Inverter simulation method, device and computer equipment

Abstract

The present application relates to an inverter simulation method, device and computer equipment. The method establishes a training data set and a test data set based on a simulation object of an inverter; trains a convolutional neural network based on the training data set and the test data set to obtain a simulation result; wherein the convolution kernel in the convolutional neural network is obtained based on a physical field model and a conditional model preset for the simulation object, thereby solving the problem of low simulation efficiency, reducing the number of iterations, and quickly obtaining a steady-state result of the simulation.

Description

Technical Field

[0001] The present application relates to the field of electrical and thermal simulation, and in particular to an inverter simulation method, device and computer equipment. Background Art

[0002] Inverter, as one of the key components of new energy power generation system, plays an important role in energy transmission, conversion and control. Its working and reliability performance are directly related to the high efficiency and high reliability operation of the whole power generation system. Power semiconductor devices are the part with the highest failure rate in inverters. Common power loss devices include field effect tubes or IGBTs, diodes, transformers, inductors, etc. Calculating the loss and junction temperature of power devices is of great significance to the evaluation of product efficiency and the prediction of the life of power devices.

[0003] The current method mainly predicts the temperature of devices in the power circuit by establishing a thermal resistance model of the power device, and calculates the efficiency of the power device and even the power circuit based on the device temperature parameters. However, the existing inverter simulation scheme has the problems of long simulation time and many iterations, resulting in low simulation efficiency. Summary of the invention

[0004] In this embodiment, an inverter simulation method, apparatus, and computer device are provided to solve the problem of low inverter simulation efficiency in the related art.

[0005] In a first aspect, an inverter simulation method is provided in this embodiment, the method comprising:

[0006] Based on the simulation object of the inverter, establish a training data set and a test data set;

[0007] Based on the training data set and the test data set, a convolutional neural network is trained to obtain simulation results;

[0008] Among them, the convolution kernel in the convolutional neural network is obtained based on the physical field model and conditional model preset for the simulation object.

[0009] In some embodiments, a training data set and a test data set are established based on a simulation object of an inverter, including:

[0010] Based on the circuit-level simulation object of the inverter, circuit structural geometry information and circuit history result information are obtained; the circuit structural geometry information includes structure information and material information of circuit components;

[0011] generating a circuit simulation input based on the circuit structured geometric information;

[0012] Generate a circuit simulation result label based on the circuit history result information;

[0013] The circuit simulation input and the circuit simulation result label are divided into a circuit training data set and a circuit test data set respectively.

[0014] In some embodiments, a training data set and a test data set are established based on a simulation object of an inverter, including:

[0015] Based on the system-level simulation object of the inverter, the system structural geometry information and the system historical result information are obtained; the system structural geometry information includes the structure information and material information of the system components;

[0016] generating a system simulation input based on the structured geometric information of the system;

[0017] Generate a system simulation result label based on the system historical result information;

[0018] The system simulation input and the system simulation result label are divided into a system training data set and a system test data set respectively.

[0019] In some embodiments, the method further comprises:

[0020] Based on the simulation input of the simulation object, a pooling rule of the convolutional neural network is defined.

[0021] In some embodiments, the method further comprises:

[0022] A linear regression relationship between an input vector and an output vector of a fully connected layer of the convolutional neural network is defined.

[0023] In some of the embodiments, when the simulation object is a circuit-level simulation object, the input vector of the fully connected layer is obtained based on a training data set and a test data set of the circuit-level simulation object;

[0024] The output vector of the fully connected layer is a simulation result of a circuit-level simulation object.

[0025] In some of the embodiments, when the simulation object is a system-level simulation object, the input vector of the fully connected layer is obtained based on a system-level training data set and a test data set;

[0026] The output vector of the fully connected layer is a system-level simulation result.

[0027] In some of the embodiments, when the simulation object includes a circuit-level simulation object and a system-level simulation object, the input vector of the fully connected layer is obtained based on a circuit-level training data set and a test data set and a system-level training data set and a test data set; the output vector of the fully connected layer is a simulation result of electrothermal fusion.

[0028] In a second aspect, an inverter simulation device is provided in this embodiment, and the device includes:

[0029] A data set module is established to establish a training data set and a test data set based on a simulation object of an inverter;

[0030] A training prediction module is used to train a convolutional neural network based on the training data set and the test data set to obtain a simulation result; wherein the convolution kernel in the convolutional neural network is obtained based on the physical field model and conditional model of the simulation object.

