Power semiconductor transient simulation precision optimization method and system based on machine learning

Abstract

The invention belongs to the technical field of power electronic simulation, and discloses a power semiconductor transient simulation precision optimization method and system based on machine learning. According to the method, mapping between a low-precision waveform and a high-precision waveform is established by using a lightweight convolutional neural network, the low-precision waveform is processed by the neural network and then is subjected to linear interpolation, and the high-precision waveform with higher accuracy compared with direct iterative calculation is obtained with lower calculation resources and time cost. Taking an IGBT device-level simulation model as an example, the transient simulation precision optimization method is successfully applied to remarkably improve the precision of the switching transient waveform. The method effectively solves the problem that the real-time simulation precision is difficult to further improve due to the contradiction between simulation precision and computing resource consumption and simulation time and the contradiction between precision and accuracy, and provides reliable and efficient technical support for improving the precision of power electronic simulation. And a new thought is developed for improving the device-level real-time simulation accuracy.

Description

Technical Field

[0001] The present invention belongs to but is not limited to the field of power electronics simulation technology, and in particular relates to a method and system for optimizing power semiconductor transient simulation accuracy based on machine learning. Background Art

[0002] With the rapid development of power electronics technology, the operating frequency of power semiconductor devices continues to increase. The switching transient process of high-frequency switching devices represented by SicMOSFET is extremely short, and even the rapid voltage and current changes are at the nanosecond level. This makes it an urgent core technical requirement in power electronics simulation to accurately describe the transient characteristics of high-frequency switching devices (such as oscillation, overshoot, spikes and other non-ideal phenomena) and accurately predict switching losses.

[0003] The simulation models of power semiconductor devices can be divided into two categories according to their precision and complexity: system level and device level. System level models are represented by average models, ideal switch models, binary resistance models, constant admittance L / C models, etc., which are characterized by focusing only on the influence of the steady-state switching of power semiconductor devices on the circuit, while ignoring the switching transient process. In contrast, device level models can describe the transient characteristics of voltage and current during the switching transient process, and can be further divided into physical models and behavioral models. The physical model mainly describes the distribution characteristics and electrical characteristics of carriers based on semiconductor physics equations, which usually involve the solution of partial differential equations or nonlinear differential equations, have high computational complexity, and the parameters involve the device manufacturing process, and the extraction process is very complicated. The behavioral model does not involve complex physical mechanisms. It uses macro models and empirical formulas to describe the switching characteristics of the device and the parameters are easier to extract. It is widely used in simulation software such as SaberRD, PSIM, and PLECS. However, due to the nonlinear voltage-current relationship of power semiconductors, the numerical solution of the behavioral model requires iterative calculation and the convergence is general.

[0004] Both physical models and behavioral models are based on numerical solution methods, so they face some common challenges in simulation. Since high-frequency switching devices have high switching frequencies and low on-state resistance, the non-ideal characteristics of their transient processes are more significant. In order to accurately solve the transient waveforms of high-frequency devices, it is necessary to establish a high-precision device-level simulation model and solve it with a sufficiently low simulation step size (one order of magnitude smaller than the transient time scale). However, doing so will inevitably lead to an increase in computing resource consumption and solution time, resulting in a contradiction between accuracy and computing resources and simulation time.

[0005] Based on the development of power electronic device-level simulation in the past five years, there are mainly the following improvement strategies to reduce resource consumption and solution time. On the one hand, the power semiconductor model is simplified and approximated to establish a simulation model with lower complexity and higher accuracy, but this often comes at the expense of accuracy. Especially when describing the fast transient characteristics of high-frequency switching devices (such as oscillation, overshoot, spikes, etc.), the simplified model may not be able to accurately capture these non-ideal phenomena. On the other hand, the simulation solution strategy is improved, such as using piecewise linearization methods to avoid iterative solutions, saving computing time costs and resource occupation, but this method is ideal, its accuracy depends on the selection of segmentation points, and may also lead to the loss of key transient characteristics. For real-time simulation, through multi-level parallel strategies, the critical path within a single calculation step of converter simulation is greatly shortened to speed up the simulation speed, but parallelization will sacrifice certain solution accuracy and numerical stability, and will occupy more computing resources. With the increasingly complex structure and higher switching frequency of power electronic systems, it is difficult to ensure and improve the accuracy of solutions by using parallelization methods. It can be seen that electronic device-level simulation also faces the contradiction between precision and accuracy.

[0006] For power electronic converters, in order to accurately obtain the transient waveforms of the high-frequency switching devices, the entire circuit must be solved with a sufficiently low step size, which brings additional computing resources and computing time consumption, seriously reduces simulation efficiency, and slows down the design process.

[0007] With the development of machine learning technology, some studies have used neural networks to approximate the nonlinear relationship between simulation input and output, and then called the neural network to obtain the transient voltage or current of the switching device. However, this method of treating the converter model as a "black box" and directly mapping the output transient waveform through input parameters has the following problems: (1) The generalization ability over a wide range of operating conditions is difficult to meet the simulation requirements; (2) The neural network is very complex, occupies a lot of resources, and takes a long time to calculate, which limits its application in simulation, especially real-time simulation; In view of the above analysis, the technical problems existing in the existing power electronic simulation technology are summarized as follows: 1. The contradiction between simulation accuracy and computing resource consumption and simulation time, especially in the simulation of fast transient processes of high-frequency switching devices. This contradiction is particularly prominent. 2. The contradiction between simulation accuracy and simulation precision. 3. It is difficult to ensure and further improve the accuracy of real-time simulation transient waveforms with existing methods. 4. In order to obtain the transient waveforms of high-frequency devices, the overall simulation efficiency of power electronic converters is reduced. 5. The existing machine learning-based methods mainly construct the mapping relationship between simulation input parameters and output transient waveforms, which have high network complexity, long calculation time, and poor generalization ability.

