Sectional type PMLSM control method

Abstract

The invention discloses a sectional type PMLSM control method, relates to the technical field of PMLSM control, and solves the technical problem that in the prior art, although the operation performance of a PMLSM can be improved to a certain extent, an effective means for directly controlling and optimizing the structural parameters of the PMLSM itself is lacked, so that global optimum of the performance of the PMLSM is often difficult to achieve. The method comprises the following steps: acquiring structure data of PMLSM; processing the structural data to obtain training data; training a support vector machine based on the training data to obtain a PMLSM thrust prediction model; multi-objective optimization is performed on the PMLSM by combining the PMLSM thrust prediction model and the NSGA-II algorithm, so that the technical problem that global optimization of the performance of the PMLSM is difficult to achieve due to lack of an effective means for directly controlling and optimizing the structural parameters of the PMLSM in the prior art is solved.

Description

Technical Field

[0001] The invention belongs to the field of PMLSM control, in particular to a segmented PMLSM control method. Background Art

[0002] With the continuous development of power electronics technology and control theory, segmented permanent magnet linear synchronous motor (PMLSM) has been widely used in high-speed machining, precision manufacturing, rail transportation and other fields due to its high efficiency and high precision.

[0003] The existing PMLSM optimization methods mainly rely on the control strategy and external environment adjustment during the operation of the motor to achieve its performance improvement. These methods include but are not limited to improving the control algorithm of the motor, optimizing the parameter configuration of the power electronic equipment, and adjusting the load characteristics and operating environment. In fact, the structural parameters of the PMLSM, such as the magnetic pole height, magnetic pole width, position and height of the side auxiliary teeth of the permanent magnet, have a crucial influence on the thrust characteristics, thrust fluctuation and overall performance of the permanent magnet linear synchronous motor. Although the existing technology can improve the operating performance of the PMLSM to a certain extent, it lacks an effective means to directly control and optimize the structural parameters of the PMLSM itself, so it is often difficult to achieve the global optimum of the PMLSM performance.

[0004] Therefore, the present invention solves the above problems by proposing a segmented PMLSM control method. Summary of the invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a segmented PMLSM control method for solving the technical problem that although the prior art can improve the operating performance of the PMLSM to a certain extent, it lacks an effective means to directly control and optimize the structural parameters of the PMLSM itself, and therefore it is often difficult to achieve the global optimal performance of the PMLSM.

[0006] To achieve the above object, a first aspect of the present invention provides a segmented PMLSM control method, comprising:

[0007] Step 1: Obtain the structural data of PMLSM;

[0008] Step 2: Obtain training data by processing the structural data;

[0009] Step 3: Train the support vector machine based on the training data to obtain the PMLSM thrust prediction model;

[0010] Step 4: Perform multi-objective optimization of PMLSM by combining the PMLSM thrust prediction model and NSGA-II algorithm.

[0011] Preferably, the obtaining of structural data of PMLSM includes:

[0012] Obtain the structural data preset before PMLSM production; wherein the structural data include: the range of the permanent magnet's pole height h and the range of the pole width ω, the range of the position d of the edge auxiliary teeth, the range of the auxiliary tooth height g, the range of the semi-closed slot width b1 and the range of the slot width b2.

[0013] It should be noted that the preset structural data is set according to the actual requirements of the PMLSM produced.

[0014] Preferably, the processing of the structural data includes:

[0015] Generate sample set based on maximum minimum Latin hypercube sampling method;

[0016] The thrust data corresponding to each sample in the sample set is obtained through finite element simulation software.

[0017] Preferably, the generating of the sample set based on the maximum-minimum Latin hypercube sampling method comprises:

[0018] By calling the lhsdesign function in MATLAB, sampling is performed from the structural data to obtain a number of samples; each sample corresponds to a structural variable data; the structural variable data refers to the specific value of each data in the structural data.

[0019] It should be noted that the structural variable data refers to the specific value of each data in the structural data. For example, in a structural variable data, the pole height h of the permanent magnet is 1.5 mm, the pole width ω is 7.7 mm, the position d of the edge auxiliary teeth is 2.1 mm, the height g of the auxiliary teeth is 1.3 mm, the slot width b1 of the semi-closed slot is 5.7 mm, and the slot width b2 is 8.1 mm.

