CN101552588B - An electric motor control device and an electric motor control system - Google Patents

Abstract

The invention provides an electric motor control device and sets best gain reduction for the input signals provided to the recognition operation. The invention comprises zero holding components (26, 27) for the periodical sampling and maintaining, delay components (28, 29) for the first sampling periodical delaying for the output signals of the zero holding components, multipliers (30, 31) for thereduction of the output signals of the delay components, a regression vector generation part (33) for the regression vector generation by adopting the output signals of the multipliers as the input, a discrete system parameter estimation part (34) for the discrete estimation of the parameter vectors by adopting the regression vector as the input and an automatic gain control part (32) which adopts the output signals of the zero holding components (26, 27) as the input signals and outputs the gain set values of the multipliers.

Description

Control device of electric motor and motor control system

Technical field

The present invention relates to control device of electric motor and motor control system that a kind of identification parameter carries out speed control.

Background technology

As the control device of electric motor that is used for industrial machinery (semiconductor-fabricating device, work mechanism, emission forming machine) etc.; Known a kind of technology is identification total the moment of inertia, a viscous friction coefficient and constant disturb mechanical system parameter such as torque outward in the real-world operation process, and this recognition result is used for the automatic setting of controller parameter such as speed control.For example, in patent documentation 1, disclose a technology, the time series data of each motor torque, acceleration and electromotor velocity when using when just changeing with counter-rotating through recurrent least square method, carries out the identification of above-mentioned mechanical system parameter.

In addition; Automatic setting technology as more general Control Parameter; The technology of non-patent literature 1 described self tuning regulator for example; Through using recurrent least square method, discern the Plant transfer function to the input signal of giving controlling object with from the time series data of the output signal of controlling object.In addition, as the algorithm that carries out above-mentioned those technological employed recurrent least square method computings accurately with the short processor of operational character length, known have a UD decomposition algorithm (UD decomposes filter), and it has detailed record in non-patent literature 2.In addition, the UD decomposition algorithm is used for the technology of identification mechanism, in patent documentation 2, obtains open.

Patent documentation 1: the spy opens the 2005-168166 communique

Patent documentation 2: the spy opens clear 64-84303 communique

Non-patent literature 1: the grand work of Suzuki " adaptive control ", CORONA society, calendar year 2001, the 1st chapter

Non-patent literature 2: meal state ocean two works " Adaptive Signal Processing algorithm ", training wind shop,, the 4th chapter in 2000

In the technology of UD decomposition algorithm, under the excessive situation of the absolute value of the input signal that gives recognition operation, might overflow in the recurrent least square method computing.Especially because of the expressible numerical value dynamic range of fixed point arithmetic is very low, under the situation of carrying out recurrent least square method, overflows and be easy to take place.

To this problem; General solution is; On the input part of recurrent least square method computing, put into attenuator (greater than 0 less than 1 decay gain), but have following problem: when the decay gain is excessive (attenuation rate is little), because of overflowing the identification error increase that causes; When the decay gain is too small (attenuation rate is big), the identification error that causes because of the input signal round-off error increases.In addition, as other solution, though also there is pair operation result to use amplitude limiter to avoid the straightforward procedure of overflowing, but can produce the saturated arithmetic eror that causes, cause the low problem of accuracy of identification.

Summary of the invention

The present invention proposes in order to address these problems a little, and problem provides a kind of control device of electric motor and motor control system, regardless of the numerical value of the motor rotary speed value of detecting, can both suppress to overflow the generation of computing.

In order to solve above-mentioned problem, control device of electric motor of the present invention,

