CN106487297A - A kind of PMSM parameter identification method based on covariance matching Unscented kalman filtering algorithm - Google Patents

Abstract

Embodiments provide a kind of permagnetic synchronous motor parameter identification method based on covariance matching Unscented kalman filtering algorithm, the method includes：Based on permagnetic synchronous motor vector controlled mathematical model, set up state observation coefficient matrix equation；Based on state observation coefficient matrix equation and default primary quantity, realize the nonlinear transformation of state observer equation according to UT conversion, realize the initialization of Unscented kalman filtering；Criterion is carried out based on covariance matching criterion method to Unscented kalman filtering divergent trend, according to criterion result, updates and revise the noise covariance in Unscented kalman filtering iterative process and measurement noise covariance；Based on updating and revised covariance data, obtain parameter of electric machine identification information.The technical scheme providing according to embodiments of the present invention, can reduce parameter of electric machine estimation difference, the stability of parameter identification in the case of lifting load disturbance.

Description

A kind of PMSM parameter identification based on covariance matching Unscented kalman filtering algorithm Method

Technical field

The present invention relates to a kind of method of motion control field, specifically one kind is using based on covariance matching no mark karr Graceful filtering algorithm realizes PMSM parameter identification.

Background technology

Permagnetic synchronous motor (PMSM) has the technology such as structure is simple, power density is big, fault rate is low and operational efficiency is high Advantage, is widely used in the fields such as joint servo driving, industrial robot.The realization of high performance PMSM control system depends on The accurate parameter of electric machine, in the case of load disturbance, electric motor load torque, speed, stator current etc. change, and lead to motor Control performance declines, and system reliability and system stability performance decline.Permanent magnet synchronous electric motor servo-controlled system is typically equipped with A series of sensors, such as velocity sensor etc., but the presence of sensor, reduce the mechanical robustness of system, increased and are System cost.Meanwhile, it is not easy to install sensor under the application scenario such as space industry.Therefore, load disturbance situation to be obtained Under high-performance PMSM joint servo control system, be necessary in running the parameter of electric machine change accurately being recognized, To control.

System identification algorithm is widely applied in the parameter identification of PMSM.Mainly there are off-line identification and on-line identification two Class.Conventional offline identification method has no load test and stall experiment, single-phase injection AC and DC current measurement, nerve net The methods such as network, genetic algorithm, ant group algorithm.Offline parameter identification needs to carry out before motor operation, the simply motor obtaining Initial operational parameter is it is impossible to obtain the change of relevant parameter in motor operation.On-line identification can obtain motor real-time parameter and become Change, at present, the method for parameter of electric machine identification mainly has method of least square (RLS), model reference adaptive method, spreading kalman filter Ripple method etc..

To measure and to estimate the mean square deviation being worth to as performance indications, Parameter Estimation Precision is subject to measurement noise to RLS algorithm Impact larger；Recursive model reference adaptive algorithm ensures identification system with Lyapunov's theory or Popov hyperstable theory Stability, is equally also subject to systematic survey effect of noise, and identification precision is not ideal enough；EKF (Extended Kalman Filter, EKF) it is popularization in nonlinear system for the Kalman filtering, but EKF introduces linearized stability, for non- The higher system of linear intensity is easily caused the decline of filter effect.In order to overcome EKF not enough, Julier etc. proposes a kind of non-thread Property filtering method, i.e. Unscented kalman filtering (Unscented Kalman Filter, UKF), using the approximate non-thread of the method for sampling Property distribution solving Nonlinear Filtering Problem.The method need not calculate Jacobi matrix, it is to avoid EKF local linearization and cause But wave filter unstable., when identifying motor model is under load disturbance, during steady-state operation there is uncertain saltus step in parameter When, either EKF or UKF makes Parameter Estimation Precision decline in some instances it may even be possible to lead to filter divergence.

Content of the invention

In view of this, one kind is exactly combined with UKF by the purpose of the present invention based on covariance matching criterion method, negative In the case of carrying disturbance, this algorithm, in filtering, by judging to filtering divergence trend, enters to the situation of filtering divergence Row judges and suppresses, and improves the robustness to Parameter Perturbation for the UKF algorithm, and is applied in PMSM parameter identification.

