A model-free robust predictive control method for dual active bridge DC-DC converter

Abstract

A model-free robust predictive control method for a dual active bridge DC-DC converter is disclosed, belonging to the field of converter technology. The method is as follows: An autoregressive moving average model is constructed and its order is determined; an extended state observer is designed and its gain is selected and discretized; after online identification and updating of the autoregressive moving average model parameters, a control signal is generated and filtered; the phase shift angle is calculated and constrained. This invention proposes a data-driven deadbeat predictive control strategy, utilizing the autoregressive moving average model to estimate the dual active bridge DC-DC converter, and identifying the model parameters online using the least squares method with a forgetting factor, thereby improving the robustness of the converter output voltage control under arbitrary disturbances.

Description

Technical Field

[0001] This invention relates to a model-free robust predictive control method for a dual active bridge DC-DC converter, belonging to the field of converter technology. Background Technology

[0002] Dual active bridge DC-DC converters are an important technology in the field of power electronics, and precise control of their output voltage is crucial for the stable operation of the system. These converters operate on a predictive principle, where the predictive controller predicts and determines the next switching state based on the current sampled value and a known discrete model, thereby achieving dynamic multi-objective optimization. However, various factors such as model inaccuracies, neglected dead time, and parameter errors often degrade the controller's performance, leading to increased steady-state error of the reference voltage and significantly impaired dynamic performance.

[0003] Existing technologies typically employ model-based compensation methods. Among these, the Disturbance Observer (DOB) approach involves designing a special observer to estimate the deviation between the nominal model and the actual controlled object, and then using this estimate for system compensation to reduce errors caused by parameter mismatch. For example, using an Extended State Observer (ESO) for lumped disturbance estimation is an effective DOB method.

[0004] On the other hand, parameter identification-based methods eliminate the impact of parameter mismatch on system performance by accurately calculating parameters online and updating the nominal model in the controller in real time. Methods such as Recursive Least Squares (RLS) can identify key parameters such as transformer leakage inductance online, significantly improving control accuracy.

[0005] However, although the above methods alleviate the problems caused by parameter mismatch to some extent, their disturbance rejection performance is still affected by the parametric model used. Model-based compensation techniques cannot completely eliminate all disturbances, especially under complex and variable real-world operating conditions, where their effectiveness is often limited.

[0006] Therefore, how to further improve the output voltage control accuracy of dual active bridge DC-DC converters, especially in the presence of parameter errors and model inaccuracies, has become a pressing technical challenge. Summary of the Invention

[0007] To address the problems existing in the background technology, the present invention provides a model-free robust predictive control method for a dual active bridge DC-DC converter.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: a model-free robust predictive control method for a dual active bridge DC-DC converter, the method comprising the following steps:

[0009] S1: Construct an autoregressive moving average model and design a data-driven model for a dual active bridge DC-DC converter using the autoregressive moving average model;

[0010] The autoregressive moving average model described in S1 is as follows:

[0011]

[0012] In formula (1):

[0013] p represents the order of the autoregressive part of the model;

[0014] q represents the order of the moving average part of the model;

[0015] θ x and φ y (x∈[0,1,...,q],y∈[1,...,p]) all represent the coefficients of the autoregressive moving average model;

[0016] ε represents the output voltage following error;

[0017] The output voltage following error at time t = k+1 is represented by: The observed value of the output voltage at time t = k+1 The reference value representing the output voltage at time t = k+1;

[0018] This represents the control signal calculated at time t = k-1 and applied to the converter at time t = k.

[0019] S2: Determine the order of the autoregressive moving average model;

[0020] S201: Determine the order q of the moving average part of the model;

[0021] S202: Determine the order p of the autoregressive component of the model:

[0022] When the partial autocorrelation coefficient between the output voltage following errors ε(k) and ε(kp) at times k and kp in the time series is less than the threshold value, it is considered that the signals corresponding to times k and kp are not correlated. Then the time difference between times k and kp is the value of the order p of the autoregressive part of the model.

[0023] S3: Design an extended state observer to suppress noise during the sampling process, reduce the order of the data-driven model in a one-step delay compensation design, and estimate the accurate predicted value of the state variable at the next sampling period t = k + 1, and then use the predicted value for coefficient identification of the data-driven model.

