A model prediction-based high-voltage multilevel converter switching frequency adaptive regulation method

Abstract

The present application relates to the field of power electronics, and more particularly to a high-voltage multilevel converter switching frequency adaptive adjustment method based on model prediction, which measures grid state variables in real time, identifies converter operation mode, selects corresponding prediction model, and predicts future state; through multi-objective rolling optimization, the optimal switching frequency is selected and applied to the PWM control link, while the feedback correction is used to correct the model parameters online, forming an adaptive closed-loop control. The present application has the advantages that: by real-time detection of grid voltage, grid-connected current, power and stability indicators and other state variables, the system operation mode is accurately identified; in the light load or oscillation risk working condition, the switching frequency is automatically increased, the control bandwidth is expanded, and the oscillation instability is suppressed; in the stable heavy load working condition, the switching frequency is reduced, and the switching loss is reduced. The method realizes the collaborative optimization of stability and efficiency, and significantly improves the operation robustness of the converter under complex working conditions.

Description

Technical Field

[0001] This invention relates to the field of power electronics technology, and in particular to a model-predictive adaptive adjustment method for the switching frequency of a high-voltage multilevel converter. Background Technology

[0002] High-voltage multi-level converters (HV-MLCs) are core power conversion devices in flexible AC / DC transmission systems (such as VSC-HVDC), large-capacity motor drives (such as medium- and high-voltage frequency converters), and new energy grid-connected devices (such as photovoltaic inverters and wind power converters). Their operational stability and efficiency are crucial for the reliable and efficient operation of the system. In practical applications, especially under conditions of complex grid impedance and frequent load fluctuations, these converters often face the contradictory problem of oscillation instability and switching losses.

[0003] From the perspective of system structure, the cable lines widely used in high-voltage power grids, the filter network on the output side of the converter, and the parasitic parameters of the transformer introduce significant capacitive impedance, resulting in an increase in the number of resonant frequency points and complex impedance characteristics, which can easily induce resonant oscillations in specific frequency bands.

[0004] From a control perspective, current engineering practices commonly employ pulse width modulation (PWM) technology based on fixed switching frequencies and corresponding linear control strategies (such as PI regulators). These methods sample and update the modulated wave within the carrier cycle, resulting in inherent control delays that severely limit the system's control bandwidth. Under light load or weak grid conditions, the system's damping characteristics deteriorate, phase margins become insufficient, and oscillations in the frequency range of hundreds to thousands of hertz are prone to occur, threatening equipment safety and grid stability.

[0005] To alleviate the aforementioned contradictions, adaptive switching frequency control is considered an effective approach. Its basic idea is to increase the switching frequency to expand the control bandwidth and enhance system damping when the system is under light load or with a high risk of oscillation; and to decrease the switching frequency to reduce switching losses and improve operating efficiency under heavy load or stable conditions. However, traditional switching frequency adjustment methods face significant engineering challenges: on the one hand, changing the switching frequency requires reconfiguring the PWM timer's period register and simultaneously adjusting all frequency-related compensator parameters in the control loop (such as PI parameters and filter coefficients); on the other hand, improper handling during frequency switching can easily cause current surges or control misalignment, making it difficult to achieve smooth and stable online switching. Summary of the Invention

[0006] The purpose of this invention is to provide a model-based adaptive adjustment method for the switching frequency of a high-voltage multilevel converter, which solves the problems of easy oscillation under light load, low efficiency under heavy load, and complex frequency switching process that relies on manual adjustment under traditional fixed frequency control. By real-time detection of system status, adaptive identification of operating mode, and automatic selection of the optimal switching frequency based on rolling prediction and multi-objective optimization, the method achieves synergistic optimization of stability and efficiency. At the same time, closed-loop correction improves the adaptability of the model, ultimately achieving the goals of suppressing oscillation, reducing losses, and simplifying engineering implementation.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A model-predictive-based adaptive switching frequency adjustment method for high-voltage multilevel converters includes: S1. Real-time measurement of the state variables of the power grid system connected to the high-voltage multilevel converter; State variables include grid voltage, grid-connected current, commanded current, output active power, output reactive power, grid oscillation flag, grid impedance, and current tracking error; S2. Identify the converter's operating mode based on status variables; Operating modes include: oscillation risk mode, light load mode, normal mode, heavy load mode, and overload or fault mode; S3. Based on the identified operating mode, select the corresponding prediction model from the pre-established piecewise linear state-space model library; S4. Based on the current state variables and the preset candidate switching frequency set, the state of the converter in the future time domain is predicted using a prediction model. S5. Through multi-objective rolling optimization, select the optimal switching frequency that minimizes the multi-objective cost function from the candidate switching frequency set. S6. Apply the optimal switching frequency to the PWM control loop of the converter, and correct the prediction model parameters online through feedback correction in the next sampling cycle to form an adaptive closed-loop control.

