Electromagnetic interference suppression method based on generative adversarial network and power control circuit

Abstract

The invention provides an electromagnetic interference suppression method based on a generative adversarial network and a power control circuit, and belongs to the technical field of electromagnetic interference suppression, and the method comprises the steps: collecting voltage and current signals and electromagnetic interference signals of a power semiconductor device, and inputting the voltage and current signals and the electromagnetic interference signals into a pre-trained electromagnetic interference suppression model, the model is constructed based on a generative adversarial network architecture, the generator is used for outputting control parameters for driving the power semiconductor device, and the discriminator is used for evaluating the difference between an electromagnetic interference value corresponding to the control parameters output by the generator and a target interference value; and converting the model output control parameter to generate a driving signal. The method has the beneficial effects that the generator parameters are adaptively adjusted according to the actual working condition and interference condition of the power semiconductor device control circuit, and the adaptability is high; eMI is quickly and accurately identified and suppressed by using deep learning, and the suppression efficiency and the overall efficiency of the system are improved; eMI radiated outwards by the circuit can be suppressed, and the problem of internal interference of the circuit can be solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electromagnetic interference suppression, and particularly relates to an electromagnetic interference suppression method based on a generative adversarial network and a power control circuit. BACKGROUND

[0002] As a commonly used power semiconductor device, an Insulated Gate Bipolar Transistor (IGBT) is widely used in power electronic systems such as frequency conversion speed regulation devices and uninterruptible power supplies due to its high input impedance, low conduction voltage drop, and fast switching speed. In real industrial production and electronic communication fields, many industrial automation production lines, power systems, and electronic communication equipment highly depend on IGBT devices to achieve efficient power conversion and control.

[0003] However, IGBT devices generate electromagnetic interference (EMI) during operation, especially during switching. These electromagnetic interferences affect the stability and reliability of the control circuit itself and have adverse effects on surrounding electronic equipment, leading to performance degradation or even failure of the equipment. If the electromagnetic interference generated by the IGBT control circuit cannot be effectively suppressed, it will cause production line failures, power system instability, and communication signal interference, resulting in significant economic losses. Therefore, it is of great value and significance to suppress electromagnetic interference of IGBT control circuits.

[0004] Currently, there are some common methods for suppressing electromagnetic interference of IGBT control circuits, but these methods have certain limitations.

[0005] One common method is to add a buffer circuit between the collector and emitter of the IGBT device. The buffer circuit is usually composed of resistors, capacitors, and diodes, and its working principle is to absorb part of the energy during IGBT switching, reduce the rate of voltage and current change, and thus reduce the generation of electromagnetic interference. However, the buffer circuit increases the complexity and cost of the circuit, and the resistors in the buffer circuit consume certain energy, reducing system efficiency. Moreover, the parameter design of the buffer circuit needs to be optimized according to specific application scenarios, and if the parameters are not properly set, it is difficult to achieve ideal electromagnetic interference suppression effect.

[0006] Another method is to wrap the IGBT control circuit with a metal shield, which reflects or absorbs electromagnetic energy, preventing electromagnetic interference from radiating outward and reducing interference with the surrounding environment. However, the design and installation of the shield need to consider space, heat dissipation and other factors, and the shielding effect is affected by factors such as material, structure and grounding. In addition, the shield can only prevent electromagnetic interference from radiating outward, and cannot solve the problem of interference inside the circuit.

[0007] Another method is to add filters such as LC filters and pi filters at the power input and signal output. Filters can attenuate specific frequency interference signals and improve the anti-interference ability of the circuit. However, the performance of the filter is limited by its component parameters and structure, and the suppression effect of high-frequency interference signals may not be ideal. At the same time, the use of filters increases the size and cost of the circuit, and also introduces certain insertion loss, affecting the overall performance of the system.

[0008] In summary, the existing electromagnetic interference suppression methods for IGBT control circuits have low efficiency, high cost, poor adaptability and other problems, which cannot meet the needs of increasingly complex power electronic systems. SUMMARY

[0009] To solve the above technical problems, the present application provides an electromagnetic interference suppression method based on a generative adversarial network. On the other hand, a power control circuit is also provided.

