Soft switching of high frequency switching transformer and distributed control method and system thereof

Abstract

The application relates to the technical field of power electronics, and discloses a soft switch of a high-frequency switching transformer and a distributed control method and system thereof. The method comprises the following steps: realizing node accurate synchronization by constructing a ring topology network, combining multi-physical quantity monitoring and parameter collaborative optimization, dynamically adjusting resonance network parameters and driving time sequence, maintaining a zero-voltage switching state, and adopting a composite optimization model to inhibit electromagnetic interference. The system comprises a network initialization module, a multi-physical quantity monitoring module, a space-time synchronization control module, a resonance parameter adjustment module, an electromagnetic optimization decision module, a dynamic adjustment module and a parameter evolution module. The application improves the stability of the soft switch through distributed network and space-time synchronization, reduces loss by adopting a dynamic optimization algorithm, enhances compatibility by constructing an electromagnetic interference optimization model, optimizes dynamic response through multi-parameter fusion and weight distribution, improves reliability through high-precision protocol and collaborative evolution, and realizes multi-dimensional intelligent collaborative control through a tensor product framework.

Description

A soft-switching method and system for a high-frequency switching transformer and its distributed control method. Technical Field

[0001] This invention relates to the field of power electronics technology, specifically to a soft-switching method and system for a high-frequency switching transformer and its distributed control. Background Technology

[0002] As power electronic equipment develops towards higher frequencies and higher power densities, high-frequency switching transformers, as core components for energy conversion, directly impact system efficiency and reliability. Traditional hard-switching technology suffers from high switching losses and severe electromagnetic interference, making it difficult to meet the demands of modern high-efficiency power conversion. Existing soft-switching solutions mostly employ centralized control architectures, which suffer from insufficient synchronization accuracy and lag in parameter adjustment when dealing with rapid load fluctuations and multi-module parallel scenarios.

[0003] In current technologies, resonant network parameters are typically designed with fixed values, which cannot adapt to changes in operating conditions over a wide load range, making it easy to disrupt zero-voltage switching conditions. Time synchronization mechanisms often rely on a single master clock, and the accumulation of phase deviations between nodes can easily lead to drive signal mismatch, causing uneven stress on the switching transistors. Electromagnetic interference suppression methods are often limited to passive filtering, lacking active optimization and control of interference sources, making it difficult to meet stringent electromagnetic compatibility standards.

[0004] Furthermore, existing control methods often employ fixed-weight strategies when dealing with multi-objective optimization problems, failing to dynamically balance the conflicting demands for efficiency improvement and electromagnetic suppression. The parameter adjustment process involves independent control dimensions lacking a collaborative optimization mechanism, thus limiting the overall system performance. Especially under extreme conditions such as high temperature and high load, existing solutions are prone to thermal runaway and oscillatory instability, hindering the long-term reliable operation of high-frequency switching transformers.

[0005] Therefore, this invention proposes a soft-switching method and system for high-frequency switching transformers and its distributed control method to address the shortcomings of existing technologies. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a soft-switching method and system for high-frequency switching transformers, which solves the problems of instability in the soft-switching state, insufficient electromagnetic interference suppression, and low accuracy of multi-node collaborative control in high-frequency switching transformers under complex operating conditions.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a soft-switching method for a high-frequency switching transformer and its distributed control method, comprising the following steps:

[0008] S1. Construct a distributed control network with a ring topology, load an improved precise time protocol and initialize the equivalent inductance and capacitance parameters of the resonant cavity, and establish a clock deviation compensation matrix between nodes.

[0009] S2. Real-time acquisition of drain-source voltage, drain current and load current change rate of each node through voltage and current sensors, synchronous monitoring of junction temperature data of power devices, and calculation of core flux change based on voltage integration method.

[0010] S3. Based on the load current change rate and magnetic flux calculation results, solve the second-order partial derivative of the phase angle between nodes to generate a synchronous drive pulse signal with timestamp calibration.

[0011] S4. Based on the timing characteristics of the synchronous drive pulse signal and the drain current change rate, the parameters of the resonant network are iteratively updated using the quasi-Newton method to maintain the zero-voltage switching state under different load conditions.

[0012] S5. Establish a control model with the square of the drain-source voltage change rate and the electromagnetic interference spectrum norm as optimization objectives, and obtain the optimal switching trajectory command by solving the Hamiltonian equation.

[0013] S6. Based on the real-time monitored junction temperature data and load current change rate, dynamically allocate the weight coefficients for efficiency optimization and electromagnetic suppression, suppress abnormal propagation through the error energy attenuation function, and perform tensor product co-evolution of resonance parameters and synchronization parameters.

[0014] Preferably, step S1 includes: constructing a distributed control network with a ring topology, loading an improved precise time protocol that includes a phase angle differential compensation mechanism, initializing the set of equivalent inductance and capacitance parameters of the resonant cavity, and establishing a clock offset compensation matrix based on the second-order mixed partial derivatives of the phase angle between nodes, wherein the construction process of the compensation matrix includes cross-differentiation operation on the phase change rate of adjacent nodes.

[0015] Preferably, step S2 includes: acquiring the instantaneous values ​​of drain-source voltage at each node using a voltage sensor, obtaining the drain current and load current change rate using a current sensor, synchronously monitoring the real-time junction temperature data of the power device, and calculating the core magnetic flux change based on the voltage integration method, wherein the calculation process of the voltage integration method includes a compensation term for the equivalent resistance of the core, and the calculation result of the magnetic flux change has a nonlinear integral relationship with the drain-source voltage sampling value.

