Predictive compensation quasi-sliding mode control method for three-phase four-leg three-level inverter

Abstract

The present application relates to the field of power electronics technology, and in particular to a predictive compensation quasi-sliding mode control method for a three-phase four-bridge-arm three-level inverter, comprising establishing a continuous state space model; discretizing the continuous state space model, constructing a discrete state space model and defining the output error of the output voltage and its derivative, defining a sliding mode function based on the output error of the output voltage and its derivative under the theory of quasi-sliding mode control; constructing a prediction model and predicting the sliding mode function value at the next moment, introducing an extended model to extract the model error, and correcting the defined sliding mode function based on the deviation between the predicted value and the model error; constructing a repetitive controller to extract the periodic error component and generate a compensation signal; fusing the compensation signal with the equivalent control quantity constructed based on the extended model to form the final control input for the drive system. The present application can greatly suppress chattering and optimize steady-state performance, ensuring system robustness while taking into account fast dynamic response.

Description

Technical Field

[0001] The present application relates to the field of power electronics technology, and in particular to a predictive compensation quasi-sliding mode control method for a three-phase four-leg three-level inverter. Background Art

[0002] With the increasing demand for high performance, high efficiency, and high power density in power electronics systems, traditional three-phase, three-leg, two-level inverters are increasingly unable to meet the increasingly complex power quality and neutral point management requirements. The three-phase, four-leg, three-level inverter, due to its structural flexibility and unique advantages in neutral current management, has become a key research direction for the next generation of high-performance inverter systems. This topology introduces a dedicated fourth leg to specifically address circulating neutral current, effectively suppressing neutral point voltage fluctuations while maintaining system symmetry and output quality. The three-level structure also offers higher voltage withstand capability, higher efficiency, lower harmonic distortion, and reduced electromagnetic interference, making it a promising candidate for high-voltage, high-power density applications.

[0003] Sliding mode control (SMC) is widely used to address common inverter control issues such as interference and modeling errors due to its robustness and dynamic response. However, the widespread chattering phenomenon in SMC has become a major obstacle to its practicality. This phenomenon, primarily caused by discontinuous sign functions in the control law, results in frequent system state switching near the sliding mode surface, resulting in high-frequency oscillations.

[0004] To mitigate chattering in SMC, various improvement strategies have been proposed. For example, the boundary layer method uses a transition region of thickness Φ near the sliding surface to smooth discontinuities in the sign function with a continuous function, thereby reducing chatter intensity. However, if Φ is not set properly, this method can easily cause steady-state errors and make it difficult to ensure system accuracy. The superwarp algorithm replaces the sign function with a continuous reaching law, achieving both chatter reduction and improved steady-state performance in some cases. However, its convergence and stability lack rigorous theoretical support, and the weight parameter setting relies on empirical experience, making it difficult to adapt to complex operating conditions. Hysteresis modulation technology introduces switching logic with a fixed bandwidth h to limit the sliding mode signal, effectively suppressing high-frequency oscillations. However, this method causes the switching frequency to vary with the system state, making it difficult to implement in hardware. Furthermore, the selection of h also directly affects the steady-state deviation. Furthermore, observer-based sliding mode control estimates the sliding surface state to suppress unmodeled dynamics and disturbances, improving control robustness. However, in engineering applications, this requires extremely high accuracy in observer modeling and estimation, resulting in poor fault tolerance.

[0005] In summary, while existing sliding mode control strategies have alleviated controller chattering to a certain extent, they still struggle to simultaneously meet the stringent requirements of multiple performance dimensions in practical engineering applications. Specifically, existing methods often introduce steady-state errors in the process of mitigating chattering, making it difficult to maintain high-precision steady-state performance while ensuring continuity. Furthermore, to improve system robustness and response speed, some control strategies incorporate observer structures or empirical weight adjustment mechanisms. While these can improve dynamic characteristics, they also introduce significant parameter sensitivity and implementation complexity. This results in control systems prone to dynamic response hysteresis or decreased robustness when faced with parameter disturbances or load changes. Therefore, in practical applications, existing sliding mode control methods struggle to achieve the dual goals of maximally suppressing chattering and optimizing steady-state performance, nor do they achieve both rapid dynamic response and robustness. Summary of the Invention

[0006] This application provides a predictive compensation quasi-sliding mode control method for a three-phase four-leg three-level inverter, which can greatly suppress chattering and optimize steady-state performance, ensuring system robustness while also taking into account fast dynamic response. This application provides the following technical solutions:

[0007] In a first aspect, the present application provides a predictive compensation quasi-sliding mode control method for a three-phase four-leg three-level inverter, the method comprising:

[0008] A continuous state-space model of multi-variable coupling is established based on the topology of a three-phase four-leg three-level inverter.

[0009] The forward Euler approximation method is used to discretize the continuous state space model and construct a discrete state space model that can be used by the digital controller;

[0010] The output error of the output voltage and its derivative are defined using a discrete state space model, and a sliding mode function is defined based on the output error of the output voltage and its derivative under the introduction of the quasi-sliding mode control theory;

[0011] Construct a prediction model and predict the sliding mode function value at the next moment. Introduce an extended discrete state space model to extract the model error. Modify the defined sliding mode function based on the deviation between the predicted value and the model error.

[0012] A repetitive controller is constructed based on the prediction error of the modified sliding mode function to extract the periodic error component and generate a compensation signal.

[0013] The compensation signal output by the repetitive controller is fused with the equivalent control quantity constructed based on the extended discrete state space model to form the final control input for the drive system.

[0014] In a specific implementation scheme, establishing a multi-variable coupled continuous state space model based on the topology of a three-phase four-leg three-level inverter includes:

[0015] The topology of the three-phase four-bridge-arm three-level inverter includes three-phase main bridge arms and a neutral bridge arm. The three-phase main bridge arms are the A-phase bridge arm, the B-phase bridge arm, and the C-phase bridge arm respectively. The neutral bridge arm is used to connect the neutral point of the load to provide a zero-sequence current loop; each bridge arm is composed of four power switching devices, and three output level states are achieved through different conduction combinations, corresponding to the positive half-bus voltage, the neutral point voltage, and the negative half-bus voltage respectively. Upper and lower capacitors connected in series are configured on the DC side to form a voltage divider structure; the output end of each bridge arm is connected to the load through a filter inductor and a filter capacitor; the neutral bridge arm is connected to the neutral point of the load via a connecting inductor.