[0020] The present invention can ensure that samples are evenly distributed in the entire range of structural variables through the maximum and minimum Latin hypercube sampling method, thereby improving the representativeness of samples and avoiding sample aggregation or sparseness that may occur in traditional random sampling methods. This evenly distributed sample set can not only better cover the possible value range of structural variables, but also improve the accuracy and generalization ability of subsequent training support vector machine models, thereby improving the reliability and effectiveness of optimization results. In addition, the evenly distributed sample set helps to reduce the amount of calculation of finite element simulation and improve the efficiency of the overall optimization process.

[0021] Preferably, obtaining thrust data corresponding to each sample in the sample set by finite element simulation software includes:

[0022] B1: Create a 2D or 3D model of the segmented PMSLM in the finite element simulation software and mark it as a finite element model;

[0023] B2: Set the structural variable data corresponding to each sample as the corresponding parameters of the finite element model to obtain the model corresponding to each sample;

[0024] B3: The model corresponding to each sample is simulated and calculated by using finite element simulation software to obtain thrust data corresponding to each sample; wherein the thrust data includes: average thrust and thrust fluctuation; the finite element simulation software includes: ANSYS, COMSOL or FEMM.

[0025] It should be noted that the structural variable data corresponding to each sample is set as the corresponding parameters of the finite element model. For example, the pole height h of the permanent magnet is 1.5 mm, the pole width ω is 7.7 mm, etc. are set as the pole height and pole width of the permanent magnet in the finite element model.

[0026] The present invention can accurately simulate the physical behavior of PMLSM through finite element simulation software (such as ANSYS, COMSOL or FEMM), ensuring the high accuracy and reliability of the simulation results; by establishing a 2D or 3D model in the finite element simulation software, and setting the structural variable data of each sample as the corresponding parameters of the model, a model corresponding to each sample can be systematically generated, thereby comprehensively covering the possible combinations of structural parameters; by performing simulation calculations on each sample model, its corresponding thrust data (including average thrust and thrust fluctuation) is obtained, providing high-quality training data for subsequent support vector machine training, these high-quality data can not only improve the prediction accuracy of the model, but also ensure that the multi-objective performance evaluation in the optimization process is more scientific and reasonable, and ultimately improve the overall performance optimization effect of PMLSM.

[0027] Preferably, the training of a support vector machine based on training data comprises:

[0028] The sample set in the training data is integrated into standard input data, and the thrust data corresponding to each sample in the sample set is integrated into standard output data;

[0029] The support vector machine is trained based on standard input data and standard output data to obtain the PMLSM thrust prediction model.

[0030] Preferably, the multi-objective optimization of PMLSM by combining the PMLSM thrust prediction model and the NSGA-II algorithm includes:

[0031] C1: Use the sample set of training data as the parent population Pt and set the algorithm parameters, including the population size N and the maximum number of iterations T.

[0032] C2: Perform non-dominated sorting and crowding calculation on the parent population Pt, and select the best individuals as the pairing pool based on the non-dominated sorting and crowding calculation results;

[0033] C3: Genetic operations are performed on the pairing pool to generate the offspring population Q; wherein the genetic operations include: selection, crossover and mutation;

[0034] C4: Merge the parent population Pt and the offspring population Q into a new population R with a population size of 2N;

[0035] C5: Perform non-dominated sorting and crowding calculation on the new population R, select N excellent individuals from it to form a new parent population, and determine whether the number of iterations reaches the maximum number of iterations T; if yes, output the optimal solution, and perform multi-objective optimization on PMLSM based on the optimal solution; if no, return to C2 iteration; the optimal solution refers to the excellent individuals finally selected.

[0036] It should be noted that the population size N refers to the number of individuals in the population, that is, the number of samples in the sample set; the individuals in the population contain a structural variable data, that is, the specific value of each data in the structural data.

