IGBT (Insulated Gate Bipolar Translator) driving control method based on secondary side hardware protection and active clamping

Abstract

The invention discloses an IGBT (Insulated Gate Bipolar Translator) driving control method based on secondary side hardware protection and active clamping, which comprises the following specific steps of: establishing an IGBT digital twin model, and updating junction temperature distribution, saturation voltage drop and theoretical boundaries of a dynamic safe working area under different working conditions in real time through model simulation; in the normal switching process of the IGBT, voltage waveforms and current change rate waveforms in the switching-on and switching-off stages of the IGBT are collected in real time, key parameters are identified, and meanwhile self-adaptive correction is carried out; according to the invention, the conversion from passive protection to active prediction can be realized, the accuracy of protection criteria is greatly improved, the voltage stress and loss of a device are remarkably reduced, false triggering and misjudgment are reduced, the self-diagnosis capability and response speed of a system are improved, the time precision and coordination of protection decisions are ensured, and the reliability of the system is improved. Thermal fatigue and breakdown risks caused by overstress are reduced, so that the service life of the device is prolonged, and the overall reliability and operation efficiency of the power module are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of circuit protection, in particular to an IGBT drive control method based on secondary side hardware protection and active clamping. BACKGROUND

[0002] Under the background of rapid development of high power density and power electronic systems, as the core power device, insulated gate bipolar transistors (IGBT) are widely used in new energy vehicle electric drive systems, photovoltaic inverters, high-speed train traction converters and industrial frequency converters. With the continuous increase of system switching frequency and voltage level, IGBTs face more complex dynamic stress environment in the working process: overcurrent, overvoltage, parasitic oscillation and thermal coupling effect, etc. The reliability of its drive and protection circuit has become a key bottleneck restricting the performance improvement of the system. The traditional IGBT drive protection scheme mainly relies on fixed threshold detection and static safe operating area (SOA) judgment. When the junction temperature, saturation voltage drop or current carrying capacity of the device changes with the environment and aging, such static criterion cannot reflect the real-time state of the device, and two extreme problems may occur: one is that the protection triggers too early, affecting the system efficiency; the other is that the response is lagging, leading to over-stress and even breakdown of the device. In addition, the conventional overvoltage suppression method mainly uses passive clamping structures such as Zener diode or TVS, which has fixed clamping threshold and limited energy absorption efficiency, and has the problems of response lag and energy over-damage under high frequency and high dv / dt conditions; therefore, it is particularly important to invent an IGBT drive control method based on secondary side hardware protection and active clamping.

[0003] The existing IGBT drive control method based on secondary side hardware protection and active clamping can realize the transition from "passive protection" to "active prediction", greatly improve the accuracy of protection criteria, significantly reduce the voltage stress and loss of the device, reduce false triggering and misjudgment, improve the self-diagnosis ability and response speed of the system, ensure the time accuracy and coordination of protection decision, reduce the risk of thermal fatigue and breakdown caused by over-stress, thereby prolonging the service life of the device and improving the overall reliability and operating efficiency of the power module; therefore, we propose an IGBT drive control method based on secondary side hardware protection and active clamping. SUMMARY

[0004] The purpose of the present application is to solve the defects in the prior art, and to propose an IGBT drive control method based on secondary side hardware protection and active clamping.

[0005] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows: The IGBT drive control method based on secondary side hardware protection and active clamping has the following specific steps: Ⅰ. Establish a digital twin model of IGBT, and update the junction temperature distribution, saturation voltage drop and theoretical boundary of the dynamic safe operating area in real time under different operating conditions through model simulation; II. During the normal switching process of IGBT, the voltage waveform and current rate of change waveform during its turn-on and turn-off phases are collected in real time, key parameters are identified, and adaptive correction is performed simultaneously. Ⅲ. Monitor the load current trajectory in real time, predict the trend of future switching cycles, and perform flexible shutdown or adjust the gate drive strength based on the current safe operating area boundary. IV. Real-time acquisition of voltage waveforms, current change rates, and gate drive circuit oscillation signals, and identification of the current event scenario through real-time fusion analysis; V. Based on the current event identification results, select the corresponding clamping mode, and at the same time measure and model the transmission delay of the isolation link under different temperatures and voltages, and perform delay compensation.

