MMT-PMSLM dynamic performance modeling optimization method and device based on PI-GAN and medium

Abstract

The invention provides a PI-GAN-based MMT-PMSLM dynamic performance modeling optimization method, equipment and a medium, and the method comprises the steps: building an MMT-PMSLM dynamic response performance analysis model based on thrust performance, mechanical and electrical time constants, motor constants and thrust bandwidth based on the requirement of a linear servo system for high dynamic performance; based on the performance analysis model, constructing an MMT-PMSLM dynamic performance prediction model based on PI-GAN; and performing iterative optimization on the dynamic performance prediction model through a dream optimization DOA algorithm to generate optimized structure parameters. The dynamic performance modeling optimization method disclosed by the invention has the characteristics of high precision, strong generalization ability and interpretability, so that the result of optimizing the motor structure by the algorithm has high reliability.

Description

Technical Field

[0001] The present invention belongs to the technical field of permanent magnet synchronous linear motors, and specifically designs a PI-GAN-based MMT-PMSLM dynamic performance modeling optimization method, system and medium. Background Art

[0002] Permanent magnet synchronous linear motors (PMSLMs) are widely used in transportation and high-precision automated manufacturing equipment, such as rail transit, industrial transportation, laser cutting machines, and 3D printing equipment, due to their fast dynamic response and precise positioning. To achieve high dynamic response and precise positioning, the PMSLM system must have lower mechanical and electrical time constants, higher thrust bandwidth characteristics, and a higher motor constant while maintaining optimal thrust performance. Thrust performance primarily consists of thrust density and thrust fluctuation rate. High thrust density, high motor constant, and low time constant mean that the servo motor system has high dynamic response capability. A wider thrust frequency bandwidth ensures that the servo motor system maintains excellent dynamic response performance over a wide speed range. Low thrust fluctuation rate ensures the machining accuracy of the dynamic positioning system.

[0003] During the design and optimization of the motor's structural design, the impact of the motor's design and parameters on the servo motor system's dynamic performance can be fully considered, thereby fundamentally improving the servo motor system's dynamic performance. Typically, innovations in permanent magnet synchronous motor design and optimization focus on changes to the primary or secondary topology and optimization of the motor's mathematical performance models. This aims to increase the motor's air gap magnetic field strength and back EMF, reduce spatial harmonics, and thus improve the motor's dynamic performance in the servo system. However, most of the newer motor topologies available are relatively complex, and compared to traditional motor structures, further consideration is still needed regarding processing and assembly technology or cost constraints. Summary of the Invention

[0004] To address the aforementioned technical issues, the present invention proposes a PI-GAN-based method, system, and medium for modeling and optimizing the dynamic performance of a moving-magnet permanent magnet synchronous linear motor (MMT-PMSLM). Specifically, this method involves establishing and optimizing the dynamic performance model of a moving-magnet permanent magnet synchronous linear motor (PMSLM) based on a physical information embedded generative adversarial network. Guided by an analytical model, this method simultaneously achieves high precision, high interpretability, and strong generalization capabilities, effectively ensuring the reliability of the MMT-PMSLM's comprehensive dynamic performance optimization results. Based on this model, a dream optimization algorithm is introduced to optimize the MMT-PMSLM's structural parameters, thereby improving its comprehensive dynamic performance.

[0005] In the first aspect, the present invention proposes a method for dynamic performance modeling and optimization of a moving magnet permanent magnet synchronous linear motor (MMT-PMSLM) based on PI-GAN. The method specifically optimizes the dynamic performance modeling of a moving magnet permanent magnet synchronous linear motor (MMT-PMSLM) based on a physical information embedding generative adversarial network (PI-GAN), and includes the following steps: Based on the linear servo system's demand for high dynamic performance, a performance analytical model of the MMT-PMSLM dynamic response is established based on thrust performance, mechanical and electrical time constants, motor constants, and thrust bandwidth. Constructing a PI-GAN-based MMT-PMSLM dynamic performance prediction model based on the performance analysis model; Iteratively optimizing the dynamic performance prediction model through a dream optimization DOA algorithm to generate optimized structural parameters; Among them, the dynamic performance prediction model includes a generator and a discriminator, taking the structural parameters of the performance analysis model as the input of the generator, and outputting the predicted values ​​based on thrust performance, mechanical and electrical time constants, motor constants and thrust bandwidth based on the generator; the discriminator is established based on the fitting performance discrimination model of the physical model embedded in the machine learning model.

[0006] Furthermore, based on the requirement of the linear servo system for high dynamic performance, the step of establishing a performance analytical model of the MMT-PMSLM dynamic response based on thrust performance, mechanical and electrical time constants, motor constants, and thrust bandwidth includes obtaining an inductance matrix of a three-phase winding by analyzing a winding inductance analytical model of the MMT-PMSLM, and performing a dynamic performance analysis on the MMT-PMSLM after decoupling current and voltage equations based on a two-dimensional analytical model of the MMT-PMSLM to generate a performance analytical model of the MMT-PMSLM dynamic response based on thrust performance, mechanical and electrical time constants, motor constants, and thrust bandwidth; Among them, thrust performance, mechanical and electrical time constants, motor constants and thrust bandwidth are specifically thrust density , thrust fluctuation , mechanical time constant , electrical time constant , motor constant and thrust bandwidth .

