Multi-objective optimization method and system for performance of hybrid excitation synchronous generator

Abstract

The invention discloses a hybrid excitation synchronous generator performance multi-objective optimization method and system, and the method comprises the steps: building a finite element simulation model of a hybrid excitation synchronous generator, and determining a motor design parameter and an optimization target quantity; designing an orthogonal experiment based on sensitivity analysis, and dividing design parameters into a high-sensitivity optimization space and a low-sensitivity optimization space; performing data preprocessing on the initial data, learning the initial data through a supervised active learning model, expanding the initial data, and finally determining a motor optimization agent model; randomly extracting a test set based on Latin hypercube sampling, and detecting the accuracy of the optimization agent model; and selecting a motor optimization design scheme by using NSGA-II and Pareto frontier evaluation. According to the method, the motor optimization model with relatively high accuracy can be obtained under the condition of relatively small finite element simulation data volume, the multi-objective optimization efficiency of the motor is improved, and the method has relatively high migration capability and strong model generalization capability.

Description

Technical Field

[0001] This invention relates to the field of hybrid excitation synchronous generator design technology, specifically to a multi-objective optimization method and system for the performance of hybrid excitation synchronous generators. Background Technology

[0002] Permanent magnet motors are widely used due to their high power density and high efficiency. However, because the air gap magnetic field of permanent magnet motors is difficult to adjust, the output voltage changes when the generator load changes, thus affecting the quality of the power supply voltage. While traditional electrically excited motors have adjustable air gap magnetic fields, their excitation losses account for a large proportion of the total losses, reducing motor efficiency. Hybrid excitation motors can effectively combine the advantages of permanent magnet motors and electrically excited motors, and also achieve brushless operation.

[0003] Hybrid-excitation motors, possessing both permanent magnets and electrically excited windings, exhibit complex air-gap magnetic fields with strong coupling and mutual constraints among their hybrid variables. Furthermore, their electromagnetic characteristics are nonlinear, making multi-objective optimization design extremely complex. Traditional motor optimization methods often employ analytical models and finite element simulation models. Analytical models have low accuracy and limited applicability; while finite element simulation models can accurately simulate the actual motor conditions, they involve significant computational burden. Therefore, achieving multi-objective optimization of the performance of hybrid-excitation synchronous generators is a pressing technical problem that needs to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a multi-objective optimization method and system for the performance of hybrid excitation synchronous generators. It mainly solves the problems of the single objective in the performance optimization of hybrid excitation synchronous generators, the time-consuming calculation of finite element models, and the low accuracy of analytical models in the existing technology, and effectively improves the comprehensive performance of hybrid excitation generators.

[0005] In a first aspect, the present invention provides a multi-objective optimization method for the performance of a hybrid excitation synchronous generator, comprising the following steps:

[0006] S1. Establish a finite element simulation model of the hybrid excitation synchronous generator, and set the optimization objective and design parameters;

[0007] S2. Determine the range of design parameters and obtain initial sample data by sampling the design parameter sample values ​​based on orthogonal experiments;

[0008] S3. Based on orthogonal experiments and sensitivity analysis, analyze the sensitivity of each design parameter to the optimization target quantity, and divide the design parameters into optimization subspaces according to the sensitivity level.

[0009] S4. Based on the supervised active learning model, the initial sample data is expanded to obtain a training set. The training set is used to fit and optimize the objective function to obtain the motor optimization surrogate model.

[0010] S5. Using Latin hypercube sampling, randomly select a test set in each optimization subspace to test the accuracy of the proposed motor optimization surrogate model;

[0011] S6. Based on the improved NSGA-II algorithm, first solve the Pareto front of the high-sensitivity layer, then fix the high-sensitivity parameters, and then solve the Pareto front of the low-sensitivity layer.

[0012] S7. Based on the Pareto front evaluation, select the Pareto front to choose the design scheme that optimizes the overall performance of the motor.