[0010] The technical problems solved by the present application can be realized by the following technical solutions:

[0011] An electromagnetic interference suppression method based on a generative adversarial network applied to a power semiconductor device, comprising:

[0012] Step S1, collecting voltage and current signals and electromagnetic interference signals of the power semiconductor device;

[0013] Step S2, inputting the collected voltage and current signals and electromagnetic interference signals into a pre-trained electromagnetic interference suppression model, the electromagnetic interference suppression model being constructed based on a generative adversarial network architecture, the generative adversarial network including a generator and a discriminator, the generator being used to output control parameters for driving the power semiconductor device, and the discriminator being used to evaluate the difference between the electromagnetic interference value corresponding to the control parameters output by the generator and the target interference value;

[0014] Step S3, converting the control parameters output by the electromagnetic interference suppression model to generate a driving signal for driving the power semiconductor device.

[0015] Preferably, the training step of the electromagnetic interference suppression model comprises:

[0016] Step A1, collect the voltage and current signals of the power semiconductor device during switching process and the electromagnetic interference signals under the preset frequency band, and construct a training data set;

[0017] Step A2, construct a generative adversarial network architecture, and train the generator and discriminator in the constructed generative adversarial network architecture according to the training data set, alternately update the model weight until the electromagnetic interference value corresponding to the output control parameter is reduced to the target interference threshold, and obtain a trained electromagnetic interference suppression model.

[0018] Preferably, the preset frequency band is 150 kHz to 1 GHz.

[0019] Preferably, the voltage and current signals of the power semiconductor device include drain-source voltage signals, collector current signals, and driving voltage signals.

[0020] Preferably, the step S2 further comprises:

[0021] The collected voltage and current signals and electromagnetic interference signals are preprocessed, and the preprocessed voltage and current signals and electromagnetic interference signals are input into the pre-trained electromagnetic interference suppression model to obtain the control parameters for driving the power semiconductor device.

[0022] Preferably, the control parameters include the driving voltage slope and the switching frequency of the power semiconductor device.

[0023] Preferably, the step S3 comprises:

[0024] Step S31, digital-to-analog conversion is performed on the control parameters output by the electromagnetic interference suppression model to obtain an analog signal;

[0025] Step S32, the rise or fall time and the switching frequency of the driving signal are adjusted according to the analog signal to generate the driving signal.

[0026] Preferably, the generative adversarial network is any one or a combination of more than one of a generative adversarial network based on a large language model, a deep convolutional generative adversarial network, and a generative adversarial network with gradient penalty.

[0027] Preferably, the power semiconductor device is an insulated gate bipolar transistor.

[0028] In another aspect, a power control circuit is also provided, which comprises a power semiconductor device, and controls the on-off state of the power semiconductor device under the action of a driving signal, and the driving signal is generated by the above-mentioned electromagnetic interference suppression method based on a generative adversarial network.

[0029] The advantages or benefits of the technical scheme of the present application are as follows:

[0030] The present application adaptively adjusts the generator parameters according to the actual working conditions and interference of the power semiconductor device control circuit, without complex parameter design and optimization, effectively suppresses different types and intensities of EMI, and has strong adaptability; at the same time, the method quickly and accurately identifies and suppresses EMI by using deep learning, without additional energy consumption, without insertion loss, and improves the suppression efficiency and overall system efficiency; in addition, the method can not only suppress the EMI radiated outward by the circuit, but also solve the internal interference problem of the circuit. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 For the preferred embodiment of the present application, a flowchart of the electromagnetic interference suppression method based on the generative adversarial network is shown.

[0032] Figure 2 For the preferred embodiment of the present application, a flowchart of the electromagnetic interference suppression model training is shown.

[0033] Figure 3 For the preferred embodiment of the present application, a flowchart of the generation of the driving signal is shown. DETAILED DESCRIPTION

[0034] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0035] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0036] The present application will be further described below with reference to the drawings and specific embodiments, but not as a limitation of the present application.