[0007] Furthermore, the steps of obtaining the inductance matrix of the three-phase winding by analyzing the winding inductance analysis model of MMT-PMSLM include: , winding mutual inductance and tank leakage inductance Get the total inductance of the motor , total motor inductance Expressed as: ; Tank leakage inductance Expressed as: ; Based on the winding self-inductance , winding mutual inductance and tank leakage inductance Get the inductance matrix of the three-phase winding : ; in, is the magnetic permeability of air, is the number of turns of each phase winding coil, is the specific leakage permeability coefficient, is the core length.

[0008] Furthermore, the steps of decoupling the current and voltage equations based on the two-dimensional analysis model of MMT-PMSLM specifically include: according to the two-dimensional analysis model of MMT-PMSLM, obtaining the air gap flux distribution in the y-axis direction in the air gap by the equivalent magnetization method and the equivalent surface current method, thereby calculating the three-phase winding back electromotive force; obtaining the response equation of the q-axis current component by the voltage-current equation in the two-phase coordinate system and performing Clark-park transformation and according to the conditional constraints of the initial current of the coil and the maximum load current, thereby realizing the decoupling of the current and voltage equations of MMT-PMSLM.

[0009] Furthermore, the steps of performing dynamic performance analysis on the MMT-PMSLM to generate a performance analytical model of the MMT-PMSLM dynamic response based on thrust performance, mechanical and electrical time constants, motor constants, and thrust bandwidth specifically include: obtaining the average thrust of the MMT-PMSLM in steady state, and obtaining the thrust density based on the ratio of the average thrust to the motor volume. ; Based on the average thrust, the thrust fluctuation of MMT-PMSLM is obtained ; Mechanical time constant , electrical time constant Reflects the speed response capability and thrust response capability of the motor, motor constant Reflects the comprehensive design level of the motor's electromagnetic and heat dissipation, thrust bandwidth It reflects the motor's ability to maintain stable output thrust within a certain speed range.

[0010] Furthermore, the step of performing dynamic performance analysis on the MMT-PMSLM to generate a performance analytical model of the MMT-PMSLM dynamic response based on thrust performance, mechanical and electrical time constants, motor constants and thrust bandwidth also includes a step of analyzing the MMT-PMSLM dynamic response based on thrust density. , thrust fluctuation , mechanical time constant , electrical time constant , motor constant and thrust bandwidth Generate comprehensive dynamic performance evaluation function of MMT-PMSLM , and is expressed as follows: ; in, The smaller the value, the better the comprehensive dynamic performance of MMT-PMSLM.

[0011] Furthermore, the loss function of the generator is expressed as: ; Among them, BCE is the cross entropy loss function, b and c are weight coefficients, is the actual value, is the predicted value, and the value of R is equal to the resistance value of each phase in MMT-PMSLM.

[0012] Furthermore, the dynamic performance prediction model is iteratively optimized by the dream optimization DOA algorithm, and the step of generating optimized structural parameters specifically includes the following steps: The multi-objective optimization problem of the MMT-PMSLM dynamic performance is transformed into a single-objective optimization problem. Based on the DOA algorithm and the PI-GAN prediction model of the motor dynamic performance, the MMT-PMSLM is iteratively optimized to generate the optimized structural parameters.

[0013] In a second aspect, the present invention also proposes an electronic device comprising: a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, it can implement the MMT-PMSLM dynamic performance modeling optimization method based on PI-GAN as described in the first aspect.

[0014] In a third aspect, the present invention further proposes a computer storage medium, wherein the computer program stored in the storage medium, when executed by a processor, can be based on the MMT-PMSLM dynamic performance modeling optimization method based on PI-GAN described in the first aspect.

[0015] This paper proposes a PI-GAN-based MMT-PMSLM dynamic performance modeling optimization method, device, and medium. The PI-GAN-based MMT-PMSLM dynamic performance prediction model established in this paper features high accuracy, strong generalization, and interpretability. These characteristics make the results of the algorithm's motor structure optimization highly reliable.

[0016] Secondly, based on the optimization results of the DOA-PI-GAN modeling optimization method, the MMT-PMSLM prototype was built and experimentally measured. The average thrust of the optimized motor increased by 64.3%, the thrust density increased by 23.5%, the thrust fluctuation was reduced to 2.59%, the motor constant increased by 61.7%, and both the thrust constant and the back EMF constant were significantly improved. The speed response time of the motor was shortened by 41.1%, the thrust response time was shortened by 10.6%, and the thrust bandwidth was extended to 37.02 Hz. Combining all the results, it was concluded that the optimized MMT-PMSLM has better thrust performance and dynamic response capability, and the results measured in all experiments are basically consistent with the high-precision FEM analysis results. All experiments further verified that the modeling optimization method proposed in this paper can accurately and effectively improve the dynamic performance of the motor.

[0017] Finally, the proposed DOA-PI-GAN modeling optimization method can make motor optimization design results more reliable than the traditional PMSLM modeling optimization method and significantly improve the dynamic performance of the linear servo system. This modeling optimization method can be applied to the optimization design of any other motor structure, providing new research ideas and methods for future motor optimization design. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Schematic diagram of the MMT-PMSLM topology structure in the MMT-PMSLM dynamic performance modeling optimization method based on PI-GAN proposed in this invention.

[0019] Figure 2 Schematic diagram of the equivalent analysis of the winding coil in the MMT-PMSLM dynamic performance modeling optimization method based on PI-GAN proposed in this invention.

[0020] Figure 3 The thrust performance comparison chart of the initial model analysis and FEM solution under MMT-PMSLM steady-state thrust.

[0021] Figure 4 Schematic diagram of the internal structure of the LSTM network.

