CN103001567A - Multi-mode inverse model identification method for speed regulating system of six-phase permanent magnet synchronous linear motor - Google Patents

Abstract

The invention discloses a multi-mode inverse model identification method for a speed regulating system of a six-phase permanent magnet synchronous linear motor. The multi-mode inverse model identification method includes that the speed regulating system of the six-phase permanent magnet synchronous linear motor is formed by process for controlling a d shaft component i<d> of primary current of the six-phase permanent magnet synchronous linear motor so that the i<d> is equal to zero, and the reversibility of a mathematical model of the speed regulating system is verified; and input space and output space of the system are divided according to a multi-mode idea, nonlinear mapping in local inverse models of the system is fit by a least square support vector machine (LSSVM), the various local inverse models are outputted in a weighted manner, and an offline initial inverse model is built. Weights of the local inverse models are adaptively adjusted by an improved recursive least square (RLS) algorithm according to deviation of output of the inverse model and input of the system, and the initial inverse model is corrected in an online manner, so that the initial inverse model can adapt to change of an object, the identification precision is improved, and the convergence rate is increased.

Description

The multimode identification of inverse model method of six-phase permanent-magnet linear synchronous motor governing system

Technical field

The present invention relates to a kind of multimode identification of inverse model method of six-phase permanent-magnet linear synchronous motor governing system, be applicable to the technical field of electric drive control.

Background technology

The multi-phase permanent linear synchronous motor has plurality of advantages as a kind of New-type electric machine: 1) high-power with the low-voltage device realization, be particularly suitable for obtaining high pressure but the powerful occasion of needs output; 2) significantly reduce motor harmonic loss, reduce motor torque ripple, improve system effectiveness and stability.Therefore the multi-phase permanent linear synchronous motor is widely used in logistics system, industrial equipment, information and automated system, traffic and the many-side such as civilian, military.

The multi-phase permanent linear synchronous motor is a multivariable close coupling non linear system, and owing to adopt direct drive mode, load disturbance, the disturbance of ripple thrust, frictional force disturbance and other uncertain disturbances can directly act on motor, have a strong impact on the control precision of motor.Traditional PID control strategy is difficult to realize the control requirement of its increasingly stringent.Neural network inverse control is as a kind of feedback linearization method of intelligence, the nonlinear function that combines the height of the clear physics conception of method of inverse, understandable characteristics and neural net approaches the advantage strong with adaptive ability, the decoupling zero of non linear system is controlled have preferably effect.Be applied to the multi-phase permanent linear synchronous motor, can realize the high accuracy control of motor.

The Neural network inverse control method is that the system's inversion model that builds is connected with original system, thereby the non linear system of complexity is approximately pseudo-linear system, and the design additional controller is realized its high performance control.Therefore, the structure of system's inversion model is most important, and increasing document is studied it, as: neural network generalized inverse model building method, but the limit of the pseudo-linear hybrid system of arbitrary disposition gained; The online adjustment strategy of system neural network inversion model is constructed first RBF neural net off-line training and is obtained initial parameter, adjusts online its network weight by the improved BP algorithm again, has strengthened the generalization ability of neural net.

Yet present research all is the static non linear mappings that approach system's inversion model with a neural net, considers that system condition is complicated, only is not enough to accurate descriptive system at the inversion model of different conditions with a network; And, although literature research the has been arranged on-line identification method of system's inversion model, because it needs to adjust in real time each weights and the threshold value of neural net inside, performance difficulty.

Summary of the invention

The present invention is take the six-phase permanent-magnet linear synchronous motor as object, purpose provides a kind of multimode identification of inverse model method that is applicable to multi-phase permanent linear synchronous motor governing system, so that adopt the multi-phase permanent linear synchronous motor of inverse system control to realize more high performance control.

The technical solution used in the present invention is: the multimode identification of inverse model method of six-phase permanent-magnet linear synchronous motor governing system comprises the steps:

1) adopts primary current d axle component i _dThe six-phase permanent-magnet linear synchronous motor governing system of=0 control mode operation, with square wave at random as input data u _qEncourage, and to input data u _q, output data ω _rSample; To the sampled data smothing filtering, ask for derivative and periodic sampling, obtain sample data

2) according to motor speed ω _rLow speed, middling speed and high velocity between, sample data is divided into three different subspaces; With sample data

