Topology control method and equipment of switching power supply and storage medium

Abstract

The invention provides a topology control method and device for a switching power supply, and a storage medium. The method comprises the following steps: collecting operation data of the switching power supply; performing Kalman filtering on the operation data to obtain first state data and second state data; selecting a corresponding dynamic model from a preset model library according to the first state data, identifying a model parameter of the dynamic model, and calculating a confidence coefficient of the dynamic model; according to the model parameters and the second state data, establishing a multi-objective optimization function in a preset prediction time domain, and solving the multi-objective optimization function to obtain a target control sequence; adjusting the control duty ratio of the switching power supply according to the target control sequence, detecting the zero-crossing time of the corresponding inductive current, and calculating the target conduction phase of the synchronous rectifier tube according to the zero-crossing time; and generating a first driving signal of the main switch tube and a second driving signal of the synchronous rectifier tube according to the target conduction phase.

Description

Technical Field

[0001] This application relates to the field of switching power supply technology, and in particular to a topology control method, device and storage medium for a switching power supply. Background Technology

[0002] Currently, traditional switching power supply control methods typically employ fixed-parameter PID controllers. However, these struggle to achieve comprehensive optimization of efficiency, output quality, and dynamic response when facing complex operating conditions such as sudden load changes and switching of operating modes. In existing technologies, once the controller parameters are determined, they are difficult to adaptively adjust according to actual operating conditions, leading to problems such as low efficiency under light loads, large ripple under heavy loads, and slow response during sudden changes. Furthermore, traditional methods lack intelligent identification capabilities for the switching power supply's operating modes, failing to differentiate control based on the different characteristics of continuous and discontinuous inductor current modes. They also fail to fully utilize synchronous rectification technology to reduce conduction losses, making it difficult for overall performance to meet the demands of high-efficiency and high-reliability applications. Summary of the Invention

[0003] This application provides a topology control method, device, and storage medium for a switching power supply, which can improve the conversion efficiency and reliability of the switching power supply.

[0004] In a first aspect, embodiments of this application provide a topology control method for a switching power supply, the method comprising: Collect operating data of the switching power supply; The running data is subjected to Kalman filtering to obtain first state data and second state data; Based on the first state data, a corresponding dynamic model is selected from the preset model library, and the model parameters of the dynamic model are identified, and the confidence level of the dynamic model is calculated. Based on the model parameters and the second state data, a multi-objective optimization function is established within a preset prediction time domain, and the multi-objective optimization function is solved to obtain the target control sequence; The control duty cycle of the switching power supply is adjusted according to the target control sequence, and the zero-crossing time of the corresponding inductor current is detected. The target conduction phase of the synchronous rectifier is calculated based on the zero-crossing time. The first drive signal for the main switch and the second drive signal for the synchronous rectifier are generated based on the target conduction phase.

[0005] Secondly, embodiments of this application provide a power supply device, the power supply device including a memory and a processor; The memory is used to store computer programs; The processor is configured to execute the computer program and, when executing the computer program, implement the topology control method of the switching power supply as described in any of the embodiments of this application.

[0006] Thirdly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to implement the topology control method for a switching power supply as described in any of the embodiments of this application.

[0007] This application provides a topology control method for a switching power supply. The method includes: acquiring operating data of the switching power supply; performing Kalman filtering on the operating data to obtain first state data and second state data; selecting a corresponding dynamic model from a preset model library based on the first state data, identifying the model parameters of the dynamic model, and calculating the confidence level of the dynamic model; establishing a multi-objective optimization function within a preset prediction time domain based on the model parameters and the second state data, solving the multi-objective optimization function, and obtaining a target control sequence; adjusting the control duty cycle of the switching power supply according to the target control sequence, detecting the zero-crossing time of the corresponding inductor current, and calculating the target conduction phase of the synchronous rectifier based on the zero-crossing time; and generating a first drive signal for the main switch and a second drive signal for the synchronous rectifier based on the target conduction phase. In the above method, obtaining the first and second state data through Kalman filtering achieves state prediction-based advance control, effectively improving the dynamic response speed of the system. By intelligently identifying the working mode and selecting the corresponding dynamic model from the model library, accurate modeling and parameter identification under different working conditions are achieved, enabling the controller to adaptively adjust. Furthermore, by establishing a multi-objective optimization function, the optimal control sequence is solved in the prediction time domain, realizing the coordinated optimization of multiple performance indicators. Additionally, by detecting the zero-crossing moment of the inductor current to calculate the target conduction phase of the synchronous rectifier, the required drive signal is generated, effectively reducing conduction losses and improving the overall conversion efficiency. Attached Figure Description

[0008] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 A schematic flowchart of a topology control method for a switching power supply provided in an embodiment of this application; Figure 2 This is a schematic block diagram illustrating a control evaluation method for a switching power supply provided in an embodiment of this application. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described below with reference to the accompanying drawings.

[0011] The terms "first" and "second," etc., used in the specification, claims, and drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0012] The term "embodiment" as used herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. 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.

[0013] It should be understood that in this application, "at least one (item)" means one or more, "more than one" means two or more, "at least two (items)" means two or three or more, and "and / or" is used to describe the relationship between related objects, indicating that there can be three relationships. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the related objects before and after are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0014] Please see Figure 1 , Figure 1 This is a schematic flowchart illustrating a topology control method for a switching power supply provided in an embodiment of this application. Figure 1 As shown, the specific steps of the topology control method for this switching power supply include: S101-S106.

[0015] S101. Collect the operating data of the switching power supply.

[0016] For example, multi-dimensional parameter information such as output voltage data, output current data, input voltage data, and switching transistor temperature data are collected. To ensure the time consistency of the acquired data, multi-channel synchronous sampling technology is adopted, so that the sampling action of each channel is triggered at the same time, avoiding data distortion caused by inconsistent sampling times. Considering that the sampling accuracy requirements of the switching power supply vary under different operating conditions, the sampling frequency can be dynamically adjusted according to the load change rate. When a large load current change rate is detected, the sampling frequency is increased to a higher level to capture rapidly changing transient processes; when the load current change rate is small, the sampling frequency is reduced to a moderate level to save computing resources and reduce power consumption. The acquired raw operating data usually contains switching noise and measurement noise, which need to be filtered to obtain accurate status information.

