CN110059348A - A kind of axial phase magnetically levitated flywheel motor suspending power numerical modeling method - Google Patents
A kind of axial phase magnetically levitated flywheel motor suspending power numerical modeling method Download PDFInfo
- Publication number
- CN110059348A CN110059348A CN201910184197.1A CN201910184197A CN110059348A CN 110059348 A CN110059348 A CN 110059348A CN 201910184197 A CN201910184197 A CN 201910184197A CN 110059348 A CN110059348 A CN 110059348A
- Authority
- CN
- China
- Prior art keywords
- suspending power
- model
- learning machine
- parameter
- extreme learning
- 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.)
- Granted
Links
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)
- Theoretical Computer Science (AREA)
- Geometry (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computer Hardware Design (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Control Of Ac Motors In General (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of axial phase magnetically levitated flywheel motor suspending power numerical modeling methods, comprising: emulation and experimental design, sample collection and processing, model off-line training, model on-line optimization.One aspect of the present invention improves model based on the extreme learning machine of principal component analysis to the adaptability and robustness of Parameters variation, realizes quick, the accurate modeling of Small Sample Database, improves the accuracy and speed of levitation force model.On the other hand differential evolution algorithm is introduced to optimize network structure, so that model built had not only been able to satisfy required precision but also realtime control requirement can be reached, avoids that hidden neuron number is excessive, the huge problem of network structure, the calculating speed for improving model makes it be more suitable the modeling of such motor.
Description
Technical field
The invention discloses a kind of axial phase magnetically levitated flywheel motor suspending power numerical modeling methods, belong to magnetic suspension and fly
The technical field of turbin generator.
Background technique
Flywheel energy storage system is a kind of physics energy storage device of energy converting between mechanical, has that specific power is big, the small in size, service life
Long, the advantages that charge and discharge are fast, cleanliness without any pollution is that a kind of researching value is high, the novel energy-storing technology that has a extensive future.It is axial
Split-phase magnetically levitated flywheel motor passes through the master of itself suspending power on the basis of being sufficiently reserved switched reluctance machines high speed good characteristic
Dynamic control, further improves motor high speed performance and operational efficiency.Flywheel energy storage is introduced into, it can be achieved that system super low-power consumption
Suspension bearing and high speed, the electrical integrated operation of efficient charge and discharge, significantly reduce system loss and volume, improve suspendability,
Critical speed and power density, are flywheel energy storage suspension bearing and one of energy conversion system preferably selects.But traditional double wrap
There is complicated electromagnetism close coupling relationship, suspending power mould in winding-magnetic circuit-electromagnetic force of group axial phase magnetically levitated flywheel motor
The analysis of type with to establish difficulty big.
Current representational suspending power modeling method has: mathematics method, support vector machines method, artificial neural network method
Deng.The wherein parameter of support vector machines model in application, the selection such as punishment parameter, kernel function is more difficult, and these are joined
Number affects precision of prediction to a certain extent again;The determination of the weight coefficient of mathematics method model is vulnerable to subjective factor
It influences;Independent learning ability is strong, memory capability is strong, non-linear parallel processing capability is strong, fault-tolerant energy because having for artificial neural network
The features such as power is strong is widely paid close attention to, wherein it is Single hidden layer feedforward neural networks (single hidden that application is wide
Layer feedforward neural network, SLFN).Many scholars attempt for extreme learning machine to be used for non-linear at present
In the forecasting research of model, but there are still following main problems: first is that when the correlation between input variable is very strong, meeting
Increase unnecessary input dimension, this operation time for not only resulting in neural network increases rapidly, it is also possible to will appear due to
Certain variables influence small and operation are caused to waste on output result;Second is that ELM will first give network before training network at random
Initial weight and threshold value, and an optimal value cannot be provided, and the selection of initial parameter can have very the final output of network
Big influence.
Summary of the invention
Goal of the invention: in order to overcome the deficiencies in the prior art, it is winged that the present invention provides a kind of axial phase magnetic suspension
Turbin generator suspending power numerical modeling method, one aspect of the present invention are based on principal component analysis (Principal component
Analysis, PCA) extreme learning machine improve model to the adaptability and robustness of Parameters variation, realize small sample number
According to quick, accurate modeling, improve the accuracy and speed of levitation force model.On the other hand differential evolution algorithm is introduced
(Differential Evolution, DE) optimizes network structure, and model built is made not only to be able to satisfy required precision but also energy
Achieve the purpose that realtime control, avoid that hidden neuron number is excessive, the huge problem of network structure improves the meter of model
Speed is calculated, it is made to be more suitable the modeling of such motor.
Technical solution: to achieve the above object, the technical solution adopted by the present invention are as follows:
A kind of axial phase magnetically levitated flywheel motor suspending power numerical modeling method, includes the following steps:
Step 1, motor finite element simulation and experimental prototype model are built according to the parameter of electric machine, analyze different operating conditions and
The suspension force characteristic of motor under operating mode.
