CN103886131A - Switch reluctance motor magnetic flux linkage online modeling method based on extreme learning machine - Google Patents

Abstract

The invention discloses a switch reluctance motor magnetic flux linkage online modeling method based on an extreme learning machine and belongs to the technical field of switch reluctance motor intelligent control. On the basis of the static data set of a switch reluctance motor, the extreme learning machine is used to built the magnetic flux linkage offline model of the switch reluctance motor, and the magnetic flux linkage online model of the switch reluctance motor is then built according real-time error adjusting. The magnetic flux linkage online model has the function of online increasing magnetic flux linkage accuracy in real time, can accurately describe the dynamic performance of the switch reluctance motor, and can adapt to different working environments of the switch reluctance motor.

Description

A kind of switched reluctance motor flux linkage line modeling method based on extreme learning machine

Technical field

The present invention relates to a kind of switched reluctance motor flux linkage line modeling method based on extreme learning machine, belong to switched reluctance machines field of intelligent control.

Background technology

Flux linkage characteristic is the key property of switched reluctance machines.The dark feature such as saturated of magnetic circuit during due to the double-salient-pole structure of switched reluctance machines and operation, accurately grasp flux linkage characteristic for optimizing design of electrical motor, improve runnability, to realize position sensorless etc. all significant.The flux linkage characteristic of switched reluctance machines can obtain by the method for FEM (finite element) calculation or experiment measuring.Finite Element Method Consideration complexity, calculated amount is large, and therefore measurement method is still the main method of obtaining at present flux linkage characteristic.

The stator magnetic linkage of switched reluctance machines is the nonlinear function of rotor-position and winding current, sets up accurate and practical flux linkage model quite concerned.Although traditional look-up table has higher precision, computation period is long, cannot meet the requirement of real-time control and design of electrical motor rapid modeling.Function analytical method is optimization system performance to a certain extent, but not strong to the adaptability of load and environmental change.Along with the development of artificial intelligence technology, the application of learning algorithm in Modeling of Switched Reluctance Motors is more and more extensive, the switched reluctance motor flux linkage modeling method being suggested has: BP neural network, RBF neural network, support vector machine etc., and BP neural network, RBF neural network need mass data to realize in switched reluctance motor flux linkage modeling, and pace of learning is slow; In the situation that support vector machine parameter being optimized without intelligent algorithm (as particle cluster algorithm, genetic algorithm etc.), it is general that support vector machine is learnt the flux linkage model precision that obtains; Extreme learning machine is a kind of single hidden layer feedforward neural network, in the magnetic linkage modeling process of realizing switched reluctance machines without mass data, study " extremely " is rapid, and the switched reluctance motor flux linkage model accuracy obtaining is very high, above reason makes the switched reluctance motor flux linkage modeling method based on extreme learning machine, has very high researching value.Simultaneously, existing switched reluctance motor flux linkage modeling is off-line modeling substantially, the magnetic linkage off-line model obtaining is difficult to adapt to the change of switched reluctance machines working environment, some scholars is being made research aspect the magnetic linkage line modeling of switched reluctance machines, wherein there is a kind of method to be: with the magnetic linkage off-line model establishing in advance, switched reluctance motor flux linkage to be predicted, predict the outcome and compare with the real-time magnetic linkage data of switched reluctance machines, the magnetic linkage off-line model real-time estimate error being established in advance, re-establish online flux linkage model according to real-time Flux estimation error, the validity of the method will inevitably require: online switched reluctance machines again modeling process needs to realize fast, this point requires particularly outstanding in the time that switched reluctance machines runs up, and this magnetic linkage line modeling method use is that pace of learning is slower, the RBF neural network that needs mass data to learn, limited to a great extent this magnetic linkage line modeling method at switched reluctance machines in higher speed, application under running up, extreme learning machine has just well solved this problem.

Summary of the invention

The object of this invention is to provide a kind of switched reluctance motor flux linkage line modeling method based on extreme learning machine.

The discrete integration formula of switched reluctance machines winding magnetic linkage is as follows:

ψ(k)＝ψ(k-1)+0.5T[u(k)-ri(k)+u(k-1)-ri(k-1)] (1)

In formula (1),

ψ phase winding magnetic linkage value,

U phase winding terminal voltage,

The T sampling period,

I phase current,

The internal resistance of r phase winding,

K sampled point sequence number.

Switched reluctance motor flux linkage model is set up required magnetic linkage data acquisition, is based on now having delivered disclosed switched reluctance motor flux linkage characteristic detection system, and it adopts step voltage to measure voltage and the electric current of phase winding, and core is DSP.In driver circuit module, 220V alternating voltage obtains needed DC voltage after the easy rectification device rectification of transformer pressure-reducing, four diodes; For maintaining voltage constant and preventing power supply and the vibration of winding formation lc circuit, the electrochemical capacitor C of a larger capacitance in parallel after power supply, disconnecting link S is capacitor C charging when closed; It is phase winding afterflow that diode VD is used in the time that MOS switching tube disconnects; R is current-limiting resistance, prevents the excessive device that burns of winding current.In detection line module, R ₁, R ₂measure for phase winding terminal voltage, resistance is larger, makes R ₁, R ₂line current is less, so that negligible on All other routes impact; R _cfor current sampling resistor.Rotor end is fixed a rotary encoder, converts position, rotor angle to electric impulse signal in real time.If by R ₂, R _cboth end voltage is designated as respectively u ₂, u _c, formula (1) can be rewritten as:

$\mathrm{\ψ}\left(k\right)=\mathrm{\ψ}(k-1)+0.5T[\frac{R_1+R_2}{R_2}{u}_2\left(k\right)-r\frac{u_c\left(k\right)}{R_C}+\frac{R_1+R_2}{R_2}{u}_2(k-1)-r\frac{u_c(k-1)}{R_C}]---\left(2\right)$

DSP, as the core of whole flux measurement system, carries out periodic sampling to all kinds of electric signal in real time, calculates real-time magnetic linkage data, is then transferred to host computer and carries out magnetic linkage modeling.

For making summary of the invention introduction below more simple and clear, clear, spy makes following parameter-definition:

Ω static data collection,

ψ _k1the real-time estimate result of the flux linkage model having established,

ψ _k0the real-time sampling result of magnetic linkage data,

δ _kthe relative error absolute value of flux linkage model real-time estimate result,

The ε flux linkage model Relative Error absolute value upper limit.

On the basis of switched reluctance machines static data collection Ω, limit of utilization learning machine successfully carries out switched reluctance motor flux linkage off-line modeling.When still switch word reluctance motor being run up with above-mentioned flux linkage characteristic detection system, carry out magnetic linkage data real-time sampling, carry out magnetic linkage comparing with the flux linkage model establishing, draw the Relative Error absolute value δ that flux linkage model is real-time _k:

${\mathrm{\δ}}_k=\left|\frac{{\mathrm{\ψ}}_{k1}-{\mathrm{\ψ}}_{k0}}{{\mathrm{\ψ}}_{k0}}\right|\×100\%---\left(3\right)$

Experimental results show that: the relative error absolute value that the flux linkage model that extreme learning machine is set up predicts the outcome can reach 0.001 order of magnitude, consider that flux linkage model that setting the is larger relative error absolute value upper limit that predicts the outcome can reduce host computer hardware resource is taken and the requirement to flux linkage model precision of prediction of actual machine operation occasion, can set flux linkage model Relative Error absolute value higher limit ε=1%～5% simultaneously.Work as δ _kwhen > ε, just the magnetic linkage sampled data in this moment is added in Ω, re-start magnetic linkage modeling, repeat constantly this operation, until δ _k≤ ε, stops the modeling again online of magnetic linkage, and flux linkage model under using the switched reluctance motor flux linkage model that now re-establishes as this working environment, to improve the adaptive faculty of flux linkage model to different operating environment.