[0016] Preferably, step S2 further includes: establishing a dynamic coupling correction model of junction temperature monitoring data and magnetic flux estimation, and using a sliding time window algorithm to perform time-domain smoothing of the load current change rate. The smoothing process includes a weighted average calculation of the current sample values ​​and a mutation suppression mechanism.

[0017] Preferably, step S3 includes: calculating the second-order mixed partial derivative of the phase angle between nodes based on the load current change rate and magnetic flux calculation results, wherein the calculation process of the partial derivative introduces a dynamic weighting factor of the load current change rate, generating a synchronous drive pulse signal with timestamp calibration, wherein the timestamp calibration includes phase lead correction based on a feedforward compensation mechanism, and the generation process of the drive pulse signal and the resonant network parameter update form a closed-loop feedback control.

[0018] Preferably, step S4 includes: based on the timing characteristics of the synchronous drive pulse signal and the drain current change rate, iteratively updating the resonant network parameters using a quasi-Newton method. The update process comprehensively calculates the efficiency gradient, electromagnetic interference gradient, and time synchronization gradient, corrects the parameters through the inverse operation of the Jacobian matrix, and dynamically adjusts the resonant cavity quality factor to maintain a zero-voltage switching state. The adjustment of the quality factor and the load current change rate form a closed-loop feedback control.

[0019] Preferably, step S5 includes: establishing a control model with the square of the drain-source voltage change rate and the electromagnetic interference spectrum norm as optimization objectives; constructing a Hamiltonian equation containing the second-order differentiability constraint of the switching trajectory; using the adjoint variable method to process the boundary conditions and solve for the optimal switching trajectory command; wherein the calculation process of the adjoint variable is data-coupled with the update of the resonant network parameters; and the generation of the optimal switching trajectory command is synchronously fed back to the timing adjustment module of the synchronous drive pulse.

[0020] Preferably, step S6 includes: based on real-time monitored junction temperature data and load current change rate, using a combination of sigmoid function and cosine modulation function to dynamically allocate weight coefficients for efficiency optimization and electromagnetic suppression, constructing a composite error energy attenuation function containing a Gaussian kernel function and frequency domain band-limiting characteristics, and suppressing abnormal propagation through convolution operation, wherein the bandwidth parameter of the Gaussian kernel function is positively correlated with the system switching frequency, and the frequency domain band-limiting characteristics are determined by the system's maximum operating frequency.

[0021] Preferably, step S6 further includes: performing tensor product co-evolution of resonance parameters and synchronization parameters, establishing parameter space mapping relationship based on node coupling matrix, and defining co-evolution rules including energy conservation constraints and phase continuity conditions, wherein the construction of the coupling matrix reflects the correlation strength of electrical parameters of adjacent nodes, and the evolution rules ensure that the parameter update process maintains the system stability boundary.

[0022] The present invention also provides a soft switching of a high-frequency switching transformer and its distributed control system, the system comprising the following modules:

[0023] The network initialization module configures the distributed control network for building a ring topology and loads the improved precision time protocol.

[0024] The multi-physical quantity monitoring module is configured to collect the drain-source voltage, drain current and load current change rate in real time, and simultaneously monitor the junction temperature data of power devices and calculate the change of magnetic core flux.

[0025] The spatiotemporal synchronization control module is configured to solve the second-order partial derivatives of the phase angle between nodes and generate a synchronization drive pulse signal for timestamp calibration.

[0026] The resonant parameter adjustment module is configured to iteratively update the resonant network parameters based on the quasi-Newton method to maintain the zero-voltage switching state.

[0027] The electromagnetic optimization decision module is configured to establish a control model with the square of the drain-source voltage change rate and the electromagnetic interference spectrum norm as optimization objectives, and to generate the optimal switching trajectory command by solving the Hamiltonian equation.

[0028] The dynamic adjustment module is configured to dynamically allocate weight coefficients for efficiency optimization and electromagnetic suppression, and suppresses abnormal propagation through a composite error energy attenuation function.

[0029] The parameter evolution module is configured to perform tensor product co-evolution of resonance parameters and synchronization parameters, thereby realizing multi-parameter spatial mapping and co-evolution.

[0030] This invention provides a soft-switching method and system for a high-frequency switching transformer, along with its distributed control method. It offers the following advantages:

[0031] 1. This invention achieves precise coordination of multi-node drive signals through a distributed control network and spatiotemporal synchronization mechanism, effectively suppressing the zero-voltage switching instability problem caused by sudden load changes. A dynamic optimization algorithm for resonant parameters is employed to ensure soft-switching characteristics are maintained over a wide load range, reducing switching losses.

[0032] 2. This invention significantly reduces high-frequency harmonic radiation by constructing an electromagnetic interference spectrum norm optimization model and combining it with frequency domain band-limiting suppression technology. Through a composite error energy attenuation function, it effectively suppresses electromagnetic interference propagation under abnormal operating conditions, thereby improving the system's electromagnetic environment adaptability.

