CN104767449A - Bearing-free asynchronous motor RBF neural network self-adaptive inverse decoupling control and parameter identification method - Google Patents

Abstract

The invention discloses a bearing-free asynchronous motor RBF neural network self-adaptive inverse decoupling control and parameter identification method. An SVPWM module, a voltage inverter, a bearing-free asynchronous motor and a load of the bearing-free asynchronous motor form a whole serving as a composite controlled object. Two radial basis function neural networks are adopted to achieve inverse control and parameter identification conducted on the composite controlled object. A self-adaptive inverse controller is formed by using an RBF neural network through learning, and is serially connected in front of the composite controlled object, errors of a feedback signal and a given signal are input into an inverse controller, and accordingly closed-loop control is formed, then a self-adaptive parameter identifier is formed by using one RBF neural network through learning and identifies output quantity speed and displacement of the composite controlled object, speed-less and displacement-free sensor control is achieved, online learning of an estimation signal is aided by means of a learning algorithm, and non-linear dynamic decoupling control of the bearing-free asynchronous motor is achieved. The bearing-free asynchronous motor RBF neural network self-adaptive inverse decoupling control and parameter identification method is high in control speed and higher in identification accuracy, and a control system is excellent.

Description

Self-bearings motors RBF neural adaptive inversion uneoupled control and parameter identification method

Technical field

The present invention is a kind of self-bearings motors RBF neural Adaptive inverse control and parameter identification method of multivariable nonlinearity, is applicable to the high performance control of self-bearings motors.Self-bearings motors inherits magnetic bearing motor advantage, features such as having without friction, without wearing and tearing, do not need lubrication and sealing, aseptic, pollution-free, the life-span is long, be highly suitable for the high-technology field injecting the high-speed driving such as high-speed precision digital control lathe, high pressure sealing pump, specialized robot, high speed gyro, satellite flywheel, high-speed aircraft and control device, supercentrifuge, high-speed flywheel energy storage, application prospect is extensive, belongs to the technical field of electric drive control equipment.

Background technology

Self-bearings motors inside has complicated electromagnetic relationship, and therefore it is the controlled device of a class multivariable, non-linear, close coupling, realize controlling accurately its radial displacement, rotating speed very difficult.To realize the stable suspersion of self-bearings motors rotor and following given rotating speed and run, just uneoupled control must be carried out to the torque force of motor and suspending power.

But the control strategy of dynamic Decoupling Control of Load Torque is the difficult point realizing self-bearings motors steady operation always.Common control has Air-gap-flux orientated and orientation on rotor flux, and experiment proves that these two kinds of methods can realize comparatively stability contorting to self-bearings motors.Although Air-gap-flux orientated control method can realize uneoupled control between electromagnetic torque and radial suspension force, this algorithm is comparatively large by the impact of the parameter of electric machine (as rotor resistance, rotor leakage inductance etc.), and existence is steady torque, and the scope of application is limited; Orientation on rotor flux method, can accomplish the decoupling zero between torque current and exciting current, but only has rotor flux to reach when stablizing and keep constant could to realize electromagnetic torque and rotor flux decoupling zero, and belonging to steady state decoupling can not realize dynamic decoupling.BP Application of Neural Network controls in self-bearings motors and obtains good control effects, but BP neural net exists convergence rate in function approximation is absorbed in the shortcomings such as local minimum slowly, easily, and does not conform to biological context in theory.

For improving the dynamic duty performance of self-bearings motors further, need to consider that the dynamic decoupling of self-bearings motors and multivariable coordinated control combine, and then Structure of need is compacter, the self-bearings motors decoupling controller of more excellent performance.

Domestic existing related application: 1) number of patent application CN20061038711.3, name is called: the control method of bearing-less AC asynchronous motor neural network inverse decoupling controller, and this patent of invention is for bearing-less AC Asynchronous Motor neural network inverse decoupling controller; 2) number of patent application CN200510038099.5, name is called: bearing-free switch magnetic-resistance motor radial neural network reversed decoupling controller and building method, and this invention is for magnetic suspension switched reluctance motor design radial neural network reversed decoupling controller; 3) number of patent application CN200510040065.X, based on neural net inverse control system for permanent-magnet synchronous motor with five degrees of freedom without bearing and control method, this invention is for the control method of permanent-magnet synchronous motor with five degrees of freedom without bearing design; 4) article numbering 0258-8013 (2004) 07-0117-05 is method for supersonic motor parameter identification based on the supersonic motor parameter identification of RBF neural; Above three invention nerve network reverse controllers used control motor thought and this patent has certain correlation, but neural net employing herein is RBF neural, different from the BP network that they adopt; There is essential distinction in the structure of motor, Mathematical Modeling, control method, control difficulty and requirement in contrast article 4 the present invention, to the design of the RBF neural adaptive inverse control of self-bearings motors at present without Patents and documents and materials.