[0037] In the preferred embodiment of the present application, in view of the above problems existing in the prior art, a method for suppressing electromagnetic interference based on a generative adversarial network is provided, which is applied to a control circuit of a power semiconductor device to suppress the electromagnetic interference (EMI) generated thereby, and improve the stability and reliability of the power electronic system.

[0038] As shown in Figure 1 The electromagnetic interference suppression method comprises the following steps:

[0039] Step S1, collecting voltage and current signals of the power semiconductor device and electromagnetic interference signals;

[0040] Step S2, input the collected voltage and current signals and electromagnetic interference signals into the pre-trained electromagnetic interference suppression model, the electromagnetic interference suppression model is constructed based on a generative adversarial network architecture, the generative adversarial network includes a generator and a discriminator, the generator is used to output control parameters for driving the power semiconductor device, and the discriminator is used to evaluate the difference between the electromagnetic interference value corresponding to the control parameters output by the generator and the target interference value.

[0041] Step S3, converting the control parameters output by the electromagnetic interference suppression model to generate a driving signal for driving the power semiconductor device.

[0042] Specifically, in view of the problems of low efficiency, high cost and poor adaptability of the existing IGBT control circuit electromagnetic interference suppression method, the embodiment constructs an electromagnetic interference suppression model based on a generative adversarial network (GAN) architecture, which includes a generator and a discriminator. The generator is used to output control parameters for driving the power semiconductor device, such as driving voltage slope and switching frequency, etc. The discriminator is used to evaluate the difference between the electromagnetic interference value corresponding to the control parameters output by the generator and the target interference value. Through the adversarial training of the generator and the discriminator, the generator can output more optimal control parameters, thereby improving the stability and reliability of the power semiconductor device control circuit, reducing the influence of electromagnetic interference on the surrounding environment, meeting the needs of increasingly complex power electronic systems, and having broad application prospects and great practical value.

[0043] The above method can adaptively adjust the parameters of the generator according to the actual working state and interference of the power semiconductor device control circuit, thereby realizing effective suppression of different types and intensities of EMI. Compared with traditional methods, it does not need complex parameter design and optimization, and has stronger adaptability.

[0044] The generative adversarial network is a deep learning model that uses the powerful computing power of deep learning to quickly and accurately identify and suppress EMI, improving the suppression efficiency. Moreover, it does not consume additional energy like a buffer circuit, nor does it introduce insertion loss like a filter, thereby improving the overall efficiency of the system.

[0045] Compared with traditional filtering techniques, this method not only suppresses the EMI radiated by the circuit, but also solves the internal interference problem of the circuit; at the same time, it can process various signals in the power semiconductor device control circuit, reducing the generation of EMI from the source.

[0046] Further, the power semiconductor device is an insulated gate bipolar transistor (IGBT).

[0047] Further, the above method can be applied to other power semiconductor devices such as metal-oxide-semiconductor field-effect transistor (MOSFET) in addition to insulated gate bipolar transistor (IGBT).

[0048] Further, as shown in Figure 2 The training step of the electromagnetic interference suppression model includes:

[0049] Step A1, collect the voltage and current signals of the power semiconductor device in the switching process and the electromagnetic interference signals in the preset frequency band, and construct a training data set;

[0050] Further, the voltage and current signals of the power semiconductor device include drain-source voltage (Vds) signal, collector current (Ic) signal, and driving voltage (Vge) signal.

[0051] Further, the preset frequency band is 150 kHz-1 GHz.

[0052] Step A2, construct a generative adversarial network architecture, and train the generator and discriminator in the constructed generative adversarial network architecture according to the training data set, alternately update the model weights until the electromagnetic interference value corresponding to the control parameter is reduced to the target interference threshold, and obtain the trained electromagnetic interference suppression model; further, the control parameter includes the driving voltage slope and the switching frequency of the power semiconductor device.

[0053] Specifically, the collector current (Ic) signal is obtained by a Hall current sensor. The Hall current sensor is deployed in the collector current path of the IGBT control circuit. When installed, the sensor is connected to the circuit stably and reliably, while avoiding the interference of external magnetic field, so as to ensure that it can accurately sense the collector current (Ic) signal.