[0017] S102. Perform Kalman filtering on the running data to obtain the first state data and the second state data.

[0018] For example, the collected operational data is input into a Kalman filter for processing. A state-space model of the switching power supply is established to describe the dynamic evolution of the state variables. The state-space model includes a state transition matrix and an observation matrix. The state transition matrix describes the transition relationship between state variables at adjacent time points, and the observation matrix describes the mapping relationship between observed values ​​and state variables. The Kalman filter calculates the prior state estimate for the current time point based on the state-space model and historical state data. This prior state estimate reflects the state information predicted based on the model. The Kalman gain is calculated using the collected operational data and the prior state estimate. The Kalman gain reflects the weight distribution relationship between observed data and model predictions in the state estimation. The prior state estimate is corrected based on the Kalman gain to obtain the posterior state estimate for the current time point as the first state data. The first state data represents the current optimal state estimate after fusing measurement information. After obtaining the first state data, the state transition matrix is ​​used to extrapolate the first state data to predict the state estimate for the next time point as the second state data. The second state data reflects the ability to predict future states based on the current state. By using Kalman filtering, not only is random noise in the operating data effectively suppressed, but also first-state data reflecting the current state and second-state data reflecting the future trend are obtained, providing dual data support for subsequent predictive control.

[0019] S103. Select the corresponding dynamic model from the preset model library based on the first state data, identify the model parameters of the dynamic model, and calculate the confidence level of the dynamic model.

[0020] For example, the current operating mode of the switching power supply is identified based on the obtained first state data. This identification is achieved by analyzing the inductor current data within the first state data. It is determined whether the inductor current remains continuous throughout the entire switching cycle. If the inductor current is always greater than zero throughout the switching cycle, the current operating mode is identified as a continuous inductor current mode. If the inductor current drops to zero during a period within the switching cycle, the proportion of this discontinuous period to the switching cycle needs to be further calculated. The proportion of the discontinuous period is compared with a preset critical proportion. When the proportion of the discontinuous period is greater than the preset critical proportion, the current operating mode is identified as a discontinuous inductor current mode; when the proportion of the discontinuous period is less than or equal to the preset critical proportion, the current operating mode is identified as a critically continuous inductor current mode. Based on the identified operating mode, a corresponding dynamic model is selected from a preset model library. This library pre-stores dynamic model structures established for different operating modes. A recursive least squares algorithm is used to identify the parameters of the selected dynamic model. Using historical input-output data within a time window, the transfer function parameters of the dynamic model are estimated by minimizing the prediction error. The transfer function parameters include key parameters such as gain coefficient, zero frequency, and pole frequency. After parameter identification is completed, the mean square error between the predicted output and the actual output of the dynamic model is calculated. The confidence level of the dynamic model is calculated based on the magnitude of the mean square error. The confidence level is inversely proportional to the mean square error. The smaller the mean square error, the higher the confidence level, indicating that the selected dynamic model describes the current working condition more accurately.

[0021] S104. Based on the model parameters and the second state data, establish a multi-objective optimization function within the preset prediction time domain, solve the multi-objective optimization function, and obtain the target control sequence.

[0022] For example, using the identified model parameters and second-state data as state prediction, a multi-objective optimization function is constructed within a preset prediction time domain to achieve synergistic optimization of multiple performance indicators. The multi-objective optimization function comprises three main components: an efficiency objective, a ripple objective, and a response speed objective. The efficiency objective is established based on switching and conduction losses within the prediction time domain, improving conversion efficiency by minimizing the weighted sum of losses. The ripple objective is established based on the degree of output voltage deviation from the reference voltage within the prediction time domain, reducing output ripple by minimizing the sum of squares of voltage deviations. The response speed objective is established based on the transition time required for the output voltage to reach the reference voltage, accelerating dynamic response by minimizing the settling time. The weighting coefficients of each objective are dynamically adjusted according to the current operating conditions. When a sudden load change is detected, the weighting coefficient of the response speed objective is increased while the weighting coefficient of the efficiency objective is decreased to prioritize fast response performance. During steady-state operation, the weighting coefficients of the efficiency and ripple objective are adjusted according to the current load rate, emphasizing efficiency optimization under light loads and ripple suppression under heavy loads. After establishing a weighted multi-objective optimization function that includes the above three objective terms, the optimization function is solved in the preset prediction time domain to obtain the target control sequence that achieves the optimal balance of multi-objective performance indicators. The target control sequence includes duty cycle control commands for multiple future control cycles.

[0023] S105. Adjust the control duty cycle of the switching power supply according to the target control sequence, detect the zero-crossing time of the corresponding inductor current, and calculate the target conduction phase of the synchronous rectifier based on the zero-crossing time.

[0024] For example, the duty cycle command corresponding to the current control cycle is extracted from the obtained target control sequence. The control duty cycle of the switching power supply is adjusted according to this command to achieve precise control of the main switch's on-time. While adjusting the duty cycle, the direction change of the inductor current is monitored in real time to capture the current zero-crossing moment. When the inductor current changes from positive to zero, this moment is recorded as the first zero-crossing moment, corresponding to the turning point where the inductor current changes from releasing energy in the positive direction to zero current. The dead-time delay is calculated based on the first zero-crossing moment and the main switch's turn-off moment. The dead-time delay is a protection interval set to prevent the main switch and synchronous rectifier from conducting simultaneously. Combining the dead-time delay and the switching characteristic parameters of the synchronous rectifier, the initial conduction phase of the synchronous rectifier is calculated. This initial conduction phase ensures that the synchronous rectifier immediately takes over the current path after the body diode begins to conduct. To further optimize conduction losses, the current conduction loss and switching loss are calculated based on the current data in the first state data, and a loss model describing the relationship between the conduction phase and the total loss is established. Under the constraints of the loss model, the target conduction phase that minimizes the total loss is solved by an iterative optimization algorithm. This target conduction phase can minimize the conduction time of the body diode while ensuring reliable commutation, thereby reducing conduction loss and improving overall conversion efficiency.