[0008] In summary, the present invention proposes a method for improving the accuracy of power semiconductor transient simulation waveforms at a relatively low cost, which is crucial for the development of power electronics simulation and power electronics technology. Summary of the invention

[0009] In view of the problems existing in the prior art, the present invention provides a method for optimizing the accuracy of power semiconductor transient simulation based on machine learning.

[0010] The present invention is implemented as follows: a method for optimizing the transient simulation accuracy of a power semiconductor based on machine learning, the method comprising: S1, analysis and design of power semiconductor switch transient waveform simulation accuracy optimization strategy; S2, design machine learning neural network structure; S3, get the dataset; S4, training the network, optimizing the network structure and training parameters, and saving the best network model; S5, processes low-precision transient waveforms based on the trained optimal network model to improve the accuracy of transient waveforms; S6, designs a simulation accuracy optimization unit for power electronic converter simulation.

[0011] Furthermore, the transient waveform accuracy optimization strategy described in S1 is inspired by the application of neural networks in the field of image super-resolution. With the help of neural networks, low-precision waveforms are reconstructed in super-resolution to obtain high-precision waveforms. Low-precision simulation consumes less computing resources and has a short simulation time, thereby avoiding further iteration resource consumption and solution time. Therefore, it is necessary to fully consider resource and time consumption so that the cost is as small as possible as compared to direct iterative solution, which is also convenient for the subsequent use of this method in various embedded devices.

[0012] Furthermore, the neural network model structure designed in S2 is constructed as follows: it has the characteristics of flexibility and simplicity, in which a convolutional network is used to improve the applicability to different inputs, and the storage resource consumption is reduced by reusing convolutional layers with the same structure. The computing resources can be sacrificed to complete the convolution operations of each layer in parallel to improve the simulation speed and reduce the simulation solution time.

[0013] Furthermore, the data set in S3 can be derived from data provided by the model manufacturer, or the model can be directly simulated under different working conditions to obtain low-precision waveforms and high-precision waveforms to construct a data set. The data set can be reasonably divided into a training set, a validation set, and a test set according to a certain ratio to enhance the generalization of the model.

[0014] Furthermore, in S4, an optimal mapping model from a low-precision waveform to a high-precision waveform is established through network training.

[0015] Furthermore, the step of improving the accuracy of the transient waveform in S5 does not sacrifice the accuracy of the waveform. For the simulation method using iterative solution, since the numerical accuracy of the current moment depends on the numerical accuracy of the previous moment, the step can improve the accuracy and precision of the waveform at the same time.

[0016] Furthermore, based on the transient waveform accuracy optimization method in S1 to S5, S6 proposes a simulation transient accuracy optimization unit structure that can be applied to various embedded device processors, providing theoretical support for the subsequent application of the method proposed in the present invention to power electronics real-time simulation practice.

[0017] Another object of the present invention is to provide a power semiconductor transient simulation accuracy optimization system based on machine learning based on the power semiconductor transient simulation accuracy optimization method based on machine learning, the system comprising: The simulation solver solves the circuit to obtain the low-precision transient waveform of the high-frequency power semiconductor device as the input of the simulation transient accuracy optimization unit.

[0018] The simulation accuracy optimization unit mainly includes four parts: data reading, caching, neural network and interpolation; Furthermore, the neural network is composed of three parts: a multiplication array, an addition tree, and an activation function. The convolution operation requires the use of more multipliers, so the multipliers can be reasonably reused between and within each layer to save computing resources. For this purpose, the network weights and biases trained on the computer are loaded into the storage unit of the processor. For the network layers of the same structure, only one hardware resource is needed, and the corresponding weights and biases are called when performing the calculation. Of course, more hardware resources can be used in parallel in exchange for less computing time. The controller is responsible for controlling the operating logic and timing of each sub-module, and plays a decisive role in whether the calculation process can be completed.

[0019] Transient waveform output: output optimized high-precision power semiconductor simulation transient waveform.

[0020] Another object of the present invention is to provide a real-time simulation of power electronics, which replaces the further iteration from low precision to high precision in high-precision simulation calculations with a flexible, simple, and time-saving neural network model, thereby reducing computing resource consumption and speeding up simulation speed, making it possible to apply the steps of the power semiconductor transient simulation accuracy optimization method based on machine learning to real-time simulation of power electronics and improve the transient accuracy of real-time simulation.

[0021] Another object of the present invention is to provide a method for improving the efficiency of offline simulation of a high-frequency converter system, characterized in that the method uses a transient waveform data set of a high-frequency semiconductor switching device constructed through a one-time multi-operating condition simulation to train a neural network model with high generalization, so that in subsequent simulations, only a low-precision step size is required to complete the simulation of the entire converter system. The high-precision transient waveform of the high-frequency semiconductor switching device can be obtained separately through the network model, thereby improving the simulation efficiency.