[0024] The extended state observer described in S3 is as follows:

[0025]

[0026] In formula (2):

[0027] Represents the first derivative of z1;

[0028] This represents the first derivative of z2;

[0029] Represents the output voltage U o Observed values;

[0030] z2 represents the observed value of the system's lumped disturbance;

[0031] Represents the input gain, where: n represents the turns ratio of the primary and secondary sides of the high-frequency transformer, U in f represents the input voltage. s L represents the control frequency. k C represents the leakage inductance of a high-frequency transformer. o Represents the output capacitor;

[0032] Both represent the gain of the extended state observer, where: ω n This represents an adjustable parameter used to determine the bandwidth of the extended state observer.

[0033] S4: Gain selection and discretization of the extended state observer;

[0034] S401: Uses control cycle T s Discretizing equation (2), we get:

[0035]

[0036] S402: Calculate the observed output voltage and the observed system lumped disturbance at time t = k+1. S5: Identify and update the parameters of the autoregressive moving average model online;

[0037] The algorithm for online identification described in S5 is as follows:

[0038]

[0039] In equation (4):

[0040] Represents a parameter vector;

[0041] The regression vector represents the time step t = k;

[0042] P represents the predicted output voltage following error at time t = k+1; P and K both represent intermediate variables.

[0043] λ represents the forgetting factor, 0≤λ≤1.

[0044] I represents the identity matrix.

[0045] S6: Generate control signals based on online identification results;

[0046] The formula for generating the control signal described in S6 is as follows:

[0047]

[0048] S7: The control signal is filtered before it is input to the converter;

[0049] The filtering formula described in S7 is as follows:

[0050]

[0051] In equation (6):

[0052] This represents the filtered control signal;

[0053] k LPF This represents the adjustable filter coefficient.

[0054] S8: Calculate and constrain the phase shift angle.

[0055] Compared with the prior art, the beneficial effects of the present invention are:

[0056] This invention proposes a data-driven deadbeat predictive control strategy. It utilizes an autoregressive moving average model to estimate the output voltage of a dual active bridge DC-DC converter and identifies the model parameters online using a least squares method with a forgetting factor, thereby improving the robustness of the converter's output voltage control under arbitrary disturbances. This approach eliminates the dependence on converter parameters in traditional methods. Based on the historical output voltage error and control output of the DAB converter, it performs online learning and adaptive adjustment. The converged model achieves robust zero steady-state error voltage control without relying on any DAB converter parameters. Even under varying operating conditions and disturbances, it maintains stable output voltage control, demonstrating broad application prospects and significant practical value. Attached Figure Description

[0057] Figure 1 This is a control block diagram of the present invention;

[0058] Figure 2This is a schematic diagram of the parameter changes of the autoregressive moving average model when the reference voltage changes stepwise from 80V to 60V.

[0059] Figure 3 The figure shows a comparison of experimental waveforms when the load resistance is reduced from 25Ω to 10Ω using a traditional PI control strategy and the present invention without parameter errors. The four waveforms from top to bottom in the figure represent the DAB converter input voltage, DAB converter port output voltage, DAB converter port output current, and high-frequency transformer internal current, respectively. The load resistance switching occurs at the point where the current waveform increases stepwise in the figure.

[0060] Figure 4 This is a schematic diagram of the experimental waveforms showing how the load resistance was reduced from 25Ω to 10Ω using the present invention, even with inductance error. Detailed Implementation

[0061] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the invention, not all embodiments. 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.

[0062] This method is applied to a dual active bridge DC-DC converter topology and uses a dual phase-shifting strategy for control, that is, changing the phase shift angle between different arms within the two H-bridges and between two different H-bridges.

[0063] The present invention discloses a model-free robust predictive control method for a dual active bridge DC-DC converter (DAB), the method comprising the following steps:

[0064] S1: Construct an autoregressive moving average (ARMA) model of appropriate order, and design a data-driven model for a dual active bridge DC-DC converter using the autoregressive moving average model;

[0065] The autoregressive moving average model described in S1 is as follows:

[0066]

[0067] In formula (1):

[0068] p represents the order of the autoregressive (AR) component of the model;

[0069] q represents the order of the moving average (MA) component of the model;

[0070] θ x and φ y(x∈[0,1,...,q],y∈[1,...,p]) all represent the coefficients of the autoregressive moving average model;

[0071] ε represents the output voltage following error;

[0072] The output voltage following error at time t = k+1 is represented by: The observed value of the output voltage at time t = k+1 The reference value representing the output voltage at time t = k+1;

[0073] This represents the control signal calculated at time t = k-1 and applied to the converter at time t = k.

