High-power IGBT (Insulated Gate Bipolar Translator) junction temperature on-line calculation method and system for deep learning driven by physical information

Abstract

The invention relates to a high-power IGBT (Insulated Gate Bipolar Translator) junction temperature online calculation method and system based on physical information-driven deep learning, and the method comprises the steps: taking online collected physical information as input through a PI-DeepONts model, and outputting the maximum junction temperature of an IGBT; the physical information comprises IGBT real-time conversion power and environment temperature data; the PI-DeepONets model comprises a backbone network and a branch network, a model operator is obtained through output approximation of the backbone network and the branch network, and the maximum junction temperature of the IGBT is output by using the model operator; the PI-DeepONets model performs model training by using a thermal dynamic parameter loss function, and takes IGBT historical conversion power, environment temperature data and an output matrix as model input in the training process, and the system is used for realizing the method. Compared with the prior art, the method has the advantages that finite element calculation and deep learning are combined on the basis of the precision of the finite element analysis method, the junction temperature calculation solving process is accelerated, and the IGBT junction temperature calculation efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of IGBT drive technology, and in particular to a high-power IGBT junction temperature online calculation method and system driven by physical information deep learning. Background Art

[0002] Maximum power point tracking (MPPT) control technology is widely used in large-scale grid-connected converters such as deep-sea wind power. However, large and frequent power modulation induces increased power loss of IGBT chips in the converter. Scientific and effective power modulation strategies have become the key to meeting the dual challenges of power guarantee and reliable operation of IGBT. Providing IGBT junction temperature data that can quickly adapt to power modulation can provide a solid basis for the scientific selection of complex strategies for thermal management systems, ensuring stable and reliable operation of wind power systems.

[0003] IGBT junction temperature calculation methods are divided into two categories: finite element analysis and thermal network model method. Among them, the advantage of finite element analysis and its derivative methods is that they can capture the chip thermal field distribution in detail, but they have high requirements on material properties and boundary conditions, and the analysis process has a large amount of calculation, which makes it difficult to meet the real-time monitoring needs; while the thermal network model method and its derivative methods perform circuit equivalence to improve the real-time calculation, but have high requirements on the precise selection of thermal resistance and thermal capacitance parameters, and require iterative updates of dynamic changes in thermal network parameters, and the inherent parameter method is difficult to meet the accuracy of monitoring needs; the engineering applicability of the above two methods cannot meet the requirements of junction temperature detection technology and temperature management system of large-scale converters.

[0004] Therefore, how to accelerate the junction temperature calculation process on the basis of achieving the accuracy of the finite element analysis method to achieve the real-time online detection efficiency requirements is a problem that needs to be solved. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a physical information driven deep learning high-power IGBT junction temperature online calculation method and system.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] According to a first aspect of the present invention, a method for online calculation of high-power IGBT junction temperature driven by deep learning using physical information is provided. The method uses a PI-DeepONets model to take online collected physical information as input and output the maximum junction temperature of the IGBT; the physical information includes real-time conversion power of the IGBT and ambient temperature data;

[0008] The PI-DeepONets model includes a trunk network and a branch network. The model operator is approximated by the output of the trunk network and the branch network, and the maximum junction temperature of the IGBT is output by the model operator.

[0009] The PI-DeepONets model uses a thermal dynamic parameter loss function for model training, and uses IGBT historical conversion power, ambient temperature data, and an output matrix as model inputs during the training process.

[0010] As a preferred technical solution, the method for obtaining the output matrix is: using a thermal finite element calculation method to calculate and process the IGBT historical conversion power and ambient temperature data; the calculation process includes:

[0011] Obtain the historical conversion power, ambient temperature and surface temperature of the IGBT under actual working conditions as input data for the thermal finite element calculation module;

[0012] The temperature distribution and thermophysical parameters of the IGBT are obtained by using a thermal finite element calculation method, and a transient temperature snapshot is generated at each moment based on the temperature distribution and thermophysical parameters;

[0013] Each of the transient thermal temperature snapshots is processed and converted into a column vector form according to a preset rule, and all column vectors are merged in time sequence to obtain a temperature-time matrix, which is the output matrix.