[0025] S106. Generate the first drive signal of the main switch and the second drive signal of the synchronous rectifier according to the target conduction phase.

[0026] For example, based on the adjusted control duty cycle and the optimized target conduction phase, a first drive signal for driving the main switch and a second drive signal for driving the synchronous rectifier are generated. The pulse width of the first drive signal is determined by the control duty cycle, which directly determines the proportion of the main switch's conduction time within one switching cycle. The conduction time of the second drive signal is determined by the target conduction phase, and the time delay of the target conduction phase relative to the main switch's turn-off time precisely controls the synchronous rectifier's turn-on timing. When generating the first and second drive signals, the dead time requirement between the two drive signals is ensured to avoid the risk of shoot-through conduction in the main switch and synchronous rectifier. After being amplified by the drive circuit, the first and second drive signals are applied to the gates of the main switch and synchronous rectifier, respectively, achieving precise control of the power switching devices. By combining the control duty cycle obtained based on multi-objective optimization with the target conduction phase obtained based on zero-crossing detection optimization, the generated first and second drive signals can simultaneously achieve a synergistic improvement in output regulation performance and loss optimization performance, enabling the switching power supply to maintain high conversion efficiency while meeting output requirements.

[0027] This application provides a topology control method for a switching power supply. The method includes: acquiring operating data of the switching power supply; performing Kalman filtering on the operating data to obtain first state data and second state data; selecting a corresponding dynamic model from a preset model library based on the first state data, identifying the model parameters of the dynamic model, and calculating the confidence level of the dynamic model; establishing a multi-objective optimization function within a preset prediction time domain based on the model parameters and the second state data, solving the multi-objective optimization function, and obtaining a target control sequence; adjusting the control duty cycle of the switching power supply according to the target control sequence, detecting the zero-crossing time of the corresponding inductor current, and calculating the target conduction phase of the synchronous rectifier based on the zero-crossing time; and generating a first drive signal for the main switch and a second drive signal for the synchronous rectifier based on the target conduction phase. In the above method, obtaining the first and second state data through Kalman filtering achieves state prediction-based advance control, effectively improving the dynamic response speed of the system. By intelligently identifying the working mode and selecting the corresponding dynamic model from the model library, accurate modeling and parameter identification under different working conditions are achieved, enabling the controller to adaptively adjust. Furthermore, by establishing a multi-objective optimization function, the optimal control sequence is solved in the prediction time domain, realizing the coordinated optimization of multiple performance indicators. Additionally, by detecting the zero-crossing moment of the inductor current to calculate the target conduction phase of the synchronous rectifier, the required drive signal is generated, effectively reducing conduction losses and improving the overall conversion efficiency.

[0028] To more clearly illustrate the technical solution of this application, the technical solution of this application will be described below through specific embodiments. It should be noted that the specific embodiments are used to expand the description of the technical solution of this application, and are not intended to limit this application.

[0029] In some embodiments, such as Figure 2 As shown, after generating the first drive signal of the main switch and the second drive signal of the synchronous rectifier according to the target conduction phase, the method further includes: S107-S109.

[0030] S107. Acquire the output voltage under the action of the drive signal, calculate the gain margin and phase margin based on the output voltage, and determine whether the gain margin and phase margin meet the stability threshold. For example, after the first and second drive signals are applied to the main switch and synchronous rectifier, the output voltage data of the switching power supply under the current control strategy is acquired. The stability characteristics of the closed-loop control are evaluated by performing a Fast Fourier Transform (FFT) analysis on the output voltage data. The acquired output voltage data is converted to the frequency domain, and the main frequency components, amplitude, and phase information of the output voltage are extracted. Based on the frequency domain analysis results, the open-loop transfer function characteristics of the switching power supply's closed-loop control are reconstructed. By analyzing the amplitude-frequency response curve of the open-loop transfer function, the phase margin is calculated at the frequency point where the amplitude crosses the zero-decibel line. The phase margin reflects the margin of the phase angle from -180 degrees; a larger phase margin indicates that the closed-loop control is further from the instability boundary. By analyzing the phase-frequency response curve of the open-loop transfer function, the gain margin is calculated at the frequency point where the phase angle reaches -180 degrees. The gain margin reflects the margin of the amplitude from the zero-decibel line; a larger gain margin indicates that the closed-loop control has a stronger tolerance to gain fluctuations. After obtaining the calculated results of the gain margin and phase margin, it is determined whether the gain margin is greater than a preset gain margin threshold and whether the phase margin is greater than a preset phase margin threshold. If both the gain margin and phase margin are greater than the preset gain margin thresholds, the gain margin and phase margin are determined to meet the stability thresholds, indicating that the current control strategy can guarantee the stable operation of the closed-loop control. If either the gain margin or phase margin is not greater than the preset gain margin threshold or the phase margin is not greater than the preset phase margin threshold, the gain margin and phase margin are determined to not meet the stability thresholds, and the model parameters need to be adjusted to improve the stability margin.

