A method, device, medium and product for electromagnetic interference prediction of a synchronous step-down circuit

Abstract

The application discloses an electromagnetic interference prediction method and device of a synchronous step-down circuit, a medium and a product, relates to the field of power electronic system electromagnetic compatibility and intelligent prediction, and comprises the following steps: acquiring and preprocessing circuit key parameters and corresponding measured electromagnetic interference frequency spectrum of the synchronous step-down circuit; an attention mechanism is introduced into a full connection structure of an HA-DNN prediction model; the preprocessed circuit key parameters are input into the HA-DNN prediction model for forward propagation, a predicted electromagnetic interference frequency spectrum is dynamically generated in combination with the attention mechanism, the predicted electromagnetic interference frequency spectrum is compared with the measured electromagnetic interference frequency spectrum, a loss function is constructed, a backward propagation algorithm is combined, a trained HA-DNN prediction model is obtained, and the output is a differential mode and common mode conducted electromagnetic interference frequency spectrum prediction value of the synchronous step-down circuit in a full frequency band of 150 kHz-30 MHz, so that the generalization ability and prediction precision are improved, and the explainability analysis of parameter influence is realized through attention weights.

Description

Technical Field

[0001] This application relates to the field of electromagnetic compatibility and intelligent prediction in power electronic systems, and in particular to an electromagnetic interference prediction method, device, medium, and product for a synchronous buck circuit. Background Technology

[0002] As power electronic systems evolve towards higher frequencies and greater integration, the nonlinear coupling effect between parasitic parameters and switching transient processes within synchronous buck circuits intensifies, leading to complex harmonic distributions in conducted electromagnetic interference (EMI). Existing EMI prediction methods for synchronous buck circuits suffer from the following technical problems: Firstly, traditional analytical modeling and numerical simulation methods struggle to balance the accuracy of representing high-frequency parasitic coupling paths with the computational efficiency of rapid iteration across multiple operating conditions, failing to meet real-time feedback requirements in engineering. Secondly, existing data-driven methods often employ traditional fully connected network structures, lacking effective explicit modeling mechanisms for the hierarchical correlations and importance of circuit input parameters. This prevents dynamic focusing on key parameters influencing EMI, resulting in insufficient feature extraction capabilities in multi-parameter strongly coupled scenarios, weak generalization ability, limited prediction accuracy, and a lack of model interpretability. Summary of the Invention

[0003] The purpose of this application is to provide a method, device, medium, and product for predicting electromagnetic interference in synchronous buck circuits, in order to solve the problems of weak generalization ability, limited prediction accuracy, and lack of model interpretability.

[0004] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a method for predicting electromagnetic interference in a synchronous buck circuit, including: The key circuit parameters of the synchronous buck circuit and the corresponding measured electromagnetic interference spectrum are obtained, and the key circuit parameters are preprocessed by data normalization to map the key circuit parameters of different dimensions to the [0,1] interval. A HA-DNN prediction model is constructed; the HA-DNN prediction model introduces an attention mechanism in a fully connected structure, and the attention mechanism is embedded between the input layer and the hidden layer; The preprocessed key circuit parameters are input into the HA-DNN prediction model for forward propagation. The influence weight of each preprocessed key circuit parameter on the electromagnetic interference spectrum features is dynamically calculated through the attention mechanism to determine the input representation after feature enhancement. This representation is then input into the hidden layer for nonlinear mapping to generate the predicted electromagnetic interference spectrum. The nonlinear mapping is used to characterize the coupling transfer function of the key circuit parameters to frequency domain interference during switching transients. Based on the error between the predicted electromagnetic interference spectrum and the measured electromagnetic interference spectrum, a loss function is constructed. The parameters of the backbone network of the HA-DNN prediction model and the trainable parameters of the attention mechanism are updated synchronously through the backpropagation algorithm. The model is iteratively trained until the convergence condition is met or the preset maximum number of iterations is reached, and the trained HA-DNN prediction model is obtained. The key circuit parameters of the synchronous buck circuit to be predicted are input into the trained HA-DNN prediction model, and the predicted values ​​of differential mode and common mode conducted electromagnetic interference spectrum of the synchronous buck circuit in the full frequency band of 150kHz-30MHz are output.

[0005] Secondly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the electromagnetic interference prediction method for the synchronous buck circuit described above.

[0006] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the electromagnetic interference prediction method for the synchronous buck circuit described above.

[0007] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the electromagnetic interference prediction method for the synchronous buck circuit described above.

