A dc-dc converter adaptive twin control method based on improved bayesian optimization

Abstract

The application provides a DC-DC converter adaptive twin control method based on improved Bayesian optimization, comprising the following steps: based on the operation mechanism of the DC-DC converter, a static digital twin mechanism model is constructed under the condition that the parameters of the converter are constant; the time-varying parameters in the physical entity of the DC-DC converter are identified by using an extended Kalman filter, and the identification result is transmitted to the static digital twin mechanism model for updating, so that a dynamic digital twin model is constructed; an improved Bayesian optimization-based controller parameter optimization algorithm is designed to adaptively modify and optimize the controller parameters; and when the control performance of the DC-DC converter is degraded, the controller is fine-tuned based on the dynamic digital twin model and the controller parameter optimization algorithm. Through the digital twin and the controller parameter optimization method, the application effectively solves the problems of difficult accurate capture of time-varying parameters and degradation of control performance in the operation of the converter.

Description

An Adaptive Twin Control Method for DC-DC Converters Based on Improved Bayesian Optimization Technical Field

[0001] This invention relates to the fields of digital twins and power electronics, and specifically to an adaptive twin control method for DC-DC converters based on improved Bayesian optimization. Background Technology

[0002] This section provides only background information relevant to this disclosure and is not necessarily prior art.

[0003] In fields such as new energy vehicles, energy storage systems, and aerospace power supply, increasingly stringent requirements are being placed on the operational stability, dynamic response speed, and energy conversion efficiency of power supply systems. As a core component for power conversion and distribution in these fields, the performance of DC-DC converters directly impacts the reliability of the overall power supply system. However, during operation, DC-DC converters are susceptible to factors such as component aging, load fluctuations, and changes in ambient temperature, leading to time-varying issues in key parameters such as inductor equivalent resistance and capacitor capacitance, resulting in low modeling accuracy. Simultaneously, conventional controllers often employ fixed parameter designs, failing to adapt to time-varying operating conditions, causing a gradual degradation in converter control performance and making it difficult to meet the high-precision power supply requirements under complex operating conditions.

[0004] Current solutions to the degradation of control performance in DC-DC converters through digital modeling include constructing digital models using mechanistic modeling, but this method has limited modeling accuracy. Controller parameter adjustment mainly relies on manual tuning or simple PID self-tuning, lacking dynamic adaptation capabilities for time-varying parameters. While these methods are feasible under simple operating conditions, they suffer from low modeling accuracy and lagging controller parameter optimization, failing to achieve stable and efficient operation throughout the converter's entire lifecycle. Therefore, achieving accurate real-time identification of time-varying parameters in DC-DC converters and adaptive dynamic optimization of controller parameters to realize efficient and stable twin control, thereby overcoming the challenge of control performance degradation under complex operating conditions, has become a critical issue urgently needing to be addressed in the current development of DC-DC converter technology.

[0005] Therefore, it is essential to introduce digital twin technology and design an efficient and feasible adaptive twin control method for DC-DC converters.

[0006] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] Purpose of the invention: The technical problem to be solved by the present invention is to provide an adaptive twin control method for DC-DC converters based on improved Bayesian optimization, which addresses the shortcomings of the existing technology.

[0008] To address the aforementioned technical problems, this invention discloses an adaptive twin control method for DC-DC converters based on improved Bayesian optimization, comprising the following steps:

[0009] Step 1: Construct a static digital twin mechanism model while keeping the converter parameters constant;

[0010] Step 2: Use extended Kalman filtering to identify time-varying parameters in the physical entity of the DC-DC converter, and pass the identification results to the static digital twin mechanism model for updating, thereby realizing the construction of the dynamic digital twin model;

[0011] Step 3: Design a DC-DC controller parameter optimization method based on improved Bayesian optimization, and use a dynamic digital twin model to adaptively modify and optimize the controller parameters under varying operating conditions.

[0012] Step 4: When the control performance of the DC-DC converter degrades, the controller parameters are fine-tuned based on the dynamic digital twin model and the controller parameter optimization algorithm.

[0013] Furthermore, the static digital twin mechanism model described in step 1 includes modeling the state equation of the DC-DC buck circuit and modeling the controller. The controller adopts PID control, and the state equation of the DC-DC buck circuit is expressed as:

[0014]

[0015] in, It is inductor current. It is the capacitor voltage. DC input voltage For inductor inductance, For capacitor capacitance, For load resistance, Duty cycle, It is the differential of the capacitor voltage. It is the differential of the inductor current.