Summary of the invention

The object of the invention is non-linear, the close coupling complication system for self-bearings motors, RBF neural Adaptive inverse control device is adopted to carry out Dynamic Nonlinear Decoupling control to suspending power, torque force and rotor flux, a kind of self-bearings motors that can make is provided to have excellent dynamic and static state performance, and there is the strong robustness of the change of the opposing parameter of electric machine and anti-disturbance, effectively can improve again every Control performance standard of self-bearings motors; In addition RBF neural Adaptive Identification device is adopted to realize the output of on-line identification self-bearings motors as radial displacement, rotating speed and magnetic linkage.

Technical scheme of the present invention is: a kind of self-bearings motors RBF neural adaptive inversion uneoupled control and parameter identification method, comprise step:

Step 1, detects voltage, electric current, tach signal with sensors, and signal sends into Flux Observation Model after 3s/2r coordinate transform, obtains magnetic linkage closed-loop control and the magnetic linkage information required for neural metwork training;

Step 2, composes in series the SVPWM voltage source inverter module one of expansion by SVPWM algoritic module one and voltage source inverter module one, SVPWM algoritic module two and voltage source inverter module two are composed in series the SVPWM voltage source inverter module two of expansion;

Step 3, build self-bearings motors and load module thereof, the SVPWM voltage source inverter module one of expansion, the SVPWM voltage source inverter module two expanded and self-bearings motors and load module thereof are integrally formed composite controlled object;

Step 4, the inverse controller of composite controlled object is built by RBF neural RBFNNC, off-line and the method that combines online is utilized to train and obtain the structure and parameter of RBF neural RBFNNC, form linear control system before the RBF neural RBFNNC trained is placed in composite controlled object, thus realize the uneoupled control to self-bearings motors;

Step 5, the identifier of composite controlled object is built by RBF neural RBFNNI, off-line and the method that combines online is utilized to train and obtain the structure and parameter of RBF neural RBFNNI, after identification precision reaches designing requirement, replace transducer to detect the signal obtained with identification signal, realize sensorless strategy.

Further, the 3s/2r coordinate transform in described step 1 can be divided into the first coordinate transform and the second coordinate transform, and described first coordinate transform is the self-bearings motors stator winding phase current i being detected by Hall current sensor _1a, i _1b, i _1cto be converted by Clark and Park conversion obtains current i under rotating coordinate system _1d, i _1q; Second coordinate transform is self-bearings motors stator winding phase voltage U being detected by Hall voltage transducer _1a, U _1b, U _1cto be converted by Clark and Park conversion obtains voltage U under rotating coordinate system _1d, U _1q.

Further, the Flux Observation Model in described step 1 comprises stator flux observer model and flux linkage observation model;

Described stator flux observer model is by current i under rotating coordinate system _1d, i _1qand voltage U _1d, U _1qstator magnetic linkage component ψ under functional transformation obtains rotating coordinate system _1d, ψ _1q:

ψ _1d＝∫(U _1d-Ri _1d)dt-L ₁i _1d

ψ _1q＝∫(U _1q-Ri _1q)dt-L ₁i _1q

Described flux linkage observation model is by current i under rotating coordinate system _1d, i _1qwith stator magnetic linkage component ψ under rotating coordinate system _1d, ψ _1qrotating coordinate system lower rotor part magnetic linkage component ψ is obtained through functional transformation _dr, ψ _qr:

$\begin{array}{c}{\mathrm{\ψ}}_{\mathrm{dr}}=i_{1d}{L}_{m1r}-T_r({\mathrm{\ω}}_r{\mathrm{\ψ}}_{1d}+\frac{{\mathrm{d\ψ}}_{\mathrm{dr}}}{\mathrm{dt}})\\ {\mathrm{\ψ}}_{\mathrm{qr}}=i_{1q}{L}_{m1r}-T_r({\mathrm{\ω}}_r{\mathrm{\ψ}}_{1q}+\frac{d{\mathrm{\ψ}}_{\mathrm{qr}}}{\mathrm{dt}})\end{array}.$

