T-type inverter current sensorless virtual vector quasi-sliding mode predictive control method

Abstract

The application provides a T-type inverter current sensor-free virtual vector quasi-sliding mode predictive control method, which directly embeds a discrete sliding mode surface into a cost function of model predictive control for a three-phase three-level T-type voltage source inverter, and fuses strong robustness of the sliding mode control and fast dynamic response of the predictive control; a virtual voltage vector set composed of multiple basic voltage vectors and having an algebraic sum of zero of a midpoint current is synthesized to realize autonomous balancing of a midpoint potential without a weighting factor; a full-dimensional state observer is designed to estimate a capacitor current, and current sensor-free operation is realized; and a three-step screening strategy is proposed to significantly reduce 69 candidate virtual voltage vectors to 14, thereby significantly reducing calculation complexity. The application solves problems of insufficient robustness, obvious chattering, complicated weighting factor setting, high dependence on sensors and heavy calculation burden in the prior art, and improves inverter output waveform quality, dynamic response, parameter robustness and engineering realizability.

Description

Technical Field

[0001] This invention relates to the field of power electronics, and more specifically to a low-complexity inverse model predictive control method and system for a T-type three-level inverter. Background Technology

[0002] Against the backdrop of the rapid development of the renewable energy industry, voltage source inverters, as the core interface equipment for interconnecting distributed energy sources such as photovoltaics and wind power with the power grid, directly determine the power quality and system stability of new energy grid connection due to their control performance. Among them, the three-phase three-level T-type voltage source inverter with LC filter has become the mainstream topology in the field of medium and low voltage new energy grid connection due to its advantages such as strong harmonic suppression capability, high output voltage quality, and excellent operational reliability. It is widely used in power electronic systems such as photovoltaic grid connection, wind power converters, and uninterruptible power supplies.

[0003] To achieve high-precision tracking control of voltage and current in inverters, advanced control strategies are crucial. Traditional linear control strategies, such as proportional-integral (PI) control, are limited by their inherent control characteristics and struggle to effectively address the strong nonlinearity of power electronic systems, random variations in grid impedance, and complex external disturbances. Their control robustness and dynamic response performance are also insufficient to meet the high requirements of new energy grid integration. Therefore, nonlinear control methods have gradually become the mainstream research and application in this field.

[0004] Sliding mode control (SMC), a typical nonlinear control method, achieves strong anti-interference capability against load changes and parameter uncertainties by driving the system state to a preset sliding surface and maintaining sliding motion. It has been widely studied in inverter control. However, in practical digital control engineering applications, since the control signal remains constant in each sampling period, ideal continuous-time sliding motion cannot be achieved. The system will generate chattering (quasi-sliding motion) with limited bandwidth near the sliding surface, which significantly reduces control accuracy. At the same time, traditional sliding mode control cannot explicitly minimize tracking error, which can easily lead to unnecessary fluctuations in inverter output and also brings greater complexity to multi-objective optimization.

[0005] Finite Control Set Model Predictive Control (FCS-MPC) can quickly determine the optimal voltage vector through online rolling optimization, minimizing tracking error with a simple control structure, effectively compensating for the aforementioned shortcomings of sliding mode control. However, traditional FCS-MPC has inherent limitations: it uses only one voltage vector per switching cycle, resulting in large inverter output ripple and inconsistent switching frequency. When applying multi-voltage vector strategies to three-phase three-level T-type inverters, neutral point potential balance becomes a critical issue that needs to be addressed. Existing neutral point potential balance schemes often achieve this by introducing weighting terms into the cost function, but the tuning process of the weighting factors is cumbersome and relies heavily on engineering experience, significantly increasing the difficulty of algorithm implementation and system debugging costs. Some studies have suppressed neutral point potential fluctuations by adjusting the duty cycle of redundant small voltage vectors, but this still does not fundamentally solve the complexity problem of multi-objective optimization.

