Work system S3-S5 motor temperature rise test method and device capable of shortening intermittent period

Abstract

The invention discloses a working system S3-S5 motor temperature rise test method and device capable of shortening the interruption period, and relates to the technical field of motor test. A long-term problem is split into a plurality of independent short-term problems by the PPINN model, and the short-term problems are supervised by a rapid coarse-grained (CG) solver, so that the high calculation efficiency of training the neural network by using small data sets can be utilized to the maximum extent, the training time can be shortened, a big data set can be decomposed into a plurality of small data sets for training by the PPINN model, and the training efficiency is improved. Therefore, under the condition of not enough GPU resources, the PINN can still be trained by using a big data set on the CPU resources, data acquired by the equivalent temperature rise test is used for prediction model training and prediction of equivalent motor temperature rise test time and a temperature rise value, only one time of measurement is needed for one time of temperature rise test, and compared with a traditional method, the experiment time can be greatly shortened, and the test efficiency is improved. The test efficiency is improved, and the labor intensity and cost are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of motor testing, and in particular to a method and device for testing the temperature rise of a motor in an S3-S5 motor operating mode with a shortened intermittent cycle. Background Art

[0002] Existing motor temperature rise prediction methods, such as the commonly used equivalent thermal circuit method and integrated parameter thermal model method, have good effects on continuous duty motors, and can more intuitively reflect the heat transfer relationship between the internal components of the motor. For motors with complex structures, simplified modeling can be performed, and the method has relatively high calculation efficiency and can obtain approximate results of temperature rise in a short time. However, due to the complex loads of intermittent periodic duty (S3-S5) motors, and the short-term heat dissipation during the power-off and stop period of each cycle, it takes more time to reach a thermal equilibrium state than continuous duty motors. The above methods have low adaptability, large error in results, and low test efficiency.

[0003] The existing finite element analysis method uses a mathematical discretization method to split the motor model into multiple small unit modules, so that the macroscopic problem of the temperature field is transformed into a microscopic problem, and the temperature distribution inside the motor can be accurately simulated. It can take into account multiple factors such as the geometric shape, material properties, boundary conditions, etc. of the motor, and has high adaptability to motors with complex structures and special working conditions. It can also consider the interaction of multiple physical fields at the same time, and the results are visualized, and the prediction results are more accurate. However, due to the need to establish a detailed motor model, including geometric models, material properties, boundary conditions, etc., the modeling process is relatively complicated and requires a lot of time and effort. In addition, the finite element analysis has a large amount of calculation, high requirements on computer performance, and a long calculation time, which is not suitable for fast and real-time temperature rise prediction.

[0004] Whether it is the equivalent thermal circuit method, the integrated parameter thermal model method or the finite element analysis method, it takes a lot of time to predict the temperature rise of the motor in the intermittent cycle working system (S3-S5). The shortcomings of the existing technology;

[0005] Therefore, a new solution to the above problems needs to be proposed. Summary of the invention

[0006] The purpose of the present invention is to provide a method and device for testing the temperature rise of an S3-S5 motor in a shortened intermittent cycle working mode, so as to solve the technical problems raised in the background technology.

[0007] To achieve the above object, the present invention provides the following technical solution: a method for testing the temperature rise of a motor in a shortened intermittent cycle working mode S3-S5, comprising at least the following steps:

[0008] S1: Build the PINN model to determine the corresponding loss function;

[0009] S2: Construct a particle swarm optimization algorithm to optimize the loss function in the PINN model;

[0010] S3: Based on the PINN model, the problem to be solved is decomposed into many independent problems, and the PPINN model is constructed. The collected data is substituted into the PPINN model, and then the PSO algorithm model is substituted into it to optimize the loss function of the PPINN model. The problem is then solved by a fast coarse-grained solver to finally obtain a temperature rise prediction model.

[0011] S4: Input the 9 parameters of any intermittent cycle working mode S3-S5 motor into the temperature rise prediction model, and the model can give the maximum temperature rise required for the equivalent temperature rise test and the time required to reach the maximum temperature rise.