[0022] Figure 5 This is a schematic diagram comparing the calculation results of different modeling methods in the MMT-PMSLM dynamic performance modeling optimization method based on PI-GAN proposed in this invention.

[0023] Figure 6 This is a schematic diagram of the iterative optimization process of solving the four optimization algorithms DOA, WOA, SSA, and PSO under the PI-GAN model in the MMT-PMSLM dynamic performance modeling optimization method based on PI-GAN proposed in the present invention.

[0024] Figure 7 Schematic diagram of the MMT-PMSLM prototype performance test experimental platform in the MMT-PMSLM dynamic performance modeling optimization method based on PI-GAN proposed in this invention.

[0025] Figure 8 The steady-state thrust performance test results of the optimized MMT-PMSLM prototype at rated current density.

[0026] Figure 9 Schematic diagram of the thrust-current characteristics of the optimized MMT-PMSLM prototype.

[0027] Figure 10 Schematic diagram showing how the thrust density and motor constant of the MMT-PMSLM can be calculated based on the thrust performance results of the MMT-PMSLM prototype.

[0028] Figure 11 This is a schematic diagram of the back electromotive force results of the prototype at different speeds measured by continuously increasing the speed.

[0029] Figure 12 Schematic diagram of the thrust and speed response experimental results of the optimized MMT-PMSLM prototype.

[0030] Figure 13 Schematic diagram of the thrust bandwidth experimental results of the optimized MMT-PMSLM prototype. DETAILED DESCRIPTION

[0031] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0032] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this invention pertains. The terms used herein in the specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Example 1

[0033] The present invention proposes a method for dynamic performance modeling and optimization of a moving magnet permanent magnet synchronous linear motor (MMT-PMSLM) based on PI-GAN. The method specifically includes the following steps: Based on the linear servo system's demand for high dynamic performance, a performance analytical model of the MMT-PMSLM dynamic response is established based on thrust performance, mechanical and electrical time constants, motor constants, and thrust bandwidth. Constructing a PI-GAN-based MMT-PMSLM dynamic performance prediction model based on the performance analysis model; Iteratively optimizing the dynamic performance prediction model through a dream optimization DOA algorithm to generate optimized structural parameters; Among them, the dynamic performance prediction model includes a generator and a discriminator, taking the structural parameters of the performance analysis model as the input of the generator, and outputting the predicted values ​​based on thrust performance, mechanical and electrical time constants, motor constants and thrust bandwidth based on the generator; the discriminator is established based on the fitting performance discrimination model of the physical model embedded in the machine learning model.

[0034] In this embodiment, if Figure 1 As shown, the MMT-PMSLM topology of this embodiment mainly consists of a dual primary stator and a secondary mover. In order to improve the thrust density of the MMT-PMSLM and balance the normal suction between the iron core and the mover, a double-sided primary winding structure is adopted. In order to reduce copper loss, improve the motor constant, and reduce the motor space harmonics, a fractional slot centralized winding structure is adopted. In order to increase the air gap magnetic field density, reduce tooth leakage magnetic flux, and ensure the sinusoidal nature of the air gap magnetic field, an I-shaped iron core structure is adopted. The mover consists of permanent magnets with staggered NS poles and an aluminum plate support workbench.

[0035] Based on the requirement of the linear servo system for high dynamic performance, the steps of establishing a performance analytical model of the MMT-PMSLM dynamic response based on thrust performance, mechanical and electrical time constants, motor constants, and thrust bandwidth include obtaining an inductance matrix of a three-phase winding by analyzing a winding inductance analytical model of the MMT-PMSLM, and performing a dynamic performance analysis on the MMT-PMSLM after decoupling current and voltage equations based on a two-dimensional analytical model of the MMT-PMSLM to generate a performance analytical model of the MMT-PMSLM dynamic response based on thrust performance, mechanical and electrical time constants, motor constants, and thrust bandwidth; Among them, thrust performance, mechanical and electrical time constants, motor constants and thrust bandwidth are specifically thrust density , thrust fluctuation , mechanical time constant , electrical time constant , motor constant and thrust bandwidth .

[0036] In this embodiment, if Figure 2 As shown in the figure, in order to establish the three-phase inductance analysis model of the motor under study, the actual racetrack-shaped coil wound by the motor is equivalent to a rectangular ring coil for analysis.

[0037] like Figure 2 As shown, the calculation formula for the resistance value of each phase is as follows: ; In the above formula, is the number of turns of each winding coil, is the distance between two adjacent cores, is the length of the core. In this paper, the length of the core is equal to the length of the permanent magnet. equal, is the width of the core, is the stack height of the coil, is the width of a single-sided coil, It is a correction factor obtained based on actual winding process experience. is 0.7.

[0038] In this embodiment, the step of obtaining the inductance matrix of the three-phase winding by analyzing the winding inductance analysis model of the MMT-PMSLM includes: , winding mutual inductance and tank leakage inductance Get the total inductance of the motor , total motor inductance Expressed as: ; Tank leakage inductance Expressed as: ; Based on the winding self-inductance , winding mutual inductance and tank leakage inductance Get the inductance matrix of the three-phase winding : ; in, is the magnetic permeability of air, is the number of turns of each phase winding coil, is the specific leakage permeability coefficient, is the core length.

[0039] It should be noted that the winding self-inductance in this embodiment , winding mutual inductance It can be calculated according to the Neumann integral formula, which is not described here in detail. It is a conventional calculation method.

