CN110059348B - Axial split-phase magnetic suspension flywheel motor suspension force numerical modeling method - Google Patents

Axial split-phase magnetic suspension flywheel motor suspension force numerical modeling method Download PDF

Info

Publication number
CN110059348B
CN110059348B CN201910184197.1A CN201910184197A CN110059348B CN 110059348 B CN110059348 B CN 110059348B CN 201910184197 A CN201910184197 A CN 201910184197A CN 110059348 B CN110059348 B CN 110059348B
Authority
CN
China
Prior art keywords
model
learning machine
extreme learning
levitation
parameters
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201910184197.1A
Other languages
Chinese (zh)
Other versions
CN110059348A (en
Inventor
朱志莹
朱金
孙玉坤
郭旋
姚郅勋
孟高军
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing Institute of Technology
Original Assignee
Nanjing Institute of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing Institute of Technology filed Critical Nanjing Institute of Technology
Priority to CN201910184197.1A priority Critical patent/CN110059348B/en
Publication of CN110059348A publication Critical patent/CN110059348A/en
Application granted granted Critical
Publication of CN110059348B publication Critical patent/CN110059348B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/17—Mechanical parametric or variational design
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/23—Design optimisation, verification or simulation using finite element methods [FEM] or finite difference methods [FDM]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/10—Numerical modelling
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
    • Y02E60/16—Mechanical energy storage, e.g. flywheels or pressurised fluids

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Geometry (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Computer Hardware Design (AREA)
  • General Engineering & Computer Science (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Control Of Ac Motors In General (AREA)

Abstract

The invention discloses a method for modeling the levitation force value of an axial split-phase magnetic levitation flywheel motor, which comprises the following steps: simulation and experimental design, sample collection and processing, offline model training and online model optimization. On the one hand, the invention improves the adaptability and the robustness of the model to parameter changes based on the extreme learning machine of principal component analysis, realizes the rapid and accurate modeling of small sample data, and improves the precision and the speed of the levitation force model. On the other hand, a differential evolution algorithm is introduced to optimize the network structure, so that the modeled model can meet the precision requirement and the control real-time requirement, the problems of excessive hidden layer neurons and huge network structure are avoided, the calculation speed of the model is improved, and the model is more suitable for modeling of the motor.