[0006] Furthermore, the closed-loop control structure of traditional inverters heavily relies on physical sensors, requiring current sensors to collect key signals such as inductor current and output current, and voltage sensors to collect signals such as DC-side voltage and filter capacitor voltage. However, physical sensors are susceptible to electromagnetic interference and temperature drift in industrial environments, and can introduce sampling errors. This not only reduces the reliability of system operation but also increases the complexity and cost of hardware circuit design. Simultaneously, sensor wiring and installation limit the miniaturization and integration of inverter designs. To reduce sensor dependence, observer technology has been gradually introduced into the inverter control field, such as sliding mode observers and full-dimensional observers, which achieve sensorless detection of key electrical signals through algorithmic estimation. However, when existing observer technology is combined with model predictive control, there are problems such as cumbersome observer gain tuning, reduced parameter robustness due to coupling with predictive control, and error superposition. At the same time, existing hybrid optimization methods that integrate sliding mode control and model predictive control either ignore the dynamic characteristics of grid voltage, have too many voltage vectors to be evaluated leading to a surge in computation, or are only applicable to two-level inverters and do not consider the core requirement of midpoint potential balance in three-level inverters, making it difficult to directly apply to the control of T-type three-level inverters.

[0007] In summary, existing control methods for T-type three-level inverters generally suffer from technical drawbacks such as cumbersome weighting factor tuning, high computational cost, significant sliding mode chattering, difficulty in balancing the midpoint potential, high sensor dependence, and insufficient parameter robustness. There is an urgent need to develop a new control method that can integrate the robustness of sliding mode control with the fast dynamic response of model predictive control, while achieving sensorless operation, self-balancing of the midpoint potential, and low computational complexity, in order to meet the application requirements of high control accuracy, high reliability, and high robustness for T-type inverters.

[0008] The shortcomings of existing technologies are as follows: Schemes based on the combination of proportional-integral linear control and modulation strategies are difficult to cope with strong system nonlinearity, grid impedance changes and external disturbances, lack robustness and have complex parameter tuning and strong coupling when performing multi-objective control; while directly applying traditional sliding mode control (SMC) has problems such as significant chattering in digital implementation, inability to explicitly minimize tracking error, and high complexity of multi-objective optimization, and it is difficult to meet the midpoint potential balance requirements of three-level inverters; if traditional finite set model predictive control (FCS-MPC) is used, it faces the defects of cumbersome tuning of midpoint potential balance weighting factor, large output ripple of single vector control, sensitivity to model parameters and high computational cost. Summary of the Invention

[0009] The objective of this invention is achieved through the following technical solutions.

[0010] This invention provides a sensorless virtual vector quasi-sliding mode predictive control method for T-type inverters. For three-phase, three-level T-type voltage source inverters, it directly embeds the discrete sliding mode surface into the cost function of model predictive control, combining the strong robustness of sliding mode control with the fast dynamic response of predictive control. By synthesizing a set of virtual voltage vectors composed of multiple basic voltage vectors with a zero algebraic sum of midpoint currents, it achieves unweighted factor-free autonomous balancing of the midpoint potential. A full-dimensional state observer is designed to estimate the capacitor current, enabling sensorless operation. A three-step screening strategy is proposed, significantly reducing the number of candidate virtual voltage vectors from 69 to 14, substantially reducing computational complexity. This invention solves the problems of insufficient robustness, significant chattering, cumbersome weighting factor tuning, high sensor dependence, and heavy computational burden in existing technologies, improving the inverter's output waveform quality, dynamic response, parameter robustness, and engineering feasibility.

[0011] Specifically, this invention aims to propose a novel sensorless virtual-vector quasi-sliding mode predictive control (VVQSMPC) method to address the shortcomings of PI linear control, traditional sliding mode control (SMC), and finite control set model predictive control (FCS-MPC). The core objective and innovation of this invention lies in integrating the robustness of sliding mode control with the fast dynamic response characteristics of model predictive control. It directly embeds the discrete sliding surface into the cost function and synthesizes a virtual voltage vector with midpoint self-balancing characteristics. Simultaneously, it introduces a full-dimensional state observer to achieve sensorless operation and simplifies calculations by designing a three-step vector selection scheme, thereby bringing about a comprehensive improvement in inverter control performance and a significant simplification of algorithm design. Specifically:

[0012] (1) Upgrade from "single control" to "integrated robustness": The proposed VVQSMPC method breaks the independent application mode of sliding mode control and model predictive control, directly embedding the discrete sliding surface into the predictive control cost function, integrating the fast dynamic tracking characteristics of MPC and the strong disturbance rejection robustness of SMC, while effectively suppressing the chattering problem of traditional sliding mode control through quasi-sliding mode motion design. Compared with the shortcomings of traditional MPC being sensitive to parameter mismatch and traditional SMC being unable to explicitly minimize tracking error, VVQSMPC significantly enhances the system's adaptability to load changes, parameter deviations and external disturbances while ensuring fast transient response.