The invention has the advantages that:

1. extreme learning machine learning process is without mass data, the model prediction precision of setting up is high, and learning process is rapid, therefore, operating limit learning machine carries out the magnetic linkage modeling of switched reluctance machines, has avoided using traditional neural network (as BP neural network, RBF neural network), support vector machine to carry out the existing problem of magnetic linkage modeling.

2. set up online switched reluctance motor flux linkage modeling method, fully taken into account the situation that may run in motor practical engineering application, can describe exactly the switched reluctance motor flux linkage characteristic in real work, there is very strong transplantability.

3. the illustrated online magnetic linkage modeling method of switched reluctance machines of this patent realizes liking entity motor, is different from the existing magnetic linkage modeling method based on the Realization of Simulation of part, has higher application.

Brief description of the drawings

Fig. 1 is the structural representation of the switched reluctance motor flux linkage characteristic detection system taking DSP as core;

Fig. 2 is detection line module;

In figure: 1, driver circuit module; 2, detection line module; 3, switched reluctance machines; 4, DSP; 5, host computer.

Embodiment

The discrete integration formula of switched reluctance machines winding magnetic linkage is as follows:

ψ(k)＝ψ(k-1)+0.5T[u(k)-ri(k)+u(k-1)-ri(k-1)] (1)

In formula (1),

ψ is phase winding magnetic linkage value,

U is phase winding terminal voltage,

T is the sampling period,

I is phase current,

R is phase winding internal resistance,

K is sampled point sequence number.

Switched reluctance motor flux linkage model is set up required magnetic linkage data acquisition, is based on now having delivered disclosed switched reluctance motor flux linkage characteristic detection system, and it adopts step voltage to measure voltage and the electric current of phase winding, and core is DSP, as shown in Figure 1.In driver circuit module 1,220V alternating voltage obtains needed DC voltage after the easy rectification device rectification of transformer pressure-reducing, four diodes; For maintaining voltage constant and preventing power supply and the vibration of winding formation lc circuit, the electrochemical capacitor C of a larger capacitance in parallel after power supply, disconnecting link S is capacitor C charging when closed; It is phase winding afterflow that diode VD is used in the time that MOS switching tube disconnects; R is current-limiting resistance, prevents the excessive device that burns of winding current.In detection line module 2, R ₁, R ₂measure for phase winding terminal voltage, resistance is larger, makes R ₁, R ₂line current is less, so that negligible on All other routes impact; R _cfor current sampling resistor.Rotor end is fixed a rotary encoder, converts position, rotor angle to electric impulse signal in real time.If by R ₂, R _cboth end voltage is designated as respectively u ₂, u _c, as shown in Figure 2, formula (1) can be rewritten as:

$\mathrm{\ψ}\left(k\right)=\mathrm{\ψ}(k-1)+0.5T[\frac{R_1+R_2}{R_2}{u}_2\left(k\right)-r\frac{u_c\left(k\right)}{R_C}+\frac{R_1+R_2}{R_2}{u}_2(k-1)-r\frac{u_c(k-1)}{R_C}]---\left(2\right)$

DSP, as the core of whole flux measurement system, carries out periodic sampling to all kinds of electric signal in real time, calculates real-time magnetic linkage data, is then transferred to host computer and carries out magnetic linkage modeling.

For making summary of the invention introduction below more simple and clear, clear, spy makes following parameter-definition:

Ω is static data collection,

ψ _k1for the real-time estimate result of the flux linkage model that established,

ψ _k0for the real-time sampling result of magnetic linkage data,

δ _kfor the relative error absolute value of flux linkage model real-time estimate result,

ε is the flux linkage model Relative Error absolute value upper limit.

On the basis of switched reluctance machines static data collection Ω, limit of utilization learning machine successfully carries out switched reluctance motor flux linkage off-line modeling.When still switch word reluctance motor being run up with above-mentioned flux linkage characteristic detection system, carry out magnetic linkage data real-time sampling, carry out magnetic linkage comparing with the flux linkage model establishing, draw the Relative Error absolute value δ that flux linkage model is real-time _k:

${\mathrm{\δ}}_k=\left|\frac{{\mathrm{\ψ}}_{k1}-{\mathrm{\ψ}}_{k0}}{{\mathrm{\ψ}}_{k0}}\right|\×100\%---\left(3\right)$

Experimental results show that: the relative error absolute value that the flux linkage model that extreme learning machine is set up predicts the outcome can reach 0.001 order of magnitude, consider that flux linkage model that setting the is larger relative error absolute value upper limit that predicts the outcome can reduce host computer hardware resource is taken and the requirement to flux linkage model precision of prediction of actual machine operation occasion, can set flux linkage model Relative Error absolute value higher limit ε=1%～5% simultaneously.Work as δ _kwhen > ε, just the magnetic linkage sampled data in this moment is added in Ω, re-start magnetic linkage modeling, repeat constantly this operation, until δ _k≤ ε, stops the modeling again online of magnetic linkage, and flux linkage model under using the switched reluctance motor flux linkage model that now re-establishes as this working environment, to improve the adaptive faculty of flux linkage model to different operating environment.

Embodiment of the present invention are specifically divided into following 4 steps:

1. the static data that uses the switched reluctance motor flux linkage characteristic detection system collection switched reluctance machines 3 taking DSP as core, obtains static data collection, and the learning training method off-line of limit of utilization learning machine is set up the flux linkage model of switched reluctance machines 3.

2., when switched reluctance machines 3 runs up, gather the real-time magnetic linkage data ψ of switched reluctance machines 3 _k0, the magnetic linkage data ψ predicting with the flux linkage model of the switched reluctance machines 3 establishing _k1contrast, according to formula

calculate the real-time Relative Error absolute value δ of flux linkage model _k.

3. set the real-time Relative Error absolute value δ of flux linkage model _khigher limit ε, ε gets a fixed value in 1%～5% scope.

4. by Relative Error absolute value δ real-time flux linkage model _kcompare with the higher limit ε setting, work as δ _kwhen > ε, add static data to concentrate the magnetic linkage data in this moment, re-establish flux linkage model; Repeat constantly this operation, until δ _k≤ ε, stops the modeling again online of magnetic linkage, and flux linkage model under using the switched reluctance motor flux linkage model that now re-establishes as this working environment.

Claims (1)

1. the switched reluctance motor flux linkage line modeling method based on extreme learning machine, is characterized in that comprising the steps:

1) the switched reluctance motor flux linkage characteristic detection system of use taking DSP as core gathers the static data of switched reluctance machines (3), obtain static data collection, the learning training method off-line of limit of utilization learning machine is set up the flux linkage model of switched reluctance machines (3);

2), when switched reluctance machines (3) runs up, gather the real-time magnetic linkage data ψ of switched reluctance machines (3) _k0, the magnetic linkage data ψ predicting with the flux linkage model of the switched reluctance machines establishing (3) _k1contrast, according to formula

calculate the real-time Relative Error absolute value δ of flux linkage model _k;

3) set the real-time Relative Error absolute value δ of flux linkage model _khigher limit ε, ε gets a fixed value in 1%～5% scope;

4) by Relative Error absolute value δ real-time flux linkage model _kcompare with the higher limit ε setting, work as δ _kwhen > ε, add static data to concentrate the magnetic linkage data in this moment, re-establish flux linkage model; Repeat constantly this operation, until δ _k≤ ε, stops the modeling again online of magnetic linkage, and flux linkage model under using the switched reluctance motor flux linkage model that now re-establishes as this working environment.

Images