Integrated permanent magnet synchronous electric drive system temperature rise calculation method based on model-data fusion optimization thermal network

Abstract

The invention relates to an integrated permanent magnet synchronous electric drive system temperature rise calculation method based on a model-data fusion optimization thermal network, and belongs to the field of integrated permanent magnet synchronous electric drive system temperature calculation. The method comprises the following steps: firstly, constructing an LPTN model of an electric drive system based on an equivalent thermal network structure of a hollow cylinder, namely a T-shaped equivalent model; solving the model by constructing a thermal conductance matrix, a node temperature correction matrix and a power loss matrix injected into each node; and a result is optimized based on a long short-term memory network-support vector regression algorithm. The method can effectively solve the problem of long time consumption caused by large sample size in the optimization design stage of finite element temperature simulation and the problem of low precision of a traditional thermal network model.

Description

Technical Field

[0001] The present invention belongs to the field of temperature calculation of integrated permanent magnet synchronous electric drive systems, and relates to a method for calculating the temperature rise of an integrated permanent magnet synchronous electric drive system that optimizes a thermal network based on model-data fusion. Background Technique

[0002] An integrated permanent magnet synchronous electric drive system is composed of a permanent magnet synchronous motor and a driver. In the field of thermal management of integrated permanent magnet synchronous electric drive systems, the existing technologies mainly focus on three types of methods: the finite element method, the lumped parameter thermal network method, and data-driven modeling. Finite element numerical simulation conducts three-dimensional transient temperature field simulation of the electric drive system through electromagnetic-thermal coupling field analysis, using a multi-physics field joint solver. Its partial differential equation model based on Maxwell's equations and Fourier's law of heat conduction can accurately analyze the temperature gradient distribution in complex structures such as the end of the stator winding and the permanent magnet air gap region. However, this method requires the establishment of a refined model containing millions of mesh elements. When solving the transient thermal equilibrium equation under non-linear boundary conditions, the single simulation time can reach several hours. For design scenarios involving multi-parameter optimization, the calculation efficiency becomes a significant bottleneck.

[0003] The lumped parameter thermal network (LPTN) method simplifies the three-dimensional heat conduction path of the motor body and the driver into an equivalent thermal resistance network, and uses the node analysis method to establish a system of ordinary differential equations. This method characterizes the key heat transfer paths such as the contact thermal resistance between the stator core and the housing and the interface thermal resistance between the IGBT module and the heat dissipation substrate through thermal resistance parameters, and theoretically can complete temperature prediction within seconds. However, it faces three main challenges in the application of integrated systems: First, the compact layout of the power module and the cooling channel leads to multi-directional heat flow coupling, making it difficult to determine the weight distribution of the axial, radial, and circumferential thermal resistances; Second, there are dynamic differences in the heat source distribution of the IGBT chips and diodes inside the inverter, making it difficult to accurately establish its thermal circuit structure and calculate the thermal resistance value therein; Third, when the motor-driver operates jointly, the interaction between the winding copper loss and the switching loss will change the overall thermal boundary conditions, and the thermal resistance coefficient database based on empirical formulas in the existing literature lacks universality.

[0004] The data-driven modeling technique provides a new idea for solving the above problems. This method extracts features from the temperature rise curves collected in experiments through machine learning algorithms such as neural networks and support vector regression, and inversely identifies the hidden thermal resistance parameters in the LPTN model. For example, a parameter identification system based on the particle swarm optimization algorithm can minimize the root mean square error between the simulated temperature and the measured data by iteratively adjusting the conduction coefficient between the nodes of the thermal network. However, this method has significant engineering limitations: First, constructing a complete training data set requires covering the entire operating condition spectrum such as rated load, overload shock, and frequent start-stop, resulting in an exponential increase in experimental costs; Second, the trained model is essentially a fitting of specific structural parameters. When design variables (such as the height of the housing heat dissipation ribs and the thickness of the winding impregnated paint) change, the model needs to be re-verified through experiments, and parametric rapid iteration cannot be achieved. More critically, when there is no physical prototype in the initial stage of system design, it is difficult for the data-driven method to establish an effective prior model to guide the thermal optimization design. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide an integrated electric drive system temperature rise calculation method based on a model-data hybrid-driven thermal network, which solves the problem that it is difficult to accurately construct the lumped parameter thermal network (LPTN) model of the integrated permanent magnet synchronous electric drive system with the traditional thermal resistance equivalent model. Specifically, by constructing a variable parameter hybrid-driven thermal network model, it effectively solves the problem of long time consumption caused by a large number of samples in the optimization design stage of finite element temperature simulation, and the problem of low accuracy of the traditional thermal network model.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] An integrated permanent magnet synchronous electric drive system temperature rise calculation method based on model-data fusion optimization of the thermal network. First, based on the equivalent thermal network structure of a hollow cylinder, that is, a T-type equivalent model, an LPTN model of the electric drive system is constructed, where LPTN represents the lumped parameter thermal network; then a thermal conductance matrix, a node temperature correction matrix, and a power loss matrix injected into each node are constructed to solve the LPTN model of the electric drive system; finally, the optimization result based on the long short-term memory network-support vector regression algorithm is used.