[0074] S2: Determine the order of the autoregressive moving average model;

[0075] S201: Determine the order q of the moving average part of the model based on the mathematical model of the DAB converter considering time delay compensation;

[0076] Take the autoregressive moving average model constructed with an autoregressive model order p=5 and a moving average model order q=2 as an example.

[0077] Since model-free predictive control follows the deadbeat principle, the current output voltage U o (k) is determined by the control voltage calculated from the two control cycles before the current time step, therefore q = 2.

[0078] S202: Determine the order p of the autoregressive component of the model:

[0079] Since the partial autocorrelation coefficient represents the correlation between the output voltage following errors ε(k) and ε(kp) at times k and kp, when the partial autocorrelation coefficient between the output voltage following errors ε(k) and ε(kp) at times k and kp in the time series is less than a threshold value (such as 0.2), it is considered that the signals corresponding to times k and kp are not correlated. Then, the time difference between times k and kp is the value of the order p of the autoregressive part of the model.

[0080] S3: Design an extended state observer (ESO) to suppress noise during the sampling process, reduce the order of the data-driven model in a one-step delay compensation design, and estimate the accurate prediction of the state variables at the next sampling period t = k + 1, and then use the prediction to identify the coefficients of the data-driven model.

[0081] The extended state observer described in S3 is as follows:

[0082]

[0083] In formula (2):

[0084] Represents the first derivative of z1;

[0085] This represents the first derivative of z2;

[0086] Represents the output voltage U o Observed values;

[0087] z2 represents the observed value of the system's lumped disturbance;

[0088] Represents the input gain, where: n represents the turns ratio of the primary and secondary sides of the high-frequency transformer, U in f represents the input voltage. s L represents the control frequency. k C represents the leakage inductance of a high-frequency transformer. o Represents the output capacitor;

[0089] Both represent the gain of the extended state observer, where: ω n This represents an adjustable parameter used to determine the bandwidth of the extended state observer.

[0090] S4: Gain selection and discretization of the extended state observer;

[0091] S401: Uses control cycle T s Discretize equation (2). In each control cycle, first determine the output voltage U at time t = k. o (k) Calculate the output voltage observation z1(k+1) at the next time step using the extended state observer. We can then obtain:

[0092]

[0093] S402: Calculate the observed output voltage and the observed system lumped disturbance at time t = k+1.

[0094] S5: The parameters of the autoregressive moving average model are identified and updated online using the least squares method with a forgetting factor;

[0095] The algorithm for online identification described in S5 is as follows:

[0096]

[0097] In equation (4):

[0098] Represents a parameter vector;

[0099] The regression vector represents the time step t = k;

[0100] This represents the predicted output voltage following error at time t=k+1;

[0101] P and K both represent intermediate variables;

[0102] λ represents the forgetting factor, 0≤λ≤1.

[0103] I represents the identity matrix.

[0104] S6: Generate control signals based on online identification results;

[0105] The formula for generating the control signal described in S6 is as follows:

[0106]

[0107] S7: The control signal is filtered before it is input to the converter to make the control smoother, suppress the output voltage jitter, and improve the voltage control performance and robustness.

[0108] Based on the observed output voltage at time t = k+1 Given the predicted output voltage z1(k+1), calculate the corresponding control signal considering one-step time delay compensation. The control signal is applied to the converter after passing through the minimum current stress dual phase-shift modulation module, achieving precise and robust voltage control.

[0109] The filtering formula described in S7 is as follows:

[0110]

[0111] In equation (6):

[0112] This represents the filtered control signal;

[0113] k LPF This represents the adjustable filter coefficient.

[0114] S8: Calculate and constrain the phase shift angle.

[0115] To control the DAB converter, the control signal should be modulated into an inner phase shift angle D1 and an outer phase shift angle D2, which can be calculated based on existing minimum current stress dual-phase shift modulation strategies. Furthermore, to prevent damage to the DAB converter, the following constraints should be applied to the phase shift angle output to limit the output range:

[0116]

[0117] To ensure the model's fitting accuracy, a model order selection method based on the inherent characteristics of the DAB converter system and time series analysis was adopted. This completely eliminates the dependence on model parameters while maintaining the inherent fast dynamic characteristics of predictive control.

[0118] like Figure 2 As shown, when the operating conditions of the dual active bridge DC-DC converter change, the parameters of the autoregressive moving average model will adaptively follow the new parameter values ​​and have a relatively fast convergence speed.