[0013] (2) The leap from "weighted tuning" to "self-balancing control": The proposed VVQSMPC method abandons the cumbersome tuning process of introducing weighting factors into the cost function to balance the midpoint potential in traditional MPC. By synthesizing 69 virtual voltage vectors with midpoint self-balancing characteristics, the midpoint current tends to zero within the control cycle, achieving unweighted factor-free autonomous balancing of the midpoint potential. Compared with traditional MPC, which requires repeated tuning of weighting factors to balance tracking accuracy and midpoint balance, VVQSMPC can achieve multi-objective collaborative control simply by reasonably combining virtual vectors, greatly reducing the engineering implementation difficulty of the algorithm.

[0014] (3) Breakthrough from "sensor dependence" to "sensorless operation": The proposed VVQSMPC method eliminates the high dependence of traditional control on physical current sensors. By designing a full-dimensional state observer, it accurately estimates key signals such as inverter capacitor current and inductor current, replacing sensor sampling values ​​in control calculations. Compared with the problems of traditional control being susceptible to electromagnetic interference and temperature drift from sensors, VVQSMPC not only simplifies hardware circuit design and reduces hardware costs, but also effectively suppresses sampling errors and improves the reliability of the system under complex operating conditions.

[0015] (4) Optimization from "loop optimization" to "tiered selection": The proposed VVQSMPC method breaks through the loop optimization mode of traditional MPC, which evaluates the cost function of all voltage vectors one by one. It designs a third-order virtual vector pre-selection scheme, which filters 69 virtual vectors in stages to only 14 candidate vectors participating in the cost function calculation. Compared with the large number of loop operations required by traditional MPC to traverse 27 real vectors, VVQSMPC significantly reduces the amount of computation through accurate vector pre-selection, and significantly improves the real-time performance of the algorithm.

[0016] The innovation of this invention lies in proposing a sensorless quasi-sliding mode predictive control method that integrates sliding mode control and model predictive control, based on virtual vectors and a full-dimensional state observer, for three-phase three-level T-type inverter topologies. This method eliminates weighting factor tuning through virtual vector self-balancing, reduces computational complexity through third-order pre-selection, achieves sensorless operation through a full-dimensional observer, and enhances control robustness through sliding mode-predictive fusion. It comprehensively addresses the technical shortcomings of insufficient robustness in PI linear control, significant chattering in traditional SMC, cumbersome and computationally expensive traditional MPC weighted tuning, and reliance on physical sensors in the control system. This provides a more efficient, reliable, and easily implemented digital control solution for T-type inverters in the medium- and low-voltage renewable energy grid-connected field. Attached Figure Description

[0017] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0018] Figure 1 A topology diagram of a three-phase three-level inverter according to an embodiment of the present invention is shown.

[0019] Figure 2 A voltage vector diagram of a three-phase three-level inverter according to an embodiment of the present invention is shown.

[0020] Figure 3 A virtual vector diagram according to an embodiment of the present invention is shown.

[0021] Figure 4 A block diagram of VVQSMPC control according to an embodiment of the present invention is shown.

[0022] Figure 5 The experimental results of the computation time for the three methods are shown in the figure.

[0023] Figure 6 The steady-state output waveforms of the three methods under linear load are shown.

[0024] Figure 7 The results of the dynamic response capability test for the three methods are shown in the figure.

[0025] Figure 8 The waveform diagram of the midpoint potential self-balancing is shown. Detailed Implementation

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

[0027] Terminology Explanation:

[0028] A three-phase three-level voltage source inverter (3P-3L-VSI) is an advanced power conversion device in the field of power electronics, widely used in medium- and high-voltage high-power scenarios, such as industrial motor drives, new energy grid connection, and energy storage systems. It controls the on / off state of 12 power switching devices to achieve the desired output voltage. 0 The three voltage levels are synthesized into a near-sine wave three-phase AC voltage, achieving efficient conversion from DC to high-quality three-phase AC. Compared to traditional two-level or three-level topologies, it has advantages such as lower output harmonic content, less electromagnetic interference, and lower filtering requirements. Due to its transformerless nature, special attention must be paid to suppressing leakage current caused by the parasitic capacitance of the photovoltaic panel to ground.