[0008] According to the specific embodiments provided in this application, this application has the following technical effects: This application constructs a Hierarchical Attention Deep Neural Network (HA-DNN) prediction model by introducing an attention mechanism into a fully connected structure. The attention mechanism is embedded between the input layer and the hidden layer. During the forward propagation of the model, the influence weights of each preprocessed key circuit parameter on the electromagnetic interference spectrum characteristics are dynamically calculated to determine the input representation after feature enhancement. This allows the model to adaptively focus on the key parameters affecting electromagnetic interference, explicitly capturing the hierarchical correlation and high-order nonlinear coupling relationships between multiple parameters. This effectively overcomes the shortcomings of existing data-driven methods in feature extraction capabilities under strongly coupled multi-parameter scenarios, improving the prediction accuracy and model generalization ability of differential-mode and common-mode conducted electromagnetic interference spectra across the 150kHz-30MHz full-band spectrum. At the same time, by dynamically allocating feature weights through the attention mechanism and synchronously optimizing them through backpropagation, the black-box limitation of traditional neural networks is broken, providing physically interpretable quantitative guidance for the sensitivity localization and optimization design of circuit parameters. This effectively solves the technical problem that existing prediction methods struggle to balance the accuracy of high-frequency parasitic coupling representation, computational efficiency, and model interpretability. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A schematic diagram of a synchronous buck circuit provided in one embodiment of this application; Figure 2 A flowchart illustrating an electromagnetic interference prediction method for a synchronous buck circuit provided in an embodiment of this application; Figure 3 A schematic diagram illustrating the training process of an HA-DNN prediction model provided in an embodiment of this application; Figure 4 This is a schematic diagram comparing the performance of different models according to an embodiment of this application; Figure 5 A schematic diagram of a simulation platform constructed according to an embodiment of this application; Figure 6 This is a schematic diagram showing the comparison between predicted and actual values ​​provided in an embodiment of this application; Figure 7 This is a schematic diagram of the mean square error curve of the HA-DNN prediction model provided in an embodiment of this application. Detailed Implementation

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0012] To make the objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0013] like Figure 1 As shown, this application embodiment provides a synchronous buck circuit, including: The DC power supply module includes a DC input power supply Vcc, which is used to provide a DC operating voltage for the synchronous buck circuit.

[0014] The LISN filter module, with its input terminal connected to the output terminal of the DC input power supply Vcc, includes LISN inductors Llisn1 and Llisn2, LISN filter capacitors C1 and C2, LISN filter resistors R1 and R2, and LISN load resistor R3. It is used to provide a standard impedance environment and filter out external power grid interference during conducted electromagnetic interference testing.

[0015] The input filtering module, with its input terminal connected to the output terminal of the LISN filtering module, includes an input filtering capacitor Cin, an equivalent series resistance Resr1, and an equivalent series inductance Lesl1, which are used for high-frequency decoupling and energy buffering of the input DC voltage.

[0016] The PCB trace module, connected in series between the input filter module and the power switch module, includes a trace resistor Rline and a trace inductance Lline, which are used to characterize the distributed impedance in the high-frequency current transmission path.

[0017] The power switching module includes an upper-side switch and a lower-side switch connected in series. The drain of the upper-side switch is connected to the output terminal of the PCB trace module. The source of the upper-side switch is connected to the drain of the lower-side switch to form a switching node. The source of the lower-side switch is grounded. Both the upper-side switch and the lower-side switch are gallium nitride power switches (EPC2045), used to achieve synchronous step-down conversion of DC voltage through high-frequency alternating on and off.

[0018] The gate drive module is electrically connected to the gates of the upper and lower switching transistors, respectively. It includes upper transistor drive resistors Rugh and Rugl, upper transistor drive parasitic inductors Lugh and Lugl, upper transistor drive bootstrap capacitor Cboot, lower transistor drive resistors Rlgh and Rlgl, and lower transistor drive parasitic inductors Llgh and Llgl. It is used to receive control signals and drive the upper and lower switching transistors to switch synchronously.

[0019] The output filtering and load module, with its input connected to the switching node, includes an output filter capacitor Co, a load inductor Lload, and a load resistor Rload. It is used to smooth and filter the pulse voltage output by the switching node and supply power to the load.