[0022] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows: First, the method for improving the accuracy of transient waveforms of power semiconductor devices based on machine learning disclosed in the present invention fully considers the defects of existing technical problems and existing solutions, conforms to the development trend of the times, adopts artificial intelligence machine learning methods, and creatively proposes the idea of ​​using neural networks to establish a low-precision transient waveform to a high-precision transient waveform mapping model, designs a simple accuracy optimization strategy, and replaces the process of further iteration of low-precision waveforms to high-precision waveforms with lightweight neural network models, thereby reducing computing resource consumption and solution time, and improving the accuracy of transient waveforms. In the embodiment, only an average network training time of 510s is required. Compared with the original low-precision waveform, the average MSE error of the IGBT switch transient waveform after accuracy optimization is reduced by more than 85%, the average RRMSE error is reduced by more than 64%, the average energy error of the turn-on process is reduced by more than 85%, and the average energy error of the turn-off process is reduced by more than 93%, with significant results.

[0023] Second, as auxiliary evidence of the inventiveness of the claims of the present invention, it is also reflected in the following important aspects: (1) The technical solution of the present invention solves a technical problem that people have been eager to solve but have never been able to solve successfully: The contradiction between accuracy and computing resources and simulation time consumption in power electronic simulation, and the contradiction between precision and accuracy. The method proposed in the present invention consumes less computing resources, improves the accuracy of transient waveforms, and reduces the time required for simulation solutions.

[0024] In order to accurately capture the transient process of high-frequency switching devices in power electronic converters, the entire system may need to be solved with a lower step size, which reduces the simulation efficiency. The method proposed in the present invention allows the power electronic converter to be solved only with a low-precision step size, and then the transient waveform accuracy of the high-frequency switching device is improved through a trained neural network model, avoiding the high-precision solution of the entire system and improving the simulation efficiency of the power electronic converter system.

[0025] Due to resource and time constraints, it is difficult for existing parallelization schemes to further improve the real-time simulation accuracy of high-frequency switching devices. The lightweight neural network model structure designed by the present invention and the waveform accuracy optimization unit that can be used for various embedded devices can replace the complex iterative process from low-precision to high-precision waveforms, and have great real-time simulation potential.

[0026] The existing methods of using neural networks to assist power electronics simulation cannot directly address the pain points. The network is complex, occupies a lot of resources, has weak generalization ability, and takes a long time to calculate. Inspired by the deep learning super-resolution technology in the image field, the present invention proposes the idea of ​​using neural networks to establish a mapping of low-precision transient waveforms to high-precision transient waveforms. Since some structures of the designed neural network can be reused and some calculations can be parallelized, the accuracy of power semiconductor transient simulation waveforms can be improved at a lower cost of resources and time, and the simulation solution time can be reduced. In the application of super-resolution technology for two-dimensional data processing, the introduction of machine learning technology has effectively improved the accuracy.

[0027] The transient waveform accuracy optimization unit designed in the present invention can be applied to various embedded devices, providing a new technical support for power electronics real-time simulation, and providing theoretical support for applying lightweight neural networks to real-time simulation to further improve the accuracy of device-level real-time simulation.

[0028] (3) The expected benefits and commercial value of the technical solution of the present invention after transformation are: Applicable to a variety of power semiconductor devices: The present invention is not only applicable to high-frequency switching devices such as SiC MOSFET and IGBT, but can also be extended to the simulation optimization of other power semiconductor devices.

[0029] Reduce simulation costs. By reducing the computing resources and time required for high-precision simulation, the simulation costs are significantly reduced, which is especially suitable for the design and optimization of large-scale power electronic systems.

[0030] Improve design efficiency: The present invention can quickly and accurately obtain the transient waveform of high-frequency power semiconductor devices, shorten the research and development cycle of power electronic systems, and improve design efficiency.

[0031] In summary, the present invention successfully solves the contradiction between the two teams in power electronic simulation technology through innovative technical solutions, and has significant technical advantages, economic benefits and social benefits. The present invention not only fills the technical gaps at home and abroad, but also provides new ideas and methods for the development of power electronic simulation technology, and has broad application prospects and important promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1It is a flow chart of a method for optimizing power semiconductor transient simulation accuracy based on machine learning provided by an embodiment of the present invention; Figure 2 It is a structural diagram of a power semiconductor transient simulation accuracy optimization system based on machine learning provided by an embodiment of the present invention; Figure 3 It is a schematic diagram of transient waveforms of IGBT switches under different working conditions with low precision provided by an embodiment of the present invention; Figure 4 It is a schematic diagram comparing a low-precision simulated current waveform, a high-precision simulated waveform, and a sampled high-precision simulated waveform during an IGBT turn-on process under a certain working condition provided by an embodiment of the present invention; Figure 5 It is a schematic diagram comparing a low-precision simulation current waveform, a high-precision simulation waveform, and a sampled high-precision simulation waveform during an IGBT turn-off process under a certain working condition provided by an embodiment of the present invention; Figure 6 It is a schematic diagram comparing a low-precision simulation voltage waveform, a high-precision simulation waveform, and a sampled high-precision simulation waveform during an IGBT turn-on process under a certain working condition provided by an embodiment of the present invention; Figure 7 It is a schematic diagram comparing a low-precision simulation voltage waveform, a high-precision simulation waveform, and a sampled high-precision simulation waveform during an IGBT turn-off process under a certain working condition provided by an embodiment of the present invention; Figure 8 Schematic diagram of the point-to-point mapping convolutional neural network structure provided by an embodiment of the present invention; Fig. 9 It is a schematic diagram of the training process of the current model of the turn-on process provided by an embodiment of the present invention; Fig.10 It is a schematic diagram of the training process of the current model of the shutdown process provided by an embodiment of the present invention; Fig.11 It is a schematic diagram of a turn-on process voltage model training process provided by an embodiment of the present invention; Fig.12 It is a schematic diagram of a voltage model training process for a shutdown process provided by an embodiment of the present invention; Fig.13 It is a comparative schematic diagram of a low-precision current waveform, a high-precision current waveform, and a current waveform after precision optimization during a turn-on process provided by an embodiment of the present invention; Fig.14 It is a comparative schematic diagram of a low-precision current waveform, a high-precision current waveform, and a current waveform after precision optimization during a shutdown process provided by an embodiment of the present invention; Fig.15 It is a schematic diagram comparing a low-precision voltage waveform, a high-precision voltage waveform, and a voltage waveform after precision optimization during a turn-on process provided by an embodiment of the present invention; Fig.16 Schematic diagram of comparison of low-precision voltage waveform, high-precision voltage waveform, and voltage waveform after precision optimization during the turn-on process provided by the embodiment of the present invention Fig.17 It is a schematic diagram comparing a network output waveform and a high-precision waveform of a current waveform in a peak region of a turn-on process under certain working conditions provided by an embodiment of the present invention; Fig.18 Schematic diagram of the structure of an upsampling convolutional neural network provided by an embodiment of the present invention; Fig.19 It is a schematic diagram comparing a network output waveform and a high-precision waveform of a current waveform in a peak region of a turn-on process under certain working conditions provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0034] like Figure 1 As shown, an embodiment of the present invention provides a method for optimizing transient simulation accuracy of a power semiconductor based on machine learning, the method comprising: S1, analysis and design of power semiconductor switch transient waveform accuracy optimization strategy; S2, design machine learning neural network structure; S3, get the dataset; S4, training the network, optimizing the network structure and training parameters, and saving the best network model; S5, processes low-precision transient waveforms based on the trained optimal network model to improve the accuracy of transient waveforms; S6, designs a simulation accuracy optimization unit for power electronic converter simulation.

