An improved robust predictive control method for renewable energy grid-connected inverters

Abstract

An improved robust predictive control method for a new energy grid-connected inverter relates to the technical field of control of new energy grid-connected inverters. The present invention designs a lookup table (LUT) based on sampled system data (current, voltage) for the implementation of the robust predictive control method, reducing the dependence of the predictive control on model parameters. Combined with the system variable gradient update equation, all system variable gradients in the lookup table are updated quickly and in real time, solving the problem that the prediction process and reference value acquisition process of the existing model predictive control method are more sensitive to system parameters and the problem of stagnation of current gradient update, reducing the output current ripple, improving the system control performance, and further improving the output performance of the grid-connected inverter. The robust predictive control method adopted by the present invention can well solve the problems of the traditional model predictive control method's strong dependence on model parameters and large output current ripple, and is particularly suitable for grid-connected power generation systems with frequently changing line impedance.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy grid-connected inverter control, and in particular to an improved robustness predictive control method for a new energy grid-connected inverter. Background Art

[0002] Grid-tied inverters, due to their simple structure and high reliability, are widely used as a key interface between renewable energy systems and independent loads or the power grid. As the integration of new energy capacity continues to increase, the proportion of traditional generators in the power grid is gradually decreasing. In this situation, grid-tied inverters will operate in voltage source mode, providing voltage support and control to the grid, helping to maintain voltage stability and quality.

[0003] In order to achieve good control of the output performance of grid-connected inverters, researchers have developed and studied a variety of control methods. Among them, model predictive control (MPC) has been widely used in the current control of grid-connected inverters due to its advantages such as simple implementation, fast response, and multi-objective collaborative control. However, due to the complexity and uncertainty of the power system, new energy grid-connected inverters face problems such as parameter changes and external interference. When the model parameters do not match the system parameters, prediction errors will occur, resulting in deterioration of the control performance of the system. In order to improve the robustness of model predictive control to parameter changes, many improved model prediction methods have been proposed at home and abroad, which are mainly divided into the following two types:

[0004] (1) Online identification of system parameters to compensate for parameter errors;

[0005] (2) Estimate and compensate for time-varying interference caused by parameter uncertainty.

[0006] Both approaches can improve the robustness of model predictive control to parameter variations; however, their implementation requires the design of complex observers and the tuning of multiple observer parameters, which results in a large computational burden.

[0007] Recently, a model-free predictive control method based on a lookup table (LUT) was proposed in [C. Lin, T. Liu, J. Yu, L. Fu, and C. Hsiao, "Model-Free Predictive Current Control for Interior Permanent-Magnet Synchronous Motor Drives Based on Current Difference Detection Technique," IEEE Transactions on Industrial Electronics, vol. 61, no. 2, pp. 667-681, Feb. 2014] and successfully implemented in a permanent magnet synchronous motor drive. This control method is simple in principle and easy to implement, completely eliminating the dependence of model predictive control on the system model. In this control strategy, the sixteen current gradients caused by eight basic voltage vectors in each control cycle are stored in two lookup tables, one for the d-axis and the other for the q-axis. Therefore, the control performance of this model-free predictive control depends heavily on the accuracy of the LUT. However, in each control cycle, only the current gradients caused by the applied voltage vectors are updated; other gradients stagnate. Long-term stagnation can lead to unreliable predictions and even affect system stability.

[0008] To improve the accuracy of the lookup table, the paper [C. Lin, J. Yu, Y. Lai and H. Yu, "Improved Model-Free Predictive Current Control for Synchronous Reluctance Motor Drives," IEEE Transactions on Industrial Electronics, vol. 63, no. 6, pp. 3942-3953, June 2016] defines a minimum update frequency for the lookup table to ensure that each gradient is updated within a specified time. In other words, if a voltage vector has not been used within a predefined frequency in the past, it must be applied in the next cycle, and its corresponding current gradient can be updated. However, frequently applying non-optimal voltage vectors can negatively impact control performance and delay lookup table updates.

[0009] In the papers [D. Da Rù, M. Polato and S. Bolognani, "Model-free predictive current control for a SynRM drive based on an effective update of measured current responses," IEEE International Symposium on Predictive Control of Electrical Drives and Power Electronics (PRECEDE), Pilsen, Czech Republic, 2017.] and [PG Carlet, F. Tinazzi, S. Bolognani and M. Zigliotto, "An Effective Model-Free Predictive Current Control for Synchronous Reluctance Motor Drives," IEEE Transactions on Industry Applications, vol. 55, no. 4, pp. 3781-3790, July-Aug. 2019.], the updated current gradients used over the past three cycles are used to estimate the remaining current gradients. This strategy works only if the voltage vectors over the past three consecutive cycles are different from each other. However, the consecutive application of three different voltage vectors is a random phenomenon, which can cause the lookup table update to stall. Furthermore, this strategy considers all possible vector sequences (up to 210), which increases the computational burden.

