Modeling method for transient state of IGBT (Insulated Gate Bipolar Translator) switch of electric driving system of new energy vehicle

Abstract

The invention discloses a modeling method for an IGBT switch transient state of an electric drive system of a new energy vehicle. The modeling method comprises the steps that S1, original waveform data and simulation waveform data are collected; s2, calculating the original waveform data and the simulation waveform data by using a linear interpolation method to obtain pre-processed original waveform data and pre-processed simulation waveform data; s3, determining a training data set of a neural network; S4, establishing an artificial neural network model of a multi-layer perceptron structure, and determining the number of neurons of each layer according to the original waveform data and the simulation waveform data; s5, dividing the training data set into a training data set and a verification data set, performing training, and determining model parameters of optimal verification loss according to the verification data set; and S6, carrying out cycle training on each training time point, storing the weight and bias of the neural network corresponding to each training time point, and forming an IGBT switch transient model of the electric drive system. The transient waveform simulation device can accurately simulate the transient waveform when the IGBT is switched on and switched off, and is suitable for simulation of a large amount of data.

Description

Technical Field

[0001] The present invention relates to the field of electric vehicles, and in particular to a modeling method for transient states of IGBT switches in electric drive systems of new energy vehicles. Background Art

[0002] High-voltage insulated gate bipolar transistors (IGBTs) are widely used in the field of electrified transportation that requires high voltage and high current. In the electric drive system of new energy vehicles, the IGBT module is set between high voltage and low voltage to isolate the electromagnetic interference between high and low voltage. By setting appropriate IGBT parameters, the coupling effect between high and low voltage can be effectively reduced, the electromagnetic interference of each cable in the electric drive system can be reduced, and the performance of the electric drive system can be optimized. Due to the influence of dead time, switching time, delay time and voltage drop, the output voltage and current will be distorted relative to the ideal switching performance. The IGBT transient parameters are crucial for scenarios such as motor control. With the rapid development of new energy vehicles, higher requirements are placed on the reliable operation of high switching speed and high switching frequency of IGBTs used in the electric drive system of new energy vehicles. By establishing a high-precision IGBT switch transient modeling to obtain IGBT transient parameters, the switching speed and frequency of IGBTs in the electric drive system of new energy vehicles can be improved.

[0003] At present, IGBT transient models are mainly divided into physical models, analytical models and behavioral models. Physical models usually require detailed and accurate device manufacturing parameters, including size, material doping concentration, etc. Due to the involvement of manufacturer's commercial secrets, many parameters are difficult to obtain, so the model is less practical; analytical models use parameters extracted from test waveforms and solve high-order physical equations. Due to the high computational complexity of the equations, the simulation efficiency is low; behavioral models use voltage and current waveforms for parameter fitting and establish corresponding equivalent circuit models. The model parameters are not related to the internal structure of the device, but the accuracy of the model is low under different working conditions. The main challenge facing IGBT transient modeling is to achieve modeling with high precision and low computational complexity. Although the physical model can most accurately represent the detailed dynamic behavior of the IGBT, as the simulation scale increases, the physical model-based modeling is often more complex and the calculation speed is very slow, so it cannot be directly implemented in large-scale simulation.

[0004] The patent application with announcement number CN114330197B discloses a method for extracting IGBT numerical model parameters based on convolutional neural network. The numerical model is constructed using TCAD numerical simulation software, and the original data set is obtained to form a training data set. The IGBT numerical model parameter extraction convolutional neural network is generated through the convolutional neural network, and the waveform extracted from the actual circuit is used as the input value to obtain the process parameters. This method of extracting parameters through neural network modeling has a slow calculation speed and low efficiency. The electric drive system of new energy vehicles requires a large number of simulations or real-time simulations. The transient parameters of the IGBT switch obtained by this method are not accurate. If the electric drive system wants to obtain high-precision parameters, it will take a long time. This method is not suitable for accurate acquisition of electric drive systems of new energy vehicles. Summary of the invention

[0005] The purpose of the present invention is to address the deficiencies in the prior art and provide a modeling method for the transient state of IGBT switches in electric drive systems of new energy vehicles, which can accurately simulate the transient waveforms when the IGBT is turned on and off, is suitable for large amounts of data and real-time simulation, and solves the problems of slow calculation speed and low accuracy when the IGBT model is applied to computer simulation.