[0013] This invention establishes a finite element simulation model of a hybrid excitation synchronous generator to determine the generator design parameters and optimization target quantities. An orthogonal experiment based on sensitivity analysis is designed, dividing the design parameters into two optimization spaces: high and low sensitivity. Initial data is preprocessed, and a supervised active learning model is used to learn from and expand the initial data, ultimately determining the generator optimization surrogate model. A test set is randomly selected based on Latin hypercube sampling to test the accuracy of the proposed optimization surrogate model. NSGA-II and Pareto frontier evaluation are used to select the generator optimization design scheme and the values ​​of each design parameter.

[0014] Furthermore, in S1, three optimization objectives and eight design parameters are set. The optimization objectives include the motor output efficiency η, the third harmonic magnetic flux density amplitude B3, and the voltage waveform distortion rate THD. The design parameters include the thickness H of the permanent magnet. pm The electrical angle θ occupied by the upper edge of the permanent magnet pm1 The electrical angle θ occupied by the lower edge of the permanent magnet pm2 Stator third harmonic winding turns N s3th Thickness H of the top of the magnetic pole cr The electrical angle θ occupied by the clipping portion cr Number of turns N in the excitation winding f Number of turns N in rotor harmonic winding r3th .

[0015] The multi-objective optimization process for motors fully considers different optimization objectives and constraints. Generally, it is desirable for the motor to have high efficiency and low voltage waveform distortion; furthermore, a higher third harmonic magnetic flux density can generate more induced electromotive force, thereby increasing excitation power.

[0016] The selection of design parameters takes into account rotor structure parameters, magnetic circuit parameters, and winding turns, which can significantly improve the overall performance of the hybrid excitation motor.

[0017] Furthermore, in S2, the design parameter sample values ​​are sampled based on orthogonal experiments, including the following steps:

[0018] S21. Divide the 8 design parameters into 3 levels;

[0019] S22. Construct the L27 orthogonal experiment;

[0020] S23. Build finite element simulation models at various levels to obtain initial sample data for motor optimization.

[0021] Furthermore, in S3, the sensitivity of each design parameter to the optimization target quantity is calculated according to the following formula:

[0022]

[0023] Where r is the sensitivity. and Let x represent the sample mean of a certain design parameter and a certain optimization objective quantity in n orthogonal experiments, respectively. i Let y be the value of a design parameter in the i-th orthogonal experiment, n be the total number of samples in the orthogonal experiment, and y be the value of a design parameter. i Let be the value of a certain optimization objective quantity in the i-th orthogonal experiment.

[0024] The design parameters are stratified according to their sensitivity levels, including the thickness H of the permanent magnet. pm Thickness H of the top of the magnetic pole cr Number of turns N of the third harmonic winding of the stator s3th The three high-sensitivity design parameters are divided into the first optimization subspace, ignoring the number of turns N of the excitation winding. f And the number of turns N of the rotor harmonic winding r3th Two low-sensitivity design parameters determine the electrical angle θ occupied by the upper edge of the permanent magnet. pm1 The electrical angle θ occupied by the lower edge of the permanent magnet pm2 The electrical angle θ occupied by the clipping portion cr The three design parameters are set to the second-level optimization subspace.

[0025] By performing sensitivity analysis and stratifying the design parameters, the order of the optimization model in each optimization subspace was reduced, the accuracy of the model was improved, the model solution rate was accelerated, and the efficiency of motor optimization was increased.

[0026] Furthermore, step S4 includes the following steps:

[0027] S41. Perform data preprocessing on the initial sample data obtained from the orthogonal experiment;

[0028] S42. A supervised active learning model is adopted to learn the initial sample data obtained from the orthogonal experiment. A new sample point is selected using the acquisition function to expand the initial sample data and form a total training set. The objective function is then trained and fitted based on Gaussian process regression.

[0029] S43. Define the optimization objective function and constraints;

[0030] Objective function and constraints for the first-level optimization subspace:

[0031]

[0032] Objective function and constraints for the second-level optimization subspace:

[0033]

[0034] Where {F1, F2, F3} is the optimization objective function.

[0035] By using a supervised active learning model, the initial finite element simulation data is learned and expanded, thereby improving the accuracy of the desired motor optimization surrogate model.

