Method and device for determining electromagnetic thermal coupling field of flywheel motor

Abstract

The invention relates to a flywheel motor electromagnetic thermal coupling field determination method and device. The method comprises the steps that an improved physical information neural network is constructed and comprises an input layer, a sharing layer and a multi-head network which are connected in sequence, a first compression excitation module is added between the sharing layer and the multi-head network, and the multi-head network comprises a magnetic field head network, a current density head network and a temperature head network; and a second compression excitation module is additionally arranged at the output end of each head network in the multi-head network. A social and cooperation process in the chimpanzee tribe is simulated to determine an optimal solution so as to obtain a model parameter initial value, the improved physical information neural network is trained by using the data set so as to obtain a trained improved physical information neural network, and loss functions during training comprise electromagnetic loss, heat conduction loss and mean square error loss; and performing electromagnetic thermal coupling field prediction by using the trained improved physical information neural network. By adopting the method, the accuracy of electromagnetic thermal coupling field prediction can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of multi-physics field coupling analysis, and in particular to a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for determining the electromagnetic-thermal coupling field of a flywheel motor. Background Art

[0002] With the development of new energy technologies, energy storage technologies have emerged. Among them, flywheel energy storage systems are an efficient way of energy storage. The performance of their core component, the flywheel motor, directly affects the energy storage capacity, charging and discharging efficiency, and stability of the entire system. By accurately calculating the electromagnetic thermal coupling field, it is possible to better optimize the design of the flywheel motor, improve energy storage and conversion efficiency, and meet the demand for efficient energy storage equipment in fields such as smart grids and electric vehicles. In the field of motor design and optimization, accurate analysis of the complex physical field distribution inside the motor is crucial to improving motor performance, reducing energy consumption, and extending motor service life. This method provides motor design engineers with a powerful tool to help develop new high-performance motors. Multi-physics field coupling analysis technology itself is an important research direction in modern engineering science, involving multiple disciplines such as electromagnetism, thermodynamics, and mechanics.

[0003] During flywheel motor operation, complex multi-physics coupling occurs within the motor, with the coupling between electromagnetic and thermal fields being particularly significant. When the motor is running, current flowing through the windings generates a magnetic field. Due to the resistance of the windings, the thermal effect of this current causes the winding temperature to rise. This temperature change, in turn, affects the electromagnetic properties of the material, such as conductivity and permeability, and thus alters the motor's electromagnetic performance, such as inductance and torque. If this interaction is not accurately analyzed and controlled, it can lead to degraded motor performance or even failure.

[0004] In early calculations of the electromagnetic-thermal coupling field of motors, the motor physical model was discretized into a finite number of units, and the numerical solution was obtained by solving the governing equations for each unit. Electromagnetic analysis determines electromagnetic performance based on Maxwell's equations, while thermal analysis solves the temperature distribution based on the heat conduction equation. The calculated losses are then loaded into the thermal analysis model as a heat source. However, this method is computationally intensive and time-consuming when processing complex models, requiring high computing resources, and it is difficult to accurately capture the nonlinear interactions between multiple physical fields.

[0005] In recent years, machine learning techniques, particularly neural networks, have been applied in the field of electric motors. Leveraging their powerful nonlinear fitting capabilities, neural networks can establish mappings between motor performance and design parameters, enabling prediction of the electromagnetic and thermal performance of motors. However, simple neural network models lack a deep understanding of physical processes, resulting in limited generalization and interpretability of complex physical field couplings. This results in low accuracy in electromagnetic-thermal coupling field predictions.

[0006] In summary, the accuracy of current electromagnetic thermal coupling field prediction needs to be improved. Summary of the Invention

[0007] Based on this, it is necessary to provide a method, device, computer equipment, computer-readable storage medium and computer program product for determining the electromagnetic thermal coupling field of a flywheel motor, which can improve the accuracy of electromagnetic thermal coupling field prediction, in order to address the above technical problems.

[0008] In a first aspect, the present application provides a method for determining the electromagnetic thermal coupling field of a flywheel motor, comprising:

[0009] Constructing a data set for an energy storage flywheel, the data set including geometric parameters, material property parameters, operating condition parameters, and electromagnetic heat output data of the flywheel motor, the electromagnetic heat output data including magnetic field strength, magnetic flux density, current density, and temperature;

[0010] Constructing an improved physical information neural network, the improved physical information neural network comprising an input layer, a shared layer, and a multi-head network connected in sequence, adding a first compression excitation module between the shared layer and the multi-head network, the multi-head network comprising a magnetic field head network, a current density head network, and a temperature head network, the input end of each head network in the multi-head network being respectively connected to the output end of the first compression excitation module, and adding a second compression excitation module to the output end of each head network in the multi-head network;

[0011] Using all groups of model parameters of the improved physical information neural network as all candidate solutions to construct an initial gorilla population, simulating the social and collaborative processes in a gorilla tribe to determine the optimal solution corresponding to the optimal individual from the gorilla population, and using the group of model parameters corresponding to the optimal solution as initial values ​​of the model parameters of the improved physical information neural network;

[0012] Training the improved physical information neural network using the data set to obtain a trained improved physical information neural network, wherein a loss function during training includes electromagnetic loss, heat conduction loss, and mean square error loss;

[0013] The geometric parameters, material property parameters, and working condition parameters of the flywheel motor collected in real time are input into the trained improved physical information neural network, and the corresponding magnetic field intensity prediction values, magnetic flux density prediction values, current density prediction values, and temperature prediction values ​​are output.

[0014] In one embodiment, the geometric parameters include stator outer diameter, stator inner diameter, rotor outer diameter, rotor inner diameter and axial length; the material characteristic parameters include magnetic permeability, electrical conductivity and thermal conductivity; and the working condition parameters include current and rotational speed.

[0015] In one embodiment, the shared layer adopts a multi-layer fully connected layer structure, which is used to perform linear transformation on the input features from the input layer and then perform activation processing to obtain the intermediate layer output, and to perform linear transformation on the intermediate layer output and then perform activation processing to obtain the shared layer output.

[0016] In one embodiment, the electromagnetic loss calculation process includes: calculating the gradient of the magnetic field intensity corresponding to each sample in the data set; obtaining the current density prediction value corresponding to each sample calculated by the current density head network; and obtaining the electromagnetic loss calculation formula based on Ampere's law in Maxwell's equations, wherein the electromagnetic loss calculation formula is:

[0017]

[0018] Where, L em is the electromagnetic loss, N is the number of samples in the data set, H i and J i are the predicted values ​​of magnetic field intensity and current density of the i-th sample, respectively. is the gradient of the magnetic field intensity of the i-th sample; the gradient of the magnetic field intensity and the predicted value of the current density are substituted into the electromagnetic loss calculation formula to calculate the electromagnetic loss.

[0019] In one embodiment, the heat conduction loss calculation process includes: calculating the thermal diffusion term value of the temperature corresponding to each sample in the data set; calculating the square value of the current density prediction value corresponding to each sample, and calculating the ratio of the square value to the conductivity to obtain the heat source term value corresponding to each sample; and obtaining a heat conduction loss calculation formula based on the heat conduction equation, wherein the heat conduction loss calculation formula is:

[0020]

[0021] Where, L th is the heat conduction loss, is the heat diffusion term, J i ·J i / σ is the heat source term, k is the thermal conductivity, T i is the temperature prediction value of the i-th sample, σ is the electrical conductivity; the thermal diffusion term value and the heat source term value are substituted into the thermal conduction loss calculation formula to calculate the thermal conduction loss.

[0022] In one embodiment, the use of the data set to train the improved physical information neural network to obtain a trained improved physical information neural network includes: dividing the data set into a training set and a validation set; using the training set to iteratively train the improved physical information neural network, wherein in each iteration, first-order moment estimation and second-order moment estimation are calculated; using a first decay rate to correct the first-order moment estimation to obtain a first-order moment estimation correction value; using a second decay rate to correct the second-order moment estimation to obtain a second-order moment estimation correction value; substituting the first-order moment estimation correction value and the second-order moment estimation correction value into a parameter update formula to obtain model parameters under the corresponding number of iterations; and updating the parameters of the improved physical information neural network to the model parameters under the corresponding number of iterations; using the validation set to verify the improved physical information neural network that has completed iterative training, and if the loss value calculated during verification stops decreasing, the training is completed.