[0037] The present invention uses a sample set of training data as a parent population and sets reasonable algorithm parameters to ensure the controllability and efficiency of the optimization process; selects the best individual as a pairing pool through non-dominated sorting and crowding calculation, effectively maintains the diversity and convergence of the population, and avoids premature convergence to a local optimal solution; generates a new offspring population by performing genetic operations such as selection, crossover and mutation on the pairing pool, thereby increasing the exploration capability of the search space; merges the parent population and the offspring population into a larger new population, and performs non-dominated sorting and crowding calculation again, and selects excellent individuals from the new parent population, thereby ensuring that the quality of each generation of population is gradually improved; and outputs the optimal solution through multiple iterations until the maximum number of iterations is reached. The optimal solution achieves global optimality in multiple objectives of average thrust and thrust fluctuation, thereby significantly improving the overall performance of the PMLSM.

[0038] Preferably, performing non-dominated sorting and crowding calculation on the parent population Pt, and selecting the best individual as the pairing pool according to the non-dominated sorting and crowding calculation results, comprises:

[0039] D1: The individuals in the parent population Pt are divided into multiple non-dominated layers by calling the non_dominated_sort function;

[0040] D2: Based on the thrust data corresponding to each sample in the sample set, the crowding degree of individuals in each non-dominated layer is calculated by calling the crowding_distance function;

[0041] D3: Sort the non-dominated layers in ascending order; the individuals with smaller numbers are more excellent;

[0042] D4: In the same non-dominated layer, individuals are sorted from large to small according to the degree of crowding; individuals with greater crowding are more excellent;

[0043] D5: Select the non-dominated layer with the smallest non-dominated layer number, and the individual with the largest crowding degree in the smallest non-dominated layer is the optimal individual.

[0044] Preferably, the non-dominated sorting and crowding calculation of the new population R to select N excellent individuals from the new population R to form a new parent population includes:

[0045] E1: Divide the individuals in the new population R into multiple non-dominated layers by calling the non_dominated_sort function;

[0046] E2: Input the individuals in the new population R into the PMLSM thrust prediction model to obtain the thrust data corresponding to the individuals in the new population R;

[0047] E3: Calculate the crowding degree of individuals in each non-dominated layer by calling the crowding_distance function based on the thrust data corresponding to the individuals in the new population R;

[0048] E4: Sort by the number of the non-dominated layer from small to large; within the same non-dominated layer, sort by the congestion degree from large to small;

[0049] E5: Extract the first N non-dominated layers by non-dominated layer number, and select the individuals with the largest crowding degree in each of the first N non-dominated layers to form a new parent population.

[0050] Preferably, the multi-objective optimization of PMLSM based on the optimal solution includes:

[0051] Extract structural variable data from the optimal solution; update the design drawings or manufacturing process of the segmented PMSLM based on the structural variable data.

[0052] It should be noted that the optimal solution refers to the optimal individual screened out after completing the iteration.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] Although the prior art can improve the operating performance of PMLSM to a certain extent, it lacks an effective means to directly control and optimize the structural parameters of the PMLSM itself, and therefore it is often difficult to achieve the technical problem of global optimization of the PMLSM performance. The present invention obtains the structural data of the PMLSM, generates a sample set by a maximum-minimum Latin hypercube sampling method, and uses finite element simulation software to obtain the thrust data corresponding to each sample to generate a training data set. A thrust prediction model of the PMLSM is constructed by training a support vector machine using the training data set. Then, the structural parameters of the PMLSM are globally optimized by combining the thrust prediction model with the NSGA-II multi-objective optimization algorithm. Excellent individuals are screened out as new parent populations by non-dominated sorting and crowding calculation, and iterative optimization is performed until a maximum number of iterations is reached to finally obtain an optimal solution. The design drawings or manufacturing processes of the PMLSM are updated based on the optimal solution, thereby achieving global optimization of the PMLSM performance. The present invention not only overcomes the limitations of the prior art, but also significantly improves the design efficiency and performance level of the PMLSM. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0056] Figure 1 A schematic diagram of the method steps of an embodiment of the present invention;

[0057] Figure 2 A schematic diagram of the positions corresponding to the PMSLM structure data of an embodiment of the present invention;

[0058] Figure 3 Schematic diagram of the steps of the PMLSM multi-objective optimization method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] See also Figure 1 The first aspect of the present invention provides a segmented PMLSM control method, comprising:

[0061] Step 1: Obtain the structural data of PMLSM;

[0062] Step 2: Obtain training data by processing the structural data;

[0063] Step 3: Train the support vector machine based on the training data to obtain the PMLSM thrust prediction model;

[0064] Step 4: Perform multi-objective optimization of PMLSM by combining the PMLSM thrust prediction model and NSGA-II algorithm.