[0006] As a further aspect of the present invention, the specific steps for establishing the IGBT digital twin model in step I are as follows: S1.1: Measure the collector-emitter voltage drop curves of IGBT under DC or low-frequency high current at multiple junction temperatures, and fit a set of parameterized nonlinear expressions representing conduction characteristics based on the measured collector-emitter voltage drop curves. At the same time, establish a corresponding nonlinear conduction model based on the parameterized nonlinear expressions. S1.2: Using pulse heating or transient power step test, record the response of the device casing or heat dissipation substrate and junction temperature over time. The thermal resistance-capacitance spectrum obtained from the test is parameterized into a multi-level Foster equivalent network through least squares fitting. Each level contains a pair of resistors and capacitors in series. Then, based on the Foster equivalent network, the corresponding thermal impedance function is established. S1.3: The conduction loss of each switching cycle is calculated using a nonlinear conduction model. The switching loss of each switching transient is calculated using the energy integration method. Based on the real-time acquired conduction loss and switching loss, the total instantaneous power is calculated. Then, the time junction temperature point is obtained by real-time numerical convolution based on the impedance function and the total instantaneous power. S1.4: Based on the preset power, a junction temperature curve is generated through predictive convolution. The predicted junction temperature curve is compared with the maximum allowable junction temperature of the device. At the same time, the timing constraints of the current IGBT drive control are established based on the collector-emitter voltage drop curve and instantaneous power at the predicted time obtained from the nonlinear conduction model. The maximum allowable current / voltage curve that satisfies all constraints within the current given time window is used as the safe operating area (SOA) boundary. Then, a complete IGBT digital twin model is established based on the nonlinear conduction model, thermal impedance function, maximum allowable junction temperature of the device, timing constraints of the current IGBT drive control, and SOA boundary.

[0007] As a further aspect of the present invention, the specific steps for identifying key parameters and performing adaptive correction in step II are as follows: S2.1: Arrange voltage sampling channels and current sensing channels respectively. Collect the voltage signal at both ends of the device through the voltage sampling channel and collect the voltage signal after isolation or shunt through the current sensing channel. Then, use bandpass filtering to remove the voltage signal and the power frequency interference and narrowband noise in the voltage signal. Then, perform discrete approximation processing on the current change rate within the preset sampling period. S2.2: Connect a controllable high-frequency test source in the gate or drive circuit, inject a sinusoidal or narrow pulse sequence of preset amplitude through the high-frequency test source, and record the test voltage on the injection port and the measured voltage and current on the device terminal. Then, divide the recorded data into multiple periodic windows and calculate the complex spectrum of the voltage and current response in each window to obtain the complex gain ratio at each frequency point. Based on the obtained complex gain ratio, calculate the small-signal equivalent impedance and transconductance at the corresponding frequency point. S2.3: During the rising and falling edge intervals of device switching, the transient segment of the voltage across the device is extracted, and the extracted transient segment is preprocessed to obtain the denoised waveform segment. Then, the waveform segment is fitted with parameters using an exponential mixture model with bias, and the time constant estimate and amplitude coefficient are obtained by nonlinear least squares method. The peak value and bandwidth of the current change rate segment are calculated. S2.4: Based on the signal equivalent impedance, transconductance, time constant, peak amplitude coefficient and bandwidth at each frequency point as observation vectors, and based on the known input quantities, a regression vector is constructed. Then, based on the observation vector and the regression vector, the RLS algorithm is used to estimate the parameter vector online and establish the corresponding covariance matrix to verify the rationality of the parameter vector generated by the online estimation. At the same time, the parameter vector that passes the verification is fed back to the IGBT digital twin model to update the model parameters.