[0010] The paper [C.Ma, H.Li, X.Yao, Z.Zhang and F.De Belie, "An Improved Model-Free Predictive Current Control With Advanced Current Gradient Updating Mechanism," IEEE Transactions on Industrial Electronics, vol. 68, no. 12, pp. 11968-11979, Dec. 2021.] analyzes the relationship between the current gradients of different voltage vectors. This relationship is then used to estimate the current gradients of all unused voltage vectors within a control cycle. However, when two consecutive base voltage vectors are identical, current gradient update stagnation still occurs. Therefore, the current gradient stagnation problem remains unresolved.

[0011] In addition, model predictive control searches for the optimal voltage vector through the value function, and parameter mismatch will lead to inaccurate reference values. Therefore, it is also necessary to improve the robustness of the predictive control method from the perspective of reference values. Summary of the Invention

[0012] For the control of the output performance of existing grid-connected inverters, if traditional model prediction is used, although the control is relatively simple, it requires accurate model parameters. When the model parameters do not match the control parameters, it is difficult to achieve good control of the output current. Therefore, some scholars have proposed a robust control method based on a lookup table, which completely eliminates the dependence of predictive control on system parameters. However, the current model-free predictive control method only outputs one voltage vector in each control cycle, resulting in large current ripple, and its current gradient update strategy cannot guarantee the real-time update of all gradients, which in turn affects the output current control of the inverter. In addition, the existing control method does not consider the impact of parameter mismatch on the accuracy of the reference value. In actual control, the reference value deviation caused by parameter error will affect the system control performance.

[0013] In view of this, the present invention proposes an improved robust predictive control method for grid-connected inverters of renewable energy. Based on sampled system data (current, voltage), a lookup table (LUT) is designed for the implementation of the robust predictive control method, reducing the predictive control's dependence on model parameters. Combined with the system variable gradient update equation, the invention achieves real-time and rapid updates of all system variable gradients in the lookup table. This solves the problem of the existing model predictive control method's prediction process and reference value acquisition process being sensitive to system parameters, as well as the problem of stagnant current gradient updates. This reduces output current ripple, improves system control performance, and further enhances the output performance of the grid-connected inverter.

[0014] In order to achieve the above object, the technical solution adopted by the present invention is:

[0015] An improved robust predictive control method for a new energy grid-connected inverter is provided, wherein the specific steps are as follows:

[0016] Step 1: Establish a mathematical model of the two-level grid-connected inverter in the αβ stationary coordinate system. According to the three-phase switch states of the two-level inverter, obtain its eight basic voltage vectors, namely u0(0,0,0), u1(1,0,0), u2(1,1,0), u3(0,1,0), u4(0,1,1), u5(0,0,1), u6(1,0,1), and u7(1,1,1).

[0017] Step 2: Calculate the voltage vector and system variable gradients (i.e., current gradient and voltage gradient) applied in the previous cycle and store them in a LUT. Use the system variable gradients (i.e., current gradient and voltage gradient) to replace the prediction model of the traditional two-level grid-connected inverter.

[0018] Step 3: Based on the mathematical relationship between different voltage vectors, the voltage vector applied in the previous cycle and its corresponding system variable gradient are used to update the system variable gradient data corresponding to other basic voltage vectors in the lookup table (LUT);

[0019] Step 4: Based on the sampled bridge arm side current and capacitor voltage, the capacitor voltage reference value and the bridge arm side current reference value are calculated using the proposed parameter-free reference value setting method;

[0020] Step 5: Add the current gradients corresponding to the eight voltage vectors to the sampled current to predict the system current and voltage, and substitute the eight prediction results into the value function equation to obtain eight value function values; select the synthetic voltage vector with the smallest value function value as the optimal voltage vector and apply it to the next control cycle of the two-level grid-connected inverter.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] 1. The method of the present invention designs a lookup table based on the current and voltage data obtained by sensor sampling, and implements a fast lookup of the system gradient stored in the table based on the relationship between different voltage vectors and their corresponding system gradients to ensure the accuracy of system prediction. This prediction process does not require any system parameters and eliminates the impact of current gradient update stagnation.