[0054] The drain-source voltage (Vds) signal and the driving voltage (Vge) signal can be obtained by a voltage probe. The voltage probe is connected to the drain and source of the IGBT device to measure the drain-source voltage (Vds); at the same time, it is connected to the driving end of the IGBT device to measure the driving voltage (Vge). In order to improve the accuracy of the data, the probe needs to be firmly connected and well contacted to prevent measurement error caused by looseness.

[0055] The electromagnetic interference signal can be obtained by an EMI receiver. The EMI receiver is set at a suitable position in the IGBT control circuit, and its parameters are adjusted to cover the frequency band of 150 kHz-1 GHz, which is used to collect the EMI signal in the frequency band.

[0056] During the acquisition process, a synchronous trigger mode is used to ensure that all devices work synchronously at a sampling frequency of 200 MHz. The synchronous trigger signal can be realized by hardware triggering or software triggering to ensure that the timing data of the collected drain-source voltage (Vds), collector current (Ic), driving voltage (Vge) and electromagnetic interference (EMI) signals are accurately corresponding.

[0057] The collected data can be stored using a high-speed data acquisition card, and the integrity and accuracy of the data are ensured during storage to avoid data loss or errors. The stored data format is uniform to facilitate subsequent analysis and processing.

[0058] The collected timing data can calculate the sampling point number by the following formula:

[0059] N = f s * t;

[0060] Wherein, N represents the sampling point number; f s represents the sampling frequency, f s is preferably 200 MHz; t represents the acquisition time.

[0061] The architecture of the generative adversarial network is constructed, the generator is used to output the IGBT control parameters, i.e. the switching frequency and the driving voltage slope; the discriminator is used to evaluate the difference between the EMI suppression effect and the target interference threshold, forming an adversarial training structure.

[0062] The input of the generator is a random noise vector, which can be sampled from a Gaussian distribution. The generator is composed of multiple fully connected layers and activation functions. The fully connected layer is used to linearly transform the input noise vector, and the activation function can be selected as the ReLU activation function. This function can effectively alleviate the gradient disappearance problem and speed up the model convergence speed. After multiple transformations, the generator outputs the IGBT control parameters, i.e. the switching frequency and the driving voltage slope.

[0063] Further, the generator includes a batch normalization layer, which makes the input data distribution of each layer more stable and improves the stability and efficiency of the training. The mathematical principle of the batch normalization layer can be expressed as:

[0064]

[0065] Wherein, x i represents the input data; i represents the data sample index; μ β and respectively represent the mean and variance of the batch β data; ∈ represents a constant to avoid a zero denominator; represents the input data after batch normalization.

[0066] The discriminator is constructed to receive two inputs, one is the EMI suppression effect data corresponding to the IGBT control parameters output by the generator, and the other is the target EMI suppression effect data, i.e. the target interference threshold. The discriminator is also composed of multiple fully connected layers, and the last layer uses a Sigmoid activation function to map the output value to the interval [0, 1] for judging whether the input data is from the generator or the target data.

[0067] The voltage and current signals collected above and electromagnetic interference (EMI) signals are used to construct a training data set. The training data set contains various types of electromagnetic interference (EMI) data collected related to the target scenario, ensuring the diversity and representativeness of the data. These data will be used as real samples to train the above-mentioned generative adversarial network architecture.

[0068] For the generator part, random noise is used as input, and a series of control parameters are output through the operation of multiple layers of neural networks. These control parameters will be used to generate simulated EMI data. For the discriminator part, real EMI data and simulated EMI data corresponding to the control parameters generated by the generator are received, and the loss is calculated by comparing the real and simulated EMI data, and the model weight is alternately updated to output parameters that stably reduce EMI to the target interference threshold.

[0069] To measure the performance of the discriminator, the discriminator uses the mean square error (MSE) loss function to calculate the loss by comparing the real and simulated EMI data, and the formula is:

[0070]

[0071] where y i represents the real EMI data value; represents the simulated EMI data value; n represents the number of data samples.