[0031] In a third aspect, the present application further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the inverter simulation method described in the first aspect is implemented.

[0032] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the inverter simulation method described in the first aspect is implemented.

[0033] Compared with the related art, the inverter simulation method, device and computer equipment provided in this embodiment establish a training data set and a test data set based on a simulation object of the inverter; train a convolutional neural network based on the training data set and the test data set to obtain simulation results; wherein the convolution kernel in the convolutional neural network is obtained based on a physical field model and a conditional model preset for the simulation object, thereby solving the problem of low simulation efficiency, reducing the number of iterations, and quickly obtaining steady-state results of the simulation.

[0034] Details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0036] Figure 1 A hardware structure block diagram of a terminal of an inverter simulation method in one embodiment;

[0037] Figure 2 is a schematic flow chart of an inverter simulation method in one embodiment;

[0038] Figure 3 A schematic diagram of a process for preparing circuit-level data and models in a preferred embodiment;

[0039] Figure 4 A schematic diagram of a process for preparing system-level data and models in a preferred embodiment;

[0040] Figure 5 A schematic diagram of a flow chart of training prediction in a preferred embodiment;

[0041] Figure 6 FIG. 4 is a structural block diagram of an inverter simulation device in an embodiment. DETAILED DESCRIPTION

[0042] In order to more clearly understand the purpose, technical solutions and advantages of the present application, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments.

[0043] Unless otherwise defined, the technical terms or scientific terms involved in this application shall have the general meaning understood by people with general skills in the technical field to which this application belongs. The words "one", "a", "the", "these" and the like in this application do not indicate a quantitative limitation, and they may be singular or plural. The terms "include", "comprise", "have" and any variants thereof involved in this application are intended to cover non-exclusive inclusions; for example, a process, method and system, product or device comprising a series of steps or modules (units) is not limited to the listed steps or modules (units), but may include unlisted steps or modules (units), or may include other steps or modules (units) inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there may be three relationships, for example, "A and / or B" may mean: A exists alone, A and B exist at the same time, and B exists alone. Generally, the character " / " indicates that the objects associated with each other are in an "or" relationship. The terms "first", "second", "third", etc. in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0044] The method embodiment provided in this embodiment can be executed in a terminal, a computer or a similar computing device. For example, running on a terminal, Figure 1 FIG. 1 is a hardware structure diagram of a terminal of the inverter simulation method of this embodiment. Figure 1 As shown, the terminal may include one or more ( Figure 1Only one is shown in the figure) a processor 102 and a memory 104 for storing data, wherein the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA. The above terminal may also include a transmission device 106 and an input and output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0045] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the inverter simulation method in this embodiment. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, to implement the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0046] The transmission device 106 is used to receive or send data via a network. The above network includes a wireless network provided by a communication provider of the terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (Radio Frequency, referred to as RF) module, which is used to communicate with the Internet wirelessly.

[0047] In this embodiment, an inverter simulation method is provided. Figure 2 is a flow chart of the inverter simulation method of this embodiment, such as Figure 2 As shown, the process includes the following steps:

[0048] Step S210: establishing a training data set and a test data set based on the simulation object of the inverter.

[0049] Specifically, the simulation object of the inverter includes at least one of a circuit-level simulation object and a system-level simulation object. The circuit-level simulation object includes an inverter circuit, which is used to implement electrical simulation of the inverter and obtain simulation results such as power and efficiency; the system-level simulation object includes an inverter product, which is used to implement thermal simulation of the inverter and obtain simulation results such as temperature and speed.

[0050] Step S220, based on the training data set and the test data set, train the convolutional neural network to obtain simulation results; wherein the convolution kernel in the convolutional neural network is obtained based on the physical field model and conditional model preset for the simulation object.

[0051] Specifically, the prediction of the inverter simulation results is achieved through the convolutional neural network (CNN) algorithm by performing convolution, pooling, full connection, classification regression and other operations. The convolution kernel during the convolution operation can be designed specifically. For example, for the circuit-level simulation object, the circuit-level convolution kernel is defined based on the circuit-level physical field model and the circuit-level conditional model to perform convolution operations on the circuit-level training data set and the test data set; for the system-level simulation object, the system-level convolution kernel is defined based on the system-level physical field model and the system-level conditional model to perform convolution operations on the system-level training data set and the test data set.