[0035] This method first analyzes and designs the simulation accuracy optimization strategy for the problem of insufficient accuracy in the transient waveform simulation process of power semiconductor switches (S1). By analyzing the existing methods for directly improving waveform accuracy, a precision optimization strategy combining direct interpolation and machine learning is proposed, ensuring that the method proposed in this paper can improve the accuracy of the waveform more accurately at a lower cost. This step provides design and optimization direction for subsequent machine learning models.

[0036] In S2, a machine learning neural network structure suitable for processing power semiconductor transient waveform data is designed. This network structure takes into account the timing and nonlinear characteristics of waveform data, and may adopt architectures such as convolutional neural network (CNN) or recurrent neural network (RNN), combined with appropriate activation functions and loss functions. Subsequently, in the S3 stage, a data set is constructed by collecting a large amount of low-precision and high-precision transient simulation data to ensure that the training data is sufficiently representative and diverse so that the network can accurately predict waveform details under various working conditions.

[0037] In S4, the obtained data set is used to train the designed neural network, and the network structure and training parameters (such as learning rate, batch size, number of iterations, etc.) are continuously adjusted to minimize the preset loss function and optimize the model performance. During the training process, the best network model is determined through cross-validation, loss curve monitoring, etc., and saved for subsequent use. This process ensures that the model has good generalization ability and high-precision waveform prediction ability, thereby providing strong support for the refined correction of low-precision simulation results.

[0038] In the S5 stage, the trained optimal network model is used to process the low-precision transient waveform, and the accuracy of the waveform details is improved by correcting and compensating the original simulation results. Finally, in S6, a simulation accuracy optimization unit for power electronic converter simulation is designed. This unit integrates the machine learning model into the overall simulation process to achieve dynamic optimization of waveform accuracy in different simulation scenarios. The overall system improves the simulation accuracy through intelligent algorithms, significantly improving the limitations of traditional simulation methods in high-dynamic, nonlinear transient processes, thereby providing more accurate data support for the design and optimization of power electronic equipment.

[0039] The transient waveform accuracy optimization strategy described in S1 is inspired by the application of neural networks in the field of image super-resolution. With the help of neural networks, low-precision waveforms are reconstructed in super-resolution to obtain high-precision waveforms. Low-precision simulation consumes less computing resources and has a short simulation time, thereby avoiding further iteration resource consumption and solution time. Therefore, it is necessary to fully consider resource and time consumption so that the cost is as small as possible as compared to direct iterative solution, which is also convenient for the subsequent use of this method in various embedded devices.

[0040] The neural network model structure designed in S2 should be as lightweight, flexible and simple as possible, with some structures being reusable and some calculations being parallelized, so as to reduce resource consumption and increase simulation speed.

[0041] The data set in S3 can be derived from data provided by the model manufacturer, or the model can be directly simulated under different working conditions to obtain low-precision waveforms and high-precision waveforms to construct a data set. The data set can be reasonably divided into a training set, a validation set, and a test set according to a certain ratio to enhance the generalization of the model.

[0042] In S4, an optimal mapping model from a low-precision waveform to a high-precision waveform is established through network training.

[0043] The step of improving the accuracy of the transient waveform in S5 does not sacrifice the accuracy of the waveform. For the simulation method using iterative solution, since the numerical accuracy of the current moment depends on the numerical accuracy of the previous moment, this step can improve the accuracy and precision of the waveform at the same time.

[0044] In S6, based on the transient simulation accuracy optimization method in S1 to S5, a transient simulation accuracy optimization unit structure that can be applied to various embedded device processors is proposed, which provides theoretical support for the subsequent application of the method proposed in the present invention to power electronics real-time simulation practice.