Description

Axial split-phase magnetic suspension flywheel motor suspension force numerical modeling method
Technical Field
The invention discloses a numerical modeling method for the levitation force of an axial split-phase magnetic levitation flywheel motor, and belongs to the technical field of magnetic levitation flywheel motors.
Background
The flywheel energy storage system is a physical energy storage device for electromechanical energy conversion, has the advantages of high specific power, small volume, long service life, quick charge and discharge, cleanness, no pollution and the like, and is a novel energy storage technology with high research value and wide application prospect. The axial split-phase magnetic suspension flywheel motor further improves the high-speed performance and the running efficiency of the motor through the active control of the self-suspension force on the basis of fully keeping the high-speed excellent characteristic of the switch reluctance motor. The flywheel energy storage device is introduced into the flywheel energy storage device, so that the suspension support with ultra-low power consumption and high-speed and high-efficiency charge-discharge integrated operation of the system can be realized, the system loss and the volume are greatly reduced, the suspension performance, the critical rotation speed and the power density are improved, and the flywheel energy storage suspension support and the energy conversion system are one of ideal choices. However, the winding-magnetic circuit-electromagnetic force of the traditional double-winding axial split-phase magnetic suspension flywheel motor has complex electromagnetic strong coupling relation, and the analysis and establishment of a suspension force model are difficult.
Representative suspension force modeling methods at present are a fuzzy mathematical method, a support vector machine method, an artificial neural network method and the like. The parameters of the model, such as penalty parameters and kernel functions, are difficult to select when the support vector machine is applied, and the parameters influence the prediction accuracy to a certain extent; the determination of the weight coefficient of the fuzzy mathematical method model is easily influenced by subjective factors; the artificial neural network is widely focused on the characteristics of strong autonomous learning capability, strong memory capability, strong nonlinear parallel processing capability, strong fault tolerance capability and the like, wherein the application of the artificial neural network is more widely single hidden layer feedforward neural network (single hidden layer feedforward neural network, SLFN). At present, many scholars try to use an extreme learning machine in the prediction research of a nonlinear model, but the main problems still exist that firstly, when the correlation between input variables is strong, unnecessary input dimensions are increased, which not only can lead to the rapid increase of the operation time of a neural network, but also can cause the operation waste due to the small influence of certain variables on an output result; secondly, ELM should randomly give the initial weight and threshold of the network before training the network, but cannot give an optimal value, and the selection of the initial parameters has a great influence on the final output of the network.
Disclosure of Invention
The invention aims to: in order to overcome the defects in the prior art, the invention provides the axial split-phase magnetic levitation flywheel motor levitation force numerical modeling method, which is based on an extreme learning machine for principal component analysis (Principal component analysis, PCA) to improve the adaptability and robustness of a model to parameter changes, realizes quick and accurate modeling of small sample data, and improves the precision and speed of a levitation force model. On the other hand, a differential evolution algorithm (Differential Evolution, DE) is introduced to optimize the network structure, so that the modeled model can meet the precision requirement and achieve the purpose of controlling the real-time property, the problems of excessive hidden layer neurons and huge network structure are avoided, the calculation speed of the model is improved, and the model is more suitable for modeling of the motor.
The technical scheme is as follows: in order to achieve the above purpose, the invention adopts the following technical scheme:
a method for modeling the levitation force value of an axial split-phase magnetic levitation flywheel motor comprises the following steps:
and step 1, constructing a motor finite element simulation and experiment prototype model according to motor parameters, and analyzing the levitation force characteristics of the motor under different operation conditions and working modes.
And 2, analyzing main factors affecting the levitation force, determining input and output variables of the extreme learning machine model, collecting sample data sets input and output under corresponding working conditions and working modes, and carrying out feature extraction and dimension reduction on the obtained data sets based on a PCA algorithm to obtain a dimension-reduced sample input data set and output set.
And 3, optimizing the weight and the threshold parameters of the extreme learning machine network by adopting a differential evolution algorithm, acquiring the optimal weight and the threshold parameters, determining the number of hidden layer nodes on the basis of the principle that the number of hidden layer nodes is smaller than the number of training samples, selecting a Sig function as an excitation function to write codes, taking a sample input dataset after dimension reduction as an input dataset of the extreme learning machine model, taking an output dataset as an output dataset of the extreme learning machine model, and training the extreme learning machine model.
And 4, after the training of the extreme learning machine model is finished, setting up an online simulation and test platform, setting the same operation working condition and operation mode as the actual operation working condition and operation mode, and verifying the training to obtain the precision of the extreme learning machine model.
Step 4-1, after the training of the extreme learning machine model is finished, obtaining a suspension force predicted value of the motor suspension force model under corresponding operation conditions and working modes
Figure BDA0001992298920000023
Setting up an actual simulation and control platform, setting up and actually carrying out the same operation working condition and operation mode, and obtaining the suspension force transmission under the actual operationAnd (5) outputting a value F.
Step 4-2, selecting a suspension force predicted value calculated by the extreme learning machine model
Figure BDA0001992298920000024
And the mean square error MSE of the actual running output value F and the decision coefficient R 2 And judging whether the constructed suspension force model meets the precision requirement or not as an evaluation index.
Figure BDA0001992298920000021
Wherein:
Figure BDA0001992298920000022
respectively the suspension force predicted values of the extreme learning machine model, F J For the actual output value under the corresponding operation condition and operation mode, L is the total training sample number, R 2 To determine the coefficient S SE Is the sum of squares of the residuals, S ST Is the sum of the total squares.
Step 4-3, determining coefficient R and mean square error MSE according to the evaluation index 2 And (3) judging: and if the precision does not meet the set requirements, changing samples, parameters, hidden node numbers and the like, and retraining the extreme learning machine model. If the precision requirement is met, the construction of the suspension force numerical model is completed.
Preferably: step 1 is performed according to the following method:
and 1-1, constructing a finite element simulation model and a test prototype model of the axial split-phase magnetic suspension flywheel motor according to motor parameters.
