A Model-Free Robust Predictive Control Method and System for Three-Level SNPC Inverters

Abstract

This invention relates to the field of three-level inverters, specifically disclosing a model-free robust predictive control method and system for three-level SNPC inverters. The method includes: real-time sampling and coordinate transformation of the inverter to obtain current and voltage components in a two-phase stationary coordinate system; establishing a hyperlocal model to unify system parameter deviations and external disturbances as a lumped disturbance, and designing a full-order sliding mode observer for real-time estimation; calculating the reference voltage vector based on the disturbance estimate and the extrapolated reference current; selecting the optimal dual-vector combination through space vector optimization, solving for the action time, and selecting redundant small vectors based on the midpoint voltage deviation to generate a PWM drive signal. This invention maintains excellent current tracking performance and low harmonic distortion even under parameter mismatch and operating condition changes, while simultaneously achieving DC-side midpoint potential balance. It has advantages such as low computational complexity, strong robustness, and ease of engineering implementation.

Description

Technical Field

[0001] This invention relates to the field of three-level inverter technology, specifically to a model-free robust predictive control method and system for a three-level SNPC inverter. Background Technology

[0002] Three-phase multilevel inverters are widely used in motor drives, grid-connected power generation, and battery energy storage systems due to their advantages such as high output voltage quality and low electromagnetic interference. Based on the traditional three-phase three-level topology, the three-level SNPC converter introduces a segmented structure of common and independent modules, significantly reducing the number of active switching devices while maintaining three-level output capability, thus lowering system cost and losses.

[0003] Existing research indicates that while Finite Control Set Model Predictive Control (FCS-MPC) offers advantages such as fast response and flexible constraint handling in power electronic systems, its control variable prediction process typically relies on object parameters, such as resistance and inductance. When the parameters used by the controller deviate from the actual object parameters—a phenomenon known as parameter mismatch—prediction errors are introduced, leading to degraded control performance. The paper "Robust Model Predictive Control for a Three-Phase PMSM Motor With Improved Control Precision," published in IEEE TIE in 2021, points out that FCS-MPC is sensitive to parameters, and parameter mismatch will cause prediction errors in the control target, thereby degrading control performance. In experimental comparisons, when there is inductor mismatch, such as L'=2L, the current ripple of the traditional MPC is significantly increased compared to the case without mismatch. It can be seen that in scenarios such as three-level SNPC inverters, if predictive control still explicitly relies on nominal filter parameters and equivalent impedance parameters, the parameter drift caused by component discreteness, temperature rise aging, and changes in operating conditions will inevitably lead to prediction bias, which manifests as increased current tracking error, increased steady-state ripple and harmonic content, and may further affect the midpoint potential balance control effect.

[0004] Therefore, it is necessary to propose a robust predictive control method applicable to three-level SNPC converters that can explicitly consider the effects of parameter mismatch, so as to improve the robustness and power quality of the system under parameter mismatch and operating condition changes. Summary of the Invention

[0005] To address the problems of high dependence on parameter accuracy and insufficient robustness in existing model predictive control technologies, this invention provides a model-free, robust predictive control method and system for three-level SNPC inverters. By constructing a hyperlocal model, the system uncertainty is aggregated into a disturbance term, and an observer is used for real-time estimation and compensation. This improves the power quality of the output current and the robustness of the system while ensuring the balance of the midpoint potential.

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

[0007] A model-free robust predictive control method for a three-level SNPC inverter includes the following steps:

[0008] S1. Sample the three-phase output current, grid-side voltage, and DC-side capacitor voltage of the inverter, and transform them to obtain the current and voltage in a two-phase stationary coordinate system;

[0009] S2. Construct a hyperlocal model of the inverter, unify the system parameter uncertainty into a lumped disturbance, and use a full-order observer to estimate the lumped disturbance in real time to obtain the disturbance estimate;

[0010] S3. Based on the current command, the reference current at future time is predicted by extrapolation, and the reference voltage vector for the next sampling period is calculated based on the hyperlocal model by combining the current sampled current and the disturbance estimate.

[0011] S4. Based on the position of the reference voltage vector in the spatial vector diagram, a set of candidate voltage vectors is selected. The optimal dual-vector combination is evaluated and selected through a cost function, and its application time is calculated. Based on the DC side midpoint voltage deviation, a redundant small vector is selected, and a PWM drive signal is generated to achieve modulation control.

[0012] As a preferred embodiment of the present invention, S1 specifically includes:

[0013] In each sampling period, the three-level SNPC inverter is sampled in real time. The three-phase output current is obtained through the current sensor, and the three-phase grid-side voltage and the DC-side upper and lower capacitor voltages are obtained through the voltage sensor.

