Llc resonant converter and data-driven control method, system, chip thereof

Abstract

The application provides an LLC resonant converter and a data-driven control method, system and chip thereof. The method comprises: collecting the output voltage, load current and rectifier bridge current of the LLC resonant converter in a current control period; calculating a reference frequency and a calculation step, and constructing a discrete frequency control set with the reference frequency and the step; determining time-varying parameters in a first-order discrete data model based on a recursive least square method; predicting the output voltage in a next control period according to the time-varying parameters corresponding to the current control period, the load current and the rectifier bridge current; calculating a corresponding value function according to the predicted output voltage and a reference voltage for each candidate frequency in the discrete frequency control set, and determining the candidate frequency that minimizes the value function as an optimal switching frequency; and generating a PWM signal according to the optimal switching frequency to drive the switching tube of the LLC resonant converter. The application improves the robustness, response efficiency and accuracy of the LLC resonant converter under parameter mismatch.

Description

Technical Field

[0001] This invention relates to the field of power electronic control technology, specifically to an LLC resonant converter and its data-driven control method, system, and chip. Background Technology

[0002] With the rapid development of data centers, new energy power generation, and electric vehicle charging, the demand for high-power-density, high-efficiency, and high-reliability DC-DC converters is becoming increasingly urgent. LLC resonant converters, due to their ability to achieve zero-voltage turn-on of the primary-side switches and zero-current turn-off of the secondary-side rectifiers, possess inherent advantages such as high efficiency, high power density, and low electromagnetic interference, making them one of the preferred topologies for medium- and high-power power supply solutions. However, the LLC resonant converter is a high-order nonlinear system, and its dynamic characteristics change significantly with variations in input voltage, load, and its own parameters, posing a significant challenge to the design of control strategies.

[0003] To improve the dynamic performance of LLC resonant converters, advanced algorithms such as model control have been introduced in existing technologies. These methods establish a mathematical model of the converter, predict the output voltage at future moments, and optimize the switching frequency online based on a value function, thereby achieving a faster dynamic response speed than traditional proportional-integral control.

[0004] However, existing model control strategies heavily rely on the accurate mathematical model of the system. In actual operation, the resonant parameters of the converter (such as inductance and capacitance) will drift with temperature, aging, and operating conditions, leading to model mismatch and causing a decline in the performance of the control system or even instability. Summary of the Invention

[0005] In view of this, it is necessary to provide an LLC resonant converter and its data-driven control method, system, and chip to solve the technical problems existing in the prior art, such as the heavy reliance on accurate mathematical models for model control of LLC resonant converters, poor parameter robustness, and the difficulty in balancing dynamic and steady-state performance.

[0006] To address the aforementioned technical problems, in a first aspect, the present invention provides a data-driven control method for an LLC resonant converter, comprising: The output voltage, load current, and rectifier bridge current of the LLC resonant converter are collected during the current control cycle. The reference frequency is calculated by a PI controller based on the error between the reference voltage and the output voltage. The step size is calculated based on the output voltage error, and a discrete frequency control set is constructed using the reference frequency and the step size; the discrete frequency control set includes at least three candidate frequencies. The time-varying parameters in the first-order discrete data model are determined based on the recursive least squares method. The first-order discrete data model is used to describe the relationship between the change in the output voltage and the load current and the rectifier bridge current. Based on the time-varying parameters corresponding to the current control cycle, the load current, and the rectifier bridge current, predict the output voltage of the next control cycle; For each candidate frequency in the discrete frequency control set, a corresponding value function is calculated based on the predicted output voltage and reference voltage, and the candidate frequency that minimizes the value function is determined as the optimal switching frequency. The PWM signal generated according to the optimal switching frequency drives the switching transistor of the LLC resonant converter.

[0007] In one possible implementation, at least three candidate frequencies in the discrete frequency control set are symmetrically distributed, and the at least three candidate frequencies include the reference frequency and candidate frequencies located on both sides of the reference frequency and differing from the reference frequency by a finite number of integer multiples of step size.

[0008] In one possible implementation, the calculation of the step size includes: The first coefficient is determined based on the processor clock cycle and the switching cycle; The second coefficient is calculated based on the adjustment coefficient and the output voltage error after limiting; if the absolute value of the difference between the reference voltage and the output voltage is less than the maximum error voltage, the output voltage error after limiting is determined to be the absolute value of the difference; if the absolute value of the difference is not less than the maximum error voltage, the output voltage error after limiting is determined to be the maximum error voltage. The step size is calculated based on the first coefficient and the second coefficient.

[0009] In one possible implementation, the first-order discrete data model is represented as: ; in, The change in the output voltage. The load current, The rectifier bridge current is... The first time-varying parameter identified in the k-th period, The second time-varying parameter is obtained for the kth period.

[0010] In one possible implementation, determining the time-varying parameters in the first-order discrete data model based on the recursive least squares method includes: The gain matrix of the current control cycle is calculated based on the forgetting factor, the covariance matrix of the previous control cycle, and the input vector of the current control cycle; the input vector is calculated based on the load current and rectifier bridge current collected in the current control cycle. The covariance matrix of the current control cycle is calculated based on the covariance matrix of the previous control cycle, the gain matrix of the current control cycle, and the input vector of the current control cycle. The parameter vector of the current control cycle is calculated based on the parameter vector of the previous control cycle, the gain matrix of the current control cycle, the input vector of the current control cycle, and the parameter vector of the previous control cycle. The first time-varying parameter and the second time-varying parameter are determined based on the parameter vector of the current control cycle.