[0065] See also Figure 2 , obtain the structural data of PMLSM, including:

[0066] Obtain the structural data preset before PMLSM production; wherein the structural data include: the range of the permanent magnet's pole height h and the range of the pole width ω, the range of the position d of the edge auxiliary teeth, the range of the auxiliary tooth height g, the range of the semi-closed slot width b1 and the range of the slot width b2.

[0067] The position of each data in the structure data is as follows Figure 2 shown.

[0068] By processing the structured data, including:

[0069] Generate sample set based on maximum minimum Latin hypercube sampling method;

[0070] The thrust data corresponding to each sample in the sample set is obtained through finite element simulation software.

[0071] Generate a sample set based on the maximum minimum Latin hypercube sampling method, including:

[0072] By calling the lhsdesign function in MATLAB, sampling is performed from the structural data to obtain a number of samples; each sample corresponds to a structural variable data; the structural variable data refers to the specific value of each data in the structural data.

[0073] The thrust data corresponding to each sample in the sample set is obtained through finite element simulation software, including:

[0074] B1: Establish a 2D or 3D model of the segmented PMSLM in the finite element simulation software and name it as the finite element model;

[0075] B2: Set the structural variable data corresponding to each sample as the corresponding parameters of the finite element model to obtain the model corresponding to each sample;

[0076] B3: The model corresponding to each sample is simulated and calculated by using finite element simulation software to obtain thrust data corresponding to each sample; wherein the thrust data includes: average thrust and thrust fluctuation; the finite element simulation software includes: ANSYS, COMSOL or FEMM.

[0077] Train the support vector machine based on the training data, including:

[0078] The sample set in the training data is integrated into standard input data, and the thrust data corresponding to each sample in the sample set is integrated into standard output data;

[0079] The support vector machine is trained based on standard input data and standard output data to obtain the PMLSM thrust prediction model.

[0080] See also Figure 3 , by combining the PMLSM thrust prediction model and the NSGA-II algorithm, multi-objective optimization of PMLSM is performed, including:

[0081] C1: Use the sample set of training data as the parent population Pt and set the algorithm parameters, including the population size N and the maximum number of iterations T.

[0082] C2: Perform non-dominated sorting and crowding calculation on the parent population Pt, and select the best individuals as the pairing pool based on the non-dominated sorting and crowding calculation results;

[0083] C3: Genetic operations are performed on the pairing pool to generate the offspring population Q; wherein the genetic operations include: selection, crossover and mutation;

[0084] C4: Merge the parent population Pt and the offspring population Q into a new population R with a population size of 2N;

[0085] C5: Perform non-dominated sorting and crowding calculation on the new population R, select N excellent individuals from it to form a new parent population, and determine whether the number of iterations reaches the maximum number of iterations T; if yes, output the optimal solution, and perform multi-objective optimization on PMLSM based on the optimal solution; if no, return to C2 iteration; the optimal solution refers to the excellent individuals finally selected.

[0086] Perform non-dominated sorting and crowding calculation on the parent population Pt, and select the best individuals as the pairing pool according to the non-dominated sorting and crowding calculation results, including:

[0087] D1: The individuals in the parent population Pt are divided into multiple non-dominated layers by calling the non_dominated_sort function;

[0088] D2: Based on the thrust data corresponding to each sample in the sample set, the crowding degree of individuals in each non-dominated layer is calculated by calling the crowding_distance function;

[0089] D3: Sort the non-dominated layers in ascending order; the individuals with smaller numbers are more excellent;

[0090] D4: In the same non-dominated layer, individuals are sorted from large to small according to the degree of crowding; individuals with greater crowding are more excellent;

[0091] D5: Select the non-dominated layer with the smallest non-dominated layer number, and the individual with the largest crowding degree in the smallest non-dominated layer is the optimal individual.