[0020] The parasitic capacitance network to ground includes the load-to-ground parasitic capacitance Clg, the drain-to-ground parasitic capacitance Cdg, and the source-to-ground parasitic capacitance Csg, which are connected in parallel across the load resistor Rload, between the drain of the upper-side switching transistor and ground, and between the source of the upper-side switching transistor and ground, respectively, and are used to characterize the ground coupling path under high-frequency switching transients.

[0021] The synchronous buck circuit generates differential-mode and common-mode conducted electromagnetic interference across the entire frequency band of 150kHz-30MHz during the switching transient process due to the nonlinear coupling of the parasitic parameters inside the PCB trace module, the power switch module, and the parasitic capacitance network to ground.

[0022] Figure 1 This paper demonstrates a constructed synchronous buck converter circuit, including its equivalent structure with key high-frequency components (including parasitic parameters), for the characteristic analysis and modeling of wide-bandgap devices in synchronous buck scenarios. This circuit is a synchronous buck converter topology that achieves voltage reduction from the input voltage Vcc to the load Rload, controlling energy transfer through high-frequency switching. This application considers the influence of topology parameters under different operating conditions, parasitic parameters of the switching transistors, key parameters of the drive circuit, parasitic parameters of PCB traces, and capacitance to ground on conducted EMI characteristics, providing a general research framework for EMI optimization design of wide-bandgap devices in voltage converter applications.

[0023] Figure 1 The parameter symbols in the synchronous buck circuit are defined as shown in Table 1.

[0024] Table 1. Parameter Symbol Definitions in Synchronous Buck Circuit

[0025] In an exemplary embodiment, the system further includes a parasitic inductance network; the parasitic inductance network includes an upper transistor drain parasitic inductance Ld and an upper transistor source parasitic inductance Ls; the upper transistor drain parasitic inductance is connected in series in the drain circuit of the upper transistor switching transistor, and the upper transistor source parasitic inductance Ls is connected in series between the switching node and the source of the lower transistor switching transistor, for characterizing the high-frequency distributed inductance effect in the switching circuit.

[0026] In an exemplary embodiment, the gate drive module further includes: a DC bus voltage source Vdc, a bus support capacitor Cdc, and a loop parasitic resistor R4 and a loop parasitic resistor R5.

[0027] The DC bus voltage source Vdc and the bus support capacitor Cdc are connected in parallel to provide a drive bias voltage for the gate drive module.

[0028] The parasitic resistances R4 and R5 are connected in series in the drive circuit to characterize the ohmic loss and high-frequency attenuation characteristics of the drive signal transmission path.

[0029] The output filtering and load module also includes: the equivalent series resistance Resr2 of the output filtering circuit and the equivalent series inductance Lesl2 of the output filtering circuit.

[0030] The equivalent series resistance Resr2 and the equivalent series resistance Lesl2 of the output filter circuit are connected in series with the output filter capacitor Co to characterize the non-ideal impedance characteristics of the output filter branch under high-frequency operating conditions.

[0031] In an exemplary embodiment, the common-mode voltage generated at the switching node is (VRh+VRL) / 2, and the differential-mode voltage is (VRh-VRL) / 2; wherein VRh and VRL are the voltages to ground of the live wire output terminal and the neutral wire output terminal of the LISN of the LISN filter module, respectively.

[0032] The differential-mode voltage propagates along the loop formed by the PCB trace module and the input filter module, while the common-mode voltage is coupled to ground via the parasitic capacitance network to ground.

[0033] The synchronous buck circuit is configured to operate at a steady-state frequency of 1MHz and a duty cycle of 45%. Under this steady-state operating condition, the switching transient of the power switch module resonates with the parasitic parameters to generate spectral data for electromagnetic interference feature mapping.

[0034] The key parameters of the synchronous buck converter include the input voltage Vcc, load resistance Rload, upper transistor drive resistors Rugh and Rugl, upper transistor trace resistance Rline, upper transistor drain parasitic inductance Ld, upper transistor source parasitic inductance Ls, and capacitance to ground Clg. The changes in the values ​​of these key parameters directly modulate the voltage change rate and loop impedance characteristics of the switching node, thereby determining the spectral amplitude distribution and harmonic intensity of the conducted electromagnetic interference.

[0035] In one exemplary embodiment, the synchronous buck circuit is also electrically or communicatively connected to an electromagnetic interference prediction system.