[0029] Sliding Mode Control (SMC) is a robust control method based on nonlinear control theory, widely used in the control of complex industrial systems subject to parameter perturbations and external disturbances. Its core idea is to design a specific sliding mode surface (SMC), typically composed of system state variables and their combinations. By switching control strategies, the system's state trajectory is forced to reach and remain on the SMC within a finite time, moving along the SMC towards the system's equilibrium point (the set target), thus achieving precise control of the system. Its control logic comprises two parts: the SMC design and the switching control law. It effectively suppresses the effects of system parameter changes and external disturbances, exhibiting strong robustness and ensuring the system can stably track the setpoint even under complex operating conditions.

[0030] Finite Control Set Model Predictive Control (FCS-MPC) is an advanced control strategy particularly suitable for power electronic converters. Its core principle lies in leveraging the finite number of switching states of the converter. Within each control cycle, it predicts the future system behavior under all possible combinations of switching states using a discrete model of the system, and directly selects the optimal switching state to apply to the converter based on a predefined cost function. This method offers fast dynamic response and is naturally suited for digital control.

[0031] Quasi-Sliding Mode Control (QSMC): An improved sliding mode control strategy. Its core lies in overcoming the difficulty of achieving ideal continuous sliding motion in traditional sliding mode control, replacing it with quasi-sliding mode motion control in the discrete domain. Quasi-sliding mode control allows the system state to fluctuate within a finite bandwidth near the sliding surface, preserving the robustness of the sliding mode core while effectively suppressing chattering. This method is adaptable to digital control implementation and can effectively improve practical control accuracy.

[0032] A. System Model

[0033] The object controlled by this invention is a three-phase three-level photovoltaic inverter, such as... Figure 1 As shown, where It is the DC side voltage. and These represent the upper and lower capacitors on the DC side, respectively. and These represent the current flowing through the upper and lower capacitors, respectively. and This constitutes an LC filter. and These are the three-phase inductor current, the current filtering capacitor current, and the output current, respectively.

[0034] B. Mathematical Model

[0035] For a three-phase three-level inverter, the four switches in each phase can output three states: "P", "O", and "N", with corresponding output voltages of respectively. 0 and Therefore, as Figure 2 As shown, the total voltage of the three-phase three-level inverter is 27 (3 3 The relationship between the output state and the switching transistor state can be expressed using a switching function. Represented as:

[0036]

[0037] Assuming the upper and lower capacitance values ​​are the same on the DC bus side, then the output voltage of each phase relative to point "O" is... This can be represented as:

[0038]

[0039] in .

[0040] In addition, the voltage of each phase relative to point "n" It can be represented as:

[0041]

[0042] The output voltage and inductor current are defined as system state variables. The system state can be obtained through the Clark transformation. Continuous time in a coordinate system:

[0043]

[0044] in yes Output current in coordinate system yes Inductor current in coordinate system yes Load terminal output voltage in coordinate system yes Inverter output voltage in coordinate system.

[0045] Discretizing the system using forward Euler techniques yields the expression for the next time step:

[0046]

[0047] in The system control cycle is represented by "(k)", which represents the current time and "(k+1)" which represents the next time.

[0048] Current flowing through neutral point O It can be represented as

[0049]

[0050] Switch state P or N ( )lead to (Without affecting the neutral point potential), the 27 basic vectors generate 7 neutral point current values, such as... Figure 2 As shown.

[0051] C. Sliding mode predictive control

[0052] First, the second-order state-space equation for output voltage tracking of a VSI system with a second-order LC filter is as follows: In the coordinate system, it is described as:

[0053]

[0054] in Indicates the reference voltage. This represents the derivative of the reference voltage.

[0055] Secondly, the sliding surface is defined as

[0056]

[0057] in It is the sliding mode coefficient that affects the dynamic response.

[0058] In order to avoid the amplification of high-frequency noise caused by the output voltage derivative in (8), the output equations (4) and (7) are substituted into (8) to obtain the discrete sliding mode function at the k-th time.