[0036] Furthermore, step S5 includes the following steps:

[0037] S51. Divide the full optimization space, which includes the first and second optimization subspaces and six design parameters, into 10 sampling layers.

[0038] S52. Randomly sample the values ​​of each design parameter in each sampling layer, where the number of turns can only be an integer.

[0039] S53. Randomly arrange and combine the extracted design parameters to obtain 10 sets of data as the test set.

[0040] Using Latin hypercube sampling, 10 test sets are randomly selected from the entire optimization space. This makes the sampled test sets highly global and can better reflect the accuracy of the solved motor optimization surrogate model in the entire optimization space.

[0041] Furthermore, step S6 includes the following steps;

[0042] S61. Data initialization, variable definition;

[0043] S62. Calculate the dominance relationship;

[0044] S63. Determine the dominance relationship and repeat until all solutions are layered.

[0045] S64. First, obtain the Pareto front of the high-sensitivity layer, then fix the high-sensitivity parameters, and then solve the Pareto for the low-sensitivity layer.

[0046] The NSGA-II algorithm is a highly efficient multi-objective optimization algorithm. Its core is to hierarchically divide the solution set through fast non-dominated sorting and identify the Pareto front. It has the characteristics of high solution efficiency and strong hierarchical ability, and is suitable for solving the Pareto front of multi-objective optimization of hybrid excitation synchronous generator performance.

[0047] Secondly, the present invention also provides a multi-objective optimization system for the performance of a hybrid excitation synchronous generator, including a simulation model module, a data acquisition module, a hierarchical partitioning module, an active learning module, a detection module, and an intelligent solution module;

[0048] Simulation Model Module: Establish a finite element simulation model of the hybrid excitation synchronous generator, and set optimization objectives and design parameters;

[0049] Data acquisition module: Determines the range of design parameter values, and obtains initial sample data by sampling the design parameter sample values ​​based on orthogonal experiments;

[0050] Hierarchical module division: Based on orthogonal experiments and sensitivity analysis, analyze the sensitivity of each design parameter to the optimization target quantity, and divide the design parameters into optimization subspaces according to the sensitivity level.

[0051] Active learning module: Based on a supervised active learning model, the initial sample data is expanded to obtain a training set, and the training set is used to fit and optimize the objective function to obtain the motor optimization surrogate model;

[0052] Detection module: Using Latin hypercube sampling, a test set is randomly selected in each optimization subspace to detect the accuracy of the proposed motor optimization surrogate model;

[0053] Intelligent solution module: Based on the improved NSGA-II algorithm, it solves the Pareto front that meets the requirements of motor optimization design. Based on the Pareto front evaluation, it selects the Pareto front and chooses the design scheme that optimizes the overall performance of the motor.

[0054] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0055] This invention is based on sensitivity analysis and performs hierarchical optimization of design parameters, giving priority to design parameters with higher sensitivity and setting them as the first-level optimization subspace. This not only reduces the order of equations in the optimization design, but also improves the accuracy of the optimization model.

[0056] Based on a supervised active learning model, the original data is learned and augmented, thereby increasing the amount of original data and improving the accuracy of the fitting optimization model without increasing the finite element simulation.

[0057] Based on Latin hypercube sampling, test groups were randomly sampled in the optimized full space, and error analysis was performed on the finite element simulation model of the test groups and the active learning fitting optimization model to ensure that the error of the active learning model meets the design requirements.

[0058] This invention combines the advantages of active learning and NSGA-II, introducing Pareto front evaluation to select the optimal motor design. A supervised active learning model is used to learn and expand the finite element simulation data of the optimized motor, increasing the amount of original finite element simulation data and improving the accuracy of the motor surrogate model. A high-order surrogate optimization model for the motor is established using Gaussian process regression. An improved NSGA-II algorithm is used to solve for the Pareto front of the motor optimization. Based on the Pareto front evaluation system, the final multi-objective optimization design scheme of the motor is selected. Using this method, a highly accurate motor optimization model can be obtained with a relatively small amount of finite element simulation data (limited initial data). The improved NSGA-II algorithm allows for faster Pareto front solving, improving the efficiency of multi-objective motor optimization and exhibiting high transferability and strong model generalization ability. Attached Figure Description