[0014] The three-phase output current and the three-phase grid-side voltage are subjected to Clark transformation to obtain the current component and grid-side voltage component in the two-phase stationary coordinate system, which are used as inputs for the subsequent hyperlocal model and full-order observer.

[0015] As a preferred embodiment of the present invention, in step S2, a hyperlocal model of the inverter is constructed, and the uncertainty of system parameters is uniformly equivalent to a lumped disturbance, specifically including:

[0016] In the aforementioned two-phase stationary coordinate system, filter inductance, resistance, grid-side impedance, midpoint voltage deviation, and power device on-state voltage drop are introduced as gain coefficients to unify and equate system uncertainties to lumped disturbances. This simplifies the model structure and decouples the parameters, thereby constructing a hyperlocal model. ;

[0017] in: This represents the output voltage of a three-level SNPC inverter in a two-phase stationary coordinate system. The output current is in a two-phase stationary coordinate system. α is the transpose of the matrix, and α is a design parameter related to the filter inductor, used to reduce the dependence of the control algorithm on actual filter inductor, resistance and other parameters.

[0018] As a preferred embodiment of the present invention, in step S2, the lumped disturbance is estimated in real time using a full-order observer to obtain the disturbance estimate, specifically including:

[0019] A full-order observer is constructed based on a hyperlocal model, employing a sliding mode observer structure. The input to the full-order observer is the sampled current. and output voltage The state variables of a full-order observer include current observations and lumped disturbances;

[0020] A full-order observer is used to estimate lumped disturbances in real time. By discretizing the data, iterative updates are performed in each sampling period using the current sampled value and the observation from the previous period, achieving real-time estimation with low storage requirements, thus obtaining the disturbance estimate. .

[0021] In a preferred embodiment of the present invention, step S3 involves predicting the reference current at future times using an extrapolation method based on the current command, specifically including:

[0022] The current command is obtained based on the grid connection command or the output of the outer current loop regulator. ;

[0023] The second-order Lagrange extrapolation method is used to predict the reference current for the next two sampling periods based on the current command of the current and the previous two sampling periods. .

[0024] As a preferred embodiment of the present invention, step S3, which involves calculating the reference voltage vector for the next sampling period based on a hyperlocal model by combining the current sampled current and the disturbance estimate, specifically includes:

[0025] Combined with the current sampling current and disturbance estimates Obtain the predicted current ;

[0026] Under the hyperlocal model, let the predicted current for the next two sampling periods be... Tracking reference current in the next two timeframes Calculate the reference voltage vector for the next sampling period. The reference voltage vector It does not explicitly depend on the nominal parameters of the filter inductor, resistor, and grid-side impedance.

[0027] In a preferred embodiment of the present invention, step S4 involves filtering a set of candidate voltage vectors based on the position of the reference voltage vector in the spatial vector diagram, evaluating and selecting the optimal dual-vector combination through a cost function, and solving for its effective time. Specifically, this includes:

[0028] Based on the position of the reference voltage vector in the space voltage vector diagram of the three-level SNPC inverter, the sector and sub-triangle region where the reference voltage vector is located are determined. The large vector, small vector and zero vector related to the sub-triangle region are combined into a candidate voltage vector set. Several candidate line segments composed of two voltage vectors are constructed with the reference voltage vector as the target.

[0029] A preset cost function is selected, and the error between the reference voltage vector and the equivalent output voltage vector is used as the index to calculate the cost value from the reference voltage vector to the midpoint of each candidate line segment. The line segment with the smallest cost function value is selected, and the voltage vectors at its two ends are used as the two active voltage vectors in this sampling period.

[0030] Based on the geometric relationships of the spatial vector diagram or the optimization conditions that make the equivalent output voltage vector approximate the reference voltage vector, solve for the action time of the two active voltage vectors. This enables dual-vector two-segment modulation.

[0031] As a preferred embodiment of the present invention, in step S4, a redundant small vector is selected based on the DC side midpoint voltage deviation to generate a PWM drive signal for modulation control, specifically including:

[0032] The midpoint voltage deviation is calculated based on the upper capacitor voltage VP and the lower capacitor voltage VN on the DC side. When there is a redundant small vector, under the premise of meeting the current control target, different redundant small vectors are selected to participate in the dual vector synthesis according to the sign of the midpoint voltage deviation in order to achieve the DC side midpoint potential balance.