Embodiments provide a kind of permagnetic synchronous motor based on covariance matching Unscented kalman filtering algorithm Parameter identification method, as shown in Figure 1, including：

Based on permagnetic synchronous motor vector controlled mathematical model, set up state observation coefficient matrix equation；

Based on state observation coefficient matrix equation and default primary quantity, realize the non-of state observer equation according to UT conversion Linear transformation, realizes the initialization of Unscented kalman filtering；

Criterion is carried out based on covariance matching criterion method to Unscented kalman filtering divergent trend, according to criterion result, Update and revise the noise covariance in Unscented kalman filtering iterative process and measurement noise covariance；

Based on updating and revised covariance data, obtain parameter of electric machine identification information.

In the above-mentioned methods, based on permagnetic synchronous motor vector controlled mathematical model, set up state observation coefficient matrix side Journey, including：

On the premise of ignoring harmonic wave, vortex and magnetic hystersis loss, based on PMSM mathematical model, its coefficient matrix can be write：

Wherein R_sIt is stator resistance；i_α、i_βFor rotor current vector α, beta -axis component；ω is stator rotating speed；L_sFor stator electricity Sense；λ is magnetic linkage；θ is rotor position angle；f₁,f₂,f₃,f₄Represent coefficient equation respectively；

State equation is chosen for：

WhereinFor subsequent time renewal amount, f is state-transition matrix, and v (t) is controlled quentity controlled variable, and δ (t) is measurement noise, B For controlled quentity controlled variable coefficient matrix；

Observational equation is chosen for：

y(t_k)=h [x (t_k)]+μ(t_k) (3)

Wherein, μ (t_k) it is observation noise, y (t_k) it is t_kMoment observed quantity exports, and h is controlled quentity controlled variable coefficient current matrix；

Choose quantity of state：

X=[i_αi_βω θ] (4)

Observed quantity：

Y=[i_αi_β] (5)

Controlled quentity controlled variable：

V=[v_αv_β] (6)

Wherein v_α, v_βIt is illustrated respectively in the voltage in α, β axle for the voltage；

In said method, based on state observation coefficient matrix equation and default primary quantity, realize State Viewpoint according to UT conversion Survey the nonlinear transformation of device equation, realize the initialization of Unscented kalman filtering, including：

The present invention according to the servo-control system model of motor, for the general step of UKF parameter identification, including：

UKF is the method that a class deterministic sampling strategy approaches nonlinear Distribution, and the core of Unscented filtering is exactly Carry out the state of nonlinear model and the recursion of error covariance and renewal by a kind of nonlinear transformation, its algorithm mainly walks Suddenly as follows：

Step 1：Initialization

Step 2：UT conversion and calculating Sigma point

If n n-dimensional random variable nStochastic variable y is a certain nonlinear function y=f (x) of x, and the statistics of x is special Property isGuaranteeing to sample, average and covariance areAnd P_xOn the premise of, select one group 2n+1 weighting point set { χ_iCome The distribution of this stochastic variable approximate.

In formula,For scale parameter, can play and adjust acting on and reducing forecast error of High Order Moment, its value Change with the difference of x distribution.It is matrix (n+k) P_xRoot mean square i-th row or i-th row.W_i ^mWith W_i ^cIt is respectively average and the weighted value of covariance；Pairing approximation Gauss distribution discrete point point set (Sigma point set), carries out non-linear change Change (after the state equation of nonlinear system and measurement equation propagation), the point set { y after being converted_i}:

y_i=f (x_i), i=1,2 ..., 2n+1 (16)

Their average and variance are weighted processing, are applied to each the Sigma point sampled, the statistics obtaining y is special Property

Initialization step mainly comprises, based on motor characteristic initial parameter value, to arrange the initial Value Operations of quantity of state observed quantity.