[0016] In a specific possible implementation scheme, establishing a multi-variable coupled continuous state space model based on the topology of a three-phase four-leg three-level inverter further includes:

[0017] Determine the corresponding relationship between the switch state and the output level of each bridge arm, and obtain the voltage value of the output voltage of each bridge arm relative to the DC midpoint;

[0018] Under the assumption that the neutral point voltage is balanced, the load line voltage is determined based on the relationship between the output voltages of each bridge arm. Based on the output structure of the three-phase bridge arm and the connection structure of the neutral bridge arm, the differential relationship between the three-phase output voltage and the filter inductor current is established.

[0019] The current balance relationship between the three-phase filter current and the output voltage is established based on the filter capacitor structure; the three-phase voltage and filter current are converted to the αβγ coordinate system through stationary coordinate transformation, and the equivalent filter inductance and filter capacitor parameters are defined;

[0020] A continuous-time state-space model is established using the voltage and current variables and equivalent filter element parameters converted to the αβγ coordinate system.

[0021] In a specific embodiment, the use of the forward Euler approximation method to discretize the continuous state space model to construct a discrete state space model that can be used by the digital controller includes:

[0022] The differential equations of the filter inductor current and filter capacitor voltage in the continuous-time state-space model are discretized based on the forward Euler approximation method, and the evolution of the state in adjacent sampling periods is expressed in differential form.

[0023] The time derivatives in the continuous equations are replaced by the corresponding differential terms to derive the discrete-time state transfer matrix, control input matrix, and corresponding disturbance compensation terms.

[0024] A discrete-time state space model is constructed by utilizing the state transfer matrix, the control input matrix and the disturbance compensation term, and combining the filter inductor current, the filter capacitor voltage, the control input voltage and the load current variables.

[0025] In a specific embodiment, the output error of the output voltage and its derivative are defined by using a discrete state space model, and the sliding mode function is defined based on the output error of the output voltage and its derivative under the theory of quasi-sliding mode control, including:

[0026] Using the discrete state space model for each coordinate component Define the tracking error of the output voltage and its derivative error;

[0027] Select two state variables and , the output error of the output voltage is defined as follows:

[0028] ;

[0029] in, represents the output voltage reference value in the αβγ reference coordinate system, is the actual output voltage value; the derivative of the error is defined as follows:

[0030] ;

[0031] in, and denote the derivatives of the reference voltage and output voltage respectively, is the filter capacitor, and They are load current and output current respectively; based on the above two state variables, construct a sliding mode function as follows:

[0032] ;

[0033] in, is the sliding mode coefficient;

[0034] Introducing the quasi-sliding mode control theory, defining the The sliding mode function value of the sampling period satisfies:

[0035] ;

[0036] in, is the sliding mode function value at the kth sampling moment, is the error bound of the sliding mode function allowed in quasi-sliding mode control.

[0037] In a specific possible implementation scheme, the prediction model is constructed and the sliding mode function value at the next moment is predicted, an extended discrete state space model is introduced to extract the model error, and the sliding mode function defined based on the deviation between the predicted value and the model error is corrected, including:

[0038] The forward Euler approximation is used to predict the system model, and the predicted state variables are substituted into the sliding mode function to obtain the sliding mode function prediction value at the next moment. :

[0039] Introducing the extended discrete state space model to define the sliding mode function error term ;

[0040] The modified sliding mode function is:

[0041] .

[0042] In a specific implementation scheme, constructing a repetitive controller based on the prediction error of the modified sliding mode function, extracting the periodic error component and generating a compensation signal includes:

[0043] Based on the error between the predicted value and the actual value of the modified sliding mode function, a repetitive controller with a periodic memory mechanism is constructed to extract the periodic error component and generate a compensation signal.

[0044] The input of the repetitive controller is the difference between the predicted value and the actual value of the modified sliding mode function. The difference is expressed as the transfer relationship between the control input disturbance and the sliding mode function error under linear approximation through small signal modeling, and the output expression of the compensation signal is determined according to the transfer coefficient;

[0045] The repetitive controller adopts a z-domain structure with a time delay link. A zero-phase low-pass filter that satisfies the normalization constraint is introduced into the internal model to construct a periodic error extraction path. An additional compensation filter is used to attenuate high-frequency interference.

[0046] The compensation signal output by the repetitive controller is subjected to time delay compensation at the end through a phase advance element, thereby forming a feedback channel that matches the equivalent control quantity.

[0047] In a specific implementation scheme, fusing the compensation signal output by the repetitive controller with the equivalent control quantity constructed based on the extended discrete state space model to form the final control input for the drive system includes:

[0048] The compensation amount of the repeat controller output Equivalent control quantity based on extended discrete state space model The fusion forms a complete control law, and the expression of the control law is as follows:

[0049] ;

[0050] in, For the control cycles are applied to the final control input of the system.

[0051] In a second aspect, the present application provides an electronic device comprising a processor and a memory; the memory stores a program, which is loaded and executed by the processor to implement a predictive compensation quasi-sliding mode control method for a three-phase four-bridge-arm three-level inverter as described in the first aspect.

[0052] In a third aspect, the present application provides a computer-readable storage medium, which stores a program. When the program is executed by a processor, it is used to implement a predictive compensation quasi-sliding mode control method for a three-phase four-bridge-arm three-level inverter as described in the first aspect.

[0053] In summary, the beneficial effects of this application include at least:

[0054] (1) By introducing the quasi-sliding mode control theory (QSMC), this application no longer relies on the symbolic function used in traditional sliding mode control. Instead, it constructs a quasi-sliding mode region, combines it with the prediction model of the sliding mode function, estimates the future value of the sliding mode function based on the current system state and the prediction model, and introduces a model error compensation term to dynamically correct the control input in real time. This method effectively avoids the high-frequency chattering problem caused by traditional symbolic functions. In the control law, QSMC not only enhances the system's adaptability to modeling uncertainty and external disturbances by accurately modeling the impact of disturbances, analyzing the system modeling error characteristics, and introducing a prediction error correction mechanism, but also avoids excessive switching control during the error convergence process, thereby making the system output smoother, significantly suppressing chattering and improving steady-state performance.

[0055] (2) This application further introduces the idea of ​​predictive control into the sliding mode control architecture, using an extended discrete state space model to predict the system state at future moments, and constructing a feedforward control quantity based on the predicted sliding mode function value, thereby realizing forward-looking control and regulation of the system's dynamic behavior. Unlike traditional passive response control, this application uses a controller to predict the evolution trend of the sliding mode function in the next step, and then calculates the equivalent control component and combines it with the error compensation mechanism constructed by the repetitive controller to achieve rapid driving of the system state and error reduction. In this process, the system still maintains good robustness to factors such as modeling errors and load disturbances, thereby significantly improving the dynamic response speed while ensuring the robustness and tracking accuracy of the system.