[0008] The prediction model is a piecewise linear state-space model that describes the relationship between switching frequency and the control bandwidth, current tracking dynamics, and switching losses of the power grid system.

[0009] In S2: The oscillation risk mode occurs when the converter's current tracking error exceeds the warning threshold, or when the grid oscillation flag is true. Light load mode is when the converter's output active power is lower than the light load power threshold and it is not in an oscillation risk mode; In normal mode, the converter's output active power is between the light load power threshold and the heavy load power threshold. Heavy load mode is when the converter's output active power exceeds the heavy load power threshold. Overload or fault mode occurs when the grid-connected current of the converter exceeds the maximum current protection threshold.

[0010] In S5, the expression for the multi-objective cost function is: ; in: This represents summation, in the future time domain from j=1 to... The above was carried out; Indicates the candidate frequency at time k The predicted current tracking error at time k+j; Indicates the candidate frequency at time k Predicted switching losses at time k+j; and This represents the weighting coefficient, which is dynamically adjusted based on the recognition working mode. This indicates the length of the future time domain and is used to define the future time range considered in the optimization.

[0011] In S5, the multi-objective rolling optimization process also includes the following constraints: The switching frequency is selected from a preset discrete candidate set; The grid-connected current of the converter does not exceed the set absolute limit; The junction temperature of the power devices in the converter shall not exceed the maximum allowable value.

[0012] In S6, feedback correction includes: Update the parameters of the grid impedance-related model based on the error between the actual measured state and the predicted state; Correct the switching loss model parameters; Adjust the boundary conditions for operation mode recognition.

[0013] The optimal switching frequency is achieved through a frequency selector. The frequency selector generates multiple clock frequencies based on the clock of the converter control system and selects the corresponding clock as the reference clock for the PWM counter according to the optimal frequency command, so as to complete the seamless switching of the switching frequency.

[0014] In S3, the prediction model is used to describe the relationship between the switching frequency and the bandwidth, current tracking dynamics, and switching losses of the converter control system.

[0015] In S4, a predictive model is used to predict the state of the converter in the future time domain, including: Predict the future state trajectory of the converter in the time domain; Evaluate the convergence trend of the current tracking error; Assess changes in the control bandwidth of the power grid system and the accumulation of switching losses.

[0016] In S3, the piecewise linear state-space model library contains multiple linearized models corresponding to different operating modes. Each model describes the mapping relationship between the switching frequency and the dynamic response and losses of the power grid system under the corresponding operating conditions.

[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. By real-time monitoring of grid voltage, grid-connected current, power, and stability indicators, the system accurately identifies its operating mode (e.g., oscillation risk, light load, normal, heavy load) and performs rolling optimization for future states based on a predictive model. Under light load or oscillation risk conditions, the system can automatically increase the switching frequency, expand the control bandwidth, and enhance system damping, thereby effectively suppressing oscillation instability. Under stable heavy load conditions, the system can reduce the switching frequency to decrease switching losses. This method achieves synergistic optimization of stability and efficiency, significantly improving the robustness of the converter under complex operating conditions. 2. Traditional switching frequency adjustment requires manual modification of the PWM counter period value and control parameters, which can easily cause current surges or control misalignment during the switching process. This invention achieves seamless switching between different switching frequencies through a frequency selector structure based on a frequency divider clock, without the need to modify the PWM counter register and control loop parameters. This significantly reduces the complexity of engineering implementation and code maintenance, and improves the reliability and smoothness of the system. 3. Feedback correction is performed in each control cycle. The model parameters (such as grid impedance, switching loss model, etc.) are corrected online by using the error between the actual state and the predicted state, and the pattern recognition boundary conditions are dynamically adjusted. This adaptive closed-loop mechanism enables the prediction model to continuously track system changes, improves the adaptability to different operating conditions and grid environments, and avoids the decline in control performance caused by model mismatch. 4. By constructing a cost function targeting current tracking error and switching losses, and dynamically adjusting the weighting coefficients according to the operating mode, this invention can simultaneously consider current tracking accuracy and system efficiency during the optimization process; it emphasizes stability in oscillation risk mode, balances tracking and losses in light load / normal mode, and prioritizes efficiency in heavy load mode, achieving multi-objective collaborative optimization; the entire control process is modularly designed into six modules: state measurement, pattern recognition, model prediction, rolling optimization, frequency application, and feedback correction, with clear logic, making it easy to implement in a digital signal processor (DSP) or field-programmable gate array (FPGA); the candidate switching frequency set adopts a discretized design, reducing the optimization computation burden and making it suitable for online real-time operation. Attached Figure Description

[0018] Figure 1This is a flowchart of a model-based adaptive frequency adjustment method for high-voltage multilevel switching.