[0008] As a further aspect of the present invention, the specific steps for real-time monitoring of the load current trajectory and prediction of future switching cycle trends in step III are as follows: S3.1: Based on the IGBT digital twin model, obtain the real-time power spectrum at the current moment, use the pre-calibrated transient thermal response function to perform convolution processing on the historical power to obtain the basic terms of the predicted value sequence of the current junction temperature trajectory, and construct the current junction temperature baseline based on the convolution result; S3.2: Based on the current safe operating area (SOA) boundary, construct the baseline SOA boundary function, taking the instantaneous voltage amplitude and junction temperature as independent variables, obtain the average current limit through the baseline SOA boundary function, and generate the corresponding baseline SOA. S3.3: Based on the baseline SOA, identify the uncertainties of covariance, measurement noise, and future load prediction error according to the parameters, construct a correction term that follows each uncertainty value and dynamically updates it. Based on a conservative factor function, adjust the baseline SOA to obtain a dynamic current boundary that is updated in real time. S3.4: Collect actual load current samples from current and historical sampling points, and use an autoregressive extrapolation model to make short-term predictions on the preset sampling step size to obtain the corresponding predicted current sequence. Compare the predicted current sequence with the dynamic current boundary at the corresponding time and find the minimum number of sampling steps that first satisfy the prediction current sequence ≥ the dynamic current boundary. S3.5: If there is a minimum number of sampling steps, calculate the gate drive strength that reduces device stress in a segmented manner within the remaining time window so that the actual current curve does not exceed the dynamic current boundary within the prediction period. If the gate drive strength exceeds the preset threshold or the time window length is lower than the preset threshold, select the flexible turn-off curve and reduce the device current to the safe area.

[0009] As a further aspect of the present invention, the specific steps for identifying the current event scenario through real-time fusion analysis in step IV are as follows: S4.1: Acquire the high-frequency voltage envelope, the derivative component of the current rate of change after bandpass, and the high-frequency oscillation pickup signal of the gate drive circuit at different times at both ends of the acquisition device. Suppress the power grid and switch base frequency interference of each group of acquisition signals through band-stop filtering. Then, perform frame processing on each group of acquisition signals and normalize each frame of acquisition signals. S4.2: Calculate the short-time peak value, envelope rise slope and spectral centroid based on the high-frequency voltage envelope; calculate the absolute value of the peak value, the frequency component of the peak value and the instantaneous bandwidth based on the derivative component after the current rate of change is bandpassed; calculate the narrowband spectral peak and harmonic ratio based on the high-frequency oscillation pickup signal; and establish the corresponding feature vectors of various types of acquired signals in each frame based on the calculation results. S4.3: Input each feature vector into the fusion classifier, and output the confidence scores of the three types of acquired signals through the linear-logistic regression network in the fusion classifier. After normalization, output the probability values ​​of each type of acquired signal. Based on the probability values ​​of each type of acquired signal, calculate the weighted average probability of each type of acquired signal within each frame time window. S4.4: If any probability value is higher than the preset category threshold and remains above the threshold for a preset duration, it is confirmed as an event of that category. Specifically, the category events include normal shutdown overshoot, short circuit fault, and parasitic oscillation.

[0010] As a further aspect of the present invention, the specific steps for measuring and modeling the transmission delay of the isolated link under different temperatures and voltages, and performing delay compensation in step V are as follows: S5.1: Set a reference test pulse sequence for calibration and periodic measurement in the primary side controller. The amplitude, pulse width and repetition period of the reference test pulse sequence can be adjusted. At the same time, the digital template of the reference test pulse sequence and the local transmission time tag are recorded in the primary side controller. S5.2: During power-on or idle maintenance periods, as well as during periodic maintenance, the primary-side controller sends a reference test pulse sequence. Upon receiving the reference test pulse sequence, the secondary-side hardware amplifies it and immediately returns the corresponding echo signal. S5.3: The primary-side controller simultaneously uses an independent sampling channel to record the transmitted waveform, calculates the cross-correlation function between the corresponding digital template of the transmitted reference test pulse sequence and the transmitted waveform, and finds its peak position to obtain the corresponding measurement delay value. S5.4: During the preparation stage or during the operation and scanning process, the delay value is measured multiple times and a collection sample set is generated. Each sample records the corresponding environmental variables. Regression fitting is performed on the data of each sample in the collection sample set to establish the corresponding delay estimation function. S5.5: During operation, the primary-side controller continuously reads the current environmental input vector, calculates the current instantaneous delay estimate through the delay estimation function, and performs smoothing filtering on the instantaneous delay estimate to obtain the corresponding smooth delay; S5.6: Before entering the protection criterion, for all arriving signals from the secondary side, generate a corresponding calibration timestamp based on the corresponding smooth delay, and use this time as the actual acquisition time of the signal.