[0059] Figure 1 Flowchart of a multi-objective optimization method for the performance of a hybrid excitation synchronous generator;

[0060] Figure 2 A 3D structural diagram of a hybrid excitation synchronous generator;

[0061] Figure 3 A schematic diagram of each design parameter;

[0062] Figure 4 Flowchart for a supervised active learning model;

[0063] Figure 5 Here is the flowchart for the NSGA-II algorithm;

[0064] Figure 6 A schematic diagram of the system structure for multi-objective performance optimization of hybrid excitation synchronous generators. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] Example 1

[0067] This embodiment presents a multi-objective optimization method for the performance of a hybrid excitation synchronous generator, such as... Figure 1 The above includes the following steps:

[0068] S1. Build a finite element simulation model, select design parameters and optimization target quantities.

[0069] In this embodiment, a 4-pole, 36-slot synchronous generator with a short-pitch double-layer lap winding armature is selected. Its finite element 3D model is shown below. Figure 2 As shown.

[0070] The main parameters of the hybrid excitation synchronous generator are shown in Table 1.

[0071] Table 1

[0072]

[0073] This embodiment uses three performance indicators—motor output efficiency η, third harmonic magnetic flux density amplitude B3, and voltage waveform distortion rate THD—as optimization targets, and the thickness H of the permanent magnet is also considered. pm The electrical angle θ occupied by the upper edge of the permanent magnet pm1 The electrical angle θ occupied by the lower edge of the permanent magnet pm2 Stator third harmonic winding turns N s3th Thickness H of the top of the magnetic pole cr The electrical angle θ occupied by the clipping portion cr Number of turns N in the excitation winding f Number of turns N in rotor harmonic winding r3th These eight parameters are used as design variables, and their design parameter diagram is shown below. Figure 3 As shown.

[0074] S2. Design an orthogonal experiment to obtain initial data. Orthogonal experiments save experimental time and cost while ensuring the reliability and reproducibility of the results. Specifically, this includes the following steps:

[0075] S21. Divide the eight design parameters into three levels, and the factors of each level are shown in Table 2.

[0076] Table 2

[0077]

[0078] S22. Based on the constructed 3-level factor, design L27(3 8 The orthogonal experiments were conducted, and the orthogonal experimental groups and experimental results are shown in Table 3.

[0079] Table 3

[0080]

[0081] The initial data were collected by designing orthogonal experiments.

[0082] S3. Based on sensitivity analysis, the design parameters are processed in layers. The design parameters with higher sensitivity are divided into the first layer of optimization subspace, and the remaining design parameters are divided into the second layer of optimization subspace. Sensitivity analysis reduces the number of equations in the optimization model and the order of the optimization objective function, thereby improving the accuracy of the optimization model and the model solution rate.

[0083] Considering the interplay of various design parameters in the design process of a hybrid excitation motor, which collectively affect the optimization objective function, the sensitivity analysis of each design parameter is performed using the following formula:

[0084]

[0085] Where r is the sensitivity. and Let x represent the sample mean of a certain design parameter and a certain optimization objective quantity in n orthogonal experiments, respectively. i Let y be the value of a certain design parameter in the i-th orthogonal experiment, and n be the total number of samples in the orthogonal experiment (n = 27). i Let be the value of a certain optimization objective quantity in the i-th orthogonal experiment;

[0086] Substituting the data, the sensitivity data for each design parameter are shown in Table 4.

[0087] By analyzing Table 4, N r3th With N f The sensitivity to the objective function is very low; therefore, N is ignored for ease of subsequent optimization. r3th With N f Design parameters. From Table 4, parameters with high sensitivity to the three optimization objectives are preferentially selected as the first optimization subspace. The design parameters for the first subspace are: the thickness H of the permanent magnet. pm Thickness H of the top of the magnetic pole cr Number of turns N of the third harmonic winding of the stator s3thThe design parameters with relatively low sensitivity are used as the second-level optimization subspace. The design parameters of the second-level optimization subspace are: the electrical angle θ occupied by the upper edge of the permanent magnet. pm1 The electrical angle θ occupied by the lower edge of the permanent magnet pm2 The electrical angle θ occupied by the clipping portion cr Through sensitivity analysis, the number of design parameters was reduced from eight to six, and the design parameters were stratified, which improved the model accuracy and the solution speed.