[0033] 3. This invention is based on the fusion of multi-physical quantity monitoring data to achieve real-time compensation and correction of key parameters such as junction temperature and magnetic flux. It employs a dynamic weight allocation mechanism to achieve an adaptive balance between efficiency optimization and electromagnetic suppression objectives, enabling rapid response to load fluctuations and changes in operating conditions.

[0034] 4. This invention improves the synchronization accuracy between nodes to the nanosecond level by introducing an improved precise time protocol and calculating the second-order partial derivative of the phase angle. Through a parameter co-evolution mechanism, a multi-parameter spatial mapping relationship is established, enhancing the system's parameter fault tolerance and long-term operational stability.

[0035] 5. This invention constructs a tensor product co-evolutionary framework, breaking through the limitations of traditional single-parameter optimization and achieving multi-dimensional joint optimization of resonance parameters and synchronization parameters. It adopts a distributed decision-making architecture, forming a closed-loop data interaction between functional modules, thereby improving the overall intelligent control level of the system. Attached Figure Description

[0036] Figure 1 is a flowchart of the method of the present invention;

[0037] Figure 2 is a system architecture diagram of the present invention. Detailed Implementation

[0038] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] Please refer to Figure 1. This embodiment of the invention provides a soft-switching method for a high-frequency switching transformer and its distributed control method, including the following steps:

[0040] S1. Construct a distributed control network with a ring topology, load an improved precise time protocol and initialize the equivalent inductance and capacitance parameters of the resonant cavity, and establish a clock deviation compensation matrix between nodes.

[0041] In this embodiment, when constructing a distributed control network with a ring topology, a closed-loop communication architecture is formed using no fewer than three control nodes, with information exchange between nodes achieved through bidirectional data channels. The improved precise time protocol, based on the standard IEEE-1588 protocol, adds a phase angle differential compensation mechanism. Specifically, this is achieved by embedding the phase change rate data field of adjacent nodes into the protocol frame structure, enabling real-time correction of dynamic phase deviations between nodes.

[0042] When initializing the resonant cavity parameters, the set of equivalent inductance parameters includes the equivalent inductance value L at the fundamental operating frequency. eq and its third harmonic component L 3rd L eq L is determined by the product of the core air gap length and the square of the number of turns in the winding. 3rd The capacitance parameters were obtained through nonlinear BH curve fitting of the magnetic core material. rWith transformer parasitic capacitance C p The parallel equivalent value constitutes the condition that satisfies C. total =C r +C p And it was calibrated by actual measurement using an impedance analyzer.

[0043] When establishing the clock skew compensation matrix, the calculation is based on the second-order mixed partial derivatives of the phase angles between nodes. The specific process is as follows: Let the phase angle of the i-th node be θ. i The phase angle between adjacent nodes is θ i-1 and θ i+1 Then the compensation matrix element M i,j Determined by the following formula:

[0044]

[0045] Where α is the correlation factor of the phase change rate between adjacent nodes, obtained through offline calibration; θ i-1 θ i+1 θ is the phase angle between adjacent nodes. i Let θ be the phase angle of the i-th node; j Let be the phase angle of the j-th node; t is the time variable. The cross-differentiation operation is implemented through the hardware acceleration unit of the digital signal processor, and the calculation cycle is synchronized with the switching frequency.

[0046] Preferably, the phase angle differential compensation mechanism is implemented in the synchronization message processing layer of the protocol stack, employing a two-stage correction strategy of pre-compensation and post-compensation: before sending the synchronization message, the compensation amount is predicted based on the historical phase change trend; after receiving the message, a secondary fine-tuning is performed based on the measured deviation. During the initialization of the resonant cavity parameters, the resonant points of the inductor and capacitor are dynamically matched using a frequency sweep method to ensure that the following conditions are met at the rated operating frequency:

[0047]

[0048] Where f0 is the rated operating frequency of the resonant cavity; L eq This is the equivalent inductance value, including the fundamental frequency and the third harmonic component; C total The total equivalent capacitance includes resonant capacitance and parasitic capacitance.

[0049] Furthermore, the ring topology network employs a redundant link design. When a single node fails, the communication path is automatically reconstructed via a bypass relay. The clock synchronization accuracy between nodes is maintained through the dynamic updating of the compensation matrix, with an update period of 1 / N of the switching cycle (N being the total number of nodes), ensuring that the phase deviation of the entire network is controlled within ±5 nanoseconds.

[0050] S2. Real-time acquisition of drain-source voltage, drain current and load current change rate of each node through voltage and current sensors, synchronous monitoring of junction temperature data of power devices, and calculation of core flux change based on voltage integration method.

[0051] In this embodiment, the instantaneous drain-source voltage V is acquired by a voltage sensor configured at the source of the power device. ds (t) The voltage sensor employs a differential sampling architecture with a common-mode rejection ratio of not less than 80dB to eliminate switching noise interference. Drain current I d The monitoring of (t) is achieved through a closed-loop Hall effect sensor, and the output of the sensor is connected to a differentiating circuit to obtain the load current change rate. The time constant of the differentiating circuit is inversely proportional to the switching period.

[0052] When simultaneously monitoring the junction temperature of power devices, a temperature-sensitive diode is embedded inside the device package. The junction temperature T is then inferred by detecting the temperature drift characteristics of the diode's forward voltage drop. j (t), the sampled data is converted into a digital signal by a Σ-Δ modulator. Preferably, the junction temperature monitoring system employs time-interleaved sampling technology to complete at least three independent measurements within a single switching cycle to eliminate transient thermal noise.