[0056] A quasi-sliding mode control strategy is introduced. By constructing a time-varying sliding mode surface and introducing a predictive model of the sliding mode function, the model error is directly quantified and compensated. In particular, errors caused by inaccurate modeling, circuit parameter disturbances, and DC bus midpoint oscillations are taken into account. Error suppression is achieved without using a sign function, significantly reducing the jitter phenomenon and improving the steady-state performance of the system. Secondly, to improve the system's responsiveness to dynamic disturbances and control accuracy, this application introduces the concept of predictive control into the sliding mode control structure. By forward-predicting and calculating the control quantity based on an extended discrete state space model, active regulation of the future evolution of the state variables is achieved, maintaining good robustness while improving the system's dynamic response speed.

[0057] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application and to implement it in accordance with the contents of the specification, the following is a detailed description of the preferred embodiments of the present application in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 1 is a topological diagram of a three-phase four-bridge-arm three-level voltage-source inverter in an embodiment of the present application.

[0059] Figure 2 It is a use case diagram of the system state variable trajectory in the embodiment of the present application.

[0060] Figure 3 It is a structural diagram of the repetitive controller in an embodiment of the present application.

[0061] Figure 4 It is a flowchart of the overall implementation method of the system in the embodiment of the present application.

[0062] Figure 5 This is a comparison chart of the interruption time of four evaluation algorithms in the verification experiment of the embodiment of this application.

[0063] Figure 6 This is an experimental waveform diagram of the steady-state changes of the four evaluation algorithms under linear load in the verification experiment of the embodiment of the present application.

[0064] Figure 7 This is an experimental waveform diagram of the dynamic changes of the four evaluation algorithms under linear load in the verification experiment of the embodiment of this application.

[0065] Figure 8 This is a harmonic spectrum diagram of the output voltage of the four evaluation algorithms under linear load in the verification experiment of the embodiment of the present application.

[0066] Figure 9 This is an experimental waveform diagram of the steady-state changes of the four evaluation algorithms under unbalanced load in the verification experiment of the embodiment of the present application.

[0067] Figure 10This is an experimental waveform diagram of the dynamic changes of the four evaluation algorithms under unbalanced load in the verification experiment of the embodiment of the present application.

[0068] Figure 11 This is a harmonic spectrum diagram of the output voltage under unbalanced load of the four evaluation algorithms in the verification experiment of the embodiment of the present application.

[0069] Figure 12 This is a voltage waveform diagram of the DC bus midpoint under load imbalance conditions in the verification experiment of the embodiment of the present application.

[0070] Figure 13 This is an experimental waveform diagram of the steady-state changes of the four evaluation algorithms under nonlinear load in the verification experiment of the embodiment of the present application.

[0071] Figure 14 Schematic diagram of the circuit of the nonlinear load in the verification experiment of the embodiment of the present application.

[0072] Figure 15 This is a harmonic spectrum diagram of the output voltage under nonlinear load of the four evaluation algorithms in the verification experiment of the embodiment of the present application.

[0073] Figure 16 It is a broken line diagram of the sensitivity analysis of the parameters of the four evaluation algorithms in the verification experiment of the embodiment of the present application.

[0074] Figure 17 It is a block diagram of an electronic device for predictive compensation quasi-sliding mode control of a three-phase four-leg three-level inverter in an embodiment of the present application. DETAILED DESCRIPTION

[0075] The following embodiments are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0076] Optionally, the present application uses the predictive compensation quasi-sliding mode control method for a three-phase four-bridge-arm three-level inverter provided in each embodiment as an example for explanation in an electronic device, where the electronic device is a terminal or a server. The terminal can be a computer, a tablet computer, etc. This embodiment does not limit the type of electronic device.

[0077] Reference Figure 1 , is a flow chart of a predictive compensation quasi-sliding mode control method for a three-phase four-leg three-level inverter provided by an embodiment of the present application. The method includes at least the following steps:

[0078] Step S101 : establishing a multi-variable coupled continuous state space model based on the topology of a three-phase four-leg three-level inverter.

[0079] In step S101, based on Figure 1Based on the topology of the three-phase, four-leg, three-level voltage-source inverter shown in Figure 1, a continuous-time state-space model is established that accurately describes the operating characteristics of the three-phase, four-leg, three-level voltage-source inverter. This model must not only fully reflect the inverter's physical circuit structure and control state but also include the coupling relationships between different electrical variables, providing an accurate and usable mathematical foundation for subsequent controller design.

[0080] Specifically, the topology of the three-phase four-bridge-arm three-level voltage-source inverter includes three-phase main bridge arms A, B, and C and an additional neutral bridge arm N. Each bridge arm consists of four switching devices, which can achieve three-level voltage output, thereby meeting the high-quality energy conversion requirements under complex load conditions. The DC side of the inverter forms a voltage divider structure through two upper and lower series capacitors, which are represented by the upper voltage and lower side voltage , under the premise of normal system operation and voltage balance, we can get: , Indicates the total voltage of the DC bus.

[0081] In each phase branch, a filter inductor L and a filter capacitor C are connected to the output end of the three-phase bridge arm to filter out the high-frequency harmonics caused by PWM control and improve the output waveform quality of the system. The fourth bridge arm (i.e., the neutral bridge arm) is used to provide a return path for the zero-sequence component of the load current. The point N is connected to the neutral point of the load through the inductor. Connected to the neutral point n of the load to achieve stable power supply under unbalanced load conditions. The filter inductor currents of each phase are recorded as , the voltages across the corresponding filter capacitors are , the current flowing through the neutral bridge arm inductor is recorded as ;

[0082] In implementation, we first need to clarify the impact of the switching state of each bridge arm on its output level. Due to the three-level topology, each bridge arm has three output states: 1, 0, and -1, indicating that the bridge arm output is positive half voltage, neutral potential, and negative half voltage. These three states are determined by the four power switches inside the bridge arm. 、 、 、 This is achieved through different conduction combinations. To this end, the switching function is introduced , used to describe the output level of each bridge arm, is defined as follows:

[0083] ;

[0084] in, , that is, each bridge arm forms a three-level output through two upper tubes and two lower tubes, forming three different level outputs according to the conduction state of the switch. Since each of the four bridge arms has three possible outputs, the overall inverter output combination is Different switch combinations.