[0052] Among them, the circuit-level physical field model is constructed based on the state loss model, the switching loss model and the junction temperature calculation model. The circuit-level conditional model includes the initial conditions and boundary conditions at the circuit level. The system-level physical field model is constructed based on the thermal model of the heat-generating components. The system-level conditional model includes the initial conditions and boundary conditions at the system level.

[0053] In this embodiment, a training data set and a test data set are established by a simulation object based on the inverter; a convolutional neural network is trained based on the training data set and the test data set to obtain simulation results; wherein, the convolution kernel in the convolutional neural network is obtained based on a physical field model and a conditional model preset for the simulation object, which solves the problem of low simulation efficiency, can reduce the number of iterations, and quickly obtain the effect of the steady-state results of the simulation.

[0054] In some embodiments, a training data set and a test data set are established based on a simulation object of an inverter, including:

[0055] Step S310, based on the circuit-level simulation object of the inverter, obtaining circuit structural geometry information and circuit history result information; the circuit structural geometry information includes structure information and material information of circuit components.

[0056] Specifically, circuit structural geometry information and circuit history result information are obtained based on existing simulation data.

[0057] Step S320: generating circuit simulation input based on the circuit structured geometry information.

[0058] Specifically, the circuit structured geometric information is analyzed, a circuit-level parameter identification vector is established, and a circuit simulation input is calculated based on the circuit-level parameter identification vector and the number of circuit elements.

[0059] Step S330: Generate a circuit simulation result label based on the circuit history result information.

[0060] Specifically, the circuit history result information can select existing simulation data or existing actual test data according to actual needs, with the purpose of correcting the simulation results through the results. The circuit simulation result label can be information such as voltage, current, output power value, temperature, etc. of each component based on the simulation results. The matrix length of the result label is adjusted and set according to the specific parameter category.

[0061] Step S340 , dividing the circuit simulation input and the circuit simulation result label into a circuit training data set and a circuit test data set respectively.

[0062] In this embodiment, the circuit structured geometric information is deconstructed to obtain circuit simulation input, and circuit simulation result labels are established through existing simulation data or existing actual test data to inject richer circuit-level information into the training data set and the test data set.

[0063] In some embodiments, a training data set and a test data set are established based on a simulation object of an inverter, including:

[0064] Step S410, based on the system-level simulation object of the inverter, obtaining system structural geometry information and system historical result information; the system structural geometry information includes structure information and material information of system components.

[0065] Specifically, system structural geometry information and circuit history result information are obtained based on existing simulation data.

[0066] Step S420: generating system simulation input based on the system structured geometric information.

[0067] Specifically, the structural geometric information of the system is analyzed, a system-level parameter identification vector is established, and a circuit simulation input is calculated based on the system-level parameter identification vector and the number of system-level components.

[0068] Step S430: Generate a system simulation result label based on the system historical result information.

[0069] Specifically, the system historical result information can select existing simulation data or existing actual test data according to actual needs, with the purpose of correcting the simulation results through the results. The system simulation result label can be based on the simulation results. The temperature, pressure, speed and other information of each component, and the matrix length is adjusted and set according to the specific parameter category.

[0070] Step S440 , dividing the system simulation input and the system simulation result label into a system training data set and a system test data set respectively.

[0071] In this embodiment, the system structured geometric information is constructed and deconstructed to obtain the system simulation input, and the system simulation result label is established through the existing simulation data or the existing actual test data to inject richer system-level information into the training data set and the test data set.

[0072] In some of these embodiments, the method further comprises:

[0073] Step S221, defining a pooling rule of a convolutional neural network based on a simulation input of a simulation object.

[0074] Specifically, the pooling layer calculation model changes according to the dimension of the simulation input after normalization. When the simulation object is a circuit-level simulation object, the simulation input is a circuit simulation input; when the simulation object is a system-level simulation object, the simulation input is a system simulation input.

[0075] In this embodiment, flexible configuration of convolutional neural network is achieved.

[0076] In some of these embodiments, the method further comprises:

[0077] Step S222, defining a linear regression relationship between an input vector and an output vector of a fully connected layer of a convolutional neural network.