Control device of electric motor possesses: speed control, and according to rotary speed value of detecting of motor and the deviation between the motor rotary speed command value, output torque current command value; Current controller according to said torque current command value and as the deviation between torque current value of detecting of the torque current composition of the drive current in the said motor, generates the control signal of the said drive current of control; With automatic tuning portion; Use motor torque value and said these both sides of the rotary speed value of detecting of said motor, adjust the Control Parameter of said speed control automatically, said control device of electric motor; Control said motor; Make said rotary speed value of detecting near said motor rotary speed command value, it is characterized in that, said automatic tuning portion comprises: the 1st low pass filter of importing said motor rotary speed value of detecting; Import the 2nd low pass filter of said motor torque value; Use the output signal of said the 1st low pass filter and the output signal of said the 2nd low pass filter, the parameter recognition device of discrete ground identification parameter vector; Parameter converter section with the discrete system/continuous system of the Control Parameter that said parameter vector is converted to said speed control; Said parameter recognition device comprises: the 1st zeroth order keeps assembly, with some cycles the output signal of said the 1st low pass filter is taken a sample; The 2nd zeroth order keeps assembly, with some cycles the output signal of said the 2nd low pass filter is taken a sample; Regression vector generation portion uses said the 1st zeroth order to keep the output signal of assembly and the output signal that said the 2nd zeroth order keeps assembly, generates regression vector; 2 multipliers keep assembly, said the 2nd zeroth order to keep assembly and their inhibit signal to said the 1st zeroth order, also have any 2 signals in the output signal of said regression vector generation portion to decay; Discrete system parameter estimation portion is according to any 2 signals in the output signal of the output signal of said 2 multipliers and said regression vector generation portion, the said parameter vector of discrete ground computing; With the automatic gain control part, use the 1st zeroth order to keep the output signal of assembly and the output signal that said the 2nd zeroth order keeps assembly, the gain of said 2 multipliers of computing.

Thus, the size of signal is just decayed by 2 multipliers, so, even if dynamic range is restricted as fixed point arithmetic, also can when carrying out recognition operation, suppress to overflow the generation of computing.In addition, owing to can set best decay gain to the rotary speed value of detecting that offers recognition operation, so, can improve accuracy of identification.Like this, just can carry out the high-precision automatic setting of Control Parameter, realize simpler high performance of control.

According to the present invention,, can both suppress to overflow the generation of computing regardless of the numerical value of the motor rotary speed value of detecting.

Description of drawings

Fig. 1 is the pie graph of the motor control system of an embodiment of the invention.

Fig. 2 is the pie graph of the parameter recognition device of the 1st execution mode of the present invention.

Fig. 3 is the pie graph of the automatic gain control part working signal generator AGI of the 1st execution mode of the present invention.

Fig. 4 is the pie graph of the automatic gain control part AGCI of the 1st execution mode of the present invention.

Fig. 5 is the pie graph of the regression vector generation portion of the 1st execution mode of the present invention.

Fig. 6 is the pie graph of discrete system parameter estimation portion.

Fig. 7 is the pie graph of fallout predictor.

Fig. 8 is the pie graph of parameter vector arithmetic unit.

Fig. 9 is the pie graph of f vector operation device.

Figure 10 is the pie graph of g vector operation device.

Figure 11 is the pie graph of α μ arithmetic unit.

Figure 12 is the pie graph of U matrix operation device.

Figure 13 is the pie graph of D matrix operation device.

Figure 14 is the pie graph of gain vector arithmetic unit.

Figure 15 is the flow chart of the action of expression the 1st execution mode.

Figure 16 is the sketch map of the action waveforms of expression the 1st execution mode.

Figure 17 is the movement oscillogram of expression the 1st execution mode and comparative example.

Figure 18 is the pie graph of the parameter recognition device of the 2nd execution mode of the present invention.

Figure 19 is the pie graph of the automatic gain control part AGC of the 2nd execution mode of the present invention.

Figure 20 is the flow chart of the action of expression the 2nd execution mode.