[0011] Compared with the prior art, the beneficial effects of the present invention are as follows: This IGBT drive control method based on secondary-side hardware protection and active clamping obtains the thermal resistance-capacitance spectrum through pulse heating or transient power step testing, establishes a multi-level Foster thermal network and the corresponding thermal impedance function, and then calculates the instantaneous power based on the real-time integration of conduction and switching losses and convolves it with the thermal impedance function to predict the junction temperature. Combining the device's allowable temperature and conduction characteristics, the safe operating area boundary is determined. Then, by injecting a high-frequency test signal, voltage and current waveforms are acquired, and features such as complex spectrum, transconductance, and time constant are extracted. A recursive least squares algorithm is used to identify model parameters in real time and update the digital twin model. Based on the updated model, future power and junction temperature evolution are predicted, the SOA boundary is dynamically updated, and short-term current extrapolation is used to determine whether an out-of-bounds violation is imminent. Based on the judgment result, [further steps are taken]. The system adjusts the gate drive or performs a flexible shutdown. Simultaneously, it acquires Vce, di / dt, and gate oscillation signals, extracts multi-dimensional features, and uses a fusion classifier to determine the current event type. Finally, a reference test pulse sequence is set in the primary-side controller. An isolation link delay model is established through echo measurement and cross-correlation analysis. During operation, real-time delay estimation and compensation are performed based on environmental parameters. This enables a shift from "passive protection" to "active prediction," significantly improving the accuracy of protection criteria, significantly reducing device voltage stress and losses, reducing false triggering and misjudgment, improving the system's self-diagnostic capabilities and response speed, ensuring the time accuracy and coordination of protection decisions, reducing the risk of thermal fatigue and breakdown caused by overstress, thereby extending device life and improving the overall reliability and operating efficiency of the power module. Attached Figure Description

[0012] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0013] Figure 1 This is a flowchart of the IGBT drive control method based on secondary-side hardware protection and active clamping proposed in this invention. Detailed Implementation

[0014] The technical solution of this invention will be described in detail below with reference to the post:

[0015] Example 1: Refer to Figure 1 The IGBT drive control method based on secondary-side hardware protection and active clamping has the following specific steps: A digital twin model of the IGBT is established, and the junction temperature distribution, saturation voltage drop, and theoretical boundary of the dynamic safe operating area are updated in real time under different operating conditions through model simulation.

[0016] Specifically, the collector-emitter voltage drop curves of the IGBT under DC or low-frequency high current at multiple junction temperatures are measured. Based on the measured collector-emitter voltage drop curves, a set of parameterized nonlinear expressions representing the conduction characteristics are fitted. Simultaneously, based on these parameterized nonlinear expressions, a corresponding nonlinear conduction model is established. Using pulse heating or transient power step testing, the response of the device casing or heat sink and junction temperature over time is recorded. The obtained thermal resistance-capacitance spectrum is parameterized into a multi-level Foster equivalent network using least-squares fitting, where each level contains a series resistor-capacitor pair. Then, based on this Foster equivalent network, a corresponding thermal impedance function is established. Using the nonlinear conduction model, the conduction loss for each switching cycle is calculated, and the switching loss for each switching transient is calculated using the energy integration method. Based on the real-time acquired conduction and switching losses, the total instantaneous power is calculated. Then, according to the impedance function and the total instantaneous power, the time junction temperature point is obtained through real-time numerical convolution. According to the preset power, the junction temperature point curve is generated through predictive convolution. The predicted junction temperature point curve is compared with the maximum allowable junction temperature of the device. At the same time, the timing constraints of the current IGBT drive control are established based on the collector-emitter voltage drop curve and instantaneous power corresponding to the predicted time obtained from the nonlinear conduction model. The maximum allowable current / voltage curve that satisfies all constraints within the current given time window is used as the safe operating area (SOA) boundary. Then, a complete IGBT digital twin model is established based on the nonlinear conduction model, thermal impedance function, maximum allowable junction temperature of the device, timing constraints of the current IGBT drive control, and SOA boundary.