[0031] S108. If satisfied, extract the feature vector of the operating condition of the switching power supply, associate the feature vector with the model parameters and store it in the empirical parameter library. For example, when the judgment result indicates that the gain margin and phase margin meet the stability threshold, feature extraction is performed on the current operating condition of the switching power supply to establish an empirical knowledge base. The feature vector of the operating condition includes statistical feature parameters in multiple dimensions, specifically including the average load rate calculated within a time window, the statistical characteristics of the load change rate, the statistical characteristics of the input voltage, and information such as ambient temperature. The average load rate is characterized by the ratio of the average output current within the time window to the rated current. The statistical characteristics of the load change rate describe the severity of load fluctuations by calculating the mean and standard deviation of the output current change rate. The statistical characteristics of the input voltage reflect the power supply input conditions by calculating the mean and fluctuation range of the input voltage. After extracting the feature vector, the similarity between this feature vector and each feature vector stored in the empirical parameter base is calculated. The similarity calculation can use metrics such as Euclidean distance or cosine similarity. The system determines whether the calculated similarity is greater than a preset similarity threshold. If the similarity is less than or equal to the preset threshold, it indicates a significant difference between the current working condition and historical working conditions in the experience parameter library. The current feature vector and its corresponding model parameters are then stored as new entries in the experience parameter library. If the similarity is greater than the preset threshold, it indicates that a similar working condition already exists in the experience parameter library. The model parameters for the corresponding entries in the experience parameter library are updated, replacing or weighting and fusing historical model parameters with the currently identified model parameters. By continuously accumulating the correlation between feature vectors and model parameters under different working conditions, the experience parameter library can provide initial parameter values ​​for subsequent encounters with similar working conditions, enabling rapid parameter retrieval based on historical experience.

[0032] S109. If not satisfied, perform conservative adjustments to the model parameters.

[0033] For example, when the judgment result indicates that the gain margin and phase margin do not meet the stability threshold, the currently used model parameters need to be conservatively adjusted to improve the stability margin. The conservative adjustment strategy is differentiated according to the specific margin deficiency. When the gain margin is not greater than the preset gain margin threshold, it indicates that the open-loop gain is too large and may cause the closed-loop control to approach the instability boundary. In this case, the gain coefficient in the model parameters is reduced to increase the gain margin by reducing the controller gain. The reduction of the gain coefficient is determined according to the degree of gain margin deficiency. When the phase margin is not greater than the preset phase margin threshold, it indicates that the phase lag is too large and may increase the oscillation risk of the closed-loop control. In this case, the zero frequency in the model parameters is reduced to increase the phase lead compensation effect by reducing the zero frequency, thereby increasing the phase margin. The reduction of the zero frequency is determined according to the degree of phase margin deficiency. The adjusted model parameters are applied to the duty cycle calculation and drive signal generation of the next control cycle. After applying the adjusted model parameters, the adjustment effect is continuously monitored, and the output voltage acquisition and gain margin and phase margin calculation processes are repeated. Through multiple iterative adjustments, until both the gain margin and phase margin meet the stability threshold requirements, the closed-loop control is rolled back from the unstable edge to the stable region. This conservative adjustment mechanism automatically triggers parameter correction when insufficient stability margin is detected, preventing oscillations or instability caused by improper parameters and ensuring the reliability and robustness of the switching power supply control.

[0034] In some embodiments, collecting operating data of the switching power supply includes: calculating the load current change rate within a preset time window; determining whether the load current change rate is greater than a first preset change rate threshold; if so, adjusting the sampling frequency to a first preset sampling frequency; if not, determining whether the load current change rate is less than a second preset change rate threshold; when the load current change rate is less than the second preset change rate threshold, adjusting the sampling frequency to a second preset sampling frequency, wherein the first preset sampling frequency is greater than the second preset sampling frequency; and synchronously collecting the output voltage, output current, input voltage, and temperature data of the switching power supply as operating data according to the adjusted sampling frequency.

[0035] For example, in the actual operation of a switching power supply, the selection of the sampling frequency has a significant impact on capturing state changes and saving computing resources. To achieve adaptive adjustment of the sampling frequency, the changes in load current are analyzed within a preset time window. The dynamic characteristics of the load are evaluated by calculating the rate of change of the load current within the time window. The preset time window can be set to 10 to 50 switching cycles. Within this time window, a sequence of sampled load current values ​​is recorded. The rate of change of the load current is calculated by statistically analyzing the current difference between adjacent sampling points. The calculated rate of change of the load current is compared with a first preset rate of change threshold, which is typically set to 5% to 10% of the rated current per millisecond. When the rate of change of the load current exceeds the first preset rate of change threshold, it indicates that the load is in a rapidly changing state. At this time, the sampling frequency is adjusted to the first preset sampling frequency, which can be set to 10 to 20 times the switching frequency. For example, when the switching frequency is 100kHz, the first preset sampling frequency can be set to 1MHz to 2MHz. When the load current change rate is not greater than the first preset change rate threshold, the system continues to determine whether the load current change rate is less than the second preset change rate threshold. The second preset change rate threshold is typically set to 1% to 2% of the rated current per millisecond. When the load current change rate is less than the second preset change rate threshold, it indicates that the load is in a slow-changing or steady-state operating state. In this case, the sampling frequency is adjusted to the second preset sampling frequency, which can be set to five to ten times the switching frequency. For example, when the switching frequency is 100kHz, the second preset sampling frequency can be set to 500kHz to 1MHz. When the load current change rate is between the second preset change rate threshold and the first preset change rate threshold, the current sampling frequency remains unchanged or a medium sampling frequency is used. Based on the adjusted sampling frequency, the synchronous sampling function of the multi-channel analog-to-digital converter is activated, triggering the sampling action of each channel simultaneously to collect the output voltage data, output current data, input voltage data, and temperature data of the switching power supply as operating data.

[0036] In some embodiments, performing Kalman filtering on the operating data to obtain first state data and second state data includes: establishing a state-space model of the switching power supply; wherein the state-space model includes a state transition matrix and an observation matrix; calculating the prior state estimate at the current moment based on the state-space model and preset historical state data; calculating the Kalman gain based on the operating data and the prior state estimate; updating the posterior state estimate at the current moment based on the Kalman gain, and determining the posterior state estimate as the first state data; calculating the state prediction at the next moment based on the posterior state estimate and the state transition matrix, and determining the state prediction as the second state data.