[0008] Furthermore, the constructed LPTN model of the electric drive system includes component thermal resistance, node temperature, and the power loss injected into each node.

[0009] The component thermal resistance includes driver thermal resistance, end cover thermal resistance, housing thermal resistance, stator thermal resistance, winding thermal resistance, end winding thermal resistance, air gap thermal resistance, rotor and permanent magnet thermal resistance, shaft thermal resistance, and convective thermal resistance.

[0010] The node temperatures include the driver temperature, end cover temperature, housing temperature, stator temperature, winding temperature, end winding temperature, rotor and permanent magnet temperature, and shaft temperature.

[0011] The power losses injected into each node include the inverter circuit power loss, stator iron loss, winding copper loss, end winding copper loss, rotor iron loss, and permanent magnet eddy current loss.

[0012] Furthermore, to solve the LPTN model of the electric drive system, a thermal conductance matrix G n×n , a node temperature matrix T n×1 , and a power loss matrix P n×1 injected into each node are constructed respectively. Among them, the thermal conductance is the reciprocal of the thermal resistance, and n is the total number of nodes in the thermal network. The node equations are written using the node voltage method:

[0013]

[0014] A correction matrix R n×n is introduced to the calculation result of the temperature. Then, the calculation formula for the corrected node temperature matrix T' n×1 is:

[0015]

[0016] Furthermore, the value of the correction matrix R n×n is obtained by training and fitting the electromagnetic-thermal coupling finite element simulation small sample data set collected by deep learning. The sample data used for training all come from the motor design schemes within a certain range of structural parameters.

[0017] Furthermore, the long short-term memory network-support vector regression algorithm specifically includes: dividing the collected finite element simulation data set and the corresponding original thermal network data set into a training set, a test set, and a validation set according to a ratio of 8:1:1, respectively setting the optimization methods of the LSTM and SVR models, and initializing the hyperparameters, setting the upper and lower limits of the optimized hyperparameters and the number of optimization iterations, where LSTM represents the long short-term memory network and SVR represents the support vector machine regression; using the training set to train the model with the given hyperparameters of the model, testing on the validation set, continuously cycling in the hyperparameter selection space, and after finding the optimal hyperparameter models M LSTM and M SVR , recording the errors E LSTM and E SVR of the validation set on the LSTM model and the SVR model respectively. Among them, M LSTM represents the optimal long short-term memory network, M SVR represents the optimal support vector machine regression, E LSTM represents the error of the long short-term memory network model, and E SVRdenote the error of the support vector machine regression model; subsequently, a threshold θ is set, and the errors E LSTM and E SVR are judged. If the errors differ greatly, the model with the smaller error is selected as the final model; if the error between the two models is less than a certain threshold, the two models are combined with weights, and the final prediction result is equal to the weighted sum of the prediction values of the two models.

[0018] The beneficial effects of the present invention are as follows:

[0019] (1) Double breakthroughs in high efficiency and high precision: By means of the model-data hybrid drive method, the present invention combines the advantages of the physical model of finite element simulation with the dynamic correction ability of data drive, effectively solving the problems of low calculation efficiency of traditional finite element simulation in the optimization design stage and large errors in the thermal resistance parameters of traditional LPTN models. Verified by experiments, the temperature prediction error of the LPTN model after introducing the correction matrix is reduced from 20% to less than 3.5%, and at the same time, the calculation speed is greatly improved compared with finite element simulation, achieving a balance between high precision and high efficiency.

[0020] (2) Accurate modeling ability for complex thermal path relationships: Aiming at the multi-component coupled heat sources (such as inverter switching losses, winding copper losses, permanent magnet eddy current losses) and special-shaped structure thermal resistances (axial / radial / circumferential thermal resistance weight differences) inside the integrated electric drive system, the present invention proposes an LPTN model based on a hollow cylinder T-shaped equivalent thermal network. Through the physical modeling of the thermal conductivity matrix and the data-driven compensation of the correction matrix, it breaks through the limitation that traditional LPTN is strongly dependent on empirical formulas for thermal resistance parameters of components such as the machine shell and the driver.