[0036] The electromagnetic interference prediction system includes a data acquisition unit, an HA-DNN prediction model unit, and an output unit. The data acquisition unit is used to acquire the key parameters of the key circuit and the corresponding measured electromagnetic interference spectrum. The HA-DNN prediction model unit is used to establish an end-to-end mapping relationship between the key parameters of the key circuit and the differential-mode and common-mode conducted electromagnetic interference spectrum of the 150kHz-30MHz full-band frequency band based on an attention mechanism. The output unit is used to output the predicted spectrum to perform EMI compliance prediction, parameter sensitivity analysis, or design space screening operations.

[0037] This application constructs a set of parameters that only require the input voltage Vcc, load resistance Rload, upper MOSFET drive resistors Ruhg and Rugl, upper MOSFET trace resistance Rline, upper MOSFET trace inductance + drain point inductance Ld, source parasitic inductance Ls, and capacitance to ground Clg to directly obtain the spectrum prediction value of conducted EMI of synchronous buck circuit.

[0038] For the prediction of conducted electromagnetic interference (EMI) in voltage converter circuits, existing technologies mainly suffer from three bottlenecks: First, analytical modeling methods require significant simplification of circuit structure and device characteristics, making it difficult to accurately characterize high-frequency parasitic parameters and nonlinear coupling effects, thus limiting prediction accuracy. Second, while numerical simulation methods can achieve relatively accurate predictions through high-precision time-domain models and electromagnetic field simulations, the computational load is enormous, and each simulation is time-consuming, making it difficult to meet the needs of rapid design iteration in multi-parameter, multi-condition optimization scenarios. Third, although data-driven methods demonstrate modeling flexibility in nonlinear scenarios, existing models lack effective modeling mechanisms for the hierarchical correlation and importance between input parameters, and are still insufficient in global spectrum mapping and high-dimensional coupling modeling. Therefore, this application addresses the above bottlenecks in the prediction of conducted EMI in existing synchronous buck converter circuits by proposing an integrated solution of "attention mechanism-optimized deep neural network modeling + full-band feature learning": (1) Based on the key circuit parameters and measured EMI data, an HA-DNN prediction model was constructed. The nonlinear coupling relationship of key parasitic parameters was automatically captured by dynamic feature weight allocation. Differential mode / common mode voltage spectrum prediction was achieved in the full frequency band of 150kHz-30MHz. The measured average prediction error was less than 3%.

[0039] (2) Using the HA-DNN prediction model as the core, an end-to-end direct mapping from key circuit parameters to the EMI spectrum is established, avoiding the repetitive calculation process of solving differential equations and iterative convergence at each time step in traditional time-domain simulation. After the model completes offline training, forward propagation only requires a few layers of matrix multiplication and nonlinear activation function operations, with a computational complexity of only O(n·h), where n is the dimension of the input parameters and h is the width of the hidden layer. Therefore, fast prediction under complex operating conditions is achieved, with a single prediction time of only milliseconds, which is nearly 10,000 times more efficient than traditional time-domain simulation, providing real-time feedback for design optimization.

[0040] (3) Through attention weight visualization analysis, the influence of different operating parameters and PCB layout parameters on EMI characteristics is revealed. Compared with the "black box" characteristics of traditional DNN, attention weight visualization analysis provides circuit designers with interpretable parameter optimization priority guidance. It can directly locate the key parameters that contribute the most to EMI, thereby optimizing the circuit design in a targeted manner, providing quantitative guidance for circuit design, and shortening the electromagnetic compatibility (EMC) rectification cycle.

[0041] For the aforementioned synchronous buck circuit, such as Figure 2 As shown, this application also provides an electromagnetic interference prediction method for a synchronous buck circuit, including: S1: Obtain the key circuit parameters of the synchronous buck circuit and the corresponding measured electromagnetic interference spectrum, and perform data normalization preprocessing on the key circuit parameters to map the key circuit parameters of different dimensions to the [0,1] interval.

[0042] S2: Construct an HA-DNN prediction model; the HA-DNN prediction model introduces an attention mechanism in a fully connected structure, which is embedded between the input layer and the hidden layer.

[0043] S3: Input the preprocessed key circuit parameters into the HA-DNN prediction model for forward propagation. Calculate the influence weights of each preprocessed key circuit parameter on the electromagnetic interference spectrum characteristics using the attention mechanism. Determine the input representation after feature enhancement and input it into the hidden layer for nonlinear mapping to generate the predicted electromagnetic interference spectrum. The nonlinear mapping is used to characterize the coupling transfer function of the key circuit parameters to frequency domain interference during switching transients.