[0078] In some of the embodiments, when the simulation object is a circuit-level simulation object, the input vector of the fully connected layer is obtained based on a training data set and a test data set of the circuit-level simulation object; and the output vector of the fully connected layer is a simulation result of the circuit-level simulation object.

[0079] Specifically, the training data set and test data set of the circuit-level simulation object can be obtained through steps S310 to S340, which include circuit simulation input and circuit simulation result labels. The circuit simulation input is sequentially subjected to convolution and pooling calculations of the convolutional neural network to obtain circuit-level features, and the circuit-level features are used as the input vector x of the fully connected layer, and the circuit simulation result label is used as the output vector y of the fully connected layer. The regression linear relationship of the fully connected layer is established as y=Wx+b, and classification is performed based on the softmax algorithm to optimize W and b, where W is the weight matrix and b is the bias vector. After the optimization is completed, the electrical simulation results of the circuit-level simulation object corresponding to the circuit simulation result label are output.

[0080] In this embodiment, accurate electrical simulation results are obtained in a targeted manner.

[0081] In some of the embodiments, when the simulation object is a system-level simulation object, the input vector of the fully connected layer is obtained based on a system-level training data set and a test data set; the output vector of the fully connected layer is a system-level simulation result.

[0082] Specifically, the training data set and test data set of the system-level simulation object can be obtained through steps S410 to S440, which include system simulation input and system simulation result labels. The system simulation input is sequentially subjected to convolution and pooling calculations of the convolutional neural network to obtain system-level features, and the system-level features are used as the input vector x of the fully connected layer, and the system simulation result label is used as the output vector y of the fully connected layer. The regression linear relationship of the fully connected layer is established as y=Wx+b, and classification is performed based on the softmax algorithm to optimize W and b, where W is the weight matrix and b is the bias vector. After the optimization is completed, the simulation results of the system-level simulation object corresponding to the system simulation result label are output.

[0083] In this embodiment, accurate thermal simulation results are obtained in a targeted manner.

[0084] In some of the embodiments, when the simulation object includes a circuit-level simulation object and a system-level simulation object, the input vector of the fully connected layer is obtained based on a circuit-level training data set and a test data set as well as a system-level training data set and a test data set; the output vector of the fully connected layer is a simulation result of electrothermal fusion.

[0085] Specifically, the circuit simulation input is sequentially subjected to convolution and pooling calculations of the convolutional neural network to obtain circuit-level features, and the system simulation input is sequentially subjected to convolution and pooling calculations of the convolutional neural network to obtain system-level features. The circuit-level features and the system-level features are summed to obtain the feature sum, which is used as the input vector x of the fully connected layer. The circuit simulation result label and the system simulation result label are summed to obtain the label sum, which is used as the output vector y of the fully connected layer. The regression linear relationship of the fully connected layer is established as y=Wx+b, and classification is performed based on the softmax algorithm to optimize W and b, where W is the weight matrix and b is the bias vector. After the optimization is completed, the output is the electrothermal fusion simulation result, and the electrothermal fusion simulation result includes the electrical simulation result and the thermal simulation result.

[0086] Since the physical phenomena in the electric field will affect the conditional parameters in the thermal simulation, and the physical field of the thermal simulation will cause the temperature characteristics of the material to change, which in turn affects the change of the initial conditions in the electric field analysis, the simulation results of the electrothermal fusion will improve the simulation accuracy compared to the separate electric simulation and thermal simulation results. Compared with the traditional electrothermal simulation, which is solved by multiple iterations of approximation and convergence, this embodiment improves the simulation efficiency, and especially simplifies the difficulty of implementing electrothermal simulation of complex inverter products.

[0087] The present embodiment is described and illustrated below through preferred embodiments.

[0088] 1. Prepare circuit-level data and models: define the input data vector set of circuit-level simulation, establish the circuit-level simulation device model matrix information of components such as field-effect transistors or IGBTs, diodes, transformers, inductors, etc., establish a condition vector set based on the input initial conditions and boundary conditions of the circuit simulation, and create a circuit simulation input parameter vector set and simulation result vector set as deep learning training and test data sets. Figure 3 As shown, specifically including:

[0089] S1.1, extract the structured geometric information of the simulation analysis data, including circuit connection relationship information and material information, and establish the parameter identification vector as:

[0090] ;

[0091] Among them: P is the parameter identification vector of each pin of a single component, x represents the x coordinate, y represents the y coordinate, z represents the z coordinate, mp represents the material property vector of the component, and m is the total number of single component parameter identification vectors; A is the total parameter identification vector of the simulation engineering data, j represents a single data structure component, and n is the total number of data structure features in the simulation engineering data. The circuit simulation structure position information is not sensitive, and the coordinate information can be converted into network information:

[0092] ;

[0093] Among them, net represents the network name of each pin in the component.