[0017] During the normal switching process of IGBT, the voltage waveform and current rate of change waveform during its turn-on and turn-off phases are collected in real time, key parameters are identified, and adaptive corrections are performed.

[0018] Specifically, voltage sampling channels and current sensing channels are set up separately. The voltage sampling channel acquires the voltage signal across the device, and the current sensing channel acquires the voltage signal after isolation or a shunt. Then, bandpass filtering is used to remove power frequency interference and narrowband noise from the voltage signal. The rate of change of current is then discretized and approximated within a preset sampling period. A controllable high-frequency test source is connected to the gate or drive circuit. A sinusoidal or narrow pulse sequence of preset amplitude is injected through the high-frequency test source, and the test voltage at the injection port and the measured voltage and current at the device terminals are recorded. The recorded data are then divided into multiple periodic windows, and the complex spectrum of the voltage and current response within each window is calculated to obtain the complex gain ratio at each frequency point. Based on the obtained complex gain ratio, the small-signal equivalent impedance and cross-current impedance at the corresponding frequency point are calculated. In the process of device switching, the transient segment of the voltage across the device is extracted during the rising and falling edges. This extracted transient segment is preprocessed to obtain a denoised waveform segment. Then, the waveform segment is fitted with parameters using a biased exponential mixture model. The time constant estimate and amplitude coefficient are obtained using the nonlinear least squares method. The peak value and bandwidth of the current rate of change segment are calculated. The signal equivalent impedance, transconductance, time constant, peak value of amplitude coefficient, and bandwidth at each frequency point are used as observation vectors. At the same time, a regression vector is constructed based on the known input quantities. Then, based on the observation vector and the regression vector, the RLS algorithm is used to estimate the parameter vector online and establish the corresponding covariance matrix. The rationality of the parameter vector generated by the online estimation is verified. The parameter vector that passes the verification is fed back to the IGBT digital twin model for model parameter update.

[0019] It monitors the load current trajectory in real time, predicts the trend of future switching cycles, and performs flexible shutdown or adjusts the gate drive strength based on the current safe operating area boundary.

[0020] Specifically, based on the IGBT digital twin model, the real-time power spectrum is obtained at the current moment. Using a pre-calibrated transient thermal response function, historical power is convolved to obtain the basic terms of the predicted value sequence of the current junction temperature trajectory. Based on the convolution result, the current junction temperature baseline is constructed. According to the current safe operating area (SOA) boundary, a baseline SOA boundary function is constructed, using instantaneous voltage amplitude and junction temperature as independent variables. The average current limit is obtained through the baseline SOA boundary function, and a corresponding baseline SOA is generated. Based on the baseline SOA, the uncertainties of covariance, measurement noise, and future load prediction error are identified according to parameters. A correction term is dynamically updated to follow each uncertainty value. Based on a conservative factor function, the baseline is adjusted downwards. The system employs line SOA to obtain real-time updated dynamic current boundaries, collects actual load current samples from current and historical sampling points, and uses an autoregressive extrapolation model to make short-term predictions for a preset sampling step size to obtain the corresponding predicted current sequence. The predicted current sequence is compared with the dynamic current boundary at the corresponding time, and the minimum sampling step size for the first time to satisfy the condition that the predicted current sequence is greater than or equal to the dynamic current boundary is found. If a minimum sampling step size exists, the gate drive strength is calculated to reduce device stress in a segmented manner within the remaining time window so that the actual current curve does not exceed the dynamic current boundary within the prediction period. If the gate drive strength exceeds a preset threshold or the time window length is less than a preset threshold, a flexible turn-off curve is selected, and the device current is reduced to a safe area.

[0021] Example 2: Refer to Figure 1 The IGBT drive control method based on secondary-side hardware protection and active clamping has the following specific steps: The system collects voltage waveforms, current change rates, and gate drive circuit oscillation signals in real time, and identifies the current event scenario through real-time fusion analysis.