[0016] Furthermore, in step 2, the time-varying parameters are identified by extending the parameters to be identified as state variables and constructing an augmented state vector based on the state equation in step 1. The time-varying parameters include capacitor capacitance, inductor inductance, and load resistance.

[0017] Among them, Euler discretization is used for the augmented state vector to construct the Jacobi matrix to transmit uncertainty.

[0018] Furthermore, the controller parameter optimization algorithm described in step 3 takes into account tracking accuracy, dynamic response, and control smoothness.

[0019] In step 3, the controller parameter optimization algorithm based on improved Bayesian optimization uses Latin hypercube sampling to select three sets of initial parameters from the feasible region of controller parameters and obtain an initial cost function data training set, which is used to construct a Gaussian process surrogate model, fit the cost function, and estimate the uncertainty.

[0020] Furthermore, in step 4, the controller parameters are fine-tuned based on the dynamic digital twin model and the controller parameter optimization algorithm. The dynamic digital twin model is used to monitor the current state and predict the response over a period of time in the future, including the prediction of capacitor voltage (i.e., output voltage) and controller output.

[0021] In step 4, the controller parameters are fine-tuned based on a dynamic digital twin model and a controller parameter optimization algorithm. An improved Bayesian optimization algorithm is used to optimize parameters based on the prediction results of the digital twin model. When the cost function is minimized, the optimal PID parameters are obtained, thus maximizing the control performance of the DC-DC converter digital model in the digital space. The obtained optimal PID parameters are then transmitted to the controller entity in the physical space using the TCP / IP protocol.

[0022] Beneficial effects:

[0023] 1. By integrating sensor signals of key operating parameters such as the output voltage and inductor current of the DC-DC converter, the values ​​of key parameters are accurately identified, thereby ensuring the accuracy of the digital twin model construction.

[0024] 2. Based on the digital twin model and the improved Bayesian optimization method for controller parameters, it can respond quickly to the situation of degraded control performance, optimize controller parameters, ensure efficient system operation, and effectively solve the problem of dynamic response lag caused by performance degradation or complex operating conditions.

[0025] 3. This invention is applicable to DC-DC converters in different scenarios such as power supply for new energy vehicles, energy storage converters, and aerospace auxiliary power supply, reducing the adaptation cost for multi-scenario applications. Attached Figure Description

[0026] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, and the advantages of the present invention in the above and / or other aspects will become clearer.

[0027] Figure 1 is a flowchart of the adaptive twin control method for DC-DC converters based on improved Bayesian optimization provided by the present invention.

[0028] Figure 2 is a schematic diagram of the DC-DC converter step-down circuit provided by the present invention.

[0029] Figure 3 is the response diagram of the dynamic digital twin model of the DC-DC converter provided by the present invention.

[0030] Figure 4 is a diagram showing the controller parameter optimization results provided by the present invention.

[0031] Figure 5 shows the results before adaptive optimization of the controller parameters provided by this invention.

[0032] Figure 6 is a graph showing the results of adaptive optimization of controller parameters provided by the present invention. Detailed Implementation

[0033] This invention provides an adaptive twin control method for DC-DC converters based on improved Bayesian optimization. It utilizes digital twin technology to achieve high-precision digital model construction and controller parameter optimization, improving the accuracy of digital modeling and maintaining continuous optimization of control performance, thus realizing twin control of the converter. The specific technical solution is as follows:

[0034] An adaptive twin control method for DC-DC converters based on improved Bayesian optimization is proposed. This method can be used for accurate digital modeling and control performance optimization of DC-DC converters. Referring to Figure 1, which is a flowchart of the adaptive twin control method for DC-DC converters based on improved Bayesian optimization, the method includes the following steps:

[0035] Step 1: Based on the operating mechanism of the DC-DC converter, a static digital twin mechanism model is constructed while ensuring constant converter parameters. A typical DC-DC buck circuit is shown in Figure 2. Based on Kirchhoff's laws, when the switch is turned on, the inductor current... and capacitor voltage It can be represented as:

[0036]

[0037] in, DC input voltage For inductor inductance, For capacitor capacitance, For load resistance, It is the differential of the capacitor voltage. It is the differential of the inductor current.