And 1-2, analyzing the relation between the levitation force and the rotor position angle, the rotor eccentricity, the exciting current and the load torque of the motor under different running states and the influence of the gyroscopic effect through different running modes of simulating and experimental design of running conditions of rotor eccentricity, magnetic circuit saturation and gyroscopic effect and levitation, electric and power generation under the constructed finite element simulation and experimental prototype model.
Preferably: step 2 is carried out according to the following method:
step 2-1, screening the most sensitive parameters from variables affecting the output performance of the levitation force as input variables by using experiments and simulation analysis, and x 1 ,x 2 ,…x b ,…,x a B is more than or equal to 1 and less than or equal to a, b is more than or equal to 1, …, a and a are the number of key parameters finally selected, and x b The position angle theta, the X axial eccentricity X, the Y axial eccentricity Y, the current I, the torque T and the working voltage U are any variable.
And 2-2, selecting a change interval of the sensitive parameters. According to the working requirements, the processing technology and the physical constraint conditions of the flywheel system, the change interval of the selected sensitive parameters is defined, the corresponding parameters are set, and the corresponding sample data set (x) is obtained by using finite element simulation calculation 1 ,x 2 ,…,x a F), where (x) 1 ,x 2 ,…,x a ) Is the input set of the model, and F is the output set.
Step 2-3, normalizing the input data set. Assume that there are n samples, each sample having m indices, each index labeled x ij I=1, 2, …, n. j=1, 2, …, m, after normalization,
Figure BDA0001992298920000031
wherein: />
Figure BDA0001992298920000032
Normalized data matrix->
Figure BDA0001992298920000033
Step 2-4, establishing a correlation coefficient matrix of the standardized matrix:
Figure BDA0001992298920000034
step 2-5, calculating the characteristic value lambda of R c And feature vector:
α c ＝(α c1 ,α c2 ,…α cm ) T ，c＝1,2,…m (2)
step 2-6, obtaining the principal component Z of the input dataset c Selecting a retention component Z according to the accumulated contribution rate beta (k) k K is equal to or greater than 1 and equal to or less than c, so that the k value is determined, and data meeting the requirement is selected as a sample input data set to be (x) 1 ,x 2 ,…,x k )。
Figure BDA0001992298920000035
Figure BDA0001992298920000041
Preferably: in the step 2-2, the key factors which are most sensitive to the influence of the suspension force performance are obtained by analyzing the change parameters of the position angle theta, the X-axis radial eccentricity X, the Y-axis radial eccentricity Y, the current I, the torque T and the working voltage U, so that the main input variables of the extreme learning machine model are determined.
Preferably: in the step 2-2, X-axis radial eccentricity X, Y-axis radial eccentricity Y, levitation winding current I, rotor position angle theta and torque T are selected as sensitivity parameters, and corresponding levitation force value F is used as corresponding output.
Preferably: step 3 is carried out according to the following method:
and 3-1, selecting respective training rules and excitation functions for different operation conditions and operation modes, and introducing a weight and a threshold value of a differential evolution algorithm optimization learning algorithm.
Step 3-2, initializing a population: randomly selecting individuals P and forming a composition of size N P And selecting the maximum evolution algebra as g max 。
P A ＝{P A1 ，P A2 ，…，P AB } (5)
Wherein: a=1, 2, … N P B=1, 2, … D, D being the number of independent variables.
Figure BDA0001992298920000042
Wherein:
Figure BDA0001992298920000043
the upper and lower limits of the A-th argument, respectively.
Step 3-3, mutation operation: random selection of three individuals from a population
Figure BDA0001992298920000044
And are not equal to each other. />
Figure BDA0001992298920000045
Wherein: v (v) r,g+1 Is a newly constructed vector. F is a scaling factor. g is algebraic.
Only one mutation strategy is shown in the formula (7), and different strategies can be set according to different requirements.
Step 3-4, cross operation: new individuals v r,g+1 And x A Discrete interleaving to obtain updated individuals u A 。
Figure BDA0001992298920000046
Wherein: crossover probability CR [0,1]. rand (A) is a random integer between [1, D ] generated by the A-th vector.
Step 3-5, selecting operation: the new generation is compared with the adaptation value f (, of the previous generation variable, the smaller value goes to the next generation, otherwise it remains.
Figure BDA0001992298920000051
Step 3-6, repeating the above mutation, crossover and selection processes until the maximum number of iterations g max And outputting and obtaining the optimal weight and the threshold parameter of the extreme learning machine network.
Step 3-7, based on the optimized weight and threshold parameters, using the hidden layer node number smaller than the training sample numberSelecting hidden node number as principle, selecting kernel function to obtain (x) 1 ,x 2 ,…,x k ) And F is an output data set for the input data set, and training to obtain an extreme learning machine model.
Preferably: according to the working requirements, the processing technology and the physical constraint conditions of the flywheel system, the optimization interval of the selected sensitive parameters is defined, corresponding parameters are set, a corresponding sample data set (x, y, I, theta, T and F) is obtained by using finite element simulation calculation, an extreme learning machine model is trained, wherein (x, y, I, theta and T) is an input set trained by the extreme learning machine model, and F is an output set.
Preferably: based on the acquired training data set (x, y, I, theta, T), feature extraction and dimension reduction of the data set are carried out by using a PCA dimension reduction algorithm, and the input data set after dimension reduction is optimized to be (x, y, I, theta) for model training.
Preferably: the Sigmoid function, the Sine function, or the Hardlim function is selected as the kernel function.
Preferably: in the steps 2-6, the corresponding component with the accumulated contribution rate beta (k) above 85% is selected as the reserved component Z k 。
Compared with the prior art, the invention has the following beneficial effects:
1) The extreme learning machine is adopted to improve the adaptability and robustness of the model to parameter changes, so that the quick and accurate modeling of small sample data is realized, and the precision and speed of the levitation force model are improved.
2) The PCA algorithm is introduced to perform feature extraction and dimension reduction on sample data, so that the problems of memory shortage or memory overflow and the like during calculation are avoided, the machine learning speed is also increased, and the overall operation efficiency is improved.
3) The network structure is optimized by using a differential evolution algorithm, so that the modeled model can meet the precision requirement and achieve the purpose of controlling the real-time property, the problem of excessive hidden layer neurons and huge network structure is avoided, and the modeling method is more suitable for modeling of the motor.
Drawings
Fig. 1 shows specific parameters and a structure diagram of the proposed axial split-phase magnetic levitation flywheel motor, wherein fig. 1 (I) shows the specific parameters of the motor, and fig. 1 (II) shows the structure of the motor.
Fig. 2 is a flow chart of a method for modeling the levitation force value of the proposed axial split-phase magnetic levitation flywheel motor.
Fig. 3 shows the relationship between levitation force and levitation winding current I.
Fig. 4 is a graph showing the relationship between the levitation force and the change in the rotor position angle θ.
Fig. 5 shows the relationship between the levitation force and the radial eccentricity X of the rotor X.