[0012] Furthermore, the construction of the PINN model to determine the corresponding loss function includes at least the following steps:

[0013] The loss function of PINN includes the mismatch between the state variables or boundary conditions and the initial conditions in the given data. The state variables or boundary conditions are BC, and the initial conditions are IC. See the following formula:

[0014]

[0015] Where: u * is the given data;

[0016] MSE {u,BC,IC} represents the mean square error for boundary conditions, initial conditions and state variables u;

[0017] N u is the sample size, which indicates the number of sample points under a specific input;

[0018] u(x,t) is the output of the neural network, representing the predicted value at position x and time t;

[0019] MSE is used to measure the difference between the predicted value and the true value;

[0020] MSE{u} is the mean square error associated with the function value. This part is usually used to measure the difference between the function value output by the neural network and the function value of the known data point. When there is a set of data points (x i ,y i ), where y i It is at point x i The true function value is u, and the output of the neural network is u θ (x i ), where θ is the parameter of the neural network, then MSE{u} is used to evaluate this difference;

[0021] MSE{BC} is the mean square error associated with boundary conditions. When the physical problem (such as partial differential equations) being processed has boundary conditions, MSE{BC} is used to measure the difference between the output of the neural network at the boundary point and the given boundary conditions.

[0022] MSE{IC} is the mean square error associated with the initial conditions. In problems involving dynamic physical processes (such as physical systems evolving over time), MSE{IC} is used to measure the difference between the output of the neural network at the initial moment and the given initial conditions.

[0023] Based on the above formula plus the residual of the partial differential equation calculated at a set of random points in the space-time domain, the residual term of the partial differential equation is as follows:

[0024]

[0025] Where: R(x,t) is a function expression of spatial variables and time variables, representing the residual term of the physical equation, N is the number of samples, and R represents the residual;

[0026] Then the loss function of PINN is trained by minimizing the total loss, and the following formula is obtained:

[0027] MSE = MSE {u，BC，IC} +MSE R

[0028] In PINN, there are solutions to the forward problem and solutions to PINN. Inverse Problem ,

[0029] For solving the forward problem, the first term represents the mismatch of the NN output u(x,t) from the boundary and / or initial conditions, i.e., MSE BC，IC ;

[0030] For solving Inverse Problem , the first term considers the mismatch u(x,t) of NN output from other datasets, namely MSE U .

[0031] Furthermore, the S2 at least includes the following steps:

[0032] S2.1: Initialize the population. This process requires initializing various parameters, which include at least the upper and lower limits of the target search space, two learning factors, the maximum number of iterations of the algorithm or the convergence accuracy point, the upper and lower limits of the speed of each particle, and the position and speed of each particle in the randomly initialized population;

[0033] S2.2: Calculate the fitness value of each particle according to the fitness function, save the optimal position of each particle, and save the best fitness value and the best position of all particles in the population;

[0034] S2.3: Update the speed and position according to the speed and position update formula;

[0035] S2.4: Calculate the fitness value of each particle after the update, and compare the best fitness value of each particle with the fitness value at its historical optimal position. If the result is better, then use its current position as the optimal position of the particle;

[0036] S2.5: Compare each particle, and compare the fitness value corresponding to its optimal position with the optimal fitness value of the population. If the result is better, update the optimal position and the optimal fitness value of the population.

[0037] S2.6: Determine whether the searched result meets the stopping condition, that is, whether the maximum number of iterations is reached or the accuracy requirement is met. If the stopping condition is met, the optimal value is output, otherwise go to S2.3 and continue running until the condition is met.

[0038] Furthermore, the process of substituting the collected data into the PPINN model and optimizing its loss function using the PSO algorithm model at least includes the following steps:

[0039] Collect data through traditional temperature rise test and equivalent temperature rise test;

[0040] The conventional temperature rise test is the maximum temperature rise obtained and the time required to reach the maximum temperature rise;

[0041] The equivalent temperature rise test is the highest temperature rise obtained by multiple measurements and the time required to reach the highest temperature rise;

[0042] Taking the collected data as input;

[0043] The PSO algorithm is used to optimize the loss function. The next step is determined based on whether the optimized parameters are appropriate. If the convergence condition is reached, the process ends; otherwise, the hidden layer is input and the cycle continues.