[0085] Then, assuming that the neutral point potential (i.e., the midpoint potential of each capacitor) has been effectively balanced through the control strategy, the output voltage of the midpoint of each bridge arm relative to the neutral point (NP) can be calculated. , expressed as a linear transformation of the switching function, namely:

[0086] ;

[0087] Next, calculate the line voltage of the three phases of the load (i.e., the voltage relative to the load neutral point n). Since the three-phase load shares a neutral point, the output voltage of the inverter three-phase bridge arm relative to the inverter bridge arm neutral point O must be subtracted from the output voltage of the N bridge arm to obtain the actual line voltage applied to the load. 、 、 as follows:

[0088] ;

[0089] in, 、 、 are the voltages of the A, B, and C phase bridge arms relative to the neutral bridge arm N, 、 、 、 are the output voltages of bridge arms A, B, C, and N relative to point O, 、 、 、 are the current output levels of bridge arms A, B, C, and N. The above conversion shows that although the three phases A, B, C, and N are independently controlled, they all share the reference potential of the fourth bridge arm N, so there is an implicit coupling relationship between the three-phase output voltages.

[0090] To further describe the dynamic electrical relationship between the inverter output and the load, Kirchhoff's voltage law is used to analyze each phase in the output LC filter circuit of each phase. The differential relationship between the output voltage and the rate of change of the filter inductor current is obtained as follows:

[0091] ;

[0092] in 、 、 They are the voltages of the load terminals of bridge arms A, B, and C relative to the load neutral point n, and are also the actual load voltages output by the filter. 、 、 are the currents in the filter inductors of bridge arms A, B, and C respectively. The above formula shows that since the three phases share the neutral inductor , the voltage compensation term is the same, which makes the voltage between the three phases also coupled to the neutral current Similarly, according to Kirchhoff's Current Law (KCL), the relationship between the filter current and the load current can be obtained as follows:

[0093] ;

[0094] in, 、 、 They are the load currents of bridge arms A, B, and C, that is, the actual current flowing into the load. In addition, the total current balance between the three phases and the neutral point satisfies:

[0095] ;

[0096] As can be seen from the above formula, the three-phase output voltage and current are intertwined in time and have a serious coupling relationship, making it difficult to effectively decouple or control them using traditional modeling.

[0097] Therefore, in order to achieve separation in control, this application further adopts a stationary coordinate system transformation, that is, transforming from the abc coordinate system to the stationary αβγ reference coordinate system, mapping the coupled three-phase variables to a set of orthogonal components to obtain a mathematical model that is easier to independently control. Specifically, the transformation matrix from the abc coordinate system to the αβγ reference coordinate system is defined as for:

[0098] ;

[0099] Through this coordinate transformation, the three-phase physical quantities are projected onto the three orthogonal axes αβγ. After the transformation, the three-phase coupling relationship is clearly split into two components: the α and β components are used to control the symmetrical components, and the γ component specifically corresponds to the zero-sequence current (introduced by the fourth bridge arm), so its dynamic characteristics can be modeled independently. Through this coordinate transformation, the filter capacitor voltage and inductor current are mapped to 、 、 and 、 、 Its dynamic relationship can be expressed as the following differential equations, and the voltage dynamic equation is as follows:

[0100] ;

[0101] The dynamic equation of the inductor current is as follows:

[0102] ;

[0103] in, 、 、 They are the three axes of the stationary orthogonal coordinate system. From the above decoupling results, we can see that the original complex coupling relationship between the three-phase voltage / current has been transformed into three sets of independent control components that do not affect each other, namely αβγ components. In particular, the neutral bridge arm inductance is introduced into the γ component. , resulting in its equivalent inductance being , thus truly reflecting the impact of the system neutral current on the system coupling. Finally, the above two differential equations can be further unified into the following general formula to facilitate the unified modeling and analysis of the subsequent control strategy:

[0104] ;

[0105] in, , After transformation Component output voltage, , After transformation The target voltage signal of the component, and the equivalent inductance satisfies:

[0106] , ;

[0107] in, 、 、 are the equivalent inductances of the α, β, and γ components, respectively. Through the above modeling process, coordinate decoupling and parameter reconstruction of the three-phase system state variables were successfully achieved, greatly simplifying the state feedback strategy and modulation logic in subsequent controller design. This completes the continuous state-space model derived from the physical circuit through time-domain modeling and coordinate transformation, laying the theoretical foundation for subsequent discretization and controller design.

[0108] Step S102: discretize the continuous state space model using the forward Euler approximation method to construct a discrete state space model that can be used by the digital controller.

[0109] In step S102, a continuous time state space model reflecting the actual operating characteristics has been established before, and decoupling is completed through coordinate transformation, so that the three-phase output of the system can be modeled separately with three components αβγ. This continuous model clearly describes the filter inductor current and filter capacitor voltage However, in digital controllers, the update of the system state must be based on discrete time steps, so it is necessary to convert the continuous state space model into a discrete model.

[0110] Specifically, this step uses the forward Euler method to perform first-order approximate discretization on the differential model. The forward Euler method is a common numerical integration method. When is small enough, the dynamic behavior of the continuous system can be well approximated. Assume that the current moment is sampling points, the time derivative is approximately in the differential form:

[0111] ;

[0112] Based on this approximation, the continuous state equation in step S101 (i.e., the coupled dynamic model of the filter capacitor voltage and the inductor current) is converted into the following discrete time form:

[0113] ;

[0114] in, Indicates the The current in the filter inductor at the sampling moment, Indicates the The voltage across the filter capacitor at each sampling moment; Indicates the control input voltage in this direction, that is, the reference voltage that the digital controller expects to output; Indicates the actual current flowing to the load on this component; 、 、 are the state transfer matrix, input control matrix and load disturbance compensation term, which are defined as follows:

[0115] ;

[0116] This discrete model maintains the state variable structure of the original continuous model: the minimum state unit in each direction is composed of "inductor current + capacitor voltage." The state transition matrix describes the intrinsic connections between the system's internal variables, the input control matrix describes the impact of the controller's control inputs on the state variables, and the load disturbance compensation term reflects the impact of the load current on the system's dynamic behavior. Ultimately, the resulting discrete state space expression is a standard second-order difference equation, making it easy to directly use in subsequent controller design, enabling precise tracking and regulation of the system state in each direction.