[0019] Figure 2 This is a flowchart for changing the switching frequency. Detailed Implementation

[0020] The present invention will now be described in detail with reference to the accompanying drawings, but it should be noted that the implementation of the present invention is not limited to the following embodiments.

[0021] The following embodiments are implemented based on the technical solution of the present invention, providing detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the following embodiments. Unless otherwise specified, the methods used in the following embodiments are conventional methods. Example 1

[0022] See Figure 1 An adaptive adjustment method for the switching frequency of a high-voltage multilevel converter based on model prediction, specifically including: S1. Status Measurement and Acquisition; At time k, the current state variables of the high-voltage multilevel converter and its grid-connected system are acquired in real time through voltage sensors, current sensors, and digital observers, and the stability of the converter system is evaluated. The system stability evaluation is performed by calculating whether the current tracking error exceeds the warning threshold, analyzing whether there are significant oscillating frequency components in the grid-connected current spectrum (such as exceeding the set threshold in a specific frequency band through FFT analysis), and determining whether the grid oscillation flag is true. If any of the above indicators are abnormal, the system is considered to have an oscillation risk.

[0023] The specific state variables measured include: Grid voltage V (valid value); Grid-connected current A (valid value); Command current A (valid value); Output active power kW; Output reactive power kvar; Power grid oscillation flag obtained through current / power spectrum analysis: (0 indicates no oscillation, 1 indicates oscillation); Estimated grid impedance Ω; Current tracking error The formula is as follows: A; The system sampling period is set to μs.

[0024] S2. Operation mode recognition and model selection; Identify the operating mode of the system based on the current state variables, and select the corresponding prediction model: S21, if or If so, it is determined to be an oscillation risk pattern; in: This indicates the high threshold for warning of current tracking error. A; S22, if And it is not in an oscillation risk mode, that is If so, it is determined to be light load mode; in: Indicates the light load power threshold. KW; S23, if If so, it is determined to be normal mode; in: Indicates the heavy load power threshold. KW.

[0025] S24, if If so, it is determined to be in overload mode; S25, if If so, it is determined to be an overload / failure mode; in: Indicates the maximum current protection threshold. A.

[0026] S26. Model selection; According to S21~S25, KW, A, Therefore, it is determined to be in normal mode; Select the corresponding model from the pre-established library of piecewise linear state-space models. This model describes the switching frequency under normal operating conditions. The state-space equation expressing the quantitative relationship between system control bandwidth, current tracking dynamic response speed, and switching losses is as follows: .

[0027] in: This represents the system state vector at time k+1; This represents the state transition matrix related to the switching frequency; This represents the system state vector at time k; This represents the input matrix related to the switching frequency; This represents the control input vector at time k; state variables Including current tracking error, system phase margin estimation, etc., coefficient matrix , Sui The changes are piecewise linearized.

[0028] S3. Future state prediction based on candidate frequency set; Current state: and candidate switching frequency control sets Input them into the selected prediction model respectively; in: A (current tracking error); A (grid-connected current); KW (output active power); kvar (output reactive power); Ω (power grid impedance); (Power grid oscillation flag); V (mains voltage); A (command current); (Candidate switching frequency set); For example, candidate switching frequency control sets kHz. These are input into the model, with preset future... System state trajectory within one step (corresponding to 1ms): ; Key areas of focus: Convergence trend of current tracking error; Estimated changes in system bandwidth; The sum of switching losses and total losses.

[0029] S4, Multi-objective Scrolling Optimization; The optimizer in the candidate frequency set Finding the optimal switching frequency To minimize the multi-objective cost function J.