Fig. 6 is a graph showing the relationship between the levitation force and the radial eccentricity Y of the rotor Y.
Fig. 7 is a graph showing the relationship between levitation force and torque winding current.
Fig. 8 is a diagram of initial sample data for extreme learning machine modeling.
Fig. 9 shows the contribution rates of the variable components in the PCA dimension reduction.
Fig. 10 is a diagram of sample data of the model training of the dimension-reduced extreme learning machine.
FIG. 11 is a graph showing the comparison of the predicted and actual values of levitation force before optimization.
FIG. 12 is a graph showing the comparison of the predicted and actual values of the levitation force after optimization.
Wherein 1 is a flywheel, 2 is an outer rotor, 3 is a suspension pole, 4 is a magnetism isolating sleeve, 5 is a torque pole, 6 is a suspension winding, 7 is a torque winding, 8 is a permanent magnet, 9 is a rotor core, and 10 is a stator core.
Detailed Description
The present invention is further illustrated in the accompanying drawings and detailed description which are to be understood as being merely illustrative of the invention and not limiting of its scope, and various equivalent modifications to the invention will fall within the scope of the appended claims to the skilled person after reading the invention.
The axial split-phase magnetic levitation flywheel motor levitation force numerical modeling method, as shown in figures 1 and 2, comprises the following steps:
step 1, building a finite element simulation and experimental prototype model of the motor, and analyzing the levitation force characteristics of the motor under different operation conditions and operation modes.
Step 1-1, optimizing to obtain optimal parameter configuration of the motor according to an optimal design method provided by the invention patent with application number 201811339664.5, constructing a three-dimensional finite element simulation calculation model of the axial split-phase magnetic suspension flywheel motor in finite element simulation software based on the optimized parameters, and simultaneously processing a test model machine model with the same parameter size, wherein specific parameters and structures of the motor are shown in fig. 1 (I) and fig. II.
And 1-2, analyzing the relation between the levitation force of the motor and parameters such as rotor position angle, rotor eccentricity, exciting current and load torque under the influence of different running states and gyroscopic effect through simulating and experimental design of running conditions such as rotor eccentricity, magnetic circuit saturation and gyroscopic effect and different running modes such as levitation, electric power and power generation.
And 2, analyzing main factors affecting the levitation force, determining input and output variables of the extreme learning machine model, collecting sample data sets input and output under corresponding working conditions and working modes, and carrying out feature extraction and dimension reduction on the obtained data sets based on a PCA algorithm.
Step 2-1, screening the most sensitive parameters from the variables affecting the output performance of the levitation force as input variables x by utilizing experiments and simulation analysis 1 ,x 2 ,…x b ,…,x a B is more than or equal to 1 and less than or equal to a, b is more than or equal to 1, …, a and a are the number of key parameters finally selected, and x b The position angle theta, the X axial eccentricity X, the Y axial eccentricity Y, the current I, the torque T and the working voltage U are any variable.
And 2-2, selecting a change interval of the sensitive parameters. The working requirement, the processing technology and the physical constraint condition of the flywheel system are considered, the change interval of the selected sensitive parameter is defined, the corresponding parameter is set, and the corresponding sample data set (x) is obtained by using finite element simulation calculation 1 ,x 2 ,…,x a F), where (x) 1 ,x 2 ,…,x a ) Is the input set of the model, and F is the output set.
Step 2-3, normalizing the input data set. Assume that there are n samples, eachThere are m indices, each of which is marked as x ij (i=1, 2, …, n, j=1, 2, …, m). After the standardization of the materials, the materials are processed,
Figure BDA0001992298920000071
wherein: />
Figure BDA0001992298920000072
Normalized data matrix->
Figure BDA0001992298920000073
Step 2-4, establishing a correlation coefficient matrix of the standardized matrix:
Figure BDA0001992298920000074
step 2-5, calculating the characteristic value lambda of R c And feature vector:
α c ＝(α c1 ,α c2 ,…α cm ) T ，c＝1,2,…m (2)
step 2-6, obtaining the principal component Z of the input dataset c Selecting the corresponding component with the accumulated contribution rate beta (k) above 85% as a reserved component Z k Thereby determining a k value, selecting data satisfying the requirement as a sample input data set as (x 1 ,x 2 ,…,x k )。
Figure BDA0001992298920000075
Figure BDA0001992298920000076
And 3, optimizing weight and threshold parameters of the extreme learning machine by adopting a differential evolution algorithm, and building and training an extreme learning machine model.
And 3-1, selecting proper training rules and excitation functions for different operation conditions and operation modes, introducing a differential evolution algorithm to optimize weights and thresholds of a learning algorithm, and writing program codes of an extreme learning machine.
Step 3-2, initializing a population: randomly selecting individuals P (meeting constraint conditions) and forming a composition of size N P And selecting the maximum evolution algebra as g max 。
P A ＝{P A1 ，P A2 ，…，P AB } (5)
Wherein: a=1, 2, … N P B=1, 2, … D, D being the number of independent variables.
Figure BDA0001992298920000081
Wherein:
Figure BDA0001992298920000082
the upper and lower limits of the A-th argument, respectively.
Step 3-3, mutation operation: random selection of three individuals from a population
Figure BDA0001992298920000083
And are not equal to each other.
Figure BDA0001992298920000084
Wherein: v r,g+1 Is a newly constructed vector. F is a scaling factor. g is algebraic.
Only one mutation strategy is shown in the formula (7), and different strategies can be set according to different requirements.
Step 3-4, cross operation: new individuals v r,g+1 (mutation operation) and x A (parent) discrete crossover to get updated individuals u i 。
Figure BDA0001992298920000085
Wherein: crossover probability CR [0,1]. rand (A) is a random integer between [1, D ] generated by the A-th vector.
Step 3-5, selecting operation: the new generation is compared with the adaptive value of the previous generation variable, and the smaller value enters the next generation, otherwise, the new generation is reserved.
Figure BDA0001992298920000086
Step 3-6, repeating the above mutation, crossover and selection processes until the maximum number of iterations g max And outputting and obtaining the optimal weight and the threshold parameter of the extreme learning machine network.
Step 3-7, selecting hidden layer node number based on the optimization weight and threshold parameters, selecting Sigmoid, sine or Hardlim as kernel function based on the principle that hidden layer node number is smaller than training sample number, and using (x) 1 ,x 2 ,…,x k ) And F is an output data set for inputting the data set, and training to obtain a suspension force model of the motor.
And step 4, after model training is finished, setting up an online simulation and test platform, setting and actually achieving the same operation working condition and operation mode, and verifying model accuracy.
Step 4-1, after model training is completed, obtaining a suspension force predicted value of the motor suspension force model under corresponding operation conditions and working modes
Figure BDA0001992298920000087
And (3) building an actual simulation and control platform, setting and actually carrying out the same operation working condition and operation mode, and obtaining a suspension force output value F under actual operation.