[0014] As a preferred technical solution, the step of generating a transient temperature snapshot includes:

[0015] Obtaining material parameters, wherein the material includes temperature-dependent materials and temperature-independent materials; the material parameters include thermal conductivity parameters, specific heat capacity and material density;

[0016] Get the historical conversion power of IGBT, and get the IGBT heat source item based on the historical conversion power. The expression is:

[0017]

[0018] Among them, q represents the heat source term, Q represents the historical conversion power of the IGBT, and V represents the volume of the IGBT;

[0019] Constructing a three-dimensional IGBT model, dividing the three-dimensional IGBT model into regular tetrahedron and hexahedron mesh elements, wherein the mesh elements are connected by nodes and the mesh elements are numbered by cells;

[0020] The temperature distribution and thermophysical parameters of each grid element are calculated based on the material parameters, historical conversion power, ambient temperature and surface temperature. The expression is:

[0021]

[0022] Among them, K represents the thermal conductivity parameter, T represents the temperature distribution, q represents the heat source term, ρ represents the material density, and C represents the specific heat capacity. It means to obtain the second-order partial derivative of the temperature division along the x-axis. It means to obtain the second-order partial derivative of the temperature division along the y-axis direction. Indicates the second-order partial derivative of the temperature division along the z-axis, k x , k y and k z Indicates the thermal conductivity of the material in the x, y, and z directions, n x 、n y and n z represents the components of the normal vector of the boundary surface in the x, y, and z directions, h is the convective heat transfer coefficient, S is the heat transfer surface area, T a Surface temperature; represents heat flux;

[0023] Sampling is performed according to the preset sampling time, and the temperature distribution and thermophysical parameters at each sampling moment are extracted to form a transient temperature snapshot at the corresponding moment.

[0024] As a preferred technical solution, the material parameter of the temperature-independent material is a fixed value, and the acquisition method is direct acquisition; the material parameter of the temperature-dependent material is a variable value, and is obtained by calculation, and the calculation expression is:

[0025]

[0026] c=a2·T temp 3 -b2T temp 2 +c2·T temp +d2,

[0027] Where k represents the thermal conductivity parameter of the temperature-dependent material, c represents the specific heat capacity of the temperature-dependent material, T temp is the ambient temperature of the IGBT, and a1, b1, c1, a2, b2, c2, and d2 represent the constant coefficients in the formula.

[0028] As a preferred technical solution, the step of merging all column vectors in time sequence to obtain a temperature-time matrix includes:

[0029] Each transient temperature snapshot is obtained, planarized, and the planarized transient temperature snapshot is rewritten into a column vector form according to the cell number of the grid element. The expression is:

[0030]

[0031] Where r represents the transient thermal temperature snapshot after planarization, sampled in the preset sampling time sequence, and t r Represented as the rth sampling moment, It is represented by the IGBT module cell number M at time t r Ambient temperature, T r Represented as time t r The ambient temperature vector represented by the finite element calculation results of the IGBT;

[0032] The ambient temperature vector at each sampling moment is obtained and combined into a temperature-time matrix in the order of the sampling moments. The temperature-time matrix is ​​the output matrix, and its expression is:

[0033]

[0034] in, Indicates that the IGBT module cell number is M at time t s Temperature, t s Indicates the last sampling time, T h Indicates that the IGBT is at an ambient temperature T temp Ambient temperature data of all units at the time.