[0023] In a second aspect, the present application further provides a device for determining an electromagnetic-thermal coupling field of a flywheel motor, comprising:

[0024] A data set construction module is used to construct a data set of the energy storage flywheel, wherein the data set includes geometric parameters, material characteristic parameters, working condition parameters and electromagnetic heat output data of the flywheel motor, wherein the electromagnetic heat output data includes magnetic field intensity, magnetic flux density, current density and temperature;

[0025] A modeling module for constructing an improved physical information neural network, wherein the improved physical information neural network includes an input layer, a shared layer, and a multi-head network connected in sequence, a first compression excitation module is added between the shared layer and the multi-head network, the multi-head network includes a magnetic field head network, a current density head network, and a temperature head network, the input end of each head network in the multi-head network is respectively connected to the output end of the first compression excitation module, and a second compression excitation module is respectively added to the output end of each head network in the multi-head network;

[0026] an optimization module, configured to use all groups of model parameters of the improved physical information neural network as all candidate solutions to construct an initial gorilla population, simulate the social and collaborative processes in the gorilla tribe, determine the optimal solution corresponding to the optimal individual from the gorilla population, and use the group of model parameters corresponding to the optimal solution as initial values ​​of the model parameters of the improved physical information neural network;

[0027] a training module, configured to train the improved physical information neural network using the data set to obtain a trained improved physical information neural network, wherein the loss function during training includes electromagnetic loss, heat conduction loss, and mean square error loss;

[0028] The prediction module is used to input the geometric parameters, material characteristic parameters, and working condition parameters of the flywheel motor collected in real time into the trained improved physical information neural network, and output the corresponding magnetic field intensity prediction value, magnetic flux density prediction value, current density prediction value, and temperature prediction value.

[0029] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0030] Constructing a data set for an energy storage flywheel, the data set including geometric parameters, material property parameters, operating condition parameters, and electromagnetic heat output data of the flywheel motor, the electromagnetic heat output data including magnetic field strength, magnetic flux density, current density, and temperature;

[0031] Constructing an improved physical information neural network, the improved physical information neural network comprising an input layer, a shared layer, and a multi-head network connected in sequence, adding a first compression excitation module between the shared layer and the multi-head network, the multi-head network comprising a magnetic field head network, a current density head network, and a temperature head network, the input end of each head network in the multi-head network being respectively connected to the output end of the first compression excitation module, and adding a second compression excitation module to the output end of each head network in the multi-head network;

[0032] Using all groups of model parameters of the improved physical information neural network as all candidate solutions to construct an initial gorilla population, simulating the social and collaborative processes in a gorilla tribe to determine the optimal solution corresponding to the optimal individual from the gorilla population, and using the group of model parameters corresponding to the optimal solution as initial values ​​of the model parameters of the improved physical information neural network;

[0033] Training the improved physical information neural network using the data set to obtain a trained improved physical information neural network, wherein a loss function during training includes electromagnetic loss, heat conduction loss, and mean square error loss;

[0034] The geometric parameters, material property parameters, and working condition parameters of the flywheel motor collected in real time are input into the trained improved physical information neural network, and the corresponding magnetic field intensity prediction values, magnetic flux density prediction values, current density prediction values, and temperature prediction values ​​are output.

[0035] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0036] Constructing a data set for an energy storage flywheel, the data set including geometric parameters, material property parameters, operating condition parameters, and electromagnetic heat output data of the flywheel motor, the electromagnetic heat output data including magnetic field strength, magnetic flux density, current density, and temperature;

[0037] Constructing an improved physical information neural network, the improved physical information neural network comprising an input layer, a shared layer, and a multi-head network connected in sequence, adding a first compression excitation module between the shared layer and the multi-head network, the multi-head network comprising a magnetic field head network, a current density head network, and a temperature head network, the input end of each head network in the multi-head network being respectively connected to the output end of the first compression excitation module, and adding a second compression excitation module to the output end of each head network in the multi-head network;

[0038] Using all groups of model parameters of the improved physical information neural network as all candidate solutions to construct an initial gorilla population, simulating the social and collaborative processes in a gorilla tribe to determine the optimal solution corresponding to the optimal individual from the gorilla population, and using the group of model parameters corresponding to the optimal solution as initial values ​​of the model parameters of the improved physical information neural network;

[0039] Training the improved physical information neural network using the data set to obtain a trained improved physical information neural network, wherein a loss function during training includes electromagnetic loss, heat conduction loss, and mean square error loss;

[0040] The geometric parameters, material property parameters, and working condition parameters of the flywheel motor collected in real time are input into the trained improved physical information neural network, and the corresponding magnetic field intensity prediction values, magnetic flux density prediction values, current density prediction values, and temperature prediction values ​​are output.

[0041] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, which, when executed by a processor, implements the following steps:

[0042] Constructing a data set for an energy storage flywheel, the data set including geometric parameters, material property parameters, operating condition parameters, and electromagnetic heat output data of the flywheel motor, the electromagnetic heat output data including magnetic field strength, magnetic flux density, current density, and temperature;

[0043] Constructing an improved physical information neural network, the improved physical information neural network comprising an input layer, a shared layer, and a multi-head network connected in sequence, adding a first compression excitation module between the shared layer and the multi-head network, the multi-head network comprising a magnetic field head network, a current density head network, and a temperature head network, the input end of each head network in the multi-head network being respectively connected to the output end of the first compression excitation module, and adding a second compression excitation module to the output end of each head network in the multi-head network;

[0044] Using all groups of model parameters of the improved physical information neural network as all candidate solutions to construct an initial gorilla population, simulating the social and collaborative processes in a gorilla tribe to determine the optimal solution corresponding to the optimal individual from the gorilla population, and using the group of model parameters corresponding to the optimal solution as initial values ​​of the model parameters of the improved physical information neural network;

[0045] Training the improved physical information neural network using the data set to obtain a trained improved physical information neural network, wherein a loss function during training includes electromagnetic loss, heat conduction loss, and mean square error loss;

[0046] The geometric parameters, material property parameters, and working condition parameters of the flywheel motor collected in real time are input into the trained improved physical information neural network, and the corresponding magnetic field intensity prediction values, magnetic flux density prediction values, current density prediction values, and temperature prediction values ​​are output.

[0047] The above-mentioned flywheel motor electromagnetic thermal coupling field determination method, device, computer equipment, computer-readable storage medium and computer program product, wherein the constructed data set includes the geometric parameters, material characteristic parameters, working condition parameters and electromagnetic thermal output data of the flywheel motor, and the electromagnetic thermal output data includes magnetic field intensity, magnetic flux density, current density and temperature; the constructed improved physical information neural network includes an input layer, a shared layer and a multi-head network connected in sequence, and a first compression excitation module is added between the shared layer and the multi-head network. The multi-head network includes a magnetic field head network, a current density head network and a temperature head network. The input end of each head network in the multi-head network is respectively connected to the output end of the first compression excitation module, and the output end of each head network in the multi-head network is respectively increased with a second compression excitation module. The excitation module is added to the first and second compressed excitation modules to reweight the features of different channels, suppress redundant channels, improve feature extraction quality, and enhance model accuracy. Furthermore, all sets of model parameters of the improved physical information neural network are used as candidate solutions to construct an initial gorilla population. The social and collaborative processes within the gorilla tribe are simulated to determine the optimal solution corresponding to the optimal individual from the gorilla population. The set of model parameters corresponding to the optimal solution is used as the initial model parameter values ​​of the improved physical information neural network. Thus, by simulating the behavior of the gorilla tribe and conducting a global search within a large candidate solution space, relatively optimal initial model parameter values ​​can be explored, providing a good foundation for subsequent model training. By setting a loss function that includes electromagnetic loss, thermal conduction loss, and mean square error loss, the model can follow the laws of physics when learning data, ensuring that the model prediction results conform to the actual physical process and increasing the credibility of the model prediction results. This results in a higher accuracy of the trained improved physical information neural network. This improved prediction accuracy of the electromagnetic-thermal coupling field is improved by using the trained improved physical information neural network. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments of the present application or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying any creative work.