[0053] When calculating the change in magnetic flux in a magnetic core using the voltage integral method, the equivalent resistance R of the magnetic core is introduced. core The compensation item is calculated using the following formula:

[0054]

[0055] Where Φ(t) is the magnetic flux of the core at time t; Φ residual The remanent magnetic flux of the magnetic core is measured during the initial demagnetization process; τ is the integration time variable. The equivalent resistance R of the magnetic core... core Establishment and junction temperature T j Association model:

[0056] R core (T j )=R0·[1+β·(T j -T0)];

[0057] Where R0 is the equivalent resistance at the reference temperature T0; β is the temperature coefficient, obtained through frequency domain impedance characteristic testing of the magnetic core material; T j T0 is the junction temperature of the power device; T0 is the reference temperature for resistor calibration.

[0058] Preferably, the nonlinear integral relationship is realized by inserting a variable gain amplifier into the integrator feedback loop, whose gain coefficient K(t) is related to the magnetic flux operating point and satisfies:

[0059]

[0060] Wherein, γ is the core saturation characteristic parameter, which is determined based on the second derivative of the BH curve of the magnetic material.

[0061] When establishing a dynamic coupling correction model between junction temperature monitoring data and magnetic flux estimation, the transfer function is constructed as follows:

[0062] Φ corr (t)=Φ(t)·[1-η·(T j (t)-T ref )];

[0063] Where η is the junction temperature-magnetic flux coupling coefficient; T ref The reference junction temperature is used. The correction model is updated in real time through recursive calculations of a digital filter, and its update cycle is synchronized with the magnetic flux calculation cycle.

[0064] When using a sliding window algorithm to smooth the load current change rate in the time domain, the window length W is defined as an integer multiple of the switching period, and the sampled values ​​within the window are distributed according to exponential weights.

[0065]

[0066] Weighting coefficient w k =e -λk The attenuation factor λ is adaptively adjusted according to the system noise level. The mutation suppression mechanism is achieved by comparing the variance change rate of adjacent windows. When a mutation event is detected, it automatically switches to the weighted mode of the previous window's historical data.

[0067] Preferably, the smoothing process is implemented in the convolution acceleration unit of the digital signal processor, using a circular buffer to store the window data. For high-frequency interference components, a digital low-pass filter is cascaded before the sliding window operation, and its cutoff frequency maintains a fixed proportional relationship with the system switching frequency.

[0068] S3. Based on the load current change rate and magnetic flux calculation results, solve the second-order partial derivative of the phase angle between nodes to generate a synchronous drive pulse signal with timestamp calibration.

[0069] In this embodiment, when solving for the second-order mixed partial derivative of the phase angle between nodes, the phase angle data θ of adjacent nodes is used. i (t) and θ j (t), combined with the load current change rate Based on the calculated results of the magnetic flux Φ(t), the following partial differential equation is constructed:

[0070]

[0071] Where w(t) is a dynamic weighting factor, with a value range of [0, 1], determined by normalizing the absolute value of the load current change rate, and its specific expression is:

[0072]

[0073] Among them, I load (t) represents the load current at time t; ∈ represents a small positive number to prevent the denominator from being zero. The calculation of the second-order mixed partial derivative is implemented by the hardware differentiating unit of the digital signal processor, and the calculation period is synchronized with the switching frequency.

[0074] When generating the synchronous drive pulse signal, a timestamp calibration mechanism is used to correct the phase lead of the pulse leading edge. The feedforward compensation mechanism calculates the correction amount Δt by predicting the phase change trend over the next N switching cycles. adv :

[0075]

[0076] Where, k p k d The feedforward compensation coefficient is obtained through offline system identification; The phase angle acceleration of the i-th node; The phase angle of the i-th node is accelerated. Preferably, the prediction algorithm uses a third-order Taylor expansion model and fits the instantaneous angular acceleration using historical phase angle data.

[0077] The closed-loop feedback control is implemented by: taking the rising edge time t of the actually generated drive pulse signal as an example. actual Compared with the theoretically calculated value t theory The deviation Δt = t actual -t theory The input is a proportional-integral controller, and the output is used to dynamically adjust the update rate of the resonant network parameters. The controller transfer function is:

[0078]

[0079] Among them, K p K i τ represents the regulator parameters; τ is the system response delay time, determined through loop stability analysis.

[0080] Preferably, the timestamp calibration process is implemented in the timing control unit of the FPGA, using a high-precision digital delay line to adjust the pulse signal at the sub-nanosecond level. The duty cycle of the driving pulse is based on the rate of change of magnetic flux. Dynamic adjustment to meet:

[0081]

[0082] Where D0 is the reference duty cycle; β is the magnetic flux coupling coefficient, which is calibrated through a magnetic core saturation characteristic experiment.

[0083] Furthermore, the data interaction between the closed-loop feedback control and the resonant network parameter update is implemented through a shared memory pool, employing a double-buffering mechanism to avoid read / write conflicts. After each parameter update, the feedforward compensation coefficient k is dynamically optimized by comparing the phase consistency error of the driving pulses of adjacent nodes. p k d This ensures that the synchronization accuracy of the entire network remains within the preset threshold.