[0045] 1. Transient waveform simulation accuracy optimization strategy There are two general methods to improve waveform accuracy: direct interpolation and curve fitting. Considering that transient processes have non-linear, discontinuous and other non-ideal characteristics, it is difficult to guarantee the accuracy of the waveform using direct interpolation, and it is difficult to find a suitable time function using curve fitting. Machine learning neural networks are very good at describing the mapping relationship between low-precision data points and high-precision data points, but establishing an accurate mapping from one point to multiple points requires a fairly complex network structure, and the training time is very long, which may be more expensive than direct iterative solutions. Therefore, it is considered to combine a simple direct interpolation method with a neural network with a strong ability to represent nonlinear relationships.

[0046] As analyzed above, for the device model using iterative solution, the low-precision waveform is often not very accurate. Assuming that the low-precision simulation step is h1 and the high-precision simulation step is h2, the neural network is used to establish a point-to-point mapping between the low-precision waveform data points and the high-precision waveform at the low-precision sampling points (that is, the high-precision waveform is sampled according to the h1 step). Therefore, after the low-precision waveform passes through the network, a high-precision waveform with a step length of h1 is obtained, and then a high-precision waveform with a step length of h2 is obtained through linear interpolation, thereby improving both "accuracy" and "precision". However, doing so is likely to miss the key points in the rapidly changing area of ​​the waveform, such as the peak point, resulting in a large error in the waveform after interpolation. Since we do not want to introduce a feature extraction unit to increase the complexity of this method, for these areas, the high-precision waveform sampling step is reduced, such as h1 / 2, and then an upsampling layer is added to the neural network to establish a mapping from n points to 2n-1 points. In theory, as long as the step length is small enough, all key points can be mapped, but doing so will further increase the complexity of the network.

[0047] 2 Neural Network Model Structure The transient waveforms of high-frequency switching devices usually have distinct characteristics. For example, the current waveform of the IGBT turn-on process always rises rapidly first, then gradually falls back to a stable value, and spikes or oscillations may occur during this period. Convolutional neural networks (CNNs) gradually learn high-level features from low-level features through multi-layer convolution and pooling operations. This hierarchical feature learning ability enables it to capture complex patterns in waveforms. At the same time, compared with other neural networks, the convolution kernels of CNN share parameters on the input data, which greatly reduces the number of model parameters and reduces the computational burden. Its lightweight characteristics make it suitable for deployment on embedded devices, which lays an important foundation for applying this method to real-time simulation.

[0048] Another advantage of CNN is that 1D convolution has no fixed requirement for the length of input data, and the length of output data is determined by the convolution parameters, which allows the same model to be used for processing different numbers of sampling points and different regions, and can also adjust the mapping of single point to single point or single point to multiple points by adjusting the parameters. Therefore, the neural network is composed of multiple layers of convolutional networks, and the upsampling network only needs to adjust the parameters of one or several layers. The specific network depth and convolution parameters need to be continuously debugged in specific tasks to find the optimal solution.

[0049] 3 Get the dataset In order to enhance the generalization of the model, offline simulation software is used to obtain low-precision and high-precision transient waveforms under various working conditions of the model and complete the sampling of high-precision waveforms. The working conditions need to cover the complete working range, and the data set is randomly divided into training set, validation set and test set.

[0050] 4 Network Training Training parameters such as learning rate, optimizer, number of training rounds, etc. need to be constantly debugged like network parameters. An early stopping strategy is adopted during the training process, that is, the training process is terminated early after the loss has not been significantly improved for a certain number of consecutive rounds. The use of the early stopping strategy can greatly save training time and computing resource consumption.

[0051] 5. Optimization of transient waveform simulation accuracy The accuracy of low-precision waveforms in the test set is optimized, and the optimized high-precision waveforms are analyzed to adjust the neural network structure parameters or neural network training parameters.

[0052] 6 Simulation Accuracy Optimization Unit Design The lightweight network makes this method easily applicable to processors of various embedded devices. The simulation solver obtains a low-precision transient waveform as the input of the optimization unit, and a high-precision waveform is obtained through calculation.

[0053] The hardware circuit of the simulation accuracy optimization unit mainly includes four parts: data reading, cache, neural network and controller. Among them, the neural network consists of three parts: multiplication array, addition tree and activation function. The convolution operation requires the use of more multipliers, so the multipliers can be reasonably reused between and within each layer to save computing resources. For this reason, the network weights and biases trained on the computer are loaded into the storage unit of the processor. For the network layer of the same structure, only one hardware resource is needed, and the corresponding weights and biases are called when performing calculations. Of course, more hardware resources can be used in parallel in exchange for less computing time.

[0054] The controller is the most important part of this unit. It is responsible for controlling the operating logic and timing of each sub-module, and plays a decisive role in whether the calculation process can be completed.

[0055] Specific application fields or related products of the present invention: like Figure 2 As shown, an embodiment of the present invention provides a power semiconductor transient simulation accuracy optimization system based on machine learning based on the power semiconductor transient simulation accuracy optimization method based on machine learning, and the system includes: The simulation solver solves the circuit to obtain the low-precision transient waveform of the high-frequency power semiconductor device as the input of the simulation transient accuracy optimization unit.

[0056] The simulation accuracy optimization unit mainly includes four parts: data reading, caching, neural network and interpolation; The controller is responsible for controlling the operating logic and timing of each module, and plays a decisive role in whether the calculation process can be completed.