[0033] Based on the two selected active voltage vectors and their active times, PWM drive signals for the three-level SNPC inverter common module and the three-phase bridge arm are generated to complete model-free predictive control for one sampling cycle, and the above steps are repeated in subsequent sampling cycles.

[0034] A model-free robust predictive control system for a three-level SNPC inverter, used to implement a model-free robust predictive control method for a three-level SNPC inverter, including:

[0035] The sampling module is used to acquire the three-phase output current, three-phase grid-side voltage and DC-side upper and lower capacitor voltage in real time, and to complete the Clark transformation to output the current and voltage components in the two-phase stationary coordinate system.

[0036] The observation module is used to receive the current component, voltage component and inverter output voltage, and to estimate the extended state in real time based on the full-order sliding mode observer constructed based on the hyperlocal model, and output the current observation value and the lumped disturbance estimate value.

[0037] The reference voltage calculation module is used to receive current commands, current observations, and lumped disturbance estimates, and uses an extrapolation algorithm to predict the future reference current and calculate the reference voltage vector.

[0038] The modulation module is used to filter the set of candidate voltage vectors based on the position of the reference voltage vector, construct candidate line segments, select the optimal applied voltage vector based on the cost function, and solve its application time.

[0039] The control module is used to convert the selected applied voltage vector and its applied time into PWM drive signals for the common module and three-phase bridge arm of the three-level SNPC inverter, and to select redundant small vectors based on the midpoint voltage deviation to achieve midpoint potential balance control.

[0040] A computer-readable storage medium storing a computer program that, when invoked by a processor, executes a model-free robust predictive control method and system for a three-level SNPC inverter. The processor is a digital signal processor, a microcontroller, or a field-programmable gate array, used to realize real-time predictive control of the three-level SNPC inverter.

[0041] Compared with the prior art, the present invention has the following advantages:

[0042] 1. This invention constructs a hyperlocal model and a full-order sliding mode observer to unify and equate nonlinear factors such as filter inductor deviation, resistance deviation, grid-side impedance uncertainty, midpoint voltage fluctuation, and power device on-state voltage drop into a lumped disturbance and perform real-time estimation. Compared with traditional predictive control methods that rely on precise parameter models, this invention fundamentally eliminates the impact of parameter mismatch on current prediction accuracy. Even under conditions such as component aging, temperature rise changes, and sudden changes in operating conditions, it maintains low current tracking error and low total harmonic distortion rate, significantly improving the system's robustness and long-term operational stability.

[0043] 2. The full-order sliding mode observer structure used in this invention is simplified, requiring only data from the current sampling period and the previous sampling period to complete the disturbance estimation iterative update, eliminating the need to store long-term historical data and greatly reducing memory usage and computational complexity. This characteristic makes the algorithm particularly suitable for high-switching-frequency, high-power three-level SNPC inverter applications, ensuring real-time control performance and dynamic response speed. Attached Figure Description

[0044] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0045] Figure 1 This is a structural diagram of a three-level SNPC inverter system;

[0046] Figure 2 This is a basic spatial vector diagram of a three-level SNPC inverter system.

[0047] Figure 3 This is a block diagram of the predictive control of the full-order sliding mode observer (SMO) of the present invention;

[0048] Figure 4 This is the overall control block diagram of the model-free predictive control method for the three-level SNPC inverter system of the present invention;

[0049] Figure 5 The graphs show the grid-connected current waveforms and THD of the conventional dual-vector MPC and the method of this invention under inductor parameter matching conditions; where (a) is the dual-vector predictive current control method and (b) is the model-free dual-vector predictive current control method based on SMO.

[0050] Figure 6 The graphs show the grid-connected current waveforms and THD of the conventional dual-vector MPC and the method of this invention under the inductor parameter mismatch condition; where (a) is the conventional dual-vector predictive current control method and (b) is the model-free dual-vector predictive current control method based on SMO.

[0051] Figure 7 The method of this invention outputs a three-phase current waveform when the load of a three-level SNPC inverter changes abruptly.

[0052] Figure 8 This describes the midpoint voltage fluctuation during the operation of a three-level SNPC inverter using the method of the present invention. Detailed Implementation

[0053] 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 only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] The concepts involved in this application will first be described with reference to the accompanying drawings. It should be noted that the following descriptions of various concepts are only for the purpose of making the content of this application easier to understand and do not constitute a limitation on the scope of protection of this application; furthermore, the embodiments and features in the embodiments of this application can be combined with each other unless otherwise specified. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0055] Example 1

[0056] like Figure 3-8 As shown, this invention provides a model-free robust predictive control method for a three-level SNPC inverter, which, based on the mathematical model of the three-level SNPC inverter, introduces a hyperlocal model. The filter inductor / resistor deviation, grid-side voltage disturbance, dead zone, and device non-ideals are all absorbed into the lumped disturbance. The control law is implemented using only design parameters. With online estimation by the observer Thus avoiding the nominal Explicit dependence of electrical parameters; when parameter drift occurs, its effects are mediated through... The real-time updates are compensated, and finally, combined with dual-vector two-stage modulation, current prediction control and midpoint potential balance are achieved. This method is applicable to... Figure 1 The three-level SNPC grid-connected inverter system shown is illustrated.