[0044] S4: Construct a loss function based on the error between the predicted electromagnetic interference spectrum and the measured electromagnetic interference spectrum, and synchronously update the backbone network parameters of the HA-DNN prediction model and the trainable parameters of the attention mechanism through the backpropagation algorithm. Iterate the training until the convergence condition is met to obtain the trained HA-DNN prediction model.

[0045] S5: Input the key circuit parameters of the synchronous buck circuit to be predicted into the trained HA-DNN prediction model, and output the predicted values ​​of differential mode and common mode conducted electromagnetic interference spectrum of the synchronous buck circuit in the full frequency band of 150kHz-30MHz.

[0046] This application proposes an attention-based optimized neural network to construct a high-precision mapping model between key circuit parameters (including Vcc, Rload, parasitic parameters Ld and Ls of the switching transistor, key parameters Rugh and Rugl of the drive circuit, parasitic parameter R4 of the PCB trace, and capacitance to ground Clg) and the conducted EMI spectrum, namely the HA-DNN prediction model. HA-DNN introduces an attention mechanism into the fully connected structure of traditional DNNs, enabling dynamic focusing on key parameters affecting EMI in synchronous buck circuits. It enhances the characteristic representation of high-frequency differential / common-mode voltage spectra, effectively solving the problems of high difficulty in EMI modeling and poor generalization ability under strong coupling of multiple parameters. This provides technical support for accurate EMI prediction and parameter optimization in high-frequency and integrated applications.

[0047] Table 2 compares the differences between Deep Neural Networks (DNNs) and Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) in terms of structural characteristics, feature interaction methods, input adaptability, parameter sensitivity, parallel capabilities, theoretical guarantees, and computational efficiency. Overall, traditional DNNs have advantages in fully connected architecture, nonlinear modeling, parallel computing, and theoretical completeness, making them suitable for modeling tasks of multi-parameter coupled systems. HA-DNNs further enhance feature selection capabilities and model interpretability, making them more suitable for engineering prediction scenarios such as electromagnetic interference, which involve large differences in input features and complex coupling relationships. Compared to traditional DNNs, HA-DNNs can reduce the dependence on the amount of training data while maintaining the same accuracy, exhibiting stronger robustness and interpretability in handling complex feature prediction tasks with multiple coupled factors.

[0048] Table 2 Comparison of DNN with CNN and RNN

[0049] In an exemplary embodiment, the key circuit parameters include: input voltage Vcc, load resistance Rload, upper transistor drive resistors Ruhg and Rugl, upper transistor trace resistance Rline, upper transistor drain parasitic inductance Ld, upper transistor source parasitic inductance Ls, and capacitance to ground Clg.

[0050] In an exemplary embodiment, S1 involves data normalization preprocessing of the key circuit parameters, specifically including: S11: Normalize the key parameters of the circuit and map them to the [0,1] interval.

[0051] S12: Perform outlier handling on the normalized data to generate preprocessed key circuit parameters.

[0052] In an exemplary embodiment, S3 dynamically calculates the influence weights of each preprocessed key circuit parameter on the electromagnetic interference spectrum characteristics through the attention mechanism, generating a predicted electromagnetic interference spectrum, specifically including: S31: Input the preprocessed key circuit parameters into the attention layer of the HA-DNN prediction model, and dynamically calculate the importance score of each preprocessed key circuit parameter through the learnable parameter matrix.

[0053] S32: Using the Sigmoid activation function, the importance score is compressed to the (0,1) interval, and competitive weight allocation is achieved through the Softmax function.

[0054] S33: For a network containing multiple attention layers, the importance of the final parameters is calculated by accumulating the weights assigned to each layer and then normalized to determine the interpretable weights and generate the predicted electromagnetic interference spectrum; the interpretable weights are the influence weights of each preprocessed circuit key parameter on the electromagnetic interference spectrum characteristics.

[0055] It should be noted that the optimization process of the learnable parameter matrix relies on backpropagation during the training phase: the predicted electromagnetic interference spectrum and the measured electromagnetic interference spectrum generated in the current round are input into the loss function, the gradient of the loss function with respect to the learnable parameter matrix is ​​calculated through the backpropagation algorithm, and the learnable parameter matrix is ​​updated along the negative gradient direction, so that the attention weight gradually converges to the optimal solution that can characterize the contribution of each circuit key parameter to the actual impact of the electromagnetic interference spectrum during the iteration process.

[0056] During the inference phase, the key parameters of the circuit to be predicted only need to be input into the trained HA-DNN prediction model. The fixed attention weights are directly generated from the trained learnable parameter matrix, and the predicted value can be output without relying on the measured electromagnetic interference spectrum.