[0035] As a preferred technical solution, the PI-DeepONets model training process includes:

[0036] Obtain the IGBT historical conversion power, ambient temperature data, output matrix, preset time point and preset position point as input data of the PI-DeepONets model;

[0037] Discretizing the IGBT historical conversion power, ambient temperature data and output matrix, inputting the discretized IGBT historical conversion power, ambient temperature data and temperature-time matrix into a branch network for processing by the branch network to output a feature vector;

[0038] Input the preset time point and position point into the backbone network and output a vector with the same dimension as the feature vector;

[0039] Approximating the eigenvector and the vector to obtain the model operator;

[0040] Based on the model operator, the total loss is calculated using the thermal dynamic parameter loss function, and the parameters of the branch network and the trunk network are updated according to the total loss;

[0041] The above process is repeated until the minimum total loss value appears, and the corresponding PI-DeepONets model is selected as the final model.

[0042] As a preferred technical solution, the backbone network and the branch network have the same structure, both of which are multi-layer structures, and the structural expression is:

[0043]

[0044] Among them, v represents the function variable input into the backbone network or branch network, represents the 0th input layer, represents the output vector of the lth layer, f (l) is the activation function of the (l)th layer, is the bias vector of the lth layer, is the weight matrix of the lth layer, and p represents the type of network;

[0045] The expression of the activation function is:

[0046] f(v)=max(0,v),

[0047] Among them, f(v) represents the activation function.

[0048] As a preferred technical solution, the expression of the thermal dynamic parameter loss function is:

[0049]

[0050] in, is the data loss function, is the loss function combined with the physical information equation. N, P, and Q are normalization factors. N represents the total number of samples for calculating residuals. P and Q represent the number of points for evaluating residuals in each sample. represents the residual of the i-th sample at the preset position point x′ and the preset time point t, j represents the index value, G θ (·) model operator, ζ represents the ζth physical residual point.

[0051] As a preferred technical solution, the expression for obtaining the maximum junction temperature is:

[0052] T jmax =G θ '(P,T m ),

[0053] Among them, T jmax is the maximum junction temperature of the IGBT, G θ ′(·) represents the trained model operator calculation, P represents the real-time conversion power obtained online, T m Indicates the ambient temperature obtained online.

[0054] According to a second aspect of the present invention, a high-power IGBT junction temperature online calculation system driven by physical information and deep learning is provided, and the system is used to implement the above method.

[0055] Compared with the prior art, the present invention has the following advantages:

[0056] 1) Combining the traditional finite element analysis method with the improved deep learning model and using physical information as model input avoids the problem of difficulty in meeting the real-time monitoring requirements due to the high requirements of the traditional finite element analysis method on material properties and boundary conditions and the large amount of calculation in the analysis process. It reduces the excessive dependence on meshing accuracy and the complex numerical solution steps, thus improving the calculation efficiency.

[0057] 2) Based on the original deep learning model loss function that only relies on data, physical information constraints are introduced to construct a dynamic loss function, which ensures the calculation accuracy of the model under complex dynamic conditions. This enables the improved model to more accurately describe the transient thermal field distribution of the IGBT module. The introduction of the dynamic function enables the model to more flexibly respond to changes in the transient temperature field when calculating the IGBT junction temperature. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a structural diagram of the PI-DeepONets model of the present invention;

[0059] Figure 2 A network structure diagram of a backbone network or a branch network of the PI-DeepONets model of the present invention;

[0060] Figure 3 This is a flow chart of the model training method of the present invention;

[0061] Figure 4 It is a comparison diagram of the PI-DeepONet model of the present invention and the traditional junction temperature calculation method; (4a) is a bar chart of the calculation time of the PI-DeepONets model of the present invention and the finite element method; (4b) is a bar chart of the performance difference between the PI-DeepONets model of the present invention and the finite element method. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0063] Unless otherwise defined, the technical terms or scientific terms involved in this application should be understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a", "the" and the like involved in this application do not indicate a quantitative limitation, and may represent the singular or plural. The terms "include", "comprise", "have" and any of their variations involved in this application are intended to cover non-exclusive inclusions; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units that are not listed, or may also include other steps or 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" can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0064] In order to solve the problem of how to accelerate the junction temperature calculation and solution process to achieve real-time online detection efficiency on the basis of achieving the accuracy of the finite element analysis method, this embodiment provides a high-power IGBT junction temperature online calculation method driven by physical information deep learning. The method uses the PI-DeepONets model to take the physical information collected online as input and output the maximum junction temperature of the IGBT, which is expressed as:

[0065] T jmax =G θ ′(P,T m ),

[0066] Among them, T jmax is the maximum junction temperature of the IGBT, G θ ′(·) represents the trained model operator calculation, P represents the real-time conversion power obtained online, T m Indicates the ambient temperature obtained online.