[0094] S1.2, establish the circuit-level model matrix set: M j =[OSJ]; j=1,2,...n.

[0095] Where: O represents the conduction loss model, S represents the switching loss model, J represents the junction temperature calculation model, and n is the total number of data structure features in the simulation engineering data.

[0096] S1.3, define initial conditions and boundary conditions:

[0097] ;

[0098] Where: S represents the set of initial values ​​val of each component at the beginning of the simulation; E i represents the set of external input IPow at the beginning of the simulation, where IPow is the input power; C represents the interaction rules between the system and the external environment, which is the controller logic CT here; E b Represents the external input signal received during the simulation, here it is the external input pulse signal Pul; C and E b Finally, discrete PWM signal matrix information is generated.

[0099] S1.4, define the simulation data set, including simulation input D and simulation result label L.

[0100] ;

[0101] Where: D is the simulation input, which is a multidimensional matrix, N represents the number of simulated components, R represents a one-dimensional vector of component structure information and material properties, Dep represents a multidimensional vector (x, y, z, mp), and R×Dep is a single component vector P in the multidimensional vector A. j The transposed matrix of . L is the simulation result label, which is a one-dimensional vector for multi-label classification. The matrix [0.001, 0.002, 0.003, … 0.1] represents the numerical value of the simulation result or the test result value, which is the loss ratio value of the circuit simulation. The matrix number set represents the loss ratio range of 0.1% to 10%. The result label can be the voltage, current, output power value, temperature and other information of each component based on the simulation results. The matrix length of the result label is adjusted and set according to the specific parameter category.

[0102] 2. Prepare system-level data and models: Define the input data vector set for system-level thermal simulation of the product, establish the thermal model matrix information of components such as controllers, field effect transistors or IGBTs, diodes, transformers, inductors, etc., establish a condition vector set based on the input initial conditions and boundary conditions of the system thermal simulation, and create a system thermal simulation input parameter vector set and simulation result vector set as deep learning training and test data sets. Figure 4 As shown, specifically including:

[0103] S2.1, extract the structured geometric information of the simulation analysis data, including the system structure connection relationship information and the material information of each component, and establish the parameter identification vector as:

[0104] ;

[0105] Where: P is the parameter identification vector of a single component, which contains the structural information of the chassis and each component, x represents the x-coordinate, y represents the y-coordinate, z represents the z-coordinate, mp represents the material property vector of the component, and m is the total number of single-component parameter identification vectors; A is the total parameter identification vector of the simulation engineering data, and n is the total number of data structure features in the simulation engineering data.

[0106] S2.2, establish the heat dissipation model of the whole system simulation components: M j =[T]; j=1,2,...n.

[0107] Where: T represents the thermal model of the heat generating component, and n is the total number of data structure features in the simulation engineering data. Here, the component thermal model can add model category parameters according to specific project characteristics.

[0108] S2.3, establish initial conditions and boundary conditions:

[0109] ;

[0110] Where: S represents the set of initial temperature values ​​Tem of each component at the beginning of the simulation, E i Represents the external input at the beginning of the simulation, here is the output power OPow of the active component; C represents the interaction rule Tm between the system and the external environment. Common heat dissipation systems include natural convection, forced convection, direct contact, heat pipes, cold storage materials, liquid cooling, phase change and other heat dissipation boundaries. The heat dissipation boundary characteristics include the range, position, size and direction of the boundary. The matrix information supports multiple heat dissipation forms for a single component; E b Represents the external input signal Ti received during the simulation process. Here, there are different input signals according to different heat dissipation boundaries, such as the inlet wind speed for forced convection; C and E b Finally, discrete heat dissipation parameter matrix information is generated.

[0111] S2.4, defines the simulation data set, including D and L parameters, which are the simulation input and simulation result labels respectively.