[0044] Furthermore, the application of the PPINN model includes at least the following steps:

[0045] S3.1: Simplify the partial differential equations solved by the fast coarse-grained solver, i.e., the CG solver, by removing the variables / multi-scaling coefficients with constants;

[0046] S3.2: CG PINN is used to serially solve the stochastic partial differential equation in the entire time domain to obtain the initial solution. By using fewer residual points and a smaller neural network to solve the stochastic partial differential equation instead of the original partial differential equation, the computational cost in CG PINN can be significantly reduced;

[0047] S3.3: Decompose the time domain into N subdomains;

[0048] Assumptions Known as t k ≤t i ≤t N (including k=0, i.e., the initial iteration), which is used as the initial condition for running N refined PINNs in parallel;

[0049] Once all t i The refined solution at can be used to calculate t i The difference between the coarse solution and the fine solution;

[0050] Then CGPINN is run continuously to update the solution u for each interface between two adjacent subdomains, i.e. The algorithm is iterated until the following conditions are met:

[0051]

[0052] Where: E tol is a user-defined tolerance.

[0053] Indicates the approximate kth iteration t i The exact solution of the moment.

[0054] E is the tolerance.

[0055] Furthermore, the temperature rise prediction model includes nine input nodes and two output nodes;

[0056] The nine input nodes include current, voltage, speed, ambient temperature, motor type, cooling method, power frequency, power factor and current temperature rise;

[0057] The two output nodes include the maximum temperature rise and the time required to reach the maximum temperature rise.

[0058] A temperature rise test device for S3-S5 motors in a shortened intermittent cycle working system, comprising at least a relationship building module, a data acquisition module, a determination module, a substitution module and a prediction module;

[0059] The relationship building module is used to encode physical conservation laws and prior physical knowledge into the neural network to establish a PINN architecture;

[0060] The data acquisition module is used to acquire necessary data such as the maximum temperature rise and the time required to reach the maximum temperature rise measured in the traditional temperature rise test and the equivalent temperature rise test, and encode it into a data set form that can be directly input;

[0061] The determination module is used to determine the loss function, determine the activation function, initialize the values ​​of the neural network, determine the boundary conditions, decompose and divide the time domain into N subdomains, and input data to build the PPINN model;

[0062] The substitution module is used to input the determined data into the prediction model;

[0063] The prediction module is used to predict the temperature rise of the intermittent cycle working mode S3-S5 motor based on the prediction model.

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] 1. The prediction model in the present invention has a greatly reduced reliance on the amount of data required for training, and a prediction model with high accuracy can be trained with less data;

[0066] 2. The PINN model in the present invention is a meshless method because no mesh is required to solve the equation, which saves a lot of effort in generating meshes in traditional PDE solvers;

[0067] 3. The PINN model in the present invention uses both data and physical constraints to further improve the prediction accuracy of the model;

[0068] 4. The PPINN model in the present invention splits a long-term problem into many independent short-term problems, which are supervised by a fast coarse-grained (CG) solver, so it can maximize the high computational efficiency of training neural networks using small data sets and shorten the training time;

[0069] 5. The PPINN model in the present invention can decompose a large data set into multiple small data sets for training. Therefore, even if there are not enough GPU resources, the large data set can still be used to train PINN on CPU resources.

[0070] 6. The data collected in the equivalent temperature rise test of the present invention is used for prediction model training to predict the equivalent motor temperature rise test time and temperature rise value. One temperature rise test only requires one measurement, which can greatly reduce the experimental time, improve the test efficiency, and reduce the labor intensity and cost compared with the traditional method (the traditional method takes about 9 hours). BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0072] Figure 1 It is a structural diagram of the PINN method of the present invention;

[0073] Figure 2It is a flow chart of the particle swarm optimization algorithm of the present invention;

[0074] Figure 3 A schematic diagram of the PPINN algorithm of the present invention;

[0075] Figure 4 It is a prediction model determined by the present invention according to the working principle of the motor. DETAILED DESCRIPTION

[0076] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0077] Embodiment 1:

[0078] A method for testing the temperature rise of an S3-S5 motor in a shortened intermittent cycle working mode, comprising at least the following steps:

[0079] See also Figure 1 , S1: Build the PINN model to determine the corresponding loss function;