[0067] The physical information includes the real-time conversion power of IGBT and the ambient temperature data; the PI-DeepONets model includes a backbone network and a branch network, and its overall structure diagram is as follows Figure 1 As shown, the model operator is approximated by the output of the main network and the branch network, and the maximum junction temperature of the IGBT is output using the model operator; the structure of the main network and the branch network is the same as shown in Figure 2 shown.

[0068] When calculating the IGBT junction temperature, the model operator of the trained PI-DeepONets model is used for calculation. Therefore, this embodiment also provides a PI-DeepONets model training method, and its flow chart is as follows: Figure 3 As shown, the specific steps include:

[0069] S1. Obtaining the historical conversion power of IGBT, ambient temperature data, input matrix, preset time point and preset position point as input data of the PI-DeepONets model, and obtaining the output matrix by: using the thermal finite element calculation method to calculate the historical conversion power of IGBT and ambient temperature data, wherein the calculation process includes:

[0070] S11, obtaining the historical conversion power, ambient temperature and surface temperature of the IGBT under actual working conditions as input data for thermal finite element calculation;

[0071] S12, using a thermal finite element calculation method to obtain the temperature distribution and thermophysical parameters of the IGBT, and generating a transient temperature snapshot at each moment based on the temperature distribution and thermophysical parameters; the step of generating the transient temperature snapshot includes:

[0072] S121, obtaining material parameters, wherein the material parameters include thermal conductivity parameters, specific heat capacity and material density; the materials include temperature-dependent materials and temperature-independent materials, and the material parameter obtaining methods of the two materials are different. The material parameters of the temperature-independent materials are directly obtained, while the material parameters of the temperature-dependent materials are obtained by calculation, and the calculation expression is:

[0073]

[0074] c=a2·T temp 3 -b2T temp 2 +c2·T temp +d2,

[0075] Where k represents the thermal conductivity parameter of the temperature-dependent material, c represents the specific heat capacity of the temperature-dependent material, T temp is the ambient temperature of the IGBT, a1, b1, c1, a2, b2, c2, d2 represent the constant coefficients in the formula;

[0076] S122, obtaining the historical conversion power of the IGBT, and obtaining the IGBT heat source item based on the historical conversion power, the expression is:

[0077]

[0078] Where, q represents the heat source term, Q represents the historical conversion power of the IGBT, and V represents the volume of the IGBT;

[0079] S123, constructing a three-dimensional IGBT model, dividing the three-dimensional IGBT model into regular tetrahedron and hexahedron mesh elements, wherein the mesh elements are connected by nodes and the mesh elements are numbered by cells;

[0080] S124. Calculate the temperature distribution and thermophysical parameters of each grid element based on the material parameters, historical conversion power, ambient temperature and surface temperature. The expression is:

[0081]

[0082] Among them, K represents the thermal conductivity parameter, T represents the temperature distribution, q represents the heat source term, ρ represents the material density, and C represents the specific heat capacity. It means to obtain the second-order partial derivative of the temperature division along the x-axis. It means to obtain the second-order partial derivative of the temperature division along the y-axis direction. Indicates the second-order partial derivative of the temperature division along the z-axis, k x , k y and k z Indicates the thermal conductivity of the material in the x, y, and z directions, n x 、n y and n z represents the components of the normal vector of the boundary surface in the x, y, and z directions, h is the convective heat transfer coefficient, S is the heat transfer surface area, T a Surface temperature; represents heat flux;

[0083] S125, setting the sampling time to 100s and performing sampling, extracting the temperature distribution and thermophysical parameters at each sampling moment to form a transient temperature snapshot at the corresponding moment.

[0084] S13, processing each of the transient thermal temperature snapshots and converting them into column vector form according to preset rules, merging all column vectors in time sequence to obtain a temperature-time matrix, and the temperature-time matrix is ​​the output matrix, and the detailed steps are:

[0085] S131, obtaining each transient temperature snapshot, performing planarization processing, and rewriting the transient temperature snapshot after planarization processing into a column vector form according to the cell sequence number of the grid element, and the expression is:

[0086]

[0087] Where r represents the transient thermal temperature snapshot after planarization, sampled in the preset sampling time sequence, and tr Represented as the rth sampling moment, It is represented by the IGBT module cell number M at time t r Ambient temperature, T r Represented as time t r The ambient temperature vector represented by the finite element calculation results of the IGBT;

[0088] S132, obtaining the ambient temperature vector at each sampling time, and merging them into a temperature-time matrix in the order of the sampling times. The temperature-time matrix is ​​the output matrix, and its expression is:

[0089]

[0090] in, Indicates that the IGBT module cell number is M at time t s Temperature, t s Indicates the last sampling time, T h Indicates that the IGBT is at an ambient temperature T temp Ambient temperature data of all units at the time.

[0091] S2, discretize the IGBT historical conversion power, ambient temperature data and output matrix, input the discretized IGBT historical conversion power, ambient temperature data and temperature-time matrix into the right branch network for processing and outputting feature vectors, and the discretized form is:

[0092] u=[u(x1),u(x2),..,u(x m )],

[0093] Among them, u represents the input of the model, u(x1),u(x2),…,u(x m ) is the discretized unit of the model input, x m is a variable; and the discretized unit is processed as follows in the branch network:

[0094]

[0095] Among them, v represents the function variable input into the backbone network or branch network, represents the 0th input layer, represents the output vector of the lth layer, f (l) is the activation function of the (l)th layer, is the bias vector of the lth layer, is the weight matrix of the lth layer, and p represents the type of network; among them, the expression of the activation function is:

[0096] f(v)=max(0,v).

[0097] S3. Input the preset time point and the preset position point into the backbone network and output a vector with the same dimension as the feature vector. Since the structure of the backbone network is the same as that of the branch network, the processing process of the preset time point and the preset position point in the backbone network is not described in detail here.

[0098] S4, approximating the eigenvector and the vector to obtain a model operator;

[0099] S5. Based on the model operator, the total loss is calculated using the thermal dynamic parameter loss function, and the parameters of the branch network and the trunk network are updated according to the total loss. The expression of the thermal dynamic parameter loss function is:

[0100]

[0101] in, is the data loss function, is the loss function combined with the physical information equation. N, P, and Q are normalization factors. N represents the total number of samples for calculating residuals. P and Q represent the number of points for evaluating residuals in each sample. represents the residual of the i-th sample at the preset position point x′ and the preset time point t, j represents the index value, G θ (·) model operator, ζ represents the ζth physical residual point.

[0102] S6. Repeat steps S1 to S5 until the minimum total loss value appears, and select the corresponding PI-DeepONets model as the final model.

[0103] This embodiment also provides a physical information driven deep learning high-power IGBT junction temperature online calculation system, which is used to implement the above method. Those skilled in the art can clearly understand that for the convenience and simplicity of description, reference can be made to the corresponding process in the above method embodiment, which will not be repeated here.

[0104] Embodiment 2:

[0105] In order to verify that the method provided in the above embodiment has better performance than the traditional method, the method provided in the present invention is compared with the traditional junction temperature calculation method in this embodiment, and the two methods are used to calculate the junction temperature of a certain IGBT module at each moment within 20s. The comparison results are as follows: Figure 4 As shown, (4a) is a bar chart comparing the calculation time of the PI-DeepONets model of the present invention and the finite element method; (4b) is a bar chart of the performance difference between the PI-DeepONets model of the present invention and the finite element method.