As input, { u _qAs output, the LSSVM off-line training is carried out in each subspace; Select Gaussian function as the LSSVM kernel function, adopt cross-validation method commonly used to obtain suitable regularization parameter γ and nuclear parameter σ, obtain respectively three LSSVM;

3) LSSVM that each is trained adds respectively two integrators and consists of local LSSVM inverse system; To local LSSVM inverse system weights initialize, with addition after three local inverse system weightings, obtain the output of the initial inversion model of six-phase permanent-magnet linear synchronous motor governing system respectively

4) utilize variable forgetting factor function lambda (k) to improving in the RLS algorithm, the RLS algorithm is improved:

$\left\{\begin{array}{c}y\left(k\right)=w^T(k-1)x\left(k\right)\\ e\left(k\right)=d\left(k\right)-y\left(k\right)\\ k\left(k\right)=\frac{p(k-1)x\left(k\right)}{\mathrm{\&lambda;}\left(k\right)+x^T\left(k\right)p(k-1)x\left(k\right)}\\ p\left(k\right)=\frac{1}{\mathrm{\&lambda;}\left(k\right)}[p(k-1)-x^T\left(k\right)k\left(k\right)p(k-1)]\\ w\left(k\right)=w(k-1)+k\left(k\right)e\left(k\right)\\ \mathrm{\&lambda;}(k+1)={\mathrm{\&lambda;}}_{\min }+(1-{\mathrm{\&lambda;}}_{\min })\exp (-\mathrm{int}\left(\mathrm{\&rho;}\right|e\left(k\right)\left|\right))\end{array}\right.$

In the formula, y is the output of modified RLS algorithm; X=[x ₁, L, x _m] ^TM input for modified RLS algorithm; W=[w ₁, L, w _m] ^TWeights for m input; D is desired output; E is desired output and the actual deviation of exporting; K is gain; P is the inverse matrix of k; Int () is bracket function; ρ is responsive gain; λ _MinMinimum value for forgetting factor.When deviation e is tending towards infinite, λ=λ _Min, so that algorithm can comparatively fast be followed the tracks of the local trend of non-stationary signal; E is tending towards at 0 o'clock, and λ=1 is to reduce parameter estimating error.

5) local LSSVM inverse system is regarded respectively as the input x=[x of modified RLS algorithm ₁, x ₂, x ₃] ^T, the weights of corresponding local LSSVM inverse system are regarded respectively the weight w of modified RLS algorithm=[w as ₁, w ₂, w ₃] ^T, the output of six-phase permanent-magnet linear synchronous motor governing system inversion model

Be the output y of modified RLS algorithm;

6) according to six-phase permanent-magnet linear synchronous motor governing system input data u _qExport with six-phase permanent-magnet linear synchronous motor governing system inversion model Deviation e, adjust online the weights of local LSSVM inverse system, so that obtain more accurately the output of six-phase permanent-magnet linear synchronous motor governing system inversion model after the weighting of local LSSVM inverse system

The invention has the beneficial effects as follows:

1) with the inner a plurality of weights of weights adjustment replacement neural net of LSSVM partial model and the adjustment of threshold value, greatly reduces the online difficulty of adjusting of inversion model, improved the real-time of modeling; And LSSVM identification precision and computational efficiency are higher, are easy to engineering and use.

2) improved RLS Weight number adaptively adjustment algorithm can be adjusted forgetting factor in real time according to the size of Identification Errors, taken into account Model Distinguish initial convergence speed, the time become follow-up control and identification precision.

3) multi-model LSSVM adopts and decomposes synthetic idea about modeling, only needs the input and output sample set of system, has preferably local fit precision and generalization ability, can obtain gratifying identification of inverse model effect.

Description of drawings

The LSSVM51 that it is good that Fig. 1 is off-line training adds the schematic diagram that two integrators consist of local LSSVM inverse system 61.

Fig. 2 is the schematic diagram of the initial inversion model of addition constructing system after local LSSVM inverse system 61, local LSSVM inverse system 62 and local LSSVM inverse system 63 weightings.

Fig. 3 is the multimode inversion model on-line identification schematic diagram of six-phase permanent-magnet linear synchronous motor governing system 1.

Embodiment

Below in conjunction with accompanying drawing the present invention is described in detail, implementation divides following 8 steps:

During 1, by secondary flux linkage orientation, the six-phase permanent-magnet linear synchronous motor adopts primary current d axle component i _d=0 control method, and combine with load and to consist of six-phase permanent-magnet linear synchronous motor governing system 1.System simplification is the non linear system of the single output of single input, and input variable is primary voltage q axle component u _q2, output variable is motor speed ω _r4.Select primary current q axle component i _q3 and motor speed ω _r4 is quantity of state, and the Mathematical Modeling that obtains is the second order state equation.