[0080] Building a PINN model to determine the corresponding loss function includes at least the following steps:

[0081] The loss function of PINN includes the mismatch between state variables or boundary conditions and initial conditions in the given data. The state variables or boundary conditions are BC and the initial conditions are IC, see the following formula:

[0082]

[0083] Where: u * is the given data;

[0084] MSE {u,BC,IC} represents the mean square error for boundary conditions, initial conditions and state variables u;

[0085] N u is the sample size, which indicates the number of sample points under a specific input;

[0086] u(x,t) is the output of the neural network, representing the predicted value at position x and time t;

[0087] MSE is used to measure the difference between the predicted value and the true value;

[0088] MSE{u} is the mean square error associated with the function value. This part is usually used to measure the difference between the function value output by the neural network and the function value of the known data point. When there is a set of data points (x i ,y i ), where y iIt is at point x i The true function value is u, and the output of the neural network is u θ (x i ), where θ is the parameter of the neural network, then MSE{u} is used to evaluate this difference;

[0089] MSE{BC} is the mean square error associated with boundary conditions. When the physical problem (such as partial differential equations) being processed has boundary conditions, MSE{BC} is used to measure the difference between the output of the neural network at the boundary point and the given boundary conditions.

[0090] MSE{IC} is the mean square error associated with the initial conditions. In problems involving dynamic physical processes (such as physical systems evolving over time), MSE{IC} is used to measure the difference between the output of the neural network at the initial moment and the given initial conditions.

[0091] Based on the above formula plus the residual of the partial differential equation calculated at a set of random points in the space-time domain, the residual term of the partial differential equation is as follows:

[0092]

[0093] Where: R(x,t) is a function expression of spatial variables and time variables, representing the residual term of the physical equation, N is the number of samples, and R represents the residual;

[0094] Then the loss function of PINN is trained by minimizing the total loss, and the following formula is obtained:

[0095] MSE = MSE {u，BC，IC} +MSE R

[0096] In PINN, there are solutions to the forward problem and solutions to PINN. Inverse Problem ,

[0097] For solving the forward problem, the first term represents the mismatch of the NN output u(x,t) from the boundary and / or initial conditions, i.e., MSE BC，IC ;

[0098] For solving Inverse Problem , the first term considers the mismatch u(x,t) of NN output from other datasets, namely MSE U ;

[0099] The PINN model depends partly on the data and partly on Partial differential equations(PDE) describes the physical field. Unlike traditional PDE solvers, although the PDE is encoded in the neural network, PINN does not need to discretize the PDE or use complex numerical algorithms to solve the equation. Instead, PINN can use automatic differentiation adopted in back-propagation to express all differential operators in the PDE and perform nonlinear mapping from input variables to output quantities by optimizing the loss function.

[0100] See also Figure 2 ,S2: Build a particle swarm optimization algorithm to optimize the loss function in the PINN model;

[0101] S2.1: Initialize the population. This process requires initializing various parameters, which include at least the upper and lower limits of the target search space, two learning factors, the maximum number of iterations of the algorithm or the convergence accuracy point, the upper and lower limits of each particle's velocity, and the position and velocity of each particle in the randomly initialized population;

[0102] S2.2: Calculate the fitness value of each particle according to the fitness function, save the optimal position of each particle, and save the best fitness value and the best position of all particles in the population;

[0103] S2.3: Update the speed and position according to the speed and position update formula;

[0104] S2.4: Calculate the fitness value of each particle after the update, and compare the best fitness value of each particle with the fitness value at its historical optimal position. If the result is better, then use its current position as the optimal position of the particle;

[0105] S2.5: Compare each particle, and compare the fitness value corresponding to its optimal position with the optimal fitness value of the population. If the result is better, update the optimal position and the optimal fitness value of the population.

[0106] S2.6: Determine whether the searched result meets the stopping condition, that is, whether the maximum number of iterations is reached or the accuracy requirement is met. If the stopping condition is met, the optimal value is output, otherwise go to S2.3 and continue running until the condition is met.