[0040] In this embodiment, based on the two-dimensional analysis model of MMT-PMSLM, the air gap flux distribution in the y-axis direction in the air gap is obtained by the equivalent magnetization method and the equivalent surface current method, so as to calculate the three-phase winding back electromotive force; the voltage-current equation in the two-phase coordinate system is obtained by performing Clark-park transformation and according to the conditional constraints of the initial current of the coil and the maximum load current, thereby realizing the decoupling of the current-voltage equation of MMT-PMSLM.

[0041] In this embodiment, it should be noted that the air gap flux distribution in the y-axis direction, the back electromotive force of the three-phase winding calculated based on the flux distribution, and the response equation of the q-axis current component obtained after the coordinate change are all based on Figure 1 and Figure 2 The whole process can be deduced through a known two-dimensional analysis model. The purpose of the present invention is to achieve the decoupling of the current-voltage equation, and the specific calculation process is not limited.

[0042] In this embodiment, the MMT-PMSLM is used as a driving motor for a high-precision linear servo system, and its output thrust performance, thrust bandwidth, motor time constant, and motor constant have always been key targets in the field of linear motors. The steps of performing dynamic performance analysis on the MMT-PMSLM to generate a performance analysis model of the MMT-PMSLM dynamic response based on thrust performance, mechanical and electrical time constants, motor constants, and thrust bandwidth specifically include: Since the thrust of a linear motor increases with the increase of its volume, in order to prevent the motor volume from expanding excessively in order to obtain greater thrust, a thrust density calculated based on the ratio of thrust to volume is introduced as an evaluation indicator, and the thrust density is obtained based on the ratio of average thrust to motor volume. ; Based on the average thrust, the thrust fluctuation of MMT-PMSLM is obtained ; Mechanical time constant , electrical time constant Reflects the speed response capability and thrust response capability of the motor, motor constant Reflects the comprehensive design level of the motor's electromagnetic and heat dissipation, thrust bandwidth It reflects the ability of the motor to maintain a stable output thrust under a certain speed range. It should be noted that in this embodiment, the thrust density is mainly obtained when the dynamic performance analysis of MMT-PMSLM is performed. , thrust fluctuation , mechanical time constant , electrical time constant , motor constant and thrust bandwidth That is, there is no limitation on the specific calculation process of each performance data. These performance data will serve as the focus of subsequent research in this embodiment. The above content also explains the role of these performance data. The following will focus on analyzing how to perform PI-GAN modeling and DOA optimization.

[0043] In order to comprehensively measure the comprehensive dynamic performance of MMT-PMSLM, it is necessary to construct a judgment function that comprehensively evaluates the various output performances of the motor based on the characteristics of each performance. and low thrust fluctuations The influence on the dynamic performance of the linear servo motor system is the greatest, so a larger weight coefficient is required. Secondly, the change of motor structural parameters affects the mechanical time constant. and electrical time constant The influence of is opposite. And under different working conditions of the actual operation of the motor, the mechanical time constant of the motor The impact on the dynamic performance of the servo system is far greater than the electrical time constant Large, so the mechanical time constant Need to be greater than the electrical time constant has a larger weight coefficient. Finally, the motor constant and thrust bandwidth The weight coefficient of MMT-PMSLM is moderate. Therefore, the comprehensive dynamic performance evaluation function of MMT-PMSLM is can be expressed as: ; The smaller the σ value is, the better the comprehensive dynamic performance of MMT-PMSLM is.

[0044] like Figure 3 As shown in the figure, the MMT-PMSLM performance analytical model established in this section is compared and verified by a high-precision finite element model. Figure 3 The following figure compares the thrust performance of the MMT-PMSLM using the analytical and FEM solutions for the initial model under steady-state thrust. The MMT-PMSLM performance results calculated using the analytical AM and finite element (FEM) modeling methods show that the analytical models for each MMT-PMSLM performance derived in this example exhibit certain errors compared to the high-precision FEM. This indicates that the AM of the MMT-PMSLM is not suitable for direct algorithm optimization. However, the computational cost of directly optimizing the algorithm for the high-precision FEM is too high, making direct algorithm optimization unsuitable.

[0045] In order to establish a high-precision prediction model, it is necessary to require that the number of samples in the sample set not only meets the training scale, but also has typicality and uniformity. Therefore, this embodiment uses the orthogonal test method to establish the sample set required for the PI-GAN model. The structural parameters that affect the output performance obtained according to the MMT-PMSLM performance analysis model are , , , , , , so a 6-factor 4-level orthogonal experimental design is established, as shown in Figure 1: ; The accuracy of the samples directly affects the accuracy of the final trained model. Therefore, this example uses a co-simulation of the finite element model and the control model to generate a sample set. The MMT-PMSLM finite element model is embedded in PID control to achieve closed-loop co-simulation, resulting in output that better reflects the motor's actual operating conditions. The sample data generated based on the FEM-control co-simulation is shown in Table 2, resulting in a total of 4096 samples as the training set for the PI-GAN model.

[0046] ; Traditional nonlinear regression models or neural network prediction models are prone to overfitting due to the lack of guidance and constraints from physical laws, resulting in poor generalization performance. Therefore, this implementation uses LSTM and physical models to interact with each other in a GAN architecture to generate a high-precision, strong generalization MMT-PMSLM dynamic performance prediction model. The proposed PI-GAN model for MMT-PMSLM dynamic performance mainly includes four aspects: the generator (G) - the establishment of an LSTM-based MMT-PMSLM dynamic performance prediction model; the discriminator (D) - the establishment of a fitting performance discrimination model based on the physical model; adversarial training using random structural parameter sets as input; and the construction of a physics-guided loss function.