[0022] Specifically, the high-frequency voltage envelope at both ends of the acquisition device at different times, the derivative component of the current rate of change after bandpass, and the high-frequency oscillation pickup signal of the gate drive circuit are collected. Band-stop filtering is used to suppress grid and switch fundamental frequency interference in each group of acquired signals. Then, each group of acquired signals is processed in frames, and each frame is normalized. Based on the high-frequency voltage envelope, the short-time peak value, envelope rise slope, and spectral centroid are calculated. Based on the derivative component of the current rate of change after bandpass, the absolute value of the peak value, the frequency component where the peak value is located, and the instantaneous bandwidth are calculated. Based on the high-frequency oscillation pickup signal, the narrowband spectral peak and harmonic ratio are calculated, and... Based on the calculation results, corresponding feature vectors for each type of acquired signal in each frame are established. Each feature vector is input into the fusion classifier, and the confidence scores of the three types of acquired signals are output through the linear-logistic regression network in the fusion classifier. After normalization processing, the probability values ​​of each type of acquired signal are output. Based on the probability values ​​of each type of acquired signal, the weighted average probability of each type of acquired signal within the time window of each frame is calculated. If any probability value is higher than the preset category threshold and remains above the threshold for a preset duration, it is confirmed as an event of that category. Specifically, the category events include normal shutdown overshoot, short circuit fault, and parasitic oscillation.

[0023] Based on the current event identification results, select the corresponding clamping mode, and simultaneously measure and model the transmission delay of the isolated link under different temperatures and voltages, and perform delay compensation.

[0024] Specifically, a reference test pulse sequence for calibration and periodic measurement is set in the primary-side controller. This reference test pulse sequence is adjustable in amplitude, pulse width, and repetition period. Simultaneously, the primary-side controller records the digital template of this reference test pulse sequence along with a local transmission time stamp. During power-on or idle maintenance periods, and during periodic maintenance, the primary-side controller sends the reference test pulse sequence. Upon receiving the reference test pulse sequence, the secondary-side hardware amplifies it and immediately returns the corresponding echo signal. The primary-side controller simultaneously records the returned waveform using an independent sampling channel, calculates the cross-correlation function between the corresponding digital template of the transmitted reference test pulse sequence and the returned waveform, and searches for its... Peak position is used to obtain the corresponding measurement delay value. During the preparation stage or operation condition scanning process, the delay value is measured multiple times and a collection sample set is generated. Each sample records the corresponding environmental variables. Regression fitting is performed on the data of each sample in the collection sample set to establish the corresponding delay estimation function. During operation, the primary side controller continuously reads the current environmental input vector and calculates the current instantaneous delay estimate through the delay estimation function. The instantaneous delay estimate is then smoothed and filtered to obtain the corresponding smooth delay. Before entering the protection criterion, for all arriving signals from the secondary side, a corresponding calibration timestamp is generated based on the corresponding smooth delay, and this time is used as the actual acquisition time of the signal.

Claims

1. An IGBT drive control method based on secondary-side hardware protection and active clamping, characterized in that, The specific steps of this control scheme are as follows: Ⅰ. Establish a digital twin model of IGBT, and update the junction temperature distribution, saturation voltage drop, and theoretical boundary of the dynamic safe operating area in real time under different operating conditions through model simulation; II. During the normal switching process of IGBT, the voltage waveform and current rate of change waveform during its turn-on and turn-off phases are collected in real time, key parameters are identified, and adaptive correction is performed simultaneously. Ⅲ. Monitor the load current trajectory in real time, predict the trend of future switching cycles, and perform flexible shutdown or adjust the gate drive strength based on the current safe operating area boundary. IV. Real-time acquisition of voltage waveforms, current change rates, and gate drive circuit oscillation signals, and identification of the current event scenario through real-time fusion analysis; V. Based on the current event identification results, select the corresponding clamping mode, and at the same time measure and model the transmission delay of the isolation link under different temperatures and voltages, and perform delay compensation.