In said method, criterion is carried out to Unscented kalman filtering divergent trend based on covariance matching criterion method, according to According to criterion result, update and revise the noise covariance in Unscented kalman filtering iterative process and measurement noise covariance, bag Include：

During aforesaid introduction, during Unscented kalman filtering application, need six parameters.Comprise original state x₀, just Beginning covariance P₀, process noise covariance Q_w, measurement noise covariance Q_vAnd non-loss transformation parameter alpha and β.With filtering data When more, x₀And P₀Impact to filter result can be ignored.Parameter alpha and β only affect High Order Nonlinear System and estimate, to filtering The correctness estimated and stability influence are less.As applied in linear system in Kalman filtering, work as Q_w、Q_vWith true When inside system, value matching degree is too low, UKF filtering algorithm can occur to degenerate and even dissipate.It is true that filtering to Unscented kalman Wave process has a major impact, and in calculating process, works as Q_w、Q_vToo little, uncertain disturbances of system and estimate that by mistake deviation can progressively increase Plus, finally deviate actual value；Work as Q_w、Q_vWhen value is too big, eventually result in filtering divergence.Therefore, in the case of load disturbance, more Newly with correction Q_w、Q_vIt is the effective ways lifting adaptivity in UKF identification process.

In disturbed deviation calculation procedure, approximately newly breath covariance is defined as：

The size that wherein N estimates for window, k is positive integer, v_iCease for new,For coefficient transposed matrix, v_iDetermined by following formula：

Wherein y_iRepresent true measurement,Represent true measurement average, self adaptation UKF optimisation criteria is to reduce to deviate Variance V_kValue,

Wherein tr () is to ask mark computing,For new breath covariance updated value；

In filtering divergence trend criterion step, a kind of method based on covariance matching criterion is carried out to filtering divergence trend Judgement has the following steps：

By in formula (22), E () is calculating of averaging, and S is an adjustability coefficients being previously set (S >=1),For residual error sequence Row,For residual sequence transposed matrix,For observed quantity average, as shown in formula (23), if this formula is false, revise P_k|k-1, Become

λ in formula_kFor adaptive weighted coefficient, χ_k|k-1For the data based on k-1 moment recursion for the k moment, χ_i,k|k-1For the i time The recursion data of point,For the state expected value in k moment, W_i ^cFor weighted mean square difference design factor, P_k|k-1Shape for the k moment State noise data, Q_k-1For k-1 moment measurement noise data；

In measurement noise correction step, when meeting the condition of convergence, using no correction Unscented kalman filtering algorithm；And sentence When determining filtering and being unsatisfactory for the condition of convergence, calculate weighted mean square difference design factor W using formula (13), (14)_i ^c, based on formula (24), profit Use adaptive weighted coefficient lambda_kGo to P_k|k-1It is modified, increase the effect of Current observation amount, so that filtering convergence；Its In, λ_kDetermined by following formula：

Wherein:

Wherein, 0 ＜ ρ ＜ 1 is attenuation quotient, and R is measured value, can improve the fast tracking capability of wave filter further, and ρ leads to Often 0.8 ± 0.15, its value is bigger, then the ratio shared by information before the k moment is less, and the impact of noise disturbance becomes for value Little, there is preferable ability of tracking for mutation noise；Meanwhile, filtering steady statue under, to gradual individually and mutation status according to Old have ability of tracking.

In said method, based on updating and revised covariance data, obtain parameter of electric machine identification information, including：

Forecast updating mainly comprises：

χ_k|k-1=f (χ_k-1) (28)

γ_i,k|k-1=h (χ_i,k|k-1) (31)

WhereinFor the state quantity data average based on k-1 moment recursion for the k moment, wherein γ_i,k|k-1Control for the k moment Parameter information,For the observed quantity data mean value based on k-1 moment recursion for the k moment；

Measurement updaue mainly comprises：

WhereinFor the state quantity data average based on k-1 moment recursion for the k moment, P_yyUpdate for observed quantity state, P_xy Update for quantity of state state, K_kKalman gain matrix, P_k|kUpdate for k moment covariance, P_k|k- 1 represents k-1 moment covariance Update.

As can be seen from the above technical solutions, the embodiment of the present invention has the advantages that：

In the technical scheme of the embodiment of the present invention, in the case of load disturbance, algorithm is according to measurement information to filtering divergence Trend is judged, and introduces decay factor suppression and dissipate, and has both ensured positive semidefinite and the orthotropicity of noise variance matrix, and simultaneously right Prediction variance matrix is monitored.Can efficiently solve and cause system noise to lead to filtering unstability to be asked due to load disturbance Topic, has certain lifting on identification precision.