[0057] Transient waveform output: output optimized high-precision power semiconductor simulation transient waveform.

[0058] An embodiment of the present invention provides a power electronics simulation, which replaces the further iteration from low precision to high precision in high-precision simulation calculations with a flexible, simple, and time-saving neural network model, thereby reducing computing resource consumption and accelerating simulation speed, making it possible to apply the steps of the power semiconductor transient simulation accuracy optimization method based on machine learning to real-time power electronics simulation, thereby improving the transient accuracy of real-time simulation.

[0059] Evidence related to the technical effects obtained by the embodiments of the present invention: The present invention takes a device-level IGBT simulation model as an example to introduce the whole process of optimizing the accuracy of the IGBT switch transient simulation waveform.

[0060] In the embodiment, the original data set is directly obtained through the offline power electronics simulation software. The switching transient characteristics of the IGBT are mainly related to the circuit input voltage and input current. The input voltage VCC range is 400-1200V, and the input current IL range is 200-1000A. In order to make the trained model have good generalization, the training set data must cover the complete working range. The voltage step is set to 16V and the current step is set to 16A. There are a total of 2601 working combinations. The low-precision transient simulation step h1 is set to 50ns, and the high-precision transient simulation step h2 is set to 5ns. Therefore, the number of data points of the high-precision waveform is ten times the number of low-precision data points. Run the simulation to obtain the switching transient waveforms under two accuracies under different working conditions. Figure 3 The low-precision IGBT voltage and current turn-on and turn-off transient waveforms under different working conditions are shown.

[0061] The high-precision waveform is sampled according to the step length h1 of the low-precision waveform to obtain a sampled high-precision waveform. At this time, the number of data points of the sampled high-precision waveform is the same as the number of data points of the low-precision waveform. Figure 4 , Figure 5 , Figure 6 , Figure 7 The following are the comparison diagrams of low-precision waveforms, original high-precision waveforms, and high-precision waveforms after sampling of four transient waveforms under the working conditions of VCC=400V and IL=200A. The following conclusions can be drawn: 1. Since the model is solved in an iterative manner, the accuracy of the low-precision waveform is also very poor. 2. The high-precision waveform after sampling is very close to the original high-precision waveform after linear interpolation, and the accuracy error is very small. Only for some areas with large slope changes, such as Figure 4The peak of the waveform has a relatively large error in the peak area because there are not enough sampling points in the peak area. Therefore, the sample points of the low-precision waveform are accurately mapped directly to the sample points of the high-precision waveform after sampling, and then through linear interpolation, the accuracy of the high-precision waveform solved by iteration can be guaranteed while optimizing the accuracy. If you want to further obtain a more accurate high-precision waveform, you can increase the number of sampling points and map the low-precision sample points to more sampling points of the high-precision waveform, but this will inevitably increase the complexity of the network and the consumption of training resources.

[0062] The low-precision one-dimensional data of the four waveforms and the sampled high-precision one-dimensional data are combined into four original data sets, and then the original data sets are randomly divided into training set, validation set and test set in the ratio of 2000:301:300.

[0063] Considering minimizing resource consumption, the number of network layers and convolution kernel size are continuously optimized through experiments, and the obtained Figure 8 As shown in the network structure, the low-precision waveform data passes through 5 layers of convolutional neural network in sequence to obtain the predicted high-precision sampling point data. The specific parameters of each convolutional layer are shown in Table 1. It can be seen that convolutional layers 2, 3, and 4 have exactly the same structure, which is conducive to reuse and saving resources.

[0064] During the convolution process, the number of data points remains unchanged through reasonable padding, so as to establish a single-point to single-point mapping. The relationship between the convolution layer input data length n1 and the output data length n2 is shown in formula (1).

[0065] (1); Where k is the size of the convolution kernel, p is the number of paddings, and s is the convolution step. Note that the padding method should use "nearest neighbor padding", that is, fill the missing values ​​with the corresponding values ​​of the nearest neighbor samples. This is because the beginning and end of the waveform are both in a steady state. If the default "zero padding" is used, incorrect features will be introduced, causing CNN to learn the wrong mapping relationship.

[0066] Table 1 Network parameters Number of input channels Number of output channels Convolution kernel size filling Step Length Convolutional layer 1 1 4 3 1 1 Convolutional Layer 2 4 4 3 1 1 Convolutional layer 3 4 4 3 1 1 Convolutional layer 4 4 4 3 1 1 Convolutional layer 5 4 1 3 1 1 For the four transient waveforms, four network models are trained respectively, with a model learning rate of 0.001, an optimizer of Adam, a loss function of MSELoss, an epoch limit of 1500, and a patience of 20, which means that when the loss does not improve significantly after 20 consecutive rounds, the model is saved and the training is stopped. This early stopping strategy can greatly save training time and computing resource consumption. The losses in the training set and the test set during the training of the four waveforms are as follows: Fig. 9 , Fig.10 , Fig.11, Fig.12 As shown in the figure, even if only the CPU is used, each model only needs an average of 449 rounds and 510 seconds to complete training.

[0067] The low-precision waveform of the test set is processed by the model to obtain the predicted waveform, and then the predicted waveform is linearly interpolated to the high-precision sample points to obtain the waveform with optimized accuracy. Each transient waveform is randomly selected from a certain working condition to observe the optimization effect, such as Fig.13 , Fig.14 , Fig.15 , Fig.16 As shown, in order to more intuitively evaluate the actual effect of this method, the root mean square error (MSE), relative root mean square error (RRMSE), relative error of energy loss during the turn-on process (ŋ1), and relative error of energy loss during the turn-off process (ŋ2) are used as evaluation indicators, and the calculation formulas are as follows: (2); (3); Among them, yj^ = {y1^, y2^, ... yn^} is the value of each point after the low-precision waveform is interpolated to the high-precision sample point or the value of each point after the low-precision waveform is optimized for accuracy, and yj is the value of each point of the corresponding high-precision waveform.