[0011] In one possible implementation, predicting the output voltage of the next control cycle based on the time-varying parameters corresponding to the current control cycle, the load current, and the rectifier bridge current includes: Based on the first time-varying parameter, the second time-varying parameter, the load current, and the rectifier bridge current of the current control cycle, the predicted voltage change for the next control cycle is calculated. The output voltage of the next control cycle is calculated based on the output voltage of the current control cycle and the predicted voltage change of the next control cycle.

[0012] In one possible implementation, the value function is expressed as: ; ; in, This is the value function value, used to evaluate the merits of candidate frequencies. For the reference voltage tracking error term, evaluate the deviation between the predicted output voltage and the reference voltage. For the frequency variation smoothing term, evaluate the deviation between the candidate frequency and the optimal frequency of the previous cycle. and They are respectively and Weighting factors This is the reference voltage for the (k+1)th control cycle (usually a constant value). The predicted output voltage for the (k+1)th control cycle. The reference frequency output by the PI controller. is the candidate frequency for the k-th period.

[0013] Secondly, the present invention also provides a data-driven control system for an LLC resonant converter, comprising: The sampling module is used to acquire the output voltage, load current, and rectifier bridge current of the LLC resonant converter during the current control cycle. The PI control module is used to calculate the reference frequency based on the error between the reference voltage and the output voltage using a PI controller. A frequency set construction module is used to calculate the step size based on the output voltage error and construct a discrete frequency control set with the reference frequency and the step size; the discrete frequency control set includes at least three candidate frequencies. The online parameter identification module is used to determine the time-varying parameters in the first-order discrete data model based on the recursive least squares method. The first-order discrete data model is used to describe the relationship between the change in the output voltage and the load current and the rectifier bridge current. The voltage prediction module is used to predict the output voltage of the next control cycle based on the time-varying parameters corresponding to the current control cycle, the load current, and the rectifier bridge current. The rolling optimization module is used to calculate the corresponding value function for each candidate frequency in the discrete frequency control set based on the predicted output voltage and reference voltage, and determine the candidate frequency that minimizes the value function as the optimal switching frequency. The PWM generation module is used to drive the switching transistors of the LLC resonant converter based on the PWM signal generated according to the optimal switching frequency.

[0014] Thirdly, the present invention also provides a control chip configured to perform the steps of the data-driven control method for the LLC resonant converter described in any of the above implementations.

[0015] Fourthly, the present invention also provides an LLC resonant converter, including the control chip described above.

[0016] The beneficial effects of this invention are as follows: The data-driven control method for LLC resonant converters provided by this invention first determines the time-varying parameters in the first-order discrete data model based on the recursive least squares method. Through rolling optimization of the value function, the optimal frequency is selected from the discrete frequency set in each control cycle, enabling rapid response to load or input voltage changes. Furthermore, since the time-varying parameters in the first-order discrete data model are determined based on the recursive least squares method, and the optimal frequency is selected from the discrete frequency set in each control cycle through rolling optimization of the value function, precise modeling is not required. Physical parameters such as Lr, Lm, and Cr of the LLC resonant cavity are completely unnecessary, and the constructed discrete frequency set contains only a small number of candidate frequencies. This fundamentally eliminates the dependence on precise mathematical models, avoids complex nonlinear optimization solutions, and significantly reduces the difficulty and workload of parameter tuning. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A schematic flowchart of an embodiment of the data-driven control method for an LLC resonant converter provided by the present invention; Figure 2 This is a schematic diagram of an embodiment of the present invention for calculating the step size; Figure 3 For the present invention Figure 1 A schematic diagram of an embodiment of S104; Figure 4 A control flow diagram of an embodiment of the data-driven control method for an LLC resonant converter provided by the present invention; Figure 5 The output voltage waveform of the data-driven control method for the LLC resonant converter provided by the present invention when the parameter mismatch is -50% and the reference voltage is 20V; Figure 6 The output voltage waveform of the data-driven control method for the LLC resonant converter provided by the present invention when the parameter mismatch is -50% and the reference voltage is 25V; Figure 7 A comparison of the output voltage error between the data-driven control method for the LLC resonant converter provided by this invention and the traditional MPC method when the parameter mismatch is -50% and the reference voltage is 20V; Figure 8 A comparison of the output voltage error between the data-driven control method for the LLC resonant converter provided by this invention and the traditional MPC method when the parameter mismatch is -50% and the reference voltage is 25V. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] In the description of the embodiments of the present invention, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0021] The terms "first," "second," etc., used in the embodiments of this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature.

[0022] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0023] Before demonstrating the embodiments, the following terms will be explained.

[0024] LLC (Inductor-Inductor-Capacitor Resonant Converter): The two L's represent the resonant inductance Lr and the magnetizing inductance Lm, respectively, and C represents the resonant capacitor Cr.

[0025] PI (Proportional-Integral Controller): This is a classic closed-loop control algorithm. Based on the error between the reference voltage and the actual output voltage, the control quantity (the reference frequency in this scheme) is calculated through the proportional term (P) and the integral term (I).

[0026] This invention provides an LLC resonant converter and its data-driven control method, system, and chip, which are described below.

[0027] Figure 1 A schematic flowchart of an embodiment of the data-driven control method for an LLC resonant converter provided by the present invention is shown below. Figure 1 As shown, the data-driven control method for the LLC resonant converter includes: S101. Acquire the output voltage, load current, and rectifier bridge current of the LLC resonant converter during the current control cycle.