[0059]

[0060] Furthermore, based on the prediction model (7) and the sliding mode function (9), the proposed prediction model based on the sliding mode function was constructed:

[0061]

[0062] Where "(k+1)" represents the next time step. Since the load current dynamics are significantly slower than the sampling frequency, the load current is considered quasi-static within one control cycle. Define the sliding mode prediction cost function:

[0063]

[0064] The reference value for the sliding mode function is .

[0065] As can be seen from (11), the control objective is to optimally achieve the equivalent control of the discrete quasi-sliding mode. Therefore, an online rolling optimization mechanism similar to that of traditional FS-MPC is adopted. Specifically, the input of the quasi-sliding mode predictive control proposed in this invention is obtained by evaluating the cost function (17) by enumerating the voltage vector of the voltage source inverter. Subsequently, the optimal vector that minimizes (11) is selected and applied to the inverter, driving the system state to reach the quasi-sliding mode in an optimal manner.

[0066] D. Full-dimensional observer

[0067] Assuming Ts is small enough, the load current can be considered as a quasi-static current within one control cycle. Substituting this condition into (4), we get:

[0068]

[0069] in, This represents the current flowing through the filter capacitor. Rearranging equation (12) yields...

[0070]

[0071] in , .

[0072] It can be seen that equation (13) establishes the theoretical equivalence between capacitor current monitoring and traditional inductor / load current detection. This equivalence maintains the observability of the system and enables sensorless operation, laying the foundation for the subsequent implementation of sliding mode predictive control.

[0073] To replace physical current sensors, we designed a full-dimensional observer:

[0074]

[0075] in express The estimated value, express The estimated value. K is the observer gain matrix, and k1 and k2 are the coefficients of the gain matrix. Discretizing equation (13) yields...

[0076]

[0077] in It is the discretized observer gain. For the error vector, and It is a discrete system matrix.

[0078] According to equation (15), by replacing the current sample with the estimated value, equation (10) can be rewritten as:

[0079]

[0080] E. Virtual voltage vector for NP potential self-balancing and three-step screening

[0081] The preceding cost function derivation assumed an ideally balanced NP potential. However, in reality, NP potential balance is a key control objective for three-level topologies. In fact, non-zero NP current... This inherently causes NP potential fluctuations, leading to the charging and discharging of the upper and lower DC link capacitors. However, if the NP current is zero, the NP potential can achieve self-balancing. In model predictive control, each actual voltage vector corresponds to a specific current. .for Figure 1 The T-type inverter shown has the following output current. The following relationship must be satisfied:

[0082]

[0083] By combining actual voltage vectors to keep the NP current zero, a new set of virtual voltage vectors can be obtained, thus ensuring a constant NP potential. Simultaneously, this process eliminates the need for cost function terms used to predict and balance the NP potential, which helps reduce computational burden.

[0084] In each sampling period Ts, the virtual vector V vx Satisfy the following relationship:

[0085]

[0086] in, It is a real vector The duration, let N be the number of real vectors, can be derived as follows: The composite vector is as follows: Figure 3 As shown, the basic principle of the combination and its effect on the NP potential can be explained as follows:

[0087] 1) Zero voltage vector sets (PPP, OOO, and NNN) and large voltage vector sets (PNN and PPN) have zero NP current, making them directly applicable to new virtual voltage vector sets.

[0088] 2) Virtual voltage vector V v13 This can be represented as a combination of two real small voltage vectors:

[0089]

[0090] Obviously, when the switch state of V3 is OON, it will generate NP current. Effect, and V 10 When the switching state is POO, it will generate a current to the NP current. Effect. Therefore, when the action time of each vector is 0.5 s T, i np When it becomes zero, the NP potential is not affected by V. v13 The impact.

[0091] 3) Similarly, the virtual voltage vector can also be represented as a combination of three real vectors. Each combination is represented by the NP currents respectively. , and It consists of real vectors, or three real vectors with zero NP current components. Each vector acts for an equal time. For example, V v12 This can be represented as a combination of three actual voltage vectors:

[0092]

[0093] 4) To further expand the number of control sets, improve control accuracy, and reduce output harmonics, a virtual voltage vector consisting of four vectors is employed. Like other vectors, the basic combination strategy is to suppress NP current within one control cycle Ts. This can be achieved by using two pairs of opposite vectors, or a pair of opposite vectors combined with two voltage vectors where the NP current is zero. Each vector has an action time of 0.25 Ts, for example, V... v9 It consists of four real vectors.