[0092] Perform non-dominated sorting and crowding calculation on the new population R, and select N excellent individuals from it to form a new parent population, including:

[0093] E1: Divide the individuals in the new population R into multiple non-dominated layers by calling the non_dominated_sort function;

[0094] E2: Input the individuals in the new population R into the PMLSM thrust prediction model to obtain the thrust data corresponding to the individuals in the new population R;

[0095] E3: Calculate the crowding degree of individuals in each non-dominated layer by calling the crowding_distance function based on the thrust data corresponding to the individuals in the new population R;

[0096] E4: Sort by the number of the non-dominated layer from small to large; within the same non-dominated layer, sort by the congestion degree from large to small;

[0097] E5: Extract the first N non-dominated layers by non-dominated layer number, and select the individuals with the largest crowding degree in each of the first N non-dominated layers to form a new parent population.

[0098] Based on the optimal solution, multi-objective optimization of PMLSM is performed, including:

[0099] Extract structural variable data from the optimal solution; update the design drawings or manufacturing process of the segmented PMSLM based on the structural variable data.

[0100] For example, suppose you need to optimize the design of a PMLSM to increase its average thrust and reduce thrust fluctuation. The specific steps are as follows:

[0101] Step 1: Obtain the structural data of PMLSM;

[0102] 1. Get the preset structure data:

[0103] The magnetic pole height h of the permanent magnet: 1.0mm to 2.0mm;

[0104] The magnetic pole width ω of the permanent magnet: 6.0 mm to 9.0 mm;

[0105] Position d of the auxiliary teeth at the edge: 1.0mm to 3.0mm;

[0106] Height g of auxiliary teeth: 1.0mm to 2.0mm;

[0107] The width b1 of the semi-closed groove is 4.0 mm to 7.0 mm;

[0108] The width b2 of the semi-closed groove is 7.0 mm to 10.0 mm;

[0109] It should be noted that these structural data are preset according to actual production requirements.

[0110] Step 2: Obtain training data by processing the structural data;

[0111] 1. Generate sample set:

[0112] Use the lhsdesign function in MATLAB to perform maximum-minimum Latin hypercube sampling from the structure data and generate 100 samples.

[0113] Each sample contains specific values ​​of the six structural variables mentioned above. For example, a sample may be: h = 1.5mm, ω = 7.7mm, d = 2.1mm, g = 1.3mm, b1 = 5.7mm, b2 = 8.1mm;

[0114] 2. Get thrust data:

[0115] A 3D model of the segmented PMSLM was established in finite element simulation software (such as ANSYS).

[0116] The structural variable data of each sample is set as the corresponding parameter of the model.

[0117] The model corresponding to each sample is simulated and calculated to obtain the thrust data of each sample, including the average thrust and thrust fluctuation.

[0118] For example, a sample thrust data might be:

[0119] Average thrust: 148.7N;

[0120] Thrust fluctuation: 27.7N;

[0121] Step 3: Train the support vector machine based on the training data to obtain the PMLSM thrust prediction model;

[0122] 1. Integrate training data:

[0123] The structural variable data of 100 samples were integrated as standard input data.

[0124] The thrust data corresponding to 100 samples are integrated into standard output data.

[0125] 2. Training Support Vector Machine:

[0126] The support vector machine (SVM) model is trained using standard input data and standard output data to obtain the PMLSM thrust prediction model.

[0127] Step 4: Multi-objective optimization of PMLSM is performed by combining the PMLSM thrust prediction model and the NSGA-II algorithm;

[0128] S1: Initialize the population:

[0129] The 100 samples of the training data are used as the parent population Pt.

[0130] Set algorithm parameters: population size (N=100), maximum number of iterations (T=500).

[0131] S2: Non-dominated sorting and congestion calculation:

[0132] Perform non-dominated sorting on the parent population Pt and divide the population into multiple non-dominated layers.