[0023] 2. The present invention adopts a predictive control method based on a lookup table, which eliminates the dependence of the prediction value calculation on parameters in the traditional model predictive control algorithm, and effectively improves the robustness of the predictive control to changes in model parameters; by increasing the dimension of the lookup table, the system current and voltage sampling data at different times are stored, and by analyzing the mathematical relationship between different voltage vectors and their corresponding system gradients, the full update formula of the current gradient and voltage gradient of the LCL type grid-connected inverter is constructed, eliminating the update stagnation of the capacitor voltage gradient of the LCL grid-connected inverter and the current gradient on the bridge arm side, and improving the prediction accuracy.

[0024] 3. The present invention designs a parameter-free reference value calculation method that eliminates the parameter dependency of the capacitor voltage reference value and the inverter-side current reference value calculation for an LCL-type voltage source inverter. A simplified VSG method is used to provide a capacitor voltage reference for the proposed method. The inverter-side current reference is derived by utilizing the equation relationship between the sampled values ​​of the capacitor current and capacitor voltage and the reference value, eliminating the parameter dependency of the capacitor voltage reference value and the grid-connected current reference value calculation for the LCL-type voltage source inverter. This method achieves stable control of active and reactive power when the grid-connected inverter is connected to the grid in voltage source mode, providing inertial support for the power grid.

[0025] 4. The robust predictive control method adopted by the present invention can effectively solve the problems of traditional model predictive control methods such as strong dependence on model parameters and large output current ripple, and is particularly suitable for grid-connected power generation systems with frequently changing line impedance. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is the grid-connected control block diagram of the grid-connected inverter in voltage source mode.

[0027] Figure 2 This is the topology diagram of the two-level grid-connected inverter of the present invention and its basic voltage vector diagram.

[0028] Figure 3 This is a schematic diagram of real-time updating of the current gradient of the present invention.

[0029] Figure 4 This is the control block diagram of the grid-connected inverter voltage source mode VSG.

[0030] Figure 5 The reference value is obtained.

[0031] Figure 6 Schematic diagram of the control flow implemented in the present invention.

[0032] Figure 7 Diagram of the experimental platform constructed for the present invention.

[0033] Figure 8 These are the experimental results of the present invention and the traditional model when the nominal active power Pn is set to 10000W and the reactive power Qn is set to OVar; among them, (a) is the traditional model predictive control method, and (b) is the improved robust predictive control method proposed in the present invention.

[0034] Figure 9 These are the experimental results of power sharing and system frequency response of the present invention and the traditional model under a reference power step; among them, (a) is the traditional model predictive control method, and (b) is the improved robust predictive control method proposed in the present invention.

[0035] Figure 10The experimental results of the present invention and the traditional model under parameter mismatch are shown in Figure 2. The active power amplitude is set to 12500W, and the mismatched L i * The values ​​of C* vary from 20% to 200% of their actual values; wherein, (a) is the traditional model predictive control method, and (b) is the improved robust predictive control method proposed in the present invention. DETAILED DESCRIPTION

[0036] The present invention is further described below in conjunction with embodiments and drawings.

[0037] When the grid-connected inverter is connected to the grid in voltage mode, the inverter effectively injects DC power into the grid by adjusting the amplitude, frequency, and phase of the output voltage. In this mode, the output current of the inverter is determined by the grid and is affected by the external load. The control block diagram using the traditional model predictive control method is as follows: Figure 1 As shown in the figure, it mainly consists of two parts: the first part is the virtual synchronous machine (VSG) part, which obtains the virtual angular frequency and amplitude of the reference voltage through active power and reactive power. At the same time, the VSG control has a virtual impedance function, thereby realizing accurate power sharing; the second part is the predictive control part, which uses the discrete model of the inverter to optimize and select the inverter output voltage vector through the model predictive control method, thereby realizing accurate and rapid control of the output voltage to meet the steady-state and dynamic response requirements of the inverter grid.

[0038] In this regard, an embodiment of the present invention provides an improved robustness prediction control method for a new energy grid-connected inverter, such as Figure 6 As shown, here, the system current and voltage sampling data of the LCL type grid-connected inverter are used to calculate the system gradient and stored in a lookup table (LUT). By analyzing the relationship between different voltage vectors and their corresponding system gradients, a real-time fast update method of the system gradient is designed, and system prediction is performed based on the system gradient, eliminating the dependence of traditional predictive control on system parameters and improving the robustness of predictive control. At the same time, a parameter-free reference value calculation method is designed, and a simplified VSG method is used to provide a capacitor voltage reference for the proposed method. The inverter side current reference is obtained by utilizing the equation relationship between the sampling values ​​of the capacitor current and capacitor voltage and the reference value, eliminating the parameter dependence of the calculation of the capacitor voltage reference value and the grid-connected current reference value of the LCL type voltage source inverter. The specific steps are as follows:

[0039] Step 1: Establish a mathematical model of the two-level grid-connected inverter in the αβ stationary coordinate system. According to the three-phase switch states of the two-level inverter, obtain its eight basic voltage vectors, namely u0(0,0,0), u1(1,0,0), u2(1,1,0), u3(0,1,0), u4(0,1,1), u5(0,0,1), u6(1,0,1), and u7(1,1,1). Please refer to Figure 2 shown.