2. The IGBT drive control method based on secondary-side hardware protection and active clamping according to claim 1, characterized in that, The specific steps for establishing the IGBT digital twin model described in Step I are as follows: S1.1: Measure the collector-emitter voltage drop curves of IGBT under DC or low-frequency high current at multiple junction temperatures, and fit a set of parameterized nonlinear expressions representing conduction characteristics based on the measured collector-emitter voltage drop curves. At the same time, establish a corresponding nonlinear conduction model based on the parameterized nonlinear expressions. S1.2: Using pulse heating or transient power step test, record the response of the device casing or heat dissipation substrate and junction temperature over time. The thermal resistance-capacitance spectrum obtained from the test is parameterized into a multi-level Foster equivalent network through least squares fitting. Each level contains a pair of resistors and capacitors in series. Then, based on the Foster equivalent network, the corresponding thermal impedance function is established. S1.3: The conduction loss of each switching cycle is calculated using a nonlinear conduction model. The switching loss of each switching transient is calculated using the energy integration method. Based on the real-time acquired conduction loss and switching loss, the total instantaneous power is calculated. Then, the time junction temperature point is obtained by real-time numerical convolution based on the impedance function and the total instantaneous power. S1.4: Based on the preset power, a junction temperature curve is generated through predictive convolution. The predicted junction temperature curve is compared with the maximum allowable junction temperature of the device. At the same time, the timing constraints of the current IGBT drive control are established based on the collector-emitter voltage drop curve and instantaneous power at the predicted time obtained from the nonlinear conduction model. The maximum allowable current / voltage curve that satisfies all constraints within the current given time window is used as the safe operating area (SOA) boundary. Then, a complete IGBT digital twin model is established based on the nonlinear conduction model, thermal impedance function, maximum allowable junction temperature of the device, timing constraints of the current IGBT drive control, and SOA boundary.

3. The IGBT drive control method based on secondary-side hardware protection and active clamping according to claim 2, characterized in that, The specific steps for identifying key parameters and performing adaptive corrections as described in step II are as follows: S2.1: Arrange voltage sampling channels and current sensing channels respectively. Collect the voltage signal at both ends of the device through the voltage sampling channel and collect the voltage signal after isolation or shunt through the current sensing channel. Then, use bandpass filtering to remove the voltage signal and the power frequency interference and narrowband noise in the voltage signal. Then, perform discrete approximation processing on the current change rate within the preset sampling period. S2.2: Connect a controllable high-frequency test source in the gate or drive circuit, inject a sinusoidal or narrow pulse sequence of preset amplitude through the high-frequency test source, and record the test voltage on the injection port and the measured voltage and current on the device terminal. Then, divide the recorded data into multiple periodic windows and calculate the complex spectrum of the voltage and current response in each window to obtain the complex gain ratio at each frequency point. Based on the obtained complex gain ratio, calculate the small-signal equivalent impedance and transconductance at the corresponding frequency point. S2.3: During the rising and falling edge intervals of device switching, the transient segment of the voltage across the device is extracted, and the extracted transient segment is preprocessed to obtain the denoised waveform segment. Then, the waveform segment is fitted with parameters using an exponential mixture model with bias, and the time constant estimate and amplitude coefficient are obtained by nonlinear least squares method. The peak value and bandwidth of the current change rate segment are calculated. S2.4: Based on the signal equivalent impedance, transconductance, time constant, peak amplitude coefficient and bandwidth at each frequency point as observation vectors, and based on the known input quantities, a regression vector is constructed. Then, based on the observation vector and the regression vector, the RLS algorithm is used to estimate the parameter vector online and establish the corresponding covariance matrix to verify the rationality of the parameter vector generated by the online estimation. At the same time, the parameter vector that passes the verification is fed back to the IGBT digital twin model to update the model parameters.