[0057] Specifically, the steps include the following:

[0058] S1. Sample the three-phase output current, grid-side voltage, and DC-side capacitor voltage of the inverter, and transform them to obtain the current and voltage in a two-phase stationary coordinate system; specifically including:

[0059] In each sampling period Internally, the three-phase output current is sampled via a Hall current sensor. The voltage of the three-phase grid side is sampled by a voltage sensor. and the voltage of the capacitor on the DC side and lower capacitor voltage , where: k corresponds to the kth sampling time or sampling point.

[0060] The three-phase output current and the three-phase grid-side voltage are subjected to Clark transformation to obtain the current components in a two-phase stationary coordinate system (αβ coordinate system). and grid-side voltage components The transformation result serves as the input for subsequent hyperlocal model establishment and full-order observers.

[0061] S2. Construct a hyperlocal model of the inverter, unifying the system parameter uncertainties into a lumped disturbance, and using a full-order observer to estimate the lumped disturbance in real time to obtain the disturbance estimate; specifically including:

[0062] S21. In a three-level SNPC grid-connected inverter, parameters such as filter inductance, resistance, and grid-side impedance exhibit certain mismatches during actual operation. Taking a two-phase stationary coordinate system as an example, the nominal model can be expressed as:

[0063] ,

[0064] in: For grid-connected current vector, This is the inverter output voltage vector. For grid-side voltage vector, This is the equivalent voltage error term caused by dead zone, voltage sampling deviation, etc. , These are the nominal filter inductor and resistor, respectively.

[0065] Considering the effects of parameter mismatch and device on-state voltage drop, the actual model can be written as:

[0066] ,

[0067] in: , For parameter deviation, The on-resistance of the power device. This is the equivalent voltage error caused by the dead zone and sampling error.

[0068] To reduce the dependence on nominal parameters, a first-order hyperlocal model gain is introduced. Rewriting the above equation, we construct the following hyperlocal model:

[0069] ,

[0070] in: This represents the output voltage of a three-level SNPC inverter in a two-phase stationary coordinate system. The output current is in a two-phase stationary coordinate system. Here, α is the transpose of the matrix, and α is a design parameter related to the filter inductor, used to reduce the dependence of the control algorithm on actual filter inductor, resistance, and other parameters. The selection of The approximate order of magnitude of, making The amplitude is smaller and the convergence is easier to observe; in engineering, it can be determined according to the nominal value of the filter inductor and is allowed to be adjusted within a certain range;

[0071] This is a lumped disturbance term used to absorb parameter mismatch and external disturbances:

[0072] ,

[0073] Through the above modeling, all parameter errors of the filter inductor, resistor, and grid-side impedance, as well as external disturbances, are absorbed into the lumped disturbance term. This lays the foundation for subsequent hyperlocal model predictive control.

[0074] S22. Based on the above hyperlocal model, design a full-order sliding mode observer (SMO) to handle lumped disturbances. Perform real-time estimation:

[0075] First, construct the extended state vector, let the extended state vector be... and its observed values ,in: and These are the current observations and the lumped disturbance observations, respectively.

[0076] The observation errors are respectively ;

[0077] The equations for a continuous-time full-order sliding mode observer are:

[0078] ,

[0079] ,

[0080] in: , For sliding mode observation gain, For grid connection angular frequency, The imaginary unit is used in... Represents the orthogonal rotation operator on the plane.

[0081] Substituting the values, we can obtain the error dynamics:

[0082] ,

[0083] Appropriate selection , This ensures that the observation error converges to zero within a finite time, thus achieving [the goal of] [the ability to] [observe] [the accuracy of] [the observation]. and Accurate estimation; select first when setting As the equivalent bandwidth parameter of the observer (a trade-off between dynamics and noise), then select It meets the requirements for upper bound of disturbance and noise immunity. .