[0040] The sampled three-phase bridge arm current i iabc , three-phase capacitor voltage v cabc Perform Clarke coordinate transformation to obtain i iαβ 、v cαβ :

[0041]

[0042] Unlike the current-source grid-connected inverter with LCL filtering, although the voltage source inverter filter also uses LCL filtering, the output current on the grid side is indirectly controlled by adjusting the power. Therefore, the system model is reduced to second order, which is expressed as:

[0043]

[0044] Among them, u iαβ =[u iα ,u iβ ] T ,u iα and u iβ Respectively represent the voltage components output by the two-level grid-connected inverter on the α-axis and β-axis; i iαβ =[i iα ,i iβ ] T ,i iα and i iβ They represent the bridge arm current components output by the grid-connected inverter on the α-axis and β-axis respectively; v cαβ =[v cα ,v cβ ] T , v cα and v cβ Represent the voltage components of the capacitor voltage on the α-axis and β-axis respectively; L i represents the inductance of the bridge arm side of the LCL filter, and C represents the capacitance of the LCL filter.

[0045] When the parameters in the mathematical model of the grid-connected inverter do not match the actual parameters, its mathematical model will become:

[0046]

[0047] Where, ΔL i and ΔC both represent the parameter errors of the LCL filter.

[0048] Discretize the above mathematical model to obtain its prediction model:

[0049]

[0050] Among them, T s Indicates the control period.

[0051] Step 2: Calculate the voltage vector and system variable gradients (current gradient, voltage gradient) applied in the previous cycle and store them in the LUT. Use the system variable gradients (current gradient and voltage gradient) to replace the prediction model of the traditional two-level grid-connected inverter.

[0052] According to the prediction model (4), when there is a parameter error in the mathematical model, the result of the prediction model will change. If the formula (4) is rewritten as the formula (5), the current gradient Δi corresponding to the voltage vector at the previous moment can be calculated. iαβ (k-1) and voltage gradient Δv cαβ (k-1):

[0053]

[0054] According to the current gradient Δi iαβ (k-1) and voltage gradient Δv cαβ (k-1), rewriting the prediction model (4) into (6) can completely eliminate the impact of parameter mismatch on the prediction model:

[0055]

[0056] Step 3: Based on the mathematical relationship between different voltage vectors, use the voltage vector applied in the previous cycle and its corresponding system variable gradient to update the system variable gradient data corresponding to other basic voltage vectors in the LUT.

[0057] According to Equation (5), voltage and current sampling can only update the gradient value under the influence of the applied vector in the previous control cycle. For the unapplied vector, the corresponding gradient value cannot be updated. This phenomenon is called stagnation, which will increase the prediction error and reduce the output voltage performance. To eliminate the stagnation phenomenon of gradient update, the following gradient update mechanism is adopted.

[0058] If the voltage vector acting at time (k-1) is u iαβ (k-1), i∈{0,1,…,7}, according to formula (5), the system gradient can be obtained as:

[0059]

[0060] When the vector u used in formula (7) iαβ (k–1) is replaced by u jαβ (k–1), j∈{0,1,…,7} and j≠i, the system gradient on the inverter side will also become the system gradient Δi corresponding to the vector ijαβ (k–1) and Δv cjαβ (k–1), which is expressed as:

[0061]

[0062] Subtracting formula (7) from formula (8) yields the system gradient relationship corresponding to the two vectors, which is expressed as:

[0063]

[0064] According to formula (9), the voltage vector u acting at time (k-1) is iαβ (k-1) corresponding system gradient Δi iiαβ (k–1), Δv ciαβ (k–1) and the rest of the basic voltage vectors u jαβ (k–1) corresponding system gradient Δi ijαβ (k–1), Δv cjαβ (k–1) has an amplitude difference, and this amplitude difference is determined by the amplitude difference of the inverter action vector and the coefficient T s / L i 、T s / C are jointly determined. In order to eliminate the coefficient T s / L i 、T s The influence of / C on the system gradient update can be expressed by pushing formula (9) forward one control cycle:

[0065]

[0066] Divide formula (9) by formula (10) to eliminate the coefficient T s / L i 、T s / C and is expressed as:

[0067]

[0068] After adjusting formula (11), the updated system gradient is obtained and expressed as:

[0069]

[0070] It should be noted that the validity of the update formula shown in formula (11) will be affected by the value of the denominator. When the denominator in the formula is equal to zero (i.e., u iαβ(k–2)-u jαβ (k–2)=0 or i iαβ (k–2)-i jαβ (k–2)=0), according to formula (9), the updated system gradient is equal to the system gradient obtained by sampling, which is expressed as:

[0071]

[0072] Based on the above derivation, the full update of the system gradient corresponding to all voltage vectors can be achieved without using system parameters, eliminating the update stagnation, such as Figure 3 shown.