[0057] The HA-DNN prediction model has the following advantages in the field of electromagnetic interference prediction: 1. Simple structure, easy to deploy, suitable for engineering applications and rapid prototyping; 2. While retaining the powerful nonlinear modeling capabilities of DNN, the model improves its accuracy in high-frequency EMI prediction by achieving adaptive focusing on key parameters through an attention mechanism; It can explicitly capture high-order coupling relationships between key circuit parameters, and is especially suitable for EMI modeling under the combined action of multiple physical quantities; It has high computational efficiency and strong parallelism, making it suitable for large-scale parameter optimization and automated design processes. Attention weights are highly interpretable and can provide circuit designers with key parameter impact analysis to assist in EMI optimization decisions.

[0058] These advantages make HA-DNN uniquely applicable in the field of electromagnetic interference prediction, particularly suitable for handling the complex EMI characteristic prediction problem caused by multi-parameter coupling in synchronous buck circuits. Therefore, this application develops an electromagnetic interference prediction method for synchronous buck circuits based on the HA-DNN algorithm. This method establishes a mapping relationship between key circuit parameters and conducted EMI spectrum, achieving accurate prediction of differential-mode / common-mode voltage across the entire frequency band of 150kHz-30MHz. After obtaining the prediction results, operations such as EMI compliance prediction, parameter sensitivity analysis to locate key influencing parameters, and rapid exploration of the design space to select the optimal parameter combination can be performed. The prediction results of this application can be practically applied to the design stages of automotive power supplies, high-frequency modular power supplies, and aerospace and communication power supplies. They can also be packaged into an automated EMI optimization tool and embedded in the EDA design flow, providing a reliable basis for circuit optimization design. The training process of the HA-DNN prediction model is as follows: Figure 3 As shown.

[0059] The following is a detailed process based on the HA-DNN prediction model: (1) Data input Input data includes key circuit parameters And the frequency domain voltage characteristic matrix Y, which represents the differential-mode and common-mode frequency domain voltage data of the synchronous buck circuit at its operating frequency and higher harmonic frequencies from 0 to 50MHz, expressed as: .

[0060] (2) Data normalization preprocessing The key circuit parameters are normalized and mapped to the [0,1] interval.

[0061] Outlier handling is performed on key circuit parameters during preprocessing, based on... Outlier data should be removed as a rule.

[0062] (3) HA-DNN prediction model Basic DNN structure: consists of an input layer, hidden layers, and an output layer.

[0063] Attention layer embedding: An attention weight layer is inserted between the input layer and the hidden layer. Through a learnable weight matrix, the influence weights of key circuit parameters on EMI spectrum characteristics are dynamically allocated.

[0064] (4) Forward propagation and dynamic adjustment of attention After preprocessing, the key circuit parameters are input into the attention layer to begin dynamic learning of feature weights. Based on a multi-layer attention mechanism, each layer accurately calculates the contribution weight of each input parameter to EMI spectrum prediction through weighted summation connections. This approach both strengthens the influence of key parameters and avoids the risk of gradient vanishing, ensuring the stability of deep network training.

[0065] When the input features pass through the attention layer, the importance scores of each key circuit parameter are dynamically calculated using the learnable parameter matrix. in: Indicates the first One parameter; Y-domain voltage characteristic matrix; Let X be the weight matrix of the parameter X, which is a trainable parameter; and They represent the first The weight matrix and bias of the conducted electromagnetic interference matrix Y corresponding to each parameter X; This represents the sigmoid activation function, which compresses the score to the (0, 1) interval.

[0066] Competitive weight allocation is achieved using the Softmax function: For a network with l attention layers, the final parameter importance is calculated by multiplying the weights of each layer: in: This represents the score of the k-th parameter in the l-th layer attention mechanism.

[0067] The interpretable weights are obtained after normalization: The trainable parameters in the above attention mechanism (including) , , (etc.) Automatically learn and optimize through the loss function and backpropagation algorithm described in the next step.

[0068] Specifically, the gradient of the loss function with respect to the prediction error is backpropagated layer by layer to the attention layer, guiding the attention parameters to be updated in the direction of reducing the prediction error, so that the model can adaptively focus on the key parameters that affect EMI.

[0069] (5) Loss function and backpropagation A loss function is constructed based on the error between the predicted EMI spectrum and the measured EMI spectrum. This loss function is used not only to update the parameters of the DNN backbone network, but also to update all trainable parameters of the attention mechanism.