Brief description

Fig. 1 is present invention method step enforcement figure；

Fig. 2 is flow chart during PMSM parameter identification for the method provided in an embodiment of the present invention；

Fig. 3 is permagnetic synchronous motor vector controlled block diagram in the embodiment of the present invention；

Fig. 4 is permagnetic synchronous motor Speed Identification comparison diagram in the embodiment of the present invention；

Fig. 5 is self adaptation UKF Identification Errors in the embodiment of the present invention

Fig. 6 is UKF Identification Errors in the embodiment of the present invention；

Fig. 7 is self adaptation UKF identification result under embodiment of the present invention load disturbance；

Specific embodiment

In order to be better understood from technical scheme, below in conjunction with the accompanying drawings the embodiment of the present invention is retouched in detail State.

It will be appreciated that described embodiment is only a part of embodiment of the present invention, rather than whole embodiment.Base Embodiment in the present invention, those of ordinary skill in the art obtained under the premise of not making creative work all its Its embodiment, broadly falls into the scope of protection of the invention.

The embodiment of the present invention provides a kind of PMSM parameter identification method based on covariance matching UKF algorithm, refer to attached Fig. 2, it is the schematic flow sheet of the provided method of the embodiment of the present invention, as shown in Fig. 2 the method comprises the following steps：

Based on permagnetic synchronous motor SVPWM mathematical model, state observer algorithm equation is set up according to its coefficient matrix.

Specifically, first, for permagnetic synchronous motor SVPWM control system, set up following state space equation model, such as Shown in accompanying drawing 2, in order to describe the control process of permagnetic synchronous motor：

Permanent-magnetic synchronous motor rotor mechanical motion equation is

In formula：ω_rFor rotor machinery angular velocity, f is frequency, θ_rFor rotor-position, F is coefficient of friction, and J is rotary inertia, T_emFor electromagnetic torque, T_LFor load torque, it is separated into the form of difference equation：

In formula：

In formulaT_SFor the systematic sampling time, determine that state equation is：

Wherein R_sIt is stator resistance；i_βFor rotor current vector α, beta -axis component；ω stator is rotating speed；L_sFor stator electricity Sense；θ is rotor position angle；

According to described state space equation, the nonlinear transformation based on Unscented kalman filtering is carrying out nonlinear model The recursion of state and error covariance and renewal；

Specifically, algorithm flow is as shown in Figure 3.In concrete application, provide and specifically join under a kind of particular motor model Count and use step, motor arrange parameter is as shown in table 1：

Table 1 permagnetic synchronous motor model machine parameter

Two kinds of motor original states are：ω=0rad/s, θ=0.5, i_a=i_b=0A, it is accurate recognition that initial parameter is chosen Key, conventional method is trial and error procedure adjustment, by Multi simulation running or test selecting the initial value of best performance.Trial and error procedure Checking shows, state variable initial value X (0) is with the inconsistent impact to state estimation of actual original state less, secondary with filtering The increase of number, the impact of initial value can be gradually reduced.Simultaneity factor debugging shows, state variable estimation difference covariance matrix P (0) Little to systematic steady state and dynamic process time effects, the change of its numerical value has certain impact to rotating speed amplitude.Test bar herein Under part, initialization of variable state X (0)=[0,0,0,0] determines needs initial condition in identification process：Estimation difference covariance square Battle array P (0)=[0.02,0.01,0.01,0.01], noise covariance matrix Q=[0.008,0.002,0.005,0.005] and survey The covariance matrix R=[0.8,0.2] of amount noise.

According to described state space equation and Initial Information, the method based on covariance matching criterion is to filtering divergence trend Judged, noise covariance and measurement noise association side when updating and revise Unscented kalman filtering application according to criterion result Difference, realizes forecast updating, measurement updaue, obtains parameter of electric machine identification information；

In concrete application, under a kind of particular motor model of above-mentioned offer, design parameter and using step as follows,

When meeting the condition of convergence, just using filtering algorithm noted earlier；And when judging that filtering is unsatisfactory for the condition of convergence, meter Calculation obtains adaptive weighted coefficient.Then gone to P with it_k|k-1It is modified, increase the effect of Current observation amount, so that filter Ripple is restrained, wherein, λ_kDetermined by following formula：