[0107] S3: Based on the PINN model, the problem to be solved is decomposed into many independent problems, and the PPINN model is constructed. The collected data is substituted into the PPINN model, and its loss function is optimized by the PSO algorithm model. Then, the fast coarse-grained solver is used to solve the problem, and finally the temperature rise prediction model is obtained;

[0108] See also Figure 3The PPINN model in S3 decomposes a long-term problem into many independent short-term problems, which are then solved by a fast coarse-grained (CG) solver. This can fully utilize the high computational efficiency of training neural networks with small data sets; compared with the original PINN model, it can achieve significant acceleration in solving long-term physical problems;

[0109] Substituting the collected data into the PPINN model and optimizing its loss function using the PSO algorithm model includes at least the following steps:

[0110] Collect data through traditional temperature rise test and equivalent temperature rise test;

[0111] The traditional temperature rise test is the maximum temperature rise obtained and the time required to reach the maximum temperature rise;

[0112] The equivalent temperature rise test is the highest temperature rise obtained from multiple measurements and the time required to reach the highest temperature rise;

[0113] Taking the collected data as input;

[0114] The PSO algorithm is used to optimize the loss function of the PPINN model. The next step is determined based on whether the optimized parameters are appropriate. If the convergence condition is reached, the process ends; otherwise, the hidden layer is input and the cycle continues.

[0115] The application of the PPINN model includes at least the following steps:

[0116] S3.1: Simplify the PDEs solved by the fast coarse-grained solver, i.e., the CG solver, by removing variables / multi-scaling coefficients with constants. For example, the nonlinear coefficients in the diffusion equation shown in the following experiments can be replaced. Remove variables / multi-scaling coefficients with constants. For example, CG PINN can be used as a fast CG solver, but standard fast finite difference solvers can also be explored. ;

[0117] S3.2: CG PINN is used to serially solve the stochastic partial differential equation in the entire time domain to obtain the initial solution. By using fewer residual points and a smaller neural network to solve the stochastic partial differential equation instead of the original partial differential equation, the computational cost in CG PINN can be significantly reduced;

[0118] S3.3: Decompose the time domain into N subdomains;

[0119] Assumptions Known as t k ≤t i ≤t N (including k=0, i.e., the initial iteration), which is used as the initial condition for running N refined PINNs in parallel;

[0120] Once all ti The refined solution at can be used to calculate t i The difference between the coarse solution and the fine solution;

[0121] Then CGPINN is run continuously to update the solution u for each interface between two adjacent subdomains, i.e. The algorithm is iterated until the following conditions are met:

[0122]

[0123] Where: E tol is a user-defined tolerance.

[0124] Indicates the approximate kth iteration t i The exact solution of the moment.

[0125] E is the tolerance.

[0126] S4: Input the 9 parameters of any intermittent cycle working mode S3-S5 motor into the temperature rise prediction model, and the model can give the maximum temperature rise required for the equivalent temperature rise test and the time required to reach the maximum temperature rise.

[0127] See also Figure 4 ,The temperature rise prediction model includes nine input nodes and two output nodes;

[0128] The nine input nodes include current, voltage, speed, ambient temperature, motor type, cooling method, power frequency, power factor and current temperature rise;

[0129] The two output nodes include the maximum temperature rise and the time required to reach the maximum temperature rise.

[0130] Embodiment 2:

[0131] Based on the above embodiment, a temperature rise test device for S3-S5 motor with shortened intermittent cycle working mode is further proposed;

[0132] A temperature rise test device for S3-S5 motors in a shortened intermittent cycle working system, comprising at least a relationship building module, a data acquisition module, a determination module, a substitution module and a prediction module;

[0133] The relationship building module is used to encode physical conservation laws and prior physical knowledge into the neural network to build the PINN architecture;

[0134] The data acquisition module is used to obtain necessary data such as the maximum temperature rise and the time required to reach the maximum temperature rise measured under the traditional temperature rise test and the equivalent temperature rise test, and encode them into a data set form that can be directly input;

[0135] The determination module is used to determine the loss function, determine the activation function, initialize the values ​​of the neural network, determine the boundary conditions, decompose and divide the time domain into N subdomains, and input data to build the PPINN model;

[0136] The substitution module is used to input the determined data into the prediction model;

[0137] The prediction module is used to predict the temperature rise of S3-S5 motors in intermittent cycle working mode based on the prediction model.