[0049] Figure 1 1 is a flow chart of a method for determining the electromagnetic-thermal coupling field of a flywheel motor according to one embodiment;

[0050] Figure 2 1 is a flow chart of a method for determining the electromagnetic-thermal coupling field of a flywheel motor according to another embodiment;

[0051] Figure 3 A schematic diagram of the change in total loss value during training of an improved physical information neural network in one embodiment;

[0052] Figure 4 A structural block diagram of a device for determining the electromagnetic thermal coupling field of a flywheel motor according to one embodiment;

[0053] Figure 5 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0055] The related electromagnetic-thermal coupling field analysis method of energy storage flywheel motor mainly relies on finite element analysis (FEA), which has the following problems:

[0056] (1) Inaccurate multi-physics coupling modeling: When a flywheel motor is running, electromagnetic, thermal and other physical fields influence and interact with each other. When establishing a multi-physics coupling model, existing technologies often simplify some complex coupling mechanisms, resulting in the model not accurately reflecting the actual physical process. For example, when considering electromagnetic thermal coupling, the heat generated by electromagnetic loss and the effect of heat on electromagnetic performance may only be described using simple linear relationships or empirical formulas, while ignoring the nonlinear factors and dynamic change characteristics that exist in reality.

[0057] (2) Insufficient simulation of high-speed rotation conditions: Flywheel motors typically operate at high speeds, and related technologies have limitations in simulating the effects of high-speed rotation. For example, high-speed rotation can cause centrifugal forces to be generated inside the motor, causing the motor structure to deform, which in turn affects the electromagnetic performance and heat transfer characteristics. However, some calculation methods may not fully consider the impact of this structural deformation on the electromagnetic-thermal coupling field, or may only perform simple approximations.

[0058] (3) Lack of accurate description of complex material properties: Flywheel motors often use some new magnetic materials and insulating materials, the properties of which are often highly nonlinear and anisotropic. When describing the properties of these complex materials, related technologies may not be able to accurately consider the changes in the material's performance under different electromagnetic and thermal conditions. For example, the magnetic permeability of some magnetic materials will change significantly with changes in temperature and magnetic field strength, and existing calculation models may not accurately capture this change pattern, thereby affecting the calculation accuracy of the electromagnetic thermal coupling field.

[0059] In response to at least one of the above problems, an embodiment of the present invention proposes a method, device, computer equipment, computer-readable storage medium and computer program product for determining the electromagnetic-thermal coupling field of a flywheel motor, which breaks the limitations of traditional methods that either focus on physical models and ignore data learning, or focus on data fitting and ignore physical principles, and gives the model powerful generalization capabilities, so that it can accurately calculate the electromagnetic-thermal coupling field when facing flywheel motors with different structures and working conditions.

[0060] In an exemplary embodiment, Figure 1 As shown, a method for determining the electromagnetic-thermal coupling field of a flywheel motor is provided. This embodiment uses the method applied to a terminal as an example for illustration. It is understood that the method can also be applied to a server, or to a system including a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0061] Step 102: construct a data set of the energy storage flywheel, which includes geometric parameters, material characteristic parameters, working condition parameters and electromagnetic heat output data of the flywheel motor, and the electromagnetic heat output data includes magnetic field strength, magnetic flux density, current density and temperature.

[0062] Geometric parameters refer to parameters related to the geometric shape of the energy storage flywheel motor. Material property parameters refer to parameters related to the material properties of the energy storage flywheel motor. Operating condition parameters refer to parameters related to the operation of the energy storage flywheel motor. Magnetic field strength refers to a vector physical quantity used to describe the source of the magnetic field. Magnetic flux density refers to a vector that describes the strength and direction of the magnetic field at a specific point in the magnetic field. Current density refers to the current intensity passing through a unit cross-sectional area. Temperature refers to the motor temperature.

[0063] Exemplarily, multiple sets of input data of the flywheel motor are obtained, each set of input data includes geometric parameters, material characteristic parameters, and working condition parameters; the electromagnetic thermal output data corresponding to each set of input data is obtained, each set of input data and the corresponding electromagnetic thermal output data constitute a sample, and multiple samples are constructed to obtain a data set of the energy storage flywheel.

[0064] Step 104, construct an improved physical information neural network, which includes an input layer, a shared layer and a multi-head network connected in sequence, and adds a first compression excitation module between the shared layer and the multi-head network. The multi-head network includes a magnetic field head network, a current density head network and a temperature head network. The input end of each head network in the multi-head network is respectively connected to the output end of the first compression excitation module, and the output end of each head network in the multi-head network is respectively added with a second compression excitation module.

[0065] Among them, the input layer refers to the starting layer of the improved physical information neural network, which is used to receive the input data of the improved physical information neural network. The shared layer refers to the intermediate layer connecting the input layer and the multi-head network respectively. The multi-head network refers to the task-specific output part of the model. The first compression excitation module and the second compression excitation module both refer to SE (Squeeze-and-Excitation) modules. The SE module is a channel attention mechanism embedded in the convolutional neural network, which can adaptively enhance the response of important feature channels and suppress secondary channels, thereby improving the model characterization ability. The magnetic field head network refers to the part that outputs the magnetic field intensity and magnetic flux density. The current density head network refers to the part that outputs the current density. The temperature head network refers to the part that outputs the motor temperature.

[0066] Exemplarily, an input layer, a shared layer and a multi-head network are obtained, the input end of the shared layer is connected to the output end of the input layer, the output end of the shared layer is connected to the input end of the multi-head network, a first compression excitation module is added between the output end of the shared layer and the input end of the multi-head network, the input end of the multi-head network is composed of the input ends of each head network, and a second compression excitation module is added to the output end of each head network in the multi-head network, thereby constructing an improved physical information neural network.

[0067] In step 106, all groups of model parameters of the improved physical information neural network are used as all candidate solutions to construct an initial gorilla population, simulate the social and collaborative processes in the gorilla tribe, determine the optimal solution corresponding to the optimal individual from the gorilla population, and use the set of model parameters corresponding to the optimal solution as the initial values ​​of the model parameters of the improved physical information neural network.

[0068] Here, model parameters refer to the numerical values ​​used to define and adjust the model's behavior in the improved physical information neural network. All candidate solutions refer to the solution set of the optimization algorithm. The initial gorilla population refers to the initial population in the gorilla tribe optimization algorithm. The optimal individual refers to the individual with the highest fitness value.

[0069] Exemplarily, a gorilla tribe optimization algorithm is selected, and all groups of model parameters of the improved physical information neural network are used as all candidate solutions to obtain an initial gorilla population, wherein each individual in the population corresponds to a candidate solution, and the fitness value of each individual is calculated. The social and collaborative processes in the gorilla tribe are simulated to determine the optimal solution corresponding to the optimal individual from the gorilla population, and then a set of model parameters corresponding to the optimal solution are used as the initial values ​​of the model parameters of the improved physical information neural network.

[0070] Step 108 : Using the data set to train the improved physical information neural network to obtain a trained improved physical information neural network, wherein the loss function during training includes electromagnetic loss, heat conduction loss, and mean square error loss.

[0071] Electromagnetic loss refers to the loss function that considers the electromagnetic physical process. Thermal conduction loss refers to the loss function that considers the temperature field physical process. Mean square error loss refers to the traditional mean square error loss function.

[0072] Exemplarily, electromagnetic loss, thermal conduction loss and mean square error loss are defined, and the electromagnetic loss, thermal conduction loss and mean square error loss are added together as the loss function of the improved physical information neural network to participate in the training. The data set is used to complete the training of the improved physical information neural network to obtain the trained improved physical information neural network.

[0073] In step 110, the geometric parameters, material characteristic parameters, and working condition parameters of the flywheel motor collected in real time are input into the trained improved physical information neural network, and the corresponding magnetic field intensity prediction value, magnetic flux density prediction value, current density prediction value, and temperature prediction value are output.

[0074] The magnetic field intensity prediction value and magnetic flux density prediction value refer to the prediction results output by the magnetic field head network in the trained improved physical information neural network. The current density prediction value refers to the prediction result output by the current density head network in the trained improved physical information neural network. The temperature prediction value refers to the prediction result output by the temperature head network in the trained improved physical information neural network.