[0133] Calculate the crowding degree of individuals in each non-dominated layer based on the thrust data corresponding to each sample;

[0134] The non-dominated layers are sorted from small to large according to their serial numbers, and the same non-dominated layer is sorted from large to small according to the congestion degree.

[0135] The individuals with the smallest sequence number and the largest crowding degree in the non-dominated layer are selected as the pairing pool.

[0136] S3: Genetic Operations:

[0137] Perform genetic operations (selection, crossover, mutation) on the pairing pool to generate the offspring population Q.

[0138] S4: Merged population:

[0139] Merge the parent population Pt and the offspring population Q into a new population R with a population size of 200.

[0140] S5: Select outstanding individuals:

[0141] The individuals in the new population R are input into the PMLSM thrust prediction model to obtain the thrust data corresponding to the individuals in the new population R;

[0142] Based on the thrust data corresponding to the individuals in the new population R, the new population R is sorted non-dominatedly and the crowding degree is calculated.

[0143] Select the top 100 excellent individuals to form a new parent population.

[0144] Determine whether the number of iterations reaches the maximum number of iterations T. If so, output the optimal solution; otherwise, return to S2 to continue iterating.

[0145] Application of the optimal solution:

[0146] 1. Extract the optimal solution:

[0147] After reaching the maximum number of iterations T, the optimal solution is output. The optimal solution is a group of individuals that perform well in average thrust and thrust fluctuation, that is, the optimal individuals.

[0148] For example, the individuals in the optimal solution, i.e., the structural variable data, are: h = 1.6 mm, ω = 7.8 mm, d = 2.2 mm, g = 1.4 mm, b1 = 5.8 mm, b2 = 8.2 mm;

[0149] The thrust data corresponding to the individuals in the optimal solution are:

[0150] Average thrust: 150.5N;

[0151] Thrust fluctuation: 22.0N;

[0152] 2. Updated design:

[0153] Based on the structural variable data in the optimal solution, the design drawings or manufacturing process of the segmented PMSLM are updated.

[0154] Prototypes are made and experimental tests are carried out to verify the effectiveness and reliability of the optimization results.

[0155] The present invention can effectively optimize the structural parameters of the PMLSM, improve its average thrust and reduce thrust fluctuation, thereby achieving global optimal performance improvement.

[0156] Working principle of the present invention:

[0157] The present invention obtains structural data of a PMLSM; obtains training data by processing the structural data; trains a support vector machine based on the training data to obtain a PMLSM thrust prediction model; and performs multi-objective optimization on the PMLSM by combining the PMLSM thrust prediction model and an NSGA-II algorithm, thereby solving the technical problem that, although the prior art can improve the operating performance of the PMLSM to a certain extent, it lacks an effective means to directly control and optimize the structural parameters of the PMLSM itself, and therefore it is often difficult to achieve the global optimum of the PMLSM performance.

[0158] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A segmented PMLSM control method, characterized in that: include: Step 1: Obtain the structural data of PMLSM; Step 2: Obtain training data by processing the structural data; Step 3: Train the support vector machine based on the training data to obtain the PMLSM thrust prediction model; Step 4: Perform multi-objective optimization of PMLSM by combining the PMLSM thrust prediction model and NSGA-II algorithm.

2. A segmented PMLSM control method according to claim 1, characterized in that: The step of obtaining the structural data of the PMLSM comprises: Obtain the structural data preset before PMLSM production; wherein the structural data include: the range of the permanent magnet's pole height h and the range of the pole width ω, the range of the position d of the edge auxiliary teeth, the range of the auxiliary tooth height g, the range of the semi-closed slot width b1 and the range of the slot width b2.

3. A segmented PMLSM control method according to claim 1, characterized in that: The processing of the structure data includes: Generate sample set based on maximum minimum Latin hypercube sampling method; The thrust data corresponding to each sample in the sample set is obtained through finite element simulation software.

4. A segmented PMLSM control method according to claim 3, characterized in that: The method of generating a sample set based on the maximum minimum Latin hypercube sampling method includes: By calling the lhsdesign function in MATLAB, sampling is performed from the structural data to obtain a number of samples; each sample corresponds to a structural variable data; the structural variable data refers to the specific value of each data in the structural data.