[0084] The above observers are based on the sampling period. Discretize the equations using the forward Euler method to obtain the discrete equations:

[0085] ,

[0086] This ensures that only the current current measurement is used in each sampling period. Control voltage The current observations can be updated using the observations from the previous cycle. and disturbance estimates This enables real-time estimation of lumped disturbances, with the observer structure as follows: Figure 3 As shown.

[0087] S3. Based on the current command, extrapolation is used to predict the reference current at future times. The reference voltage vector for the next sampling period is calculated using a hyperlocal model, combining the current sampled current with the disturbance estimate. Specifically, this includes:

[0088] S31. Based on the grid-connected operation target or the output of the upper-level power controller, let the current command given by the outer loop in the two-phase stationary coordinate system be... To improve the accuracy of model-free prediction, second-order Lagrange extrapolation is used to obtain the reference current for the next two sampling periods:

[0089] ,

[0090] in , This is the current command for the first two sampling cycles.

[0091] S32. Based on the hyperlocal model, observations are used for lumped disturbances. Under the alternative conditions, the current prediction equation can be obtained:

[0092] ,

[0093] To achieve zero steady-state error current tracking, let the predicted current at time k+2 be... Tracking reference current in the next two timeframes The reference voltage vector that can be derived is:

[0094] ,

[0095] Since the disturbance estimation term has already absorbed the errors of parameters such as filter inductance, resistance, and grid-side impedance, the resulting... It no longer explicitly depends on nominal parameters such as L and R, thus maintaining good current tracking performance even under parameter mismatch and operating condition changes.

[0096] S4. Based on the position of the reference voltage vector in the spatial vector diagram, a set of candidate voltage vectors is selected. The optimal dual-vector combination is evaluated using a cost function, and its application time is calculated. Redundant small vectors are selected based on the DC-side midpoint voltage deviation. A PWM drive signal is generated to achieve modulation control. Specifically, this includes:

[0097] S41. After determining the reference voltage vector Subsequently, based on its Figure 2 The position in the three-level SNPC space voltage vector diagram is shown, i.e., the sector (sectors 1-6) and the sub-triangle region (each sector contains 4 sub-regions). The large vector, small vector and zero vector in the corresponding region are selected to construct a candidate voltage vector set. .

[0098] For any single candidate vector The cost function is defined as the Euclidean distance between the reference voltage and the vector: ;

[0099] Construct several pairs of vectors from the candidate set. , Assuming that the duration of action of the line segment is inversely proportional to the cost function, the duration distribution of action of the two vectors can be obtained as follows: ,in: The sampling period;

[0100] When it is necessary to superimpose the zero vector, the duration of the zero vector's action can be set. .

[0101] For each pair of candidate bivectors, calculate its combined voltage:

[0102] ,

[0103] and with Selecting the optimal dual vector based on the minimum criterion and its duration of action .

[0104] S42. To achieve midpoint potential balance, utilize the sampled data... Calculate the midpoint voltage deviation .

[0105] In the presence of small vector redundancy, according to Different symbol selections involve redundant small vectors. Synthesis: When When choosing a small vector that is beneficial to increasing the voltage of the upper capacitor, priority should be given to selecting the small vector. When selecting a small vector that is beneficial to increasing the lower capacitor voltage, the midpoint potential is automatically balanced.

[0106] The final selected dual vectors and their operating times are mapped to the specific switching states of the three-level SNPC common module and the three-phase bridge arm. The gate drive signals of each power device are generated through PWM or timing comparison to complete the predictive control and midpoint potential balance of the three-level SNPC inverter.

[0107] In summary, this embodiment combines hyperlocal models, full-order observers, and dual-vector hyperlocal model predictive control. Under the conditions of filter parameter deviation and grid-side impedance uncertainty, it can significantly reduce grid-connected current harmonics and ripple, and improve the system's robustness to parameter mismatch and external disturbances.

[0108] To further verify the technical effectiveness of the method described in this invention, a simulation platform for a three-level SNPC inverter closed-loop control system was built using simulation software. The traditional FCS-MPC dual-vector predictive current control method in the prior art was selected as a comparison object, and a comparative experiment was conducted with the dual-vector model-free predictive current control method based on a full-order sliding mode observer (SMO) proposed in this invention under the same conditions. Both methods used the same sampling period. Switching frequency DC bus voltage Current reference The filter structure and its nominal parameters are kept consistent except for the "parameter mismatch" setting required for comparison. The simulation parameters are shown in Table 1.