[0119] like Figure 3 As shown, although both the traditional PI control strategy and the method of this invention can achieve zero steady-state error output voltage following, the transient recovery time of this invention is 4.1 ms, exhibiting the characteristics of fast dynamics in predictive control. In contrast, the transient recovery time of the traditional PI control strategy is 20.5 ms, demonstrating the superiority of the method in this invention.

[0120] like Figure 4 As shown, the present invention can completely eliminate errors, has good steady-state and dynamic tracking performance, a transient recovery time of 4.2ms, and is completely unaffected by changes or inaccuracies in converter parameters, thus significantly improving the robustness of the system.

[0121] This invention only requires the dual active bridge DC-DC converter to have a voltage sensor, instead of a current sensor as required in existing methods. Furthermore, it can improve the robustness of the predictive controller output voltage control without additional hardware design, and reduce hardware costs.

[0122] The algorithm can be programmed and implemented in a digital control chip (DSP), and can be applied to general-purpose dual active bridge DC-DC converter controllers, which has extremely high application and economic value.

[0123] The algorithm of this invention only requires the current voltage of the converter, and the controller does not need to measure any converter parameters in advance.

[0124] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0125] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A model-free robust predictive control method for a dual active bridge DC-DC converter, characterized in that: The method includes the following steps: S1: Construct an autoregressive moving average model and design a data-driven model for a dual active bridge DC-DC converter using the autoregressive moving average model; The autoregressive moving average model described in S1 is as follows: （1） In formula (1): p represents the order of the autoregressive part of the model; q represents the order of the moving average part of the model; and All of these represent the coefficients of the autoregressive moving average model; This represents the output voltage following error; The output voltage following error at time t=k+1 is represented by: The observed value of the output voltage at time t=k+1, The reference value representing the output voltage at time t=k+1; This represents the control signal calculated at time t=k-1 and applied to the converter at time t=k; S2: Determine the order of the autoregressive moving average model; S3: Design an extended state observer to suppress noise during the sampling process, reduce the order of the data-driven model in a one-step delay compensation design, and estimate the accurate predicted value of the state variable at time t=k+1 of the next sampling period, and then use the predicted value for coefficient identification of the data-driven model. S4: Gain selection and discretization of the extended state observer; S5: Online identification and updating of autoregressive moving average model parameters; S6: Generate control signals based on online identification results; S7: The control signal is filtered before it is input to the converter; S8: Calculate and constrain the phase shift angle.

2. The model-free robust predictive control method for a dual active bridge DC-DC converter according to claim 1, characterized in that: S2 includes the following steps: S201: Determine the order q of the moving average part of the model; S202: Determine the order p of the autoregressive component of the model: When the output voltage tracking error is between time k and time kp in the time series and When the partial autocorrelation coefficient between the two signals is less than the threshold value, it is considered that the signals corresponding to time k and time kp are not correlated. Then, the time difference between time k and time kp is the value of the order p of the autoregressive part of the model.

3. The model-free robust predictive control method for a dual active bridge DC-DC converter according to claim 2, characterized in that: The extended state observer described in S3 is as follows: （2） In formula (2): represent The first derivative; represent The first derivative; Represents output voltage Observed values; Observations representing the lumped disturbances of the system; Represents the input gain, where: n represents the turns ratio of the primary and secondary sides of the high-frequency transformer, U in f represents the input voltage. s L represents the control frequency. k C represents the leakage inductance of a high-frequency transformer. o Represents the output capacitor; Both represent the gain of the extended state observer, where: This represents an adjustable parameter used to determine the bandwidth of the extended state observer.

4. The model-free robust predictive control method for a dual active bridge DC-DC converter according to claim 3, characterized in that: S4 includes the following steps: S401: Uses control cycle T s Discretizing equation (2), we get: （3） S402: Calculate the observed output voltage and the observed system lumped disturbance at time t=k+1.

5. The model-free robust predictive control method for a dual active bridge DC-DC converter according to claim 4, characterized in that: The algorithm for online identification described in S5 is as follows: （4） In equation (4): Represents a parameter vector; The regression vector represents the time step t=k; This represents the predicted output voltage following error at time t=k+1; P and K both represent intermediate variables; To represent the forgetting factor, 0 ≤ ≤1; Represents the identity matrix.

6. The model-free robust predictive control method for a dual active bridge DC-DC converter according to claim 5, characterized in that: The formula for generating the control signal described in S6 is as follows: （5）。 7. The model-free robust predictive control method for a dual active bridge DC-DC converter according to claim 6, characterized in that: The filtering formula described in S7 is as follows: （6） In formula (6): This represents the filtered control signal; This represents the adjustable filter coefficient.

Images