[0068] (4); (5); Among them, t1 and t2 are defined as the time required for the IGBT collector-emitter voltage to reach 90% of the stable value and the time required for the voltage to reach 10% of the stable value, respectively. t3 and t4 are defined as the time required for the IGBT collector-emitter voltage to reach 10% of the final value and the time required to reach 90% of the final value, respectively. The numerators above formulas (4) and (5) calculate the turn-on or turn-off losses of the low-precision waveform or the oversampled waveform, respectively, and the denominators below the formulas calculate the turn-on / turn-off losses of the corresponding high-precision waveforms, respectively.

[0069] The above indicators of the four waveforms in the test set before and after precision optimization are calculated respectively. In order to ensure the consistency of scale, the low-precision waveforms in the calculation are actually directly linearly interpolated to high-precision sample points. The average MSE, RRMSE, ŋ1, ŋ2 and improvement percentage of the 300 random working conditions in the test set are shown in Table 2, Table 3, Table 4, Table 5, and Table 6. It can be seen that the waveform after the precision is optimized by the method of the present invention is also more accurate. Compared with the high-precision waveform solved by direct iteration, the largest average RRMSE error occurs in the turn-on voltage waveform, which is only 0.84%. Compared with the low-precision waveform, the accuracy is greatly improved, and the average improvement percentage under the MSE indicator is greater than 90%. Overall, the energy loss prediction of the switching process is also more accurate, and the energy loss errors of the turn-on and turn-off processes are only 2.21% and 0.22%.

[0070] In summary, the method proposed in the present invention effectively improves the precision and accuracy of transient waveforms of power semiconductor switch simulation at a relatively low cost of resources and time, and solves two pairs of contradictions existing in current power electronics simulation.

[0071] Table 2 Turn-on process current waveform accuracy optimization accuracy improvement effect, index index Waveform MSE RRMSE Low-resolution waveform 187.70 2.14% Accuracy Optimized Waveform 18.27 0. 76% Percentage increase 90.26% 64.49% Table 3 Accuracy improvement effect of current waveform optimization during shutdown process index index Waveform MSE RRMSE Low-resolution waveform 3191.49 14.88% Accuracy Optimized Waveform 8.53 0. 79% Percentage increase 99.73% 94.69% Table 4 Turn-on process voltage waveform accuracy optimization accuracy improvement effect index index Waveform MSE RRMSE Low-resolution waveform 129.81 3.60% Accuracy Optimized Waveform 7.11 0. 84% Percentage increase 94.52% 76.67% Table 5 Accuracy improvement effect of voltage waveform optimization during shutdown process index index Waveform MSE RRMSE Low-resolution waveform 3245.46 8.42% Accuracy Optimized Waveform 8.24 0. 45% Percentage increase 99.75% 94.66% Table 6 Effect of improving energy loss prediction accuracy before and after precision optimization index index Waveform <![CDATA[ŋ1]]> <![CDATA[ŋ2]]> Low-resolution waveform 15.08% 3.60% Accuracy Optimized Waveform 2.21% 0.22% Percentage increase 85.34% 93.89% In the previous network structure, through reasonable padding, the data length of the convolution process remains unchanged, thereby establishing a single-point mapping between the low-precision waveform and the high-precision waveform at the low-precision sampling point. This method is simple and direct, but it has a great disadvantage: it is easy to miss some key points in areas with large change rates, such as the peak points of transient waveforms, resulting in Fig.17 The situation shown, Fig.17 This is a comparison chart of the network model output and the high-precision waveform in the peak area of ​​the IGBT turn-on process current waveform under certain working conditions. It can be seen that since the peak point of the high-precision waveform does not correspond to the sampling point, the interpolation will cause a large error at this time. Accurately predicting the peak is very important for power electronics simulation.

[0072] In response to the above problems, an improvement method is proposed. For areas with large transient waveform change rates, deconvolution (upsampling) is used to establish a mapping from one point to multiple high-precision waveform sample points. The structure of the upsampling convolutional neural network is as follows: Fig.18 Compared to Figure 8 The improved network changes the 4th convolution unit into an upsampling unit, and the rest of the structure remains unchanged. The upsampling unit consists of a deconvolution layer and an activation function relu. The parameters of the deconvolution layer are shown in the following table. Through reasonable step size and padding, the deconvolution layer doubles the data length, so that a mapping from 1 point to 3 points can be established.

[0073] Table 7 Deconvolution layer parameters Number of input channels Number of output channels Convolution kernel size filling Step Length Deconvolution Layer 4 4 3 1 2 Through the optimization of the upsampling network, Fig.17 The error is improved, and the improved waveform is as follows Fig.19 As shown in the figure, the error in the peak area is significantly reduced, but the mapping accuracy of some points is not high. It can be predicted that as long as the number of mapping points continues to increase, the error in the peak area can continue to decrease, but the cost is that under the same network structure, the task complexity becomes higher and the mapping accuracy will decrease.