Wherein:

Wherein, 0 ＜ ρ ＜ 1 is attenuation quotient, can improve the fast tracking capability of wave filter further, the usual value of ρ exists 0.9 about, its value is bigger, then the ratio shared by information before the k moment is less, and the impact of noise disturbance diminishes, for prominent Becoming noise has preferable ability of tracking；Meanwhile, under filtering steady statue, to gradual individually and mutation status, still there is tracking Ability.Fig. 5 and Fig. 6 is self adaptation UKF Identification Errors and the contrast of UKF Identification Errors in the embodiment of the present invention；Fig. 7 is that the present invention is real Apply self adaptation UKF identification result under a load disturbance.

Claims (5)

1. a kind of permagnetic synchronous motor (PMSM) parameter identification method based on covariance matching Unscented kalman filtering algorithm, its It is characterised by, methods described includes：

Based on permagnetic synchronous motor vector controlled mathematical model, set up state observation coefficient matrix equation；

Based on state observation coefficient matrix equation and default primary quantity, realize the non-linear of state observer equation according to UT conversion Conversion, realizes the initialization of Unscented kalman filtering；

Criterion is carried out based on covariance matching criterion method to Unscented kalman filtering divergent trend, according to criterion result, updates With the noise covariance revised in Unscented kalman filtering iterative process and measurement noise covariance；

Based on updating and revised covariance data, obtain parameter of electric machine identification information.

2. method according to claim 1 it is characterised in that described based on permagnetic synchronous motor vector controlled mathematical modulo Type, sets up state observation coefficient matrix equation, operates including following：

Based on permagnetic synchronous motor mathematical model, its coefficient matrix can be write：

$f\left(x\right)=\left[\begin{array}{c}{f}_1\\ f_2\\ f_3\\ f_4\end{array}\right]=\left[\begin{array}{c}-\frac{R_s{i}_{\mathrm{\α}}}{L_s}+\frac{\mathrm{\ω}\mathrm{\λ}}{L_s}\·sin\mathrm{\θ}\\ -\frac{R_s{i}_{\mathrm{\β}}}{L_s}+\frac{\mathrm{\ω}\mathrm{\λ}}{L_s}\·cos\mathrm{\θ}\\ 0\\ \mathrm{\ω}\end{array}\right]---\left(1\right)$

Wherein R_sIt is stator resistance；i_α、i_βFor rotor current vector in α axle, beta -axis component；ω is stator rotating speed；L_sFor stator electricity Sense；λ is magnetic linkage；θ is rotor position angle；f₁,f₂,f₃,f₄Represent coefficient equation respectively.

3. method according to claim 1 is it is characterised in that described based on state observation coefficient matrix equation and default Primary quantity, realizes the nonlinear transformation of state observer equation, realizes the initialization of Unscented kalman filtering according to UT conversion, bag Include following operation：

Key step mainly comprises：Choose setting original state, UT conversion, three steps of initialization；

Based on the general step of algorithm, the quantity of state during parameter identification is chosen for：

State equation is chosen for：

$\stackrel{\·}{x}\left(t\right)=f\[x\left(t\right)\]+B\·v\left(t\right)+\mathrm{\δ}\left(t\right)---\left(2\right)$

WhereinFor subsequent time renewal amount, x (t) is t measurement amount, and f is state-transition matrix, and v (t) is controlled quentity controlled variable, δ T () is measurement noise, B is controlled quentity controlled variable coefficient matrix；

Observational equation is chosen for：

y(t_k)=h [x (t_k)]+μ(t_k) (3)

Wherein, μ (t_k) it is observation noise, y (t_k) it is t_kMoment observed quantity exports, and h is controlled quentity controlled variable coefficient current matrix；

Choose quantity of state：

X=[i_αi_βω θ] (4)

Observed quantity：

Y=[i_αi_β] (5)

Controlled quentity controlled variable：

V=[v_αv_β] (6)

$B=\left[\begin{array}{cc}\frac{1}{L_S}& 0\\ 0& \frac{1}{L_s}\\ 0& 0\\ 0& 0\end{array}\right]---\left(7\right)$

$h\left(x\right)=\left[\begin{array}{c}{i}_{\mathrm{\α}}\\ i_{\mathrm{\β}}\end{array}\right]---\left(8\right)$

Wherein v_α, v_βRepresent the component of voltage in α axle, β axle for the voltage respectively；

UT conversion mainly comprises：Using the point (Sigma points) of given average under the conditions of original state and variance, by this A little points pass through nonlinear transformation one by one, then approach the probability density characteristic of nonlinear transformation again with the point that these obtain；

Initialization step mainly comprises, based on motor characteristic initial parameter value, to arrange the initial Value Operations of quantity of state observed quantity.