[0038] When the switch is closed, the inductor current... and capacitor voltage It can be represented as:

[0039]

[0040] Let the duty cycle of the switch be Using the state-space averaging method, the state equations of the DC-DC buck circuit can be expressed as follows: (Weighted average of two linear state equations over one switching cycle)

[0041]

[0042] The controller uses PID control, which can be expressed as:

[0043]

[0044] in, for Time controller output, This refers to the deviation between the target value and the actual value of the output voltage. , and These are the proportional coefficient, integral coefficient, and differential coefficient, respectively.

[0045] The above process is used to construct a static digital twin model that includes a controller and a DC-DC converter.

[0046] Step 2 involves using an extended Kalman filter to identify time-varying parameters in the physical entity of the DC-DC converter, and then passing the identification results to the static digital twin mechanism model for updating, thus constructing a dynamic digital twin model. Due to the actual operation of the capacitor... ,inductance and resistance It may change over time, and at the same time Fluctuations may also occur; therefore, parameter identification is necessary to ensure that the digital twin model of the DC-DC converter maintains consistency with the output of the physical system. Extended Kalman filtering employs local linearity to address nonlinear problems.

[0047] The parameters to be identified are extended as state variables, and an augmented state vector is constructed based on the state equation from step 1. , can be represented as:

[0048]

[0049] Applying Euler discretization to the augmented state vector, assuming the parameters change slowly, it can be expressed as:

[0050]

[0051] Therefore, the discrete-time state estimate can be obtained as follows:

[0052]

[0053] in, Based on Moment Time-state estimation, for Time-state estimation, Sampling time.

[0054] By constructing the Jacobi matrix Transmitting uncertainty:

[0055]

[0056] Covariance propagation is:

[0057]

[0058] in, It is the predicted value of the prior covariance. It is the posterior covariance. It is the process noise covariance matrix.

[0059] Time observation value The inductor current and capacitor voltage can be expressed as:

[0060]

[0061] Calculate Kalman gain :

[0062]

[0063] in, For the measurement matrix, set as follows: , For measuring noise.

[0064] A state update can be represented as:

[0065]

[0066] in, for Time-state estimation.

[0067] The covariance update process can be represented as:

[0068]

[0069] in, For posterior covariance, It is an identity matrix.

[0070] The above process will identify the key parameter, capacitance. ,inductance ,resistance and input voltage The output of the static digital twin mechanism model is passed to the physical transformer to form a dynamic digital twin model.

[0071] Step 3: Design a DC-DC controller parameter optimization method based on improved Bayesian optimization, and use a dynamic digital twin model to adaptively modify and optimize the controller parameters under varying operating conditions.

[0072] Considering tracking accuracy, dynamic response, control smoothness, etc., multi-objective cost function Designed as follows:

[0073]

[0074] in, For the PID parameters to be optimized, To predict the number of steps, covering the entire dynamic response process of the system, The time-weighted coefficients, which enhance the weight of recent errors, are set to... , Sampling time; This is the error penalty coefficient, used to ensure tracking accuracy. For the first The deviation between the target value and the actual value of the step output voltage; The penalty coefficient is used to control the quantity and suppress excessive duty cycle, thereby reducing switching losses. For the first Step controller output; The penalty coefficient for the rate of change of the control quantity. The rate of change of the controller output is defined as follows: ; The overshoot penalty coefficient, For the first Step overshoot, defined as , This is the reference voltage.

[0075] Latin hypercube sampling method is used to extract parameters from the feasible region of the controller. Three sets of initial parameters were selected to obtain the initial cost function training set, which was used to construct a Gaussian process surrogate model, fit the cost function, and estimate the uncertainty. It can be represented as

[0076] (1.15)

[0077] in, Follow the mean The variance is It follows a normal distribution.

[0078] The design aims to improve the data acquisition function as follows:

[0079]

[0080] in, The standard normal cumulative distribution function is... It is the standard normal probability density function. As a standardized variable, reflecting the gap between the predictive performance and the current best performance, it can be expressed as:

[0081]

[0082] in, This represents the optimal cost function value in the training set.

[0083] Based on the desired improvement of the acquisition function, the next experimental parameters can be obtained. Select as

[0084]

[0085] Experiment with the selected parameters in the dynamic digital twin model and obtain new cost values. The Gaussian process surrogate model is updated, and the optimal parameters are output after the maximum number of trials is reached. The optimization process is now complete.