[0028] It should be noted that at the beginning of each control cycle, the digital controller (such as an NXP series processor) acquires the output voltage, load current, and rectifier bridge current of the LLC resonant converter through an analog-to-digital converter (ADC). The load current can be obtained through a current sensor or an output sampling resistor, while the rectifier bridge current can be obtained by sampling the current on the output side of the rectifier bridge.

[0029] S102. The reference frequency is calculated by a PI controller based on the error between the reference voltage and the output voltage.

[0030] It should be noted that the error between the set reference voltage and the current output voltage is calculated as follows: (Equation 1); in, For output reference voltage, The actual output voltage sampled in the k-th control cycle. This is the maximum error voltage (used to limit voltage error). This represents the output voltage error after amplitude limiting. The error is then fed into a PI controller. The PI controller outputs a proportional term Kp·e(k) and an integral term Ki·Σe(k). These two terms are added together to obtain the reference frequency f_center. The parameters Kp and Ki are adjusted empirically.

[0031] S103. Calculate the step size based on the output voltage error, and construct a discrete frequency control set using the reference frequency and the step size; the discrete frequency control set includes at least three candidate frequencies.

[0032] It should be noted that the frequency adjustment step size is adaptively calculated based on the magnitude of the error between the output voltage and the reference voltage in the current control cycle. This step size changes nonlinearly with the voltage error; it is appropriately increased when the error is large to accelerate the dynamic response, and decreased when the error is small to ensure steady-state accuracy. Subsequently, using the reference frequency output by the PI controller as the center, and combining this adaptive step size, a discrete frequency control set containing at least three candidate frequencies is constructed.

[0033] S104. Determine the time-varying parameters in the first-order discrete data model based on the recursive least squares method. The first-order discrete data model is used to describe the relationship between the change in the output voltage and the load current and the rectifier bridge current.

[0034] It should be noted that within each control cycle, recursive least squares (RLS) with a forgetting factor is used to identify the time-varying parameters in the first-order discrete data model online in real time. This first-order discrete data model is used to characterize the dynamic behavior of the LLC resonant converter near the current operating point, specifically describing the linear mapping relationship between the output voltage change and the load current and rectifier bridge current. By recursively updating the parameter estimates, this method can adapt to model changes caused by factors such as load disturbances, temperature drift, or component aging, thereby achieving accurate capture and modeling of the system's dynamic characteristics without relying on any physical resonant parameters (such as resonant inductance, magnetizing inductance, and resonant capacitor).

[0035] S105. Based on the time-varying parameters corresponding to the current control cycle, the load current, and the rectifier bridge current, predict the output voltage of the next control cycle.

[0036] It should be noted that: by utilizing the time-varying parameters obtained online through the recursive least squares method in the current control cycle, combined with the load current and rectifier bridge current collected in the current control cycle, the output voltage of the next control cycle is predicted based on the constructed first-order discrete data model. This prediction process is completely independent of any physical parameters of the LLC resonant converter (such as resonant inductance, magnetizing inductance, resonant capacitor, etc.), achieving model-free voltage prediction and providing accurate target reference values ​​for subsequent frequency rolling optimization.

[0037] S106. For each candidate frequency in the discrete frequency control set, calculate the corresponding value function based on the predicted output voltage and reference voltage, and determine the candidate frequency that minimizes the value function as the optimal switching frequency.

[0038] It should be noted that for each candidate frequency in the discrete frequency control set, the corresponding value function is calculated by combining the output voltage predicted for the next moment in the current control cycle with the set reference voltage. By comparing the value function values ​​of all candidate frequencies, the candidate frequency that minimizes the value function is selected as the optimal switching frequency for the current control cycle. This ensures that the output voltage quickly tracks the reference value while also taking into account the smooth change of the switching frequency.

[0039] S107. Drive the switching transistor of the LLC resonant converter with the PWM signal generated according to the optimal switching frequency.

[0040] It should be noted that, based on the optimal switching frequency determined by minimizing the value function, the PWM modulation module inside the digital controller (such as a microprocessor or DSP) is configured to generate a complementary PWM signal that matches the optimal switching frequency. This PWM signal, after power amplification and level conversion by a drive circuit (such as an isolated gate driver), is applied to the control terminal of the primary-side switching transistor (typically a MOSFET or IGBT) of the LLC resonant converter to control the switching transistor's turn-on and turn-off timing, thereby achieving closed-loop regulation of the converter's output voltage. By repeating the above process in each control cycle, the switching frequency is ensured to always approach the optimal value under the current operating conditions, enabling the system to possess both fast dynamic response and stable steady-state performance.

[0041] In summary, the data-driven control method for LLC resonant converters provided in this embodiment of the invention first determines the time-varying parameters in the first-order discrete data model based on the recursive least squares method. Through rolling optimization using a value function, the optimal frequency is selected from the discrete frequency set in each control cycle, enabling rapid response to load or input voltage changes. Furthermore, since the time-varying parameters in the first-order discrete data model are determined based on the recursive least squares method, and the selection of the optimal frequency from the discrete frequency set in each control cycle through rolling optimization using a value function does not require precise modeling, it completely eliminates the need for physical parameters such as Lr, Lm, and Cr of the LLC resonant cavity. Moreover, the constructed discrete frequency set contains only a small number of candidate frequencies, fundamentally eliminating the dependence on precise mathematical models, avoiding complex nonlinear optimization solutions, and significantly reducing the difficulty and workload of parameter tuning.