4. method according to claim 1, the method based on covariance matching criterion judges to filtering divergence trend, Update according to criterion result and revise the noise covariance in Unscented kalman filtering iterative process and measurement noise covariance, main Comprise disturbed deviation calculating, filtering divergence trend criterion, three steps of measurement noise correction, specifically include following operation：

In disturbed deviation calculation procedure, approximately newly breath covariance is defined as：

$S_k=\frac{1}{N}\underset{i=k-N}{\overset{k-1}{\Σ}}{v}_i{v}_i^T---\left(9\right)$

The size that wherein N estimates for window, k is positive integer, v_iCease for new,For coefficient transposed matrix, v_iDetermined by following formula：

$v_i=y_i-{\stackrel{\‾}{y}}_{i-1}---\left(10\right)$

Wherein y_iRepresent true measurement,Represent true measurement average, self adaptation UKF optimisation criteria is to reduce disturbed deviation V_kValue,

Wherein tr () is to ask mark computing,For new breath covariance updated value；

In filtering divergence trend criterion step, a kind of method based on covariance matching criterion judges to filtering divergence trend Have the following steps：

${\tilde{y}}_k^T\tilde{y}\≤S\×tr\[E\left(\tilde{y}{\tilde{y}}_k^T\right)\]---\left(12\right)$

${\tilde{y}}_k=y_k-h\left({\stackrel{\‾}{x}}_{k|k-1}\right)---\left(13\right)$

By in formula (13), E () is calculating of averaging, and S is an adjustability coefficients being previously set (S >=1),For residual sequence, For residual sequence transposed matrix,For observed quantity average, as shown in formula (13), if this formula is false, revise P_k|k-1So as to become For

$P_{k|k-1}={\mathrm{\λ}}_k\underset{i=0}{\overset{2n}{\mathrm{\Σ}}}{W}_i^c\[{\mathrm{\χ}}_{i,k|k-1}-{\stackrel{\‾}{x}}_{k|k-1}\]\×{\[{\mathrm{\χ}}_{i,k|k-1}-{\stackrel{\‾}{x}}_{k|k-1}\]}^T+Q_{k-1}---\left(14\right)$

λ in formula_kFor adaptive weighted coefficient, χ_k|k-1For the data based on k-1 moment recursion for the k moment, χ_i,k|k-1For i time point Recursion data,For the state expected value in k moment, W_i ^cFor weighted mean square difference design factor, P_k|k-1State for the k moment is made an uproar Sound data, Q_k-1For k-1 moment measurement noise data；

In measurement noise correction step, when meeting the condition of convergence, using no correction Unscented kalman filtering algorithm；And judge to filter When ripple is unsatisfactory for the condition of convergence, using formula (14), using adaptive weighted coefficient to P_k|k-1It is modified, increase Current observation amount Effect so that filtering convergence；Wherein, λ_kDetermined by following formula：

${\mathrm{\&lambda;}}_k=\left\{\begin{array}{cc}{\mathrm{\&lambda;}}_0,& {\mathrm{\&lambda;}}_0\&GreaterEqual;1\\ 1,& {\mathrm{\&lambda;}}_0<1\end{array}\right.---\left(15\right)$

Wherein:

${\mathrm{\&lambda;}}_0=\frac{tr{(C_{0,k}-R)}^T}{tr(\underset{i=0}{\overset{2n}{\mathrm{\&Sigma;}}}{W}_i^c\&lsqb;{\mathrm{\&gamma;}}_{i,k|k-1}-{\stackrel{\&OverBar;}{y}}_{k|k-1}\&rsqb;\&times;{\&lsqb;{\mathrm{\&gamma;}}_{i,k|k-1}-{\stackrel{\&OverBar;}{y}}_{k|k-1}\&rsqb;}^T)}---\left(16\right)$