[0109] Table 1 Simulation Parameters

[0110]

[0111] This invention sets out the following two typical operating conditions:

[0112] (1) Parameter adaptation condition: The controller parameters are consistent with the parameters of the controlled object, which is used to compare the steady-state power quality of the two methods under normal conditions;

[0113] (2) Inductance parameter mismatch condition: The inductance parameter used by the controller is 1.5 times the actual inductance, which is used to compare the robustness of the two methods under parameter mismatch conditions.

[0114] Under each operating condition, after the system reaches steady state, three-phase grid-connected current data for two consecutive fundamental cycles are extracted for FFT harmonic analysis, and THD is calculated using a unified FFT setting. The comparison results of key indicators for the two methods under different operating conditions are summarized in Table 2.

[0115] Table 2 Comparison of Key Indicators

[0116] Operating conditions index Traditional dual-vector MPC Method of the present invention Improvement Parameter matching conditions THD (%) 1.50 1.24 A decrease of approximately 17.3%. Inductor mismatch condition (1.5 times actual value) THD (%) 2.06 1.26 A decrease of approximately 38.8%

[0117] in, Figure 5 Figures (a) and (b) present simulation results of the traditional dual-vector predictive current control method and the model-free predictive current control method based on SMO of this invention under parameter adaptation conditions. Figure 5 It can be seen that both methods can achieve stable tracking, but the present invention has smaller current ripple and lower harmonic content, and the THD is further reduced from about 1.50% to about 1.24%.

[0118] Furthermore, Figure 6 (a) and (b) respectively present the comparative results of the two methods under the condition of inductance parameter mismatch (the controller inductance parameter is 1.5 times the actual inductance). Figure 6 As shown, traditional methods, due to their explicit dependence on inductor parameters in prediction calculations, introduce prediction biases and exacerbate current distortion due to mismatch, increasing THD to approximately 2.06%. In contrast, this invention uses a hyperlocal model to equate parameter biases and other uncertainties to lumped disturbances, which are then compensated for online by SMO estimation. Therefore, under the same mismatch conditions, it maintains a better current waveform, with a THD of approximately 1.26%, indicating that this invention has stronger parameter robustness. Furthermore, Figure 7 As shown, when the load of a three-level SNPC inverter undergoes a sudden change, the method of the present invention exhibits rapid output current response and a smooth transition process, demonstrating excellent dynamic performance. Figure 8 The DC-side upper and lower capacitor voltage fluctuations during operation of the three-level SNPC inverter shown are controlled within... Within 0.5V, the midpoint potential balance control effect is better.

[0119] The above results demonstrate that the model-free robust predictive control method for three-level SNPC inverters proposed in this invention can effectively reduce the output current harmonics of the SNPC inverter, improve the robustness of the system under parameter mismatch and operating condition changes, and simultaneously achieve fast dynamic response and good DC-side midpoint voltage balance control.

[0120] Example 2

[0121] like Figure 4 As shown, a model-free robust predictive control system for a three-level SNPC inverter is provided to implement a model-free robust predictive control method for a three-level SNPC inverter, comprising:

[0122] The sampling module is used to sample the three-level SNPC inverter system in real time to obtain the three-phase output current, three-phase grid-side voltage, and DC-side upper and lower capacitor voltages. The three-phase output current and three-phase grid-side voltage are transformed by Clark transformation to form the current in a two-phase stationary coordinate system. and grid-side voltage ;

[0123] The observation module is used to measure the current in a two-phase stationary coordinate system. Grid-side voltage and inverter output voltage The system takes a full-order sliding mode observer as input, estimates the extended state in real time within a hyperlocal model framework, and outputs a lumped perturbation estimate. and current observations ;

[0124] Among them, the observation module uses a full-order sliding mode observer to correct lumped disturbances by utilizing current observation errors, so as to improve the observation accuracy of parameter mismatch and external disturbances.

[0125] The reference voltage calculation module is used to calculate the voltage based on a hyperlocal model, combined with current commands. The reference current at future times is extrapolated and predicted. Then, using the current sampled current and the estimated lumped disturbance, the reference voltage vector for the next sampling period is calculated. The reference voltage vector is used to compensate for parameter mismatch and external disturbances, and its calculation process does not explicitly depend on the nominal parameters of the filter inductor, resistor and grid-side impedance.

[0126] The modulation module is used to filter the set of candidate voltage vectors based on the position of the reference voltage vector, construct candidate line segments, select the optimal applied voltage vector based on the cost function, and solve its application time.

[0127] The control module is used to convert the selected applied voltage vector and its applied time into PWM drive signals for the common module and three-phase bridge arm of the three-level SNPC inverter, and to select redundant small vectors based on the midpoint voltage deviation to achieve midpoint potential balance control, which is used to control the operation of the three-level SNPC inverter system.