[0042] In some embodiments of the present invention, at least three candidate frequencies in the discrete frequency control set are symmetrically distributed, and the at least three candidate frequencies include the reference frequency and candidate frequencies located on both sides of the reference frequency and differing from the reference frequency by a finite number of integer multiples of step size.

[0043] It should be noted that the discrete frequency control set expands symmetrically to both sides of the reference frequency f_center output by the PI controller, forming a set containing an odd number of candidate frequencies. Specifically, the set includes the reference frequency f_center itself, and several candidate frequencies located to the left (below the reference frequency) and to the right (above the reference frequency) of f_center. The number of candidate frequencies on the left and right sides is equal, and the frequency difference between each candidate frequency and the reference frequency f_center is an integer multiple of the step size. The total number of candidate frequencies is 2N+1, where N is a positive integer, and the discrete frequency control set contains at least 3 candidate frequencies (when N=1).

[0044] In this embodiment, the at least three candidate frequencies in the discrete frequency control set are symmetrically distributed and finite, so that each rolling optimization only requires evaluating a very small number of frequency points. Compared with unlimited continuous frequency adjustment or large-scale search, this significantly reduces the number of calculations of the value function, making it suitable for real-time execution by a low-cost digital controller. Furthermore, the reference frequency is located at the center and extends symmetrically on both sides, ensuring that the controller's adjustment range and sensitivity remain consistent regardless of whether the switching frequency needs to be increased or decreased. This avoids the unidirectional adjustment bias that may be caused by the asymmetric set, giving the system the same dynamic response capability under both positive and negative errors, and ensuring the unbiasedness of frequency adjustment. Furthermore, since the symmetrical structure depends only on the step size, and the step size can be adaptively adjusted in real time according to the voltage error, the discrete frequency control set can dynamically shrink or expand. When the error is large, the step size is large, covering a wider frequency range to quickly approach the target; when the error is small, the step size is small, achieving fine adjustment while maintaining symmetry, taking into account both dynamic and steady-state performance.

[0045] In some embodiments of the present invention, such as Figure 2 As shown, the calculation of the step size includes: S201. Determine the first coefficient based on the processor clock cycle and the switching cycle; S202. Calculate the second coefficient based on the adjustment coefficient and the output voltage error after limiting; if the absolute value of the difference between the reference voltage and the output voltage is less than the maximum error voltage, determine the output voltage error after limiting as the absolute value of the difference; if the absolute value of the difference is not less than the maximum error voltage, determine the output voltage error after limiting as the maximum error voltage. S203. The step size is calculated based on the first coefficient and the second coefficient.

[0046] It should be noted that in digital controllers (such as NXP series DSPs or MCUs), two key timing parameters are pre-acquired: the processor clock cycle and the switching cycle. Using the reference frequency output by the PI controller as the center frequency, a discrete frequency control set is established. This set contains the reference frequency and a finite number of adjacent frequency values, which can be expressed as: [f_center - Δf, f_center, f_center + Δf], where Δf is the adaptive frequency (i.e., the step size). The step size can be calculated using the following formula: (Equation 2); in, Step size, For the switching cycle, For the NXP processor's clock cycle, This is an adjustment coefficient (used to adjust the degree of influence of voltage error on the step size). This represents the output voltage error after amplitude limiting.

[0047] In this embodiment, a large step size and wide frequency adjustment range are used to quickly reduce errors when there are large errors; a small step size and fine-tuned frequency are used when there are small errors to avoid overshoot and oscillation, achieving zero steady-state error tracking. Furthermore, the square of the error is used to make the step size's sensitivity to error increase sharply as the error increases, significantly shortening the dynamic recovery time.

[0048] In some embodiments of the present invention, the first-order discrete data model is represented as: ; in, The change in the output voltage. The load current, The rectifier bridge current is... The first time-varying parameter identified in the k-th period, The second time-varying parameter is obtained for the kth period.

[0049] It should be noted that the LLC resonant converter is a seventh-order nonlinear system, and its accurate modeling is very complex. The full-bridge converter, resonant circuit, and rectifier diodes in the LLC resonant converter can be considered as a single system composed of an equivalent inductor. and output capacitor The filter is connected to a voltage source controlled by a switching frequency. The seventh-order model can be simplified to a second-order model by using the rectifier bridge current.

[0050] (Equation 3); in, This is the rectifier current. The actual output voltage sampled in the k-th control cycle. For equivalent inductance, It is a controlled voltage source, the value of which is controlled by the switching frequency. For output capacitor, For output load current, This refers to the transformer turns ratio (the ratio of the primary winding to the secondary winding). It is the magnetizing inductor (magnetizing inductor of LLC resonant converter). This is the resonant inductor (the resonant inductor of an LLC resonant converter).

[0051] Due to the controlled voltage source The transfer function of the voltage source with frequency f is nonlinear, and Taylor series can be used to apply this to the controlled voltage source. After linearization, the following formula is obtained: (Equation 4); in, It is a controlled voltage source. This is the primary voltage of the transformer. This refers to the transformer turns ratio (the ratio of the primary winding to the secondary winding). For switching frequency, These are the linearization coefficients (obtained from Taylor expansion).