Among the figure: 1-motor, 2-load (driven object load), 3-connection shaft, 4-electric power converter; The 5-velocity calculator, 6-current detector, 7,10, the 124-subtracter, 8-current controller; The 9-speed control, 12-storage gain, 13-integrator, 14,16,93,146,165,168,185,226,231,256, the 259-adder; 11,15, the 17-proportional gain, 18,275, the 279-gain, 19,350-parameter recognition device, 21, the 22-low pass filter; 23-parameter recognition device working signal generator (IDA), 24-automatic gain control part working signal generator (AGI), 25-discrete system/continuous system parameter converter section, 26, the 27-zeroth order keeps assembly; 28,29,71,73,75,86,113,166,169,257,276, the 280-1 Delay Element of taking a sample, 30,31,91,92,145,164,167,184,205,207,225,230,255,258,274, the 278-multiplier, 32-automatic gain control part (AGCI), 33-regression vector generation portion; 34-discrete system parameter estimation portion, 35-discrete system parameter estimation portion's working signal generator (PVAI), 36-first-order lag filter, 37-automatic tuning portion; The 38-position detector, 39-position detector, 70-parameter recognition device working signal Id_act, 72-OR logic element; The 74-XOR logic element, 76-automatic gain control part working signal Agc_act, 84,85,374,375-signed magnitude arithmetic(al) device, the 90-minimum value is relatively selected element; 89,96,228,273,277,294, the 295-divider, 94-extracting operation device, 114,229-sign-inverted device, 123-fallout predictor; 125-parameter vector arithmetic unit, 127-f vector operation device, 128-g vector operation device, 129-α μ arithmetic unit; 130-U matrix operation device, 131-D matrix operation device, 132-gain vector arithmetic unit; 115,142,143,163,170,183,186,203,204,209,236,237,235,254,260,272,281,293,296,373-vectorization mark, 100-control device of electric motor, 200-motor control system; 206,208,227, the 232-amplitude limiter, 224-forgetting factor λ, Iq ^*-torque current command value, the Iq-torque current value of detecting, I^q-torque current presumed value, Ie-current deviation, τ _M-motor torque, τ _SMotor torque instruction before the-viscous friction compensation, τ M ^*The instruction of-motor torque, τ _d^-viscous friction compensating torque, ω _M-motor rotary speed the value of detecting, ω _M ^*-electromotor velocity command value, ω e-electromotor velocity deviation, total the moment of inertia presumed value of J^-mechanical system, D^-viscous friction coefficient presumed value, τ _MFMotor torque after the-filtration, ω _MFMotor rotary speed value of the detecting after-filtration, τ _MS(N)-and motor torque sampling value after filtering, ω _MS(N)-and electromotor velocity sampling value after filtering, τ _MS(N-1)-and motor torque 1 sampling length of delay after filtering, ω _MS(N-1)-and motor torque 1 sampling length of delay after filtering, K _g, K _α, K _{G α}-decay gain, τ _M(N)-and decay back motor torque, ω _M(N)-and decay back electromotor velocity, Φ (N)-regression vector, θ ^ (N-1)-parameter vector is presumed value last time, Agc_act-automatic gain control part working signal, Id_act-parameter recognition device working signal, Pv_actI-discrete system parameter estimation portion working signal.

Embodiment

(basic principle)

Even, also can suppress to overflow the generation of computing, the input signal of giving recognition operation is inserted the variable attenuation gain of variable set point in order to carry out the recurrent least square method computing.With the value of variable attenuation gain, be set at the maximum of overflowing generation (minimal attenuation rate) that can be suppressed in the recurrent least square method computing again.Like this, just can take into account following two aspects: make the identification error that causes because of the input signal round-off error minimize and suppress to overflow the generation of computing.

At first, the algorithm of recognition operation is described, then, the set point calculation method that variable attenuation is gained describes.Suppose: be u, establish value from the output signal of identifying object when being y when establishing, can show the transfer function of the identifying object that will obtain with (1) formula toward the value of the input signal of identifying object.

(mathematical formulae 1)

$\frac{y}{u}=\frac{b_0+b_1{z}^{-1}+\&CenterDot;\&CenterDot;\&CenterDot;+b_{\mathrm{nb}}{z}^{-\mathrm{<figure-callout\; id="1"\; label="nb"\; filenames="H200810184719XE00011.png,H200810184719XE00021.png"\; state="\{\{state\}\}">nb</figure-callout>}}}{1+a_1{z}^{-1}+\&CenterDot;\&CenterDot;\&CenterDot;+a_{\mathrm{na}}{z}^{-\mathrm{na}}}---\left(1\right)$

At this moment, regression vector Φ ^*(N) with the definition of (2) formula, parameter vector θ ^ (N) is with the definition of (3) formula.