[0084] S4. Based on the timing characteristics of the synchronous drive pulse signal and the drain current change rate, the parameters of the resonant network are iteratively updated using the quasi-Newton method to maintain the zero-voltage switching state under different load conditions.

[0085] In this embodiment, when iteratively updating the resonant network parameters using the quasi-Newton method, the optimization objective function F(x) is defined as a weighted sum of efficiency, electromagnetic interference, and timing synchronization indices, where the parameter vector x = {L} eq C total ,Q} includes the equivalent inductance, total capacitance, and quality factor. The specific expression of the objective function is:

[0086] F(x)=α·η(x)+β·||EMI(x)||2+γ·Δt sync (x);

[0087] Where α, β, and γ are weighting coefficients; η is the calculated conversion efficiency; ||EMI||² represents the L2 norm of the electromagnetic interference spectrum; Δt sync This refers to the synchronization time deviation between nodes.

[0088] Efficiency gradient The calculation is achieved through the perturbation observation method. While keeping the switching frequency constant, a small perturbation δx is applied to the resonant cavity parameters, and the change in efficiency Δη is measured.

[0089]

[0090] Electromagnetic interference gradient The acquisition utilizes a near-field probe array to sample the spatial radiation field distribution, and extracts the energy change rate of characteristic frequency bands through Fast Fourier Transform. Time-synchronized gradient... It is calculated from the time difference of the rising edge of the driving pulses of adjacent nodes.

[0091] During the inverse operation of the Jacobian matrix, an approximate matrix H of the Hessian matrix is ​​constructed. k The BFGS update formula is used at this time:

[0092]

[0093] Among them, H k This is the approximation of the Hessian matrix for the k-th iteration; s k =x k+1 -x k Update the vector for the parameters. This represents the gradient change vector. Matrix inversion is performed using Cholesky decomposition to ensure numerical stability.

[0094] When dynamically adjusting the resonant cavity quality factor Q, establish a relationship with the rate of change of load current. Feedback relationship:

[0095]

[0096] Where Q0 is the baseline quality factor; κ is the adjustment amplitude coefficient; and ξ is the slope control factor, calibrated through load transient response testing. The closed-loop feedback control is implemented by adjusting the damping resistance value of the resonant network using a digital potentiometer to ensure that the zero-voltage switching condition V is maintained during load abrupt changes. ds (t switch ) = 0.

[0097] Preferably, the iteration step size of the quasi-Newton method is determined using the Armijo line search strategy, satisfying the sufficient descent condition:

[0098]

[0099] in, The search direction is represented by c ∈ (0,1); the attenuation coefficient is represented by λ. k This is a dynamic step size. The iteration termination condition is set to the rate of change of the objective function being less than a preset threshold for three consecutive iterations.

[0100] Furthermore, the parameter update process forms a data interaction channel with the spatiotemporal synchronization control module. When the quality factor adjustment is detected to exceed the safety threshold, an emergency correction mechanism for the synchronization pulse phase angle is triggered. The safety threshold is determined based on the maximum allowable current stress of the resonant cavity and the thermal design margin.

[0101] S5. Establish a control model with the square of the drain-source voltage change rate and the electromagnetic interference spectrum norm as optimization objectives, and obtain the optimal switching trajectory command by solving the Hamiltonian equation.

[0102] In this embodiment, when establishing the control model, the optimization objective function J is defined as a weighted combination of the time integral of the square of the drain-source voltage change rate and the electromagnetic interference spectrum norm:

[0103]

[0104] Where t0 is the optimization start time; λ is the drain-source voltage change rate; λ is the weighting coefficient. Represents the Fourier transform; ||·|| ∞ This is the maximum amplitude norm in the frequency domain. The objective function simultaneously constrains the switching transient characteristics in both the time and frequency domains, ensuring that soft-switching conditions meet electromagnetic compatibility requirements.

[0105] When constructing Hamilton's equations, state variables x = [V] are introduced. ds ,I d ,Φ] T The control variable u(t) (duty cycle of the driving pulse) is included, along with an additional constraint on the second-order differentiability of the switching trajectory:

[0106]

[0107] Among them, K max The upper limit of the acceleration due to changes in the control quantity is determined by the safe operating area of ​​the power device. Hamiltonian function. Defined as:

[0108]

[0109] Where p is the adjoint variable vector; f(x,u) is the system state equation.

[0110] When using the adjoint variable method to solve the problem, the adjoint equation is first established:

[0111]

[0112] Boundary condition processing employs the virtual terminal method, at the time domain endpoint t. f Extended virtual state satisfies p(t) f =0. During the solution process, the update of the accompanying variable and the iteration of the resonant network parameters form data coupling, specifically through the exchange of the quality factor Q and equivalent inductance L in the shared memory area. eq Real-time values.

[0113] Preferably, the generation of the optimal switching trajectory command employs a spectral method to discretize the Hamiltonian equation, transforming the continuous-time problem into a nonlinear programming problem. At time-domain grid point t... kAt kΔt, the Legendre polynomial is used to approximate the control variable u(t), and the optimal solution is obtained by solving the KKT conditions using the interior point method.

[0114] The specific implementation of the data coupling mechanism is as follows: after each resonant parameter update, the Jacobian matrix of the system state equation f(x,u) is recalculated. And update the coefficient matrix of the adjoint equation. Simultaneously, the optimal switching trajectory command u is... * The derivative information of (t) is fed back to the spatiotemporal synchronization control module to adjust the rising edge slope of the drive pulse.