Step 2, the principal element of analyzing influence suspending power, determine extreme learning machine model outputs and inputs variable, adopts
Collect the sample data set inputted under corresponding operating condition and operating mode with output, feature is carried out to data acquired collection based on PCA algorithm
It extracts and dimensionality reduction, sample input data set and output after obtaining dimensionality reduction collects.
Step 3, using the weight and threshold parameter of differential evolution algorithm optimization extreme learning machine network, best initial weights are obtained
Number of training is less than using node in hidden layer and determines node in hidden layer as principle and on this basis with threshold parameter, choosing
Sig function is selected as excitation function and writes code, using the sample input data set after dimensionality reduction as the input of extreme learning machine model
Data set, output integrate as the output data set of extreme learning machine model, training extreme learning machine model.
Step 4, extreme learning machine model training finishes, and builds in-circuit emulation and test platform, setting and practical identical fortune
Row operating condition and operational mode, verifying training obtain the precision of extreme learning machine model.
Step 4-1, after extreme learning machine model training, the motor obtained under corresponding operating condition, operating mode is outstanding
The suspending power predicted value of buoyancy modelBuild actual emulation and control platform, setting and practical identical operating condition and operation mould
Formula obtains the suspending power output valve F under actual motion.
Step 4-2 chooses the suspending power predicted value that extreme learning machine model calculatesIt is square with actual motion output valve F
Error MSE and coefficient of determination R2As evaluation index, judge whether constructed levitation force model meets required precision.
In formula:The respectively suspending power predicted value of extreme learning machine model, FJFor under corresponding operating condition, operating mode
Real output value, L be total training sample number, R2For the coefficient of determination, SSEFor residual sum of squares (RSS), SSTFor total sum of squares.
Step 4-3, according to evaluation index mean square error MSE and coefficient of determination R2Determined: if precision is unsatisfactory for institute
The requirement of setting then changes sample, parameter, the number of hidden nodes etc., re -training extreme learning machine model.If meeting precision to want
It asks, then completes the building of suspending power numerical model.
Preferred: step 1 carries out as follows:
Step 1-1, according to the parameter of electric machine construct the axial phase magnetically levitated flywheel motor limit element artificial module and
Experimental prototype model.
Step 1-2, under the finite element simulation and experimental prototype model of building by emulation with experimental design rotor eccentricity,
The operating condition and suspension of magnetic circuit saturation and gyroscopic effect, electronic, power generation different working modes, are analyzed in different operation shapes
The suspending power of motor and rotor position angle, rotor eccentricity, exciting current and load torque relationship under the influence of state and gyroscopic effect.
Preferred: step 2 carries out as follows:
Step 2-1 is filtered out most sensitive using experiment and simulation analysis from the variable for influencing suspending power output performance
Parameter as input variable, x1,x2,…xb,…,xa, 1≤b≤a, b=1 ..., a, a are the key parameter number finally chosen
Mesh, xbFor angular position theta, X axis bias x, any variable in Y-axis bias y, electric current I, torque T and operating voltage U.
Step 2-2 chooses the constant interval of sensitive parameter.According to fly wheel system job requirement, processing technology and physics
Constraint condition, the constant interval of clear selected sensitive parameter, is arranged relevant parameter, obtains with finite element stimulation corresponding
Sample data set (x1,x2,…,xa, F), wherein (x1,x2,…,xa) be model input set, F be output collection.
Step 2-3, is standardized input data set.Assuming that there is n sample, each sample has m index, each finger
Labeled as xij, i=1,2 ..., n.J=1,2 ..., m, after standardization,Its
In:Standardized data matrix
Step 2-4 establishes the correlation matrix of normalized matrix:
Step 2-5 calculates the eigenvalue λ of RcAnd feature vector:
αc=(αc1,αc2,…αcm)T, c=1,2 ... m (2)
Step 2-6 seeks the principal component Z of input data setc, chosen according to contribution rate of accumulative total β (k) and retain ingredient Zk, 1≤k
≤ c, so that it is determined that k value, it is (x that the data met the requirements, which are selected as sample input data set,1,x2,…,xk)。
It is preferred: with angular position theta, X-axis radial disbalance x, Y-axis radial disbalance y, electric current I, torque T and work in step 2-2
Voltage U is running parameter, and analysis, which is obtained, influences most sensitive key factor on suspending power performance, so that it is determined that extreme learning machine mould
The main input variable of type.
It is preferred: X-axis radial disbalance x, Y-axis radial disbalance y, suspending windings electric current I, rotor-position are chosen in step 2-2
Angle θ, torque T are responsive parameter, and corresponding suspending power numerical value F is as corresponding output.
Preferred: step 3 carries out as follows:
Step 3-1 chooses respective trained rule and excitation function, draws for different operating condition and operating mode
Enter the weight and threshold value of differential evolution algorithm optimized learning algorithm.
Step 3-2, initialization of population: random selection individual P, and forming size is NPPopulation, and choose maximum evolution
Algebra is gmax。
PA={ PA1, PA2..., PAB} (5)
In formula: A=1,2 ... NP, B=1,2 ... D, D are the number of arguments.
In formula:The upper and lower limit of respectively the A independent variable.
Mutation operation: step 3-3 randomly chooses three individuals from populationAnd it is not mutually equal.