2, adopt the Interactor algorithm to output variable ω _r4 ask local derviation, until the aobvious input variable u that contains _q2.According to the inverse system principle as can be known, the inverse system of six-phase permanent-magnet linear synchronous motor governing system 1 correspondence exists, and can determine that the input variable of its inverse system is motor speed ω _r4 second dervative Output variable is primary voltage q axle component u _q2, and expression formula is $u_q=\mathrm{\&phi;}({\mathrm{\&omega;}}_r,{\stackrel{\&CenterDot;}{\mathrm{\&omega;}}}_r,{\stackrel{\&CenterDot;\&CenterDot;}{\mathrm{\&omega;}}}_r).$

3, six-phase permanent-magnet linear synchronous motor governing system 1 is adopted primary current d axle component i _d=0 control mode operation, with the at random square wave of realistic range of operation as input variable u _q2 come excitation system, and to input data u _q2, output data ω _r4 sample.To the sampled data smothing filtering, ask for derivative and periodic sampling, obtain sample data

4, according to motor speed ω _rBetween 4 low speed, middling speed and high velocity, sample data is divided into three different subspaces.With

As input, { u _qAs output, the LSSVM off-line training is carried out in each subspace.Select Gaussian function as the LSSVM kernel function, adopt cross-validation method commonly used to obtain suitable regularization parameter γ and nuclear parameter σ, obtain respectively LSSVM51, LSSVM52 and LSSVM53.

5, the LSSVM51 that off-line training is good, LSSVM52 and LSSVM53 add two integrators, consist of respectively local LSSVM inverse system 61(as shown in Figure 1), local LSSVM inverse system 62 and local LSSVM inverse system 63.The structure of the initial inversion model of system as shown in Figure 2, respectively to the weight w of local LSSVM inverse system 61, local LSSVM inverse system 62 and local LSSVM inverse system 63 ₁71, w ₂72 and w ₃73 initializes with addition after three local inverse system weightings, obtain the output of the initial inversion model of system

6, utilize variable forgetting factor function lambda (k) to improving in the RLS algorithm, the mathematic(al) representation of the RLS algorithm that is improved:

$\left\{\begin{array}{c}y\left(k\right)=w^T(k-1)x\left(k\right)\\ e\left(k\right)=d\left(k\right)-y\left(k\right)\\ k\left(k\right)=\frac{p(k-1)x\left(k\right)}{\mathrm{\&lambda;}\left(k\right)+x^T\left(k\right)p(k-1)x\left(k\right)}\\ p\left(k\right)=\frac{1}{\mathrm{\&lambda;}\left(k\right)}[p(k-1)-x^T\left(k\right)k\left(k\right)p(k-1)]\\ w\left(k\right)=w(k-1)+k\left(k\right)e\left(k\right)\\ \mathrm{\&lambda;}(k+1)={\mathrm{\&lambda;}}_{\min }+(1-{\mathrm{\&lambda;}}_{\min })\exp (-\mathrm{int}\left(\mathrm{\&rho;}\right|e\left(k\right)\left|\right))\end{array}\right.$

In the formula, y is the output of modified RLS algorithm; X=[x ₁, L, x _m] ^TM input for modified RLS algorithm; W=[w ₁, L, w _m] ^TWeights for m input; D is desired output; E is desired output and the actual deviation of exporting; K is gain; P is the inverse matrix of k; Int () is bracket function; ρ is responsive gain; λ _MinMinimum value for forgetting factor.When deviation e is tending towards infinite, λ=λ _Min, so that algorithm can comparatively fast be followed the tracks of the local trend of non-stationary signal; E is tending towards at 0 o'clock, and λ=1 is to reduce parameter estimating error.

7, the input x=[x that local LSSVM inverse system 61, local LSSVM inverse system 62 and local LSSVM inverse system 63 is regarded respectively as modified RLS algorithm ₁, x ₂, x ₃] ^T, corresponding weight w ₁71, w ₂72 and w ₃73 weight w of regarding respectively modified RLS algorithm as=[w ₁, w ₂, w ₃] ^T, the output of system's inversion model

Be the output y of modified RLS algorithm.

