μF to 12.5μF (i.e., 0.5C), the RMSE of the traditional MPC method increases from 3.47V to 8.04V; however, due to the presence of the capacitor parameter in the value function evaluation formula, the RMSE of the method proposed in this invention increases from 3.66V to 4.45V, which is 3.59V lower than that of the traditional MPC method. Figure 11e and 11f As shown, when the control parameters of the inductor and capacitor are simultaneously reduced by 50% (i.e., 0.5L) i At 0.5°C, the RMSE of the method proposed in this invention is 4.12°C lower than that of the traditional MPC method.

[0137] In summary, this invention addresses the problem of the strong dependence on parameter accuracy in traditional MPC methods by proposing a model-free predictive control method for LC-filtered voltage source inverters. This invention utilizes voltage and current gradients for prediction calculations, exhibiting good parameter robustness. It significantly reduces inverter output voltage errors when parameters change, improving voltage quality, and achieving voltage performance comparable to traditional MPC methods under accurate parameter conditions. Furthermore, this invention proposes an advanced voltage and current gradient update method that updates voltage and current gradients under all vector actions in real time, eliminating stagnation and its associated prediction spikes, further improving output voltage quality.

[0138] In another aspect, the present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of any of the methods described above.

[0139] In another aspect, the present invention also discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of any of the methods described above.

[0140] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform the steps of any of the methods described in the above embodiments.

[0141] It is understood that the system provided in the embodiments of the present invention corresponds to the method provided in the embodiments of the present invention, and the explanation, examples and beneficial effects of the relevant content can be referred to the corresponding parts of the above methods.

[0142] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0143] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0144] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A model-free predictive control method for an LC-filtered voltage source inverter, characterized in that, Includes the following steps, Step 1: Establish a mathematical model of the LC filter-type voltage source inverter in the stationary coordinate system. Based on the state of its three-phase switching transistors, obtain its eight basic voltage vectors, which are the first basic voltage vectors. The second fundamental voltage vector The third fundamental voltage vector The fourth fundamental voltage vector The fifth fundamental voltage vector The sixth fundamental voltage vector The seventh fundamental voltage vector The eighth basic voltage vector ; Step 2: Reconstruct the state-space equation of the LC-filtered VSI to achieve real-time updates of the capacitor voltage and inductor current gradients; Step 3: Based on the updated capacitor voltage and inductor current gradients from Step 2, as well as the sampled capacitor voltage and inverter-side inductor current, predict the capacitor voltage and inductor current at future moments, and calculate the reference capacitor voltage and inverter-side inductor current. Step 4: Substitute the predicted capacitor voltage and predicted inductor current calculated in Step 3 into the value function, and select a weighting factor to eliminate the coupling effect caused by LC filtering; The predicted capacitor voltage and predicted inductor current corresponding to the eight basic voltage vectors are substituted into the value function for evaluation, and the vector with the smallest value function value is selected as the optimal vector for application in the next control cycle.

2. The model-free predictive control method for an LC filter-type voltage source inverter according to claim 1, characterized in that: Step one specifically includes, The sampled three-phase capacitor voltage v ca v cb v cc Three-phase inverter side inductor current i ia i ib i ic and three-phase load current i oa i ob i oc The capacitor voltage v in the stationary coordinate system is obtained by performing a Clark coordinate transformation. cα v cβ Inverter-side inductor current i in stationary coordinate system iα i iβ and the load current i in the stationary coordinate system oα i oβ : (1) (2) (3) The mathematical model of LC-filtered VSI in the stationary coordinate system is as follows: (4) Among them, L i C is the inverter-side filter inductor; C is the filter capacitor; v iα v iβ The fundamental voltage vector in the stationary coordinate system; According to the zero-order hold method, the predicted capacitor voltage and the predicted inverter-side inductor current at time k+1 are: (5) Where, Φ 11 Φ is the first coefficient in the first row of the first coefficient matrix. 12 Φ is the second coefficient in the first row of the first coefficient matrix. 21 Φ is the first coefficient in the second row of the first coefficient matrix. 22 Γ is the second coefficient in the second row of the first coefficient matrix. 11 Γ is the first coefficient in the first row of the second coefficient matrix. 12 Γ is the second coefficient in the first row of the second coefficient matrix. 21 Γ is the first coefficient in the second row of the second coefficient matrix. 22 For the second coefficient in the second row of the second coefficient matrix, calculate and represent the first coefficient matrix Φ and the second coefficient matrix Γ as follows: (6) T is the resonant frequency of the LC filter. s To control the cycle; The capacitor voltage gradient and inductor current gradient sampled at time k are expressed as follows: (7) in, v cα (k–1) v cβ (k–1) represents the capacitor voltage gradient in the stationary coordinate system; i iα (k–1) i iβ (k–1) is the inductor current gradient in the stationary coordinate system, which is generated by the basic voltage vector acting at time k–1. At the same time, the sampled voltage and current gradients must correspond to the applied vector and be stored in the LUT for voltage and current prediction.