[0094]

[0095] Based on the above synthesis principle, a total of 69 vectors were synthesized, an increase of 155% compared to the traditional 27 real vectors. All synthesized vectors are as follows: Figure 3 As shown.

[0096] according to Figure 3 It can be seen that although the new vector set increases the number of voltage vectors, it also introduces a heavier computational burden. To address this issue, this invention proposes a three-step screening method, as follows:

[0097] The first step is to evaluate six representative vectors (Vv12, Vv23, Vv34, Vv45, Vv56, Vv66) for each sector to find the dummy vector that minimizes the cost function. The second step is to select a smaller sector containing three representative vectors. The third step is to select the best dummy vector from the remaining 5 or 7 VVs in the smaller sector. For clarity, Table I lists the representative vectors evaluated at each level of the three-level VVQSMPC. For example, if VV12, Vv23, Vv34, Vv45, Vv56, Vv66 are selected in stage I... v12 Then, candidate vectors in sector I, including Vv9, Vv10, and Vv14, are evaluated. Then, assuming Vv10 is selected, Vv5, Vv15, Vv6, Vv7, and Vv11 are evaluated. Clearly, the three-step screening proposed in this invention reduces the number of candidate virtual voltage vectors from 69 to 14, significantly alleviating the computational burden.

[0098] Table I. Three-Step Virtual Voltage Vector Lookup Table

[0099]

[0100] In summary, the overall control block diagram of the sensorless virtual vector quasi-sliding mode predictive control proposed in this invention is as follows: Figure 4 As shown. First, the three-phase output voltage sampling signal is... Transformation to In the coordinate system, the observer predicts the capacitor current at the next moment, and then predicts the sliding surface at the next moment based on the sampled value and the observed value. The virtual vector corresponding to the minimum cost function is found through three-step screening, and finally the switching transistor is controlled by the SPWM modulation wave.

[0101] To verify the effectiveness of the VVQSMPC proposed in this invention, a three-phase three-level inverter experiment was constructed. Experimental parameters are detailed in Table II.

[0102] Table II. System Parameters

[0103]

[0104] Meanwhile, the sensorless VVQSMPC algorithm proposed in this invention is compared with two other methods, as explained below:

[0105] Traditional MPC: Traditional MPC employs a finite control set predictive optimization structure. Based on the system's discrete mathematical model, it selects the optimal switching state through rolling optimization. Specifically, to ensure voltage balance, it directly predicts the three-phase output voltage and current, selecting the switching sequence that minimizes the cost function to achieve closed-loop control.

[0106] Sensor-based VVQSMPC: The proposed control algorithm is implemented using a physical current sensor instead of an observer as an idealized reference.

[0107] Experimental effects of the present invention

[0108] A. Calculation time

[0109] The computation time of the three methods was compared on an inverter platform with a sampling period of 62.5 μs. The experimental results are as follows: Figure 5 As shown, the execution time of all three control methods does not exceed 60% of the control cycle, indicating that all tested algorithms have relatively low computational burden. Among the three methods, traditional model predictive control takes the longest time, at 33.63 μs. Sensor-based VVQSMPC takes 22.98 μs. The sensorless VVQSMPC proposed in this invention takes 26.86 μs, a slight increase of 3.88 μs due to the calculation of the full-order observer.

[0110] B. Output waveform quality

[0111] Figure 6 Steady-state output waveforms for three methods under linear loads are presented. (a) is the traditional MPC; (b) is the VVQSMPC with a sensor; and (c) is the algorithm proposed in this invention. Linear load With a resistance of 30Ω, the total harmonic distortion (THD) of the traditional MPC, the VVQSMPC with sensor, and the algorithm proposed in this invention are 4.46%, 2.47%, and 2.21%, respectively. Experimental results show that the algorithm proposed in this invention has the lowest THD and the best output waveform quality.

[0112] C. Dynamic Testing

[0113] To test the dynamic response capabilities of the three methods, the linear load was stepped from 30Ω to 80Ω. The experimental results are as follows: Figure 7 As shown, (a) is the traditional MPC; (b) is the VVQSMPC with sensor; and (c) is the algorithm proposed in this invention. The three methods exhibit similar dynamic characteristics, with the output voltage remaining almost constant. At the step time, the NP potential fluctuation for all three methods is approximately 5 V. This observation indicates that the three methods demonstrate good dynamic response to linear load changes.