[0070] Based on the gradient descent algorithm, the gradient of the loss function with respect to the DNN network parameters and attention layer weights is calculated.

[0071] The parameters are updated along the negative gradient direction, and the formula simplifies to: in: Parameters to be updated; The learning rate; For loss function pairs The gradient.

[0072] (6) Convergence judgment and model output Based on the loss function and gradient update rule in (5), this step optimizes the model parameters through iterative training. In each iteration, the forward propagation, loss calculation, backpropagation, and parameter update operations are repeatedly executed to continuously iterate the training and monitor the loss value on the validation set.

[0073] When the loss value stops decreasing after several consecutive iterations, the model is deemed to have reached the convergence condition, and training is stopped.

[0074] The HA-DNN algorithm, with corresponding programs written in MATLAB and verified in conjunction with the equivalent model of the synchronous buck circuit mentioned above, can achieve the following core functions: accurately predict the conducted EMI spectrum in the 150kHz-30MHz frequency band by inputting key circuit parameters, with the prediction error controlled within 3%; at the same time, by using the feature importance matrix output by the attention weight layer, the influence of each input parameter on the key EMI frequency points is quantified, providing flexible and practical design guidance for engineering applications.

[0075] (7) Model performance verification To verify the effectiveness of HA-DNN, this application conducts comparative experiments with traditional DNN, CNN, Long Short-Term Memory (LSTM) networks, and Gradient Boosting Regression Tree (GBRT) on the aforementioned dataset. Figure 4 As shown, the root mean square error (RMSE) of the HA-DNN prediction model is only 0.426, which is significantly better than the comparison model. This result shows that the attention mechanism can effectively focus on the key parameters affecting high-frequency EMI and improve the prediction ability of the model in complex coupling scenarios.

[0076] The prediction method provided in this application is a prediction method for EMI of synchronous buck circuits based on the HA-DNN prediction model. It is applicable to EMC evaluation and forward design optimization of high-frequency voltage conversion circuits and power electronic conversion systems.

[0077] To verify the effectiveness of the proposed model, this application constructs a dual-platform verification system combining simulation and hardware to comprehensively evaluate the proposed HA-DNN prediction model from two dimensions.

[0078] The simulation platform has a synchronous buck circuit with the same topology as the hardware platform, such as... Figure 5 As shown. The simulation experiment uses the EPC2045 gallium nitride (GaN) power device as the switch. Its main advantages include: ultra-fast switching speed, which can fully excite high-frequency EMI characteristics; low parasitic parameters, making the circuit resonance characteristics highly sensitive to layout parameters, which is beneficial for examining EMI behavior under multi-parameter coupling; and support for MHz-level switching frequencies, providing an ideal platform for studying the EMI characteristics of wide-bandgap devices under high-frequency operating conditions. The simulation parameters are set as follows: the switching frequency is set to 1MHz, the duty cycle is adjusted to 45%, and the transient simulation command is used to set the running time covering the circuit startup and steady-state phases to accurately simulate the actual operating conditions of the synchronous buck circuit.

[0079] Then, a set of key circuit parameters were randomly selected. , that is, Vcc=24V, Rload=10Ω, Rugh=5Ω, Rugl=5Ω, R4=20×10-3Ω, Ld=0.43×10 -9 H, Ls = 0.075 × 10 -9 H, Clg = 1 × 10 -12 F is used to predict conducted EMI. The comparison between the model proposed in this application and the experimental results is as follows: Figure 6 As shown. This application selects the average relative error between the predicted EMI and the actual EMI as the error criterion, with a mean square error (RMSE) of 0.47. Figure 7 As shown, the model proposed in this application can accurately predict conducted EMI under synchronous buck circuits.

[0080] In summary, this application has the following advantages: The model boasts dual advantages in accuracy and efficiency: the model's conducted EMI prediction error is less than 3% across the entire frequency band, and once the model is trained, a single EMI prediction calculation takes only milliseconds.

[0081] Interpretability of parameter effects: The contribution of key circuit parameters to conducted EMI is quantified by using an attention weight matrix, providing priority guidance for the optimization of key circuit parameters.

[0082] Strong engineering adaptability: The model architecture is compatible with various half-bridge topologies such as synchronous buck, boost, buck-boost, etc., and can be directly migrated to EMI prediction scenarios with different wide bandgap devices.