[0115] Furthermore, the electromagnetic interference spectral norm is calculated using a windowed FFT technique, with the Blackman-Harris window selected as the window function to suppress spectral leakage. The key monitoring range of the frequency band is set to the odd harmonic region of the switching frequency, and the infinite norm is calculated after extracting the characteristic frequency components through bandpass filtering.

[0116] Preferably, the implementation of the quadratic differentiability constraint is achieved by introducing a slack variable ξ(t), transforming the original constraint into:

[0117]

[0118] Where u(t) is the control variable.

[0119] The constraints are incorporated into the objective function using the obstacle function method, forming an unconstrained optimization problem. During the iterative solution process, an adaptive adjustment strategy for the obstacle parameters is employed to balance convergence speed and computational accuracy.

[0120] The time-domain waveform of the optimal switching trajectory command is generated by combining a lookup table with real-time interpolation, and waveform storage and DAC output control are implemented in the FPGA. The command update cycle and the resonant parameter adjustment cycle are kept as integer multiples to ensure the consistency of the system's dynamic response.

[0121] S6. Based on the real-time monitored junction temperature data and load current change rate, dynamically allocate the weight coefficients of efficiency optimization and electromagnetic suppression, suppress abnormal propagation through the error energy attenuation function, and perform tensor product co-evolution of resonance parameters and synchronization parameters.

[0122] In this embodiment, when dynamically allocating the efficiency optimization and electromagnetic suppression weight coefficients, a combination function is constructed:

[0123] w eff (t)=σ(α·T j (t))·cos(2πf m t);

[0124] w emi(t)=1-w eff (t);

[0125] Among them, w eff (t) represents the efficiency optimization weighting coefficient; σ(·) is the sigmoid function; α is the junction temperature sensitivity coefficient; f m The modulation frequency is related to the system switching frequency f. sw Satisfy f m =f sw / N (N is an integer); T j (t) represents the real-time junction temperature. The junction temperature data T j (t) comes from the online monitoring system of the power device and is mapped to the weight domain through the temperature-efficiency transfer function.

[0126] When constructing the composite error energy attenuation function, a time-frequency domain joint processing function is defined:

[0127]

[0128] Where E(t) is the error energy at time t; ∈(τ) is the time-domain error energy; G(t; σ) is the Gaussian kernel function, and its bandwidth parameter is... (k is the scaling factor); σ is the Gaussian kernel bandwidth parameter; B(f) is the frequency domain band-limiting function:

[0129]

[0130] Among them, f max This is the system's maximum operating frequency. The convolution operation is implemented through a combination of an analog integrator circuit and a digital FIR filter to suppress the propagation of abnormal oscillations caused by sudden load changes.

[0131] When performing tensor product co-evolution, the resonant parameter vector R = [L eq C total [,Q] and the synchronization parameter vector S=[θ,Δt] adv ,k p The tensor product space of ]:

[0132]

[0133] Where R is the resonant parameter vector; S is the synchronization parameter vector; This is a tensor product operator that generates a 9×1 parameter space vector.

[0134] The node coupling matrix M is constructed to reflect the strength of parameter correlation, and the matrix elements M ij Determined by the correlation coefficient of electrical parameters of adjacent nodes:

[0135]

[0136] Where Cov represents the covariance; σ is the standard deviation. The co-evolutionary rule includes:

[0137] Energy conservation constraint:

[0138]

[0139] Phase continuity condition:

[0140] |θ i (t+Δt)-θ i (t)|≤Δθ max ;

[0141] Wherein, δP max The allowable power deviation threshold; Δθ max This is the maximum phase jump angle.

[0142] Preferably, the co-evolutionary process employs a genetic algorithm framework, defining a fitness function:

[0143]

[0144] Among them, T opt This is the historically optimal parameter tensor. The mutation operation is performed under the constraint of the coupling matrix, ensuring that the parameter update direction decreases along the energy loss gradient.

[0145] The frequency domain band-limiting characteristic is achieved by cascading multiple elliptic filters to form a filter bank, with each filter having a stopband attenuation of no less than 60 dB. Preferably, the bandwidth parameter σ of the Gaussian kernel function is dynamically adjusted according to the real-time switching frequency to satisfy σ·f sw =C (C is a constant), ensuring that the time-frequency resolution product remains constant.

[0146] Furthermore, the parameter space mapping relationship is implemented through a self-organizing feature map neural network, projecting the high-dimensional tensor space onto a two-dimensional grid to visualize and monitor the parameter evolution trajectory. When a parameter combination is detected to be approaching the stability boundary, a parameter rollback mechanism is triggered to load historical safety configurations.

[0147] The co-evolutionary rules are implemented through a constrained optimization algorithm, verifying energy conservation and phase continuity conditions after each parameter update. For candidate solutions that violate the constraints, a projection method is used to map them to the feasible region boundary, while the number of violations is recorded to adjust the learning rate of the evolutionary strategy.

[0148] Referring to Figure 2, the present invention also provides a soft-switching method for a high-frequency switching transformer and its distributed control system, the system comprising the following modules:

[0149] The network initialization module configures the distributed control network for building a ring topology and loads the improved precision time protocol.