[0030] Define a multi-objective cost function: in: This indicates that in the prediction time domain j=1 to Summation operation on; Based on candidate frequency at time k The predicted current tracking error at time k+j; Based on candidate frequency at time k Predicted switching losses at time k+j; and These are the weighting coefficients. , (Focusing on tracking performance in normal mode); The weighting coefficients are dynamically adjusted according to the identified working mode in order to achieve different control objectives; The constraints are: in: Indicates the absolute limit of grid-connected current. A; This indicates the junction temperature of the IGBT power devices in the converter; Indicates the maximum allowable junction temperature. ℃; Represents the set of candidate switching frequencies; The cost function is minimized at kHz, therefore we choose... kHz; For example, calculate separately kHz The value (J(f(m)) is the cost function value corresponding to the switching frequency f(m), simply referred to as the frequency cost value), is compared to obtain Minimum: ; At other frequencies All values ​​are greater than this value.

[0031] The constraints mean that the switching frequency can only be selected from the candidate set, and the current should not exceed the absolute limit. junction temperature Not exceeding the allowable value .

[0032] S5, Optimal Frequency Application; The optimized switching frequency obtained The kHz frequency is used in the PWM control stage of the converter. By adjusting the reference clock of the PWM counter through a frequency selector, the switching frequency can be smoothly switched.

[0033] S6. Feedback correction and model update; At time k+1, measure the new system state. And compared with the predicted value The prediction error is obtained through comparison. This error is then used to fine-tune and correct the piecewise linear model parameters online. Update grid impedance Relevant model parameters; Correct the switching loss model parameters to more accurately predict efficiency at different frequencies; Adjust the boundary conditions for pattern recognition (such as adjusting the light load threshold according to actual operation) to make the pattern division more consistent with the actual system.

[0034] Based on the prediction error, the model parameters Online corrections are performed, and the corrected model is used for prediction optimization in the next cycle to form adaptive closed-loop control.

[0035] See Figure 2 System clock MHz is divided to obtain different clock frequencies: ; The frequency selector obtains the optimal frequency command based on the model prediction method. The actual clock frequency is selected to obtain reference clocks for PWM counters of different frequencies. The synchronization control instruction determines when the optimal frequency instruction is applied to the frequency selector. Different reference clock frequencies are applied to the PWM counters, resulting in carrier waves of different frequencies, ultimately leading to different switching frequencies. Therefore, changing the switching frequency does not require changing the PWM counter's count value or modifying other control parameters, simplifying the code writing complexity.

[0036] This invention accurately identifies system operating modes (such as oscillation risk, light load, normal, heavy load, etc.) by real-time detection of state variables such as grid voltage, grid-connected current, power, and stability indicators, and performs rolling optimization of future states based on a predictive model. Under light load or oscillation risk conditions, the system can automatically increase the switching frequency, expand the control bandwidth, and enhance system damping, thereby effectively suppressing oscillation instability. Under stable heavy load conditions, the system can reduce the switching frequency to reduce switching losses. This method achieves synergistic optimization of stability and efficiency, significantly improving the robustness of the converter under complex operating conditions. Traditional switching frequency regulation requires manual modification of the PWM counter period value and control parameters, which can easily cause current surges or control misalignment during the switching process. This invention, through a frequency selector structure, achieves seamless switching between different switching frequencies based on a frequency-divided clock, without requiring modification of the PWM counter register and control loop parameters. This significantly reduces the complexity of engineering implementation and code maintenance, and improves the reliability and smoothness of the system. Feedback correction is performed in each control cycle, using the error between the actual state and the predicted state to correct model parameters (such as grid impedance, switching loss model, etc.) online, and dynamically adjusting the pattern recognition boundary conditions. This adaptive closed-loop mechanism enables the prediction model to continuously track system changes, improving its adaptability to different operating conditions. The invention adapts well to various power grid environments, avoiding performance degradation due to model mismatch. By constructing a cost function targeting current tracking error and switching losses, and dynamically adjusting weighting coefficients based on operating modes, the invention can simultaneously consider current tracking accuracy and system efficiency during optimization. It prioritizes stability under oscillation risk modes, balances tracking and losses under light / normal load modes, and prioritizes efficiency under heavy load modes, achieving multi-objective collaborative optimization. The entire control process is modularly designed into six modules: state measurement, pattern recognition, model prediction, rolling optimization, frequency application, and feedback correction. The logic is clear and easy to implement in a digital signal processor (DSP) or field-programmable gate array (FPGA). The candidate switching frequency set is discretized, reducing the computational burden of optimization and making it suitable for online real-time operation.