[0088] Table 4

[0089]

[0090] S4. Based on a supervised active learning model, the initial data is learned and effectively augmented using a data acquisition function to increase the initial data volume. The augmented data is then used as the training set to fit an optimization objective function, thereby obtaining the motor optimization surrogate model. The supervised active learning flowchart is as follows: Figure 4 As shown. The steps are as follows:

[0091] S41. Preprocess the initial finite element simulation data obtained through the L27 orthogonal experiment.

[0092] S42. An approximate objective function is constructed through iterative learning of the initial data, and new sample points are generated in the three-dimensional candidate network based on the acquisition function.

[0093] S43. Update the initial finite element simulation data by adding newly generated simulation sample points after active learning to form a total training set. Use the training set to fit a new optimization objective function.

[0094] S44. Define the optimization objective function and constraints;

[0095] Objective function and constraints for the first-level optimization subspace:

[0096]

[0097] Objective function and constraints for the second-level optimization subspace:

[0098]

[0099] Where {F1, F2, F3} is the optimization objective function.

[0100] The objective function and its constraints in each optimization subspace are obtained through supervised active learning.

[0101] S5. Based on Latin hypercube sampling, 10 sets of data are randomly selected as the test set to test the accuracy of the proposed motor optimization surrogate model in the full optimization space. Compared with traditional Monte Carlo sampling, Latin hypercube sampling can better cover the entire parameter space with the same number of samples, making it suitable for high-dimensional problems. The specific steps are as follows:

[0102] S51. Divide the full optimization space of the 6 design parameters into 10 sampling layers;

[0103] S52. Randomly sample the values ​​of each design parameter in each sampling layer, where the number of turns can only be an integer.

[0104] S53. Randomly arrange and combine the extracted design parameters to obtain 10 sets of data as the test set.

[0105] S6. Use NSGA-II and Pareto frontiers to solve for the optimal design of the hybrid excitation motor. The NSGA-II flowchart is shown below. Figure 5 As shown, the steps are as follows:

[0106] S61. Initialize the data required by the algorithm and define the various variables;

[0107] S62. Calculate the dominance relationship and identify the first non-dominance relationship;

[0108] S63. Determine the dominance relationship and repeat until all solutions are layered.

[0109] S64. First, obtain the Pareto front of the high-sensitivity layer, then fix the high-sensitivity parameters, and then solve the Pareto for the low-sensitivity layer.

[0110] The final optimized design scheme of the hybrid excitation synchronous generator was obtained by solving the problem. The comparison of the design parameters and motor performance before and after optimization is shown in Table 5.

[0111] Table 5

[0112]

[0113] This embodiment uses finite element simulation for verification. An optimized finite element simulation model of the hybrid excitation motor was built, and the finite element simulation results are shown in Table 5. The specific process is as follows:

[0114] (1) Motor efficiency;

[0115] Before optimization, the efficiency of the hybrid excitation motor was 88.13%, and after optimization, the efficiency of the hybrid excitation motor increased to 91.32%, representing an efficiency improvement of 3.19%.

[0116] (2) Third harmonic magnetic flux density;

[0117] The amplitude of the third harmonic magnetic flux density of the motor did not change much before and after optimization. Before optimization, the amplitude of the third harmonic magnetic flux density of the hybrid excitation motor was 0.4057T, and after optimization, the amplitude of the third harmonic magnetic flux density of the hybrid excitation motor was 0.3959T. The amplitude of the third harmonic magnetic flux density decreased slightly by 0.0098T.

[0118] (3) Voltage waveform distortion rate;

[0119] Before optimization, the voltage waveform distortion rate of the hybrid excitation motor was 6.38% (rated load). (Lag) The voltage waveform distortion rate of the optimized hybrid excitation motor is 5.36% (rated load). (Lag) The voltage waveform distortion rate decreased by 1.02% after optimization, and the no-load voltage distortion rate was 2.23% after optimization. The voltage waveform distortion rate of the motor after optimization was significantly reduced compared with that before optimization.