[0006] The purpose of the present invention is to adopt the following scheme to achieve: a modeling method for the transient state of IGBT switch of the electric drive system of new energy vehicles, which specifically includes the following steps:

[0007] S1: Establish the IGBT physical model of the electric drive system of new energy vehicles, and use LTspice to collect the temperature T, DC supply voltage Vcc and load current I of the IGBT under different working conditions L As the original waveform data, Python-Ltspice is used for joint simulation to obtain the simulation waveform data of collector-emitter voltage Vce and collector current Ic under different working conditions of IGBT;

[0008] S2: Determine a fixed time step, and use linear interpolation to calculate the temperature T, DC supply voltage Vcc and load current I corresponding to each time point under the fixed time step. L , obtain the pre-processed original waveform data; use the linear interpolation method to calculate the collector-emitter voltage Vce and collector current Ic corresponding to each time point under a fixed time step, and obtain the pre-processed simulation waveform data;

[0009] S3: Determine a training data set for the neural network, wherein the training data set includes an input sample data set and an output sample data set:

[0010] Divide the preprocessed raw waveform data into arrays to form an input sample data set;

[0011] Determine the fluctuation time range of the collector-emitter voltage Vce and the collector current Ic when the IGBT is turned on and off according to the preprocessed original waveform data and the preprocessed simulation waveform data, and determine the training time point and the number of training time points within the fluctuation time range according to the fixed time step in S2;

[0012] The collector-emitter voltage Vce and the collector current Ic at the same training time point in the preprocessed simulation waveform data are used as an output data set of the neural network model to form an output sample data set;

[0013] S4: Establish an artificial neural network model of a multilayer perceptron structure, wherein the input layer includes three neurons; the hidden layer includes a first hidden layer and a second hidden layer, the first hidden layer includes 10 to 30 neurons, and the second hidden layer includes 5 to 10 neurons; the output layer includes two neurons;

[0014] S5: normalizing the training data set and dividing it into a training data set and a validation data set, putting the training data set into the artificial neural network model for training, and determining the model parameters of the optimal validation loss of the artificial neural network model according to the validation data set and saving the model parameters;

[0015] S6: Perform cyclic training for each training time point, and save the weights and biases of the neural network corresponding to each training time point, obtain the neural network model parameters at each training time point, and form an IGBT switch transient model of the electric drive system.

[0016] Preferably, the three neurons in the input layer correspond to the temperature T, the DC supply voltage Vcc and the load current I respectively. L Three neurons; the two neurons in the output layer correspond to two output data sets, emitter voltage Vce and collector current Ic respectively.

[0017] Preferably, the number of neurons in the first hidden layer is 16, and the number of neurons in the second hidden layer is 8.

[0018] Preferably, each sample of the input sample data set is a temperature T, a DC supply voltage Vcc and a load current I under the same working condition. L .

[0019] Preferably, the ratio of the training data set to the validation data set is 9:1.

[0020] The advantages of the present invention are:

[0021] 1. The transient modeling method of IGBT switch in the electric drive system of new energy vehicles can greatly improve the simulation efficiency while ensuring the simulation accuracy. It can be directly applied to system-level simulation, avoiding the complex calculation process and non-convergence phenomenon of traditional circuit simulation software. Therefore, this method is suitable for situations where a large number of simulations or real-time simulations are required, meeting the needs of large-scale simulations or real-time simulations of electric drive systems of new energy vehicles;

[0022] 2. The IGBT switch transient model can realize the IGBT switch transient simulation on a nanosecond high-resolution time scale, and can accurately simulate the transient waveforms when the IGBT is turned on and off, and the calculation speed will not be very slow as the simulation scale increases;

[0023] 3. The IGBT switch transient modeling method based on neural network training can avoid the use of traditional experimental measurement methods, save time and money for model construction, and provide an alternative solution for extracting process parameters for some units that lack professional experimental equipment or conditions.