3. The model-free predictive control method for an LC filter-type voltage source inverter according to claim 2, characterized in that: Step two specifically includes, Equation (5) is reconstructed as follows: (8) Because in equation (8): v cα (k) – v cα (k–1)= v cα (k–1), v cβ (k) – v cβ (k–1)= v cβ (k–1) and i iα (k) –i iα (k–1)= i iα (k–1), i iβ (k) – i iβ (k–1)= i iβ (k–1); when the basic voltage vector v in equation (8) iα (k–1), v iβ (k–1) becomes the residual vector v jα (k–1), v jβ At (k–1), the voltage gradient corresponding to its residual vector v cjα (k–1) v cjβ (k–1) and the current gradient corresponding to the residual vector i ijα (k–1) i ijβ (k–1) is represented as: (9) Where j∈{0, 1,…7} and j≠i; by subtracting formula (8) and formula (9), we get: (10) To eliminate parameter Γ 21 and Γ 11 The voltage-current gradient relationship at time k–2 is obtained through a one-step recursion and expressed as: (11) Dividing equations (10) and (11) yields the remaining application vector v. jα (k–1), v jβ The capacitor voltage gradient under the action of (k–1) v cjα (k–1) and v cjβ (k–1) and inductor current gradient ( i ijα (k–1) and i ijβ (k–1)) and expressed as: (12) According to equation (12), the calculation of the current gradient is affected by the value of the denominator in the formula; when the denominator is not zero, it simplifies to: (13) When the denominator is 0, according to formula (10), when the coordinate components of a vector are equal, their corresponding voltage and current gradients are also equal; the updated formula (12) is re-expressed as: (14) It achieves real-time updates of all voltage and current gradients in each control cycle, completely eliminating stagnation.

4. The model-free predictive control method for an LC filter-type voltage source inverter according to claim 3, characterized in that: Step three specifically includes: Based on the voltage and current gradients of each vector action, the predicted voltage and current at time k+1 are expressed as follows: (15) To compensate for the one-step control delay, the predicted voltage and current at time k+1 are expressed as: (16) in, v cα (k+1) v cβ (k+1) and i iα (k+1) i iβ (k+1) represents the voltage and current gradients under the action of the eight basic voltage vectors; Voltage reference and current reference are represented as follows: (17) Among them, v c ref As a reference for capacitor voltage, i i ref V serves as the reference for the inverter-side inductor current. ref ω is the reference amplitude of the capacitor voltage. ref This is the reference angular frequency of the capacitor voltage.

5. The model-free predictive control method for an LC filter-type voltage source inverter according to claim 4, characterized in that: Step four is described in detail below: Substituting the voltage and current gradients corresponding to the eight basic voltage vectors into equation (18), and then substituting the resulting eight voltage and current predictions into the bi-objective value function for evaluation, the value function is expressed as: (18) Among them, g d For a dual-objective value function, g v Let g be the capacitance voltage value function. i λ is the value function of the inverter-side inductor current, and λ is the weighting factor. (19) Among them, v cα ref v cβ ref For the capacitor voltage reference in the stationary coordinate system, i iα ref i iβ ref The inductor current reference is in a stationary coordinate system; Finally, the vector that minimizes the value function is selected as the optimal vector and applied in the next control cycle.

6. A computer-readable storage device storing a computer program that, when executed by a processor, causes the processor to perform the steps of the method as claimed in any one of claims 1 to 5.

Images