[0047] like Figure 4 As shown, the LSTM-based MMT-PMSLM dynamic performance prediction model is established: the LSTM network structure is as follows Figure 4 As shown, LSTM is used to directly perform regression fitting on the sample data in Table 2, where The input to the network can be expressed as: ; Among them, such as Figure 4 As shown, and tanh is the activation function, is the output of the previous hidden layer, is the memory state of the previous hidden layer neuron. and The current hidden layer state is obtained by weighted summation of the activation function and three forget gates and neuronal memory states .

[0048] Finally, the output H of the model can be expressed as: ; in The predicted values ​​of various MMT-PMSLM performances output by the sample.

[0049] This paper uses MSE as the loss function for optimizing LSTM prediction, which can be defined as: ; Where n is the total number of data, is the actual value, is the predicted value.

[0050] In this embodiment, the discriminator (D) is established based on the physical model fitting performance discrimination model: the physical model can be embedded in the machine learning model to enhance the physical interpretation and generalization ability of the model. In the PI-GAN architecture of this embodiment, the various performance output results of MMT-PMSLM (thrust density , thrust fluctuation , mechanical time constant , electrical time constant , motor constant and thrust bandwidth ) is used as physical information to form a discriminator to determine whether the output of the generator LSTM achieves fitting accuracy and conforms to physical laws. To a certain extent, it can represent the fitting accuracy and generalization ability of the model, between [0-1]. The larger the value, the better the model fit. The calculation formula is as follows: ; in is the average value of the actual sample value. Therefore, the judgment condition of the discriminator can be expressed as: ; In this embodiment, the random structural parameter set verifies that an excellent motor performance prediction model should have the ability to output high-precision outputs for any structural parameter combination. Therefore, this embodiment randomly samples the structural parameters to generate some samples and inputs them into the generator and discriminator respectively. The predicted value output by the generator is , the output of the parsing function in the discriminator is the true value According to the judgment condition of the discriminator, the discriminator judges the predicted value of the output under the random input of the generator The accuracy is checked to see if it meets the standard. If not, the generator is back-propagated and optimized by adding a physics-guided loss function. To prevent the physical model from being contaminated, the discriminator is not back-propagated and optimized in this model. Secondly, in the construction of the physics-guided loss function, this embodiment adds a physics-guided loss to the loss function used for back-propagation optimization of the generator model to enhance the generalization ability of the model. Therefore, the loss function of the generator can be expressed as: ; Among them, BCE is the cross entropy loss function, b and c are weight coefficients, is the actual value, is the predicted value, and the value of R is equal to the resistance value of each phase in MMT-PMSLM.

[0051] Finally, the generator loss function is minimized through the loop Backpropagation optimization is performed to make the generator model generate outputs that conform to physical laws until the structural parameter set of the random input can completely pass the discriminator.

[0052] At this point, the MMT-PMSLM dynamic performance prediction model based on PI-GAN has been constructed.

[0053] Since the motor structural parameters selected during the optimization algorithm search process are random, the model needs to have high precision and strong generalization ability to ensure the subsequent effective search for the optimal objective function value. To verify the high precision and strong generalization ability of the proposed model, this embodiment further expands the selection range of the structural parameters of the test set samples as shown in Table 3: ; 200 cases were randomly sampled within the above structural parameter range to form the test sample of the PI-GAN model, as shown in Table 4: ; Figure 5 These are the calculation results of four MMT-PMSLM performance prediction models based on FEM, PI-GAN, PINN, and LSTM on the test samples.

[0054] In terms of computational accuracy, the PI-GAN method achieves higher accuracy on test samples because it incorporates physical model guidance into its high-precision prediction model. This indicates that the PI-GAN method has stronger generalization capabilities. The PINN method directly incorporates physical loss guidance into the neural network. While this enhances generalization capabilities, it also reduces model accuracy during training. The LSTM method, lacking physical guidance, exhibits poor accuracy on test samples, indicating poor generalization capabilities.

[0055] In terms of computational cost, the computation time of PI-GAN, PINN, and LSTM is similar. The computation time of PI-GAN for 200 cases does not exceed 1 minute, which is completely acceptable.

[0056] Therefore, the PI-GAN motor performance calculation model proposed in this embodiment can have stronger generalization ability and interpretability while maintaining fast calculation and high accuracy.

[0057] In this embodiment, the mapping relationship between the motor's structural parameters and output performance is complex and nonlinear. Compared to classic algorithms such as the Whale Optimization Algorithm (WOA), the Sparrow Search Algorithm (SSA), and the Particle Swarm Optimization Algorithm (PSO), the Dream Optimization Algorithm (DOA) has higher convergence and reliability, and better global search capabilities in multi-modal problems. Therefore, DOA has a faster convergence speed and higher solution quality for the multi-objective optimization problem of the dynamic performance of the MMT-PMSLM. This embodiment will iteratively optimize the MMT-PMSLM based on the DOA and the PI-GAN prediction model of the motor's dynamic performance to solve the globally optimal motor structural parameters.