[0052] Therefore, the seventh-order model can be simplified to a second-order linear model: (Equation 5); Although the second-order linear model reduces the order, it still contains... , Given these parameters, discretizing the above second-order linear model yields the following second-order difference equation: (Formula 6); in, Let be the rectifier bridge current in the kth control cycle. This represents the rectifier bridge current during the (k+1)th control cycle. The output voltage of the kth control cycle. The output voltage of the (k+1)th control cycle. For the controlled voltage source in the k-th control cycle, This is the output load resistor.

[0053] Analysis of the second-order discrete mathematical model of the LLC resonant converter as shown in Equation (7) reveals that the load current and rectifier bridge current directly affect the output voltage change, while the frequency indirectly affects the output voltage change by influencing the rectifier bridge current. To further reduce computational complexity and facilitate data-driven operation, a first-order discrete data model is constructed to describe the system dynamics. Therefore, the second-order difference equation can be further simplified, and the first-order discrete data model of the LLC resonant converter is constructed as shown in Equation 7 below: (Equation 7); (Equation 8); in, The first time-varying parameter identified in the k-th control cycle. The second time-varying parameter identified in the k-th control cycle. and This is the set of unknown and disturbance terms in the system. Although the first-order discrete data model shown in Equation 7 above does not explicitly include the switching frequency, the time-varying parameters... and The effects of frequency changes and system operating point are implicitly included, and the effect of frequency regulation can be captured through online updates in each control cycle.

[0054] In this embodiment, the first-order discrete data model of the LLC resonant converter does not contain any LLC resonant cavity parameters (Lr, Lm, Cr), nor does it require information such as transformer turns ratio and equivalent resistance. Even if these parameters drift due to temperature, aging, or batch differences, the first-order discrete data model can still achieve high-performance model-free predictive control of the LLC resonant converter by adaptively tracking with RLS, thus eliminating dependence on physical parameters and ensuring prediction accuracy. With extremely low computational cost and parameter tuning difficulty, this achieves high-performance model-free predictive control of the LLC resonant converter.

[0055] In some embodiments of the present invention, such as Figure 3 As shown, step S104 includes: S301. The gain matrix of the current control cycle is calculated based on the forgetting factor, the covariance matrix of the previous control cycle, and the input vector of the current control cycle; the input vector is calculated based on the load current and rectifier bridge current collected in the current control cycle. S302. Calculate the covariance matrix of the current control cycle based on the covariance matrix of the previous control cycle, the gain matrix of the current control cycle, and the input vector of the current control cycle. S303. Calculate the parameter vector of the current control cycle based on the parameter vector of the previous control cycle, the gain matrix of the current control cycle, the input vector of the current control cycle, and the parameter vector of the previous control cycle. S304. Determine the first time-varying parameter and the second time-varying parameter based on the parameter vector of the current control cycle.

[0056] It should be noted that: due to the first time-varying parameter Second time-varying parameter Since the parameters are time-varying, recursive least squares (RLS) is used for online real-time identification within each control cycle. The core of RLS lies in recursively updating the parameter estimates based on the latest input-output data of the system, thereby minimizing the prediction error. The parameter identification process of RLS is as follows: Equation 8 above can be expressed in matrix form as shown in Equation 9 below: (Equation 9); (Equation 10); in, Let T be the input vector (data vector) and T be the transpose symbol. For parameter vectors, The output quantity is the change in output voltage ΔVo(k) during the k-th control cycle.

[0057] According to Equation 4, the parameter identification process can be expressed as: (Equation 11); in, This is the gain matrix (Kalman gain) for the k-th control cycle. Let be the covariance matrix of the k-th control cycle. Let be the covariance matrix of the (k-1)th control cycle. Let be the input vector for the k-th control cycle. Let be the parameter vector for the k-th control cycle. Let be the parameter vector for the (k-1)th control cycle. Forgetting factor, It is the output quantity (i.e., the output voltage change ΔVo(k) in the kth control cycle).

[0058] Parameter vector of the current control cycle It is a 2×1 column vector, whose first element is the first time-varying parameter. The second element is the second time-varying parameter. . and The initial values ​​are set empirically. Utilizing the first-order discrete data model parameters obtained through online identification, the so-called model-free voltage prediction does not rely on the physical parameter model. Instead, it uses a data-driven model to directly predict the output voltage at future moments, thus completely avoiding dependence on traditional physical model parameters (such as Lr, Lm, Cr, etc.).

[0059] In this embodiment, RLS updates parameters using the latest data in each control cycle, enabling rapid response to load changes, input voltage fluctuations, and resonant parameter drift, achieving real-time adaptive tracking. Furthermore, the RLS algorithm eliminates the need for offline modeling; it automatically adjusts parameters through continuous recursion, ensuring that the first-order discrete data model is always the optimal model in terms of minimizing prediction error, thus guaranteeing the performance of model-free predictive control under all operating conditions. Moreover, the forgetting factor combined with the RLS algorithm provides a solid data-driven foundation for model-free predictive control of LLC resonant converters, eliminating dependence on the converter's physical model.

[0060] In some embodiments of the present invention, step S105 includes: S401. Based on the first time-varying parameter, the second time-varying parameter, the load current, and the rectifier bridge current of the current control cycle, calculate the predicted voltage change for the next control cycle. S402. The output voltage of the next control cycle is calculated based on the output voltage of the current control cycle and the predicted voltage change of the next control cycle.

[0061] It should be noted that: based on the time-varying parameters identified in real time and ， Combination k Output voltage sampling at all times and voltage differential ,available k Output voltage at +1 time for: (Equation 12); in, This represents the predicted output voltage change during the (k+1)th control cycle. The output voltage for the predicted (k+1)th control cycle.