[0037] For example, to extract accurate state information and achieve state prediction from noisy operating data, a state-space model describing the dynamic characteristics of the switching power supply is established. The state-space model is in discrete-time form. The state equations describe the evolution of state variables between adjacent sampling times, and the observation equations describe the mapping relationship between observed values ​​and state variables. The state transition matrix is ​​derived based on the circuit topology and operating principle of the switching power supply. For a Buck converter, the state variables can be selected as inductor current and output capacitor voltage. The elements of the state transition matrix are determined by circuit parameters such as inductance value, capacitance value, switching cycle, and load resistance. The observation matrix describes the linear relationship between measurable observations such as output voltage and output current and state variables. After establishing the state-space model, the prior state estimate for the current moment is calculated based on the state-space model and preset historical state data. The historical state data refers to the posterior state estimate updated by Kalman filtering at the previous sampling time. The prior state estimate is obtained by multiplying the historical state data by the state transition matrix, reflecting the current state information based on model prediction. The Kalman gain is calculated using the collected operational data and the calculated prior state estimate. The calculation of the Kalman gain involves the prior estimation error covariance matrix, the observation matrix, and the observation noise covariance matrix. The Kalman gain reflects the allocation of confidence between the model prediction and the actual observation in state estimation. When the observation noise is low, the Kalman gain is high, and the state estimation relies more on the observation data. When the model uncertainty is low, the Kalman gain is low, and the state estimation relies more on the model prediction. The prior state estimate is corrected based on the calculated Kalman gain by adding the product of the Kalman gain and the innovation (observation value minus predicted observation value) to obtain the posterior state estimate at the current time step. This posterior state estimate is designated as the first state data, representing the optimal estimate of the current state after fusing observation information. After obtaining the first state data, extrapolation is performed using the first state data and the state transition matrix. Multiplying the first state data by the state transition matrix yields the predicted state value for the next time step. This predicted state value is designated as the second state data, reflecting the predictive ability for the state at the next sampling time step.

[0038] In some embodiments, the first state data includes: inductor current sequence values; selecting a corresponding dynamic model from a preset model library based on the first state data; identifying the model parameters of the dynamic model; and calculating the confidence level of the dynamic model, including: determining whether the inductor current sequence values ​​remain continuous throughout the entire switching cycle; if so, selecting a dynamic model corresponding to the continuous inductor current mode from the preset model library; if not, calculating the proportion of the discontinuous period of the inductor current sequence values ​​to the switching cycle, and determining whether the proportion is greater than a preset critical proportion; when the proportion is greater than the preset critical proportion, selecting a dynamic model corresponding to the discontinuous inductor current mode from the preset model library; when the proportion is not greater than the preset critical proportion, selecting a dynamic model corresponding to the critical continuous inductor current mode from the preset model library; using a recursive least squares algorithm to identify the transfer function parameters of the selected dynamic model based on historical input and output data; wherein the transfer function parameters include gain coefficient, zero frequency, and pole frequency; calculating the mean square error between the predicted output and the actual output of the dynamic model; and calculating the confidence level based on the mean square error; wherein the confidence level is inversely proportional to the mean square error.

[0039] For example, the first state data includes inductor current sequence values, which record the numerical sequence of inductor current changes over time within one or more switching cycles. The current operating mode of the switching power supply is identified by analyzing the continuity characteristics of the inductor current sequence values. To determine whether the inductor current sequence values ​​remain continuous throughout the entire switching cycle, a specific method is to check whether there are sampling points in the inductor current sequence values ​​that drop to zero or close to zero. When the inductor current sequence values ​​remain positive throughout the entire switching cycle, it indicates that the inductor current is in continuous mode, and the dynamic model corresponding to the continuous inductor current mode is selected from a preset model library. The dynamic model for the continuous inductor current mode is typically described using a second-order transfer function, containing one zero and two complex conjugate poles. When there are periods in the inductor current sequence values ​​that drop to zero, it indicates that the inductor current is discontinuous, and it is necessary to further calculate the proportion of the discontinuous period to the switching cycle. The proportion of the discontinuous period to the switching cycle is calculated by counting the number of sampling points in the inductor current sequence values ​​that are less than a preset current threshold (e.g., one percent of the rated current), and the ratio of this number to the total number of sampling points in the switching cycle. The proportion of the discontinuous period to the switching cycle is compared with a preset critical proportion, which can be set to 10% to 20%. When the proportion of the discontinuous period to the switching cycle is greater than the preset critical proportion, it indicates that the inductor current is in a significantly discontinuous operating state. A dynamic model corresponding to the discontinuous inductor current mode is selected from the preset model library. The dynamic model for the discontinuous inductor current mode considers the characteristic of the output capacitor supplying power to the load alone during the current discontinuity period; the model order and parameters differ significantly from the continuous mode. When the proportion of the discontinuous period to the switching cycle is not greater than the preset critical proportion, it indicates that the inductor current is in a critical state between the continuous and discontinuous modes. A dynamic model corresponding to the critical continuous mode of the inductor current is selected from the preset model library. After selecting the dynamic model, a recursive least squares algorithm is used to identify the parameters of the selected dynamic model. The recursive least squares algorithm utilizes historical input-output data, including the duty cycle input sequence and output voltage output sequence within a past time window. The transfer function parameters, including the gain coefficient, zero frequency, and pole frequency, are estimated by minimizing the squared error between the predicted output and the actual output. The gain coefficient reflects the steady-state response of the output voltage to changes in the duty cycle. The zero-point frequency affects the speed of the dynamic response, while the pole frequency determines the bandwidth and stability of the closed-loop control. After identifying the transfer function parameters, the predicted output is calculated using the identified dynamic model. The predicted output is then compared with the actual output, and the mean square error (MSE) between the two is calculated. The MSE is obtained by summing the squares of the errors at each sampling point in the prediction error sequence and taking the average.The confidence level of the dynamic model is calculated based on the mean squared error. The confidence level is inversely proportional to the mean squared error. An exponential function can be used to establish the mapping relationship between the confidence level and the mean squared error, so that the smaller the mean squared error, the closer the confidence level is to 1, and the larger the mean squared error, the closer the confidence level is to 0.