4. The IGBT drive control method based on secondary-side hardware protection and active clamping according to claim 3, characterized in that, The specific steps for real-time monitoring of the load current trajectory and prediction of future switching cycle trends described in step III are as follows: S3.1: Based on the IGBT digital twin model, obtain the real-time power spectrum at the current moment, use the pre-calibrated transient thermal response function to perform convolution processing on the historical power to obtain the basic terms of the predicted value sequence of the current junction temperature trajectory, and construct the current junction temperature baseline based on the convolution result; S3.2: Based on the current safe operating area (SOA) boundary, construct the baseline SOA boundary function, taking the instantaneous voltage amplitude and junction temperature as independent variables, obtain the average current limit through the baseline SOA boundary function, and generate the corresponding baseline SOA. S3.3: Based on the baseline SOA, identify the uncertainties of covariance, measurement noise, and future load prediction error according to the parameters, construct a correction term that follows each uncertainty value and dynamically updates it. Based on a conservative factor function, adjust the baseline SOA to obtain a dynamic current boundary that is updated in real time. S3.4: Collect actual load current samples from current and historical sampling points, and use an autoregressive extrapolation model to make short-term predictions on the preset sampling step size to obtain the corresponding predicted current sequence. Compare the predicted current sequence with the dynamic current boundary at the corresponding time and find the minimum number of sampling steps that first satisfy the prediction current sequence ≥ the dynamic current boundary. S3.5: If there is a minimum number of sampling steps, calculate the gate drive strength that reduces device stress in a segmented manner within the remaining time window so that the actual current curve does not exceed the dynamic current boundary within the prediction period. If the gate drive strength exceeds the preset threshold or the time window length is lower than the preset threshold, select the flexible turn-off curve and reduce the device current to the safe area.

5. The IGBT drive control method based on secondary-side hardware protection and active clamping according to claim 1, characterized in that, The specific steps for identifying the current event scenario through real-time fusion analysis described in step IV are as follows: S4.1: Acquire the high-frequency voltage envelope, the derivative component of the current rate of change after bandpass, and the high-frequency oscillation pickup signal of the gate drive circuit at different times at both ends of the acquisition device. Suppress the power grid and switch base frequency interference of each group of acquisition signals through band-stop filtering. Then, perform frame processing on each group of acquisition signals and normalize each frame of acquisition signals. S4.2: Calculate the short-time peak value, envelope rise slope and spectral centroid based on the high-frequency voltage envelope; calculate the absolute value of the peak value, the frequency component of the peak value and the instantaneous bandwidth based on the derivative component after the current rate of change is bandpassed; calculate the narrowband spectral peak and harmonic ratio based on the high-frequency oscillation pickup signal; and establish the corresponding feature vectors of various types of acquired signals in each frame based on the calculation results. S4.3: Input each feature vector into the fusion classifier, and output the confidence scores of the three types of acquired signals through the linear-logistic regression network in the fusion classifier. After normalization, output the probability values ​​of each type of acquired signal. Based on the probability values ​​of each type of acquired signal, calculate the weighted average probability of each type of acquired signal within each frame time window. S4.4: If any probability value is higher than the preset category threshold and remains above the threshold for a preset duration, it is confirmed as an event of that category. Specifically, the category events include normal shutdown overshoot, short circuit fault, and parasitic oscillation.

6. The IGBT drive control method based on secondary-side hardware protection and active clamping according to claim 5, characterized in that, The specific steps for measuring and modeling the transmission delay of the isolated link under different temperatures and voltages, and performing delay compensation as described in step V, are as follows: S5.1: Set a reference test pulse sequence for calibration and periodic measurement in the primary side controller. The amplitude, pulse width and repetition period of the reference test pulse sequence can be adjusted. At the same time, the digital template of the reference test pulse sequence and the local transmission time tag are recorded in the primary side controller. S5.2: During power-on or idle maintenance periods, as well as during periodic maintenance, the primary-side controller sends a reference test pulse sequence. Upon receiving the reference test pulse sequence, the secondary-side hardware amplifies it and immediately returns the corresponding echo signal. S5.3: The primary-side controller simultaneously uses an independent sampling channel to record the transmitted waveform, calculates the cross-correlation function between the corresponding digital template of the transmitted reference test pulse sequence and the transmitted waveform, and finds its peak position to obtain the corresponding measurement delay value. S5.4: During the preparation stage or during the operation and scanning process, the delay value is measured multiple times and a collection sample set is generated. Each sample records the corresponding environmental variables. Regression fitting is performed on the data of each sample in the collection sample set to establish the corresponding delay estimation function. S5.5: During operation, the primary-side controller continuously reads the current environmental input vector, calculates the current instantaneous delay estimate through the delay estimation function, and performs smoothing filtering on the instantaneous delay estimate to obtain the corresponding smooth delay; S5.6: Before entering the protection criterion, for all arriving signals from the secondary side, generate a corresponding calibration timestamp based on the corresponding smooth delay, and use this time as the actual acquisition time of the signal.

Images