[0062] In this embodiment, the prediction formula for the output voltage of the next control cycle does not contain any physical parameters of the LLC resonant converter (Lr, Lm, Cr, n, etc.), but only relies on the time-varying parameters identified in real time. and Even if the resonant parameters drift, the prediction formula remains unchanged, and automatic compensation through RLS parameter updates fundamentally solves the model mismatch problem. Furthermore, since the time-varying parameters are optimal parameters in the sense of minimizing prediction error, they can truly reflect the system dynamics, thus improving the accuracy of the prediction. Further, the prediction of the future output voltage of the LLC resonant converter is accurate and computationally low.

[0063] In some embodiments of the present invention, the value function is expressed as: ; ; in, This is the value function value, used to evaluate the merits of candidate frequencies. For the reference voltage tracking error term, evaluate the deviation between the predicted output voltage and the reference voltage. For the frequency variation smoothing term, evaluate the deviation between the candidate frequency and the optimal frequency of the previous cycle. and They are respectively and Weighting factors This is the reference voltage for the (k+1)th control cycle (usually a constant value). The predicted output voltage for the (k+1)th control cycle. The reference frequency output by the PI controller. The candidate frequency for the k-th cycle is selected, and the candidate frequency that minimizes the value function is finally chosen as the switching frequency.

[0064] It should be noted that, in order to select the candidate frequency with the minimum value function from the discrete frequency control set as the optimal switching frequency, the value function is defined as: (Equation 13); (Equation 14); in, This is the value function value, used to evaluate the merits of candidate frequencies. For the reference voltage tracking error term, evaluate the deviation between the predicted output voltage and the reference voltage. For the frequency variation smoothing term, evaluate the deviation between the candidate frequency and the optimal frequency of the previous cycle. and They are respectively and Weighting factors This is the reference voltage for the (k+1)th control cycle (usually a constant value). The predicted output voltage for the (k+1)th control cycle. The reference frequency output by the PI controller. The candidate frequency for the k-th cycle is selected, and the candidate frequency that minimizes the value function shown in Equation 13 is chosen as the optimal switching frequency.

[0065] In this embodiment, the voltage prediction calculation for the future control cycle proposed by Equation 13 does not require any model parameters, thus achieving model-free voltage prediction calculation and improving the robustness of the system. Furthermore, the value function shown in Equation 13 simultaneously considers voltage tracking accuracy and frequency smoothness, and by adjusting the weighting factors, it can flexibly meet the needs of different application scenarios. Moreover, since the discrete frequency control set contains only a finite number (usually 3-7) of candidate frequencies, the value function calculation only involves addition and multiplication, requiring no iterative optimization or differentiation. This simplifies the calculation, enhances real-time performance, and provides an efficient, robust, and easily implemented frequency selection mechanism for LLC resonant converters. It serves as a crucial bridge connecting model-free voltage prediction and PWM output, ensuring excellent system performance under both dynamic and steady-state conditions.

[0066] The specific steps of the technical solution adopted in this invention are as follows: The PI controller outputs a reference frequency as the center frequency, and a discrete frequency control set is constructed based on the center frequency; wherein, the parameters Kp and Ki are adjusted empirically, and the control frequency of the predictive controller is synchronized with the PI controller. The model-free predictive controller is responsible for improving the robustness and dynamic performance of the system. Figure 4 As shown, Figure 4 The following functional modules are included from top to bottom and left to right: The main circuit of the LLC resonant converter, located on the left side of the block diagram, includes primary-side switching transistors (Q1-Q4, full-bridge or half-bridge), resonant cavities (Lr, Lm, Cr), a transformer, a secondary-side rectifier bridge (diode or synchronous rectifier), and an output filter capacitor Co. This module receives PWM drive signals and outputs a DC voltage Vo.

[0067] Signal sampling module: Three sampling branches are drawn from the main circuit, namely output voltage Vo(k) sampling (via voltage divider or isolation sensor), load current IR(k) sampling (current sensor in series in the output circuit), and rectifier bridge current IBr(k) sampling (current sensor at the output end of the secondary rectifier bridge). The sampled values ​​are sent to the controller.

[0068] PI controller: Input reference voltage Vref and error between Vo(k), output reference frequency f_center.

[0069] Adaptive step size calculation and discrete frequency set construction module: Receives the output voltage error and hardware parameters (Tc, Ts), calculates the step size Δf, and generates a candidate frequency set {f_center-Δf,f_center,f_center+Δf} with f_center as the center.

[0070] The RLS online parameter identification module takes the current load current IR(k), rectifier bridge current IBr(k), and output voltage change ΔVo(k)=Vo(k)-Vo(k-1) as input for the current cycle and updates the parameter vector in real time using a recursive least squares method with a forgetting factor.

[0071] Model-free voltage prediction module: Using the RLS output A(k) and B(k) and the currently sampled IR(k) and IBr(k), calculate the predicted voltage change ΔVo_pred=A(k)·IR(k)+B(k)·IBr(k), and then add it to the current Vo(k) to obtain the predicted output voltage Vop(k+1) for the next cycle.

[0072] Value function calculation and rolling optimization module: Receives Vop(k+1), candidate frequency set, and the optimal frequency f_prev_opt of the previous cycle. Calculates the value function g for each candidate frequency and selects the candidate frequency that minimizes g as the optimal switching frequency f_opt.