[0075] For example, the geometric parameters, material property parameters, and working condition parameters of the flywheel motor are acquired in real time, and are input into the trained improved physical information neural network for prediction, and the corresponding magnetic field strength prediction value, magnetic flux density prediction value, current density prediction value, and temperature prediction value are output.

[0076] In the above-mentioned method for determining the electromagnetic-thermal coupling field of a flywheel motor, the constructed data set includes geometric parameters, material characteristic parameters, working condition parameters and electromagnetic thermal output data of the flywheel motor, and the electromagnetic thermal output data includes magnetic field intensity, magnetic flux density, current density and temperature; the constructed improved physical information neural network includes an input layer, a shared layer and a multi-head network connected in sequence, and a first compression excitation module is added between the shared layer and the multi-head network. The multi-head network includes a magnetic field head network, a current density head network and a temperature head network. The input end of each head network in the multi-head network is respectively connected to the output end of the first compression excitation module, and the output end of each head network in the multi-head network is respectively added with a second compression excitation module; by adding the first compression excitation module and The second compression excitation module reweights the features of different channels, suppresses redundant channels, improves feature extraction quality, and enhances the accuracy of the improved physical information neural network. Furthermore, all sets of model parameters of the improved physical information neural network are used as candidate solutions to construct an initial gorilla population. The social and collaborative processes within the gorilla tribe are simulated to determine the optimal solution corresponding to the best individual within the gorilla population. The set of model parameters corresponding to the optimal solution is used as the initial values ​​of the model parameters of the improved physical information neural network. Thus, by simulating the behavior of the gorilla tribe and conducting a global search within a large candidate solution space, relatively optimal initial values ​​of the model parameters can be explored, providing a good foundation for subsequent model training. By setting a loss function that includes electromagnetic loss, thermal conduction loss, and mean square error loss, the improved physical information neural network is able to follow the laws of physics when learning data, ensuring that its predictions are consistent with actual physical processes and increasing the credibility of its predictions. This results in a higher accuracy of the trained improved physical information neural network. Consequently, predictions using this trained improved physical information neural network improve the accuracy of electromagnetic-thermal coupling field predictions.

[0077] In an exemplary embodiment, the geometric parameters include stator outer diameter, stator inner diameter, rotor outer diameter, rotor inner diameter and axial length; the material property parameters include magnetic permeability, electrical conductivity and thermal conductivity; and the operating condition parameters include current and rotational speed.

[0078] Among them, the stator outer diameter refers to the circumferential diameter size of the outer side of the motor stator (stationary part) core. The stator inner diameter refers to the diameter size of the inner circumference of the motor stator core. The rotor outer diameter refers to the maximum rotation diameter of the flywheel rotor body (including the permanent magnet integrated structure). The rotor inner diameter refers to the diameter of the rotor center axis hole. The axial length refers to the overall size of the motor body along its rotation axis (perpendicular to the direction of the flywheel rotation plane). Magnetic permeability is a parameter that characterizes the ability of a material to be magnetized in a magnetic field. Electrical conductivity is a parameter that describes the ability of a material to conduct electric current. Thermal conductivity is a parameter that characterizes the ability of a material to conduct heat. Current is the ratio of the amount of charge passing through the cross-section of the conductor to the time taken. The rotational speed refers to the rate at which the energy storage flywheel body rotates around the central axis.

[0079] In this embodiment, by considering various parameters that affect the electromagnetic thermal output data, such as the stator outer diameter, stator inner diameter, rotor outer diameter, rotor inner diameter, axial length, magnetic permeability, electrical conductivity, thermal conductivity, current and speed, and participating in subsequent training, the trained model can accurately calculate the electromagnetic thermal coupling field for flywheel motors with different geometric structures, different material properties and different working conditions.

[0080] In an exemplary embodiment, the shared layer adopts a multi-layer fully connected layer structure, which is used to perform linear transformation on the input features from the input layer and then perform activation processing to obtain the intermediate layer output, and to perform linear transformation on the intermediate layer output and then perform activation processing to obtain the shared layer output.

[0081] A multi-layer fully connected layer architecture consists of multiple fully connected layers. Linear transformation refers to a processing method that uses vector addition and scalar multiplication. Activation processing introduces nonlinear capabilities through activation functions.

[0082] In this embodiment, by adopting a multi-layer fully connected layer structure and performing multiple linear transformations and activation processing on the input features from the input layer, the nonlinear expression capability of the features can be gradually enhanced, so as to convert the input features of the input layer into more advanced abstract features, thereby better improving the expression capability of the model.

[0083] In an exemplary embodiment, the electromagnetic loss calculation process includes: calculating the gradient of the magnetic field intensity corresponding to each sample in the data set; obtaining the current density prediction value corresponding to each sample calculated by the current density head network; and obtaining the electromagnetic loss calculation formula based on Ampere's law in Maxwell's equations. The electromagnetic loss calculation formula is:

[0084]

[0085] Where, L em is the electromagnetic loss, N is the number of samples in the data set, Hi and J i are the predicted values ​​of magnetic field intensity and current density of the i-th sample, respectively. is the gradient of the magnetic field intensity of the i-th sample; the gradient of the magnetic field intensity and the predicted value of the current density are substituted into the electromagnetic loss calculation formula to calculate the electromagnetic loss.

[0086] The gradient of the magnetic field intensity refers to the rate of change of the magnetic field intensity along a specific direction in space. Ampere's law, the core equation in Maxwell's equations, describes the relationship between magnetic field, electric current, and changing electric field.

[0087] In this embodiment, the electromagnetic loss calculation formula is obtained by considering Ampere's law in Maxwell's equations, which can fully consider the physical process of the electromagnetic field and improve the fitting degree of the improved physical information neural network to the electromagnetic physics laws.

[0088] In an exemplary embodiment, the heat conduction loss calculation process includes: calculating the thermal diffusion term value of the temperature corresponding to each sample in the data set; calculating the square value of the current density prediction value corresponding to each sample, and calculating the ratio of the square value to the conductivity to obtain the heat source term value corresponding to each sample; and obtaining the heat conduction loss calculation formula based on the heat conduction equation. The heat conduction loss calculation formula is:

[0089]

[0090] Where, L th is the heat conduction loss, is the heat diffusion term, J i ·J i / σ is the heat source term, k is the thermal conductivity, T i is the temperature prediction value of the i-th sample, σ is the electrical conductivity; the heat diffusion term and the heat source term are substituted into the heat conduction loss calculation formula to calculate the heat conduction loss.

[0091] The heat conduction equation describes the transfer of heat within a medium. The heat diffusion term describes the spontaneous diffusion of heat. The heat source term characterizes the input and output of external energy.

[0092] In this embodiment, the heat conduction loss calculation formula is obtained by considering the heat conduction equation, which can fully consider the physical process of the temperature field and enhance the simulation ability of the physical information neural network on the thermal physical process.

[0093] In an exemplary embodiment, a data set is used to train an improved physical information neural network to obtain a trained improved physical information neural network, including: dividing the data set into a training set and a validation set; using the training set to iteratively train the improved physical information neural network, wherein in each iteration, first-order moment estimation and second-order moment estimation are calculated; using a first decay rate to correct the first-order moment estimation to obtain a first-order moment estimation correction value; using a second decay rate to correct the second-order moment estimation to obtain a second-order moment estimation correction value; substituting the first-order moment estimation correction value and the second-order moment estimation correction value into a parameter update formula to obtain model parameters under a corresponding number of iterations; and updating the parameters of the improved physical information neural network to the model parameters under the corresponding number of iterations; using the validation set to verify the improved physical information neural network that has completed iterative training, and if the loss value calculated during verification stops decreasing, the training is completed.

[0094] The first-order moment estimate is the exponentially weighted moving average of the slope of the total loss function. The second-order moment estimate is the exponentially weighted moving average of the square of the slope of the total loss function. The parameter update formula includes a minimum constant ∈.

[0095] In this embodiment, the first-order moment estimate and the second-order moment estimate are obtained by utilizing the slope of the total loss function, and the first-order moment estimate and the second-order moment estimate are corrected. Then, the first-order moment estimate correction value and the second-order moment estimate correction value are used to update the model parameters, which can avoid the initial update amplitude being too small. In addition, the introduction of a very small constant ∈ can improve the calculation stability.