In formula: νr,g+1For the vector of neotectonics.F is zoom factor.G is algebra.
A kind of Mutation Strategy is only indicated in formula (7), different requirements can set different strategy.
Step 3-4, crossover operation: new individual vr,g+1With xAThe individual u that discrete crossover is updatedA。
In formula: crossover probability CR ∈ [0,1].Rand (A) is the random integers between [1, D] that the A vector generates.
Step 3-5, selection operation: the Cenozoic compared with the adaptive value f () of previous generation's variable, is worth small person and enters the next generation, no
Then retain.
Step 3-6 repeats the above mutation, intersection and selection course, until maximum number of iterations gmax, export and obtain the limit
The best initial weights and threshold parameter of learning machine network.
Step 3-7 herein on the basis of weight and threshold parameter obtained by optimization, is less than number of training with node in hidden layer
The number of hidden nodes is chosen for principle, kernel function is selected, with (x1,x2,…,xk) it is input data set, F is output data set, training
Obtain extreme learning machine model.
It is preferred: according to fly wheel system job requirement, processing technology and physical constraint condition, clear selected sensitive parameter
Optimization section, be arranged relevant parameter, with finite element stimulation obtain corresponding sample data set (x, y, I, θ, T,
F), training extreme learning machine model, wherein (x, y, I, θ, T) is the input set of extreme learning machine model training, F is output collection.
It is preferred: based on acquired training dataset (x, y, I, θ, T), to carry out data set with PCA dimension-reduction algorithm
Feature extraction and dimensionality reduction, the input data set after dimensionality reduction are optimised for the training that (x, y, I, θ) carries out model.
It is preferred: to select Sigmoid function, Sine function or Hardlim function as kernel function.
It is preferred: in step 2-6, to choose the tie element of contribution rate of accumulative total β (k) on 85% as reservation ingredient
Zk。
The present invention compared with prior art, has the advantages that
1) model is improved using extreme learning machine to the adaptability and robustness of Parameters variation, realizes Small Sample Database
Quick, accurate modeling, improve the accuracy and speed of levitation force model.
2) it introduces PCA algorithm and feature extraction and dimensionality reduction is carried out to sample data, to avoid due to low memory or meter
The problems such as memory occurred when calculation overflows, also accelerates the speed of machine learning, improves whole operation efficiency.
3) network structure is optimized with differential evolution algorithm, model built is made not only to be able to satisfy required precision but also can reach
To the purpose of realtime control, avoid that hidden neuron number is excessive, and the huge problem of network structure makes it be more suitable such
Motor modeling.
Detailed description of the invention
Fig. 1 show mentioned axial phase magnetically levitated flywheel motor design parameter and structure chart, and wherein Fig. 1 (I) indicates electricity
Machine design parameter, Fig. 1 (II) indicate electric machine structure.
Fig. 2 show mentioned axial phase magnetically levitated flywheel motor suspending power numerical modeling method flow chart.
Fig. 3 show the relational graph that suspending power changes with suspending windings electric current I.
Fig. 4 show the relational graph that suspending power changes with rotor position angle θ.
Fig. 5 show the relational graph that suspending power changes with rotor X-axis radial disbalance x.
Fig. 6 show the relational graph that suspending power changes with rotor Y-axis radial disbalance y.
Fig. 7 show the relational graph that suspending power changes with torque winding current.
Fig. 8 show the initial sample data figure that extreme learning machine model is practiced.
Fig. 9 show the contribution rate of each variable ingredient under PCA dimensionality reduction.
Figure 10 show the sample data figure of extreme learning machine model white silk after dimensionality reduction.
Figure 11 show suspending power predicted value and actual value comparative result figure before optimization.
Figure 12 show suspending power predicted value and actual value comparative result figure after optimization.
Wherein, 1 is flywheel, and 2 be outer rotor, and 3 be suspension pole, and 4 be magnetic shield, and 5 be torque pole, and 6 be suspending windings, and 7 are
Torque winding, 8 be permanent magnet, and 9 be rotor core, and 10 be stator core.
Specific embodiment
In the following with reference to the drawings and specific embodiments, the present invention is furture elucidated, it should be understood that these examples are merely to illustrate this
It invents rather than limits the scope of the invention, after the present invention has been read, those skilled in the art are to of the invention various
The modification of equivalent form falls within the application range as defined in the appended claims.
A kind of axial phase magnetically levitated flywheel motor suspending power numerical modeling method, as shown in Figure 1, 2, including walks as follows
It is rapid:
Step 1, motor finite element simulation and experimental prototype model are built, is analyzed electric under different operating conditions and operating mode
The suspension force characteristic of machine.
Step 1-1 optimizes to obtain according to the mentioned optimum design method of patent of invention application No. is 201811339664.5
The configuration of motor optimized parameter, constructs the axial phase magnetically levitated flywheel based on the parameter after optimization in finite element emulation software
The threedimensional FEM computation model of motor, the experimental prototype model of one identical parameters size of simultaneous processing, motor are specific
Shown in parameter and structure such as Fig. 1 (I), (II).