[0024] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is the principle diagram of the IGBT double pulse test simulation circuit of the present invention;

[0026] Figure 2 Schematic diagram of the gate voltage Vg waveform of the present invention;

[0027] Figure 3 It is a schematic diagram of the LTspice-Python joint simulation structure of the present invention;

[0028] Figure 4 It is the transient current waveform diagram of IGBT turn-on and turn-off under different working conditions of the present invention;

[0029] Figure 5 It is the transient voltage waveform diagram of IGBT turn-on and turn-off under different working conditions of the present invention;

[0030] Figure 6 It is a schematic diagram of the neural network structure of the present invention;

[0031] Figure 7 This is a comparison diagram of the prediction effect of the collector current Ic when the IGBT of the present invention is turned on;

[0032] Figure 8 This is a comparison diagram of the predicted effect of the collector-emitter voltage Vce when the IGBT of the present invention is turned on;

[0033] Fig. 9 This is a comparison diagram of the prediction effect of the collector current Ic when the IGBT of the present invention is turned off;

[0034] Fig.10 This is a comparison diagram of the prediction effect of the collector-emitter voltage Vce when the IGBT of the present invention is turned off;

[0035] Fig.11 is a flow chart of the present invention;

[0036] Fig.12 This is a schematic diagram of the IGBT model of the electric drive system of the present invention. DETAILED DESCRIPTION

[0037] See also Figures 1 to 12 A modeling method for transient state of IGBT switch of electric drive system of new energy vehicle, specifically comprising the following steps:

[0038] S1: Establish an IGBT physical model of the electric drive system of a new energy vehicle. The IGBT physical model is obtained from the IGBT manufacturer and imported into the circuit simulation software LTspice to establish an IGBT double-pulse simulation test circuit. The simulation test circuit is as follows: Figure 1 LTspice is used to collect the temperature T, DC supply voltage Vcc and load current I of the IGBT under different working conditions. L As the original waveform data; in the double pulse simulation test circuit, set the initial temperature value T to 20°C, the DC supply voltage Vcc to 400V, and the load current I L The gate resistance Rg is constant at 20Ω; the gate voltage Vg is constant as a square wave of one cycle, and the square wave waveform is as follows: Figure 2 As shown; the simulation time is 15μs, where the maximum simulation step in the circuit simulation software LTspice is set to 2ns;

[0039] According to the IGBT double pulse test circuit, a netlist source file is created, a Python-LTspice joint simulation script is established, and Python-Ltspice is used for joint simulation to obtain the simulation waveform data of the collector-emitter voltage Vce and collector current Ic under different working conditions of the IGBT. The specific implementation is as follows: first determine the working parameter range of the IGBT, as shown in Table 1. According to the working parameter range of the IGBT, the Python-LTspice joint simulation script automatically modifies the working parameters in the IGBT switch transient simulation netlist before each simulation. Then, the subprocess.call() statement is used to run the modified netlist source file, and then the Python script is used to read the .raw file generated by the LTspice software (for example, l=ltspice.Ltspice(file_path); l.parse()), and the original waveform data of the IGBT collector-emitter voltage Vce and collector current Ic at the corresponding time obtained by each operation are extracted. Among them, the 1365 IGBT operating conditions are based on the DC supply voltage Vcc, load current I in Table 1. L And the values ​​obtained by different combinations of the temperature T of the device IGBT. The LTspice-Python joint simulation flow chart is shown in Figure 3.

[0040] Table 1 IGBT operating parameter range

[0041]

[0042] S2: Determine a fixed time step according to the system performance of the new energy vehicle under electric drive. In this embodiment, the fixed time step is determined to be 5ns, and the original waveform data is calculated using linear interpolation to obtain the temperature T, DC power supply voltage Vcc and load current I corresponding to each time point under the fixed time step. L , get the pre-processed original waveform data; use the linear interpolation method to calculate the collector-emitter voltage Vce and collector current Ic corresponding to each time point under a fixed time step, get the pre-processed simulation waveform data, save the pre-processed original waveform data and pre-processed simulation waveform data in an Excel table, and form the following Figure 4 The waveform of the transient collector-emitter voltage when GBT is turned on and off under different working conditions is as follows: Figure 5 Waveforms of GBT turn-on and turn-off transient collector current under different working conditions;