[0040] In some embodiments, after calculating the confidence level based on the mean squared error, the method further includes: determining whether the confidence level is less than a preset confidence threshold; when the confidence level is less than the preset confidence threshold, selecting a candidate dynamic model from a preset model library, and re-executing the identification of transfer function parameters and the calculation of confidence level; when the preset model library contains dynamic models corresponding to multiple working modes, calculating the weighting coefficient of each dynamic model, performing weighted fusion of the model parameters of each dynamic model, and obtaining fused model parameters, wherein the weighting coefficient is determined based on the confidence level of each dynamic model.

[0041] For example, the first state data includes inductor current sequence values, which record the numerical sequence of inductor current changes over time within one or more switching cycles. The current operating mode of the switching power supply is identified by analyzing the continuity characteristics of the inductor current sequence values. To determine whether the inductor current sequence values ​​remain continuous throughout the entire switching cycle, a specific method is to check whether there are sampling points in the inductor current sequence values ​​that drop to zero or close to zero. When the inductor current sequence values ​​remain positive throughout the entire switching cycle, it indicates that the inductor current is in continuous mode, and the dynamic model corresponding to the continuous inductor current mode is selected from a preset model library. The dynamic model for the continuous inductor current mode is typically described using a second-order transfer function, containing one zero and two complex conjugate poles. When there are periods in the inductor current sequence values ​​that drop to zero, it indicates that the inductor current is discontinuous, and it is necessary to further calculate the proportion of the discontinuous period to the switching cycle. The proportion of the discontinuous period to the switching cycle is calculated by counting the number of sampling points in the inductor current sequence values ​​that are less than a preset current threshold (e.g., one percent of the rated current), and the ratio of this number to the total number of sampling points in the switching cycle. The proportion of the discontinuous period to the switching cycle is compared with a preset critical proportion, which can be set to 10% to 20%. When the proportion of the discontinuous period to the switching cycle is greater than the preset critical proportion, it indicates that the inductor current is in a significantly discontinuous operating state. A dynamic model corresponding to the discontinuous inductor current mode is selected from the preset model library. The dynamic model for the discontinuous inductor current mode considers the characteristic of the output capacitor supplying power to the load alone during the current discontinuity period; the model order and parameters differ significantly from the continuous mode. When the proportion of the discontinuous period to the switching cycle is not greater than the preset critical proportion, it indicates that the inductor current is in a critical state between the continuous and discontinuous modes. A dynamic model corresponding to the critical continuous mode of the inductor current is selected from the preset model library. After selecting the dynamic model, a recursive least squares algorithm is used to identify the parameters of the selected dynamic model. The recursive least squares algorithm utilizes historical input-output data, including the duty cycle input sequence and output voltage output sequence within a past time window. The transfer function parameters, including the gain coefficient, zero frequency, and pole frequency, are estimated by minimizing the squared error between the predicted output and the actual output. The gain coefficient reflects the steady-state response of the output voltage to changes in the duty cycle. The zero-point frequency affects the speed of the dynamic response, while the pole frequency determines the bandwidth and stability of the closed-loop control. After identifying the transfer function parameters, the predicted output is calculated using the identified dynamic model. The predicted output is then compared with the actual output, and the mean square error (MSE) between the two is calculated. The MSE is obtained by summing the squares of the errors at each sampling point in the prediction error sequence and taking the average.The confidence level of the dynamic model is calculated based on the mean squared error. The confidence level is inversely proportional to the mean squared error. An exponential function can be used to establish the mapping relationship between the confidence level and the mean squared error, so that the smaller the mean squared error, the closer the confidence level is to 1, and the larger the mean squared error, the closer the confidence level is to 0.

[0042] In some embodiments, a multi-objective optimization function is established within a preset prediction time domain, and the multi-objective optimization function is solved to obtain a target control sequence. This includes: establishing an efficiency target term based on predicted switching losses and predicted conduction losses within the prediction time domain; establishing a ripple target term based on the degree to which the output voltage deviates from the reference voltage within the prediction time domain; establishing a response speed target term based on the time required for the output voltage to reach the reference voltage; determining whether the switching power supply is in a sudden load change state; if so, increasing the weight coefficient of the response speed target term and decreasing the weight coefficient of the efficiency target term; if not, adjusting the weight coefficients of the efficiency target term and the ripple target term according to the current load rate; establishing a weighted multi-objective optimization function containing the efficiency target term, the ripple target term, and the response speed target term; solving the weighted multi-objective optimization function to obtain the target control sequence.