[0073] PWM generation module: Based on the timer configuration of f_opt, it generates a complementary PWM signal with dead time to drive the switching transistors of the LLC main circuit.

[0074] The modules are connected via a data bus or hardwire to form a closed-loop control circuit.

[0075] exist Figure 4 In the control flowchart of the LLC resonant converter shown, the dashed lines represent the drive signals of the four MOSFETs on the primary side of the LLC resonant converter, typically labeled Q1, Q2, Q3, and Q4 in a full-bridge topology. The red dashed lines represent the control signal path from the PWM generation module output to the gate of the MOSFETs, with the following meanings: Q1 and Q2 form one bridge arm (upper MOSFET Q1, lower MOSFET Q2), with complementary drive signals (with dead time); Q3 and Q4 form another bridge arm (upper MOSFET Q3, lower MOSFET Q4), with similarly complementary drive signals, and their phases are offset from Q1 / Q2 (usually by 180° phase shift) to achieve the square wave excitation required for LLC resonance. The PWM signals generated according to the optimal switching frequency drive the MOSFETs of the LLC resonant converter, corresponding to PWM1 to PWM4 drive signals in the diagram.

[0076] This invention can be simulated using MATLAB R2023 software. The experimental parameters of the converter in the simulation are: resonant inductance Lr = 18.2 μH, magnetizing inductance Lm = 127.4 μH, and capacitance C = 217.4 nF. To verify the robustness of this invention under parameter uncertainties, the drift of the resonant network parameters due to temperature rise during actual operation is simulated. The resonant inductance Lr, magnetizing inductance Lm, and resonant capacitance Cr are simultaneously deviated from their nominal values ​​by ±50%, and the output voltage control accuracy of the control method of this invention is tested.

[0077] Simulation results are as follows Figure 5 As shown: with a reference voltage of 20V and a parameter mismatch of -50%, the output voltage error of the traditional MPC method is 10V, while the output voltage error of this invention remains at 0.085V. Simulation results are as follows. Figure 6 As shown: when the reference voltage is 25V, the output voltage error of the traditional MPC method is 12.45V, while the output voltage error of the present invention remains at 0.5V. Figure 7 and Figure 8 The output voltage error of the traditional MPC method and the present invention under parameter mismatch of ±50% was compared. The above results demonstrate the strong robustness of the present invention to parameter variations.

[0078] The PI-model-free predictive control method proposed in this invention effectively solves the technical challenge of balancing dynamic performance and parameter robustness in traditional control strategies, providing a high-performance control solution for LLC resonant converters. This invention constructs a two-layer PI-model-free predictive control architecture. The PI controller outputs a reference switching frequency, responsible for maintaining the global stability of the system; the model-free predictive controller is responsible for improving system robustness and dynamic performance, and is divided into two decoupled sub-modules: an online parameter identification module first identifies model parameters in real time using a reduced-order first-order discrete data model, and then employs a recursive least squares method with a forgetting factor, thereby achieving independent prediction of future output voltage; the frequency rolling optimization module generates a discrete frequency control set centered on the PI output reference frequency, and achieves real-time dynamic correction of the PI reference frequency through online evaluation of the value function and rolling optimization. Through this architecture, this invention significantly improves the dynamic response speed and control accuracy while ensuring high robustness of the system under uncertain conditions such as model parameter mismatch. It aims to achieve fast dynamic response, strong parameter robustness, and high steady-state accuracy without relying on a precise mathematical model.

[0079] To better implement the data-driven control method for the LLC resonant converter in this embodiment of the invention, based on the data-driven control method for the LLC resonant converter, this embodiment of the invention also provides a data-driven control system for the LLC resonant converter, which includes: The sampling module is used to acquire the output voltage, load current, and rectifier bridge current of the LLC resonant converter during the current control cycle. The PI control module is used to calculate the reference frequency based on the error between the reference voltage and the output voltage using a PI controller. A frequency set construction module is used to calculate the step size based on the output voltage error and construct a discrete frequency control set with the reference frequency and the step size; the discrete frequency control set includes at least three candidate frequencies. The online parameter identification module is used to determine the time-varying parameters in the first-order discrete data model based on the recursive least squares method. The first-order discrete data model is used to describe the relationship between the change in the output voltage and the load current and the rectifier bridge current. The voltage prediction module is used to predict the output voltage of the next control cycle based on the time-varying parameters corresponding to the current control cycle, the load current, and the rectifier bridge current. The rolling optimization module is used to calculate the corresponding value function for each candidate frequency in the discrete frequency control set based on the predicted output voltage and reference voltage, and determine the candidate frequency that minimizes the value function as the optimal switching frequency. The PWM generation module is used to drive the switching transistors of the LLC resonant converter based on the PWM signal generated according to the optimal switching frequency.

[0080] The data-driven control system for the LLC resonant converter provided in the above embodiments can realize the technical solutions described in the above embodiments of the data-driven control method for the LLC resonant converter. The specific implementation principles of each module or unit can be found in the corresponding content in the above embodiments of the data-driven control method for the LLC resonant converter, and will not be repeated here.

[0081] The present invention also provides a control chip configured to execute the steps of the data-driven control method for the LLC resonant converter described in any of the above implementations.

[0082] Accordingly, this application also provides an LLC resonant converter, including the control chip described above.