[0058] In order to constrain the three-dimensional size of the motor, the width of the permanent magnet Set to a fixed value to obtain effective optimization results. This embodiment uses the comprehensive dynamic performance evaluation function : ; The multi-objective optimization problem of MMT-PMSLM dynamic performance is transformed into a single-objective optimization problem. Therefore, based on the PI-GAN prediction model of motor dynamic performance, the mathematical description of the MMT-PMSLM optimization problem is: ; In this embodiment, the DOA optimization process mainly includes the following three stages: (1) Initialization phase: Similar to other meta-heuristic algorithms, in the initialization phase, DOA needs to first generate a random population in the search space as the initial population to start the optimization process of the algorithm. The calculation formula for the initial population is as follows: ; in, is the i-th individual in the population, and are the lower limit and upper limit of the motor structure parameter range respectively, rand is a random number [0-1], and N is the number of individuals, which is set to 50 in this embodiment.

[0059] (2) Exploration phase: The number of iterations from 0 to 0.9gmax is the exploration phase. Each iteration is considered a dreaming behavior, and the optimal solution is sought through continuous iteration. First, the population is divided into q groups. Before each dream, all individuals in each group are shown their best dream in the past, that is, the best individual in the past iteration. They will remember the position of the best individual before dreaming and reset their own position to the position of the best individual.

[0060] ; Among them, q = 1, 2, 3, 4, 5. Each individual will randomly forget some dimensional information when dreaming, and randomly select kq dimensions and update the position in these dimensions.

[0061] ; Wherein randi(a,b) is a random integer selected in the range of a to b, Dim is the total dimension of the independent variable, and in this embodiment, the total dimension of the structural parameter variable is 6.

[0062] The forgetting and replenishing strategy allows individuals to forget and self-organize position information in the forgetting dimension. The formula for updating the position is as follows: ; in .

[0063] In order to prevent falling into local optimality, DOA adopts a dream sharing strategy to enhance the ability to escape from local optimality. This strategy allows individuals to randomly obtain the positions of other individuals in the forgotten dimension. The formula for updating the position is as follows: ; Where m is a natural number randomly selected in the range [1,50] for each dimension.

[0064] (3) Development phase: The number of iterations from 0.9gmax to gmax is the development phase. Before each dream, the best individual in the previous iteration is shown to all individuals. All individuals in the population have the same number of forgetting dimensions. .

[0065] ; Randomly select from the total dimensions And update the position in these dimensions.

[0066] ; The forgetting and replenishing strategies are updated in this phase to: ; in In order to avoid DOA falling into local optimum and premature convergence, this embodiment sets the maximum number of iterations to .

[0067] In this embodiment, Figure 6 This example demonstrates the iterative optimization process for solving the mathematical description of the MMT-PMSLM optimization problem using four optimization algorithms: Direct Array (DOA), WoA, SSA, and PSO. A comparison of the iterative processes reveals that the DOA algorithm employed in this example achieves higher solution quality and faster convergence than other algorithms in optimizing the comprehensive dynamic performance of the MMT-PMSLM. Table 5 compares the structural parameters and performance of the MMT-PMSLM before and after optimization.

[0068] ; After finite element FEM verification, the average thrust of the MMT-PMSLM optimized by the DOA-PI-GAN method was increased from 969.9N to 1612.5N, an increase of 66.3%, and the thrust density was increased from 1.155N / cm 3 Increased to 1.441N / cm 3 , increased by 24.8%, thrust fluctuation was reduced from 21.9% to 2.32%, mechanical time constant was reduced from 25.3ms to 12.7ms, reduced by 49.8%, electrical time constant was reduced from 4.7ms to 4.2ms, reduced by 10.6%, motor constant was reduced from 31.72N / Increased to 45.36 N / , an improvement of 43%, and the thrust bandwidth was expanded from 35.56Hz to 37.11Hz. By comparing the comprehensive dynamic performance evaluation index σ of the MMT-PMSLM before and after optimization, it can be seen that the dynamic performance of the optimized MMT-PMSLM has been greatly improved, proving the effectiveness and superiority of the DOA-PI-GAN method for optimizing the dynamic performance of the MMT-PMSLM. Example 2

[0069] In order to further verify the effectiveness and superiority of the DOA-PI-GAN modeling optimization MMT-PMSLM method proposed in this embodiment, this embodiment manufactured an MMT-PMSLM experimental prototype based on the DOA optimization results. Figure 7 The MMT-PMSLM prototype performance test platform was demonstrated. It primarily includes a cSPACE hardware-in-the-loop control system, a driver board, and a magnetic grating position sensor to implement three closed-loop motor motion. The force loading system performs load testing using the MMT-PMSLM's load platform.

[0070] Thrust performance and motor constant experimental verification: The steady-state thrust performance test results of the optimized MMT-PMSLM prototype at rated current density are as follows: Figure 8 As shown in the figure, the measured thrust performance output of the MMT-PMSLM prototype is essentially consistent with the optimized FEM analysis results. The average thrust from the FEM analysis of the MMT-PMSLM before optimization was 969.9 N, with a thrust fluctuation of 21.9%. The average thrust from the FEM analysis after optimization was 1612.5 N, with a thrust fluctuation of 2.32%. The measured average thrust from the optimized prototype was 1593.2 N, with a thrust fluctuation of 2.59%. The slight discrepancy between the measured results and the FEM analysis is primarily due to manufacturing errors and friction losses in the rotor's motion.

[0071] The thrust-current characteristics of the prototype after MMT-PMSLM optimization are as follows Figure 9 Experimental results show that the optimized MMT-PMSLM exhibits a good linear relationship between current and thrust in the range of 0 to 5 A. The measured thrust constant of the optimized MMT-PMSLM is 318.6 N / A, a significant improvement over the initial FEM analysis result of 193.9 N / A.