[0117] Step S103 : defining an output error of the output voltage and its derivative using a discrete state space model, and defining a sliding mode function based on the output error of the output voltage and its derivative by introducing the theory of quasi-sliding mode control.

[0118] In step S103, first, based on the discrete state space model obtained in step S102, for each coordinate component The tracking error of the output voltage and its derivative error are defined to construct the sliding mode function, and on this basis, the mathematical form of quasi-sliding mode control (QSMC) is introduced.

[0119] Specifically, in order to achieve precise control of the inverter output voltage, two state variables are first selected: and To describe the voltage error and its changing trend, the output error of the output voltage is defined as follows:

[0120] ;

[0121] in, represents the output voltage reference value in the αβγ reference coordinate system, is the actual output voltage value.

[0122] Furthermore, the derivative of the error is defined as follows:

[0123] ;

[0124] in, and denote the derivatives of the reference voltage and output voltage respectively, is the filter capacitor, and are the load current and output current respectively.

[0125] Based on the above two state variables, the sliding mode function is constructed using the traditional linear form as follows:

[0126] ;

[0127] in, is a sliding mode coefficient greater than 0, which is used to adjust the position of the sliding surface and the stability of the system. To ensure the stability of the closed-loop system, according to Lyapunov stability theory, the following inequality must be satisfied:

[0128] ;

[0129] in, is the sliding mode function The time derivative, is a positive scalar that adjusts the sliding mode speed and anti-disturbance capability. However, in traditional sliding mode control (SMC), the sign function is used to implement switching control, but due to the switching frequency limitation and the non-ideality of the actual system, it often leads to chattering phenomenon, such as Figure 2In order to alleviate this problem, the quasi-sliding mode control (QSMC) theory is introduced and the definition of The sliding mode function value of the sampling period satisfies:

[0130] ;

[0131] in, is the sliding mode function value at the kth sampling moment, is the error bound of the sliding mode function allowed in quasi-sliding mode control (QSMC), which is a positive number greater than zero. This constraint indicates that the trajectory of the system state is allowed to fluctuate within a bounded neighborhood around the sliding surface, without strictly reaching zero, thereby avoiding frequent switching and significantly reducing chattering, such as Figure 2 In summary, step S103 defines the output voltage error and its derivative based on the discrete model, and constructs the sliding mode function in combination with the quasi-sliding mode control concept, providing a theoretical basis for the design of the subsequent sliding mode control algorithm.

[0132] Step S104: construct a prediction model and predict the sliding mode function value at the next moment, introduce an extended discrete state space model to extract the model error, and modify the defined sliding mode function based on the deviation between the predicted value and the model error.

[0133] In step S104, first, in order to ensure the stability of the quasi-sliding mode control, a reaching law needs to be designed to keep the absolute value of the sliding mode function within the allowable error range. The specific reaching law requirements are as follows:

[0134] like In the quasi-sliding mode state, Should be kept in quasi-sliding mode state. Not in quasi-sliding mode state, Should be compared Closer to the sliding mode state and does not cross this area.

[0135] In order to realize the above convergence law, it is necessary to accurately predict the sliding mode function at the next sampling moment The application draws on the idea of ​​deadbeat control (DB) in QSMC and uses forward Euler approximation to predict the system model. The constructed prediction model is as follows:

[0136] ;

[0137] in, and They are The inductor current and output voltage of a sampling period, is the sampling period, and They are The output current and control input voltage of the sampling period; since the output current The change in a sampling period is small, so , that is, using the current sample value to approximate the predicted value. Substitute the above predicted state variables into the sliding mode function definition to obtain the sliding mode function predicted value at the next moment:

[0138] ;

[0139] in, Output voltage reference value The above expression can be used to solve the system control input However, due to the errors in the model, it cannot be used directly. Considering the various effects of parameter uncertainty, measurement error, modeling deviation and disturbance in the actual system, if these effects are ignored, the sliding mode function prediction value will deviate from the true value, which will affect the correct implementation of the reaching law. To this end, this application corrects the state prediction of the system based on the extended model. The extended model is as follows:

[0140] ;

[0141] in, Represents the natural evolution relationship matrix of the system state (filter current and output voltage) within a sampling period, A column vector representing the direct action path and proportional relationship of the control input on the system state variables, The coupling coefficient vector represents the degree of influence of the external load current on the system state, Represents the uncertainty part of the state evolution relationship matrix due to parameter errors and imperfect modeling, It represents the modeling error caused by factors such as device non-ideality or delay in the control input path. It represents the deviation term or uncertainty term in modeling the impact of load-side disturbance on system state. Indicates the deviation of the DC bus voltage, Indicates the interference from the DC bus neutral point (point O), is the external interference term, Representation and column vector Orthogonal component vectors.

[0142] In order to quantitatively describe the model error, the model error term is further extracted independently to obtain:

[0143] ;

[0144] in,

[0145] ;

[0146] It is worth noting that the main components of the modeling error are usually the grid fundamental frequency and its higher harmonics, which appear as periodic sinusoidal disturbances. Some offsets may also come from DC offsets caused by system delays and sampling errors. In order to accurately characterize the impact of errors on the sliding mode function prediction, the sliding mode function error term is defined as:

[0147] ;

[0148] Therefore, the modified sliding mode function is:

[0149] ;

[0150] in, represents the extended sliding mode function defined after accounting for the error, which is the new sliding mode approach target. Therefore, unlike the fixed sliding surface in traditional SMC, the sliding surface in the QSMC strategy is a time-varying sliding surface that changes dynamically with disturbances and parameter uncertainties. This allows the control system to adaptively adjust its control target and more effectively respond to modeling errors and external disturbances.

[0151] Step S105 : construct a repetitive controller based on the prediction error of the corrected sliding mode function, extract the periodic error component and generate a compensation signal.

[0152] In step S105, a sliding mode function is constructed that takes into account the influence of model error. After that, however, due to the existence of model mismatch, interference and unmodeled dynamics in the system, the control effect of the actual system may still be affected. In order to further reduce the prediction error of the sliding mode function, this step introduces a repetitive controller as a compensation mechanism to compensate the equivalent control quantity constructed based on the prediction model. The basic idea of ​​the repetitive controller is to introduce a feedback mechanism with periodic memory to enable the system to effectively suppress periodic interference, especially the fundamental frequency interference and its integer multiple frequency harmonic components. Its control structure in the z domain is as follows Figure 3 As shown, is the number of samples in one cycle of the output voltage, and is a low-pass filter, is a phase-leading element, is the accused, is the output of the controller.