[0083] The LLC resonant converter and its data-driven control method, system, and chip provided by this invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A data-driven control method for an LLC resonant converter, characterized in that, include: The output voltage, load current, and rectifier bridge current of the LLC resonant converter are collected during the current control cycle. The reference frequency is calculated by a PI controller based on the error between the reference voltage and the output voltage. The step size is calculated based on the output voltage error, and a discrete frequency control set is constructed using the reference frequency and the step size; the discrete frequency control set includes at least three candidate frequencies. The time-varying parameters in the first-order discrete data model are determined based on the recursive least squares method. The first-order discrete data model is used to describe the relationship between the change in the output voltage and the load current and the rectifier bridge current. Based on the time-varying parameters corresponding to the current control cycle, the load current, and the rectifier bridge current, predict the output voltage of the next control cycle; For each candidate frequency in the discrete frequency control set, a corresponding value function is calculated based on the predicted output voltage and reference voltage, and the candidate frequency that minimizes the value function is determined as the optimal switching frequency. The PWM signal generated according to the optimal switching frequency drives the switching transistor of the LLC resonant converter.

2. The data-driven control method for an LLC resonant converter according to claim 1, characterized in that, The discrete frequency control set has at least three candidate frequencies that are symmetrically distributed. The at least three candidate frequencies include the reference frequency and candidate frequencies located on both sides of the reference frequency and differing from the reference frequency by a finite number of integer multiples of step size.

3. The data-driven control method for an LLC resonant converter according to claim 2, characterized in that, The calculation of the step size includes: The first coefficient is determined based on the processor clock cycle and the switching cycle; The second coefficient is calculated based on the adjustment coefficient and the output voltage error after limiting; if the absolute value of the difference between the reference voltage and the output voltage is less than the maximum error voltage, the output voltage error after limiting is determined to be the absolute value of the difference; if the absolute value of the difference is not less than the maximum error voltage, the output voltage error after limiting is determined to be the maximum error voltage. The step size is calculated based on the first coefficient and the second coefficient.

4. The data-driven control method for an LLC resonant converter according to claim 1, characterized in that, The first-order discrete data model is represented as follows: ； in, The change in the output voltage. The load current, The rectifier bridge current is... The first time-varying parameter identified in the k-th period, The second time-varying parameter is obtained for the kth period.

5. The data-driven control method for an LLC resonant converter according to claim 4, characterized in that, The determination of time-varying parameters in a first-order discrete data model based on the recursive least squares method includes: The gain matrix of the current control cycle is calculated based on the forgetting factor, the covariance matrix of the previous control cycle, and the input vector of the current control cycle; the input vector is calculated based on the load current and rectifier bridge current collected in the current control cycle. The covariance matrix of the current control cycle is calculated based on the covariance matrix of the previous control cycle, the gain matrix of the current control cycle, and the input vector of the current control cycle. The parameter vector of the current control cycle is calculated based on the parameter vector of the previous control cycle, the gain matrix of the current control cycle, the input vector of the current control cycle, and the parameter vector of the previous control cycle. The first time-varying parameter and the second time-varying parameter are determined based on the parameter vector of the current control cycle.

6. The data-driven control method for an LLC resonant converter according to claim 5, characterized in that, The step of predicting the output voltage of the next control cycle based on the time-varying parameters corresponding to the current control cycle, the load current, and the rectifier bridge current includes: Based on the first time-varying parameter, the second time-varying parameter, the load current, and the rectifier bridge current of the current control cycle, the predicted voltage change for the next control cycle is calculated. The output voltage of the next control cycle is calculated based on the output voltage of the current control cycle and the predicted voltage change of the next control cycle.

7. The data-driven control method for an LLC resonant converter according to any one of claims 1 to 6, characterized in that, The value function is expressed as: ； ； in, The value is a value function used to evaluate the merits of candidate frequencies. For the reference voltage tracking error term, evaluate the deviation between the predicted output voltage and the reference voltage. For the frequency variation smoothing term, evaluate the deviation between the candidate frequency and the optimal frequency of the previous cycle. and They are respectively and Weighting factors This is the reference voltage for the (k+1)th control cycle. The predicted output voltage for the (k+1)th control cycle. The reference frequency output by the PI controller. is the candidate frequency for the k-th period.

8. A data-driven control system for an LLC resonant converter, characterized in that, include: The sampling module is used to acquire the output voltage, load current, and rectifier bridge current of the LLC resonant converter during the current control cycle. The PI control module is used to calculate the reference frequency based on the error between the reference voltage and the output voltage using a PI controller. A frequency set construction module is used to calculate the step size based on the output voltage error and construct a discrete frequency control set with the reference frequency and the step size; the discrete frequency control set includes at least three candidate frequencies. The online parameter identification module is used to determine the time-varying parameters in the first-order discrete data model based on the recursive least squares method. The first-order discrete data model is used to describe the relationship between the change in the output voltage and the load current and the rectifier bridge current. The voltage prediction module is used to predict the output voltage of the next control cycle based on the time-varying parameters corresponding to the current control cycle, the load current, and the rectifier bridge current. The rolling optimization module is used to calculate the corresponding value function for each candidate frequency in the discrete frequency control set based on the predicted output voltage and reference voltage, and determine the candidate frequency that minimizes the value function as the optimal switching frequency. The PWM generation module is used to drive the switching transistors of the LLC resonant converter based on the PWM signal generated according to the optimal switching frequency.

9. A control chip, characterized in that, The control chip is configured to perform the data-driven control method for the LLC resonant converter as described in any one of claims 1 to 7.

10. An LLC resonant converter, characterized in that, Includes the control chip as described in claim 9.

Images