8, as shown in Figure 3, according to the input u of system _q2 with the output of system inversion model

Deviation e9, adjust online weight w ₁71, w ₂72 and w ₃73, so that obtain more accurately system's inversion model output after local LSSVM inverse system 61, local LSSVM inverse system 62 and local LSSVM inverse system 63 weightings

When the present invention pressed secondary flux linkage orientation, the six-phase permanent-magnet linear synchronous motor adopted primary current d axle component i _d=0 control method consists of six-phase permanent-magnet linear synchronous motor governing system, and checking governing system Mathematical Modeling is reversible.According to multi-model thought, system's input-output space is divided, utilize the Nonlinear Mapping in the local inversion model of least square method supporting vector machine (LSSVM) match system, with each local inversion model weighting output, set up the initial inversion model of off-line.According to the deviation of inversion model output with system's input, adjust local inversion model weights by improved recurrence least square (RLS) algorithm self adaptation, to initial inversion model on-line correction, make the variation of its adaption object, improve identification precision and convergence rate.

Claims (1)

1. the multimode identification of inverse model method of a six-phase permanent-magnet linear synchronous motor governing system is characterized in that comprising the steps:

1) adopts primary current d axle component i _d=0 control mode operation six-phase permanent-magnet linear synchronous motor governing system, with square wave at random as input data u _qEncourage, and to input data u _q, output data ω _rSample; To the sampled data smothing filtering, ask for derivative and periodic sampling, obtain sample data

2) according to motor speed ω _rLow speed, middling speed and high velocity between, sample data is divided into three different subspaces; With sample data

As input, { u _qAs output, the LSSVM off-line training is carried out in each subspace; Select Gaussian function as the LSSVM kernel function, adopt cross-validation method commonly used to obtain suitable regularization parameter γ and nuclear parameter σ, obtain respectively three LSSVM that train;

3) LSSVM that each is trained adds respectively two integrators and consists of local LSSVM inverse system; To local LSSVM inverse system weights initialize, with addition after three local inverse system weightings, obtain the output of the initial inversion model of six-phase permanent-magnet linear synchronous motor governing system respectively

4) utilize variable forgetting factor function lambda (k) to improving in the RLS algorithm, the RLS algorithm is improved:

$\left\{\begin{array}{c}y\left(k\right)=w^T(k-1)x\left(k\right)\\ e\left(k\right)=d\left(k\right)-y\left(k\right)\\ k\left(k\right)=\frac{p(k-1)x\left(k\right)}{\mathrm{\&lambda;}\left(k\right)+x^T\left(k\right)p(k-1)x\left(k\right)}\\ p\left(k\right)=\frac{1}{\mathrm{\&lambda;}\left(k\right)}[p(k-1)-x^T\left(k\right)k\left(k\right)p(k-1)]\\ w\left(k\right)=w(k-1)+k\left(k\right)e\left(k\right)\\ \mathrm{\&lambda;}(k+1)={\mathrm{\&lambda;}}_{\min }+(1-{\mathrm{\&lambda;}}_{\min })\exp (-\mathrm{int}\left(\mathrm{\&rho;}\right|e\left(k\right)\left|\right))\end{array}\right.$

In the formula, y is the output of modified RLS algorithm; X=[x ₁, x ₂, x ₃] ^TM input for modified RLS algorithm; W=[w ₁, w ₂, w ₃] ^TM the weights of inputting for modified RLS algorithm; D is desired output; E is desired output and the actual deviation of exporting; K is gain; P is the inverse matrix of k; Int () is bracket function; ρ is responsive gain; λ _MinMinimum value for forgetting factor: when deviation e is tending towards infinite, λ=λ _MinE is tending towards at 0 o'clock, λ=1;

5) local LSSVM inverse system is regarded respectively as the input x=[x of modified RLS algorithm ₁, x ₂, x ₃] ^T, the weights of corresponding local LSSVM inverse system are regarded respectively the weight w of modified RLS algorithm=[w as ₁, w ₂, w ₃] ^T, the output of six-phase permanent-magnet linear synchronous motor governing system inversion model

Be the output y of modified RLS algorithm;

6) according to six-phase permanent-magnet linear synchronous motor governing system input data u _qExport with six-phase permanent-magnet linear synchronous motor governing system inversion model Deviation e, adjust online the weights of local LSSVM inverse system, so that obtain more accurately the output of six-phase permanent-magnet linear synchronous motor governing system inversion model after the weighting of local LSSVM inverse system

Images