[0021] (3) Intelligent dynamic correction and multi-condition adaptability: The introduction of the LSTM-SVR hybrid algorithm combines the ability of the long short-term memory network to capture the transient temperature rise time series characteristics (such as the thermal accumulation effect under frequent start-stop conditions) and the high-efficiency fitting of support vector regression for non-linear thermal resistance relationships, realizing adaptive temperature prediction for complex conditions (such as overload impact, variable speed operation). Through the dynamic weighted strategy of the error threshold, weighted prediction is adopted when the error of the training set is less than the threshold θ, reducing the average prediction error of the model under multiple conditions to less than 3%, which is significantly better than the single-algorithm model.

[0022] (4) Small-sample data drive and rapid iterative optimization: The generation of the correction matrix only requires small-sample data of finite element simulation. Through deep learning, global compensation is carried out on the thermal resistance parameter deviation, avoiding the dependence of traditional data-driven methods on full-condition experimental data. At the same time, the correction matrix can be dynamically updated with the adjustment of motor structure parameters (such as air gap width, permanent magnet thickness), supporting rapid iteration in the design optimization stage and solving the defects of the fixed structure and non-reusability of traditional data-driven models.

[0023] Other advantages, objects and features of the present invention will be set forth in part in the following description, and in part will be obvious to those skilled in the art upon examination of the following, or may be learned from the practice of the present invention. The objects and other advantages of the present invention may be realized and obtained by the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to make the objects, technical solutions and advantages of the present invention clearer, the present invention will be described in detail preferably with reference to the accompanying drawings, where:

[0025] Figure 1 is the overall flowchart of the integrated electric drive system temperature rise calculation method based on the model-data hybrid-driven thermal network of the present invention;

[0026] Figure 2 is the thermal resistance equivalent model of a general hollow cylinder;

[0027] Figure 3 is the three-dimensional model of the permanent magnet synchronous electric drive system;

[0028] Figure 4 is the LPTN model of the permanent magnet synchronous electric drive system;

[0029] Figure 5 is the result optimization flowchart based on LSTM-SVR;

[0030] Figure 6 is the schematic diagram of network weighted combination;

[0031] Figure 7 is the comparison and error between the original LPTN result and the FEA result;

[0032] Figure 8 is the comparison and error between the LPTN result and the FEA result after introducing the correction matrix. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] The following specific examples illustrate the embodiments of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the drawings provided in the following embodiments only illustrate the basic concept of the present invention schematically, and the following embodiments and the features in the embodiments can be combined with each other without conflict.

[0034] Among them, the accompanying drawings are only for illustrative purposes, showing only schematic diagrams rather than actual drawings, and should not be construed as limiting the present invention; in order to better illustrate the embodiments of the present invention, some components in the accompanying drawings will be omitted, enlarged or reduced, which do not represent the dimensions of the actual product; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the accompanying drawings may be omitted.

[0035] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components; in the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "left", "right", "front", "rear", etc. indicating the orientation or positional relationship, they are based on the orientation or positional relationship shown in the accompanying drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, the terms describing the positional relationship in the accompanying drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those of ordinary skill in the art, the specific meanings of the above terms can be understood according to specific circumstances.

[0036] Please refer to Figures 1 to 8 , the embodiments of the present invention provide an integrated electric drive system temperature rise calculation method based on a model-data hybrid-driven thermal network. First, based on the equivalent thermal network structure of a hollow cylinder, that is, the T-type equivalent model, the LPTN model of the electric drive system is constructed. Since most of the components of a rotating motor are basically cylindrical in shape, the T-type equivalent model is very suitable for constructing the LPTN of the motor.

[0037] The housing, stator, rotor, etc. of the electric drive system can all be approximately considered as Figure 2 (a) The hollow cylinder in, and can be equivalent to Figure 2 (b) The circuit shown, where R 1a , R 2a , R 3a represent the thermal resistance in the axial direction, and R 1r , R 2r and R 3r represent the thermal resistance in the radial direction. The thermal resistance values are calculated by formula (1); T1 and T2 are the temperatures of the outer ring and inner ring surfaces respectively; T3 and T4 are the temperatures of two axial cross-sections respectively.