[0072] According to the thrust performance results of the MMT-PMSLM prototype, the thrust density and motor constant of the MMT-PMSLM can be calculated as follows: Figure 10 The thrust density and motor constant of the prototype are 1.426N / cm 3 and 44.94N / , which is basically consistent with the FEM calculation results after optimization.

[0073] By comparing and analyzing the measured results of the thrust performance of the MMT-PMSLM prototype with the FEM results before and after optimization, the DOA-PI-GAN modeling optimization method proposed in this embodiment increased the measured average thrust of the motor by 64.3%, reduced the thrust fluctuation to 2.59%, increased the thrust density by 23.5%, and increased the motor constant by 41.7%. The experiment verified that the thrust performance of the optimized MMT-PMSLM meets the application requirements of high-precision machining platforms. It should be noted that this embodiment also carried out experimental verification of no-load back electromotive force and back electromotive force constant, experimental verification of mechanical and electrical time constants, and experimental verification of thrust bandwidth. It can be seen that the back electromotive force results of the prototype at different speeds are measured by continuously increasing the speed. Figure 11 As shown in the figure, the back electromotive force constant of the MMT-PMSLM after optimization is 241.7V / m / s, which is significantly improved compared to the 147.1V / m / s of the FEM analysis result before optimization. The motor optimized by the DOA-PI-GAN optimization method proposed in this embodiment not only improves the fundamental amplitude of the back electromotive force but also weakens the odd harmonics, and has good sinusoidal characteristics. The thrust and speed response experimental results of the optimized MMT-PMSLM prototype are shown in the figure. Figure 12As shown. The mechanical time constant represents the time required for the motor speed to rise from standstill to 63.2% of the no-load speed. The measured mechanical time constant is 14.9ms, which is slightly different from the 12.7ms of the optimized FEM analysis result. The reason is the hardware delay of the test system. This difference is acceptable. The electrical time constant represents the time required for the current to rise from 0 to 63.2% of the rated current, that is, the time required to reach 63.2% of the rated steady-state thrust. The measured electrical time constant is 4.2ms, which is consistent with the optimized FEM analysis result. By comparing the measured results of the thrust and speed response of the MMT-PMSLM prototype with the FEM results before and after optimization, the DOA-PI-GAN optimization method proposed in this embodiment reduces the measured thrust response time and speed response time of the motor by 10.6% and 41.1%, respectively. The experiment verifies that the optimized MMT-PMSLM has better dynamic performance. The experimental results of the thrust bandwidth of the optimized MMT-PMSLM prototype are shown in Figure 13 The measured thrust bandwidth of the optimized MMT-PMSLM is 37.02 Hz, which is consistent with the optimized FEM analysis result of 37.11 Hz. The experiment verified that the optimized MMT-PMSLM has a larger thrust bandwidth, which means the motor has the ability to achieve faster speeds and maintain greater thrust.

[0074] Based on all the above experimental verifications, combined with the comprehensive dynamic performance evaluation function of MMT-PMSLM It can be concluded that the comprehensive dynamic performance of the MMT-PMSLM after optimization using the DOA-PI-GAN modeling optimization method improved by 92.9%. Based on the above prototype experimental results, the correctness and superiority of the DOA-PI-GAN optimization method proposed in this embodiment are verified. Example 3

[0075] This embodiment provides an electronic device corresponding to the PI-GAN-based MMT-PMSLM dynamic performance modeling optimization method provided in Example 1. The electronic device can be, for example, a mobile phone, a laptop computer, a tablet computer, a desktop computer, etc., to execute the method of Example 1.

[0076] The processing device includes a processor, a memory, a communication interface, and a bus. The processor, memory, and communication interface are connected via the bus to facilitate communication between them. The memory stores a computer program executable on the processor. When the processor executes the computer program, it executes the PI-GAN-based MMT-PMSLM dynamic performance modeling and optimization method provided in Example 1.

[0077] In some embodiments, the memory may be a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory.

[0078] In other embodiments, the processor may be a central processing unit (CPU), a digital signal processor (DSP), or other general-purpose processors, which are not limited herein. Example 4

[0079] The PI-GAN-based MMT-PMSLM dynamic performance modeling optimization method of this embodiment 1 can be specifically implemented as a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing the PI-GAN-based MMT-PMSLM dynamic performance modeling optimization method described in this embodiment 1.

[0080] Computer readable storage media can be tangible devices that hold and store instructions used by instruction execution devices. Computer readable storage media can be, for example, but not limited to, electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any combination thereof.

[0081] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of the systems, methods, and computer program products according to various embodiments of the present application. Each box in the flowchart or block diagram may represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function.

[0082] 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 it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

[0083] The above embodiments are only used to illustrate the present invention, wherein the structure, connection mode and manufacturing process of each component can be changed. Any equivalent transformations and improvements based on the technical solution of the present invention should not be excluded from the scope of protection of the present invention.

Claims

1. The MMT-PMSLM dynamic performance modeling optimization method based on PI-GAN is characterized by: The method specifically involves modeling and optimizing the dynamic performance of a moving magnet permanent magnet synchronous linear motor (MMT-PMSLM) based on a physical information embedding generative adversarial network (PI-GAN), and includes the following steps: Based on the linear servo system's demand for high dynamic performance, a performance analytical model of the MMT-PMSLM dynamic response is established based on thrust performance, mechanical and electrical time constants, motor constants, and thrust bandwidth. Constructing a PI-GAN-based MMT-PMSLM dynamic performance prediction model based on the performance analysis model; Iteratively optimizing the dynamic performance prediction model through a dream optimization DOA algorithm to generate optimized structural parameters; Among them, the dynamic performance prediction model includes a generator and a discriminator, taking the structural parameters of the performance analysis model as the input of the generator, and outputting the predicted values ​​based on thrust performance, mechanical and electrical time constants, motor constants and thrust bandwidth based on the generator; the discriminator is established based on the fitting performance discrimination model of the physical model embedded in the machine learning model.