[0072] During the training process, the generator and the discriminator are trained in an adversarial manner, and the training is alternately performed. First, the weights of the generator are fixed, and the weights of the discriminator are updated so that it can better distinguish between real and simulated data. Then, the weights of the discriminator are fixed, and the weights of the generator are updated so that the generator generates more realistic simulated data to deceive the discriminator. The process of alternately updating is repeated until the control parameters output by the generator are stable and the simulated EMI data generated by the generator reduces the actual EMI value below the target threshold. At this time, the training is completed.

[0073] The goal of the generator is to generate IGBT control parameters that can deceive the discriminator into judging as target data, while the goal of the discriminator is to accurately distinguish between the data generated by the generator and the target data. Through continuous iterative training, the IGBT control parameters generated by the generator can better achieve EMI suppression effect.

[0074] Further, step S2 also includes:

[0075] The collected voltage and current signals and electromagnetic interference signals are preprocessed, and the preprocessed voltage and current signals and electromagnetic interference signals are input into the pre-trained electromagnetic interference suppression model to obtain control parameters for driving the power semiconductor device.

[0076] Specifically, after collecting the drain-source voltage Vds, collector current Ic, drive voltage Vge, and electromagnetic interference value of the control circuit, the collected data is preliminarily preprocessed to remove possible noise interference, and a digital filtering algorithm such as Kalman filtering can be used to improve the accuracy of the data.

[0077] Next, the preprocessed data is input into the pre-trained electromagnetic interference suppression model, and the optimal control parameters are calculated in real time. The model quickly outputs optimal IGBT drive slope (such as 1.2V / ns), switching frequency (such as 18kHz), and other parameters.

[0078] During the internal calculation of the model, complex nonlinear mapping will be performed according to the input data, for example, through a series of neuron calculations, the core calculation of which can be represented by the formula:

[0079]

[0080] where y represents the model output; x i represents the input data; w i represents the weight, b represents the bias, and f represents the activation function.

[0081] Finally, the model quickly outputs optimal IGBT drive slope and switching frequency and other control parameters, which will be used to adjust the working state of the IGBT in real time to achieve the best circuit performance and electromagnetic compatibility.

[0082] Further, as shown in Figure 3 step S3 includes:

[0083] Step S31, digital-to-analog conversion is performed on the control parameters output by the electromagnetic interference suppression model to obtain an analog signal;

[0084] Step S32, the rise or fall time and switching frequency of the drive signal are adjusted according to the analog signal to generate a drive signal.

[0085] Specifically, the control parameters output by the model are usually a series of discrete numerical values, which represent key information for adjusting the circuit system. In order for these digital parameters to function in traditional analog circuits, a 16-bit D / A converter is needed.

[0086] 16-bit D / A converter has high resolution, which can accurately convert digital signal to analog signal. Its conversion process is based on the principle of quantization, which maps the digital signal to the corresponding analog voltage value according to a certain quantization interval. The conversion formula is:

[0087]

[0088] where V out represents the output analog voltage; V ref represents the reference voltage; D represents the input 16-bit digital signal.

[0089] The digital control parameters output by the model are converted to analog signals by a 16-bit D / A converter, and the converted analog signals are sent to an IGBT drive chip, such as an IR2110 drive chip. In the IR2110 drive chip, the analog signal will act on the internal control circuit, and the rising and falling time of the drive signal will be adjusted by changing the charging and discharging time of the internal capacitor. The adjustment of the rising and falling time can effectively reduce the voltage and current change rate during the IGBT switching process, thereby reducing electromagnetic interference.

[0090] At the same time, the analog signal is sent to the PWM controller, and the PWM controller dynamically modifies the switching frequency according to the received analog signal, which can avoid strong electromagnetic radiation of the circuit at certain specific frequencies, thereby realizing real-time suppression of EMI.

[0091] Further, the electromagnetic interference suppression model can be a neural network model based on deep learning, such as a convolutional neural network (CNN) or a recurrent neural network (RNN).