[0138] Embodiment three:

[0139] This embodiment further proposes a computer-readable storage medium based on the above embodiment;

[0140] A computer-readable storage medium stores computer instructions that can be read by a computer. After the prediction model is built, the maximum temperature rise of the motor of this type and the time required to reach the maximum temperature rise can be predicted through computer instructions by simply inputting current, voltage, speed, ambient temperature, motor type, cooling method, power frequency, power factor and current temperature rise.

[0141] In summary:

[0142] The present invention proposes a method for shortening the temperature rise detection of intermittent cycle working system (S3-S5) motors. Data is collected by carrying out traditional temperature rise tests (the highest temperature rise obtained and the time required to reach the highest temperature rise) and equivalent temperature rise tests (the highest temperature rise obtained by multiple measurements and the time required to reach the highest temperature rise) on intermittent cycle working system (S3-S5) motors.

[0143] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A method for testing the temperature rise of a motor in a shortened intermittent cycle working system S3-S5, characterized in that: At least the following steps are included: S1: Build the PINN model to determine the corresponding loss function; S2: Construct a particle swarm optimization algorithm to optimize the loss function in the PINN model; S3: Based on the PINN model, the problem to be solved is decomposed into many independent problems, and the PPINN model is constructed. The collected data is substituted into the PPINN model, and then the PSO algorithm model is substituted into it to optimize the loss function of the PPINN model. The problem is then solved by a fast coarse-grained solver to finally obtain a temperature rise prediction model. S4: Input the 9 parameters of any intermittent cycle working mode S3-S5 motor into the temperature rise prediction model, and the model can give the maximum temperature rise that can be achieved in the equivalent temperature rise test and the time required to reach the maximum temperature rise.

2. A method for testing the temperature rise of a motor in a shortened intermittent cycle working mode S3-S5 according to claim 1, characterized in that: The construction of the PINN model to determine the corresponding loss function includes at least the following steps: The loss function of PINN includes the mismatch between the state variables or boundary conditions and the initial conditions in the given data. The state variables or boundary conditions are BC, and the initial conditions are IC. See the following formula: Where: u * is the given data; MSE {u,BC,IC} represents the mean square error for boundary conditions, initial conditions and state variables u; N u is the sample size, which indicates the number of sample points under a specific input; u(x,t) is the output of the neural network, representing the predicted value at position x and time t; MSE is used to measure the difference between the predicted value and the true value; MSE{u} is the mean square error associated with the function value. This part is usually used to measure the difference between the function value output by the neural network and the function value of the known data point. When there is a set of data points (x i ,y i ), where y i It is at point x i The true function value is u, and the output of the neural network is u θ (x i ), where θ is the parameter of the neural network, then MSE{u} is used to evaluate this difference; MSE{BC} is the mean square error associated with boundary conditions. When the physical problem being processed has boundary conditions, MSE{BC} is used to measure the difference between the output of the neural network at the boundary point and the given boundary conditions; MSE{IC} is the mean square error associated with the initial conditions. In problems involving dynamic physical processes, MSE{IC} is used to measure the difference between the output of the neural network at the initial moment and the given initial conditions; Based on the above formula plus the residual of the partial differential equation calculated at a set of random points in the space-time domain, the residual term of the partial differential equation is as follows: Where: R(x,t) is a function expression of spatial variables and time variables, representing the residual term of the physical equation, N is the number of samples, and R represents the residual; Then the loss function of PINN is trained by minimizing the total loss, and the following formula is obtained: MSE=MSE {u，BC，IC} +MSE R In PINN, there are solutions to the forward problem and solutions to the PINN inverse problem. For solving the forward problem, the first term represents the mismatch of the NN output u(x,t) from the boundary and / or initial conditions, i.e., MSE BC，IC ; For solving the inverse problem, the first term considers the mismatch u(x, t) of the NN output from other datasets, namely MSE U .