[0153] In the implementation, the input of the repetitive controller is the sliding mode function prediction value With actual value In order to simplify the controller implementation and accurately establish the quantitative relationship between input and output, the small signal modeling method is adopted. The system is linearized near , and the transfer relationship between the disturbance component and the control input disturbance is derived:

[0154] ;

[0155] set up:

[0156] ;

[0157] but:

[0158] ;

[0159] in, is the expression of the sliding mode function error in the z domain, is the output signal of the repetitive controller, Represents the z-transform variable, corresponding to the delay operation of one control cycle, is the gain factor between the sliding mode function error and the controller output.

[0160] To enhance the stability of the system, the internal model introduced in the repetitive controller uses a zero-phase low-pass filter of the following form:

[0161] ;

[0162] The filter satisfies the normalization constraint:

[0163] ;

[0164] in, 、 are all real coefficients used to determine the weights of filter terms of each order. All coefficients are non-negative. Indicates that the current signal is pushed forward Sampling period, Indicates that the current signal is pushed forward Sampling period, is the order of the filter. Its function is to weaken the unstable gain existing in the ideal repetitive control and improve the robustness of the system to model errors and high-frequency interference. In addition, the compensation filter used is usually designed as a low-pass second-order system to further smooth the error response and avoid false operations caused by high-frequency components. In actual control, since the input of the repetitive controller is the sliding mode function error of the previous control cycle, there is a delay of one sampling cycle. In order to ensure the control effect, a phase advancer needs to be introduced at the end of the controller for corresponding compensation. Finally, the output of the repetitive controller is It will be used to compensate for the part of the equivalent control quantity that is omitted due to modeling errors, thereby improving the accuracy and stability of the entire control system.

[0165] Step S106: Fusing the compensation signal output by the repetitive controller with the equivalent control quantity constructed based on the extended discrete state space model to form a final control input for the drive system.

[0166] In step S106, the output compensation of the repetitive controller is completed. After the calculation of , this step will compare it with the equivalent control quantity constructed based on the extended discrete state space model The fusion forms a complete control law for direct drive system. The expression of the control law is as follows:

[0167] ;

[0168] in, For the The final control input applied to the system is is the equivalent control term derived from the system prediction model under quasi-sliding mode conditions, is the output of the repetitive controller, which is used to compensate for system model errors and disturbances. The above structure reflects the two-level control concept of this control strategy: the first level obtains equivalent control quantity based on model prediction to ensure the basic response of the system; the second level uses the repetitive controller to perform feedback correction to enhance the robustness and anti-interference ability of the system. The fused control input This input is directly applied to the target control object, the power control module on the inverter side. This input not only ensures the system approaches the sliding mode surface but also significantly reduces steady-state errors caused by periodic disturbances or model errors, thereby achieving higher-precision and more robust control. This completes the control input design process, providing the foundation for the subsequent actual control signal output and serving as the basis for driving the next stage of the execution unit.

[0169] In summary, combined with Figure 4 First, to address the chattering problem caused by the introduction of sign functions in traditional sliding mode control, this application introduces a quasi-sliding mode control strategy. By constructing a time-varying sliding surface and introducing a predictive model of the sliding mode function, the model error is directly quantified and compensated. In particular, the errors caused by inaccurate modeling, circuit parameter disturbances, and DC bus midpoint oscillation are taken into account. This achieves error suppression without using sign functions, significantly reduces chattering, and improves the steady-state performance of the system. Secondly, to improve the system's responsiveness to dynamic disturbances and control accuracy, this application introduces the concept of predictive control into the sliding mode control structure. By forward-predicting and calculating the control quantity based on the extended discrete state space model, it achieves active regulation of the future evolution of the state variables, improving the system's dynamic response speed while maintaining good robustness.

[0170] Furthermore, to demonstrate the technical benefits of this application and verify the superiority and effectiveness of the proposed RQSMC strategy, various experimental evaluations were conducted on a 3kW 3P4L-3L inverter prototype. The test platform utilized a Texas Instruments TMS320F28374S digital signal processor (DSP) to execute the control framework. This DSP integrates multiple high-speed analog-to-digital (A / D) conversion interfaces, high-resolution enhanced pulse-width modulation (PWM) channels with programmable parameters, and auxiliary peripheral modules tailored for advanced power electronics applications. Combined with the processor's high clock frequency, these integrated features make it particularly suitable for advanced power electronics implementations. Key system parameters are systematically summarized in Table 1 below.

[0171]

[0172] In the verification experiments, four control strategies were implemented and compared on the experimental platform. Each strategy is defined as follows: Traditional PI control: Traditional PI control follows a classic dual-loop control structure, with the outer loop controlling the output voltage and the inner loop regulating the inductor current. Specifically, to ensure voltage balance, the three-phase voltages and currents are decomposed into instantaneous positive-sequence, negative-sequence, and zero-sequence components, which are independently controlled in their respective reference frames. Traditional sliding mode control (SMC): The state variables selected for the sliding mode control are the same as those in the proposed algorithm, and a boundary layer-based reaching law is employed to suppress chattering. Quasi-sliding mode control (QMPC): QSMC is a controller based solely on equivalent control without the need for additional repetitive control compensation. In principle, as long as the prediction model is sufficiently accurate, error quantization becomes unnecessary, and precise control can be achieved using the discrete sliding mode algorithm itself. The purpose of incorporating this algorithm into the experiments is to demonstrate the inherent control performance of the QSMC controller and the necessity of error quantization through comparative experiments with the proposed RQSMC algorithm. Repeated quasi-sliding mode control (RQSMC): The algorithm proposed in this application.

[0173] Specifically, first, the time complexity of the four methods was evaluated to compare their respective computational requirements. Operationally, the measured execution time was defined as the time interval from the completion of sampling to the successful update of the period register of the PWM module. In order to isolate the intrinsic computational efficiency of the core algorithm, the time consumption associated with auxiliary processes (including signal conditioning routines, software protection mechanisms, and other peripheral operations unrelated to the core algorithm) was excluded. The experimental results are shown in Figure 2. Figure 5 As shown, it is shown that the addition of the repetitive controller does not significantly increase the computational burden.