[0043] S3: Determine a training data set for the neural network, wherein the training data set includes an input sample data set and an output sample data set:

[0044] The pre-processed raw waveform data is divided into arrays, and the temperature T, DC supply voltage Vcc and load current I under the same working condition are LDivide into the same array to form an input sample data set;

[0045] Determine the fluctuation time range of the collector-emitter voltage Vce and the collector current Ic when the IGBT is turned on and off according to the pre-processed original waveform data and the pre-processed simulation waveform data obtained in S2; determine the training time points and the number of training time points within the fluctuation time range with the fixed time step in S2 as the time interval;

[0046] according to Figure 4 As shown, it can be determined that the time range for the IGBT to generate fluctuations in the collector-emitter voltage Vce and collector current Ic when it is turned on is 5205ns to 5950ns (duration is 750ns), and the time step of 5ns in S2 is used as the time interval, and the number of training time points in the turn-on range is determined to be 150; the time range for the collector-emitter voltage Vce and collector current Ic to generate fluctuations when it is turned off is 10800ns to 13295ns (duration is 2500ns), and the time step of 5ns in S2 is used as the time interval, and the number of training time points in the turn-off range is determined to be 500;

[0047] The collector-emitter voltage Vce and collector current Ic at the same training time point in the preprocessed simulation waveform data are used as an output data set of the neural network model. For example, 10800ns is the first time point in the IGBT turn-on data set, and the values ​​of the collector-emitter voltage Vce and collector current Ic of 1365 IGBT operating conditions at 10800ns are used as the output data set of the first neural network training. The output data sets at different training time points form an output sample data set.

[0048] The number of training time points during the fluctuation time of the collector-emitter voltage Vce and collector current Ic when the IGBT is turned on and off is determined to be 150 and 500 respectively, so 150 neural network models are needed to predict the values ​​of the collector-emitter voltage Vce and collector current Ic at each time point when the IGBT is turned on, and 500 neural network models are needed to predict the values ​​of the collector-emitter voltage Vce and collector current Ic at each training time point when the IGBT is turned off. Among them, each neural network model has the same structure, but the weights and bias parameters obtained after model training are different.

[0049] S4: Establish an artificial neural network model with a multi-layer perceptron structure, in which the input layer includes three neurons, which correspond to the three characteristic temperatures T, DC supply voltage Vcc and load current I of the input data respectively. LThe hidden layer includes a first hidden layer and a second hidden layer. The first hidden layer includes 10 to 30 neurons, and the second hidden layer includes 5 to 10 neurons. In this embodiment, the number of neurons in the first hidden layer is 16, and the number of neurons in the second hidden layer is 8; the output layer includes two neurons, and the two neurons correspond to the emitter voltage Vce and the collector current Ic of the two output data sets respectively;

[0050] S5: After normalizing the input sample data set and the output sample data set in the training data set, randomly divide the training data set and the validation data set into a ratio of 9:1, put the training data set into the artificial neural network model for training, and determine the model parameters of the optimal validation loss of the artificial neural network model according to the validation data set and save the model parameters;

[0051] S6: Perform cyclic training for each training time point, and save the weights and biases of the neural network corresponding to each training time point, obtain the neural network model parameters at each training time point, and form an IGBT switch transient model of the electric drive system.

[0052] The obtained IGBT switch transient model of the electric drive system is used for model prediction. The model parameters at each training time point are used to predict the temperature of 35°C, the DC supply voltage Vcc of 455V and the load current I L are the Vce and Ic values ​​at the corresponding time points when the IGBT is turned on and off at 90A (i.e., the input data of the model);

[0053] For the model at each training time point, the trained neural network model is used to predict the normalized input data according to the input data to obtain the normalized output result, and the normalized output result is reversed back to the original data range. Then, the predicted data of the collector-emitter voltage Vce and collector current Ic corresponding to all training time points of the IGBT under this working condition are obtained. Finally, these data are combined into an output to obtain the turn-on and turn-off transient waveforms of the IGBT.