Step 1-2 by the operating conditions such as emulation and experimental design rotor eccentricity, magnetic circuit saturation and gyroscopic effect and hangs
It is floating, electronic, power generation etc. different working modes, analysis under the influence of different operating statuses and gyroscopic effect the suspending power of motor with
The relationship of the parameters such as rotor position angle, rotor eccentricity, exciting current and load torque.
Step 2, the principal element of analyzing influence suspending power, determine extreme learning machine model outputs and inputs variable, adopts
Collect the sample data set inputted under corresponding operating condition and operating mode with output, feature is carried out to data acquired collection based on PCA algorithm
Extraction and dimensionality reduction.
Step 2-1 is filtered out most sensitive using experiment and simulation analysis from the variable for influencing suspending power output performance
Parameter as input variable x1,x2,…xb,…,xa, 1≤b≤a, b=1 ..., a, a are the key parameter number finally chosen
Mesh, xbFor angular position theta, X axis bias x, any one variable in Y-axis bias y, electric current I, torque T and operating voltage U.
Step 2-2 chooses the constant interval of sensitive parameter.Consider fly wheel system job requirement, processing technology and physics
Constraint condition, the constant interval of clear selected sensitive parameter, is arranged relevant parameter, obtains with finite element stimulation corresponding
Sample data set (x1,x2,…,xa, F), wherein (x1,x2,…,xa) be model input set, F be output collection.
Step 2-3, is standardized input data set.Assuming that there is n sample, each sample has m index, each finger
Labeled as xij(i=1,2 ..., n, j=1,2 ..., m).After standardization,Its
In:Standardized data matrix
Step 2-4 establishes the correlation matrix of normalized matrix:
Step 2-5 calculates the eigenvalue λ of RcAnd feature vector:
αc=(αc1,αc2,…αcm)T, c=1,2 ... m (2)
Step 2-6 seeks the principal component Z of input data setc, choose contribution rate of accumulative total β (k) correspondence on 85% at
It is allocated as to retain ingredient Zk, so that it is determined that k value, it is (x that the data met the requirements, which are selected as sample input data set,1,x2,…,
xk)。
Step 3, it using the weight and threshold parameter of differential evolution algorithm optimization extreme learning machine, builds and trains the limit
Habit machine model.
Step 3-1 chooses suitable training rule and excitation function, draws for different operating condition and operating mode
The weight and threshold value for entering differential evolution algorithm optimized learning algorithm, write the program code of extreme learning machine.
Step 3-2, initialization of population: random selection individual P (meeting constraint condition), and forming size is NPPopulation,
And choosing maximum evolutionary generation is gmax。
PA={ PA1, PA2..., PAB} (5)
In formula: A=1,2 ... NP, B=1,2 ... D, D are the number of arguments.
In formula:The upper and lower limit of respectively the A independent variable.
Mutation operation: step 3-3 randomly chooses three individuals from populationAnd it is not mutually equal.
In formula: vr,g+1For the vector of neotectonics.F is zoom factor.G is algebra.
A kind of Mutation Strategy is only indicated in formula (7), different requirements can set different strategy.
Step 3-4, crossover operation: new individual vr,g+1(mutation operation) and xAThe individual that (parent) discrete crossover is updated
ui。
In formula: crossover probability CR ∈ [0,1].Rand (A) is the random integers between [1, D] that the A vector generates.
Step 3-5, selection operation: the Cenozoic compared with the adaptive value of previous generation's variable, is worth small person and enters the next generation, otherwise protect
It stays.
Step 3-6 repeats the above mutation, intersection and selection course, until maximum number of iterations gmax, export and obtain the limit
The best initial weights and threshold parameter of learning machine network.
Step 3-7 herein on the basis of weight and threshold parameter obtained by optimization, is less than number of training with node in hidden layer
The number of hidden nodes is chosen for principle, selects Sigmoid, Sine or Hardlim as kernel function, with (x1,x2,…,xk) it is input
Data set, F are output data set, and training obtains the levitation force model of motor.
Step 4, model training finishes, and builds in-circuit emulation and test platform, setting and practical identical operating condition and fortune
Row mode verifies model accuracy.
Step 4-1 after model training, obtains the motor levitation force model under corresponding operating condition, operating mode
Suspending power predicted valueActual emulation and control platform, setting and practical identical operating condition and operational mode are built, is obtained real
Suspending power output valve F under the operation of border.
Step 4-2 chooses the suspending power predicted value that extreme learning machine model calculatesIt is square with actual motion output valve F
Error MSE and coefficient of determination R2As evaluation index, judge whether constructed levitation force model meets required precision.
In formula:The respectively suspending power predicted value of extreme learning machine model, FJFor under corresponding operating condition, operating mode
Real output value, L be total training sample number, R2For the coefficient of determination, wherein SSEFor residual sum of squares (RSS), SSTIt is total square
With.
Step 4-3, according to evaluation index mean square error MSE and coefficient of determination R2Determined: if precision is unsatisfactory for institute
The requirement of setting then changes sample, parameter, the number of hidden nodes etc., re -training extreme learning machine model.If meeting precision to want
It asks, then completes the building of suspending power numerical model.