[0073] Step 4: Based on the sampled bridge arm side current and capacitor voltage, the capacitor voltage reference value and the bridge arm side current reference value are calculated using the proposed parameter-free reference value setting method.

[0074] After replacing the traditional prediction model with the proposed method, the influence of parameters on the prediction control can be eliminated. In this section, the present invention uses a simplified VSG method to provide a capacitor voltage reference for the proposed method, achieving stable control of active and reactive power when the grid-connected inverter is connected to the grid in voltage source mode, and providing inertia support for the grid. The simplified VSG control block diagram is shown in the figure. Figure 4 As shown in the figure, it mainly includes the following four parts: ① speed regulator; ② swing equation of inertia simulation; ③ reactive power control; ④ virtual impedance loop.

[0075] ① Speed ​​​​regulator: The main goal of the speed ​​regulator is to adjust the active power according to the frequency deviation. The essence of the speed ​​regulator is achieved through ω-P droop control, which is expressed as:

[0076] P in =P n -k ω (ω m -ω n ) (14)

[0077] Where, P in With ω m are the virtual active power and virtual angular frequency of VSG, k ω is the droop coefficient of ω-P droop control, P n With ω n are the nominal active power and nominal angular frequency, respectively, where ω n =2πf n .

[0078] ② Swing equation for inertia simulation: The main goal of the swing equation is to simulate the inertia and damping characteristics of the synchronous generator by introducing the rotor motion equation, which is expressed as:

[0079]

[0080]

[0081] Where, P out is the output active power of VSG, D is the damping factor of VSG, J is the virtual inertia matrix of VSG, ω c are the cutoff frequency, damping factor, and virtual moment of inertia of the low-pass filter.

[0082] ③ Reactive power control: In VSG, the essence of reactive power control is achieved through QV droop control, which is expressed as:

[0083] V ref =V n -k q (Q out -Q n ) (17)

[0084]

[0085] Where V ref , is the voltage reference amplitude, V n is the nominal voltage amplitude, k q is the droop coefficient of QV droop control, Q out is the output reactive power, Q n is the nominal reactive power.

[0086] ④ Virtual impedance loop: Different line impedances may affect the accuracy of power sharing. To mitigate this adverse effect, VSGs usually use virtual impedance to correct the output impedance and achieve efficient power distribution. Therefore, the inner loop voltage reference value with a virtual impedance loop is expressed as:

[0087]

[0088] Where, v cαβ ref is the capacitor voltage reference of the proposed MFPC, Z v The virtual impedance is preset for damping synchronous resonance. Since the capacitor voltage reference is a general sinusoidal variable, the virtual impedance is expressed as:

[0089] Z v =R v +jωL v (20)

[0090] It can be seen from formula (21) that the calculation of the inverter side current reference value in the traditional model predictive control is affected by the filter capacitor parameters.

[0091]

[0092] In order to eliminate the influence of system parameters on reference calculation, this section proposes a new reference calculation method. The essence of reference calculation is to obtain the inverter side current reference by using the relationship between capacitor current and capacitor voltage.

[0093] like Figure 5 As shown, the system sampling data of the voltage source grid-connected inverter meets the following requirements:

[0094]

[0095] Formula (21) and formula (22) are transformed and divided to eliminate the coefficient ωC and expressed as:

[0096]

[0097] Generally, the influence of system parameter changes on the inverter-side current reference value can be eliminated by (23). However, when the denominator of (23) is close to zero, it cannot be used to calculate the system reference value. Another method for calculating the system reference value is needed.

[0098]

[0099] According to formula (23) and formula (24), the calculation process of the proposed inverter side current reference value does not require the use of any system parameters, thereby eliminating the influence of parameter errors on the reference calculation and improving the parameter robustness of the system control.

[0100] Step 5: Add the current gradients corresponding to the eight voltage vectors to the sampled current to predict the system current and voltage. Substitute the eight prediction results into the merit function equation to obtain eight merit function values. The resulting voltage vector with the smallest merit function value is selected as the optimal voltage vector and applied to the next control cycle of the two-level grid-connected inverter.