[0096] Exemplarily, a method for determining the electromagnetic-thermal coupling field of a flywheel motor specifically includes:

[0097] Step 202: construct a data set of the energy storage flywheel, the data set including geometric parameters, material characteristic parameters, working condition parameters and electromagnetic heat output data of the flywheel motor, the electromagnetic heat output data including magnetic field strength, magnetic flux density, current density and temperature.

[0098] In step 202 , a data set of the energy storage flywheel may be acquired through a finite element model.

[0099] Specifically, by constructing a finite element model of the flywheel motor, the input data and output data of the finite element model are set, the input data includes the geometric parameters of the flywheel motor (the geometric parameters include the stator outer diameter D s-out 、Stator inner diameter D s-in , rotor outer diameter D r-out 、Rotor inner diameter D r-in, axial length L, material characteristic parameters (material characteristic parameters include magnetic permeability μ, electrical conductivity σ, thermal conductivity k and working condition parameters (working condition parameters include current I, speed n). Output data include magnetic field intensity, magnetic flux density, current density and temperature. In addition, it is necessary to set the Maxwell equations and heat conduction equations as constraints of the finite element model, use the finite element model to simulate the motor operating environment, analyze the electromagnetic heat transfer process (including magnetic field distribution, current density changes and temperature field formation process under different working conditions), and use electromagnetic analysis, heat conduction analysis and multi-physics field coupling analysis to study the role of Maxwell equations and heat conduction equations in the motor model (i.e., the finite element model of the flywheel motor), determine the change law of electromagnetic thermal physical quantities under different parameters, and provide data support for the subsequent construction of a multi-physics field coupling model of the electromagnetic thermal coupling field (i.e., the subsequent improved physical information neural network), and obtain relevant simulation data (i.e., electromagnetic thermal output data).

[0100] In addition, the computational accuracy of the finite element model is closely related to the quality of the meshing. Although the computational effort increases with mesh refinement, it can more accurately simulate the actual situation. In this embodiment, based on the flywheel motor structure and physical field distribution, the finite element model can be divided into electromagnetic components such as the stator, rotor, air gap, and windings, as well as heat conduction components such as solid materials and heat dissipation media. The dimensions of each component of the finite element model can be obtained by actually measuring the flywheel motor, and the material property parameters can be obtained by consulting data or experimental determination.

[0101] In the study of electromagnetic-thermal coupling field characteristics, it is known that the electromagnetic field and temperature field influence each other. For example, the Joule heat generated by the current changes the material temperature, which in turn affects the electromagnetic properties of the material. For example, after determining the operating frequency and heat dissipation conditions, under typical operating conditions, a point-by-point scanning simulation of the electromagnetic thermal physical quantities in each area within the flywheel motor is performed. For example, the temperature field is simulated every 5°C within the range of 20°C to 100°C to obtain the electromagnetic-thermal coupling field distribution data (i.e., electromagnetic thermal output data) under different operating conditions (i.e., different input data).

[0102] In this embodiment, different input data are set to simulate different operating conditions of the flywheel motor, thereby obtaining corresponding electromagnetic thermal coupling field distribution data. The input data under each operating condition and the corresponding electromagnetic thermal coupling field distribution data constitute a sample, and the samples corresponding to all operating conditions are constructed to obtain a data set for the energy storage flywheel, thereby completing Figure 2 The data collection steps are shown.

[0103] Step 204, construct an improved physical information neural network, which includes an input layer, a shared layer and a multi-head network connected in sequence, and adds a first compression excitation module between the shared layer and the multi-head network. The multi-head network includes a magnetic field head network, a current density head network and a temperature head network. The input end of each head network in the multi-head network is respectively connected to the output end of the first compression excitation module, and the output end of each head network in the multi-head network is respectively added with a second compression excitation module.

[0104] Specifically, the improved physical information neural network is also called a multi-head physical information neural network model based on the SE mechanism. The multi-head physical information neural network model includes an input layer, a shared layer and a multi-head network.

[0105] Input layer: The input layer integrates the aforementioned multivariate input features (i.e., the input data in the sample). The input dimension is determined by the number of parameters in the specific input feature. If the number of parameters in the input feature is m, the input dimension is m. The input layer passes the raw data (i.e., the input data of the improved physical information neural network, also represented as the input feature vector X) to the shared layer, laying the foundation for subsequent operations.

[0106] Shared layer: The shared layer adopts a multi-layer fully connected layer structure. The input feature vector X undergoes a linear transformation using the weight matrix W1 and the bias vector b1, and then a nonlinear factor is introduced through the ReLU activation function to obtain the intermediate layer output h1, i.e., h1 = ReLU(W1x+b1). Next, the intermediate layer output h1 undergoes a linear transformation using the weight matrix W2 and the bias vector b2 and the ReLU activation function again to obtain the shared layer output h2, i.e., h2 = ReLU(W2h1+b2). ​​On this basis, the shared layer is improved, and an SE module is innovatively embedded at the output end of the shared layer. After the shared layer output h2 enters the SE module, it first performs global information aggregation on the feature channels through an adaptive average pooling operation, compresses the feature map of each channel into a single value, and obtains the global feature information of each channel. Assume that the data dimension input to the SE module is h2=ReLU(W2h1+b2) (the features after the fully connected layer can be regarded as height=width=1, height is height, width is width). After adaptive average pooling, the data dimension becomes [batch, channels, 1, 1]. Batch is the number of samples. Channels is the number of feature channels. Subsequently, these global feature information passes through two fully connected layers. The first fully connected layer reduces the number of channels to the original 1 / reduction (reduction is a hyperparameter, usually set to 16), and the activation function uses ReLU. Then the second fully connected layer restores the number of channels to the original number of channels, and the activation function uses Sigmoid to generate channel attention weights. Finally, by multiplying the generated channel attention weights with the original input features, the features of different channels are re-weighted, key feature channels are highlighted, redundant channels are suppressed, and the quality and efficiency of feature extraction are improved. The shared layer output h2 of the shared layer can be expressed as the target shared layer output after processing by the SE module.

[0107] Multi-head network: The multi-head network includes a magnetic field head network, a current density head network, and a temperature head network.

[0108] Among them, the magnetic field head network: The magnetic field head network focuses on predicting the magnetic field strength and magnetic flux density. The target shared layer output First, the weight matrix W of the magnetic field head network is H With the bias vector b H The linear transformation of the magnetic field intensity prediction value H is obtained, that is, Then, the target shared layer output Then through the weight matrix W B With the bias vector b B The linear transformation of the magnetic flux density is obtained to obtain the predicted value B, that is, The SE mechanism is integrated into the magnetic field head network (i.e., the SE module is embedded at the output end of the magnetic field head network). After the linear transformation of the magnetic field intensity prediction value and the magnetic flux density prediction value obtained above, the obtained feature data is input into the SE module. The SE module follows similar operations as those in the shared layer. It first performs adaptive average pooling to obtain global features, then generates channel attention weights through the fully connected layer, and finally multiplies the input features to the features output by the shared layer (i.e., the target shared layer output). ) to further screen and strengthen the data and improve the accuracy of prediction of magnetic field related physical quantities.

[0109] Current density head network: responsible for predicting current density. The target shared layer output The weight matrix W passing through the current density head network J With the bias vector b J The linear transformation of is used to obtain the predicted value of current density J, that is, The SE module is embedded at the output of the current density head network to process the current density prediction value J. Through operations such as adaptive average pooling, weight generation by the fully connected layer, and weight-feature multiplication in the SE module, the improved physical information neural network can more accurately capture key features related to current density and optimize prediction results.

[0110] Temperature head network: used to predict temperature. The target shared layer output of the shared layer output The weight matrix W of the temperature head network T With the bias vector b T The linear transformation of , we get the temperature prediction value T, that is, The SE module is embedded at the output end of the temperature head network. The SE module performs a series of operations on the temperature prediction value, enabling the improved physical information neural network to better focus on the core features that affect the temperature distribution and improve the accuracy of temperature prediction.