[0092] Further, the generative adversarial network is any one or a combination of the following: a generative adversarial network based on a large language model, a deep convolutional generative adversarial network, and a generative adversarial network with gradient penalty.

[0093] Specifically, the generative adversarial network can be a large language model-based generative adversarial network, such as Qwen-2 GAN. Qwen-2 GAN is a deep learning model, and the Qwen-2 GAN model architecture constructs an adversarial training structure of a generator and a discriminator. In IGBT control circuit EMI suppression, the characteristics of the generative adversarial network are used. The generator takes the original signal of the IGBT control circuit (such as the drain-source voltage Vds, the collector current Ic, the drive voltage Vge, and the electromagnetic interference value) as the input. The generator generates an output as close as possible to the interference-free signal by learning a large number of interference-free signal samples, that is, the switching frequency, the drive voltage slope, and other control parameters. The discriminator judges whether the input signal is a real interference-free signal or a signal generated by the generator by comparing the real EMI data (150 kHz-1 GHz frequency band, 200 MHz sampling) with the simulated EMI data. The difference between the suppression effect and the target is evaluated using MSE as the loss function, and the weights are updated alternately to realize adversarial learning. Through the continuous confrontation and learning of the generator and the discriminator, the generator can gradually improve the ability to generate interference-free signals.

[0094] Further, a deep convolutional generative adversarial network (DCGAN) can be used instead of Qwen-2GAN. DCGAN replaces the fully connected layer of the traditional GAN with a convolutional layer to extract local mutation features in time series data (such as IGBT switching transient Vds peak and EMI peak). In application, the generator inputs real-time circuit data, extracts features through 3-4 layers of convolutional layers, and outputs switching frequency and drive voltage slope. The discriminator also uses a convolutional layer to compare the local differences between real and simulated EMI data, such as peak amplitude and duration to calculate the loss. The training process is consistent with Qwen-2 GAN, and the same precision control parameter output can be achieved.

[0095] Further, a WGAN-GP with gradient penalty can be used instead of Qwen-2GAN. WGAN-GP replaces MSE or JS divergence with Wasserstein distance to calculate loss, solving the problem of unstable training and easy to fall into local optimum of traditional GAN. In application, only the loss function of the discriminator is replaced from MSE to Wasserstein distance, and a gradient penalty term is added to ensure smooth training process; the generator still outputs IGBT control parameters, and the discriminator evaluates EMI suppression effect, which can avoid the parameter fluctuation problem of Qwen-2 GAN in EMI threshold harsh scene.

[0096] Further, LSTM (Long Short Term Memory) can be used instead of Qwen-2 GAN. LSTM captures the long-range dependence of time series data (such as the time correlation between IGBT switching frequency and EMI peak value) through a gating mechanism (input gate, forget gate, output gate), and the model structure is more compact and has better real-time performance than GAN. When applied, an LSTM feature extraction layer and a fully connected output layer are constructed, the input is the leakage source voltage Vds, the collector current Ic, the drive voltage Vge and the electromagnetic interference value of the past 10-20 sampling points, and the output is the switching frequency and the drive voltage slope at the next moment; without adversarial training, the model is optimized only through regression loss (such as MSE), which is suitable for scenarios with smaller sample size or higher real-time requirements, such as small and medium-sized industrial motor drive systems.

[0097] In another aspect, a power control circuit is also provided, which includes a power semiconductor device, and a drive signal is used to control the on-off state of the power semiconductor device. The drive signal is generated by the above-mentioned electromagnetic interference suppression method based on a generative adversarial network.

[0098] Specifically, the power control circuit generates a drive signal by using the above-mentioned electromagnetic interference suppression method based on a generative adversarial network to control the on-off state of the power semiconductor device. Through the generation and discrimination mechanism of the generative adversarial network, the electromagnetic interference signal in the insulated gate bipolar transistor (IGBT) control circuit is analyzed and processed, and an effective suppression strategy is generated to reduce the electromagnetic interference level.