[0112] ;

[0113] Where: D is the simulation input, which is a three-dimensional matrix, N represents the number of simulated components, R represents the one-dimensional vector of component structure information and material properties, Dep represents the multi-dimensional vector (x, y, z, mp), and R×Dep is the single component vector P in the multi-dimensional vector A. j The transposed matrix of ; L is the simulation result label, which is a one-dimensional vector for multi-label classification, which can be the temperature, pressure, speed, etc. of the sample point of interest. The matrix [N×[1,2,3,…200] represents the simulation temperature value or test result value of the simulation result, which is the temperature value of each component of the system simulation. The matrix number set represents the temperature value range of 1~200. The result label can be the temperature, pressure, speed and other information of each component based on the simulation result. The matrix length is adjusted and set according to the specific parameter category.

[0114] 3. Training and prediction: Use CNN algorithm to train simulation data set, extract the features of simulation data through convolution and pooling, and use Softmax activation function to perform multi-classification and regression to predict simulation results under different simulation data and conditions, so as to improve simulation efficiency. Figure 5 As shown, specifically including:

[0115] S3.1, normalize the input and output data: normalize the R matrix in the input data D. Normalization mainly includes matrix filling, coordinate transformation, numerical transformation, matrix transformation, etc.

[0116] ;

[0117] Among them: R' is the normalized matrix, R is the initial matrix, R matrix supports multiple different normalization calculations to achieve consistency of subsequent calculations, Z represents the matrix that supplements 0 when the number of columns of R matrix is ​​insufficient, i represents the number of columns of the supplemented 0 matrix, I is the defined matrix length, m is the number of columns of R matrix; T is the initial matrix of coordinate transformation, rate is the unit conversion value; matrix transformation is a multi-dimensional transformation of R matrix. The data normalization algorithm is not limited to the calculation method described above.

[0118] S3.2, define the convolution kernel: define the convolution kernel parameters according to the initial conditions and boundary conditions of the simulation. Different simulation types have different convolution kernels. The simulation convolution kernel is defined as:

[0119] ;

[0120] Where: Mm represents the component model, I represents the initial condition, S represents the state of the system at the beginning of the simulation, and E i represents the external input signal received at the beginning of the simulation; B represents the boundary conditions, C represents the interaction rules between the system and the external environment, and E represents the boundary conditions. b Represents the external input signal received during the simulation.

[0121] The simulation conditions corresponding to the matrix parameters in the convolution kernel are standardized, and different convolution kernels have different simulation feature extraction capabilities; before the convolution calculation, the data edges are filled according to the size of the convolution kernel.

[0122] S3.3, define pooling rules: define the pooling layer strategy according to the calculation needs, and the pooling layer calculation model is:

[0123] ;

[0124] in: y l i represents the value of the output feature at position i, y i’ l-1 Represents the input features within the pooling window, and max represents the pooling operation type. The pooling layer calculation model changes according to the dimension of the R' matrix.

[0125] S3.4, calculate the regression parameters of the fully connected layer and perform classification using the softmax algorithm.

[0126] Define the input vector of the fully connected layer as x, the output vector as y, and establish the regression linear relationship of the fully connected layer as:

[0127] ;

[0128] Among them: W is called the weight matrix, that is, the lines connecting neurons, and b is called the bias vector. N represents the number of components, (R×Dep) represents the one-dimensional input vector of a single component after the full connection layer conversion, x represents the full connection layer input vector, and y represents the full connection layer output vector, that is, the prediction result vector. Here, if the x and y vectors only contain circuit-level input vector information, only the circuit-level simulation results are predicted; if the electrothermal simulation results need to be predicted, it is necessary to sum the x and y vectors separately, and then perform classification and regression operations.

[0129] Use the softmax algorithm to mathematically transform the y vector to obtain the probability value P k , the mapping relationship is as follows:

[0130] ;

[0131] Where: i represents the index of vector y, K represents the dimension of vector y, and the denominator is the sum of the exponential values ​​of all elements.

[0132] In this preferred embodiment, by deconstructing the data structure and establishing the data structure information matrix relationship, a convolution kernel is established according to the physical field component model, the initial conditions and boundary conditions of the physical field, and multiple convolution kernels are defined. The weight and bias of the vector are calculated through convolution, pooling, classification, regression and other steps, thereby realizing the prediction of the simulation results. This method is not only applicable to a single product and a single physical field, but also to multi-physical field simulation evaluations at multiple different product levels.