Step 4-2, selecting a suspension force predicted value calculated by the extreme learning machine model
Figure BDA0001992298920000088
And the mean square error MSE of the actual running output value F and the decision coefficient R 2 And judging whether the constructed suspension force model meets the precision requirement or not as an evaluation index.
Figure BDA0001992298920000091
Wherein:
Figure BDA0001992298920000092
respectively the suspension force predicted values of the extreme learning machine model, F J For the actual output value under the corresponding operation condition and operation mode, L is the total training sample number, R 2 To determine the coefficient, wherein S SE Is the sum of squares of the residuals, S ST Is the sum of the total squares.
Step 4-3, determining coefficient R and mean square error MSE according to the evaluation index 2 And (3) judging: and if the precision does not meet the set requirements, changing samples, parameters, hidden node numbers and the like, and retraining the extreme learning machine model. If the precision requirement is met, the construction of the suspension force numerical model is completed.
Simulation of
1. Taking a motor disclosed in a document with a Chinese patent application number of CN201610864124.3 and a name of an axial split-phase inner stator permanent magnet bias magnetic suspension switch reluctance flywheel motor as an example, a three-dimensional finite element model and an online simulation platform of the motor are constructed, and the suspension force characteristics of the motor are analyzed under a radial eccentric operation condition and a suspension operation mode as an example. The parameters and specific structures of the constructed motor are shown in fig. 1 (I) and (II).
2. The key factors which are most sensitive to the influence of the levitation force performance are analyzed and obtained by taking the position angle theta, the X-axis radial eccentricity X, the Y-axis radial eccentricity Y, the current I, the torque T and the working voltage U as the change parameters, so that the main input variables of the extreme learning machine model are determined, and the change curves of the motor levitation force along with the corresponding parameters are shown in figures 3-7. As can be seen from fig. 3, 5 and 6, the levitation force monotonically increases with the increase of the radial eccentricity x of the current I, X shaft and the radial eccentricity Y of the Y shaft, and the change is obvious. As can be seen from fig. 4, the levitation force varies with the variation of the rotor position angle θ, and increases in succession. As can be seen from fig. 7, when the levitation winding current is 1.5A and the torque winding current is 1 to 4A, the levitation force varies with the variation of the torque current, and thus it can be known that the levitation force has a certain influence on the output of the motor torque T on the levitation force. The linear relation between the voltages at two ends of the motor and the currents fed into the levitation and torque windings is considered, so that the X-axis radial eccentricity X, the Y-axis radial eccentricity Y, the levitation winding current I, the rotor position angle theta and the torque T are selected as sensitivity parameters, and the corresponding levitation force value F is used as corresponding output for reducing the dimension of calculated data and accelerating the machine learning efficiency.
3. The working requirements, the processing technology and the physical constraint conditions of the flywheel system are considered, the optimization interval of the selected sensitive parameters is defined, corresponding parameters are set, a corresponding sample data set (x, y, I, theta, T and F) is obtained through finite element simulation calculation, an extreme learning machine model is trained, wherein (x, y, I, theta and T) is an input set trained by the extreme learning machine model, F is an output set, and the initial data set is shown in fig. 8.
4. Based on the obtained training data set (X, Y, I, θ, T, F), feature extraction and dimension reduction of the data set are performed by using a PCA dimension reduction algorithm, fig. 9 is a feature value and cumulative contribution ratio of corresponding data, and each principal component with the cumulative contribution ratio greater than 85% is taken, that is, the X-axis radial eccentricity X, Y-axis radial eccentricity Y, the levitation winding current I, and the rotor position angle θ are taken as reserved principal components. Thus, the reduced-dimension input dataset is optimized to (x, y, I, θ) for model training.
5. After a training input data set is determined, a differential evolution algorithm is adopted to perform network structure optimization, an optimal weight and a threshold parameter are obtained, on the basis, the number of hidden layer nodes is determined on the basis that the number of hidden layer nodes is smaller than the number of training samples, a Sig function is selected as an excitation function to write codes, the reduced dimension (x, y, I, theta) is taken as the input data set, F is taken as the output data set, and a levitation force model of the motor is trained and built. Sample data of the extreme learning machine model training after the dimension reduction is shown in fig. 10.
6. Predicting the suspension force value obtained by the extreme learning machine
Figure BDA0001992298920000101
And (5) performing accuracy verification on the actual output F under the same operation working condition and operation mode. As can be seen from FIGS. 11 and 12, the suspension force numerical model obtained by training the unoptimized ELM modelThe error is larger, the mean square error MSE= 73.3454, and the coefficient R is determined 2 = 0.96956. After the PCA and the differential evolution algorithm are optimized, the mean square error MSE= 35.9348 between the predicted value and the actual value of the obtained suspension force numerical model is used for determining the coefficient R 2 0.990788, the precision requirement of the constructed levitation force model is met, the precision and accuracy are greatly improved, and the superiority of the method is verified.
The embodiment only carries out the numerical modeling of the levitation force of the 12/12-structure axial split-phase magnetic levitation switch reluctance flywheel motor under the radial eccentric operation working condition and the levitation working mode, and the other structure magnetic levitation switch reluctance motors can carry out the numerical modeling of the levitation force under other different operation working conditions and working modes by utilizing the technical scheme of the invention.
The invention discloses a method for modeling the levitation force value of an axial split-phase magnetic levitation flywheel motor, which mainly comprises the following steps: simulation and experimental design, sample collection and processing, offline model training and online model optimization. On the one hand, the invention improves the adaptability and the robustness of the model to parameter changes based on the extreme learning machine of principal component analysis (Principal component analysis, PCA), realizes the rapid and accurate modeling of small sample data, and improves the precision and the speed of the levitation force model. On the other hand, a differential evolution algorithm (Differential Evolution, DE) is introduced to optimize the network structure, so that the modeled model can meet the precision requirement and control real-time requirement, the problems of excessive hidden layer neurons and huge network structure are avoided, and the calculation speed of the model is improved, so that the model is more suitable for modeling of the motor.
The foregoing is only a preferred embodiment of the invention, it being noted that: it will be apparent to those skilled in the art that various modifications and adaptations can be made without departing from the principles of the present invention, and such modifications and adaptations are intended to be comprehended within the scope of the invention.