$C_{0,k}=\left\{\begin{array}{cc}{\tilde{y}}_k^T\tilde{y},& k=1\\ \frac{{\mathrm{\&rho;C}}_{0,k}+y{\tilde{y}}_k^T}{1+\mathrm{\&rho;}},& k>1\end{array}\right.---\left(17\right)$

Wherein, 0 ＜ ρ ＜ 1 is attenuation quotient, and R is measured value, can improve the fast tracking capability of wave filter further, ρ generally takes 0.8 ± 0.15, its value is bigger, then the ratio shared by information before the k moment is less, and the impact of noise disturbance diminishes for value, There is preferable ability of tracking for mutation noise；Meanwhile, under filtering steady statue, gradual individually and mutation status are remained unchanged and has There is ability of tracking.

5. method according to claim 1, based on updating and revised covariance data, obtains parameter of electric machine identification letter Breath, operates including following：

Forecast updating mainly comprises：

χ_k|k-1=f (χ_k-1) (18)

${\stackrel{\&OverBar;}{\mathrm{\&chi;}}}_{k|k-1}=\underset{i=0}{\overset{2n}{\mathrm{\&Sigma;}}}{W}_i^c{\mathrm{\&chi;}}_{i,k|k-1}---\left(19\right)$

$P_{k|k-1}=\underset{i=0}{\overset{2n}{\mathrm{\&Sigma;}}}{W}_i^c\&lsqb;{\mathrm{\&chi;}}_{i,k|k-1}-{\stackrel{\&OverBar;}{x}}_{k|k-1}\&rsqb;\&times;{\&lsqb;{\mathrm{\&chi;}}_{i,k|k-1}-{\stackrel{\&OverBar;}{x}}_{k|k-1}\&rsqb;}^T+Q---\left(20\right)$

γ_i,k|k-1=h (χ_i,k|k-1) (21)

${\stackrel{\&OverBar;}{y}}_{k|k-1}=\underset{i=0}{\overset{2n}{\mathrm{\&Sigma;}}}{W}_i^c{\mathrm{\&gamma;}}_{i,k|k-1}---\left(22\right)$

WhereinFor the state quantity data average based on k-1 moment recursion for the k moment, wherein γ_i,k|k-1For k moment control parameter Information,For the observed quantity data mean value based on k-1 moment recursion for the k moment；

Measurement updaue mainly comprises：

$P_{yy}=\underset{i=0}{\overset{2n}{\mathrm{\&Sigma;}}}{W}_i^c\&lsqb;{\mathrm{\&gamma;}}_{i,k|k-1}-{\stackrel{\&OverBar;}{y}}_{k|k-1}\&rsqb;\&times;{\&lsqb;{\mathrm{\&gamma;}}_{i,k|k-1}-{\stackrel{\&OverBar;}{y}}_{k|k-1}\&rsqb;}^T+R---\left(23\right)$

$P_{xy}=\underset{i=0}{\overset{2n}{\&Sigma;}}{W}_i^c\&lsqb;{\mathrm{\&chi;}}_{i,k|k-1}-{\stackrel{\&OverBar;}{x}}_{k|k-1}\&rsqb;\&times;{\&lsqb;{\mathrm{\&chi;}}_{i,k|k-1}-{\stackrel{\&OverBar;}{x}}_{k|k-1}\&rsqb;}^T---\left(24\right)$

${\stackrel{\&OverBar;}{x}}_{k|k}={\stackrel{\&OverBar;}{x}}_{k|k-1}+K_k(y_k-{\stackrel{\&OverBar;}{y}}_{k|k-1})---\left(25\right)$

$P_{k|k}=P_{k|k-1}-K_k{P}_{yy}{K}_k^T---\left(26\right)$

WhereinFor the state quantity data average based on k-1 moment recursion for the k moment, P_yyUpdate for observed quantity state, P_xyFor shape State amount state updates, K_kKalman gain matrix, P_k|kUpdate for k moment covariance, P_k|k-1Represent k-1 moment covariance to update.