[0099] In an industrial motor drive system, the IGBT control circuit frequently performs high-speed switching operations, which can easily generate strong electromagnetic interference, affecting the normal operation of surrounding electronic equipment. The method of the present application can effectively reduce such interference; in the power electronic system of a new energy vehicle, the IGBT control circuit is crucial for motor control, and suppressing EMI can ensure the stable operation of in-vehicle electronic equipment and the safety performance of the vehicle. The method can be widely applied to devices involving IGBT control circuits in the fields of power systems, industrial automation, new energy vehicles, etc., to improve the electromagnetic compatibility and operating performance of the devices.

[0100] The above is only the preferred embodiment of the present application, and does not limit the implementation and protection scope of the present application. For those skilled in the art, it should be realized that any equivalent replacement and obvious change made by applying the contents of the present application and drawings should be included in the protection scope of the present application.

Claims

1. An electromagnetic interference suppression method based on generative adversarial networks, characterized in that, Applications in power semiconductor devices, including: Step S1: Collect the voltage and current signals and electromagnetic interference signals of the power semiconductor device; Step S2: The collected voltage and current signals and electromagnetic interference signals are input into a pre-trained electromagnetic interference suppression model. The electromagnetic interference suppression model is built based on a generative adversarial network architecture. The generative adversarial network includes a generator and a discriminator. The generator is used to output control parameters for driving power semiconductor devices. The discriminator is used to evaluate the difference between the electromagnetic interference value corresponding to the control parameters output by the generator and the target interference value. Step S3: Convert the control parameters output by the electromagnetic interference suppression model to generate a drive signal for driving the power semiconductor device.

2. The electromagnetic interference suppression method based on generative adversarial networks according to claim 1, characterized in that, The training steps for the electromagnetic interference suppression model include: Step A1: Collect the voltage and current signals of the power semiconductor device during the switching process and the electromagnetic interference signals under the preset frequency band, and construct a training dataset; Step A2: Construct a generative adversarial network architecture, and train the generator and discriminator in the constructed generative adversarial network architecture according to the training dataset. Alternately update the model weights until the electromagnetic interference value corresponding to the output control parameter is reduced to the target interference threshold, and obtain the trained electromagnetic interference suppression model.

3. The electromagnetic interference suppression method based on generative adversarial networks according to claim 1, characterized in that, The preset frequency band is 150kHz to 1GHz.

4. The electromagnetic interference suppression method based on generative adversarial networks according to claim 1, characterized in that, The voltage and current signals of the power semiconductor device include drain-source voltage signal, collector current signal, and drive voltage signal.

5. The electromagnetic interference suppression method based on generative adversarial networks according to claim 1, characterized in that, Step S2 also includes: The acquired voltage and current signals and electromagnetic interference signals are preprocessed, and the preprocessed voltage and current signals and electromagnetic interference signals are input into a pre-trained electromagnetic interference suppression model to obtain control parameters for driving power semiconductor devices.

6. The electromagnetic interference suppression method based on generative adversarial networks according to claim 1, characterized in that, The control parameters include the driving voltage slope and switching frequency of the power semiconductor device.

7. The electromagnetic interference suppression method based on generative adversarial networks according to claim 1, characterized in that, Step S3 includes: Step S31: Convert the control parameters output by the electromagnetic interference suppression model from digital to analog to obtain an analog signal; Step S32: Adjust the rise or fall time and switching frequency of the drive signal according to the analog signal to generate the drive signal.

8. The electromagnetic interference suppression method based on generative adversarial networks according to claim 1, characterized in that, The generative adversarial network is any one or more combinations of generative adversarial networks based on large language models, deep convolutional generative adversarial networks, and generative adversarial networks with gradient penalties.

9. The electromagnetic interference suppression method based on generative adversarial networks according to claim 1, characterized in that, The power semiconductor device is an insulated gate bipolar transistor.

10. A power control circuit, characterized in that, The device includes a power semiconductor device, and the on / off state of the power semiconductor device is controlled by a driving signal, wherein the driving signal is generated using the electromagnetic interference suppression method based on generative adversarial networks as described in any one of claims 1-9.

Images