2. The MMT-PMSLM dynamic performance modeling optimization method based on PI-GAN according to claim 1 is characterized in that: Based on the requirement of the linear servo system for high dynamic performance, the steps of establishing a performance analytical model of the MMT-PMSLM dynamic response based on thrust performance, mechanical and electrical time constants, motor constants, and thrust bandwidth include obtaining an inductance matrix of a three-phase winding by analyzing a winding inductance analytical model of the MMT-PMSLM, and performing a dynamic performance analysis on the MMT-PMSLM after decoupling current and voltage equations based on a two-dimensional analytical model of the MMT-PMSLM to generate a performance analytical model of the MMT-PMSLM dynamic response based on thrust performance, mechanical and electrical time constants, motor constants, and thrust bandwidth; Among them, thrust performance, mechanical and electrical time constants, motor constants and thrust bandwidth are specifically thrust density , thrust fluctuation , mechanical time constant , electrical time constant , motor constant and thrust bandwidth .

3. The MMT-PMSLM dynamic performance modeling optimization method based on PI-GAN according to claim 2 is characterized in that, The steps of obtaining the inductance matrix of the three-phase winding by analyzing the winding inductance analysis model of MMT-PMSLM include: , winding mutual inductance and tank leakage inductance The total motor inductance L is obtained and is expressed as: ； Tank leakage inductance Expressed as: ； Based on the winding self-inductance , winding mutual inductance and tank leakage inductance Get the inductance matrix of the three-phase winding : ； in, is the magnetic permeability of air, is the number of coils of each phase winding, is the specific leakage permeability coefficient, and l is the core length.

4. The MMT-PMSLM dynamic performance modeling optimization method based on PI-GAN according to claim 3 is characterized in that: The steps for decoupling the current and voltage equations based on the two-dimensional analysis model of MMT-PMSLM specifically include: according to the two-dimensional analysis model of MMT-PMSLM, the air gap flux distribution in the y-axis direction in the air gap is obtained by the equivalent magnetization method and the equivalent surface current method, thereby calculating the three-phase winding back electromotive force; through the voltage and current equations in the two-phase coordinate system and performing Clark-park transformation and according to the conditional constraints of the initial current and maximum load current of the coil, the response equation of the q-axis current component is obtained, thereby realizing the decoupling of the current and voltage equations of MMT-PMSLM.

5. The MMT-PMSLM dynamic performance modeling optimization method based on PI-GAN according to claim 4 is characterized in that, The steps of performing dynamic performance analysis on the MMT-PMSLM to generate a performance analytical model of the MMT-PMSLM dynamic response based on thrust performance, mechanical and electrical time constants, motor constants, and thrust bandwidth specifically include: obtaining the average thrust of the MMT-PMSLM in steady state, and obtaining the thrust density based on the ratio of the average thrust to the motor volume. ; Based on the average thrust, the thrust fluctuation of MMT-PMSLM is obtained ; Mechanical time constant , electrical time constant Reflects the speed response capability and thrust response capability of the motor, motor constant Reflects the comprehensive design level of the motor's electromagnetic and heat dissipation, thrust bandwidth It reflects the motor's ability to maintain stable output thrust within a certain speed range.

6. The MMT-PMSLM dynamic performance modeling optimization method based on PI-GAN according to claim 5 is characterized in that: The step of performing dynamic performance analysis on the MMT-PMSLM to generate a performance analytical model of the MMT-PMSLM dynamic response based on thrust performance, mechanical and electrical time constants, motor constants and thrust bandwidth also includes a step of analyzing the MMT-PMSLM dynamic performance based on thrust density. , thrust fluctuation , mechanical time constant , electrical time constant , motor constant and thrust bandwidth Generate comprehensive dynamic performance evaluation function of MMT-PMSLM , and is expressed as follows: ； in, The smaller the value, the better the comprehensive dynamic performance of MMT-PMSLM.

7. The MMT-PMSLM dynamic performance modeling optimization method based on PI-GAN according to claim 6 is characterized in that: The loss function of the generator is expressed as: ； Among them, BCE is the cross entropy loss function, b and c are weight coefficients, is the actual value, is the predicted value, and the value of R is equal to the resistance value of each phase in MMT-PMSLM.

8. The MMT-PMSLM dynamic performance modeling optimization method based on PI-GAN according to claim 6 is characterized in that: The step of iteratively optimizing the dynamic performance prediction model by the dream optimization DOA algorithm and generating optimized structural parameters specifically includes the following steps: The multi-objective optimization problem of the MMT-PMSLM dynamic performance is transformed into a single-objective optimization problem. Based on the DOA algorithm and the PI-GAN prediction model of the motor dynamic performance, the MMT-PMSLM is iteratively optimized to generate the optimized structural parameters.

9. An electronic device, characterized in that: include: A memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the MMT-PMSLM dynamic performance modeling optimization method based on PI-GAN can be implemented as described in any one of claims 1 to 8.

10. A computer storage medium for receiving a user input program, characterized in that: When the computer program stored in the storage medium is executed by the processor, it can be based on the MMT-PMSLM dynamic performance modeling optimization method based on PI-GAN according to any one of claims 1 to 8.

Images