[0111] The above input layer, the shared layer with the SE module and the multi-head network with the SE module are connected to complete Figure 2 The construction of a multi-head physical information neural network model based on the SE mechanism is shown in the figure. Among them, the SE module added to the shared layer is also called the first compression excitation module, and the SE module added to the multi-head network is also called the second compression excitation module.

[0112] In step 206, all groups of model parameters of the improved physical information neural network are used as all candidate solutions to construct an initial gorilla population, simulate the social and collaborative processes in the gorilla tribe, determine the optimal solution corresponding to the optimal individual from the gorilla population, and use the set of model parameters corresponding to the optimal solution as the initial values ​​of the model parameters of the improved physical information neural network.

[0113] Step 208 : Using the data set to train the improved physical information neural network to obtain a trained improved physical information neural network, wherein the loss function during training includes electromagnetic loss, heat conduction loss, and mean square error loss.

[0114] The model training process corresponding to step 206 and step 208 is described in detail below.

[0115] First, as Figure 2 The definition of the electromagnetic-thermal coupling field loss function during the training of the improved physical information neural network is shown. The electromagnetic-thermal coupling field loss function (referred to as the loss function) during training includes electromagnetic loss, thermal conduction loss, and mean square error.

[0116] Electromagnetic loss: Calculate electromagnetic loss based on Maxwell's equations. Specifically, Ampere's law For example, the loss is defined by calculating its residual. In actual calculations, due to the limitations of numerical calculations, for discrete data points, the gradient is approximated. For example, in one-dimensional case, the gradient of the magnetic field intensity can be approximately calculated by the central difference method as (Where, is the gradient of the magnetic field intensity of the i-th sample, H i+1 is the predicted value of the magnetic field intensity of the i+1th sample, H i-1 is the predicted value of the magnetic field intensity of the i-1th sample, and Δx is the spatial step length). Then the electromagnetic loss L em The calculation formula is Where N is the number of samples, H i and J i are the predicted magnetic field intensity and current density for the i-th sample, respectively. After introducing the SE mechanism, the model learns the relevant features of magnetic field and current density more efficiently, and can more accurately capture key information related to electromagnetics. This helps to more accurately calculate electromagnetic losses and improves the fit of the physical information neural network to the laws of electromagnetic physics.

[0117] Heat conduction loss: According to the heat conduction equation To calculate the heat conduction loss. Also in numerical calculations, approximate methods are used to calculate the temperature gradient and divergence. For example, in two-dimensional cases, the second-order derivative of the temperature prediction value can be approximated by the finite difference method. The calculation formula for heat conduction loss is Among them L th is the heat conduction loss, is the second-order derivative of the temperature prediction value, which is used as the heat diffusion term, T i is the temperature prediction value of the i-th sample. i ·J i / σ is the heat source term, σ is the electrical conductivity, and k is the thermal conductivity coefficient. The SE mechanism enables the model to more sensitively perceive key factors when processing temperature-related features. By screening and strengthening these features, it optimizes the calculation of heat conduction losses and enhances the physical information neural network's ability to simulate thermophysical processes.

[0118] Total loss function: The total loss function integrates electromagnetic loss, thermal conduction loss, and traditional mean square error loss. The mean square error loss is used to measure the difference between the model's predicted value and the true value. The calculation formula of the total loss function L is: where y true，i is the actual current density of the i-th sample, y pred，i is the predicted value of the current density of the i-th sample. The SE mechanism improves the accuracy of the prediction of each physical quantity as a whole, making the improved physical information neural network more accurate in calculating electromagnetic loss, thermal conduction loss, and mean square error loss, thereby optimizing the calculation of the total loss function and prompting the improved physical information neural network to iterate in a more optimal direction. The convergence curve of the value of the total electromagnetic heat loss function of the neural network (i.e., the total loss value) is shown in the figure below. Figure 3 When the total loss function, electromagnetic loss, heat conduction loss, and mean square error loss are all less than the corresponding thresholds, the training is completed.

[0119] During the training process, an optimization algorithm is selected and model training and verification are performed.

[0120] Specifically, if Figure 2 As shown, the embodiment of the present invention adopts ADAM optimizer (adaptive moment estimation) combined with GTO (Gorilla Troops Optimizer) to update model parameters and calculate prediction values, so as to give full play to the advantages of both and improve the training effect of the physical information neural network.

[0121] Among them, the ADAM optimizer can significantly improve the efficiency and stability of model training by dynamically adjusting the learning rate. Specifically, the first-order moment estimate m of the tth iteration is calculated. t and the second-order moment estimate v of the t-th iteration t The first-order moment estimate m of the tth iteration t is the gradient g t The exponentially weighted moving average of t =β1m t-1 +(1-β1)g t , m t-1 is the first-order moment estimate of the t-1th iteration, and the second-order moment estimate v of the tth iteration t is the gradient g t The exponentially weighted moving average of the square is calculated as Where β1 is the first attenuation rate, β2 is the second attenuation rate, usually β1 = 0.9, β2 = 0.999; g t is the gradient at the current moment of the t-th iteration, corresponding to the slope of the total loss function.

[0122] Specifically, the first-order moment estimate and the second-order moment estimate are corrected to obtain the corrected first-order moment estimate (i.e., the first-order moment estimate correction value) and the corrected second-order moment estimate (i.e., the second-order moment estimation correction value), the calculation formulas are Where t is the number of iterations.

[0123] Finally, the model parameters θ of the tth iteration are updated according to the revised first-order moment estimate and second-order moment estimate t , the calculation formula is Where α is the learning rate, ∈ is a very small constant (also called a very small constant) used to avoid the denominator being zero, usually ∈ = 10 -8 θ t-1 are the model parameters for the t-1th iteration.

[0124] Specifically, the GTO algorithm simulates the social and collaborative processes in a gorilla tribe to find the optimal solution. The main steps in improving the physical information neural network model training are as follows:

[0125] Specifically, a population is initialized (i.e., multiple candidate solutions), each corresponding to a set of model parameter values. The model parameters are expanded into vector form in a specific manner to construct an initial gorilla population (referred to as the initial population). The position of each individual gorilla is a vector, which is expanded from a set of model parameter values. At the same time, the initial optimal solution (corresponding to the silverback gorilla) is determined. Generally, the solution corresponding to an individual in the initial population is selected as the starting optimal solution, and its fitness is calculated.

[0126] Specifically, the iterative process then begins. In each iteration, a preliminary candidate solution (GX) is generated for each individual in the population (i.e., the position vector of the gorilla corresponding to each candidate solution) according to the GTO algorithm rules. For example, by judging the size relationship between the randomly generated parameter C and the preset value W, it is decided whether to adopt the follow-the-silverback strategy or the compete-for-adult-female strategy to generate the preliminary candidate solution. If C ≥ W, the follow-the-silverback strategy is adopted, and the preliminary candidate solution is calculated using the relevant formula; if C < W, the compete-for-adult-female strategy is adopted, and the preliminary candidate solution is generated according to the corresponding formula. After the preliminary candidate solution is generated, it is applied to the model parameters and the fitness of the preliminary candidate solution is calculated (i.e., the total loss value obtained by the total loss function). If the fitness of the preliminary candidate solution is better than the fitness of the current individual, the current individual is updated to the preliminary candidate solution. At the same time, if the fitness of the preliminary candidate solution is better than the fitness of the current best solution (the silverback gorilla), the best solution is updated to the preliminary candidate solution.

[0127] Specifically, the data obtained using the finite element model in the above data collection step is divided into a training set and a validation set according to a certain ratio (for example, 80% training set and 20% validation set). In the early stage of training, the GTO algorithm is first used to iteratively optimize the model a certain number of times. By simulating the behavior of a gorilla tribe, a global search is performed within a larger solution space to explore a better range of model parameter values, laying a good foundation for subsequent optimization. After the GTO algorithm completes the early optimization, the GTO algorithm determines a set of model parameters corresponding to the optimal solution corresponding to the optimal individual as the initial values ​​of the model parameters of the improved physical information neural network, and then enters the training phase combined with the ADAM optimizer. In this phase, the model is iteratively trained using the data from the training set. In each iteration, the model's predicted value is calculated through forward propagation, and the total loss value is calculated based on the above-mentioned total loss function. The model parameters are then updated using the ADAM optimizer based on the total loss value through backpropagation. At the same time, after each training cycle, the improved physical information neural network is verified using the validation set data. The loss value and other evaluation metrics (such as mean squared error and mean absolute error) on the validation set are calculated to monitor the model's generalization ability and performance. If the loss value on the validation set stops decreasing or begins to increase, it indicates that the improved physical information neural network may be overfitting. In this case, training can be stopped early and the current optimal model parameters can be saved. The current optimal model parameters are assigned to the corresponding parameters in the improved physical information neural network to obtain the trained improved physical information neural network.