Emulation
It 1, is CN201610864124.3 with Chinese Patent Application No., entitled " a kind of axial phase inner stator permanent magnetism is inclined
Set magnetic levitation switch magnetic resistance fly-wheel motor " document disclosed in for motor, construct motor three-dimensional finite element model and online
Emulation platform specifically analyzes its suspension force characteristic under radial disbalance operating condition, suspension operating mode.Constructed
Shown in the parameter of electric machine and specific structure such as Fig. 1 (I), (II).
It 2, is variation ginseng with angular position theta, X-axis radial disbalance x, Y-axis radial disbalance y, electric current I, torque T and operating voltage U
Number, analysis, which is obtained, influences most sensitive key factor on suspending power performance, so that it is determined that the main input of extreme learning machine model
Variable, Fig. 3-Fig. 7 show motor suspending power with the change curve of relevant parameter.By Fig. 3, Fig. 5, Fig. 6 as it can be seen that suspending power is with electricity
Flow I, X-axis radial disbalance x, the increase of Y-axis radial disbalance y and be increased monotonically, variation is obvious.From fig. 4, it can be seen that suspending power is by rotor
The variation of angular position theta and change, first subtract increase afterwards therewith.As seen from Figure 7, when suspending windings electric current is 1.5A, torque winding current
When logical 1-4A, suspending power changes with the variation of torque torque current, it is possible thereby to learn suspending power for motor torque T for
The output of suspending power has certain influence.The voltage for considering motor both ends exists linear with the electric current for being passed through suspension and torque winding
Relationship, therefore be to reduce to calculate data dimension, accelerate machine learning efficiency, chooses X-axis radial disbalance x, Y-axis radial disbalance y, hangs
Floating winding current I, rotor position angle θ, torque T are responsive parameter, and corresponding suspending power numerical value F is as corresponding output.
3, consider fly wheel system job requirement, processing technology and physical constraint condition, define the excellent of selected sensitive parameter
Change section, relevant parameter is set, is obtained corresponding sample data set (x, y, I, θ, T, F) with finite element stimulation, instructs
Practice extreme learning machine model, wherein (x, y, I, θ, T) is the input set of extreme learning machine model training, F is output collection, initial number
It is as shown in Figure 8 according to collection.
4, based on acquired training dataset (x, y, I, θ, T, F), the feature of data set is carried out with PCA dimension-reduction algorithm
Extract and dimensionality reduction, Fig. 9 be corresponding data characteristic value and accumulation contribution rate, take contribution rate of accumulative total greater than 85% it is each it is main at
Point, that is, take X-axis radial disbalance x, Y-axis radial disbalance y, suspending windings electric current I and rotor position angle θ as reservation principal component.Cause
This, the input data set after dimensionality reduction is optimised for the training that (x, y, I, θ) carries out model.
5, after determining training input data set, Topological expansion is carried out using differential evolution algorithm, obtains best initial weights
Number of training is less than using node in hidden layer and determines node in hidden layer as principle and on this basis with threshold parameter, choosing
Sig function is selected as excitation function and writes code, with (x, y, I, the θ) after dimensionality reduction for input data set, F is output data set,
The levitation force model of training building motor.The sample data that extreme learning machine model is practiced after dimensionality reduction is as shown in Figure 10.
6, by the resulting suspending power predicted value of extreme learning machineWith the reality output under identical operating condition and operating mode
F carries out precision test.By Figure 11 and Figure 12 as it can be seen that the resulting suspending power numerical model error of not optimized ELM model training
It is larger, mean square error MSE=73.3454, coefficient of determination R2=0.96956.And pass through PCA of the present invention and differential evolution
After algorithm optimization, mean square error MSE=35.9348 between gained suspending power numerical model predicted value and actual value determines system
Number R2=0.990788, meet the required precision of constructed levitation force model, and accuracy and accuracy are greatly improved,
Demonstrate the superiority of the method for the invention.
Above-described embodiment is only 12/12 structure axial phase magnetic levitation switch magnetic resistance fly-wheel motor in radial disbalance operation work
Suspending power numerical modeling under condition, suspension operating mode, remaining structure magnetic suspension switched reluctance motor is in other different operations
It can use technical solution of the present invention under operating condition and operating mode and carry out suspending power numerical modeling.
Axial phase magnetically levitated flywheel motor suspending power numerical modeling method key step of the present invention includes: emulation and experiment
Design, sample collection and processing, model off-line training, model on-line optimization.One aspect of the present invention is based on principal component analysis
The extreme learning machine of (Principal component analysis, PCA) come improve model to the adaptability of Parameters variation and
Robustness realizes quick, the accurate modeling of Small Sample Database, improves the accuracy and speed of levitation force model.On the other hand
It introduces differential evolution algorithm (Differential Evolution, DE) to optimize network structure, makes model built can
Realtime control requirement can be reached again by meeting required precision, avoid that hidden neuron number is excessive, and network structure is huge to ask
Topic, improves the calculating speed of model, it is made to be more suitable the modeling of such motor.