Claims (10)

1. The method for modeling the levitation force value of the axial split-phase magnetic levitation flywheel motor is characterized by comprising the following steps of:
step 1, constructing a motor finite element simulation and experiment prototype model according to motor parameters, and analyzing the levitation force characteristics of the motor under different operation conditions and operation modes;
step 2, analyzing main factors affecting the levitation force, determining input and output variables of an extreme learning machine model, collecting sample data sets input and output under corresponding working conditions and working modes, and carrying out feature extraction and dimension reduction on the obtained data sets based on a PCA algorithm to obtain a dimension-reduced sample input data set and output set;
step 3, optimizing weight and threshold parameters of the extreme learning machine network by adopting a differential evolution algorithm, acquiring optimal weight and threshold parameters, determining the number of hidden layer nodes on the basis of the weight and threshold parameters by taking the number of hidden layer nodes as a principle that the number of hidden layer nodes is smaller than the number of training samples, selecting a Sigmoid function as an excitation function to write codes, taking a sample input dataset after dimension reduction as an input dataset of the extreme learning machine model, taking an output dataset as an output dataset of the extreme learning machine model, and training the extreme learning machine model;
step 4, after the training of the extreme learning machine model is finished, setting up an online simulation and test platform, setting the same operation working condition and operation mode as the actual operation working condition and operation mode, and verifying the training to obtain the precision of the extreme learning machine model;
step 4-1, after the training of the extreme learning machine model is finished, obtaining a suspension force predicted value of the motor suspension force model under corresponding operation conditions and working modes
Figure FDA0004051826380000011
Setting up an actual simulation and control platform, setting up and actually carrying out the same operation working condition and operation mode, and obtaining a suspension force output value F under actual operation;
step 4-2, selecting a suspension force predicted value calculated by the extreme learning machine model
Figure FDA0004051826380000012
And the mean square error MSE of the actual running output value F and the decision coefficient R 2 As an evaluation index, judging whether the constructed suspension force model meets the precision requirement;
Figure FDA0004051826380000013
wherein:
Figure FDA0004051826380000014
respectively the suspension force predicted values of the extreme learning machine model, F J For the actual output value under the corresponding operation condition and operation mode, L is the total training sample number, R 2 To determine the coefficient S SE Is the sum of squares of the residuals, S ST Is the sum of the total squares;
step 4-3, determining coefficient R and mean square error MSE according to the evaluation index 2 And (3) judging: if the precision does not meet the set requirements, changing the number of samples, parameters and hidden layer nodes, and retraining an extreme learning machine model; if the precision requirement is met, the construction of the suspension force numerical model is completed.
2. The method for modeling the levitation force value of the axial split-phase magnetic levitation flywheel motor according to claim 1, which is characterized by comprising the following steps: step 1 is performed according to the following method:
step 1-1, constructing a three-dimensional finite element simulation model of an axial split-phase magnetic suspension flywheel motor in finite element simulation software according to motor parameters, and simultaneously processing a test prototype model with the same parameter size;
and 1-2, analyzing the relation between the levitation force and the rotor position angle, the rotor eccentricity, the exciting current and the load torque of the motor under different running states and the influence of the gyroscopic effect through different running modes of simulating and experimental design of running conditions of rotor eccentricity, magnetic circuit saturation and gyroscopic effect and levitation, electric and power generation under the constructed finite element simulation model and experimental prototype model.
3. The method for modeling the levitation force value of the axial split-phase magnetic levitation flywheel motor according to claim 2, which is characterized by comprising the following steps: step 2 is carried out according to the following method:
step 2-1, screening the most sensitive parameters from the variables affecting the output performance of the levitation force as input variables x by utilizing experiments and simulation analysis 1 ,x 2 ,…x b ,…,x a B is more than or equal to 1 and less than or equal to a, b is more than or equal to 1, …, a and a are the number of key parameters finally selected, and x b The position angle theta, the X axial eccentricity X, the Y axial eccentricity Y, the current I, the torque T and the working voltage U are any variable;
step 2-2, selecting a change interval of the sensitive parameters; according to the working requirements, the processing technology and the physical constraint conditions of the flywheel system, the change interval of the selected sensitive parameters is defined, the corresponding parameters are set, and the corresponding sample data set (x) is obtained by using finite element simulation calculation 1 ,x 2 ,…,x a F), where (x) 1 ,x 2 ,…,x a ) The input set is the model, and F is the output set;
step 2-3, normalizing the input data set; assume that there are n samples, each sample having m indices, each index labeled x ij I=1, 2, …, n; j=1, 2, …, m, after normalization,
Figure FDA0004051826380000021
wherein:
Figure FDA0004051826380000022
normalized data matrix->
Figure FDA0004051826380000023
Step 2-4, establishing a correlation coefficient matrix of the standardized matrix:
Figure FDA0004051826380000024
step 2-5, calculating the characteristic value lambda of R c And feature vector:
α c ＝(α c1 ,α c2 ,…α cm ) T ，c＝1,2,…m (2)
step 2-6, obtaining the principal component Z of the input dataset c Selecting a retention component Z according to the accumulated contribution rate beta (k) k ，
K is equal to or less than 1 and equal to or less than c, so that the k value is determined, and data meeting the requirement is selected as a sample input data set (x 1 ,x 2 ,…,x k )；
Figure FDA0004051826380000031
Figure FDA0004051826380000032
4. The method for modeling the levitation force value of the axial split-phase magnetic levitation flywheel motor according to claim 3, wherein the method comprises the following steps of: in the step 2-2, the key factors which are most sensitive to the influence of the suspension force performance are obtained by analyzing the change parameters of the position angle theta, the X-axis radial eccentricity X, the Y-axis radial eccentricity Y, the current I, the torque T and the working voltage U, so that the main input variables of the extreme learning machine model are determined.
5. The method for modeling the levitation force value of the axial split-phase magnetic levitation flywheel motor according to claim 4, which is characterized by comprising the following steps: in the step 2-2, X-axis radial eccentricity X, Y-axis radial eccentricity Y, levitation winding current I, rotor position angle theta and torque T are selected as sensitivity parameters, and corresponding levitation force value F is used as corresponding output.
6. The method for modeling the levitation force value of the axial split-phase magnetic levitation flywheel motor according to claim 5, which is characterized by comprising the following steps: step 3 is carried out according to the following method:
step 3-1, selecting corresponding training rules and excitation functions for different operation conditions and operation modes, introducing a weight and a threshold value of a differential evolution algorithm optimization learning algorithm, and writing program codes of an extreme learning machine;
step 3-2, initial populationAnd (3) initializing: randomly selecting individuals P and forming a composition of size N P Is determined to have the maximum evolution algebra g max ；
P A ＝{P A1 ，P A2 ，…，P AB } (5)
Wherein: a=1, 2, … N P The method comprises the steps of carrying out a first treatment on the surface of the B=1, 2, … D, D being the number of independent variables;
Figure FDA0004051826380000033
wherein:
Figure FDA0004051826380000034
the upper and lower limits of the A-th independent variable are respectively;
step 3-3, mutation operation: random selection of three individuals from a population
Figure FDA0004051826380000035
And are not equal to each other;
Figure FDA0004051826380000036
wherein: v (v) r,g+1 Is a newly constructed vector; f is a scaling factor; g is algebraic;
step 3-4, cross operation: variation operation v r,g+1 With parent x A Discrete interleaving to obtain updated individuals u A ；
Figure FDA0004051826380000041
Wherein: cross probability CR e 0, 1; rand (A) is a random integer between [1, D ] generated by the A-th vector;
step 3-5, selecting operation: comparing the adaptation value f (·) of the new generation and the previous generation variables, if the adaptation value f (·) is smaller than the adaptation value f (·) of the previous generation, entering the next generation, otherwise, keeping;
Figure FDA0004051826380000042
step 3-6, repeating the above mutation operation, crossover operation and selection operation until the maximum iteration number g max Outputting and obtaining the optimal weight and threshold parameters of the extreme learning machine network;
step 3-7, selecting hidden layer node number based on the optimized weight and threshold parameters and with hidden layer node number smaller than training sample number, selecting kernel function, and obtaining (x) 1 ,x 2 ,…,x k ) And F is an output data set for the input data set, and training to obtain an extreme learning machine model.
7. The method for modeling the levitation force value of the axial split-phase magnetic levitation flywheel motor according to claim 6, which is characterized by comprising the following steps: according to the working requirements, the processing technology and the physical constraint conditions of the flywheel system, the optimization interval of the selected sensitive parameters is defined, corresponding parameters are set, a corresponding sample data set (x, y, I, theta, T and F) is obtained by using finite element simulation calculation, an extreme learning machine model is trained, wherein (x, y, I, theta and T) is an input set trained by the extreme learning machine model, and F is an output set.
8. The method for modeling the levitation force value of the axial split-phase magnetic levitation flywheel motor according to claim 7, wherein the method comprises the following steps of: based on the acquired training data set (x, y, I, theta, T), feature extraction and dimension reduction of the data set are carried out by using a PCA dimension reduction algorithm, and the input data set after dimension reduction is optimized to be (x, y, I, theta) for model training.
9. The method for modeling the levitation force value of the axial split-phase magnetic levitation flywheel motor according to claim 8, which is characterized by comprising the following steps: the Sigmoid function, the Sine function, or the Hardlim function is selected as the kernel function.
10. The method for modeling the levitation force value of the axial split-phase magnetic levitation flywheel motor according to claim 9, which is characterized by comprising the following steps:in the steps 2-6, the corresponding component with the accumulated contribution rate beta (k) above 85% is selected as the reserved component Z k 。
CN201910184197.1A 2019-03-12 2019-03-12 Axial split-phase magnetic suspension flywheel motor suspension force numerical modeling method Active CN110059348B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910184197.1A CN110059348B (en) 2019-03-12 2019-03-12 Axial split-phase magnetic suspension flywheel motor suspension force numerical modeling method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910184197.1A CN110059348B (en) 2019-03-12 2019-03-12 Axial split-phase magnetic suspension flywheel motor suspension force numerical modeling method