[0128] In step 210, the geometric parameters, material characteristic parameters, and working condition parameters of the flywheel motor collected in real time are input into the trained improved physical information neural network, and the corresponding magnetic field intensity prediction value, magnetic flux density prediction value, current density prediction value, and temperature prediction value are output.

[0129] In an exemplary embodiment, to achieve efficient and intelligent calculation of the electromagnetic-thermal coupling field of a flywheel motor, after obtaining a trained improved physical information neural network, an intelligent computing system integrating data-driven and physical models can be constructed. This system should have functions such as data management, model training, predictive analysis, and visualization. During electromagnetic-thermal analysis, this intelligent computing system can access the trained improved physical information neural network, obtain real-time geometric parameters, material properties, and operating condition parameters of the flywheel motor, input them into the retrieved trained improved physical information neural network, and output corresponding predicted values ​​for magnetic field intensity, magnetic flux density, current density, and temperature. These predicted values ​​are then displayed.

[0130] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0131] The flywheel motor electromagnetic thermal coupling field determination method of the present invention is a flywheel motor electromagnetic thermal coupling field determination method based on the SE-PINN method, also known as a flywheel motor electromagnetic thermal coupling field calculation method based on the multi-head physical information neural network technology introducing the SE mechanism. The method is mainly used in technical fields such as energy storage and conversion, motor design and optimization, and multi-physics field coupling analysis. This calculation method integrates electromagnetic and thermodynamic theories and uses advanced numerical calculation technology to achieve an in-depth understanding and accurate simulation of the interaction of multiple physical fields inside the flywheel motor. Specifically, based on the SE-PINN method, the present invention integrates basic physical laws such as the Maxwell equations in electromagnetism and the heat conduction equation in thermodynamics into the neural network training process. By constructing a special loss function, the neural network strictly follows the physical laws for prediction and calculation while learning data features. This fusion innovation breaks the limitations of traditional methods that either focus on physical models and neglect data learning, or focus on data fitting and neglect physical principles, and gives the model a strong generalization ability, so that it can accurately calculate the electromagnetic thermal coupling field when facing flywheel motors with different structures and working conditions.

[0132] The method of the present invention experimentally collects geometric parameters, material properties, operating conditions, and corresponding electromagnetic thermal output data for various flywheel motors. The input layer integrates the multivariate input features to determine the input dimension. After extracting features through multi-layer full connectivity and Reinforced Luminance (ReLU) activation in the shared layer, an SE mechanism is embedded to reweight channel features through operations such as adaptive average pooling. Electromagnetic loss and thermal conduction loss are defined using approximate calculation methods based on Maxwell's equations and the heat conduction equation, respectively. The Adam optimizer is used to calculate first-order and second-order moment estimates and perform bias correction, updating model parameters according to a specific formula. The model is trained using a training set, with predicted values ​​calculated by forward propagation and gradients calculated by backpropagation based on the total loss function. The Adam optimizer, combined with the GTO algorithm, updates parameters based on this information. The SE mechanism helps the model quickly identify and focus on important features. A validation set is used to evaluate the model's accuracy and generalization. This method overcomes the limitations of traditional methods, enabling the model to accurately calculate the electromagnetic thermal coupling field of flywheel motors under different structures and operating conditions based on physical laws and data characteristics, avoiding frequent adjustments due to changes in model structure or operating conditions, greatly enhancing the model's versatility. For example, even when faced with newly developed, irregular-shaped flywheel motors, the model can accurately calculate based on the physical principles and data characteristics it already understands, eliminating the need to rebuild the model. The neural network adheres to the laws of physics when learning data, ensuring that the calculated results conform to actual physical processes and avoid any contradictions with common physics. This makes the results more reliable and provides a solid data foundation for subsequent motor design and optimization. It can quickly adapt to new operating conditions and materials, providing strong support for the performance analysis of flywheel motors in diverse application scenarios across fields such as smart grids and electric vehicles, meeting the specific performance requirements of different scenarios.

[0133] Based on the same inventive concept, embodiments of the present application also provide a flywheel motor electromagnetic thermal coupling field determination device for implementing the aforementioned flywheel motor electromagnetic thermal coupling field determination method. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more embodiments of the flywheel motor electromagnetic thermal coupling field determination device provided below can be found in the aforementioned limitations of the flywheel motor electromagnetic thermal coupling field determination method and are not further elaborated here.

[0134] In an exemplary embodiment, Figure 4 As shown, a device for determining the electromagnetic-thermal coupling field of a flywheel motor is provided, comprising: a data set construction module, a modeling module, an optimization module, a training module, and a prediction module, wherein:

[0135] A data set construction module is used to construct a data set for the energy storage flywheel. The data set includes geometric parameters, material property parameters, working condition parameters, and electromagnetic heat output data of the flywheel motor. The electromagnetic heat output data includes magnetic field intensity, magnetic flux density, current density, and temperature.

[0136] A modeling module is used to construct an improved physical information neural network, the improved physical information neural network includes an input layer, a shared layer and a multi-head network connected in sequence, a first compression excitation module is added between the shared layer and the multi-head network, the multi-head network includes a magnetic field head network, a current density head network and a temperature head network, the input end of each head network in the multi-head network is respectively connected to the output end of the first compression excitation module, and a second compression excitation module is respectively added to the output end of each head network in the multi-head network;

[0137] an optimization module for using all groups of model parameters of the improved physical information neural network as all candidate solutions to construct an initial gorilla population, simulating the social and collaborative processes in the gorilla tribe to determine the optimal solution corresponding to the optimal individual from the gorilla population, and using the set of model parameters corresponding to the optimal solution as initial values ​​of the model parameters of the improved physical information neural network;

[0138] A training module, for training the improved physical information neural network using the data set to obtain a trained improved physical information neural network, wherein the loss functions during training include electromagnetic loss, heat conduction loss, and mean square error loss;

[0139] The prediction module is used to input the geometric parameters, material characteristic parameters, and working condition parameters of the flywheel motor collected in real time into the trained improved physical information neural network, and output the corresponding magnetic field intensity prediction value, magnetic flux density prediction value, current density prediction value, and temperature prediction value.

[0140] In an exemplary embodiment, in the data set construction module, the geometric parameters include stator outer diameter, stator inner diameter, rotor outer diameter, rotor inner diameter and axial length; the material property parameters include magnetic permeability, electrical conductivity and thermal conductivity; and the working condition parameters include current and speed.

[0141] In an exemplary embodiment, in the modeling module, the shared layer adopts a multi-layer fully connected layer structure, and the multi-layer fully connected layer structure is used to perform linear transformation on the input features from the input layer and then perform activation processing to obtain the intermediate layer output, and to perform linear transformation on the intermediate layer output and then perform activation processing to obtain the shared layer output.

[0142] In an exemplary embodiment, in the training module, the electromagnetic loss calculation process includes: calculating the gradient of the magnetic field intensity corresponding to each sample in the data set; obtaining the current density prediction value corresponding to each sample calculated by the current density head network; and obtaining the electromagnetic loss calculation formula based on Ampere's law in Maxwell's equations. The electromagnetic loss calculation formula is:

[0143]

[0144] Where, Lem is the electromagnetic loss, N is the number of samples in the data set, H i and J i are the predicted values ​​of magnetic field intensity and current density of the i-th sample, respectively. is the gradient of the magnetic field intensity of the i-th sample; the gradient of the magnetic field intensity and the predicted value of the current density are substituted into the electromagnetic loss calculation formula to calculate the electromagnetic loss.