The above is only a preferred embodiment of the present invention, it should be pointed out that: for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered
It is considered as protection scope of the present invention.
Claims (10)
1. a kind of axial phase magnetically levitated flywheel motor suspending power numerical modeling method, which comprises the steps of:
Step 1, motor finite element simulation and experimental prototype model are built according to the parameter of electric machine, analyzes different operating conditions and work
The suspension force characteristic of motor under mode;
Step 2, the principal element of analyzing influence suspending power determines the variable that outputs and inputs of extreme learning machine model, acquires phase
The sample data set inputted under operating condition and operating mode with output is answered, feature extraction is carried out to data acquired collection based on PCA algorithm
And dimensionality reduction, sample input data set and output collection after obtaining dimensionality reduction;
Step 3, using the weight and threshold parameter of differential evolution algorithm optimization extreme learning machine network, best initial weights and threshold are obtained
Value parameter, and on this basis, number of training is less than using node in hidden layer and determines node in hidden layer as principle, is selected
The functions such as Sigmod write code as excitation function, using the sample input data set after dimensionality reduction as the defeated of extreme learning machine model
Enter data set, output integrates as the output data set of extreme learning machine model, training extreme learning machine model;
Step 4, extreme learning machine model training finishes, and builds in-circuit emulation and test platform, setting and practical identical operation work
Condition and operational mode, verifying training obtain the precision of extreme learning machine model;
Step 4-1 after extreme learning machine model training, obtains the motor suspending power under corresponding operating condition, operating mode
The suspending power predicted value of modelActual emulation and control platform, setting and practical identical operating condition and operational mode are built,
Obtain the suspending power output valve F under actual motion;
Step 4-2 chooses the suspending power predicted value that extreme learning machine model calculatesWith the mean square error of actual motion output valve F
MSE and coefficient of determination R2As evaluation index, judge whether constructed levitation force model meets required precision;
In formula:The respectively suspending power predicted value of extreme learning machine model, FJFor the reality under corresponding operating condition, operating mode
Border output valve, L are total training sample number, R2For the coefficient of determination, SSEFor residual sum of squares (RSS), SSTFor total sum of squares;
Step 4-3, according to evaluation index mean square error MSE and coefficient of determination R2Determined: if precision be unsatisfactory for it is set
It is required that then change sample, parameter, the number of hidden nodes etc., re -training extreme learning machine model;If meeting required precision,
Complete the building of suspending power numerical model.
2. axial phase magnetically levitated flywheel motor suspending power numerical modeling method according to claim 1, it is characterised in that: step
Rapid 1 carries out as follows:
Step 1-1 constructs the three of axial phase magnetically levitated flywheel motor according to parameter of electric machine building in finite element emulation software
Tie up limit element artificial module, the experimental prototype model of one identical parameters size of simultaneous processing;
Step 1-2, under the limit element artificial module of building and experimental prototype model by emulation with experimental design rotor eccentricity,
The operating condition and suspension of magnetic circuit saturation and gyroscopic effect, electronic, power generation different working modes, are analyzed in different operation shapes
The suspending power of motor and rotor position angle, rotor eccentricity, exciting current and load torque relationship under the influence of state and gyroscopic effect.
3. axial phase magnetically levitated flywheel motor suspending power numerical modeling method according to claim 2, it is characterised in that: step
Rapid 2 carry out as follows:
Step 2-1 filters out most sensitive ginseng from the variable for influencing suspending power output performance using experiment and simulation analysis
Number is used as input variable x1,x2,…xb,…,xa, 1≤b≤a, b=1 ..., a, a are the key parameter number finally chosen, xbFor
Angular position theta, X axis bias x, any variable in Y-axis bias y, electric current I, torque T and operating voltage U;
Step 2-2 chooses the constant interval of sensitive parameter;According to fly wheel system job requirement, processing technology and physical constraint
Condition, the constant interval of clear selected sensitive parameter, is arranged relevant parameter, corresponding sample is obtained with finite element stimulation
Notebook data collection (x1,x2,…,xa, F), wherein (x1,x2,…,xa) be model input set, F be output collection;
Step 2-3, is standardized input data set;Assuming that there is n sample, each sample has m index, each digit synbol
For xij, i=1,2 ..., n;J=1,2 ..., m, after standardization,Wherein:Standardized data matrix
Step 2-4 establishes the correlation matrix of normalized matrix:
Step 2-5 calculates the eigenvalue λ of RcAnd feature vector:
αc=(αc1,αc2,…αcm)T, c=1,2 ... m (2)
Step 2-6 seeks the principal component Z of input data setc, chosen according to contribution rate of accumulative total β (k) and retain ingredient Zk, 1≤k≤c,
So that it is determined that the data met the requirements are selected as sample input data set (x by k value1,x2,…,xk);
4. axial phase magnetically levitated flywheel motor suspending power numerical modeling method according to claim 3, it is characterised in that: step
Using angular position theta, X-axis radial disbalance x, Y-axis radial disbalance y, electric current I, torque T and operating voltage U as running parameter in rapid 2-2,
Analysis, which is obtained, influences most sensitive key factor on suspending power performance, so that it is determined that the main input of extreme learning machine model becomes
Amount.