[0101] The cost function of capacitor voltage tracking and inverter side current tracking is expressed as:

[0102]

[0103] Then, the cost function of the inverter is expressed as:

[0104] J=λ v J v +λ i J i (26)

[0105] The value function values ​​corresponding to the eight basic voltage vectors are calculated according to the value function (26), and the voltage vector with the smallest value function value is selected as the optimal vector to be used in the next control cycle.

[0106] Specific experiments:

[0107] Experimental platform such as Figure 7 The experimental setup includes a VSI, TMS32F28335, AC voltage source, DC voltage source, and LCL filter. The sampling frequency is set to 20 kHz. The experimental parameters are shown in Table 1.

[0108] Table 1 System parameters

[0109]

[0110] In order to verify the steady-state performance of the present invention, Figure 8 Given the nominal active power P n is set to 10000W, the reactive power Q n Experimental results with Var set to 0. Figure 8 (a) shows the predicted and reference ripples of the inverter current and capacitor voltage under the traditional model predictive control method, where the prediction and reference are calculated based on system parameters. It can be seen that under the traditional model predictive control method, the root mean square error (RMSE) of the inverter current prediction error is 1.81A, and the RMSE of the capacitor voltage prediction error is 4.40V. Figure 8 (b) shows the predicted and reference ripples of the inverter current and capacitor voltage under the improved robust predictive control method proposed in this invention, where the prediction and reference are generated by the proposed MFP and PLR methods, respectively. The RMSE of the inverter current prediction error is 0.54 A, and the RMS of the capacitor voltage prediction error is 2.05 V, verifying the effectiveness and correctness of the proposed method.

[0111] In order to evaluate the dynamic performance under precise parameters, Figure 9 The experimental results of the present invention on power sharing and system frequency response under reference power steps are shown; here, the nominal active power is from 0 W to 5000 W, and then from 5000 W to 10000 W. The amplitude of the reactive power is set to 0Var. Figure 9 (a) shows the experimental results of the traditional model predictive control method. The response time for the nominal power steps from 0W to 5000W and 5000W to 10000W is 237ms and 216ms, respectively. Figure 9 (b) shows the experimental results of the improved robust predictive control method proposed in this paper. The proposed method has a response time of 251ms and 238ms for nominal power steps from 0W to 5000W and from 5000W to 10000W, respectively. This shows that the dynamic response of the proposed method is similar to that of traditional model predictive control methods.

[0112] Figure 10The experimental results of different control methods under the condition of parameter mismatch are shown. The amplitude of active power is set to 12500W and the mismatch L i * and C* vary from 20% to 200% of their actual values. Figure 10 (a) shows the grid connection point voltage v under the traditional model predictive control method c It can be observed that when the mismatch value is large, the grid voltage v c The ripple is obviously distorted. When the parameter mismatch is 200% of the actual value (L i * =2L i , C*=2C), the grid voltage prediction error is 35.59V, and THD increases from 0.75% in steady state to 6.24%. When the mismatch value is 20% of the actual value (L i * =0.2L i , C*=0.2C), the grid connection point voltage prediction error is 28.94, and the THD increases from 0.75% in steady state to 3.55%. Figure 10 (b) shows the grid connection point voltage v under the improved robust predictive control method proposed in this invention c It can be seen that the grid connection point voltage prediction error is always less than 2V, and the THD is always less than 1%. Compared with the traditional model predictive control method under parameter mismatch, it has more stable control effect and lower error, and has excellent parameter robustness.

[0113] In summary, the present invention proposes an improved robust predictive control method for a new energy grid-connected inverter, aiming to enhance the parameter robustness of traditional predictive control and reduce current ripple. The method uses the current data and voltage data obtained by sensor sampling to design a lookup table, and based on the relationship between different voltage vectors and their corresponding system gradients, the system gradient is stored in a fast lookup table to ensure the accuracy of system prediction. The prediction process does not require any system parameters and eliminates the impact of stagnation in current gradient updates. At the same time, the present invention also designs a parameter-free reference value calculation method, which adopts a simplified VSG method to provide a capacitor voltage reference. By utilizing the equation relationship between the sampled values ​​of capacitor current and capacitor voltage and the reference value, the inverter side current reference is obtained, thereby eliminating the parameter dependence of the calculation of the capacitor voltage reference value and the grid-connected current reference value of the LCL type voltage source inverter. The robust predictive control method proposed by the present invention overcomes the problems of strong dependence on model parameters and large output current ripple of the traditional model predictive control method. It is particularly suitable for grid-connected power generation systems with frequent changes in line impedance. By introducing an improved control algorithm and a parameter-free reference value calculation method, this method can improve the stability and control accuracy of the system, thereby providing a reliable solution for the application of new energy grid-connected inverters.