[0128] Example 3

[0129] A system based on the control method, comprising:

[0130] The three-level SNPC inverter main circuit and control subsystem include a DC-side capacitor and DC power supply, a level generation unit, and a three-phase inverter bridge arm and filter connected to the grid or load. The control subsystem uses the aforementioned super local predictive control method to drive the level generation unit and generate corresponding switching control signals to control the inverter main circuit.

[0131] Example 4

[0132] A computer-readable storage medium storing a computer program, which, when called by a processor, executes a model-free robust predictive control method and system for a three-level SNPC inverter, wherein the processor is a digital signal processor, a microcontroller, or a field-programmable gate array, and is used to realize real-time predictive control of a three-level SNPC inverter.

[0133] As can be seen from the above description, the embodiments of the present invention achieve the following technical effects:

[0134] This invention effectively models uncertainties in inverter systems (such as filter inductance, resistance, grid-side impedance, and nonlinearity of switching devices) as lumped disturbances by introducing a hyperlocal model and a full-order sliding mode observer (SMO). The system performs real-time estimation of the current. Compared to traditional control methods based on precise parameter models, this invention can still maintain low current tracking error and small total harmonic distortion (THD) under parameter mismatch, component aging, and external disturbances, significantly improving the robustness and adaptability of the system.

[0135] The full-order sliding mode observer in this invention reduces reliance on long-term historical data through a simplified observer structure. Compared to traditional methods, the observer only relies on data from the current and previous sampling periods for disturbance estimation, significantly reducing storage requirements and system computational complexity. It is particularly suitable for applications requiring real-time control and high dynamic performance, such as high-power inverters and grid-connected systems.

[0136] The embodiments and / or implementation methods described above are merely preferred embodiments and / or implementation methods for implementing the technology of the present invention, and are not intended to limit the implementation methods of the technology of the present invention in any way. Any person skilled in the art can make some modifications or alterations to other equivalent embodiments without departing from the scope of the technical means disclosed in the content of the present invention, but they should still be regarded as the technology or embodiments that are substantially the same as the present invention.

[0137] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are only preferred embodiments of this application. It should be noted that due to the limitations of written expression, while there are objectively infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this application, and can also combine the above technical features in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. A model-free robust predictive control method for a three-level SNPC inverter, characterized in that, include: The three-phase output current, grid-side voltage, and DC-side capacitor voltage of the sampled inverter are transformed to obtain the current and voltage in a two-phase stationary coordinate system. A hyperlocal model of the inverter is constructed, and the uncertainty of system parameters is uniformly equivalent to a lumped disturbance. The lumped disturbance is estimated in real time using a full-order observer to obtain the disturbance estimate. Based on the current command, the reference current at future times is predicted using extrapolation. The reference voltage vector for the next sampling period is calculated based on the hyperlocal model by combining the current sampled current with the disturbance estimate. Based on the position of the reference voltage vector in the spatial vector diagram, a set of candidate voltage vectors is selected, the optimal dual-vector combination is evaluated and selected by the cost function and its action time is solved, and redundant small vectors are selected based on the DC side midpoint voltage deviation to generate a PWM drive signal to achieve modulation control. Based on the current command, the reference current at future times is predicted using an extrapolation method, specifically including: The current command is obtained based on the grid connection command or the output of the outer current loop regulator. ; The second-order Lagrange extrapolation method is used to predict the reference current for the next two sampling periods based on the current command of the current and the previous two sampling periods. ; Combining the current sampled current with the disturbance estimate, the reference voltage vector for the next sampling period is calculated based on a hyperlocal model, specifically including: Combined with the current sampling current and disturbance estimates Obtain the predicted current ; Under the hyperlocal model, let the predicted current for the next two sampling periods be... Tracking reference current in the next two timeframes Calculate the reference voltage vector for the next sampling period. The reference voltage vector It does not explicitly depend on the nominal parameters of the filter inductor, resistor, and grid-side impedance.

2. The model-free robust predictive control method for a three-level SNPC inverter according to claim 1, characterized in that, The three-phase output current, grid-side voltage, and DC-side capacitor voltage of the sampling inverter are transformed to obtain current and voltage in a two-phase stationary coordinate system, specifically including: In each sampling period, the three-level SNPC inverter is sampled in real time. The three-phase output current is obtained through the current sensor, and the three-phase grid-side voltage and the DC-side upper and lower capacitor voltages are obtained through the voltage sensor. The three-phase output current and the three-phase grid-side voltage are subjected to Clark transformation to obtain the current component and grid-side voltage component in a two-phase stationary coordinate system.

3. The model-free robust predictive control method for a three-level SNPC inverter according to claim 2, characterized in that, Constructing a hyperlocal model of the inverter, unifying the system parameter uncertainties into lumped disturbances, specifically includes: In the aforementioned two-phase stationary coordinate system, the deviations of the real-time sampling parameters and the influence of non-ideal factors are uniformly and equivalently represented as a lumped disturbance. Thus, a hyperlocal model is constructed. ; in: This represents the output voltage of a three-level SNPC inverter in a two-phase stationary coordinate system. The output current is in a two-phase stationary coordinate system. This is the transpose of the matrix. α These are the design parameters related to the filter inductor.

4. The model-free robust predictive control method for a three-level SNPC inverter according to claim 3, characterized in that, The lumped disturbance is estimated in real time using a full-order observer to obtain the disturbance estimate, specifically including: A full-order observer is constructed based on a hyperlocal model, employing a sliding mode observer structure. The input to the full-order observer is the sampled current. and output voltage The state variables of a full-order observer include current observations and lumped disturbances; A full-order observer is used to estimate the lumped disturbance in real time. Through discretization, the disturbance estimate is obtained by iteratively updating the data in each sampling period using the current sampled value and the observation from the previous period. .

5. The model-free robust predictive control method for a three-level SNPC inverter according to claim 4, characterized in that, Based on the position of the reference voltage vector in the spatial vector diagram, a set of candidate voltage vectors is selected. The optimal combination of two vectors is evaluated using a cost function, and its duration is calculated. Specifically, this includes: Based on the position of the reference voltage vector in the space voltage vector diagram of the three-level SNPC inverter, the sector and sub-triangle region where the reference voltage vector is located are determined. The large vector, small vector and zero vector related to the sub-triangle region are combined into a candidate voltage vector set. Several candidate line segments composed of two voltage vectors are constructed with the reference voltage vector as the target. A preset cost function is selected, and the error between the reference voltage vector and the equivalent output voltage vector is used as the index to calculate the cost value from the reference voltage vector to the midpoint of each candidate line segment. The line segment with the smallest cost function value is selected, and the voltage vectors at its two ends are used as the two active voltage vectors in this sampling period. Based on the geometric relationships of the spatial vector diagram or the optimization conditions that make the equivalent output voltage vector approximate the reference voltage vector, solve for the action time of the two active voltage vectors. .

6. The model-free robust predictive control method for a three-level SNPC inverter according to claim 5, characterized in that, Based on the DC side midpoint voltage deviation, a redundant small vector is selected to generate a PWM drive signal for modulation control, specifically including: The midpoint voltage deviation is calculated based on the upper capacitor voltage VP and the lower capacitor voltage VN on the DC side. When there is a redundant small vector, under the premise of meeting the current control target, different redundant small vectors are selected to participate in the dual vector synthesis according to the sign of the midpoint voltage deviation. Based on the two selected active voltage vectors and their active times, PWM drive signals for the three-level SNPC inverter common module and the three-phase bridge arms are generated to complete model-free predictive control for one sampling cycle.

7. A model-free robust predictive control system for a three-level SNPC inverter, characterized in that, A method for implementing a model-free robust predictive control of a three-level SNPC inverter as described in any one of claims 1-6 includes: The sampling module is used to acquire the three-phase output current, three-phase grid-side voltage and DC-side upper and lower capacitor voltage in real time, and to complete the Clark transformation to output the current and voltage components in the two-phase stationary coordinate system. The observation module is used to receive the current component, voltage component and inverter output voltage, and to estimate the extended state in real time based on the full-order sliding mode observer constructed based on the hyperlocal model, and output the current observation value and the lumped disturbance estimate value. The reference voltage calculation module is used to receive current commands, current observations, and lumped disturbance estimates, and uses an extrapolation algorithm to predict the future reference current and calculate the reference voltage vector. The modulation module is used to filter the set of candidate voltage vectors based on the position of the reference voltage vector, construct candidate line segments, select the optimal applied voltage vector based on the cost function, and solve its application time. The control module is used to convert the selected applied voltage vector and its applied time into PWM drive signals for the common module and three-phase bridge arm of the three-level SNPC inverter, and to select redundant small vectors based on the midpoint voltage deviation to achieve midpoint potential balance control.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is called by the processor, it executes a model-free robust predictive control method for a three-level SNPC inverter as described in any one of claims 1-6. The processor is a digital signal processor, a microcontroller, or a field-programmable gate array, used to realize real-time predictive control of the three-level SNPC inverter.

Images