[0106] As shown in Figure (4a), when calculating the junction temperature of the relative object, the PI-DeepONets model provided in the above embodiment takes 4.6 seconds, while the finite element method takes 138.4 seconds to calculate the junction temperature. It can be seen that the PI-DeepONets model has higher calculation efficiency than the finite element calculation model.

[0107] In Figure (4b), the mean, standard deviation, minimum and maximum values ​​of the PI-DeepONets model and the finite element method are used as indicators for evaluating the performance, and a bar graph is drawn using the differences between the performance indicators of the PI-DeepONets model and the finite element method. It can be seen that the method provided in the above embodiment has various error performance controls within a better range compared to finite element calculations, and has good performance and accuracy.

[0108] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A physical information driven deep learning high-power IGBT junction temperature online calculation method, characterized in that: The method uses the PI-DeepONets model to take the physical information collected online as input and output the maximum junction temperature of the IGBT; the physical information includes the real-time conversion power of the IGBT and the ambient temperature data; The PI-DeepONets model includes a trunk network and a branch network. The model operator is approximated by the output of the trunk network and the branch network, and the maximum junction temperature of the IGBT is output by the model operator. The PI-DeepONets model uses a thermal dynamic parameter loss function for model training, and uses IGBT historical conversion power, ambient temperature data, and an output matrix as model inputs during the training process.

2. According to the physical information driven deep learning high-power IGBT junction temperature online calculation method of claim 1, it is characterized in that: The method for obtaining the output matrix is: using a thermal finite element calculation method to calculate and process the IGBT historical conversion power and ambient temperature data; the calculation process includes: Obtain the historical conversion power, ambient temperature and surface temperature of the IGBT under actual working conditions as input data for the thermal finite element calculation module; The temperature distribution and thermophysical parameters of the IGBT are obtained by using a thermal finite element calculation method, and a transient temperature snapshot at each moment is generated based on the temperature distribution and thermophysical parameters; Each of the transient thermal temperature snapshots is processed and converted into a column vector form according to a preset rule, and all column vectors are merged in time sequence to obtain a temperature-time matrix, which is the output matrix.

3. According to claim 2, a physical information driven deep learning high-power IGBT junction temperature online calculation method is characterized in that: The step of generating a transient temperature snapshot comprises: Obtaining material parameters, wherein the material includes temperature-dependent materials and temperature-independent materials; the material parameters include thermal conductivity parameters, specific heat capacity and material density; Get the historical conversion power of IGBT, and get the IGBT heat source item based on the historical conversion power. The expression is: Among them, q represents the heat source term, Q represents the historical conversion power of the IGBT, and V represents the volume of the IGBT; Constructing a three-dimensional IGBT model, dividing the three-dimensional IGBT model into regular tetrahedron and hexahedron mesh elements, wherein the mesh elements are connected by nodes and the mesh elements are numbered by cells; The temperature distribution and thermophysical parameters of each grid element are calculated based on the material parameters, historical conversion power, ambient temperature and surface temperature. The expression is: Among them, K represents the thermal conductivity parameter, T represents the temperature distribution, q represents the heat source term, p represents the material density, and C represents the specific heat capacity. It means to obtain the second-order partial derivative of the temperature division along the x-axis. It means to obtain the second-order partial derivative of the temperature division along the y-axis direction. Indicates the second-order partial derivative of the temperature division along the z-axis, k x , k y and k z Indicates the thermal conductivity of the material in the x, y, and z directions, n x 、n y and n z represents the components of the normal vector of the boundary surface in the x, y, and z directions, h is the convective heat transfer coefficient, S is the heat transfer surface area, T a Surface temperature; represents heat flux; Sampling is performed according to the preset sampling time, and the temperature distribution and thermophysical parameters at each sampling moment are extracted to form a transient temperature snapshot at the corresponding moment.