3. The method for testing the temperature rise of a motor in a shortened intermittent cycle working mode S3-S5 according to claim 1, characterized in that: The S2 at least comprises the following steps: S2.1: Initialize the population. This process requires initializing various parameters, which include at least the upper and lower limits of the target search space, two learning factors, the maximum number of iterations of the algorithm or the convergence accuracy point, the upper and lower limits of the speed of each particle, and the position and speed of each particle in the randomly initialized population; S2.2: Calculate the fitness value of each particle according to the fitness function, save the optimal position of each particle, and save the best fitness value and the best position of all particles in the population; S2.3: Update the speed and position according to the speed and position update formula; S2.4: Calculate the fitness value of each particle after the update, and compare the best fitness value of each particle with the fitness value at its historical optimal position. If the result is better, then use its current position as the optimal position of the particle; S2.5: Compare each particle, and compare the fitness value corresponding to its optimal position with the optimal fitness value of the population. If the result is better, update the optimal position and the optimal fitness value of the population. S2.6: Determine whether the searched result meets the stopping condition, that is, whether the maximum number of iterations is reached or the accuracy requirement is met. If the stopping condition is met, the optimal value is output, otherwise go to S2.3 and continue running until the condition is met.

4. The method for testing the temperature rise of a motor in a shortened intermittent cycle working mode S3-S5 according to claim 1, characterized in that: Substituting the collected data into the PPINN model and optimizing its loss function using the PSO algorithm at least includes the following steps: Collect data through traditional temperature rise test and equivalent temperature rise test; The conventional temperature rise test is the maximum temperature rise obtained and the time required to reach the maximum temperature rise; The equivalent temperature rise test is the highest temperature rise obtained by multiple measurements and the time required to reach the highest temperature rise; Taking the collected data as input; The PSO algorithm is used to optimize the loss function. The next step is determined based on whether the optimized parameters are appropriate. If the convergence condition is reached, the process ends; otherwise, the hidden layer is input and the cycle continues.

5. A method for testing the temperature rise of a motor in a shortened intermittent cycle working mode S3-S5 according to claim 4, characterized in that: The application of the PPINN model includes at least the following steps: S3.1: Simplify the partial differential equations solved by the fast coarse-grained solver, i.e., the CG solver, by removing the variables / multi-scaling coefficients with constants; S3.2: CG PINN is used to serially solve the stochastic partial differential equation in the entire time domain to obtain the initial solution. By using fewer residual points and a smaller neural network to solve the stochastic partial differential equation instead of the original partial differential equation, the computational cost in CG PINN is significantly reduced; S3.3: Decompose the time domain into N subdomains; Assumptions Known as t k ≤t i ≤t N (including k=0, i.e., the initial iteration), which is used as the initial condition for running N refined PINNs in parallel; Once all t i The refined solution at t i The difference between the coarse solution and the fine solution; Then CGPINN is run continuously to update the solution u for each interface between two adjacent subdomains, i.e. The algorithm is iterated until the following conditions are met: Where: E tol is the user-defined tolerance; Indicates the approximate k-th iteration t i The exact solution of the moment; E is the tolerance.

6. The method for testing the temperature rise of a motor in a shortened intermittent cycle working mode S3-S5 according to claim 1, characterized in that: The temperature rise prediction model includes nine input nodes and two output nodes; The nine input nodes include current, voltage, speed, ambient temperature, motor type, cooling method, power frequency, power factor and current temperature rise; The two output nodes include the maximum temperature rise and the time required to reach the maximum temperature rise.

7. A temperature rise test device for a motor with a shortened intermittent cycle working system S3-S5, used in a temperature rise test method for a motor with a shortened intermittent cycle working system S3-S5 as described in any one of claims 1-6, characterized in that: At least includes a relationship building module, a data acquisition module, a determination module, a substitution module and a prediction module; The relationship building module is used to encode physical conservation laws and prior physical knowledge into the neural network to establish a PINN architecture; The data acquisition module is used to acquire necessary data such as the maximum temperature rise and the time required to reach the maximum temperature rise measured in the traditional temperature rise test and the equivalent temperature rise test, and encode it into a data set form that can be directly input; The determination module is used to determine the loss function, determine the activation function, initialize the values ​​of the neural network, determine the boundary conditions, decompose and divide the time domain into N subdomains, and input data to build the PPINN model; The substitution module is used to input the determined data into the prediction model; The prediction module is used to predict the temperature rise of the intermittent cycle working mode S3-S5 motor based on the prediction model.

Images