[0145] In an exemplary embodiment, the heat conduction loss calculation process in the training module includes: calculating the thermal diffusion term value of the temperature corresponding to each sample in the data set; calculating the square value of the current density prediction value corresponding to each sample, and calculating the ratio of the square value to the conductivity to obtain the heat source term value corresponding to each sample; and obtaining the heat conduction loss calculation formula based on the heat conduction equation. The heat conduction loss calculation formula is:

[0146]

[0147] Where, L th is the heat conduction loss, is the heat diffusion term, J i ·J i / σ is the heat source term, k is the thermal conductivity, T i is the temperature prediction value of the i-th sample, σ is the electrical conductivity; the heat diffusion term and the heat source term are substituted into the heat conduction loss calculation formula to calculate the heat conduction loss.

[0148] In an exemplary embodiment, in a training module, a data set is used to train an improved physical information neural network to obtain a trained improved physical information neural network, including: dividing the data set into a training set and a validation set; using the training set to iteratively train the improved physical information neural network, wherein in each iteration, first-order moment estimation and second-order moment estimation are calculated; using a first decay rate to correct the first-order moment estimation to obtain a first-order moment estimation correction value; using a second decay rate to correct the second-order moment estimation to obtain a second-order moment estimation correction value; substituting the first-order moment estimation correction value and the second-order moment estimation correction value into a parameter update formula to obtain model parameters under a corresponding number of iterations; and updating the parameters of the improved physical information neural network to the model parameters under a corresponding number of iterations; using a validation set to verify the improved physical information neural network that has completed iterative training, and if the loss value calculated during verification stops decreasing, the training is completed.

[0149] Each module in the aforementioned flywheel motor electromagnetic-thermal coupling field determination device can be implemented in whole or in part via software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in hardware form, or can be stored in a computer device memory in software form, so that the processor can call and execute the corresponding operations of each module.

[0150] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 5 As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface, the display unit and the input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, near field communication (NFC) or other technologies. When the computer program is executed by the processor, a method for determining the electromagnetic thermal coupling field of a flywheel motor is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device casing, or an external keyboard, touchpad or mouse.

[0151] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0152] In an exemplary embodiment, a computer device is further provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0153] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0154] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0155] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile memory and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processor involved in the various embodiments provided herein may be, but are not limited to, a general-purpose processor, a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), a programmable logic unit (PLC), a data processing logic unit based on quantum computing, an artificial intelligence (AI) processor, and the like.

[0156] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0157] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A method for determining the electromagnetic thermal coupling field of a flywheel motor, characterized in that: The method comprises: Constructing a data set for an energy storage flywheel, the data set including geometric parameters, material property parameters, operating condition parameters, and electromagnetic heat output data of the flywheel motor, the electromagnetic heat output data including magnetic field strength, magnetic flux density, current density, and temperature; Constructing an improved physical information neural network, the improved physical information neural network comprising an input layer, a shared layer, and a multi-head network connected in sequence, adding a first compression excitation module between the shared layer and the multi-head network, the multi-head network comprising a magnetic field head network, a current density head network, and a temperature head network, the input end of each head network in the multi-head network being respectively connected to the output end of the first compression excitation module, and adding a second compression excitation module to the output end of each head network in the multi-head network; Using all groups of model parameters of the improved physical information neural network as all candidate solutions to construct an initial gorilla population, simulating the social and collaborative processes in a gorilla tribe to determine the optimal solution corresponding to the optimal individual from the gorilla population, and using the group of model parameters corresponding to the optimal solution as initial values ​​of the model parameters of the improved physical information neural network; Training the improved physical information neural network using the data set to obtain a trained improved physical information neural network, wherein a loss function during training includes electromagnetic loss, heat conduction loss, and mean square error loss; The geometric parameters, material property parameters, and working condition parameters of the flywheel motor collected in real time are input into the trained improved physical information neural network, and the corresponding magnetic field intensity prediction values, magnetic flux density prediction values, current density prediction values, and temperature prediction values ​​are output.

2. The method according to claim 1, characterized in that The geometric parameters include stator outer diameter, stator inner diameter, rotor outer diameter, rotor inner diameter and axial length; the material characteristic parameters include magnetic permeability, electrical conductivity and thermal conductivity; and the working condition parameters include current and rotational speed.

3. The method according to claim 1, characterized in that The shared layer adopts a multi-layer fully connected layer structure, which is used to perform linear transformation on the input features from the input layer and then perform activation processing to obtain the intermediate layer output, and to perform linear transformation on the intermediate layer output and then perform activation processing to obtain the shared layer output.

4. The method according to claim 1, wherein The calculation process of the electromagnetic loss includes: For each sample in the data set, calculating the gradient of the magnetic field intensity corresponding to each sample; Get the current density prediction value corresponding to each sample calculated by the current density head network; According to Ampere's law in Maxwell's equations, the electromagnetic loss calculation formula is obtained: Where, L em is the electromagnetic loss, N is the number of samples in the data set, H i and J i are the predicted values ​​of magnetic field intensity and current density of the i-th sample, respectively. is the gradient of the magnetic field intensity of the i-th sample; The gradient of the magnetic field intensity and the predicted value of the current density are substituted into the electromagnetic loss calculation formula to calculate the electromagnetic loss.

5. The method according to claim 4, characterized in that The calculation process of the heat conduction loss includes: For each sample in the data set, calculating the thermal diffusion term value of the temperature corresponding to each sample; Calculating the square value of the current density prediction value corresponding to each sample, and calculating the ratio of the square value to the conductivity to obtain the heat source term value corresponding to each sample; The heat conduction loss calculation formula is obtained according to the heat conduction equation. The heat conduction loss calculation formula is: Where, L th is the heat conduction loss, is the heat diffusion term, J i ·J i / σ is the heat source term, k is the thermal conductivity, T i is the temperature prediction value of the i-th sample, σ is the conductivity; The heat diffusion term value and the heat source term value are substituted into the heat conduction loss calculation formula to calculate the heat conduction loss.

6. The method according to claim 1, characterized in that The step of training the improved physical information neural network using the data set to obtain a trained improved physical information neural network includes: Dividing the data set into a training set and a validation set; Iteratively training the improved physical information neural network using the training set, wherein in each iteration, a first-order moment estimate and a second-order moment estimate are calculated; the first-order moment estimate is corrected using a first decay rate to obtain a first-order moment estimate correction value; the second-order moment estimate is corrected using a second decay rate to obtain a second-order moment estimate correction value; the first-order moment estimate correction value and the second-order moment estimate correction value are substituted into a parameter update formula to obtain model parameters for a corresponding number of iterations; and the parameters of the improved physical information neural network are updated to the model parameters for the corresponding number of iterations; The improved physical information neural network that has completed iterative training is verified using the verification set. If the loss value calculated during verification stops decreasing, the training is completed.

7. A device for determining the electromagnetic thermal coupling field of a flywheel motor, characterized in that: The device comprises: A data set construction module is used to construct a data set of the energy storage flywheel, wherein the data set includes geometric parameters, material characteristic parameters, working condition parameters and electromagnetic heat output data of the flywheel motor, wherein the electromagnetic heat output data includes magnetic field intensity, magnetic flux density, current density and temperature; A modeling module for constructing an improved physical information neural network, wherein the improved physical information neural network includes an input layer, a shared layer, and a multi-head network connected in sequence, a first compression excitation module is added between the shared layer and the multi-head network, the multi-head network includes a magnetic field head network, a current density head network, and a temperature head network, the input end of each head network in the multi-head network is respectively connected to the output end of the first compression excitation module, and a second compression excitation module is respectively added to the output end of each head network in the multi-head network; an optimization module, configured to use all groups of model parameters of the improved physical information neural network as all candidate solutions to construct an initial gorilla population, simulate the social and collaborative processes in the gorilla tribe, determine the optimal solution corresponding to the optimal individual from the gorilla population, and use the group of model parameters corresponding to the optimal solution as initial values ​​of the model parameters of the improved physical information neural network; a training module, configured to train the improved physical information neural network using the data set to obtain a trained improved physical information neural network, wherein the loss function during training includes electromagnetic loss, heat conduction loss, and mean square error loss; The prediction module is used to input the geometric parameters, material characteristic parameters, and working condition parameters of the flywheel motor collected in real time into the trained improved physical information neural network, and output the corresponding magnetic field intensity prediction value, magnetic flux density prediction value, current density prediction value, and temperature prediction value.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images