Publications (2)

Publication Number Publication Date
CN110059348A CN110059348A (en) 2019-07-26
CN110059348B true CN110059348B (en) 2023-04-25

Family

ID=67316964

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910184197.1A Active CN110059348B (en) 2019-03-12 2019-03-12 Axial split-phase magnetic suspension flywheel motor suspension force numerical modeling method

Country Status (1)

Country Link
CN (1) CN110059348B (en)

Families Citing this family (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110661390B (en) * 2019-09-24 2021-05-25 江苏大学 Accurate modeling method for suspension force of 12/14 pole magnetic suspension switched reluctance motor
CN111245318B (en) * 2020-01-18 2021-11-02 浙江大学 Radial force accurate compensation decoupling control method for bearingless permanent magnet synchronous motor
CN113222307A (en) * 2020-01-21 2021-08-06 厦门邑通软件科技有限公司 Simulation method, system and equipment for generating operation behavior record set
CN113011064B (en) * 2021-03-10 2023-08-11 南京工程学院 Magnetic saturation calculation method for magnetic suspension flywheel motor core
CN113281996B (en) * 2021-04-02 2022-07-22 江苏大学 Control method of vehicle-mounted magnetic suspension flywheel battery system
CN113128108B (en) * 2021-04-07 2023-05-26 汕头大学 Method for determining diameter of jet grouting pile based on differential evolution artificial intelligence
CN113392557B (en) * 2021-06-18 2023-08-11 南京工程学院 Method for establishing axial split-phase magnetic suspension flywheel motor suspension force model
CN113610116B (en) * 2021-07-14 2023-07-04 上海工程技术大学 Fault diagnosis method for optimizing support vector machine by self-adaptive differential evolution algorithm
CN113719283B (en) * 2021-09-07 2023-01-17 武汉理工大学 Method and device for predicting working hours of mine rock drilling equipment
CN113761660A (en) * 2021-09-10 2021-12-07 南京工程学院 Vehicle-mounted flywheel dynamic modeling method based on data driving and mechanism model fusion
CN114002963A (en) * 2021-10-29 2022-02-01 南京工程学院 Multi-working-condition suspension support system dynamic modeling method
CN117195752B (en) * 2023-11-07 2024-01-26 苏州保邦电气有限公司 Modeling method of axial split-phase hybrid excitation type magnetic levitation motor

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104283393A (en) * 2014-09-25 2015-01-14 南京工程学院 Method for optimizing structure parameter of single-winding magnetic suspension switch reluctance machine
CN109245449A (en) * 2018-11-12 2019-01-18 南京工程学院 A kind of optimum design method of axial phase magnetic levitation switch magnetic resistance fly-wheel motor

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104283393A (en) * 2014-09-25 2015-01-14 南京工程学院 Method for optimizing structure parameter of single-winding magnetic suspension switch reluctance machine
CN109245449A (en) * 2018-11-12 2019-01-18 南京工程学院 A kind of optimum design method of axial phase magnetic levitation switch magnetic resistance fly-wheel motor

Also Published As

Publication number Publication date
CN110059348A (en) 2019-07-26

Similar Documents

Publication Publication Date Title
CN110059348B (en) Axial split-phase magnetic suspension flywheel motor suspension force numerical modeling method
CN109324291B (en) Prediction method for predicting service life of proton exchange membrane fuel cell
CN109492748B (en) Method for establishing medium-and-long-term load prediction model of power system based on convolutional neural network
CN114253157B (en) Motor multi-parameter optimization method and system based on second-order sensitivity analysis
CN116842459B (en) Electric energy metering fault diagnosis method and diagnosis terminal based on small sample learning
CN113376540B (en) LSTM battery health state estimation method based on evolutionary attention mechanism
CN113449919B (en) Power consumption prediction method and system based on feature and trend perception
CN113869616A (en) Self-adaptive residential user load prediction method
CN116757057A (en) Air quality prediction method based on PSO-GA-LSTM model
CN115730635A (en) Electric vehicle load prediction method
CN109840335A (en) Based on the radial forging pit prediction optimization method for strengthening T-S fuzzy neural network
CN116345555A (en) CNN-ISCA-LSTM model-based short-term photovoltaic power generation power prediction method
CN109408896B (en) Multi-element intelligent real-time monitoring method for anaerobic sewage treatment gas production
CN113762591A (en) Short-term electric quantity prediction method and system based on GRU and multi-core SVM counterstudy
CN117312835A (en) Fuel cell performance prediction method based on mixed probability data driving model
CN116613745A (en) PSO-ELM electric vehicle charging load prediction method based on variation modal decomposition
CN110261773B (en) Aviation generator fault symptom extraction method and system
CN112014757A (en) Battery SOH estimation method integrating capacity increment analysis and genetic wavelet neural network
CN116412162A (en) Digital twin model-based magnetic suspension blower fault diagnosis method and system
CN117131667A (en) Optimization method and system for notch permanent magnet type hybrid excitation doubly salient motor
CN114266186B (en) Air conditioner energy consumption data missing value filling method, equipment and medium
CN113408160B (en) Motor parameter design method based on multi-objective optimization
CN111859574B (en) Synchronous reluctance motor rotor optimization design method for reducing torque pulsation
CN115764855A (en) Real-time adjustable capacity and available electric quantity prediction method for electric vehicle quick charging station
CN115459982A (en) Power network false data injection attack detection method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant