CN104767449B  Selfbearings motors RBF neural adaptive inversion decoupling control and parameter identification method  Google Patents
Selfbearings motors RBF neural adaptive inversion decoupling control and parameter identification method Download PDFInfo
 Publication number
 CN104767449B CN104767449B CN201510092881.9A CN201510092881A CN104767449B CN 104767449 B CN104767449 B CN 104767449B CN 201510092881 A CN201510092881 A CN 201510092881A CN 104767449 B CN104767449 B CN 104767449B
 Authority
 CN
 China
 Prior art keywords
 mrow
 msub
 self
 rbf neural
 control
 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.)
 Expired  Fee Related
Links
 230000001537 neural Effects 0.000 title claims abstract description 78
 230000003044 adaptive Effects 0.000 title claims abstract description 35
 239000002131 composite material Substances 0.000 claims abstract description 26
 238000006073 displacement reaction Methods 0.000 claims abstract description 13
 230000004907 flux Effects 0.000 claims description 27
 238000004804 winding Methods 0.000 claims description 19
 238000006243 chemical reactions Methods 0.000 claims description 11
 239000000725 suspensions Substances 0.000 claims description 10
 238000000034 methods Methods 0.000 claims description 6
 230000001131 transforming Effects 0.000 claims description 6
 239000000203 mixtures Substances 0.000 claims description 4
 239000010410 layers Substances 0.000 claims 2
 210000004218 Nerve Net Anatomy 0.000 claims 1
 239000003570 air Substances 0.000 claims 1
 239000007789 gases Substances 0.000 claims 1
 230000035699 permeability Effects 0.000 claims 1
 230000002411 adverse Effects 0.000 abstract description 4
 230000005611 electricity Effects 0.000 description 7
 238000010586 diagram Methods 0.000 description 5
 230000001264 neutralization Effects 0.000 description 3
 238000010276 construction Methods 0.000 description 2
 230000001276 controlling effect Effects 0.000 description 2
 230000000875 corresponding Effects 0.000 description 2
 230000001808 coupling Effects 0.000 description 2
 238000010168 coupling process Methods 0.000 description 2
 238000005859 coupling reaction Methods 0.000 description 2
 238000005516 engineering process Methods 0.000 description 2
 230000001105 regulatory Effects 0.000 description 2
 241000208340 Araliaceae Species 0.000 description 1
 241000406668 Loxodonta cyclotis Species 0.000 description 1
 235000003140 Panax quinquefolius Nutrition 0.000 description 1
 230000003321 amplification Effects 0.000 description 1
 238000004364 calculation method Methods 0.000 description 1
 150000001875 compounds Chemical class 0.000 description 1
 230000003750 conditioning Effects 0.000 description 1
 238000001514 detection method Methods 0.000 description 1
 230000000694 effects Effects 0.000 description 1
 238000005265 energy consumption Methods 0.000 description 1
 238000004146 energy storage Methods 0.000 description 1
 238000007716 flux method Methods 0.000 description 1
 235000005035 ginseng Nutrition 0.000 description 1
 235000008434 ginseng Nutrition 0.000 description 1
 239000000463 material Substances 0.000 description 1
 230000004048 modification Effects 0.000 description 1
 238000006011 modification reaction Methods 0.000 description 1
 238000003199 nucleic acid amplification method Methods 0.000 description 1
 238000005457 optimization Methods 0.000 description 1
 238000007789 sealing Methods 0.000 description 1
 230000003068 static Effects 0.000 description 1
 238000005728 strengthening Methods 0.000 description 1
 230000001360 synchronised Effects 0.000 description 1
 230000017105 transposition Effects 0.000 description 1
Abstract
The invention discloses selfbearings motors RBF neural adaptive inversion decoupling control and parameter identification method, and composite controlled object is formed as an entirety by SVPWM modules, voltage source inverter, selfbearings motors and its load；The Adverse control and parameter identification to complex controll object are realized using two radial basis function neural networks；Adaptive inverse control device is formed by learning with RBF neural, before controller is serially connected in composite controlled object, inverse controller input is feedback signal and the error of Setting signal, thus forms closedloop control；Again parameter adaptive identifier is formed by learning with a RBF neural, composite controlled object output quantity speed and displacement are recognized, realize and controlled without speed and without gap sensors, and estimation signal is helped into online study by learning algorithm, realize and selfbearings motors Dynamic Nonlinear Decoupling is controlled.It controls speed soon and identification precision is higher, and control system is excellent.
Description
Technical field
The present invention is the selfbearings motors RBF neural Adaptive inverse control and ginseng of a kind of multivariable nonlinearity
Number discrimination method, suitable for the high performance control of selfbearings motors.Selfbearings motors inherit magnetic bearing electricity
Machine advantage, have the characteristics that without friction, without wear, be not required to lubricate and seal, sterile, pollutionfree, long lifespan, be highly suitable for noting
Enter highspeed precision digital control lathe, high pressure sealing pump, specialized robot, high speed gyro, satellite flywheel, highspeed aircraft and control dress
Put, the hightechnology field of the highspeed driving such as supercentrifuge, highspeed flywheel energy storage, application prospect is extensive, belongs to Electric Drive
The technical field of control device.
Background technology
There is complicated electromagnetic relationship, therefore it is a kind of multivariable, nonlinear, strong coupling inside selfbearings motors
The controlled device of conjunction, to realize its radial displacement, rotating speed are accurately controlled it is extremely difficult.To realize to the asynchronous electricity of bearingfree
The stable suspersion of motivation rotor and follow given rotating speed to run, must just carry out decoupling control to the torque force and suspending power of motor
System.
But the control strategy of dynamic Decoupling Control of Load Torque is always the difficult point for realizing selfbearings motors steady operation.
Common control has Airgapflux orientated and orientation on rotor flux, and experiment proves that both approaches can be to the asynchronous electricity of bearingfree
Motivation realizes relatively stable control.Although Airgapflux orientated control method can be realized between electromagnetic torque and radial suspension force
Decoupling control, but this algorithm is had a great influence by the parameter of electric machine (such as rotor resistance, rotor leakage inductance), and it is steady turn to exist
Square, the scope of application are limited；Orientation on rotor flux method, can accomplish the decoupling between torque current and exciting current, but only
There is rotor flux to reach and stablize and could be realized when keeping constant electromagnetic torque and rotor flux decoupling, belong to steady state decoupling not
It can realize dynamic decoupling.BP neural network controls applied to selfbearings motors and obtains preferable control effect, but BP
Neutral net in terms of function approximation there are convergence rate it is slow, be easily absorbed in local minimum the shortcomings of, and in theory with biological context
It is not consistent.
To further improve the dynamic duty performance of selfbearings motors, it is necessary to consider selfbearings motors
Dynamic decoupling and multivariable coordinated control are combined, and then Structure of need is compacter, the asynchronous electricity of the bearingfree of more excellent performance
Motivation decoupling controller.
The country is existing to obtain related application：1) number of patent application CN20061038711.3, it is entitled：Bearingless AC
The control method of asynchronous motor neural network inverse decoupling controller, the invention patent are set for bearingless AC asynchronous motor
Count neural network inverse decoupling controller；2) number of patent application CN200510038099.5, it is entitled：Bearingfree switch magneticresistance is electronic
Machine radial neural network reversed decoupling controller and building method, the invention are radially neural for magnetic suspension switched reluctance motor design
Network inverse decoupling controller；3) number of patent application CN200510040065.X, based on nerve network reverse five degrees of freedom without bearing forever
Magneticsynchro electric machine control system and control method, the invention are directed to the controlling party of permanentmagnet synchronous motor with five degrees of freedom without bearing design
Method；4) supersonic motor parameter identifications of article numbering 02588013 (2004) 07011705 based on RBF neural is pin
To the method for supersonic motor parameter identification；Nerve network reverse controller control motor thought is special with this used in three above invention
Profit has certain correlation, but neutral net is using RBF neural herein, and the BP networks used with them are not
Together；The present invention of article 4 is contrasted in the structure of motor, mathematical model, control method, control difficulty and requirement there are essential distinction,
Design to the RBF neural adaptive inverse control of selfbearings motors is provided without related patents and document at present
Material.
The content of the invention
The purpose of the present invention is nonlinear, the close coupling complication system for selfbearings motors, to suspending power, turns
Square power and rotor flux carry out Dynamic Nonlinear Decoupling control using RBF neural Adaptive inverse control device, there is provided a kind of
It can make selfbearings motors that there is excellent dynamic and static state performance, and with the change of the resistance parameter of electric machine and antiloading
The strong robustness of disturbance, and every Control performance standard of selfbearings motors can be effectively improved；In addition using RBF god
Output such as radial displacement, rotating speed and the magnetic linkage of online identification selfbearings motors are realized through network selfadapting identifier.
The technical scheme is that：A kind of selfbearings motors RBF neural adaptive inversion decoupling control and
Parameter identification method, including step：
Step 1, it is sent into sensors detection voltage, electric current, tach signal, signal after 3s/2r coordinate transforms
Flux Observation Model, to obtain magnetic linkage closedloop control and the required magnetic linkage information of neural metwork training；
Step 2, SVPWM algoritic modules one and voltage source inverter module one are composed in series to the SVPWM voltagetypes of extension
Inverter module one, the SVPWM voltagetypes that SVPWM algoritic modules two and voltage source inverter module two are composed in series to extension are inverse
Become device module two；
Step 3, selfbearings motors and its load module are built, by the SVPWM voltage source inverter modules of extension
First, the SVPWM voltage source inverters module two of extension and selfbearings motors and its load module are as an entirety
Form composite controlled object；
Step 4, the inverse controller of composite controlled object is built by RBF neural RBFNNC, using offline and online
The method being combined is trained and obtains the structure and parameter of RBF neural RBFNNC, by trained RBF neural
RBFNNC forms linear control system before being placed in composite controlled object, so as to fulfill the decoupling control to selfbearings motors
System；
Step 5, the identifier of composite controlled object is built by RBF neural RBFNNI, utilizes offline and online phase
With reference to method training and obtain the structure and parameter of RBF neural RBFNNI, after identification precision reaches design requirement, use
Identification signal replaces the signal that sensor detects, and realizes sensorless strategy.
Further, the 3s/2r coordinate transforms in the step 1 can be divided into the first coordinate transform and the second coordinate transform, institute
It is the selfbearings motors stator winding phase current i detected by Hall current sensor to state the first coordinate transform_{1a}、
i_{1b}、i_{1c}Convert to obtain electric current i under rotating coordinate system by Clark conversion and Park_{1d}、i_{1q}；Second coordinate transform is by suddenly
You detect selfbearings motors stator winding phase voltage U by voltage sensor_{1a}、U_{1b}、U_{1c}By Clark conversion and
Park converts to obtain voltage U under rotating coordinate system_{1d}、U_{1q}。
Further, the Flux Observation Model in the step 1 includes stator flux observer model and flux linkage observation mould
Type；
The stator flux observer model is by electric current i under rotating coordinate system_{1d}、i_{1q}With voltage U_{1d}、U_{1q}Obtained through functional transformation
Stator magnetic linkage component ψ under to rotating coordinate system_{1d}、ψ_{1q}：
ψ_{1d}=∫ (U_{1d}Ri_{1d})dtL_{1}i_{1d}
ψ_{1q}=∫ (U_{1q}Ri_{1q})dtL_{1}i_{1q}
The flux linkage observation model is by electric current i under rotating coordinate system_{1d}、i_{1q}With stator magnetic linkage under rotating coordinate system
Component ψ_{1d}、ψ_{1q}Rotating coordinate system lower rotor part magnetic linkage component ψ is obtained through functional transformation_{dr}、ψ_{qr}：
Further, in the step 2, the SVPWM algoritic modules one are by given voltage signal U_{α1s}*、U_{β1s}* be converted to and account for
Sky is than signal S_{a1s}、S_{b1s}、S_{c1s}, duty cycle signals are output to voltage source inverter one and produce voltage signal U_{a1s}、U_{b1s}、U_{c1s}Come
Control torque winding system；
The SVPWM algoritic modules two are by given voltage signal U_{α2s}*、U_{β2s}* duty cycle signals S is converted to_{a2s}、S_{b2s}、
S_{c2s}, duty cycle signals are output to voltage source inverter two and produce voltage signal U_{a2s}、U_{b2s}、U_{c2s}To control suspending windings system
System.
Further, the torque winding system mathematical model of selfbearings motors and its load module in the step 3
For common cage type asynchronous motor mathematical model；The suspending windings system suspension power of selfbearings motors and its load module
Math equation is as follows：
F_{x}=M ( i_{d1s}i_{d2s}+i_{q1s}i_{q2s})
F_{y}=M (i_{q1s}i_{d2s}+i_{d1s}i_{q2s})
M mutual inductances between torque winding and suspending windings in formula；
The suspending windings system state equation equation of selfbearings motors and its load module is as follows：
M is rotor quality in formula；F_{sx}、F_{sy}Intrinsic Maxwell force, its expression formula are：
F_{sx}=k_{s}x
F_{sy}=k_{s}y
In formulaFor radial displacement rigidity；R is rotor radius；L is rotor shaft length；μ_{0}For air magnetic conductance
Rate；δ is gas length；K is decay factor, generally takes 0.3.
Further, in the step 45, RBF neural RBFNNC and RBF neural RBFNNI structure and parameters are true
Fixed offline training method is：
The number and its center and width of hidden node are determined by offline training, and is calculated between hidden layer and output layer
Connection weight initial value, the input sample of RBFNNC is displacement X, Y, rotational speed omega_{r}, magnetic linkage ψ_{r}Setpoint and actual value mistake
Difference and error signal pass through e_{c}Output valve { the e of function module_{1}, e_{2}, e_{3}, e_{4}, e_{c1}, e_{c2}, e_{c3}, e_{c4}, output sample is process
{ the U of coordinate transform_{1α}, U_{1β}, U_{2α}, U_{2β}}；The input sample of RBFNNI outputs and inputs for the composite controlled object after delay
{U_{1α}, U_{1β}, U_{2α}, U_{2β}, X, Y, ω_{r}, ψ_{r}, output sample exports { X ', Y ', ω for the composite controlled object of identification_{r}', ψ_{r}′}；
Increase number of nodes according to certain rules selfadaptive in the training process, and output signal will be made according to rule
Deleted with too small Hidden unit, effectively realize that mission nonlinear maps with minimum Hidden unit.
Further, in the step 45, RBF neural RBFNNC and RBF neural RBFNNI structure and parameters are true
Least square method of recursion learning rules are used in fixed online training method.
The advantage of the invention is that：
1. selfbearings motors had both inherited magnetic bearing supporting motor advantage, solve traditional supports structure and answer
It is miscellaneous, volume is big, of high cost, efficiency is low, high failure rate shortcoming, and more reasonable than magnetic axis bearing motor, more practical.1) significantly
Axial space is shortened, improves axial utilization rate, the limitation of highpower and ultrahigh rotating speed can be broken through；2) this selfbearings motors
Radial suspension control system in power amplification circuit use the three phase power inverter circuit based on SVPWM algorithms so that motor
Control low in energy consumption, torque pulsation is small, improves the sine degree of phase current, reduces the THD of electric current.
2. by RBF neural adaptive inversion selfbearings motors are carried out torque force and radial suspension force it
Between dynamic decoupling so as to fulfill position system, rotor speed and magnetic linkage control while, this method relative to such as BP nerve
Network is inverse the characteristics of its is exclusive：RBF neural has good biological context and Function approximation capabilities, not only structure letter
List, fast convergence rate, generalization ability are strong, and have global optimum and Property of Optimal Approximation, further simplify control device network
Structure.
3. the characteristics of prominent in the present invention is to use RBF neural Adaptive Identification device, pass through the voltage electricity easily detected
Stream signal recognizes radial displacement, rotating speed and the magnetic linkage letter of selfbearings motors in the case of it need not know motor accurate parameters
Breath, and identification precision is high, so as to fulfill sensorless strategy, reduces system cost, improves system reliability, especially suitable for
Adverse circumstances and system requirements high field are closed.
4. the present invention increases number of nodes according to certain rules selfadaptive in the training process, and will be to defeated according to rule
Go out the too small Hidden unit of signal function to delete, to ensure that network structure is simple, compact, effectively realized with minimum Hidden unit
Mission nonlinear maps.Had using least square method of recursion study has supervision ground by online training by least square method of recursion
Regulating networks connection weight, the advantages of strengthening the robustness of inverse system.
Selfbearings motors RBF neural of the present invention based on RBF neural adaptive inversion construction is adaptive
Adverse control system, improves selfbearings motors control performance, and is equally applicable to other bearingfree motors control system
System and all types of electric machine control systems of magnetic bearing supporting.So the application prospect of this control method is very wide, for it
Also there is important application value for his bearingfree motor.
Brief description of the drawings
Fig. 1 is the schematic diagram of selfbearings motors rotor flux observer；
Fig. 2 be by the SVPWM voltage source inverters module one that extends and the SVPWM voltage source inverters module two of extension with
And the schematic diagram of selfbearings motors and its load module as an entirety composition composite controlled object；
Fig. 3 is RBF neural schematic internal view and isoboles；
Fig. 4 is that RBF neural inverse network and RBF neural parameter identification network collectively form compound input and output
Practise sample schematic diagram；
Fig. 5 is selfbearings motors RBF neural adaptive inversion decoupling and controlling system block diagram；
Fig. 6 is selfbearings motors RBF neural Adaptive inverse control and parameter identification system block diagram.
Embodiment
Embodiments of the present invention are：Become first using common electric current, voltage, speed, Flux Observation Model and Park
A flux observer with Clark conversion compositions is changed, to estimate the rotor of the selfbearings motors needed for magnetic linkage closedloop
Magnetic linkage information.Two SVPWM and voltage source inverter module, and selfbearings motors and its load module one are acted as
Composite controlled object is formed for an entirety, the controlled volume of composite controlled object is selfbearings motors rotor radial position
Shifting, rotating speed and magnetic linkage；Build the inverse controller of composite controlled object using RBF neural, inverse controller input be to
The error signal for determining signal and feedback signal forms closed loop；In addition realized using a RBF neural RBFNNI to controlled pair
Rotating speed and the displacement signal identification of elephant；Two RBF neurals all use three layers of feed forward type network, including input layer (8 sections
Point), hidden layer and output layer (4 nodes), wherein hidden layer uses radial basis function, the side being combined by offline and online study
Formula realizes network structure initialization and rightvalue optimization；Finally recognize RBF neural adaptive inversion system and autoadaptive parameter,
It is adaptive that two SVPWM and voltage source inverter module collectively form RBF neural adaptive inversion selfbearings motors
Adverse control system realizes the independent control to selfbearings motors torque force and radial suspension force, so as to fulfill motor dynamics
Decoupling and object parameters identification.
The embodiment of the present invention is further illustrated below in conjunction with the accompanying drawings.
As shown in Figure 1, construction selfbearings motors flux observer：By two coordinate transforms 11,12, stator magnetic linkage
Observation model 13 and rotor flux identification model 14 form；One of coordinate transform is by selfbearings motors stator
Winding phase current i_{1a}、i_{1b}、i_{1c}The winding phase current i of selfbearings motors is gathered by 3s/2r conversion 11_{1d}、i_{1q}；Separately
It is selfbearings motors stator winding phase voltage U that one 3s/2r, which converts 12 coordinate transforms,_{1a}、U_{1b}、U_{1c}Become by 3s/2r
12 are changed to gather the winding phase voltage U of selfbearings motors_{1d}、U_{1q}；Then module is recognized by corresponding magnetic linkage to obtain
Required magnetic linkage value.The stator flux observer model 13 is by electric current i under rotating coordinate system_{1d}、i_{1q}With voltage U_{1d}、U_{1q}Through letter
Transformation of variables obtains stator magnetic linkage component ψ under rotating coordinate system_{1d}、ψ_{1q}：
ψ_{1d}=∫ (U_{1d}Ri_{1d})dtL_{1}i_{1d}
ψ_{1q}=∫ (U_{1q}Ri_{1q})dtL_{1}i_{1q}
The flux linkage observation model 14 is by electric current i under rotating coordinate system_{1d}、i_{1q}With stator magnet under rotating coordinate system
Chain component ψ_{1d}、ψ_{1q}Rotating coordinate system lower rotor part magnetic linkage component ψ is obtained through functional transformation_{dr}、ψ_{qr}：
In Fig. 2, SVPWM algoritic modules 1 and SVPWM algoritic modules 2 31 are common SVPWM algorithms；Voltagetype inversion
Device 1 and voltage source inverter 2 32 are voltagetype power inverter IPM；Given voltage signal U_{α1s}*、U_{β1s}* calculated through SVPWM
Method module 1 obtains duty cycle signals S_{a1s}、S_{b1s}、S_{c1s}, duty cycle signals be output to voltage source inverter 1 produce electricity
Press signal U_{a1s}、U_{b1s}、U_{c1s}To control torque system；Given voltage signal U_{α2s}*、U_{β2s}* obtained through SVPWM algoritic modules 2 31
Duty cycle signals S_{a2s}、S_{b2s}、S_{c2s}, duty cycle signals are output to voltage source inverter 2 32 and produce voltage signal U_{a2s}、U_{b2s}、
U_{c2s}To control suspension system.
As shown in Figure 2 by two SVPWM21,31 and voltage source inverter module 22,32, and selfbearings motors
And its load module 1 forms composite controlled object 4 together as an entirety, the controlled volume input of composite controlled object is yes
{U_{1α}, U_{1β}, U_{2α}, U_{2β}Voltage signal, the displacement of selfbearings motors rotor radial, rotating speed and magnetic linkage are as output.
The core design of the present invention is the design and learning method of RBF neural.Initially set up RBF as shown in Figure 3
Neutral net RBFNN 75, as seen from the figure input layer 71 have 8 nodes, hidden layer 72 is the node that radial basis function is formed, radially
Basic function uses gaussian kernel function, and the output of ith of Hidden unit is：
H in formula_{i}For the output of ith of hidden node；X is input vector；C_{i}For the center of ith of hidden node；b_{i}For this
Hidden layer width；  *   it is euclideam norm；4 output nodes 74 are obtained by add operation；Established according to above method
RBF neural RBFNNC 51 and RBF neural RBFNNI 52.
RBF neural is combined by offline learning and online study and is trained, and training sample is as shown in Figure 4：
The input sample of RBFNNC 51 is displacement X, Y, rotational speed omega_{r}, magnetic linkage ψ_{r}Setpoint and value of feedback error signal and error signal
Through e_{c}Function：
e_{ci}(t)=e_{i}(t)e_{i}(t1) (i=1,2,3,4)
Output valve { e_{1}, e_{2}, e_{3}, e_{4}, e_{c1}, e_{c2}, e_{c3}, e_{c4}, output sample is the { U by coordinate transform_{1α}, U_{1β}, U_{2α},
U_{2β}}；The input sample of RBFNNI52 outputs and inputs { U for the composite controlled object of delay_{1α}, U_{1β}, U_{2α}, U_{2β}, X, Y, ω_{r},
ψ_{r}, output sample exports { X ', Y ', ω for the composite controlled object of identification_{r}', ψ_{r}′}.Offline training is first carried out, by RBF nerves
The Hidden unit initial value of network is set as zero, and hidden node is adaptively added according to " novelty " condition, is deleted using one kind
Strategy, as study constantly reduces to a certain extent sluggish knot removal into being about to those contributions to output, with true
It is simple, compact to protect network structure, the Nonlinear Mapping of system is effectively realized with minimum Hidden unit, offline training determines network
Structure simultaneously initializes hidden layer center and output weights；By online study, pass through gradient descent method online amending
Network parameters, make networkadaptive environmental change.
It is described effectively to realize that the specific algorithm that mission nonlinear maps is as follows with minimum Hidden unit：
For ith of learning sample (x (i), Z (i)),
Step 7.1, c is initialized_{j}For any real number, each Hidden unit output of RBF neural is calculated respectivelyWith
Network output y (i)：
Step 7.2, calculation error   E (i)  =  Z (i)y (i)  , Z (i) exports for target in formula, i.e., system is through adopting
Record after sample conditioning, y (i) is network reality output, and calculates the distance of sample and already present hidden layer：
d_{j}=  x (i)c_{j} , j=1,2 ..., m
M is already present Hidden unit number in formula；
Make d_{min}=min (d_{j})
Step 7.3, if   E_{i}  ＞ ε, d_{min}＞ λ (i), then：
λ (i)=max (λ_{max}γ^{i}, λ_{min})
ε is the desired precision of network in formula；The fitting precision of network when λ (i) is ith of input, with the carry out λ of study
(i) from λ_{max}It is reduced to λ_{min}；γ is decay factor, and 0 ＜ γ ＜ 1, then increase a Hidden unit, its parameter：
c_{k}=x_{i}
C in formula_{j}For the center of the p Hidden unit nearest from input sample, p=2 is taken here；
Step 7.4, otherwise, just least square method of recursion regulating networks connection weight is pressed；
Step 7.5, if all meeting for contact n sample of input：
σ is predefined constant in formula, if formula conditional when i=1,2 meets at the same time, when jth hidden node is deleted；
Step 7.6, i+1 group sample is inputted, is repeated the above process.
In the step 45, RBF neural RBFNNC51 and RBF neural RBFNNI52 structure and parameters determine
Online training method use least square method of recursion learning rules：
Step 8.1, kth group is inputted, reestablishes network output equation：
ω (k), u (k) are respectively weighted vector and radial basis function vector in formula, and H represents conjugate transposition；
Step 8.2, P (0)=δ is made^{1}I, ω (0)=0；
Step 8.3, calculate
ζ (k)=y (k)ω^{H}(k1)u(k)
ω (k)=ω (k1)+v (k) ζ^{*}(k)
P (k)=η^{1}P(k1)η^{1}v(k)uH(k)P(k1)
δ is on the occasion of small constant in formula；η is forgetting factor, 0≤η≤1；* complex conjugate is represented.
As shown in Figure 5 by the RBF neural established against 50 and two extend SVPWM voltage source inverters 2,3 knots
Close and form RBF neural Adaptive inverse control device 6, be placed on before selfbearings motors, realize decoupling control.
As Fig. 6 forms control system：By RBF neural RBFNNC51 and RBF neural RBFNNI52, two
SVPWM 21,31 and voltage source inverter module 22,32 collectively form RBF neural adaptive inversion selfbearings motors
Adaptive inverse control and parameter identification system 7.
The present invention can be realized according to abovementioned attached drawing and its correlation step.
The technical principle of the present invention is summarized further below.
The principle of the present invention is to change traditional selfbearings motors to use rotor fieldoriented and Airgapflux orientated
The strategy of decoupling control, design invention is a kind of to use RBF neural adaptive inverse control to selfbearings motors
Carry out Dynamic Nonlinear Decoupling control.
The adaptive Inverted control system of RBF neural and parameter identification method of the present invention is adaptive using RBF neural
Answer inverse controller to replace the team in existing decoupling control method to answer inverse system model, compensate for such as Airgapflux orientated, rotor
The dynamic decoupling of torque force and suspending power has been better achieved in the deficiency of field orientation and BP neural network reversed decoupling, this method,
Online identification precision is high, and reducing sensor use reduces cost, strengthens governing system reliability, while make the asynchronous electricity of bearingfree
Motivation governing system has stronger antiinterference and robustness.
The RBF neural Adaptive inverse control of selfbearings motors and the control method of parameter identification system are：
First using common electric current, voltage, speed, Flux Observation Model and Park conversion and a magnetic linkage of Clark conversion compositions
Observer, to estimate the rotor flux information of the selfbearings motors needed for magnetic linkage closedloop.By two SVPWM and voltagetype
Inverter module, and selfbearings motors and its load module form composite controlled object together as an entirety,
The controlled volume of composite controlled object is the displacement of selfbearings motors rotor radial, rotating speed and magnetic linkage；Using a RBF nerve
The inverse system of network struction composite controlled object, inverse system input are formed for the error signal of Setting signal and feedback signal and closed
Ring, realizes decoupling control between torque force and radial suspension force；In addition realized using a RBF neural without speed and without position
Displacement sensor controls, and realizes and selfbearings motors Dynamic Nonlinear Decoupling is controlled；It is finally that RBF neural is adaptive
Inverse controller and autoadaptive parameter identifier, that two SVPWM and voltage source inverter module collectively forms RBF neural is adaptive
It should be realized against selfbearings motors adaptive inverse control to selfbearings motors torque force and radial suspension force
Independent control, so as to fulfill rotor stable suspersion and operation.
Wherein abovementioned flux observer is recognized by two coordinate transforms, stator flux observer model and rotor flux
Model forms；One of coordinate transform is by selfbearings motors stator winding phase current i_{1a}、i_{1b}、i_{1c}Pass through Clark
Conversion and Park become the winding phase current i for bringing collection selfbearings motors_{1d}、i_{1q}；Another coordinate transform is nothing
Bearing asynchronous motor stator winding phase voltage U_{1a}、U_{1b}、U_{1c}It is different that collection bearingfree is brought by Clark conversion and Park changes
Walk the winding phase voltage U of motor_{1d}、U_{1q}；Then module is recognized by corresponding magnetic linkage to obtain required magnetic linkage value.
It should be understood that abovementioned example of applying is only illustrative of the invention and is not intended to limit the scope of the invention, the present invention is being read
Afterwards, modification of the those skilled in the art to the various equivalent forms of the present invention falls within the application appended claims and is limited
Scope.
Claims (7)
1. selfbearings motors RBF neural adaptive inversion decoupling control and parameter identification method, it is characterised in that bag
Include step：
Step 1, with sensors detection voltage, electric current, tach signal, signal magnetic linkage is sent into after 3s/2r coordinate transforms
Observation model, to obtain magnetic linkage closedloop control and the required magnetic linkage information of neural metwork training；
Step 2, SVPWM algoritic modules one (21) and voltage source inverter module one (22) are composed in series to the SVPWM electricity of extension
Die mould inverter module one (2), extension is composed in series by SVPWM algoritic modules two (31) and voltage source inverter module two (32)
SVPWM voltage source inverters module two (3)；
Step 3, selfbearings motors and its load module (1) are built, by the SVPWM voltage source inverters module one of extension
(2), the SVPWM voltage source inverters module two (3) of extension and selfbearings motors and its load module (1) are used as one
A entirety composition composite controlled object (4)；
Step 4, the inverse controller of composite controlled object (4) is built by RBF neural RBFNNC (51), using it is offline and
The method that line is combined is trained and obtains the structure and parameter of RBF neural RBFNNC (51), by trained RBF nerve nets
Network RBFNNC (51) forms linear control system before being placed in composite controlled object (4), so as to fulfill to bearingfree asynchronous electric
The decoupling control of machine；
Step 5, the identifier of composite controlled object (4) is built by RBF neural RBFNNI (52), using offline and online
The method being combined is trained and obtains the structure and parameter of RBF neural RBFNNI (52), and reaching design in identification precision will
After asking, the signal that sensor detects is replaced with identification signal, realizes sensorless strategy.
2. selfbearings motors RBF neural adaptive inversion decoupling control according to claim 1 and parameter are distinguished
Knowledge method, it is characterised in that the 3s/2r coordinate transforms in the step 1 can be divided into the first coordinate transform (11) and the second coordinate
Convert (12), first coordinate transform (11) is the selfbearings motors stator detected by Hall current sensor
Winding phase current i_{1a}、i_{1b}、i_{1c}Convert to obtain electric current i under rotating coordinate system by Clark conversion and Park_{1d}、i_{1q}；Second sits
Mark conversion (12) is that handle detects selfbearings motors stator winding phase voltage U by Hall voltage sensor_{1a}、U_{1b}、U_{1c}It is logical
Cross Clark conversion and Park converts to obtain voltage U under rotating coordinate system_{1d}、U_{1q}。
3. selfbearings motors RBF neural adaptive inversion decoupling control according to claim 2 and parameter are distinguished
Knowledge method, it is characterised in that the Flux Observation Model in the step 1 includes stator flux observer model (13) and rotor flux
Observation model (14)；
The stator flux observer model (13) is by electric current i under rotating coordinate system_{1d}、i_{1q}With voltage U_{1d}、U_{1q}Obtained through functional transformation
Stator magnetic linkage component ψ under to rotating coordinate system_{1d}、ψ_{1q}：
ψ_{1d}=∫ (U_{1d}Ri_{1d})dtL_{1}i_{1d}
ψ_{1q}=∫ (U_{1q}Ri_{1q})dtL_{1}i_{1q}
The flux linkage observation model (14) is by electric current i under rotating coordinate system_{1d}、i_{1q}With stator magnetic linkage under rotating coordinate system point
Measure ψ_{1d}、ψ_{1q}Rotating coordinate system lower rotor part magnetic linkage component ψ is obtained through functional transformation_{dr}、ψ_{qr}：
<mrow>
<msub>
<mi>&psi;</mi>
<mrow>
<mi>d</mi>
<mi>r</mi>
</mrow>
</msub>
<mo>=</mo>
<msub>
<mi>i</mi>
<mrow>
<mn>1</mn>
<mi>d</mi>
</mrow>
</msub>
<msub>
<mi>L</mi>
<mrow>
<mi>m</mi>
<mn>1</mn>
<mi>r</mi>
</mrow>
</msub>
<mo></mo>
<msub>
<mi>T</mi>
<mi>r</mi>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>&omega;</mi>
<mi>r</mi>
</msub>
<msub>
<mi>&psi;</mi>
<mrow>
<mn>1</mn>
<mi>d</mi>
</mrow>
</msub>
<mo>+</mo>
<mfrac>
<mrow>
<msub>
<mi>d&psi;</mi>
<mrow>
<mi>d</mi>
<mi>r</mi>
</mrow>
</msub>
</mrow>
<mrow>
<mi>d</mi>
<mi>t</mi>
</mrow>
</mfrac>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msub>
<mi>&psi;</mi>
<mrow>
<mi>q</mi>
<mi>r</mi>
</mrow>
</msub>
<mo>=</mo>
<msub>
<mi>i</mi>
<mrow>
<mn>1</mn>
<mi>q</mi>
</mrow>
</msub>
<msub>
<mi>L</mi>
<mrow>
<mi>m</mi>
<mn>1</mn>
<mi>r</mi>
</mrow>
</msub>
<mo></mo>
<msub>
<mi>T</mi>
<mi>r</mi>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>&omega;</mi>
<mi>r</mi>
</msub>
<msub>
<mi>&psi;</mi>
<mrow>
<mn>1</mn>
<mi>q</mi>
</mrow>
</msub>
<mo>+</mo>
<mfrac>
<mrow>
<msub>
<mi>d&psi;</mi>
<mrow>
<mi>q</mi>
<mi>r</mi>
</mrow>
</msub>
</mrow>
<mrow>
<mi>d</mi>
<mi>t</mi>
</mrow>
</mfrac>
<mo>)</mo>
</mrow>
<mo>.</mo>
</mrow>
4. selfbearings motors RBF neural adaptive inversion decoupling control according to claim 1 and parameter are distinguished
Knowledge method, it is characterised in that in the step 2,
The SVPWM algoritic modules one (21) are by given voltage signal U_{α1s}*、U_{β1s}* duty cycle signals S is converted to_{a1s}、S_{b1s}、
S_{c1s}, duty cycle signals are output to voltage source inverter module one (22) and produce voltage signal U_{a1s}、U_{b1s}、U_{c1s}To control torque
Winding system；
The SVPWM algoritic modules two (31) are by given voltage signal U_{α2s}*、U_{β2s}* duty cycle signals S is converted to_{a2s}、S_{b2s}、
S_{c2s}, duty cycle signals are output to voltage source inverter module two (32) and produce voltage signal U_{a2s}、U_{b2s}、U_{c2s}Suspend to control
Winding system.
5. selfbearings motors RBF neural adaptive inversion decoupling control according to claim 1 and parameter are distinguished
Knowledge method, it is characterised in that the torque winding system number of selfbearings motors and its load module (1) in the step 3
Model is common cage type asynchronous motor mathematical model；The suspending windings of selfbearings motors and its load module (1)
System suspension power math equation is as follows：
F_{x}=M ( i_{d1s}i_{d2s}+i_{q1s}i_{q2s})
F_{y}=M (i_{q1s}i_{d2s}+i_{d1s}i_{q2s})
M mutual inductances between torque winding and suspending windings in formula；
The suspending windings system state equation equation of selfbearings motors and its load module (1) is as follows：
<mrow>
<mi>m</mi>
<mover>
<mi>x</mi>
<mo>&CenterDot;&CenterDot;</mo>
</mover>
<mo>=</mo>
<msub>
<mi>F</mi>
<mi>x</mi>
</msub>
<mo></mo>
<msub>
<mi>F</mi>
<mrow>
<mi>s</mi>
<mi>x</mi>
</mrow>
</msub>
</mrow>
<mrow>
<mi>m</mi>
<mover>
<mi>y</mi>
<mo>&CenterDot;&CenterDot;</mo>
</mover>
<mo>=</mo>
<msub>
<mi>F</mi>
<mi>y</mi>
</msub>
<mo></mo>
<msub>
<mi>F</mi>
<mrow>
<mi>s</mi>
<mi>y</mi>
</mrow>
</msub>
</mrow>
M is rotor quality in formula；F_{sx}、F_{sy}Intrinsic Maxwell force, its expression formula are：
F_{sx}=k_{s}x
F_{sy}=k_{s}y
In formulaFor radial displacement rigidity；R is rotor radius；L is rotor shaft length；μ_{0}For air permeability；δ is
Gas length；K is decay factor, generally takes 0.3.
6. selfbearings motors RBF neural adaptive inversion decoupling control according to claim 1 and parameter are distinguished
Knowledge method, it is characterised in that in the step 45, RBF neural RBFNNC (51) and RBF neural RBFNNI (52)
The offline training method that structure and parameter determines is：
The number and its center and width of hidden node are determined by offline training, and calculates the company between hidden layer and output layer
Connect the initial value of power, the input sample of RBFNNC (51) is displacement X, Y, rotational speed omega_{r}, magnetic linkage ψ_{r}Setpoint and actual value mistake
Difference and error signal pass through e_{c}Output valve { the e of function module_{1},e_{2},e_{3},e_{4},e_{c1},e_{c2},e_{c3},e_{c4}, output sample is process
{ the U of coordinate transform_{1α},U_{1β},U_{2α},U_{2β}}；The input sample of RBFNNI (52) for composite controlled object after delay input and
Export { U_{1α},U_{1β},U_{2α},U_{2β},X,Y,ω_{r},ψ_{r}, output sample exports { X', Y', ω for the composite controlled object of identification_{r}',
ψ_{r}'}；
Increase number of nodes according to certain rules selfadaptive in the training process, and will be to exporting signal function mistake according to rule
Small Hidden unit is deleted, and effectively realizes that mission nonlinear maps with minimum Hidden unit.
7. selfbearings motors RBF neural adaptive inversion decoupling control according to claim 1 and parameter are distinguished
Knowledge method, it is characterised in that in the step 45, RBF neural RBFNNC (51) and RBF neural RBFNNI (52)
Least square method of recursion learning rules are used in the online training method that structure and parameter determines.
Priority Applications (1)
Application Number  Priority Date  Filing Date  Title 

CN201510092881.9A CN104767449B (en)  20150302  20150302  Selfbearings motors RBF neural adaptive inversion decoupling control and parameter identification method 
Applications Claiming Priority (1)
Application Number  Priority Date  Filing Date  Title 

CN201510092881.9A CN104767449B (en)  20150302  20150302  Selfbearings motors RBF neural adaptive inversion decoupling control and parameter identification method 
Publications (2)
Publication Number  Publication Date 

CN104767449A CN104767449A (en)  20150708 
CN104767449B true CN104767449B (en)  20180424 
Family
ID=53649112
Family Applications (1)
Application Number  Title  Priority Date  Filing Date 

CN201510092881.9A Expired  Fee Related CN104767449B (en)  20150302  20150302  Selfbearings motors RBF neural adaptive inversion decoupling control and parameter identification method 
Country Status (1)
Country  Link 

CN (1)  CN104767449B (en) 
Cited By (1)
Publication number  Priority date  Publication date  Assignee  Title 

CN110345013A (en) *  20190724  20191018  曲阜师范大学  Magnetic suspension vertical shaft Wind turbines control method based on Neural Network model predictive control 
Families Citing this family (13)
Publication number  Priority date  Publication date  Assignee  Title 

CN105227024B (en) *  20151028  20171110  广东美的制冷设备有限公司  A kind of rotor flux online identification method and system of permagnetic synchronous motor 
CN106788051B (en) *  20161123  20190205  江苏大学  A kind of inductiontype bearingless motor method for estimating rotating speed 
CN106788028B (en) *  20161220  20181214  江苏大学  Bearingfree permanent magnet synchronous motor intensified learning controller and its building method 
WO2018120000A1 (en) *  20161230  20180705  Nokia Technologies Oy  Artificial neural network 
CN106950822B (en) *  20170322  20191115  北京化工大学  The discrimination method of multivariable Hammerstein model under complicated heavytailed influence of noise 
CN107370429B (en) *  20170629  20200331  江苏大学  Fuzzy neural network inverse decoupling controller for bearingless permanent magnet synchronous motor 
CN108828941A (en) *  20180615  20181116  北京空天技术研究所  Separation control method based on parameter identification 
CN108646571B (en) *  20180712  20201030  北京航空航天大学  Highprecision position identification method for gyro frame servo system 
CN109687798A (en) *  20190108  20190426  江苏大学  Mix stator magnetic suspension switched reluctance motor suspension system high_order sliding mode control method 
CN110061676B (en) *  20190304  20200828  江苏大学  Bearingless permanent magnet synchronous motor controller based on flux linkage observer 
CN110071675A (en) *  20190417  20190730  清华大学  The design method and system of Amorphous Metal Motor electric and magnetic oscillation control 
CN110504878A (en) *  20190712  20191126  杭州洲钜电子科技有限公司  A kind of bearingfree permanent magnet synchronous motor rotor speed and displacement flexible measurement method 
CN111245318A (en) *  20200118  20200605  浙江大学  Radial force accurate compensation decoupling control method for bearingless permanent magnet synchronous motor 
Citations (4)
Publication number  Priority date  Publication date  Assignee  Title 

DE19505506A1 (en) *  19950210  19960822  Daimler Benz Ag  Use of observer model for induction motor torque estimation 
CN1655438A (en) *  20050311  20050817  江苏大学  Magnetic levitation switch reluctance motor radial neural network reversed decoupling controller and method for constructing same 
CN1845449A (en) *  20060308  20061011  江苏大学  Method for controlling bearingless AC asynchronous motor neural network inverse decoupling controller 
CN102739150A (en) *  20120620  20121017  哈尔滨工业大学  Parameter identification control device and control method of sensorless permanent magnet synchronous motor 

2015
 20150302 CN CN201510092881.9A patent/CN104767449B/en not_active Expired  Fee Related
Patent Citations (4)
Publication number  Priority date  Publication date  Assignee  Title 

DE19505506A1 (en) *  19950210  19960822  Daimler Benz Ag  Use of observer model for induction motor torque estimation 
CN1655438A (en) *  20050311  20050817  江苏大学  Magnetic levitation switch reluctance motor radial neural network reversed decoupling controller and method for constructing same 
CN1845449A (en) *  20060308  20061011  江苏大学  Method for controlling bearingless AC asynchronous motor neural network inverse decoupling controller 
CN102739150A (en) *  20120620  20121017  哈尔滨工业大学  Parameter identification control device and control method of sensorless permanent magnet synchronous motor 
NonPatent Citations (1)
Title 

基于RBF神经网络的多变量系统PID解耦控制;焦竹青等;《系统仿真学报》;20080228;第20卷(第3期);第627634页 * 
Cited By (1)
Publication number  Priority date  Publication date  Assignee  Title 

CN110345013A (en) *  20190724  20191018  曲阜师范大学  Magnetic suspension vertical shaft Wind turbines control method based on Neural Network model predictive control 
Also Published As
Publication number  Publication date 

CN104767449A (en)  20150708 
Similar Documents
Publication  Publication Date  Title 

Teja et al.  A new model reference adaptive controller for four quadrant vector controlled induction motor drives  
Glumineau et al.  Sensorless AC electric motor control  
Vieira et al.  Sensorless slidingmode rotor speed observer of induction machines based on magnetizing current estimation  
Barkat et al.  Noninteracting adaptive control of PMSM using interval type2 fuzzy logic systems  
Boukas et al.  Highperformance induction motor speed control using exact feedback linearization with state and state derivative feedback  
Chaoui et al.  Adaptive fuzzy logic control of permanent magnet synchronous machines with nonlinear friction  
Panchade et al.  A survey on sliding mode control strategies for induction motors  
CN201910764U (en)  Permanent magnet synchronous motor (PMSM) direct torque control system based on terminal sliding mode  
Li et al.  On the rejection of internal and external disturbances in a wind energy conversion system with directdriven PMSG  
CN100536311C (en)  Neural network generalized inverse permanent magnetism synchronous machine decoupling controller structure method without bearing  
Fan et al.  Design, analysis, and sensorless control of a selfdecelerating permanentmagnet inwheel motor  
JP4724070B2 (en)  Electric motor control device  
Luo et al.  Sensorless stator field orientation controlled induction motor drive with a fuzzy speed controller  
CN103532459A (en)  Linear servo motor control method for numericallycontrolled machine tool driving  
Sun et al.  Nonlinear flux linkage modeling of a bearingless permanent magnet synchronous motor based on AWLSSVM regression algorithm  
Rehman et al.  A fuzzy learning—Sliding mode controller for direct fieldoriented induction machines  
CN110289795B (en)  Permanent magnet synchronous motor control system and control method for electric automobile  
CN100573370C (en)  Based on neural network inverse control system for permanentmagnet synchronous motor with five degrees of freedom without bearing and control method  
CN100433537C (en)  Method for controlling bearingless AC asynchronous motor neural network inverse decoupling controller  
CN103401493B (en)  Permanent magnet synchronization motor spindle driving control system and method  
CN101951222A (en)  Control method of brushless doublefed motor and application thereof  
CN104378038B (en)  Permanent magnet synchronous motor parameter identification method based on artificial neural network  
CN106612089B (en)  Permanentmagnet synchronous motor with five degrees of freedom without bearing intensified learning controller and building method  
Zhang et al.  Radial magnetic bearings: An overview  
CN101917150A (en)  Robust controller of permanent magnet synchronous motor based on fuzzyneural network generalized inverse and construction method thereof 
Legal Events
Date  Code  Title  Description 

PB01  Publication  
C06  Publication  
SE01  Entry into force of request for substantive examination  
EXSB  Decision made by sipo to initiate substantive examination  
GR01  Patent grant  
GR01  Patent grant  
CF01  Termination of patent right due to nonpayment of annual fee 
Granted publication date: 20180424 Termination date: 20200302 

CF01  Termination of patent right due to nonpayment of annual fee 