[0133] It should be noted that the steps shown in the above process or the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0134] In this embodiment, an inverter simulation device is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. The terms "module", "unit", "sub-unit", etc. used below can implement a combination of software and / or hardware for a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0135] Figure 6 is a structural block diagram of the inverter simulation device of this embodiment, such as Figure 6 As shown, the device includes: a data set establishment module 10 and a training prediction module 20.

[0136] The data set establishment module 10 is used to establish a training data set and a test data set based on the simulation object of the inverter.

[0137] The training prediction module 20 is used to train a convolutional neural network based on a training data set and a test data set to obtain a simulation result; wherein the convolution kernel in the convolutional neural network is obtained based on a physical field model and a conditional model of the simulation object.

[0138] In some of the embodiments, a training data set and a test data set are established based on a simulation object of an inverter, including: obtaining circuit structured geometry information and circuit history result information based on a circuit-level simulation object of the inverter; the circuit structured geometry information includes structure information and material information of circuit components; generating a circuit simulation input based on the circuit structured geometry information; generating a circuit simulation result label based on the circuit history result information; and dividing the circuit simulation input and the circuit simulation result label into a circuit training data set and a circuit test data set, respectively.

[0139] In some of the embodiments, a training data set and a test data set are established based on a simulation object of an inverter, including: obtaining system structural geometry information and system historical result information based on a system-level simulation object of the inverter; the system structural geometry information includes structural information and material information of system components; generating a system simulation input based on the system structural geometry information; generating a system simulation result label based on the system historical result information; and dividing the system simulation input and the system simulation result label into a system training data set and a system test data set, respectively.

[0140] In some of the embodiments, the device also includes a pooling rule definition module for defining a pooling rule of a convolutional neural network based on a simulation input of a simulation object.

[0141] In some of these embodiments, the device also includes a fully connected layer definition module for defining a linear regression relationship between an input vector and an output vector of a fully connected layer of a convolutional neural network.

[0142] In some of the embodiments, when the simulation object is a circuit-level simulation object, the input vector of the fully connected layer is obtained based on a training data set and a test data set of the circuit-level simulation object; the output vector of the fully connected layer is a simulation result of the circuit-level simulation object.

[0143] In some of the embodiments, when the simulation object is a system-level simulation object, the input vector of the fully connected layer is obtained based on a system-level training data set and a test data set; the output vector of the fully connected layer is a system-level simulation result.

[0144] In some of the embodiments, when the simulation object includes a circuit-level simulation object and a system-level simulation object, the input vector of the fully connected layer is obtained based on a circuit-level training data set and a test data set as well as a system-level training data set and a test data set; the output vector of the fully connected layer is a simulation result of electrothermal fusion.

[0145] It should be noted that the above modules can be functional modules or program modules, and can be implemented by software or hardware. For modules implemented by hardware, the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0146] In this embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0147] Optionally, the computer device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0148] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementation modes, and will not be repeated in this embodiment.

[0149] In addition, in combination with the inverter simulation method provided in the above embodiments, a storage medium may be provided in this embodiment to implement the method. The storage medium stores a computer program, and when the computer program is executed by a processor, any one of the inverter simulation methods in the above embodiments is implemented.

[0150] It should be understood that the specific embodiments described herein are only used to explain the application, rather than to limit it. Based on the embodiments provided in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the protection scope of this application.

[0151] Obviously, the drawings are only some examples or embodiments of the present application. For ordinary technicians in the field, the present application can also be applied to other similar situations based on these drawings without creative work. In addition, it is understandable that although the work done in this development process may be complicated and lengthy, for ordinary technicians in the field, certain changes in design, manufacturing or production based on the technical content disclosed in this application are only conventional technical means and should not be regarded as insufficient content disclosed in this application.

[0152] The term "embodiment" in this application refers to a specific feature, structure or characteristic described in conjunction with the embodiment that can be included in at least one embodiment of the present application. The appearance of this phrase in various locations in the specification does not necessarily mean the same embodiment, nor does it mean that it is mutually exclusive with other embodiments and is independent or optional. It is clearly or implicitly understood by those of ordinary skill in the art that the embodiments described in this application can be combined with other embodiments without conflict.