[0120] Example 2

[0121] Unlike Example 1, this example uses different intelligent algorithms to solve the Pareto front, such as particle swarm optimization, differential evolution, and gray wolf algorithm.

[0122] Example 3

[0123] This embodiment is a multi-objective performance optimization system for a hybrid excitation synchronous generator, the specific structure of which is as follows: Figure 6 As shown, it includes the following:

[0124] Simulation Model Module: Establish a finite element simulation model of the hybrid excitation synchronous generator, and set optimization objectives and design parameters;

[0125] Data acquisition module: Determines the range of design parameter values, and obtains initial sample data by sampling the design parameter sample values ​​based on orthogonal experiments;

[0126] Hierarchical module division: Based on orthogonal experiments and sensitivity analysis, analyze the sensitivity of each design parameter to the optimization target quantity, and divide the design parameters into optimization subspaces according to the sensitivity level.

[0127] Active learning module: Based on a supervised active learning model, the initial sample data is expanded to obtain a training set, and the training set is used to fit and optimize the objective function to obtain the motor optimization surrogate model;

[0128] Detection module: Using Latin hypercube sampling, a test set is randomly selected in each optimization subspace to detect the accuracy of the proposed motor optimization surrogate model;

[0129] Intelligent solution module: Based on the improved NSGA-II algorithm, it solves the Pareto front that meets the requirements of motor optimization design. Based on the Pareto front evaluation, it selects the Pareto front and chooses the design scheme that optimizes the overall performance of the motor.

[0130] The specific working process of each module is shown in Example 1.

[0131] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0132] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A multi-objective optimization method for the performance of a hybrid excitation synchronous generator, characterized in that: Includes the following steps: S1. Establish a finite element simulation model of the hybrid excitation synchronous generator, and set the optimization objective and design parameters; S2. Determine the range of design parameters and obtain initial sample data by sampling the design parameter sample values ​​based on orthogonal experiments; S3. Based on orthogonal experiments and sensitivity analysis, analyze the sensitivity of each design parameter to the optimization target quantity, and divide the design parameters into optimization subspaces according to the sensitivity level. S4. Based on the supervised active learning model, the initial sample data is expanded to obtain a training set. The training set is used to fit and optimize the objective function to obtain the motor optimization surrogate model. S5. Using Latin hypercube sampling, randomly select a test set in each optimization subspace to test the accuracy of the proposed motor optimization surrogate model; S6. Based on the improved NSGA-II algorithm, first solve the Pareto front of the high-sensitivity layer, then fix the high-sensitivity parameters and solve the Pareto front of the low-sensitivity layer. S7. Based on Pareto front evaluation, the Pareto front of the high-sensitivity layer is first selected, then the high-sensitivity parameters are fixed, and the Pareto front of the low-sensitivity layer is selected to finally obtain the design scheme with the best overall motor performance.

2. The multi-objective optimization method for the performance of a hybrid excitation synchronous generator according to claim 1, characterized in that: In S1, three optimization objectives and eight design parameters are set. The optimization objectives include motor output efficiency η, third harmonic magnetic flux density amplitude B3, and voltage waveform distortion rate THD. The design parameters include the thickness H of the permanent magnet. pm The electrical angle θ occupied by the upper edge of the permanent magnet pm1 The electrical angle θ occupied by the lower edge of the permanent magnet pm2 Stator third harmonic winding turns N s3th Thickness H of the top of the magnetic pole cr The electrical angle θ occupied by the clipping portion cr Number of turns N in the excitation winding f Number of turns N in rotor harmonic winding r3th .

3. The multi-objective optimization method for the performance of a hybrid excitation synchronous generator according to claim 2, characterized in that: In S2, the design parameter sample values ​​are sampled based on orthogonal experiments, including the following steps: S21. Divide the 8 design parameters into 3 levels; S22. Construct the L27 orthogonal experiment; S23. Build finite element simulation models at various levels to obtain initial sample data for motor optimization.