[0114] D. Midpoint potential self-equilibrium

[0115] To further evaluate the self-balancing capability of the neutral point voltage of the virtual vector synthesized by the algorithm proposed in this invention, a 12V DC offset was first added to the sampling signal to simulate the NP potential imbalance. Subsequently, the offset was eliminated, and the NP potential waveform was observed. Experimental results are as follows: Figure 8 As shown.

[0116] The experimental results show that the virtual vector control algorithm proposed in this invention can balance the midpoint potential within a finite time (420ms), thus verifying the correctness of the previous theoretical derivation. This algorithm achieves self-balancing of the NP potential, thereby optimizing the cost function and reducing computational complexity.

[0117] D. Parameter mismatch

[0118] In practical applications, parameter mismatch is difficult to avoid due to factors such as component heating, aging, and electromagnetic saturation. Given that LC filter parameter mismatch significantly degrades PI control performance and may cause system instability, this invention analyzes only two MPC-based algorithms by comparing the output voltage THD. Specifically, the capacitor and inductor parameters are modified in the code, while the actual capacitor and inductor parameters remain unchanged. Parameter mismatch and The definition of is:

[0119]

[0120] in, and These are actual parameter values. and The parameter values ​​used in the calculation.

[0121] The experimental results are shown in Table III. It can be seen that the traditional MPC is most sensitive to parameter mismatch. However, even with parameter mismatch, the VVQSMPC proposed in this invention still achieves the lowest total harmonic distortion (THD) and exhibits excellent robustness.

[0122] Table III. Parameter Mismatch Analysis

[0123]

[0124] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A sensorless virtual vector quasi-sliding mode predictive control method for a T-type inverter, characterized in that, The method, applied to a three-phase three-level T-type voltage source inverter, includes: The three-phase output voltage signal of the inverter is acquired and a coordinate transformation is performed to obtain the output voltage value in the αβ coordinate system; The output voltage value in the αβ coordinate system is input to a preset full-dimensional state observer to obtain the estimated value of the capacitor current in the next control cycle. Based on the output voltage sampling value of the current control cycle, the estimated capacitor current value of the next control cycle, and the preset discrete sliding surface function, the sliding surface function value of the next control cycle is predicted. A cost function is constructed, which is used to characterize the deviation between the sliding surface function value and the sliding surface reference value in the next control cycle; A preset three-step screening strategy is used to select candidate virtual voltage vectors from a preset set of virtual voltage vectors. Each virtual voltage vector in the set of virtual voltage vectors is synthesized from at least two basic voltage vectors, and the algebraic sum of the midpoint current of each virtual voltage vector in one control cycle is zero. The selected candidate virtual voltage vectors are substituted into the cost function for calculation, and the candidate virtual voltage vector that minimizes the cost function value is selected as the optimal virtual voltage vector. Based on the optimal virtual voltage vector, a corresponding switching control signal is generated to control the power switching devices of the inverter.

2. The method according to claim 1, characterized in that, The preset set of virtual voltage vectors is constructed in the following way: The basic voltage vector with zero midpoint current is selected and directly used as the virtual voltage vector. Two fundamental voltage vectors with opposite midpoint current effects are combined and their durations are made equal to synthesize a virtual voltage vector. The three basic voltage vectors are combined such that the algebraic sum of the midpoint currents of the three basic voltage vectors is zero, and the duration of each basic voltage vector is equal, to synthesize a virtual voltage vector. The four basic voltage vectors are combined such that the algebraic sum of the midpoint currents of the four basic voltage vectors is zero, and the duration of each basic voltage vector is equal, to synthesize a virtual voltage vector.

3. The method according to claim 1, characterized in that, The three-step screening strategy includes: The first step is to divide the voltage vector plane into multiple sectors, evaluate the pre-defined representative virtual voltage vectors in each sector, select the representative virtual voltage vector that minimizes the cost function value, and determine its sector. The second step is to evaluate the pre-defined representative virtual voltage vector of the small sector within the sector determined in the first step, select the representative virtual voltage vector of the small sector that minimizes the cost function value, and determine the small sector in which it is located. The third step is to evaluate the remaining candidate virtual voltage vectors in the small sector determined in the second step, excluding the already evaluated vectors, and select the virtual voltage vector that minimizes the cost function value as the optimal virtual voltage vector.