[0153] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of patent protection. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the scope of protection of the present application. Therefore, the scope of protection of the present application shall be subject to the attached claims.

Claims

1. An inverter simulation method, characterized in that: The method comprises: Based on the simulation object of the inverter, establish a training data set and a test data set; Based on the training data set and the test data set, a linear regression relationship between an input vector and an output vector of a fully connected layer of a convolutional neural network is defined, and the convolutional neural network is trained to obtain a simulation result; Wherein, the convolution kernel in the convolutional neural network is obtained based on a physical field model and a conditional model preset for the simulation object; Wherein, when simulating electrothermal fusion, the simulation object includes a circuit-level simulation object and a system-level simulation object, and the training data set and the test data set include a circuit simulation input and a corresponding circuit simulation result label and a system simulation input and a corresponding system simulation result label; The circuit simulation input is sequentially subjected to convolution and pooling calculations of the convolutional neural network to obtain circuit-level features, and the system simulation input is sequentially subjected to convolution and pooling calculations of the convolutional neural network to obtain system-level features. The circuit-level features and the system-level features are summed to obtain a feature sum, and the feature sum is used as the input vector of the fully connected layer; the circuit simulation result label and the system simulation result label are summed to obtain a label sum, and the label sum is used as the output vector of the fully connected layer to establish a regression linear relationship of the fully connected layer.

2. The inverter simulation method according to claim 1, characterized in that: Based on the simulation object of the inverter, establish the training data set and test data set, including: Based on the circuit-level simulation object of the inverter, circuit structural geometry information and circuit history result information are obtained; the circuit structural geometry information includes structure information and material information of circuit components; generating a circuit simulation input based on the circuit structured geometric information; Generate a circuit simulation result label based on the circuit history result information; The circuit simulation input and the circuit simulation result label are divided into a circuit training data set and a circuit test data set respectively.

3. The inverter simulation method according to claim 1, characterized in that: Based on the simulation object of the inverter, establish the training data set and test data set, including: Based on the system-level simulation object of the inverter, the system structural geometry information and the system historical result information are obtained; the system structural geometry information includes the structure information and material information of the system components; generating a system simulation input based on the structured geometric information of the system; Generate a system simulation result label based on the system historical result information; The system simulation input and the system simulation result label are divided into a system training data set and a system test data set respectively.

4. The inverter simulation method according to claim 1, characterized in that: The method further comprises: Based on the simulation input of the simulation object, a pooling rule of the convolutional neural network is defined.

5. The inverter simulation method according to claim 1, characterized in that: When the simulation object is a circuit-level simulation object, the input vector of the fully connected layer is obtained based on a training data set and a test data set of the circuit-level simulation object; The output vector of the fully connected layer is a simulation result of a circuit-level simulation object.

6. The inverter simulation method according to claim 1, characterized in that: When the simulation object is a system-level simulation object, the input vector of the fully connected layer is obtained based on a system-level training data set and a test data set; The output vector of the fully connected layer is a system-level simulation result.

7. An inverter simulation device, characterized in that: The device comprises: A data set module is established to establish a training data set and a test data set based on a simulation object of an inverter; A training prediction module, used to define a linear regression relationship between an input vector and an output vector of a fully connected layer of a convolutional neural network based on the training data set and the test data set, train the convolutional neural network, and obtain a simulation result; wherein the convolution kernel in the convolutional neural network is obtained based on a physical field model and a conditional model of the simulation object; Wherein, when simulating electrothermal fusion, the simulation object includes a circuit-level simulation object and a system-level simulation object, and the training data set and the test data set include a circuit simulation input and a corresponding circuit simulation result label and a system simulation input and a corresponding system simulation result label; The circuit simulation input is sequentially subjected to convolution and pooling calculations of the convolutional neural network to obtain circuit-level features, and the system simulation input is sequentially subjected to convolution and pooling calculations of the convolutional neural network to obtain system-level features. The circuit-level features and the system-level features are summed to obtain a feature sum, and the feature sum is used as the input vector of the fully connected layer; the circuit simulation result label and the system simulation result label are summed to obtain a label sum, and the label sum is used as the output vector of the fully connected layer to establish a regression linear relationship of the fully connected layer.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the inverter simulation method according to any one of claims 1 to 6 when executing the computer program.

Images