4. The multi-objective optimization method for the performance of a hybrid excitation synchronous generator according to claim 3, characterized in that: In S3, the sensitivity of each design parameter to the optimization objective quantity is calculated using the following formula: Where r is the sensitivity. and Let x represent the sample mean of a certain design parameter and a certain optimization objective quantity in n orthogonal experiments, respectively. i Let y be the value of a design parameter in the i-th orthogonal experiment, n be the total number of samples in the orthogonal experiment, and y be the value of a design parameter. i Let be the value of a certain optimization objective quantity in the i-th orthogonal experiment; The design parameters are stratified according to their sensitivity levels, including the thickness H of the permanent magnet. pm Thickness H of the top of the magnetic pole cr Number of turns N of the third harmonic winding of the stator s3th The three high-sensitivity design parameters are divided into the first optimization subspace, ignoring the number of turns N of the excitation winding. f And the number of turns N of the rotor harmonic winding r3th Two low-sensitivity design parameters determine the electrical angle θ occupied by the upper edge of the permanent magnet. pm1 The electrical angle θ occupied by the lower edge of the permanent magnet pm2 The electrical angle θ occupied by the clipping portion cr The three design parameters are set to the second-level optimization subspace.

5. The multi-objective optimization method for the performance of a hybrid excitation synchronous generator according to claim 4, characterized in that: S4 includes the following steps: S41. Perform data preprocessing on the initial sample data obtained from the orthogonal experiment; S42. A supervised active learning model is adopted to learn the initial sample data obtained from the orthogonal experiment. A new sample point is selected using the acquisition function to expand the initial sample data and form a total training set. The objective function is then trained and fitted based on Gaussian process regression. S43. Define the optimization objective function and constraints; Objective function and constraints for the first-level optimization subspace: Objective function and constraints for the second-level optimization subspace: Where {F1, F2, F3} is the optimization objective function.

6. The multi-objective optimization method for the performance of a hybrid excitation synchronous generator according to claim 5, characterized in that: S5 includes the following steps: S51. Divide the full optimization space, which includes the first and second optimization subspaces and six design parameters, into 10 sampling layers. S52. Randomly sample the values ​​of each design parameter in each sampling layer, where the number of turns can only be an integer. S53. Randomly arrange and combine the extracted design parameters to obtain 10 sets of data as the test set.

7. The multi-objective optimization method for the performance of a hybrid excitation synchronous generator according to claim 1, characterized in that: S6 includes the following steps; S61. Data initialization, variable definition; S62. Calculate the dominance relationship; S63. Determine the dominance relationship and repeat until all solutions are layered. S64. First, obtain the Pareto front of the high-sensitivity layer, then fix the high-sensitivity parameters, and then solve the Pareto for the low-sensitivity layer.

8. A multi-objective performance optimization system for a hybrid excitation synchronous generator, used to implement the optimization method according to any one of claims 1-7, characterized in that: It includes a simulation model module, a data acquisition module, a hierarchical partitioning module, an active learning module, a detection module, and an intelligent solution module; Simulation Model Module: Establish a finite element simulation model of the hybrid excitation synchronous generator, and set optimization objectives and design parameters; Data acquisition module: Determines the range of design parameter values, and obtains initial sample data by sampling the design parameter sample values ​​based on orthogonal experiments; Hierarchical module division: Based on orthogonal experiments and sensitivity analysis, analyze the sensitivity of each design parameter to the optimization target quantity, and divide the design parameters into optimization subspaces according to the sensitivity level. Active learning module: Based on a supervised active learning model, the initial sample data is expanded to obtain a training set, and the training set is used to fit and optimize the objective function to obtain the motor optimization surrogate model; Detection module: Using Latin hypercube sampling, a test set is randomly selected in each optimization subspace to detect the accuracy of the proposed motor optimization surrogate model; Intelligent solution module: Based on the improved NSGA-II algorithm, it solves the Pareto front that meets the requirements of motor optimization design. Based on the Pareto front evaluation, it selects the Pareto front and chooses the design scheme that optimizes the overall performance of the motor.

Images