5. axial phase magnetically levitated flywheel motor suspending power numerical modeling method according to claim 4, it is characterised in that: step
It is sensibility that X-axis radial disbalance x, Y-axis radial disbalance y, suspending windings electric current I, rotor position angle θ, torque T are chosen in rapid 2-2
Parameter, corresponding suspending power numerical value F is as corresponding output.
6. axial phase magnetically levitated flywheel motor suspending power numerical modeling method according to claim 5, it is characterised in that: step
Rapid 3 carry out as follows:
Step 3-1 chooses corresponding trained rule and excitation function for different operating condition and operating mode, and it is poor to introduce
The weight and threshold value for dividing evolution algorithm optimized learning algorithm, write the program code of extreme learning machine;
Step 3-2, initialization of population: random selection individual P, and forming size is NPPopulation, determine maximum evolutionary generation be
gmax;
PA={ PA1, PA2..., PAB} (5)
In formula: A=1,2 ... NP;B=1,2 ... D, D are the number of arguments;
In formula:The upper and lower limit of respectively the A independent variable;
Mutation operation: step 3-3 randomly chooses three individuals from populationAnd it is not mutually equal;
In formula: νr,g+1For the vector of neotectonics;F is zoom factor;G is algebra;
Step 3-4, crossover operation: mutation operation νr,g+1With parent xAThe individual u that discrete crossover is updatedA;
In formula: crossover probability CR ∈ [0,1];Rand (A) is the random integers between [1, D] that the A vector generates;
Step 3-5, selection operation: the Cenozoic compared with the adaptive value f () of previous generation's variable, is worth small person and enters the next generation, otherwise
Retain;
Step 3-6 repeats the above mutation, intersection and selection course, until maximum number of iterations gmax, export and obtain limit study
The best initial weights and threshold parameter of machine network;
Step 3-7, herein on the basis of weight and threshold parameter obtained by optimization, being less than number of training with node in hidden layer is original
The number of hidden nodes is then chosen, kernel function is selected, with (x1,x2,…,xk) it is input data set, F is output data set, and training obtains
Extreme learning machine model.
7. axial phase magnetically levitated flywheel motor suspending power numerical modeling method according to claim 6, it is characterised in that: root
According to fly wheel system job requirement, processing technology and physical constraint condition, phase is arranged in the optimization section of clear selected sensitive parameter
Parameter is answered, is obtained corresponding sample data set (x, y, I, θ, T, F) with finite element stimulation, training extreme learning machine mould
Type, wherein (x, y, I, θ, T) is the input set of extreme learning machine model training, F is output collection.
8. axial phase magnetically levitated flywheel motor suspending power numerical modeling method according to claim 7, it is characterised in that: base
In acquired training dataset (x, y, I, θ, T), feature extraction and the dimensionality reduction of data set, drop are carried out with PCA dimension-reduction algorithm
Input data set after dimension is optimised for the training that (x, y, I, θ) carries out model.
9. axial phase magnetically levitated flywheel motor suspending power numerical modeling method according to claim 8, it is characterised in that: choosing
Sigmoid function, Sine function or Hardlim function are selected as kernel function.
10. axial phase magnetically levitated flywheel motor suspending power numerical modeling method according to claim 9, it is characterised in that:
In step 2-6, the tie element of contribution rate of accumulative total β (k) on 85% is chosen as reservation ingredient Zk。
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 true CN110059348A (en) | 2019-07-26 |
CN110059348B 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) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110661390A (en) * | 2019-09-24 | 2020-01-07 | 江苏大学 | Accurate modeling method for suspension force of 12/14 pole magnetic suspension switched reluctance motor |
CN111245318A (en) * | 2020-01-18 | 2020-06-05 | 浙江大学 | Radial force accurate compensation decoupling control method for bearingless permanent magnet synchronous motor |
CN113011064A (en) * | 2021-03-10 | 2021-06-22 | 南京工程学院 | Magnetic saturation calculation method for magnetic suspension flywheel motor iron core |
CN113128108A (en) * | 2021-04-07 | 2021-07-16 | 汕头大学 | Method for determining diameter of jet grouting pile based on differential evolution artificial intelligence |
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 |
CN113281996A (en) * | 2021-04-02 | 2021-08-20 | 江苏大学 | Control method of vehicle-mounted magnetic suspension flywheel battery system |
CN113392557A (en) * | 2021-06-18 | 2021-09-14 | 南京工程学院 | Method for establishing suspension force model of axial split-phase magnetic suspension flywheel motor |
CN113610116A (en) * | 2021-07-14 | 2021-11-05 | 上海工程技术大学 | Fault diagnosis method for adaptive differential evolution algorithm optimized support vector machine |
CN113719283A (en) * | 2021-09-07 | 2021-11-30 | 武汉理工大学 | Method and device for predicting working hours of mine rock drilling equipment |
JP2023040995A (en) * | 2021-09-10 | 2023-03-23 | 南京工程学院 | On-vehicle flywheel dynamic modeling method based on data driven and mechanism model combination |