4. The method according to claim 1, characterized in that, The full-dimensional state observer is constructed based on the system's state-space model, and its gain matrix coefficients are tuned according to preset rules. It is used to estimate the inverter's capacitor current and inductor current based on the output voltage value in the αβ coordinate system.

5. The method according to claim 1, characterized in that, The discrete sliding surface function is constructed based on the output voltage tracking error of the inverter and its derivative. By introducing the mathematical model of the inverter, the derivative term of the output voltage is converted into a measurable or estimable state variable to avoid high-frequency noise amplification.

6. A sensorless virtual vector quasi-sliding mode predictive control system for a T-type inverter, characterized in that, The system, applied to a three-phase, three-level T-type voltage source inverter, includes: The signal acquisition and transformation module is used to acquire the three-phase output voltage signal of the inverter and perform coordinate transformation to obtain the output voltage value in the αβ coordinate system. The full-dimensional state observer module is used to estimate the capacitor current estimate for the next control cycle based on the output voltage value in the αβ coordinate system. The sliding surface prediction module is used to predict the sliding surface function value of the next control cycle based on the output voltage sampling value of the current control cycle, the estimated value of the capacitor current of the next control cycle, and the preset discrete sliding surface function. A cost function construction module is used to construct a cost function, which is used to characterize the deviation between the sliding surface function value and the sliding surface reference value in the next control cycle. The three-step vector filtering module is used to filter candidate virtual voltage vectors from a preset set of virtual voltage vectors using a preset three-step filtering strategy. Each virtual voltage vector in the set of virtual voltage vectors is synthesized from at least two basic voltage vectors, and the algebraic sum of the midpoint current of each virtual voltage vector in one control cycle is zero. The rolling optimization module is used to substitute the selected candidate virtual voltage vectors into the cost function for calculation, and select the candidate virtual voltage vector that minimizes the cost function value as the optimal virtual voltage vector. The pulse generation module is used to generate corresponding switching control signals based on the optimal virtual voltage vector to control the power switching devices of the inverter.

7. The system according to claim 6, characterized in that, The preset set of virtual voltage vectors is constructed in the following way: A basic voltage vector with zero midpoint current is selected and directly used as a virtual voltage vector; two basic voltage vectors with opposite midpoint current effects are combined and their durations are made equal to synthesize a virtual voltage vector. The three basic voltage vectors are combined such that the algebraic sum of the midpoint currents of the three basic voltage vectors is zero, and the duration of each basic voltage vector is equal, to synthesize a virtual voltage vector. The four basic voltage vectors are combined such that the algebraic sum of the midpoint currents of the four basic voltage vectors is zero, and the duration of each basic voltage vector is equal, to synthesize a virtual voltage vector.

8. The system according to claim 6, characterized in that, The three-step vector filtering module includes: The first-level sector filtering unit is used to divide the voltage vector plane into multiple sectors, evaluate the preset representative virtual voltage vectors in each sector, select the representative virtual voltage vector that minimizes the cost function value, and determine the sector in which it is located. The second-level small sector filtering unit is used to evaluate the preset small sector representative virtual voltage vector in the sector determined by the first-level sector filtering unit, select the small sector representative virtual voltage vector that minimizes the cost function value, and determine the small sector in which it is located. The third-level candidate vector filtering unit is used to evaluate the remaining candidate virtual voltage vectors in the small sector determined by the second-level small sector filtering unit, excluding the vectors that have already been evaluated, and select the virtual voltage vector that minimizes the cost function value as the optimal virtual voltage vector.

9. The system according to claim 6, characterized in that, The full-dimensional state observer module is constructed based on the system's state-space model, and its gain matrix coefficients are tuned according to preset rules. It is used to estimate the capacitor current and inductor current of the inverter based on the output voltage value in the αβ coordinate system.

10. The system according to claim 6, characterized in that, The discrete sliding surface function is constructed based on the output voltage tracking error of the inverter and its derivative. By introducing the mathematical model of the inverter, the derivative term of the output voltage is converted into a measurable or estimable state variable to avoid high-frequency noise amplification.

Images