Claims

1. A model-predictive-based adaptive switching frequency adjustment method for high-voltage multilevel converters, characterized in that, include: S1. Real-time measurement of the state variables of the power grid system connected to the high-voltage multilevel converter; State variables include grid voltage, grid-connected current, commanded current, output active power, output reactive power, grid oscillation flag, grid impedance, and current tracking error; S2. Identify the converter's operating mode based on status variables; Operating modes include: oscillation risk mode, light load mode, normal mode, heavy load mode, and overload or fault mode; S3. Based on the identified operating mode, select the corresponding prediction model from the pre-established piecewise linear state-space model library; S4. Based on the current state variables and the preset candidate switching frequency set, the state of the converter in the future time domain is predicted using a prediction model. S5. Through multi-objective rolling optimization, select the optimal switching frequency that minimizes the multi-objective cost function from the candidate switching frequency set. S6. Apply the optimal switching frequency to the PWM control loop of the converter, and correct the prediction model parameters online through feedback correction in the next sampling cycle to form an adaptive closed-loop control.

2. The adaptive adjustment method for switching frequency of a high-voltage multilevel converter based on model prediction according to claim 1, characterized in that, The prediction model described is a piecewise linear state-space model that describes the relationship between switching frequency and the power grid system control bandwidth, current tracking dynamic characteristics, and switching losses.

3. The adaptive adjustment method for switching frequency of a high-voltage multilevel converter based on model prediction according to claim 1, characterized in that, In S2: The oscillation risk mode occurs when the converter's current tracking error exceeds the warning threshold, or when the grid oscillation flag is true. Light load mode is when the converter's output active power is lower than the light load power threshold and it is not in an oscillation risk mode; In normal mode, the converter's output active power is between the light load power threshold and the heavy load power threshold. Heavy load mode is when the converter's output active power exceeds the heavy load power threshold. Overload or fault mode occurs when the grid-connected current of the converter exceeds the maximum current protection threshold.

4. The adaptive adjustment method for switching frequency of a high-voltage multilevel converter based on model prediction according to claim 1, characterized in that, In S5, the expression for the multi-objective cost function is: ； in: This represents summation, in the future time domain from j=1 to... The above was carried out; Indicates the candidate frequency at time k The predicted current tracking error at time k+j; Indicates the candidate frequency at time k Predicted switching losses at time k+j; and This represents the weighting coefficient, which is dynamically adjusted based on the recognition working mode. This indicates the length of the future time domain and is used to define the future time range considered in the optimization.

5. The adaptive adjustment method for switching frequency of a high-voltage multilevel converter based on model prediction according to claim 4, characterized in that, The multi-objective rolling optimization process also includes the following constraints: The switching frequency is selected from a preset discrete candidate set; The grid-connected current of the converter does not exceed the set absolute limit; The junction temperature of the power devices in the converter shall not exceed the maximum allowable value.

6. The adaptive adjustment method for switching frequency of a high-voltage multilevel converter based on model prediction according to claim 1, characterized in that, In S6, the feedback correction includes: Update the parameters of the grid impedance-related model based on the error between the actual measured state and the predicted state; Correct the switching loss model parameters; Adjust the boundary conditions for operation mode recognition.

7. The adaptive adjustment method for switching frequency of a high-voltage multilevel converter based on model prediction according to claim 1, characterized in that, The optimal switching frequency is achieved through a frequency selector. The frequency selector generates multiple clock frequencies based on the clock of the converter control system and selects the corresponding clock as the reference clock for the PWM counter according to the optimal frequency command, so as to complete the seamless switching of the switching frequency.

8. The adaptive adjustment method for switching frequency of a high-voltage multilevel converter based on model prediction according to claim 1, characterized in that, In S3, the prediction model is used to describe the relationship between the switching frequency and the bandwidth, current tracking dynamics, and switching losses of the converter control system.

9. The adaptive adjustment method for switching frequency of a high-voltage multilevel converter based on model prediction according to claim 1, characterized in that, In S4, the prediction of the converter's state in the future time domain using a prediction model includes: Predict the future state trajectory of the converter in the time domain; Evaluate the convergence trend of the current tracking error; Assess changes in the control bandwidth of the power grid system and the accumulation of switching losses.

10. The adaptive adjustment method for switching frequency of a high-voltage multilevel converter based on model prediction according to claim 1, characterized in that, In S3, the piecewise linear state-space model library contains multiple linearized models corresponding to different operating modes. Each model describes the mapping relationship between the switching frequency and the dynamic response and losses of the power grid system under the corresponding operating conditions.

Images