[0150] A ring topology is used to deploy distributed control nodes, with redundant communication links forming bidirectional data channels between nodes. An improved time synchronization mechanism is integrated at the protocol stack level to support dynamic compensation for phase deviation and clock drift correction. Upon node startup, a resonant cavity parameter scan is automatically performed to establish an initial inductor-capacitor matching table, and the master / slave node roles are determined through a negotiation algorithm.

[0151] The multi-physical quantity monitoring module is configured to collect the drain-source voltage, drain current and load current change rate in real time, and simultaneously monitor the junction temperature data of power devices and calculate the change of magnetic core flux.

[0152] It integrates a high-precision sensor array, including an isolated voltage probe, a zero-flux current sensor, and an infrared thermal imaging unit, enabling simultaneous acquisition of electrical and thermal parameters. A built-in magnetic flux calculation engine employs an improved integration algorithm to eliminate measurement deviations caused by core losses, outputting a real-time magnetic flux change rate data stream. An adaptive filtering unit is configured to dynamically adjust the sampling frequency and signal bandwidth according to operating conditions.

[0153] The spatiotemporal synchronization control module is configured to solve the second-order partial derivatives of the phase angle between nodes and generate a synchronization drive pulse signal for timestamp calibration.

[0154] The phase angle processing unit is deployed based on a distributed computing architecture, and a global phase relationship matrix is ​​constructed through data exchange between adjacent nodes. The drive pulse generation unit integrates a programmable delay line, supports sub-nanosecond timestamp calibration, and has dual-mode operation capabilities of feedforward compensation and feedback adjustment. An impedance matching network is configured at the synchronization signal output end to ensure signal integrity transmission to the power stage.

[0155] The resonant parameter adjustment module is configured to iteratively update the resonant network parameters based on the quasi-Newton method to maintain the zero-voltage switching state.

[0156] A built-in parameter optimization engine uses a hybrid optimization algorithm to adjust the equivalent values ​​of resonant network components in real time. A parameter safety monitoring unit is configured to trigger a rapid correction mechanism when a resonant point offset exceeds a threshold. It also works in conjunction with the thermal management module to dynamically adjust the upper limit of the quality factor based on the heatsink temperature, preventing component damage from overstress.

[0157] The electromagnetic optimization decision module is configured to establish a control model with the square of the drain-source voltage change rate and the electromagnetic interference spectrum norm as optimization objectives, and to generate the optimal switching trajectory command by solving the Hamiltonian equation.

[0158] An integrated multi-objective optimization solver is used to construct a joint time-frequency domain analysis model. The switch trajectory planning unit adopts a model predictive control framework, combining historical data and real-time state predictions to generate the optimal driving sequence. A spectrum analysis front-end is configured to capture near-field radiation characteristics through a multi-channel data acquisition card, providing feedback input for the optimization model.

[0159] The dynamic adjustment module is configured to dynamically allocate weight coefficients for efficiency optimization and electromagnetic suppression, and suppresses abnormal propagation through a composite error energy attenuation function.

[0160] The weight allocation unit employs a nonlinear function generator to dynamically adjust and optimize target priorities based on operational status characteristics. The anomaly suppression channel features a multi-stage filtering network, including time-domain convolution kernels and frequency-domain limiters, to achieve wideband interference suppression. It also includes a built-in self-diagnostic function that can identify oscillation modes and automatically switch suppression strategies.

[0161] The parameter evolution module is configured to perform tensor product co-evolution of resonance parameters and synchronization parameters, realizing multi-parameter spatial mapping and co-evolution.

[0162] A co-evolutionary algorithm framework is deployed, and a multi-dimensional parameter space mapping relationship library is established. The evolutionary process is supervised by a stability constraint controller to ensure that parameter updates conform to the system dynamics. An evolutionary history database is configured to support rapid backtracking and loading of optimal parameter combinations, enhancing the system's resilience to disturbances.

[0163] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A distributed control method for soft switching of a high-frequency switching transformer, characterized in that, Includes the following steps: S1. Construct a distributed control network with a ring topology, load an improved precise time protocol, initialize the equivalent inductance and capacitance parameters of the resonant cavity, and establish a clock deviation compensation matrix between nodes; S2. Real-time acquisition of drain-source voltage, drain current, and load current change rate of each node using voltage and current sensors, synchronous monitoring of power device junction temperature data, and calculation of core flux change based on voltage integration method; S3. Solve for the second-order partial derivative of the phase angle between nodes based on the load current change rate and flux calculation results, and generate a synchronous drive pulse signal with timestamp calibration; wherein, based on the load current change rate and flux calculation results, the second-order mixed partial derivative of the phase angle between nodes is solved, wherein the calculation process of the partial derivative introduces a dynamic weighting factor of the load current change rate, and a synchronous drive pulse signal with timestamp calibration is generated, wherein the timestamp calibration includes phase lead correction based on feedforward compensation mechanism, and the generation process of the drive pulse signal and the resonant network parameter update form a closed-loop feedback control; S4. Based on the timing characteristics of the synchronous drive pulse signal and the drain current change rate, the resonant network parameters are iteratively updated using the quasi-Newton method to maintain the zero-voltage switching state under different load conditions. S5. A control model is established with the square of the drain-source voltage change rate and the electromagnetic interference spectrum norm as optimization objectives. The optimal switching trajectory command is obtained by solving the Hamiltonian equation. S6. Based on the real-time monitored junction temperature data and load current change rate, the weight coefficients for efficiency optimization and electromagnetic suppression are dynamically allocated. Abnormal propagation is suppressed through the error energy attenuation function, and the tensor product of the resonant parameters and synchronization parameters is co-evolved.