[0114] The above content is merely an example and explanation of the concept of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in a similar manner. As long as they do not deviate from the concept of the invention or exceed the scope defined by the claims, they should all fall within the scope of protection of the present invention.

Claims

1. An improved robust predictive control method for renewable energy grid-connected inverters, characterized in that: The specific steps are as follows: Step 1: Establish a mathematical model of the two-level grid-connected inverter in the αβ stationary coordinate system. According to the three-phase switch states of the two-level inverter, obtain its eight basic voltage vectors, namely u0(0,0,0), u1(1,0,0), u2(1,1,0), u3(0,1,0), u4(0,1,1), u5(0,0,1), u6(1,0,1), and u7(1,1,1). Step 2: Calculate the voltage vector and system variable gradients applied in the previous cycle, i.e., current gradient and voltage gradient, and store them in the LUT; The prediction model of the traditional two-level grid-connected inverter is replaced by the system variable gradient, i.e., the current gradient and the voltage gradient; Step 3: Based on the mathematical relationship between different voltage vectors, the voltage vector applied in the previous cycle and its corresponding system variable gradient are used to update the system variable gradient data corresponding to other basic voltage vectors in the lookup table (LUT); The detailed steps are as follows: In order to eliminate the stagnation phenomenon of gradient update, the following gradient update mechanism is adopted; If the voltage vector acting at time (k-1) is u iαβ (k-1), i∈{0,1,…,7}, the system gradient is: Where Δi iαβ (k-1) is the current gradient corresponding to the voltage vector at the previous moment, Δv cαβ (k-1) is the voltage gradient corresponding to the voltage vector at the previous moment; When the vector u used in formula (7) iαβ (k–1) is replaced by u jαβ (k–1), j∈{0,1,…,7} and j≠i, the system gradient on the inverter side will also become the system gradient Δi corresponding to the vector ijαβ (k–1) and Δv cjαβ (k–1), which is expressed as: Subtracting formula (7) from formula (8) yields the system gradient relationship corresponding to the two vectors, which is expressed as: According to formula (9), the voltage vector u acting at time (k-1) is iαβ (k-1) corresponding system gradient Δi iiαβ (k–1), Δv ciαβ (k–1) and the rest of the basic voltage vectors u jαβ (k–1) corresponding system gradient Δi ijαβ (k–1), Δv cjαβ (k–1) has an amplitude difference, and this amplitude difference is determined by the amplitude difference of the inverter action vector and the coefficient T s / L i 、T s / C jointly determined; in order to eliminate the coefficient T s / L i 、T s The influence of / C on the system gradient update is to push formula (9) forward by one control cycle and express it as: Divide formula (9) by formula (10) to eliminate the coefficient T s / L i 、T s / C and is expressed as: After adjusting formula (11), the updated system gradient is obtained and expressed as: The validity of the update formula shown in formula (11) will be affected by the value of the denominator. When the denominator in the formula is equal to zero (i.e., u iαβ (k–2)-u jαβ (k–2)=0 or i iαβ (k–2)-i jαβ (k–2)=0), according to formula (9), the updated system gradient is equal to the system gradient obtained by sampling, which is expressed as: Based on the above derivation, it is possible to fully update the system gradients corresponding to all voltage vectors without using system parameters, eliminating their update stagnation; Step 4: Based on the sampled bridge arm side current and capacitor voltage, the capacitor voltage reference value and the bridge arm side current reference value are calculated using the proposed parameter-free reference value setting method; Step 5: Add the current gradients corresponding to the eight voltage vectors to the sampled current to predict the system current and voltage, and substitute the eight prediction results into the value function equation to obtain eight value function values; select the synthetic voltage vector with the smallest value function value as the optimal voltage vector and apply it to the next control cycle of the two-level grid-connected inverter.

2. The improved robust predictive control method for a new energy grid-connected inverter according to claim 1, characterized in that: Step 1: The detailed steps are as follows: The sampled three-phase bridge arm current i iabc , three-phase capacitor voltage v cabc Perform Clarke coordinate transformation to obtain i iαβ 、v cα β: Unlike the current-source grid-connected inverter with LCL filtering, although the voltage source inverter filter also uses LCL filtering, the output current on the grid side is indirectly controlled by adjusting the power. Therefore, the system model is reduced to second order, which is expressed as: Among them, u iαβ =[u iα ,u iβ ] T ,u iα and u iβ Respectively represent the voltage components output by the two-level grid-connected inverter on the α-axis and β-axis; i iαβ =[i iα ,i iβ ] T ,i iα and i iβ They represent the bridge arm current components output by the grid-connected inverter on the α-axis and β-axis respectively; v cαβ =[v cα ,v cβ ] T , v cα and v cβ Represent the voltage components of the capacitor voltage on the α-axis and β-axis respectively; L i Indicates the inductance of the bridge arm side of the LCL filter, and C indicates the capacitance of the LCL filter; When the parameters in the mathematical model of the grid-connected inverter do not match the actual parameters, its mathematical model will become: Where, ΔL i and ΔC both represent the parameter errors of the LCL filter; Discretize the above mathematical model to obtain its prediction model: Among them, T s Indicates the control period.