[0074] The network structure proposed in this embodiment can be used for real-time simulation in theory. For the internal structure of the single-point mapping network, convolution units 2, 3, and 4 are exactly the same. For the upsampling convolution network, convolution units 1, 2, 3, and 5 are exactly the same as the corresponding units of the single-point mapping network. This design provides more flexibility for the simulation calculation process, whether it is to reuse FPGA hardware to save resources or to consume more resources for parallel computing, there are more options.

[0075] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0076] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A method for optimizing power semiconductor transient simulation accuracy based on machine learning, characterized in that: The method includes: S1, analyze the simulation accuracy optimization strategy, and improve the waveform accuracy by combining the direct interpolation method and machine learning strategy to solve the problem of insufficient accuracy in the transient waveform simulation of power semiconductor switches; S2, design a machine learning neural network structure suitable for processing transient waveform data of power semiconductors, which takes into account the timing and nonlinear characteristics of waveform data, and adopts a convolutional neural network, a recurrent neural network or a combination thereof, combined with appropriate activation functions and loss functions, to enhance the ability to extract transient waveform features; S3, collect a large amount of low-precision and high-precision transient simulation data to build a data set to ensure that the training data is sufficiently representative and diverse so that the designed neural network can accurately predict waveform details under various working conditions; S4, training the neural network model based on the constructed data set, optimizing the model performance by adjusting the network structure and training parameters, including learning rate, batch size, and number of iterations, and determining the best network model through cross-validation and loss curve monitoring to ensure the generalization ability of the model and high-precision waveform prediction ability; S5, using the trained neural network model to correct and compensate the low-precision transient waveform to improve the accuracy of the simulation waveform, so that the corrected simulation result is more consistent with the high-precision simulation result; S6, designs and integrates a simulation accuracy optimization unit, which contains the trained neural network model and is embedded in the power electronic converter simulation process to achieve dynamic optimization of waveform accuracy in different simulation scenarios, so as to improve the accuracy limitations of traditional simulation methods in highly dynamic and nonlinear transient processes, and provide more accurate data support for the design and optimization of power electronic equipment.

2. The power semiconductor transient simulation accuracy optimization method based on machine learning according to claim 1 is characterized in that: The transient waveform accuracy optimization strategy in S1 is inspired by the application of neural networks in the field of image super-resolution. With the help of neural networks, low-precision waveforms are reconstructed with super-resolution to obtain high-precision waveforms. Low-precision simulation consumes less computing resources and has a short simulation time, thereby avoiding further iterative resource consumption and solution time. Therefore, it is necessary to fully consider resource and time consumption so that the cost is as small as possible from direct iterative solution, which is also convenient for using this method in various embedded devices.

3. The power semiconductor transient simulation accuracy optimization method based on machine learning according to claim 1 is characterized in that: The neural network model structure described in S2 is constructed as follows: it has the characteristics of flexibility and simplicity, wherein a convolutional network is used to improve the applicability to different inputs, and convolutional layers with the same structure are reused to reduce storage resource consumption. Computing resources can be sacrificed to complete convolution operations at each layer in parallel to improve simulation speed and reduce simulation solution time.

4. The method for optimizing power semiconductor transient simulation accuracy based on machine learning according to claim 1, characterized in that: The data set in S3 can come from the data provided by the model manufacturer, or the model can be directly simulated offline to obtain low-precision waveforms and corresponding high-precision waveforms to construct a data set; the data set can be reasonably divided into training sets, validation sets and test sets to enhance the generalization of the model.

5. The power semiconductor transient simulation accuracy optimization method based on machine learning according to claim 1 is characterized in that: In S4, a mapping model from a low-precision waveform to a high-precision waveform is established through network training.

6. The power semiconductor transient simulation accuracy optimization method based on machine learning according to claim 1 is characterized in that: In S6, based on the transient simulation accuracy optimization method in S1 to S5, a simulation accuracy optimization unit structure that can be applied to various embedded device processors is proposed, which provides theoretical support for the subsequent application of the method proposed in the present invention to power electronics real-time simulation practice.

7. A system for optimizing power semiconductor transient simulation accuracy based on machine learning based on the method for optimizing power semiconductor transient simulation accuracy based on machine learning as claimed in any one of claims 1 to 6, characterized in that: The system includes: A simulation solver solves the circuit to obtain a low-precision transient waveform of a high-frequency power semiconductor device as input to a simulation transient accuracy optimization unit; The simulation accuracy optimization unit mainly includes four parts: data reading, caching, neural network and interpolation; The controller is responsible for controlling the operating logic and timing of each sub-module, and plays a decisive role in whether the calculation process can be completed.

8. The power semiconductor transient simulation accuracy optimization system based on machine learning according to claim 7 is characterized in that: The neural network consists of three parts: a multiplication array, an addition tree, and an activation function. The convolution operation requires the use of more multipliers, so the multipliers can be reasonably reused between and within each layer to save computing resources. To this end, the network weights and biases trained on the computer are loaded into the storage unit of the processor. For network layers with the same structure, only hardware resources need to be occupied once, and the corresponding weights and biases are called when performing calculations. Of course, more hardware resources can be used in parallel in exchange for less computing time.

9. A method for improving the offline simulation efficiency of a high-frequency converter system based on the power semiconductor transient simulation accuracy optimization method based on machine learning as described in any one of claims 1 to 6, characterized in that: This method uses a data set constructed through a one-time multi-operating condition simulation to train a neural network model with high generalization, so that in subsequent simulations, only a low-precision step size is needed to complete the simulation of the entire converter system. The high-precision transient waveform of the high-frequency semiconductor switch can be obtained through the network model.

Images