Further, in described step 2, described SVPWM algoritic module one will to determining voltage signal U _{α 1s}*, U _{β 1s}* duty cycle signals S is converted to _a1s, S _b1s, S _c1s, duty cycle signals is outputted to voltage source inverter one and produces voltage signal U _a1s, U _b1s, U _c1scarry out controlling torque winding system;

Described SVPWM algoritic module two will to determining voltage signal U _{α 2s}*, U _{β 2s}* duty cycle signals S is converted to _a2s, S _b2s, S _c2s, duty cycle signals is outputted to voltage source inverter two and produces voltage signal U _a2s, U _b2s, U _c2scontrol suspending windings system.

Further, in described step 3, the torque winding system Mathematical Modeling of self-bearings motors and load module thereof is common cage type asynchronous motor Mathematical Modeling; The suspending windings system suspension power math equation of self-bearings motors and load module thereof is as follows:

F _x＝M(-i _d1si _d2s+i _q1si _q2s)

F _y＝M(i _q1si _d2s+i _d1si _q2s)

In formula, M is coefficient of mutual inductance between torque winding and suspending windings;

The suspending windings system state equation equation of self-bearings motors and load module thereof is as follows:

$\begin{array}{c}m\stackrel{\·\·}{x}=F_x-F_{\mathrm{sx}}\\ m\stackrel{\·\·}{y}=F_y-F_{\mathrm{sy}}\end{array}$

In formula, m is rotor quality; F _sx, F _syintrinsic Maxwell force, its expression formula is:

F _sx＝k _sx

F _sy＝k _sy

In formula for radial displacement rigidity; R is rotor radius; L is armature spindle length; μ ₀for air permeability; δ is gas length; K is decay factor, generally gets 0.3.

Further, in described step 4-5, the off-line training method that RBF neural RBFNNC and RBF neural RBFNNI structure and parameter are determined is:

By the number of off-line training determination hidden node and center thereof and width, and calculate the initial value of the connection weight between hidden layer and output layer, the input amendment of RBFNNC is displacement X, Y, rotational speed omega _r, magnetic linkage ψ _rset-point and the error of actual value and error signal through e _coutput valve { the e of function module ₁, e ₂, e ₃, e ₄, e _c1, e _c2, e _c3, e _c4, output sample is the { U through coordinate transform _{1 α}, U _{1 β}, U _{2 α}, U _{2 β}; The input amendment of RBFNNI is the input and output { U of the composite controlled object after time delay _{1 α}, U _{1 β}, U _{2 α}, U _{2 β}, X, Y, ω _r, ψ _r, output sample be identification composite controlled object export X ', Y ', ω _r', ψ _r';

Increase nodes according to certain rules self-adaptive in the training process, and delete acting on too small Hidden unit to output signal according to rule, effectively realize mission nonlinear with minimum Hidden unit and map.

Further, in described step 4-5, the on-line training method that RBF neural RBFNNC and RBF neural RBFNNI structure and parameter are determined adopts least square method of recursion learning rules.

The invention has the advantages that:

1. self-bearings motors had both inherited magnetic bearing supporting motor advantage, solves traditional supports complex structure, volume is large, cost is high, efficiency is low, failure rate high shortcoming, more more reasonable, more practical than magnetic bearing motor again.1) significantly shorten axial space, improve axial utilance, high-power and ultrahigh rotating speed restriction can be broken through; 2) in the radial suspension control system of this self-bearings motors, power amplification circuit adopts the three phase power inverter circuit based on SVPWM algorithm, and make Electric Machine Control low in energy consumption, torque pulsation is little, improves the sine degree of phase current, reduces the THD of electric current.

2. the dynamic decoupling between torque force and radial suspension force is being carried out to self-bearings motors thus while realizing position system, rotor speed and magnetic linkage control by RBF neural adaptive inversion, the feature that the method has it exclusive relative to such as BP nerve network reverse: RBF neural has good biological context sum functions approximation capability, not only structure is simple, fast convergence rate, generalization ability are strong, and there is global optimum and Property of Optimal Approximation, simplify controller network structure further.

3. feature outstanding in the present invention adopts RBF neural Adaptive Identification device, by the voltage and current signal that easily detects in the radial displacement without the need to knowing identification self-bearings motors in motor accurate parameters situation, rotating speed and magnetic linkage information, and identification precision is high, thus realize sensorless strategy, reduce system cost, improve system reliability, be specially adapted to adverse circumstances and the conjunction of system requirements high field.

4. the present invention increases nodes according to certain rules self-adaptive in the training process, and delete acting on too small Hidden unit to output signal according to rule, to guarantee that network configuration is simple, compact, effectively realize mission nonlinear with minimum Hidden unit and map.Adopt least square method of recursion study to have and have supervision ground regulating networks connection weight by on-line training by least square method of recursion, strengthen the advantage of the robustness of inverse system.

The present invention is based on the self-bearings motors RBF neural adaptive inverse control of RBF neural adaptive inversion structure, improve self-bearings motors control performance, and be equally applicable to all types of electric machine control systems of other bearing-free motor control system and magnetic bearing supporting.So the application prospect of this control method is very wide, for other bearing-free motors, also there is important using value.

Accompanying drawing explanation

Fig. 1 is the schematic diagram of self-bearings motors rotor flux observer;

Fig. 2 integrally forms the schematic diagram of composite controlled object by the SVPWM voltage source inverter module two of the SVPWM voltage source inverter module one expanded and expansion and self-bearings motors and load module thereof;

Fig. 3 is RBF neural schematic internal view and isoboles;

Fig. 4 is that RBF neural inverse network and RBF neural parameter identification network form compound input and output learning sample schematic diagram jointly;

Fig. 5 is self-bearings motors RBF neural adaptive inversion decoupling and controlling system block diagram;

Fig. 6 is self-bearings motors RBF neural Adaptive inverse control and parameter identification system block diagram.

Embodiment

Embodiments of the present invention are: first adopt conventional electric current, voltage, speed, Flux Observation Model and Park to convert and convert with Clark the flux observer formed, estimate the rotor flux information of the self-bearings motors needed for magnetic linkage closed-loop.By two SVPWM and voltage source inverter module, and self-bearings motors and load module thereof integrally form composite controlled object together, and the controlled volume of composite controlled object is the displacement of self-bearings motors rotor radial, rotating speed and magnetic linkage; Adopt a RBF neural to build the inverse controller of composite controlled object, the error signal that inverse controller is input as Setting signal and feedback signal forms closed loop; Adopt a RBF neural RBFNNI realization to the rotating speed of controlled device and displacement signal identification in addition; Two RBF neural all adopt three layers of feed forward type network, comprise input layer (8 nodes), hidden layer and output layer (4 nodes), wherein hidden layer uses RBF, and the mode combined by off-line and on-line study realizes network configuration initialization and right-value optimization; Finally RBF neural adaptive inversion system and auto-adaptive parameter identification, two SVPWM and voltage source inverter module are formed jointly RBF neural adaptive inversion self-bearings motors adaptive inverse control to realize controlling the independence of self-bearings motors torque force and radial suspension force, thus realize motor dynamics decoupling zero and object parameters identification.

The specific embodiment of the present invention is further illustrated below in conjunction with accompanying drawing.

As shown in Figure 1, self-bearings motors flux observer is constructed: by two coordinate transforms 11,12, stator flux observer model 13 and rotor flux identification model 14 form; One of them coordinate transform is by self-bearings motors stator winding phase current i _1a, i _1b, i _1cthe winding phase current i of self-bearings motors is gathered by 3s/2r conversion 11 _1d, i _1q; It is self-bearings motors stator winding phase voltage U that another 3s/2r converts 12 coordinate transforms _1a, U _1b, U _1cthe winding phase voltage U of self-bearings motors is gathered by 3s/2r conversion 12 _1d, U _1q; Then required magnetic linkage value is obtained by corresponding magnetic linkage recognition module.Described stator flux observer model 13 is by current i under rotating coordinate system _1d, i _1qand voltage U _1d, U _1qstator magnetic linkage component ψ under functional transformation obtains rotating coordinate system _1d, ψ _1q:

ψ _1d＝∫(U _1d-Ri _1d)dt-L ₁i _1d

ψ _1q＝∫(U _1q-Ri _1q)dt-L ₁i _1q

Described flux linkage observation model 14 is by current i under rotating coordinate system _1d, i _1qwith stator magnetic linkage component ψ under rotating coordinate system _1d, ψ _1qrotating coordinate system lower rotor part magnetic linkage component ψ is obtained through functional transformation _dr, ψ _qr:

$\begin{array}{c}{\mathrm{\ψ}}_{\mathrm{dr}}=i_{1d}{L}_{m1r}-T_r({\mathrm{\ω}}_r{\mathrm{\ψ}}_{1d}+\frac{{\mathrm{d\ψ}}_{\mathrm{dr}}}{\mathrm{dt}})\\ {\mathrm{\ψ}}_{\mathrm{qr}}=i_{1q}{L}_{m1r}-T_r({\mathrm{\ω}}_r{\mathrm{\ψ}}_{1q}+\frac{d{\mathrm{\ψ}}_{\mathrm{qr}}}{\mathrm{dt}})\end{array}.$

In Fig. 2, SVPWM algoritic module 1 and SVPWM algoritic module 2 31 are conventional SVPWM algorithm; Voltage source inverter 1 and voltage source inverter 2 32 are voltage-type power inverter IPM; To determining voltage signal U _{α 1s}*, U _{β 1s}* duty cycle signals S is obtained through SVPWM algoritic module 1 _a1s, S _b1s, S _c1s, duty cycle signals outputted to voltage source inverter 1 and produces voltage signal U _a1s, U _b1s, U _c1scarry out controlling torque system; To determining voltage signal U _{α 2s}*, U _{β 2s}* duty cycle signals S is obtained through SVPWM algoritic module 2 31 _a2s, S _b2s, S _c2s, duty cycle signals is outputted to voltage source inverter 2 32 and produces voltage signal U _a2s, U _b2s, U _c2scontrol suspension system.

As shown in Figure 2 by two SVPWM21,31 and voltage source inverter module 22,32, and self-bearings motors and load module 1 thereof integrally form composite controlled object 4 together, and it is { U that the controlled volume of composite controlled object is input as _{1 α}, U _{1 β}, U _{2 α}, U _{2 β}voltage signal, the displacement of self-bearings motors rotor radial, rotating speed and magnetic linkage are as output.

Core design of the present invention is design and the learning method of RBF neural.First set up RBF neural RBFNN 75 as shown in Figure 3, input layer 71 has 8 nodes as seen from the figure, the node that hidden layer 72 is formed for RBF, and RBF adopts gaussian kernel function, and the output of i-th Hidden unit is:

$h_i=\exp (-\frac{\left|\right|x-c_i\left|\right|}{2}),(i=\mathrm{1,2},...,n)$

H in formula _ibe the output of i-th hidden node; X is input vector; C _iit is the center of i-th hidden node; b _ifor this hidden layer width; || * || be euclideam norm; 4 output nodes 74 are obtained through add operation; According to above method establishment RBF neural RBFNNC 51 and RBF neural RBFNNI 52.

RBF neural is combined by off-line learning and on-line study and trains, training sample as shown in Figure 4: the input amendment of RBFNNC 51 is displacement X, Y, rotational speed omega _r, magnetic linkage ψ _rset-point and value of feedback error signal and error signal through e _cfunction:

e _ci(t)＝e _i(t)-e _i(t-1) (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 through coordinate transform _{1 α}, U _{1 β}, U _{2 α}, U _{2 β}; The input amendment of RBFNNI52 is the input and output { U of the composite controlled object of time delay _{1 α}, U _{1 β}, U _{2 α}, U _{2 β}, X, Y, ω _r, ψ _r, output sample be identification composite controlled object export X ', Y ', ω _r', ψ _r'.First carry out off-line training, the Hidden unit initial value of RBF neural is set as zero, hidden node is added according to " novelty " condition self adaptation, adopt a kind of deletion strategy, along with study constantly carries out those to be reduced to a certain degree sluggish knot removal to the contribution exported, to guarantee that network configuration is simple, compact, effectively realize the Nonlinear Mapping of system with minimum Hidden unit, off-line training determination network configuration also carries out initialization to hidden layer center and output weights; Through on-line study, by gradient descent method on-line amending network parameters, make network-adaptive environmental change.

Described with minimum Hidden unit effectively realize mission nonlinear map specific algorithm as follows:

For i-th learning sample (x (i), Z (i)),

Step 7.1, initialization c _jfor any real number, each Hidden unit calculating RBF neural respectively exports y (i) is exported with network:

Step 7.2, the error of calculation || E (i) ||=|| Z (i)-y (i) ||, in formula, Z (i) is target output, the i.e. record of system after sampling conditioning, y (i) is the actual output of network, and calculates the distance of sample and already present hidden layer:

d _j＝||x(i)-c _j||，j＝1，2，...，m

In formula, m is already present Hidden unit number;

Make d _min=min (d _j)

Step 7.3, if || E _i|| > ε, d _min> λ (i), then:

λ(i)＝max(λ _maxγ ⁱ，λ _min)

In formula, ε is the precision that network is expected; The fitting precision of network when λ (i) is i-th input, carries out λ (i) from λ along with what learn _maxbe reduced to λ _min; γ is decay factor, 0 < γ < 1, then increase a Hidden unit, its parameter:

c _k＝x _i

${\mathrm{\&sigma;}}_k=\frac{1}{p}{\left(\underset{j=1}{\overset{p}{\mathrm{\&Sigma;}}}{\left|\right|x_i-c_j\left|\right|}^2\right)}^{\frac{1}{2}}$

C in formula _jfor the center from nearest p the Hidden unit of input amendment, get p=2 here;

Step 7.4, otherwise, just press least square method of recursion regulating networks connection weight;

Step 7.5, if all satisfied for contact input n sample:

In formula, σ is predefined constant, if formula conditional is at i=1,2 when meeting simultaneously, when a jth hidden node is deleted;

Step 7.6, input the i-th+1 group sample, repeats said process.

In described step 4-5, the on-line training method that RBF neural RBFNNC51 and RBF neural RBFNNI52 structure and parameter are determined adopts least square method of recursion learning rules:

Step 8.1, to the input of kth group, re-establishes network output equation:

In formula, ω (k), u (k) are respectively weighted vector and RBF vector, and H represents conjugate transpose;

Step 8.2, makes P (0)=δ ^-1i, ω (0)=0;

Step 8.3, calculates $v\left(k\right)=\frac{{\mathrm{\&eta;}}^{-1}P(k-1)u\left(k\right)}{1+{\mathrm{\&eta;}}^{-1}{u}^H\left(k\right)P(k-1)u\left(k\right)}$

ζ(k)＝y(k)-ω ^H(k-1)u(k)

ω(k)＝ω(k-1)+v(k)ζ ^*(k)

P(k)＝η ^-1P(k-1)-η ^-1v(k)uH(k)P(k-1)

In formula, δ is on the occasion of little constant; η is forgetting factor, 0≤η≤1; * complex conjugate is represented.

As shown in Figure 5 the SVPWM voltage source inverter 2,3 that the RBF neural established is expanded against 50 and two is bonded RBF neural Adaptive inverse control device 6, before being placed on self-bearings motors, realizes uneoupled control.

As Fig. 6 forms control system: RBF neural RBFNNC51 and RBF neural RBFNNI52, two SVPWM 21,31 and voltage source inverter module 22,32 are formed RBF neural adaptive inversion self-bearings motors Adaptive inverse control and parameter identification system 7 jointly.

Just the present invention can be realized according to above-mentioned accompanying drawing and correlation step thereof.

Further know-why of the present invention is summarized below.

Principle of the present invention changes traditional self-bearings motors to adopt strategy that is rotor field-oriented and Air-gap-flux orientated uneoupled control, and a kind of RBF neural adaptive inverse control that adopts of design invention carries out Dynamic Nonlinear Decoupling control to self-bearings motors.

RBF neural adaptive inversion Systematical control of the present invention and parameter identification method use RBF neural Adaptive inverse control device to replace the team in existing decoupling control method to answer inverse system model, compensate for the deficiency of such as Air-gap-flux orientated, rotor field-oriented and BP neural net reversed decoupling, the method achieves the dynamic decoupling of torque force and suspending power better, on-line identification precision is high, reduce transducer use to reduce costs, governing system reliability is strengthened, makes self-bearings motors governing system have stronger anti-interference and robustness simultaneously.

The RBF neural Adaptive inverse control of self-bearings motors and the control method of parameter identification system are: first adopt conventional electric current, voltage, speed, Flux Observation Model and Park to convert and convert with Clark the flux observer formed, estimate the rotor flux information of the self-bearings motors needed for magnetic linkage closed-loop.By two SVPWM and voltage source inverter module, and self-bearings motors and load module thereof integrally form composite controlled object together, and the controlled volume of composite controlled object is the displacement of self-bearings motors rotor radial, rotating speed and magnetic linkage; Adopt a RBF neural to build the inverse system of composite controlled object, the error signal that inverse system is input as Setting signal and feedback signal forms closed loop, realizes uneoupled control between torque force and radial suspension force; Adopt a RBF neural to realize controlling without speed and without gap sensors in addition, realize controlling self-bearings motors Dynamic Nonlinear Decoupling; Finally RBF neural Adaptive inverse control device and auto-adaptive parameter identifier, two SVPWM and voltage source inverter module are formed jointly RBF neural adaptive inversion self-bearings motors adaptive inverse control to realize controlling the independence of self-bearings motors torque force and radial suspension force, thus realize rotor stable suspersion and operation.

Wherein above-mentioned flux observer is made up of two coordinate transforms, stator flux observer model and rotor flux identification model; One of them coordinate transform is by self-bearings motors stator winding phase current i _1a, i _1b, i _1cto be converted by Clark and Park conversion gathers the winding phase current i of self-bearings motors _1d, i _1q; Another coordinate transform is self-bearings motors stator winding phase voltage U _1a, U _1b, U _1cto be converted by Clark and Park conversion gathers the winding phase voltage U of self-bearings motors _1d, U _1q; Then required magnetic linkage value is obtained by corresponding magnetic linkage recognition module.

Should understand above-mentioned example of executing only to be not used in for illustration of the present invention and to limit the scope of the invention, after having read the present invention, the amendment of those skilled in the art to the various equivalent form of value of the present invention has all fallen within the application's claims limited range.

Claims (7)

1. self-bearings motors RBF neural adaptive inversion uneoupled control and parameter identification method, is characterized in that, comprises step:

Step 1, detects voltage, electric current, tach signal with sensors, and signal sends into Flux Observation Model after 3s/2r coordinate transform, obtains magnetic linkage closed-loop control and the magnetic linkage information required for neural metwork training;

Step 2, SVPWM algoritic module one (21) and voltage source inverter module one (22) are composed in series the SVPWM voltage source inverter module one (2) of expansion, SVPWM algoritic module two (31) and voltage source inverter module two (32) are composed in series the SVPWM voltage source inverter module two (3) of expansion;

Step 3, build self-bearings motors and load module (1) thereof, the SVPWM voltage source inverter module one (2) of expansion, the SVPWM voltage source inverter module two (3) expanded and self-bearings motors and load module (1) thereof are integrally formed composite controlled object (4);

Step 4, the inverse controller of composite controlled object (4) is built by RBF neural RBFNNC (51), off-line and the method that combines online is utilized to train and obtain the structure and parameter of RBF neural RBFNNC (51), form linear control system before the RBF neural RBFNNC trained (51) being placed in composite controlled object (4), thus realize the uneoupled control to self-bearings motors;

Step 5, the identifier of composite controlled object (4) is built by RBF neural RBFNNI (52), off-line and the method that combines online is utilized to train and obtain the structure and parameter of RBF neural RBFNNI (52), after identification precision reaches designing requirement, replace transducer to detect the signal obtained with identification signal, realize sensorless strategy.

2. self-bearings motors RBF neural adaptive inversion uneoupled control according to claim 1 and parameter identification method, it is characterized in that, 3s/2r coordinate transform in described step 1 can be divided into the first coordinate transform (11) and the second coordinate transform (12), and described first coordinate transform (11) is the self-bearings motors stator winding phase current i being detected by Hall current sensor _1a, i _1b, i _1cto be converted by Clark and Park conversion obtains current i under rotating coordinate system _1d, i _1q; Second coordinate transform (12) is self-bearings motors stator winding phase voltage U being detected by Hall voltage transducer _1a, U _1b, U _1cto be converted by Clark and Park conversion obtains voltage U under rotating coordinate system _1d, U _1q.

3. self-bearings motors RBF neural adaptive inversion uneoupled control according to claim 2 and parameter identification method, it is characterized in that, the Flux Observation Model in described step 1 comprises stator flux observer model (13) and flux linkage observation model (14);

Described stator flux observer model (13) is by current i under rotating coordinate system _1d, i _1qand voltage U _1d, U _1qstator magnetic linkage component ψ under functional transformation obtains rotating coordinate system _1d, ψ _1q:

ψ _1d＝∫(U _1d-Ri _1d)dt-L ₁i _1d

ψ _1q＝∫(U _1q-Ri _1q)dt-L ₁i _1q

Described flux linkage observation model (14) is by current i under rotating coordinate system _1d, i _1qwith stator magnetic linkage component ψ under rotating coordinate system _1d, ψ _1qrotating coordinate system lower rotor part magnetic linkage component ψ is obtained through functional transformation _dr, ψ _qr:

${\mathrm{\&psi;}}_{\mathrm{dr}}=i_{1d}{L}_{m1r}-T_r({\mathrm{\&omega;}}_r{\mathrm{\&psi;}}_{1d}+\frac{d{\mathrm{\&psi;}}_{\mathrm{dr}}}{\mathrm{dt}})$

${\mathrm{\&psi;}}_{\mathrm{qr}}=i_{1q}{L}_{m1r}-T_r({\mathrm{\&psi;}}_r{\mathrm{\&psi;}}_{1q}+\frac{d{\mathrm{\&psi;}}_{\mathrm{qr}}}{\mathrm{dt}}).$

4. self-bearings motors RBF neural adaptive inversion uneoupled control according to claim 1 and parameter identification method, is characterized in that, in described step 2,

Described SVPWM algoritic module one (21) will to determining voltage signal U _{α 1s}*, U _{β 1s}* duty cycle signals S is converted to _a1s, S _b1s, S _c1s, duty cycle signals is outputted to voltage source inverter one (22) and produces voltage signal U _a1s, U _b1s, U _c1scarry out controlling torque winding system;

Described SVPWM algoritic module two (31) will to determining voltage signal U _{α 2s}*, U _{β 2s}* duty cycle signals S is converted to _a2s, S _b2s, S _c2s, duty cycle signals is outputted to voltage source inverter two (32) and produces voltage signal U _a2s, U _b2s, U _c2scontrol suspending windings system.

5. self-bearings motors RBF neural adaptive inversion uneoupled control according to claim 1 and parameter identification method, it is characterized in that, in described step 3, the torque winding system Mathematical Modeling of self-bearings motors and load module (1) thereof is common cage type asynchronous motor Mathematical Modeling; The suspending windings system suspension power math equation of self-bearings motors and load module (1) thereof is as follows:

F _x＝M(-i _d1si _d2s+i _q1si _q2s)

F _y＝M(i _q1si _d2s+i _d1si _q2s)

In formula, M is coefficient of mutual inductance between torque winding and suspending windings;

The suspending windings system state equation equation of self-bearings motors and load module (1) thereof is as follows:

$m\stackrel{..}{x}=F_x-F_{\mathrm{sx}}$

$m\stackrel{..}{y}=F_y-F_{\mathrm{sy}}$

In formula, m is rotor quality; F _sx, F _syintrinsic Maxwell force, its expression formula is:

F _sx＝k _sx

F _sy＝k _sy

In formula for radial displacement rigidity; R is rotor radius; L is armature spindle length; μ ₀for air permeability; δ is gas length; K is decay factor, generally gets 0.3.

6. self-bearings motors RBF neural adaptive inversion uneoupled control according to claim 1 and parameter identification method, it is characterized in that, in described step 4-5, the off-line training method that RBF neural RBFNNC (51) and RBF neural RBFNNI (52) structure and parameter are determined is:

By the number of off-line training determination hidden node and center thereof and width, and calculate the initial value of the connection weight between hidden layer and output layer, the input amendment of RBFNNC (51) is displacement X, Y, rotational speed omega _r, magnetic linkage ψ _rset-point and the error of actual value and error signal through e _coutput valve { the e of function module ₁, e ₂, e ₃, e ₄, e _c1, e _c2, e _c3, e _c4, output sample is the { U through coordinate transform _{1 α}, U _{1 β}, U _{2 α}, U _{2 β}; The input amendment of RBFNNI (52) is the input and output { U of the composite controlled object after time delay _{1 α}, U _{1 β}, U _{2 α}, U _{2 β}, X, Y, ω _r, ψ _r, output sample be identification composite controlled object export X ', Y ', ω _r', ψ _r';

Increase nodes according to certain rules self-adaptive in the training process, and delete acting on too small Hidden unit to output signal according to rule, effectively realize mission nonlinear with minimum Hidden unit and map.

7. self-bearings motors RBF neural adaptive inversion uneoupled control according to claim 1 and parameter identification method, it is characterized in that, in described step 4-5, the on-line training method that RBF neural RBFNNC (51) and RBF neural RBFNNI (52) structure and parameter are determined adopts least square method of recursion learning rules.