3. The improved robust predictive control method for a new energy grid-connected inverter according to claim 2, characterized in that: The detailed steps for step 2 are as follows: According to the prediction model (4), when there is a parameter error in the mathematical model, the result of the prediction model will change. If the formula (4) is rewritten as the formula (5), the current gradient Δi corresponding to the voltage vector at the previous moment can be calculated. iαβ (k-1) and voltage gradient Δv cαβ (k-1): According to the current gradient Δi iαβ (k-1) and voltage gradient Δv cαβ (k-1), rewriting the prediction model (4) into (6) can completely eliminate the impact of parameter mismatch on the prediction model:

4. The improved robust predictive control method for a new energy grid-connected inverter according to claim 3, characterized in that: Step 4: Detailed steps are as follows: The simplified VSG control system achieves stable control of active and reactive power when the grid-connected inverter is connected to the grid in voltage source mode, providing inertial support for the grid. It consists of four parts: ① speed regulator; ② swing equation for inertia simulation; ③ reactive power control; and ④ virtual impedance loop. ① Speed ​​​​regulator: The main goal of the speed ​​regulator is to adjust the active power according to the frequency deviation. The essence of the speed ​​regulator is achieved through ω-P droop control, which is expressed as: P in =P n -k ω (oh m -oh n ) (14) Where, P in With ω m are the virtual active power and virtual angular frequency of VSG, k ω is the droop coefficient of ω-P droop control, P n With ω n are the nominal active power and nominal angular frequency, respectively, where ω n =2πf n ; ② Swing equation for inertia simulation: The main goal of the swing equation is to simulate the inertia and damping characteristics of the synchronous generator by introducing the rotor motion equation, which is expressed as: Where, P out is the output active power of VSG, D is the damping factor of VSG, J is the virtual inertia matrix of VSG, ω c are the cutoff frequency, damping factor and virtual moment of inertia of the low-pass filter; ③ Reactive power control: In VSG, the essence of reactive power control is achieved through QV droop control, which is expressed as: V ref =V n -k q (Q out -Q n )(17) Where V ref , is the voltage reference amplitude, V n is the nominal voltage amplitude, k q is the droop coefficient of QV droop control, Q out is the output reactive power, Q n is the nominal reactive power; ④ Virtual impedance loop: Different line impedances may affect the accuracy of power sharing. To mitigate this adverse effect, VSGs usually use virtual impedance to correct the output impedance and achieve efficient power distribution. Therefore, the inner loop voltage reference value with a virtual impedance loop is expressed as: Where, v cαβ ref is the capacitor voltage reference of the proposed MFPC, Z v The virtual impedance is preset for damping synchronous resonance. Since the capacitor voltage reference is a general sinusoidal variable, the virtual impedance is expressed as: Z v =R v +jωL v (20) From formula (21), it can be seen that the calculation of the inverter side current reference value in the traditional model predictive control is affected by the filter capacitor parameters; In order to eliminate the influence of system parameters on reference calculation, a new reference calculation method is proposed. The essence is to use the relationship between capacitor current and capacitor voltage to obtain the inverter side current reference; The system sampling data of the voltage source grid-connected inverter meets the following requirements: Formula (21) and formula (22) are transformed and divided to eliminate the coefficient ωC and expressed as: Generally, the influence of system parameter changes on the inverter-side current reference value can be eliminated by (23); however, when the denominator of (23) is close to zero, it cannot be used to calculate the system reference value; another system reference value calculation method is needed; According to formula (23) and formula (24), the calculation process of the proposed inverter side current reference value does not require the use of any system parameters, thereby eliminating the influence of parameter errors on the reference calculation and improving the parameter robustness of the system control.

5. The improved robust predictive control method for a new energy grid-connected inverter according to claim 4, characterized in that: Step 5: The detailed steps are as follows: The cost function of capacitor voltage tracking and inverter side current tracking is expressed as: Then, the cost function of the inverter is expressed as: J=λ v J v +λ i J i (26) The value function values ​​corresponding to the eight basic voltage vectors are calculated according to the value function (26), and the voltage vector with the smallest value function value is selected as the optimal vector to be used in the next control cycle.

Images