2. The distributed control method for soft switching of a high-frequency switching transformer according to claim 1, characterized in that, Step S1 includes: constructing a distributed control network with a ring topology, loading an improved precise time protocol that includes a phase angle differential compensation mechanism, initializing the set of equivalent inductance and capacitance parameters of the resonant cavity, and establishing a clock deviation compensation matrix based on the second-order mixed partial derivatives of the phase angle between nodes, wherein the construction process of the compensation matrix includes cross-differentiation operation on the phase change rate of adjacent nodes.

3. The distributed control method for soft switching of a high-frequency switching transformer according to claim 1, characterized in that, Step S2 includes: acquiring instantaneous values ​​of drain-source voltage at each node using a voltage sensor, obtaining drain current and load current change rate using a current sensor, synchronously monitoring real-time junction temperature data of power devices, and calculating core flux change based on voltage integration method. The calculation process of voltage integration method includes a compensation term for the equivalent resistance of the core, and the calculation result of the flux change has a nonlinear integral relationship with the drain-source voltage sampling value.

4. The distributed control method for soft switching of a high-frequency switching transformer according to claim 3, characterized in that, Step S2 further includes: establishing a dynamic coupling correction model of junction temperature monitoring data and magnetic flux estimation, and using a sliding time window algorithm to perform time-domain smoothing of the load current change rate. The smoothing process includes weighted average calculation of current sampling values ​​and a mutation suppression mechanism.

5. The distributed control method for soft switching of a high-frequency switching transformer according to claim 1, characterized in that, Step S4 includes: based on the timing characteristics of the synchronous drive pulse signal and the drain current change rate, the parameters of the resonant network are iteratively updated using the quasi-Newton method. The update process comprehensively calculates the efficiency gradient, electromagnetic interference gradient, and time synchronization gradient, and performs parameter correction through the inverse operation of the Jacobian matrix. The quality factor of the resonant cavity is dynamically adjusted to maintain the zero-voltage switching state. The adjustment of the quality factor and the load current change rate form a closed-loop feedback control.

6. The distributed control method for soft switching of a high-frequency switching transformer according to claim 1, characterized in that, Step S5 includes: establishing a control model with the square of the drain-source voltage change rate and the electromagnetic interference spectrum norm as optimization objectives; constructing a Hamiltonian equation containing the second-order differentiability constraint of the switching trajectory; using the adjoint variable method to process the boundary conditions and solve for the optimal switching trajectory command; wherein the calculation process of the adjoint variable is data-coupled with the update of the resonant network parameters; and the generation of the optimal switching trajectory command is synchronously fed back to the timing adjustment module of the synchronous drive pulse.

7. The distributed control method for soft switching of a high-frequency switching transformer according to claim 1, characterized in that, Step S6 includes: based on real-time monitored junction temperature data and load current change rate, using a combination of sigmoid function and cosine modulation function to dynamically allocate weight coefficients for efficiency optimization and electromagnetic suppression, constructing a composite error energy attenuation function containing Gaussian kernel function and frequency domain band-limiting characteristics, and suppressing abnormal propagation through convolution operation, wherein the bandwidth parameter of the Gaussian kernel function is positively correlated with the system switching frequency, and the frequency domain band-limiting characteristics are determined by the system's maximum operating frequency.

8. The distributed control method for soft switching of a high-frequency switching transformer according to claim 1, characterized in that, Step S6 further includes: performing tensor product co-evolution of resonance parameters and synchronization parameters, establishing parameter space mapping relationship based on node coupling matrix, and defining co-evolution rules including energy conservation constraints and phase continuity conditions, wherein the construction of the coupling matrix reflects the correlation strength of electrical parameters of adjacent nodes, and the evolution rules ensure that the parameter update process maintains the system stability boundary.

9. A distributed control system for soft switching of a high-frequency switching transformer, applied to the method described in any one of claims 1-8, characterized in that, The system includes the following modules: a network initialization module, configured to construct a distributed control network with a ring topology and load an improved precise time protocol; a multi-physical quantity monitoring module, configured to acquire drain-source voltage, drain current, and load current change rate in real time, synchronously monitor junction temperature data of power devices, and calculate core flux change; and a spatiotemporal synchronization control module, configured to solve for the second-order partial derivatives of the phase angle between nodes and generate a timestamp-calibrated synchronization drive pulse signal. The resonant parameter adjustment module is configured to iteratively update the resonant network parameters based on the quasi-Newton method to maintain the zero-voltage switching state. The electromagnetic optimization decision module is configured to establish a control model with the square of the drain-source voltage change rate and the electromagnetic interference spectrum norm as optimization objectives, and generate the optimal switching trajectory command by solving the Hamiltonian equation; the dynamic adjustment module is configured to dynamically allocate the weight coefficients for efficiency optimization and electromagnetic suppression, and suppress abnormal propagation through a composite error energy attenuation function; the parameter evolution module is configured to perform the tensor product co-evolution of resonance parameters and synchronization parameters, realizing multi-parameter spatial mapping and symbiotic evolution.

Images