4. According to claim 3, a physical information driven deep learning high-power IGBT junction temperature online calculation method is characterized in that: The material parameters of the temperature-independent material are fixed values, and the acquisition method is direct acquisition; the material parameters of the temperature-dependent material are variable values, and are obtained by calculation, and the calculation expression is: c=a2·T temp 3 -b2T temp 2 +c2·T temp +d2, Where k represents the thermal conductivity parameter of the temperature-dependent material, c represents the specific heat capacity of the temperature-dependent material, T temp is the ambient temperature of the IGBT, and a1, b1, c1, a2, b2, c2, and d2 represent the constant coefficients in the formula.

5. The method for online calculation of high-power IGBT junction temperature based on physical information driven deep learning according to claim 3 is characterized in that: The step of merging all column vectors in time sequence to obtain a temperature-time matrix comprises: Each transient temperature snapshot is obtained, planarized, and the planarized transient temperature snapshot is rewritten into a column vector form according to the cell number of the grid element. The expression is: Where r represents the transient thermal temperature snapshot after planarization, sampled in the preset sampling time sequence, and t r Represented as the rth sampling moment, It is represented by the IGBT module cell number M at time t r Ambient temperature, T r Represented as time t r The ambient temperature vector represented by the finite element calculation results of the IGBT; The ambient temperature vector at each sampling moment is obtained and combined into a temperature-time matrix in the order of the sampling moments. The temperature-time matrix is ​​the output matrix, and its expression is: in, Indicates that the IGBT module cell number is M at time t s Temperature, t s Indicates the last sampling time, T h Indicates that the IGBT is at an ambient temperature T temp Ambient temperature data of all units at the time.

6. The method for online calculation of high-power IGBT junction temperature based on physical information driven deep learning according to claim 2, characterized in that: The PI-DeepONets model training process includes: Obtain the IGBT historical conversion power, ambient temperature data, output matrix, preset time point and preset position point as input data of the PI-DeepONets model; Discretizing the IGBT historical conversion power, ambient temperature data and output matrix, inputting the discretized IGBT historical conversion power, ambient temperature data and temperature-time matrix into a branch network for processing by the branch network to output a feature vector; Input the preset time point and the preset position point into the backbone network and output a vector with the same dimension as the feature vector; Approximating the eigenvector and the vector to obtain the model operator; Based on the model operator, the total loss is calculated using the thermal dynamic parameter loss function, and the parameters of the branch network and the trunk network are updated according to the total loss; The above process is repeated until the minimum total loss value appears, and the corresponding PI-DeepONets model is selected as the final model.

7. The method for online calculation of high-power IGBT junction temperature based on physical information driven deep learning according to claim 6, characterized in that: The backbone network and the branch network have the same structure, both of which are multi-layer structures, and the structural expression is: Among them, v represents the function variable input into the backbone network or branch network, represents the 0th input layer, Represents the output vector of the ,th layer, f (l) is the activation function of the (l)th layer, is the bias vector of the,th layer, is the weight matrix of the ,th layer, and p represents the type of network; The expression of the activation function is: f(v)=max(0,v), Among them, f(v) represents the activation function.

8. The method for online calculation of high-power IGBT junction temperature based on physical information driven deep learning according to claim 6, characterized in that: The expression of the thermal dynamic parameter loss function is: in, is the data loss function, is the loss function combined with the physical information equation. N, P, and Q are normalization factors. N represents the total number of samples for calculating residuals. P and Q represent the number of points for evaluating residuals in each sample. represents the residual of the i-th sample at the preset position point x′ and the preset time point t, j represents the index value, G θ (·) model operator, ζ represents the ζth physical residual point.

9. The method for online calculation of high-power IGBT junction temperature based on physical information driven deep learning according to claim 1, characterized in that: The expression for obtaining the maximum junction temperature is: T jmax =G θ ′(P,T m ), Among them, T jmax is the maximum junction temperature of the IGBT, G θ ′(·) represents the trained model operator calculation, P represents the real-time conversion power obtained online, T m Indicates the ambient temperature obtained online.

10. A physical information driven deep learning high-power IGBT junction temperature online calculation system, characterized in that: The system is used to implement the method according to any one of claims 1 to 9.

Images