[0038]

[0039] Among them, r2 and r1 represent the radii of the inner ring and outer ring respectively; k a , k r represent the axial thermal conductivity and radial thermal conductivity respectively, and l represents the length of the hollow cylinder.

[0040] The integrated electric drive system analyzed in this embodiment is composed of a motor, a driver, and a reducer, as shown in Figure 3 . According to the T-type equivalent thermal network model of a hollow cylinder, an LPTN model of the electric drive system as shown in Figure 4 is established. Figure 4 Among them, R1 to R2 are the thermal resistances of the driver, R3 to R4, R 11 to R 12 are the thermal resistances of the end covers, R5 to R 10 are the thermal resistances of the housing, R 13 to R 18 are the thermal resistances of the stator, R 19 to R 23 are the thermal resistances of the windings, R 24 to R 27 are the thermal resistances of the end windings, R 28 to R 30 are the air-gap thermal resistances, R 31 to R 36 are the thermal resistances of the rotor and permanent magnets, R 37 to R 39 are the thermal resistances of the shaft, R 40 to R 53 are the convective thermal resistances; T1 is the temperature of the driver, T2 and T3 are the temperatures of the end covers, T4 is the temperature of the housing, T5 is the temperature of the stator, T6 is the temperature of the windings, T7 and T8 are the temperatures of the end windings, T9 is the temperature of the rotor and permanent magnets, T 10 is the temperature of the shaft; P1 is the power loss of the inverter circuit, P2 is the iron loss of the stator, P3 is the copper loss of the windings, P4 and P5 are the copper losses of the end windings, and P6 is the iron loss of the rotor and the eddy current loss of the permanent magnets.

[0041] To solve this thermal network, a thermal conductance matrix G n×n , a node temperature matrix T n×1 , and a power loss matrix P n×1 injected into each node are constructed respectively. Among them, the thermal conductance is the reciprocal of the thermal resistance, and n is the total number of nodes in the thermal network. The node equation is written using the node voltage method:

[0042]

[0043] Considering that Figure 4 in the LPTN model shown in n×n it is difficult to accurately calculate some of the thermal resistance parameters, which will cause a large error in the calculation result T of formula (2). Therefore, the present invention proposes to introduce a correction matrix R n×n to the calculation result of the temperature, and formula (2) can be changed to:

[0044]

[0045] The proposed correction matrix R n×n, whose value is obtained by training and fitting a small sample dataset of electromagnetic-thermal coupling finite element simulation collected by deep learning. To make the correction matrix applicable to the correction of LPTN temperature results under various motor structure parameters, the sample data used for training are all from motor design schemes within a certain range of structure parameters. Therefore, the fitted R n×n can accurately correct the LPTN temperature result T of any design scheme within a certain range, making the corrected result T' close to the finite element simulation result, so as to accurately replace the finite element simulation.

[0046] To improve the prediction ability of the thermal network, this embodiment proposes a result optimization method based on the LSTM-SVR (Long Short-Term Memory Network-Support Vector Regression) algorithm, and its process is as Figure 5 shown. The LSTM-SVR algorithm combines the powerful modeling ability of LSTM for time series data and the efficient processing ability of SVR for nonlinear regression problems, can learn complex nonlinear relationships from sample data, and achieve precise optimization of the thermal network calculation results, thus significantly improving the prediction accuracy of the thermal network. Figure 6 is the schematic diagram of the weighted combination process of LSTM and SVR.

[0047] The collected finite element simulation dataset and the corresponding original thermal network dataset are divided into a training set, a test set, and a validation set according to the ratio of 8:1:1. The optimization methods of the LSTM and SVR models are respectively set, and the hyperparameters are initialized, and the upper and lower limits of the optimization hyperparameters and the optimization iteration times are set. The hyperparameters of the given model are used to train the model with the training set, and tested on the validation set, continuously cycling in the hyperparameter selection space. After finding the optimal hyperparameter models M LSTM and M SVR , record the errors E LSTM and E SVR on the validation set. Subsequently, a threshold θ is set, and the errors E LSTM and E SVR are judged. If the errors differ greatly, the model with the smaller error is selected as the final model; if the error between the two models is less than a certain threshold, the two models are weighted combined, and the final prediction result is equal to the weighted sum of the prediction values of the two models.

[0048] To verify the accuracy of the correction matrix for correcting the thermal network temperature of the designed models within the specified range, 10 groups of LPTN calculation results and FEA calculation results before and after introducing the correction matrix are collected within the design range, and the result comparison and result error are respectively as Figure 7 and Figure 8 shown.