[0043] After obtaining the model parameters and second-state data, a multi-objective optimization function is established within a preset prediction time domain to achieve synergistic optimization of multiple performance indicators. The preset prediction time domain can be set to 5 to 20 control cycles, and the corresponding control duty cycle needs to be determined for each control cycle within the prediction time domain. An efficiency objective term is established based on the predicted switching loss and predicted conduction loss within the prediction time domain. The predicted switching loss is calculated based on the switching characteristic parameters of the switching transistor and the predicted switching current, while the predicted conduction loss is calculated based on the on-resistance of the switching transistor and the predicted effective value of the conduction current. The efficiency objective term is represented as the sum of the switching loss and conduction loss for each control cycle within the prediction time domain. The total energy loss is reduced by minimizing the efficiency objective term. A ripple objective term is established based on the degree to which the output voltage deviates from the reference voltage within the prediction time domain. In each control cycle within the prediction time domain, the output voltage value is predicted using the dynamic model and the control duty cycle, and the deviation between the predicted output voltage value and the reference voltage is calculated. The ripple objective term is represented as the sum of the squares of the output voltage deviations for each control cycle within the prediction time domain. The output voltage ripple is reduced by minimizing the ripple objective term. A response speed target is established based on the time required for the output voltage to reach the reference voltage. This target measures the adjustment time needed to transition from the current output voltage state to the reference voltage. The response speed target can be expressed as the number of control cycles required for the output voltage to enter the reference voltage error band (e.g., ±1% of the reference voltage). After establishing these three target items, the weighting coefficients of each target item need to be adjusted according to the current operating state of the switching power supply. The switching power supply is then assessed for a sudden load change, determined by whether the load current change rate exceeds the sudden change threshold or whether the load current change amplitude exceeds a preset percentage of the rated current. When the switching power supply is in a sudden load change state, rapid response becomes the primary performance requirement. Increasing the weighting coefficient of the response speed target item can raise it from 0.2 in steady state to 0.6, while decreasing the weighting coefficient of the efficiency target item can lower it from 0.5 in steady state to 0.2. The weighting coefficient of the ripple target item remains at a moderate level, for example, 0.2. When the switching power supply is not in a sudden load change state, the weighting coefficients of the efficiency and ripple target items are adjusted according to the current load rate. The current load rate is calculated as the ratio of output current to rated current. When the current load rate is less than 30%, the system is in a light load operation state. The weighting factor for the efficiency objective is increased to 0.6, the weighting factor for the ripple objective is decreased to 0.2, and the weighting factor for the response speed objective is set to 0.2. When the current load rate is greater than 70%, the system is in a heavy load operation state. The weighting factor for the ripple objective is increased to 0.6, and the weighting factor for the efficiency objective is decreased to 0.2. A weighted multi-objective optimization function is established, comprising the efficiency, ripple, and response speed objectives. This function is expressed as a weighted sum of the three objective terms multiplied by their respective weighting factors.To solve the weighted multi-objective optimization function, under the constraints of duty cycle (duty cycle range between 0 and 1) and current constraint (inductor current does not exceed saturation current), optimization algorithms such as quadratic programming or gradient descent are used to calculate the duty cycle sequence that minimizes the weighted multi-objective optimization function. This duty cycle sequence is the target control sequence.

[0044] In some embodiments, detecting the zero-crossing moment of the corresponding inductor current and calculating the target conduction phase of the synchronous rectifier based on the zero-crossing moment includes: real-time monitoring of the direction change of the inductor current; recording the first zero-crossing moment when the inductor current changes from positive to zero; calculating the dead-time delay based on the first zero-crossing moment and the turn-off moment of the main switch; calculating the initial conduction phase of the synchronous rectifier based on the dead-time delay and the switching characteristic parameters of the synchronous rectifier; calculating the current conduction loss and the current switching loss based on the current data in the first state data; establishing a loss model based on the initial conduction phase, the current conduction loss, and the current switching loss, the loss model being used to describe the relationship between the conduction phase and the total loss; and obtaining the target conduction phase that minimizes the total loss by solving an iterative optimization algorithm under the constraints of the loss model.

[0045] For example, after adjusting the control duty cycle of the switching power supply according to the target control sequence, it is necessary to optimize the turn-on time of the synchronous rectifier to reduce conduction losses. The direction change of the inductor current is monitored in real time, and the real-time value and sign of the inductor current are obtained through a current sensor or current sampling circuit. When the inductor current is detected to change from positive to zero, it indicates that the inductor's energy release process has ended and it is about to enter the freewheeling stage or intermittent state; this moment is recorded as the first zero-crossing moment. The detection of the first zero-crossing moment can be achieved by judging whether the inductor current sampling value changes from greater than zero to less than or equal to zero. To improve detection accuracy, a zero-crossing comparator or interpolation algorithm can be used to accurately locate the zero-crossing moment. The dead time delay is calculated based on the first zero-crossing moment and the turn-off moment of the main switch. The turn-off moment of the main switch refers to the moment when the main switch drive signal changes from high level to low level. The dead time delay is equal to the time difference between the first zero-crossing moment and the turn-off moment of the main switch, reflecting the time elapsed from the main switch being turned off to the inductor current dropping to zero. The initial conduction phase of the synchronous rectifier (SRRC) is calculated based on the dead-time delay and the switching characteristic parameters of the SRRC, including the turn-on delay and rise time. The initial conduction phase is set to a short delay after the first zero-crossing moment, slightly longer than the turn-on delay of the SRRC, ensuring that the SRRC can promptly take over the current path after the body diode begins to conduct. To further optimize conduction losses, the current conduction loss and current switching loss are calculated based on the current data in the first state data. The current conduction loss is calculated based on the on-resistance of the SRRC and the effective value of the current flowing through it, while the current switching loss is calculated based on the switching characteristic parameters and switching frequency of the SRRC. A loss model is established based on the initial conduction phase, current conduction loss, and current switching loss, describing the functional relationship between the conduction phase and total loss. When the SRRC conduction phase is advanced, the body diode conduction time is shortened, and the conduction loss decreases. However, if the conduction phase is too advanced, it may increase switching losses or introduce other losses. The loss model comprehensively considers the variation of conduction loss and switching loss with the conduction phase, establishing the total loss as a function of the conduction phase. Under the constraints of the loss model, an iterative optimization algorithm is used to find the target conduction phase that minimizes the total loss. The iterative optimization algorithm can employ Newton's method or the golden section method, searching for the optimal solution within the feasible region near the initial conduction phase, calculating the total loss corresponding to different conduction phases, and finding the target conduction phase that minimizes the total loss.

[0046] This application provides a power supply device, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and, when executing the computer program, implement the topology control method of the switching power supply as described in any of the embodiments of this application.

[0047] This application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it enables the processor to implement a topology control method for a switching power supply as described in any of the embodiments of this application.

[0048] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A topology control method for a switching power supply, characterized in that, The method includes: Collect operating data of the switching power supply; The running data is subjected to Kalman filtering to obtain first state data and second state data; Based on the first state data, a corresponding dynamic model is selected from the preset model library, and the model parameters of the dynamic model are identified, and the confidence level of the dynamic model is calculated. Based on the model parameters and the second state data, a multi-objective optimization function is established within a preset prediction time domain, and the multi-objective optimization function is solved to obtain the target control sequence; The control duty cycle of the switching power supply is adjusted according to the target control sequence, and the zero-crossing time of the corresponding inductor current is detected. The target conduction phase of the synchronous rectifier is calculated based on the zero-crossing time. The first drive signal for the main switch and the second drive signal for the synchronous rectifier are generated based on the target conduction phase.