[0086] Step 4: When the control performance of the DC-DC converter degrades, the controller parameters are fine-tuned based on the dynamic digital twin model and the controller parameter optimization algorithm. When the performance of the DC-DC converter entity in the physical space degrades or the operating conditions change, the controller parameters can no longer achieve fast, accurate, and stable precise control. The physical entity controller parameters are adjusted based on the dynamic digital twin model constructed in Step 2 and the optimization algorithm designed in Step 3.

[0087] Using a dynamic digital twin model to monitor the current state and predict the response over a future period, including capacitor voltage (i.e., output voltage) prediction. and controller output Based on the known reference voltage, the cost function can be calculated. , , and The predicted values ​​are then used to calculate the cost function.

[0088] An improved Bayesian optimization algorithm is used to optimize parameters based on the prediction results of the digital twin model. The optimal PID parameters are obtained when the cost function is minimized, thus maximizing the control performance of the DC-DC converter digital model in the digital space. Based on this, the optimal PID parameters are transmitted to the controller entity in the physical space using the TCP / IP protocol. The specific process is as follows:

[0089] The PID parameters output by the optimization algorithm are encapsulated according to a preset data format (such as JSON or binary frame structure) to ensure the integrity of the parameter information. A connection-oriented communication link between the host computer and the physical controller is established based on the TCP protocol. The TCP protocol's three-way handshake mechanism ensures the reliability of the connection, and the IP protocol is used to complete the routing and forwarding of data frames: the host computer acts as a client, sending the encapsulated parameter frame to the physical controller's IP address and preset port through the local network interface; the physical controller acts as a server, receiving the data frame on the listening port, extracting the PID parameters through the protocol parsing module, and verifying the data integrity. If the verification passes, the parameter register of the local control algorithm is updated to achieve real-time updates of the control strategy.

[0090] Figure 3 shows the response diagram of the dynamic digital twin model constructed in step 2, including inductor current, capacitor voltage and duty cycle. It can be clearly seen that the actual value is consistent with the tracked value (output of the digital twin model), indicating that the dynamic digital twin model can accurately characterize the state of the physical converter.

[0091] Figure 4 shows the controller parameter optimization algorithm designed in step 3, which optimizes the controller parameters KP, KI, and KD and makes corresponding adjustments. Figure 5 shows the simulation results before the controller adaptive optimization. When the simulation time is 2.5 seconds, the input voltage changes, the system response is slow, and the control performance deteriorates. Figure 6 shows the simulation results after the controller adaptive optimization. When the simulation time is 2.5 seconds, the optimization algorithm in step 3 is triggered, and the optimal controller parameters obtained in step 3 are passed to the physical object. It can be clearly seen that the system responds quickly and the control performance is optimized.

[0092] In summary, this invention provides an adaptive twin control method for DC-DC converters based on improved Bayesian optimization. This method can construct a high-precision dynamic digital twin model, and perform adaptive optimization of the controller based on this model, ensuring that the system always maintains good control performance. Furthermore, the method provided by this invention has strong generalization capabilities and is applicable to various types of adaptive controller optimization problems.

[0093] In its specific implementation, this application provides a computer storage medium and a corresponding data processing unit. The computer storage medium is capable of storing a computer program, which, when executed by the data processing unit, can run the invention's content regarding an improved Bayesian optimization-based adaptive twin control method for DC-DC converters, as well as some or all of the steps in various embodiments. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0094] Those skilled in the art will clearly understand that the technical solutions in the embodiments of the present invention can be implemented using computer programs and their corresponding general-purpose hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of computer programs, i.e., software products. These computer program software products can be stored in a storage medium and include several instructions to cause a device containing a data processing unit (which may be a personal computer, server, microcontroller, MUU, or network device, etc.) to execute the methods described in various embodiments or certain parts of the embodiments of the present invention.

[0095] This invention provides an idea and method for adaptive twin control of DC-DC converters based on improved Bayesian optimization. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment of the invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.