5. A segmented PMLSM control method according to claim 3, characterized in that: The method of obtaining thrust data corresponding to each sample in the sample set by using finite element simulation software includes: B1: Create a 2D or 3D model of the segmented PMSLM in the finite element simulation software and mark it as a finite element model; B2: Set the structural variable data corresponding to each sample as the corresponding parameters of the finite element model to obtain the model corresponding to each sample; B3: The model corresponding to each sample is simulated and calculated by using finite element simulation software to obtain thrust data corresponding to each sample; wherein the thrust data includes: average thrust and thrust fluctuation; the finite element simulation software includes: ANSYS, COMSOL or FEMM.

6. A segmented PMLSM control method according to claim 1, characterized in that: The step of training a support vector machine based on training data includes: The sample set in the training data is integrated into standard input data, and the thrust data corresponding to each sample in the sample set is integrated into standard output data; The support vector machine is trained based on standard input data and standard output data to obtain the PMLSM thrust prediction model.

7. A segmented PMLSM control method according to claim 1, characterized in that: The multi-objective optimization of PMLSM by combining the PMLSM thrust prediction model and the NSGA-II algorithm includes: C1: Use the sample set of training data as the parent population Pt and set the algorithm parameters, including the population size N and the maximum number of iterations T. C2: Perform non-dominated sorting and crowding calculation on the parent population Pt, and select the best individuals as the pairing pool based on the non-dominated sorting and crowding calculation results; C3: Genetic operations are performed on the pairing pool to generate the offspring population Q; wherein the genetic operations include: selection, crossover and mutation; C4: Merge the parent population Pt and the offspring population Q into a new population R with a population size of 2N; C5: Perform non-dominated sorting and crowding calculation on the new population R, select N excellent individuals from it to form a new parent population, and determine whether the number of iterations reaches the maximum number of iterations T; if yes, output the optimal solution, and perform multi-objective optimization on PMLSM based on the optimal solution; if no, return to C2 iteration; the optimal solution refers to the excellent individuals finally selected.

8. A segmented PMLSM control method according to claim 7, characterized in that: The non-dominated sorting and crowding calculation are performed on the parent population Pt, and the optimal individuals are selected as the pairing pool according to the non-dominated sorting and crowding calculation results, including: D1: The individuals in the parent population Pt are divided into multiple non-dominated layers by calling the non_dominated_sort function; D2: Based on the thrust data corresponding to each sample in the sample set, the crowding degree of individuals in each non-dominated layer is calculated by calling the crowding_distance function; D3: Sort the non-dominated layers in ascending order; the individuals with smaller numbers are more excellent; D4: In the same non-dominated layer, individuals are sorted from large to small according to the degree of crowding; individuals with greater crowding are more excellent; D5: Select the non-dominated layer with the smallest non-dominated layer number, and the individual with the largest crowding degree in the smallest non-dominated layer is the optimal individual.

9. A segmented PMLSM control method according to claim 7, characterized in that: The non-dominated sorting and crowding calculation of the new population R are performed to select N excellent individuals from the new population R to form a new parent population, including: E1: Divide the individuals in the new population R into multiple non-dominated layers by calling the non_dominated_sort function; E2: Input the individuals in the new population R into the PMLSM thrust prediction model to obtain the thrust data corresponding to the individuals in the new population R; E3: Calculate the crowding degree of individuals in each non-dominated layer by calling the crowding_distance function based on the thrust data corresponding to the individuals in the new population R; E4: Sort by the number of the non-dominated layer from small to large; within the same non-dominated layer, sort by the congestion degree from large to small; E5: Extract the first N non-dominated layers by non-dominated layer number, and select the individuals with the largest crowding degree in each of the first N non-dominated layers to form a new parent population.

10. A segmented PMLSM control method according to claim 7, characterized in that: The multi-objective optimization of PMLSM based on the optimal solution includes: Extract structural variable data from the optimal solution; update the design drawings or manufacturing process of the segmented PMSLM based on the structural variable data.

Images