[0083] High technical scalability: In the future, it can be extended to introduce dependency modeling for complex operating conditions such as temperature, and combined with iterative optimization of attention mechanism to continuously improve the EMI prediction accuracy of wide bandgap devices in extreme scenarios.

[0084] In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments. The computer device can be a server or a terminal. The computer device includes a processor, memory, an input / output interface (I / O), and a communication interface. The processor, memory, and I / O are connected via a system bus, and the communication interface is connected to the system bus via the I / O interface. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device stores data to be processed. The I / O interface of the computer device is used for exchanging information between the processor and external devices. The communication interface of the computer device is used for communicating with an external terminal via a network connection. When the computer program is executed by the processor, it implements the above-described methods.

[0085] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0086] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0087] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0088] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by hardware related to computer program instructions. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0089] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0090] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0091] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting electromagnetic interference in a synchronous buck converter, characterized in that, include: The key circuit parameters of the synchronous buck circuit and the corresponding measured electromagnetic interference spectrum are obtained, and the key circuit parameters are preprocessed by data normalization to map the key circuit parameters of different dimensions to the [0,1] interval. A HA-DNN prediction model is constructed; the HA-DNN prediction model introduces an attention mechanism in a fully connected structure, and the attention mechanism is embedded between the input layer and the hidden layer; The preprocessed key circuit parameters are input into the HA-DNN prediction model for forward propagation. The influence weight of each preprocessed key circuit parameter on the electromagnetic interference spectrum features is dynamically calculated through the attention mechanism to determine the input representation after feature enhancement. This representation is then input into the hidden layer for nonlinear mapping to generate the predicted electromagnetic interference spectrum. The nonlinear mapping is used to characterize the coupling transfer function of the key circuit parameters to frequency domain interference during switching transients. Based on the error between the predicted electromagnetic interference spectrum and the measured electromagnetic interference spectrum, a loss function is constructed. The parameters of the backbone network of the HA-DNN prediction model and the trainable parameters of the attention mechanism are updated synchronously through the backpropagation algorithm. The model is iteratively trained until the convergence condition is met or the preset maximum number of iterations is reached, and the trained HA-DNN prediction model is obtained. The key circuit parameters of the synchronous buck circuit to be predicted are input into the trained HA-DNN prediction model, and the predicted values ​​of differential mode and common mode conducted electromagnetic interference spectrum of the synchronous buck circuit in the full frequency band of 150kHz-30MHz are output.

2. The electromagnetic interference prediction method for the synchronous buck circuit according to claim 1, characterized in that, The key parameters of the circuit include: input voltage Vcc, load resistance Rload, upper transistor drive resistors Ruhg and Rugl, upper transistor trace resistance Rline, upper transistor drain parasitic inductance Ld, upper transistor source parasitic inductance Ls, and capacitance to ground Clg.

3. The electromagnetic interference prediction method for the synchronous buck circuit according to claim 1, characterized in that, The influence weights of each preprocessed key circuit parameter on the electromagnetic interference spectrum characteristics are dynamically calculated using the attention mechanism to generate a predicted electromagnetic interference spectrum, specifically including: The preprocessed key circuit parameters are input into the attention layer of the HA-DNN prediction model, and the importance score of each preprocessed key circuit parameter is dynamically calculated through the learnable parameter matrix. The importance score is compressed to the (0,1) interval using the Sigmoid activation function, and competitive weight allocation is achieved through the Softmax function. For a network containing multiple attention layers, the importance of the final parameters is calculated by accumulating the weights assigned to each layer and then normalized to determine the interpretable weights, thereby generating the predicted electromagnetic interference spectrum. The interpretable weights are the influence weights of each preprocessed key circuit parameter on the electromagnetic interference spectrum characteristics.

4. The electromagnetic interference prediction method for the synchronous buck circuit according to claim 1, characterized in that, The loss function is used to calculate the error between the predicted electromagnetic interference spectrum and the measured electromagnetic interference spectrum. The backpropagation algorithm updates parameters along the negative gradient direction. In each iteration, it repeatedly performs forward propagation, loss calculation, backpropagation, and parameter update operations, and monitors the validation set loss value. When the validation set loss value no longer decreases for several consecutive iterations, the HA-DNN prediction model is determined to have reached the convergence condition.

5. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the electromagnetic interference prediction method for the synchronous buck circuit according to any one of claims 1-4.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the electromagnetic interference prediction method for the synchronous buck circuit according to any one of claims 1-4.

7. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the electromagnetic interference prediction method for the synchronous buck circuit according to any one of claims 1-4.

Images