Claims

1. An adaptive twin control method for DC-DC converters based on improved Bayesian optimization, characterized in that, Includes the following steps: Step 1: Construct a static digital twin mechanism model while ensuring constant converter parameters. Step 2: Identify time-varying parameters in the physical entity of the DC-DC converter using extended Kalman filtering, and pass the identification results to the static digital twin mechanism model for updating, thus constructing a dynamic digital twin model. Step 3: Design a DC-DC controller parameter optimization method based on improved Bayesian optimization, and use the dynamic digital twin model to adaptively modify and optimize controller parameters under varying operating conditions; considering tracking accuracy, dynamic response, control smoothness, etc., a multi-objective cost function is employed. Designed as follows: ,in, For the PID parameters to be optimized, To predict the number of steps, covering the entire dynamic response process of the system, The time-weighted coefficients, which enhance the weight of recent errors, are set to... ， Sampling time; This is the error penalty coefficient, used to ensure tracking accuracy. For the first The deviation between the target value and the actual value of the step output voltage; The penalty coefficient is used to control the quantity and suppress excessive duty cycle, thereby reducing switching losses. For the first Step controller output; The penalty coefficient for the rate of change of the control quantity. The rate of change of the controller output is defined as follows: ； The overshoot penalty coefficient, For the first Step overshoot, defined as ， The reference voltage is used; the Latin hypercube sampling method is employed to obtain the controller parameters from the feasible region. Three sets of initial parameters were selected to obtain the initial cost function training set, which was used to construct a Gaussian process surrogate model, fit the cost function, and estimate the uncertainty. Represented as in, Follow the mean The variance is The normal distribution is obtained; the desired improvement of the acquisition function is as follows: in, The standard normal cumulative distribution function is... It is the standard normal probability density function. For standardized variables, reflecting the gap between predictive performance and current best performance, it is expressed as: in, The optimal cost function value is obtained from the training set; the acquisition function is improved according to the desired result to obtain the next experimental parameters. Select as Experiment with the selected parameters in the dynamic digital twin model and obtain new cost values. The Gaussian process surrogate model is updated, and the optimal parameters are output after the maximum number of trials is reached. The optimization process ends; Step 4: When the control performance of the DC-DC converter degrades, the controller parameters are fine-tuned based on the dynamic digital twin model and the controller parameter optimization algorithm.

2. The adaptive twin control method for a DC-DC converter based on improved Bayesian optimization according to claim 1, characterized in that, The static digital twin mechanism model described in step 1 includes state equation modeling of the DC-DC buck circuit and controller modeling. The controller adopts PID control, and the state equation of the DC-DC buck circuit is expressed as: ,in, It is inductor current. It is the capacitor voltage. DC input voltage For inductor inductance, For capacitor capacitance, For load resistance, It is the differential of the capacitor voltage. It is the differential of the inductor current.

3. The adaptive twin control method for a DC-DC converter based on improved Bayesian optimization according to claim 2, characterized in that, Step 2 involves identifying the time-varying parameters, extending the parameters to be identified as state variables, and constructing an augmented state vector based on the state equation from step 1. The time-varying parameters include capacitor capacitance, inductor inductance, and load resistance.

4. The adaptive twin control method for a DC-DC converter based on improved Bayesian optimization according to claim 3, characterized in that, Step 2 involves identifying the time-varying parameters and using Euler discretization for the augmented state vector.

5. The adaptive twin control method for a DC-DC converter based on improved Bayesian optimization according to claim 4, characterized in that, Step 2 describes using extended Kalman filtering to identify time-varying parameters in the physical entity of the DC-DC converter and constructing a Jacobian matrix to transmit uncertainty.

6. The adaptive twin control method for a DC-DC converter based on improved Bayesian optimization according to claim 2, characterized in that, Step 4 describes the fine-tuning of controller parameters based on a dynamic digital twin model and a controller parameter optimization algorithm. The dynamic digital twin model is used to monitor the current state and predict the response over a future period, including the prediction of capacitor voltage (output voltage) and controller output.

7. The adaptive twin control method for a DC-DC converter based on improved Bayesian optimization according to claim 6, characterized in that, Step 4 describes the fine-tuning of controller parameters based on a dynamic digital twin model and a controller parameter optimization algorithm. An improved Bayesian optimization algorithm is used to optimize parameters based on the prediction results of the digital twin model. When the cost function is minimized, the optimal PID parameters are obtained, thus maximizing the control performance of the DC-DC converter digital model in the digital space.

8. The adaptive twin control method for a DC-DC converter based on improved Bayesian optimization according to claim 7, characterized in that, The optimal PID parameters are transmitted to the controller entity in the physical space using the TCP / IP protocol.

Images