2. The topology control method for a switching power supply as described in claim 1, characterized in that, After generating the first drive signal for the main switch and the second drive signal for the synchronous rectifier based on the target conduction phase, the method further includes: The output voltage under the action of the driving signal is collected, the gain margin and phase margin are calculated based on the output voltage, and it is determined whether the gain margin and the phase margin meet the stability threshold. If the conditions are met, extract the feature vector of the operating condition of the switching power supply, associate the feature vector with the model parameters and store it in the empirical parameter library; If the conditions are not met, conservative adjustments are made to the model parameters.

3. The topology control method for a switching power supply according to claim 1, characterized in that, The collected operating data of the switching power supply includes: Calculate the rate of change of load current within a preset time window; Determine whether the rate of change of the load current is greater than a first preset rate of change threshold; If so, adjust the sampling frequency to the first preset sampling frequency; If not, determine whether the load current change rate is less than the second preset change rate threshold; When the load current change rate is less than the second preset change rate threshold, the sampling frequency is adjusted to the second preset sampling frequency, wherein the first preset sampling frequency is greater than the second preset sampling frequency; Based on the adjusted sampling frequency, the output voltage, output current, input voltage, and temperature data of the switching power supply are synchronously collected as the operating data.

4. The topology control method for a switching power supply according to claim 1, characterized in that, The step of performing Kalman filtering on the running data to obtain first state data and second state data includes: Establish a state-space model of the switching power supply; wherein, the state-space model includes a state transition matrix and an observation matrix; Based on the state space model and the preset historical state data, calculate the prior state estimate for the current moment; Calculate the Kalman gain based on the operational data and the prior state estimate; The posterior state estimate at the current moment is updated based on the Kalman gain, and the posterior state estimate is determined as the first state data. Based on the posterior state estimate and the state transition matrix, the predicted state value for the next time step is calculated, and the predicted state value is determined as the second state data.

5. The topology control method for a switching power supply according to claim 1, characterized in that, The first state data includes: inductor current sequence values. The step of selecting a corresponding dynamic model from a preset model library based on the first state data, identifying the model parameters of the dynamic model, and calculating the confidence level of the dynamic model includes: Determine whether the inductor current sequence remains continuous throughout the entire switching cycle; If so, select the dynamic model corresponding to the continuous inductor current mode from the preset model library; If not, calculate the proportion of the discontinuous period of the inductor current sequence value to the switching cycle, and determine whether the proportion is greater than a preset critical proportion. When the ratio is greater than the preset critical ratio, a dynamic model corresponding to the discontinuous inductor current mode is selected from the preset model library. When the ratio is not greater than the preset critical ratio, select the dynamic model corresponding to the critical continuous mode of inductor current from the preset model library. A recursive least squares algorithm is used to identify the transfer function parameters of the selected dynamic model based on historical input and output data; wherein, the transfer function parameters include the gain coefficient, zero frequency, and pole frequency; Calculate the mean square error between the predicted output and the actual output of the dynamic model; The confidence level is calculated based on the mean square error; wherein the confidence level is inversely proportional to the mean square error.

6. The topology control method for a switching power supply according to claim 5, characterized in that, After calculating the confidence level based on the mean square error, the method further includes: Determine whether the confidence level is less than a preset confidence threshold; When the confidence level is less than the preset confidence threshold, a candidate dynamic model is selected from the preset model library, and the identification of the transfer function parameters and the calculation of the confidence level are re-executed. When the preset model library contains dynamic models corresponding to multiple working modes, the weighting coefficient of each dynamic model is calculated, and the model parameters of each dynamic model are weighted and fused to obtain fused model parameters. The weighting coefficient is determined according to the confidence level of each dynamic model.

7. The topology control method for a switching power supply according to claim 1, characterized in that, The step of establishing a multi-objective optimization function within a preset prediction time domain, solving the multi-objective optimization function, and obtaining the target control sequence includes: An efficiency target term is established based on the predicted switching loss and predicted conduction loss in the predicted time domain. A ripple target term is established based on the degree to which the output voltage deviates from the reference voltage in the predicted time domain; Establish a response speed target based on the time required for the output voltage to reach the reference voltage; Determine whether the switching power supply is under a sudden load change. If so, increase the weighting coefficient of the response speed target item and decrease the weighting coefficient of the efficiency target item; If not, adjust the weighting coefficients of the efficiency target and the ripple target based on the current load rate; A weighted multi-objective optimization function is established, which includes the efficiency objective, the ripple objective, and the response speed objective. The weighted multi-objective optimization function is solved to obtain the objective control sequence.

8. The topology control method for a switching power supply according to claim 1, characterized in that, The zero-crossing time of the detected inductor current, and the calculation of the target conduction phase of the synchronous rectifier based on the zero-crossing time, including: Real-time monitoring of changes in the direction of the inductor current; When the inductor current is detected to change from positive to zero, the first zero-crossing moment is recorded; Calculate the dead time delay based on the first zero-crossing moment and the turn-off moment of the main switch. The initial conduction phase of the synchronous rectifier is calculated based on the dead time and the switching characteristic parameters of the synchronous rectifier. Calculate the current conduction loss and the current switching loss based on the current data in the first state data; A loss model is established based on the initial conduction phase, the current conduction loss, and the current switching loss. The loss model is used to describe the relationship between the conduction phase and the total loss. Under the constraints of the loss model, the target conduction phase that minimizes the total loss is obtained by solving an iterative optimization algorithm.

9. A power supply device, characterized in that, The power supply device includes a memory and a processor; The memory is used to store computer programs; The processor is configured to execute the computer program and, in executing the computer program, implement the topology control method for the switching power supply as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to implement the topology control method for the switching power supply as described in any one of claims 1 to 7.

Images