CN117195752A (en) * | 2023-11-07 | 2023-12-08 | 苏州保邦电气有限公司 | Modeling method of axial split-phase hybrid excitation type magnetic levitation motor |
Citations (2)
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 |
-
2019
- 2019-03-12 CN CN201910184197.1A patent/CN110059348B/en active Active
Patent Citations (2)
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 |
Cited By (19)
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 |
CN110661390A (en) * | 2019-09-24 | 2020-01-07 | 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 |
CN111245318A (en) * | 2020-01-18 | 2020-06-05 | 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 |
CN113011064A (en) * | 2021-03-10 | 2021-06-22 | 南京工程学院 | Magnetic saturation calculation method for magnetic suspension flywheel motor iron core |
CN113011064B (en) * | 2021-03-10 | 2023-08-11 | 南京工程学院 | Magnetic saturation calculation method for magnetic suspension flywheel motor core |
CN113281996A (en) * | 2021-04-02 | 2021-08-20 | 江苏大学 | Control method of vehicle-mounted magnetic suspension flywheel battery system |
CN113128108A (en) * | 2021-04-07 | 2021-07-16 | 汕头大学 | Method for determining diameter of jet grouting pile based on differential evolution artificial intelligence |
CN113392557A (en) * | 2021-06-18 | 2021-09-14 | 南京工程学院 | Method for establishing suspension force model of axial split-phase magnetic suspension flywheel motor |
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 |
CN113610116A (en) * | 2021-07-14 | 2021-11-05 | 上海工程技术大学 | Fault diagnosis method for adaptive differential evolution algorithm optimized support vector machine |
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 |
CN113719283A (en) * | 2021-09-07 | 2021-11-30 | 武汉理工大学 | Method and device for predicting working hours of mine rock drilling equipment |
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 |
JP2023040995A (en) * | 2021-09-10 | 2023-03-23 | 南京工程学院 | On-vehicle flywheel dynamic modeling method based on data driven and mechanism model combination |
JP7311189B2 (en) | 2021-09-10 | 2023-07-19 | 南京工程学院 | Automotive Flywheel Dynamic Modeling Method Based on Combination of Data-Driven and Mechanism Models |
CN117195752A (en) * | 2023-11-07 | 2023-12-08 | 苏州保邦电气有限公司 | Modeling method of axial split-phase hybrid excitation type magnetic levitation motor |
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 |
Also Published As
Publication number | Publication date |
---|---|
CN110059348B (en) | 2023-04-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110059348A (en) | A kind of axial phase magnetically levitated flywheel motor suspending power numerical modeling method | |
CN107103332A (en) | A kind of Method Using Relevance Vector Machine sorting technique towards large-scale dataset | |
CN107423839A (en) | A kind of method of the intelligent building microgrid load prediction based on deep learning | |
Grimaccia et al. | Design of tubular permanent magnet generators for vehicle energy harvesting by means of social network optimization | |
Li et al. | A nonlinear flux linkage model for bearingless induction motor based on GWO-LSSVM | |
CN104899431A (en) | LSSVM fluctuating wind speed prediction method based on integration of ant colony and particle swarm | |
CN109255726A (en) | A kind of ultra-short term wind power prediction method of Hybrid Intelligent Technology | |
CN105260786A (en) | Comprehensive optimization method of simulation credibility evaluation model of electric propulsion system | |
CN110535149A (en) | A kind of area Gong Biantai electric load three-phase imbalance prediction technique | |
Tian et al. | Daily power demand prediction for buildings at a large scale using a hybrid of physics-based model and generative adversarial network | |
CN109670655A (en) | A kind of electric system multi-objective particle swarm optimization dispatching method | |
Zhang et al. | An efficient multi-objective bayesian optimization approach for the automated analytical design of switched reluctance machines | |
Li et al. | Deep echo state networks with multi-span features for nonlinear time series prediction | |
CN117436595A (en) | New energy automobile energy consumption prediction method, device, equipment and storage medium | |
Chen et al. | A Spark-based Ant Lion algorithm for parameters optimization of random forest in credit classification | |
CN110635474B (en) | Power grid dynamic trajectory trend prediction method based on long-term and short-term memory network | |
Cuevas et al. | Recent metaheuristics algorithms for parameter identification | |
CN116842459A (en) | Electric energy metering fault diagnosis method and diagnosis terminal based on small sample learning | |
CN114372634B (en) | Short-term wind power prediction method and system based on neural network imbalance optimization | |
Khosla et al. | A comparison of computational efforts between particle swarm optimization and genetic algorithm for identification of fuzzy models | |
Zou et al. | An optimization control of thermal power combustion based on reinforcement learning | |
CN116164039A (en) | Construction method of radial six-pole hybrid magnetic bearing displacement-free sensor control system | |
Elwer et al. | Intelligent fuzzy controller using particle swarm optimization for control of permanent magnet synchronous motor for electric vehicle | |
Xu et al. | Electromagnetic buffer optimization based on Nash game | |
Hu et al. | Magnetic force prediction of hybrid magnet with Halbach array using generalized regression neural network optimized by a modified Aquila Optimizer |
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 |