[0054] The IGBT switch transient waveform obtained by LTspice software simulation is compared with the IGBT switch transient waveform obtained by the electric drive system IGBT switch transient model in this embodiment. Figure 7-10 As shown in the figure, it can be seen from the comparison that the waveform predicted by the transient model of the IGBT switch of the electric drive system is highly consistent with the waveform obtained by the LTspice software, and the simulation time is shortened by half. In order to more intuitively evaluate the accuracy of the model, the relative root mean square error is used to evaluate the model. Table 2 shows the relative root mean square error of the current and voltage under the IGBT switching state.

[0055] Table 2 Relative RMS error of current and voltage in IGBT switching transient state

[0056] IGBT parameters Relative root mean square error (%) Turn-on current 0.56 Turn-on voltage 1.04 Shutdown current 1.05 Shutoff voltage 0.42

[0057] Based on the established IGBT switch transient model of the electric drive system, the following Fig.12 The IGBT model of the electric drive system shown in the figure analyzes the impact of IGBT on the new energy electric drive system model, and reduces the impact of high and low cables and IGBT in the electric drive system on the entire electric drive system.

[0058] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modification made to the present invention by those skilled in the art without departing from the spirit of the present invention shall fall within the protection scope of the present invention.

Claims

1. A modeling method for transient state of IGBT switch in electric drive system of new energy vehicle, characterized in that: The specific steps include: S1: Establish the IGBT physical model of the electric drive system of new energy vehicles, and use LTspice to collect the temperature T, DC supply voltage Vcc and load current I of the IGBT under different working conditions L As the original waveform data, Python-Ltspice is used for joint simulation to obtain the simulation waveform data of collector-emitter voltage Vce and collector current Ic under different working conditions of IGBT; S2: Determine a fixed time step, and use linear interpolation to calculate the temperature T, DC supply voltage Vcc and load current I corresponding to each time point under the fixed time step. L , get the preprocessed raw waveform data; The simulated waveform data is calculated using a linear interpolation method to obtain the collector-emitter voltage Vce and collector current Ic corresponding to each time point under a fixed time step, and the pre-processed simulated waveform data is obtained; S3: Determine a training data set for the neural network, wherein the training data set includes an input sample data set and an output sample data set: Divide the preprocessed raw waveform data into arrays to form an input sample data set; Determine the fluctuation time range of the collector-emitter voltage Vce and the collector current Ic when the IGBT is turned on and off according to the preprocessed original waveform data and the preprocessed simulation waveform data, and determine the training time point and the number of training time points within the fluctuation time range according to the fixed time step in S2; The collector-emitter voltage Vce and the collector current Ic at the same training time point in the preprocessed simulation waveform data are used as an output data set of the neural network model to form an output sample data set; S4: Establish an artificial neural network model of a multilayer perceptron structure, wherein the input layer includes three neurons; the hidden layer includes a first hidden layer and a second hidden layer, the first hidden layer includes 10 to 30 neurons, and the second hidden layer includes 5 to 10 neurons; the output layer includes two neurons; S5: normalizing the training data set and dividing it into a training data set and a validation data set, putting the training data set into the artificial neural network model for training, and determining the model parameters of the optimal validation loss of the artificial neural network model according to the validation data set and saving the model parameters; S6: Perform cyclic training for each training time point, and save the weights and biases of the neural network corresponding to each training time point, obtain the neural network model parameters at each training time point, and form an IGBT switch transient model of the electric drive system.

2. The modeling method for IGBT switch transient state of electric drive system of new energy vehicle according to claim 1 is characterized in that: The three neurons in the input layer correspond to temperature T, DC supply voltage Vcc and load current I respectively. L Three neurons; the two neurons in the output layer correspond to two output data sets, emitter voltage Vce and collector current Ic respectively.

3. The modeling method for IGBT switch transient state of electric drive system of new energy vehicle according to claim 1 is characterized in that: The number of neurons in the first hidden layer is 16, and the number of neurons in the second hidden layer is 8.

4. The modeling method for transient state of IGBT switch of electric drive system of new energy vehicle according to claim 1 is characterized in that: Each sample of the input sample data set is a temperature T, a DC supply voltage Vcc and a load current I under the same working condition. L .

5. The modeling method for transient state of IGBT switch of electric drive system of new energy vehicle according to claim 1, characterized in that: The ratio of the training dataset to the validation dataset is shown to be 9:1.

Images