[0174] exist Figure 6 and Figure 7 In Figure 2, the experimental waveforms of the four algorithms under linear load are depicted. These waveforms include the three-phase output voltage and the B-phase output current. Figure 6The steady-state performance is shown, where all four algorithms show good steady-state capability. By performing fast Fourier transform (FFT) analysis on the experimental data acquired from the oscilloscope, Figure 8 The harmonic spectrum of the output voltage is presented. Experimental results show that when the system is connected to a linear balanced load, the proposed algorithm exhibits the minimum harmonic distortion, while the traditional SMC method exhibits the most harmonics. Figure 7 Experimental results are presented, describing a step change in linear load impedance. During this transient load transition, the output voltage fluctuations are negligible for all four methods.

[0175] Figure 9 、 10 Figures 1 and 2 show experimental results when the inverter is supplying a single-phase load. Phase A of the inverter is supplied with a 36.3Ω resistive load, while phases B and C are unloaded. Due to the unbalanced load, an inrush current flows through the fourth wire between the load neutral point and the midpoint of the inverter's fourth bridge arm, causing a voltage drop and distorting the symmetrical output voltage. In this study, a dedicated neutral point balancing control strategy was not designed. Instead, a standard PI algorithm was used to regulate the DC voltage balance, intentionally allowing for a certain degree of disturbance in the DC bus neutral point voltage. This approach effectively demonstrates the inherent disturbance rejection capability of the proposed algorithm. Figure 12 The voltage waveform at the DC bus midpoint under unbalanced load conditions is shown. The voltage exhibits 50Hz fluctuations with an amplitude of ±6V, confirming the previous analysis of DC voltage disturbances. Figure 9 A steady-state performance analysis is presented, where the voltage waveform exhibits significant distortion under the traditional PI algorithm. In contrast, the proposed method demonstrates stronger resilience to single-phase load disturbances. Figure 11 The THD analysis results in Figure 3 show that both the conventional SMC and the proposed algorithm exhibit higher output voltage quality than the conventional PI and QSMC methods. However, the conventional SMC and QSMC exhibit larger steady-state errors compared to PI and RQSMC, indicating poor tracking performance under unbalanced load conditions. Figure 10 As shown in Figure 3, the PI-regulated voltage suffers severe distortion during a sudden single-phase load transient. In contrast, the three compared algorithms maintain higher voltage stability under the same transient load conditions. Experimental studies validate the algorithm's ability to simultaneously achieve fast dynamic response and excellent steady-state voltage tracking performance.

[0176] exist Figure 13 In, it shows Figure 14 The steady-state experimental waveform under nonlinear load is shown. It is worth noting that due to the linear characteristics of traditional PI control, it exhibits significant distortion when faced with nonlinear loads. In contrast, the SMC-based method maintains good steady-state performance while effectively tracking the reference voltage. In addition, Figure 15 The harmonic spectrum of the output voltage is presented. Experimental results show that the SMC-based approach exhibits superior ability to handle nonlinear loads compared to a traditional PI controller. Among the three SMC-based approaches, the proposed algorithm exhibits the lowest total harmonic distortion (3.66%). This demonstrates that the proposed approach achieves the best steady-state performance compared to traditional PI, traditional SMC, and QSMC.

[0177] Reference Figure 16 , is a broken line diagram of the sensitivity analysis of the parameters of the four evaluation algorithms in the verification experiment. Figure 16 It can be seen that all three methods maintain satisfactory performance even under severe parameter mismatch conditions. Specifically, the proposed method has poor parameter mismatch recovery ability compared to the other two methods.

[0178] It is worth noting that for traditional SMC, the steady-state error of the output voltage actually decreases when the parameter value increases. This is because when the control strategy is essentially dependent on the system model parameters, parameter changes under specific load conditions will directly affect the output characteristics. However, it is not recommended to arbitrarily increase the parameters because such empirical adjustments lack universality under different operating conditions. In parameter mismatch analysis, system performance should not be judged solely by and THD, but must be fully evaluated through sensitivity assessment. Specifically, The gradient of THD with respect to parameter perturbations is used as the key robustness indicator. Smaller gradients correspond to lower parameter dependence and enhanced algorithm robustness, reflecting better adaptability to model uncertainty. Obviously, considering Compared to QSMC and RQSMC, the traditional SMC exhibits significantly higher sensitivity to parameter mismatch in voltage tracking accuracy. Furthermore, the proposed algorithm significantly reduces steady-state error compared to QSMC. However, THD analysis shows that the waveform quality of the proposed algorithm is more susceptible to parameter variations than the other two methods when considering the THD gradient, especially when the model inductance exceeds the actual value. This is because the repetitive control algorithm integrated in the proposed method is sensitive to parameter variations due to its inherent linear control characteristics. However, numerically, the THD of the proposed algorithm remains comparable to that of the traditional SMC even when the model inductance reaches 1.5 times the actual value. Therefore, as long as the parameter variation remains within ±50% of the actual value, the overall performance of the proposed algorithm still outperforms the traditional SMC, demonstrating its excellent applicability. Experimental results further demonstrate that the proposed algorithm requires only a baseline prediction model, a simple error compensation mechanism, and a simple neutral point control method to demonstrate robust performance and low tracking error under various operating conditions.

[0179] Figure 17 4 is a block diagram of an electronic device provided in one embodiment of the present application. The device includes at least a processor 401 and a memory 402.

[0180] The processor 401 may include one or more processing cores, and the processor 401 may also include a main processor and a coprocessor. The memory 402 may include one or more computer-readable storage media. In some embodiments, the non-transitory computer-readable storage medium in the memory 402 is used to store at least one instruction, which is executed by the processor 401 to implement the predictive compensation quasi-sliding mode control method for a three-phase four-bridge-leg three-level inverter provided in the method embodiment of the present application.

[0181] Of course, the electronic device may also include fewer or more components, which is not limited in this embodiment.

[0182] Optionally, the present application also provides a computer-readable storage medium, which stores a program, and the program is loaded and executed by a processor to implement the predictive compensation quasi-sliding mode control method for a three-phase four-bridge-arm three-level inverter of the above method embodiment.

[0183] Optionally, the present application also provides a computer product, which includes a computer-readable storage medium, in which a program is stored, and the program is loaded and executed by a processor to implement the predictive compensation quasi-sliding mode control method for a three-phase four-bridge-arm three-level inverter of the above-mentioned method embodiment.