[0049] From Figure 7It can be seen that there is a large difference between the LPTN temperature calculation results without introducing the correction matrix and the finite element simulation results, and the maximum error is close to 20%. The main reason is that it is difficult to accurately calculate the thermal resistance values of components such as the casing and the driver, resulting in a large error in the final calculation results of the model. And according to Figure 8 It can be seen that the maximum error between the thermal network calculation results after introducing the correction matrix and the finite element simulation does not exceed 3.5%, achieving a good calculation effect and effectively realizing the replacement of the finite element simulation.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the purpose and scope of the present technical solution, and they should all be covered by the scope of the claims of the present invention.

Claims

1. An integrated permanent magnet synchronous electric drive system temperature rise calculation method based on model-data fusion to optimize the thermal network, characterized in that First, based on the equivalent thermal network structure of the hollow cylinder, i.e., the T-shaped equivalent model, the LPTN model of the electric drive system is constructed, where LPTN represents the lumped parameter thermal network; then, the thermal conductance matrix, the node temperature correction matrix, and the power loss matrix injected into each node are constructed to solve the LPTN model of the electric drive system; finally, the optimization result is obtained by using the long short-term memory network-support vector regression algorithm.

2. The temperature rise calculation method of the integrated permanent magnet synchronous electric drive system according to claim 1, wherein The constructed LPTN model of the electric drive system includes component thermal resistance, node temperature, and the power loss injected into each node.

3. The method for calculating the temperature rise of the integrated permanent magnet synchronous electric drive system according to claim 1, wherein, The component thermal resistance includes driver thermal resistance, end cover thermal resistance, housing thermal resistance, stator thermal resistance, winding thermal resistance, end winding thermal resistance, air gap thermal resistance, rotor and permanent magnet thermal resistance, shaft thermal resistance, and convective thermal resistance.

4. The method for calculating the temperature rise of the integrated permanent magnet synchronous electric drive system according to claim 1, wherein The node temperature includes driver temperature, end cover temperature, housing temperature, stator temperature, winding temperature, end winding temperature, rotor and permanent magnet temperature, and shaft temperature.

5. The temperature rise calculation method of the integrated permanent magnet synchronous electric drive system according to claim 1, characterized in that, The power loss injected into each node includes inverter circuit power loss, stator iron loss, winding copper loss, end winding copper loss, rotor iron loss, and permanent magnet eddy current loss.

6. The method for calculating the temperature rise of the integrated permanent magnet synchronous electric drive system according to any one of claims 1 to 5, characterized in that To solve the LPTN model of the electric drive system, a thermal conductance matrix G n×n , a node temperature matrix T n×1 and a power loss matrix P n×1 injected into each node are constructed. Among them, the thermal conductance is the reciprocal of the thermal resistance, and n is the total number of nodes in the thermal network. The node equation is written using the node voltage method: Introduce the correction matrix R to the calculation result of temperature n×n , then the corrected node temperature matrix T' n×1 The calculation formula is as follows:

7. The method for calculating the temperature rise of the integrated permanent magnet synchronous electric drive system according to claim 6, characterized in that, The correction matrix R n×n is obtained by training and fitting a small sample dataset of electromagnetic-thermal coupling finite element simulations collected by deep learning. The sample data used for training are all from motor design schemes within the range of structural parameters.

8. The temperature rise calculation method of the integrated permanent magnet synchronous electric drive system according to claim 1, characterized in that, The long short-term memory network-support vector regression algorithm specifically includes: dividing the collected finite element simulation data set and the corresponding original thermal network data set into a training set, a test set, and a validation set according to a ratio of 8:1:1, respectively setting the optimization methods of the LSTM and SVR models, initializing the hyperparameters, setting the upper and lower limits of the optimized hyperparameters and the number of optimization iterations, where LSTM represents the long short-term memory network and SVR represents support vector machine regression; using the training set to train the model with the given hyperparameters of the model, testing on the validation set, continuously cycling in the hyperparameter selection space, and finding the best hyperparameter model M LSTM and M SVR After that, record the errors E LSTM and E SVR of the validation set on the LSTM model and the SVR model respectively, where M LSTM represents the optimal long short-term memory network, and M SVR represents the optimal support vector machine regression; then set a threshold θ, and judge the errors E LSTM and E SVR If the errors are quite different, select the model with the smaller error as the final model; if the error between the two models is less than the threshold, perform a weight combination on the two models, and the final prediction result is equal to the weighted sum of the prediction values of the two models.

Images