[0184] The above embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A predictive compensation quasi-sliding mode control method for a three-phase four-leg three-level inverter, characterized in that: The method comprises: A continuous state-space model of multi-variable coupling is established based on the topology of a three-phase four-leg three-level inverter. The forward Euler approximation method is used to discretize the continuous state space model and construct a discrete state space model that can be used by the digital controller; The output error of the output voltage and its derivative are defined using a discrete state space model. The sliding mode function is defined based on the output error of the output voltage and its derivative under the introduction of the quasi-sliding mode control theory, including: Using the discrete state space model for each coordinate component Define the tracking error of the output voltage and its derivative error; select two state variables and , the output error of the output voltage is defined as follows: ； in, represents the output voltage reference value in the αβγ reference coordinate system, is the actual output voltage value; the derivative of the error is defined as follows: ； in, and denote the derivatives of the reference voltage and output voltage respectively, is the filter capacitor, and They are load current and output current respectively; based on the above two state variables, construct a sliding mode function as follows: ； in, is the sliding mode coefficient; introduce the quasi-sliding mode control theory and define The sliding mode function value of the sampling period satisfies: ； in, is the sliding mode function value at the kth sampling moment, is the error bound of the sliding mode function allowed in quasi-sliding mode control; Construct a prediction model and predict the sliding mode function value at the next moment. Introduce an extended discrete state space model to extract the model error. Based on the deviation between the predicted value and the model error, correct the defined sliding mode function, including: The forward Euler approximation is used to predict the system model, and the predicted state variables are substituted into the sliding mode function to obtain the sliding mode function prediction value at the next moment. :Introduce the extended discrete state space model to define the sliding mode function error term ; The modified sliding mode function is: ； A repetitive controller is constructed based on the prediction error of the modified sliding mode function to extract the periodic error component and generate a compensation signal. The compensation signal output by the repetitive controller is fused with the equivalent control quantity constructed based on the extended discrete state space model to form the final control input for the drive system.

2. The predictive compensation quasi-sliding mode control method for a three-phase four-leg three-level inverter according to claim 1, characterized in that: The method of establishing a multi-variable coupled continuous state space model based on the topology of a three-phase four-leg three-level inverter includes: The topology of the three-phase four-bridge-arm three-level inverter includes three-phase main bridge arms and a neutral bridge arm. The three-phase main bridge arms are the A-phase bridge arm, the B-phase bridge arm, and the C-phase bridge arm respectively. The neutral bridge arm is used to connect the neutral point of the load to provide a zero-sequence current loop; each bridge arm is composed of four power switching devices, and three output level states are achieved through different conduction combinations, corresponding to the positive half-bus voltage, the neutral point voltage, and the negative half-bus voltage respectively. Upper and lower capacitors connected in series are configured on the DC side to form a voltage divider structure; the output end of each bridge arm is connected to the load through a filter inductor and a filter capacitor; the neutral bridge arm is connected to the neutral point of the load via a connecting inductor.

3. The predictive compensation quasi-sliding mode control method for a three-phase four-leg three-level inverter according to claim 2, characterized in that: The multi-variable coupled continuous state space model based on the topology of the three-phase four-bridge-leg three-level inverter further includes: Determine the corresponding relationship between the switch state and the output level of each bridge arm, and obtain the voltage value of the output voltage of each bridge arm relative to the DC midpoint; Under the assumption that the neutral point voltage is balanced, the load line voltage is determined based on the relationship between the output voltages of each bridge arm. Based on the output structure of the three-phase bridge arm and the connection structure of the neutral bridge arm, the differential relationship between the three-phase output voltage and the filter inductor current is established. The current balance relationship between the three-phase filter current and the output voltage is established based on the filter capacitor structure; the three-phase voltage and filter current are converted to the αβγ coordinate system through stationary coordinate transformation, and the equivalent filter inductance and filter capacitor parameters are defined; A continuous-time state-space model is established using the voltage and current variables and equivalent filter element parameters converted to the αβγ coordinate system.

4. The predictive compensation quasi-sliding mode control method for a three-phase four-leg three-level inverter according to claim 1, characterized in that: The forward Euler approximation method is used to discretize the continuous state space model to construct a discrete state space model that can be used by the digital controller. The differential equations of the filter inductor current and filter capacitor voltage in the continuous-time state-space model are discretized based on the forward Euler approximation method, and the evolution of the state in adjacent sampling periods is expressed in differential form. The time derivatives in the continuous equations are replaced by the corresponding differential terms to derive the discrete-time state transfer matrix, control input matrix, and corresponding disturbance compensation terms. A discrete-time state space model is constructed by utilizing the state transfer matrix, the control input matrix and the disturbance compensation term, and combining the filter inductor current, the filter capacitor voltage, the control input voltage and the load current variables.

5. The predictive compensation quasi-sliding mode control method for a three-phase four-leg three-level inverter according to claim 1, characterized in that: The method of constructing a repetitive controller based on the prediction error of the modified sliding mode function, extracting the periodic error component and generating a compensation signal includes: Based on the error between the predicted value and the actual value of the modified sliding mode function, a repetitive controller with a periodic memory mechanism is constructed to extract the periodic error component and generate a compensation signal. The input of the repetitive controller is the difference between the predicted value and the actual value of the modified sliding mode function. The difference is expressed as the transfer relationship between the control input disturbance and the sliding mode function error under linear approximation through small signal modeling, and the output expression of the compensation signal is determined according to the transfer coefficient; The repetitive controller adopts a z-domain structure with a time delay link. A zero-phase low-pass filter that satisfies the normalization constraint is introduced into the internal model to construct a periodic error extraction path. An additional compensation filter is used to attenuate high-frequency interference. The compensation signal output by the repetitive controller is subjected to time delay compensation at the end through a phase advance element, thereby forming a feedback channel that matches the equivalent control quantity.

6. The predictive compensation quasi-sliding mode control method for a three-phase four-leg three-level inverter according to claim 1, characterized in that: The method of fusing the compensation signal output by the repetitive controller with the equivalent control quantity constructed based on the extended discrete state space model to form a final control input for the drive system includes: The compensation amount of the repeat controller output Equivalent control quantity based on extended discrete state space model The fusion forms a complete control law, and the expression of the control law is as follows: ； in, For the control cycles are applied to the final control input of the system.

7. An electronic device, characterized in that: The device includes a processor and a memory; the memory stores a program, and the program is loaded and executed by the processor to implement a predictive compensation quasi-sliding mode control method for a three-phase four-bridge-leg three-level inverter as described in any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that The storage medium stores a program, which, when executed by a processor, is used to implement a predictive compensation quasi-sliding mode control method for a three-phase four-leg three-level inverter according to any one of claims 1 to 6.

Images