(mathematical formulae 2)

Φ ^*(N)：＝[-y(N-1)，…，-y(N-na)，u(N)，u(N-1)，…，u(N-nb)] ^T(2)

$\widehat{\mathrm{\&theta;}}\left(N\right):={[{\hat{a}}_1\left(N\right),\&CenterDot;\&CenterDot;\&CenterDot;,{\hat{a}}_{\mathrm{na}}\left(N\right),{\hat{b}}_0\left(N\right),{\hat{b}}_1\left(N\right),\&CenterDot;\&CenterDot;\&CenterDot;,{\hat{b}}_{\mathrm{nb}}\left(N\right)]}^T---\left(3\right)$

(2) Φ of formula ^*(N) be a kind of vector, inscape is for since 1 y and the current u before the nb sampling that are sampled to before the na sampling, and the θ ^ (N) of (3) formula is a kind of vector, and inscape is the coefficient recognition result of the N sampling time point of (1) formula.In addition, when establishing unknown parameter quantity when being p, they are the p dimensional vector.

Here, the predicated error ε of N sampling time point (N) is defined by (4) formula.

(mathematical formulae 3)

$\mathrm{\&epsiv;}\left(N\right):=y\left(N\right)-{\mathrm{\&Phi;}}^{*T}\left(N\right)\widehat{\mathrm{\&theta;}}(N-1)---\left(4\right)$

(4) ε of formula (N), be illustrated in the observed y of N sampling time point (N) pairing, with according to parameter vector θ ^ (N-1) and regression vector Φ in (N-1) sampling time point identification ^*Error between the predicted value of the y that (N) tries to achieve (N).At this moment, parameter vector θ ^ (N) is an initial value with θ ^ (0)=0 (null vector), through repeat the computing of (5) formula in each sampling period, converges to true value.This is the said content of non-patent literature 2 (4.57) formulas.

(mathematical formulae 4)

$\widehat{\mathrm{\&theta;}}\left(N\right)=\widehat{\mathrm{\&theta;}}(N-1)+k\left(N\right)\mathrm{\&epsiv;}\left(N\right)---\left(5\right)$

Below, quote non-patent literature 2, the computational methods based on the UD decomposition algorithm of the needed gain vector k of (5) formula computing (N) are described.The characteristic of UD decomposition algorithm is: upgrading the diagonal angle composition shown in (6) formula in each N sampling period is 1 ((the diagonal matrix D (N) of p * p) shown in the upper triangular matrix U (N) of p * p) and (7); Calculate gain vector k (N) simultaneously, the initial value of matrix U (N), D (N) is obtained by (8) formula, (9) formula respectively.In addition, known: the diagonal angle key element initial value γ of diagonal matrix D (N) set for more big on the occasion of, parameter vector θ ^ (N) restrains to true value more.

(mathematical formulae 5)

(mathematical formulae 6)

U (0)=I (unit matrix of I:p * p) (8)

D (0)=γ I (unit matrix of I:p * p, γ: bigger positive constant) (9)

In update algorithm, each N sampling period is all carried out the computing of (10) formula, (11) formula, SUB 1, uses formula (19) to obtain k (N) then.The computing of so-called SUB1, be with forgetting factor λ (1 below, near 1 positive number) for α 0 (N), then for j=1,2 ..., p carries out the computing of (13) formula, (14) formula, (15) formula, (16) formula and SUB2.The computing of so-called SUB2 is exactly to i=1, and 2 ..., j-1 carries out the computing of (17) formula, (18) formula.

(mathematical formulae 7)

f(N)：＝[f ₁(N)，f ₂(N)，…，f _p(N)] ^T＝U ^T(N-1)Φ ^*(N)(10)

(mathematical formulae 8)

g(N)：＝[g ₁(N)，g ₂(N)，…，g _p(N)] ^T＝D(N-1)f(N)(11)

(mathematical formulae 9)

α ₀(N)＝λ(12)

(mathematical formulae 10)

α _j(N)＝α _j-1(N)+f _j(N)g _j(N)(13)

(mathematical formulae 11)

${\mathrm{\&mu;}}_j\left(N\right)=-\frac{f_j\left(N\right)}{{\mathrm{\&alpha;}}_{j-1}\left(N\right)}---\left(14\right)$

(mathematical formulae 12)

${\left(D\left(N\right)\right)}_{\mathrm{jj}}=\frac{{\mathrm{\&alpha;}}_{j-1}}{\mathrm{\&lambda;}{\mathrm{\&alpha;}}_j\left(N\right)}{\left(D(N-1)\right)}_{\mathrm{jj}}---\left(15\right)$

(mathematical formulae 13)

v _j(N)＝g _j(N)(16)

(mathematical formulae 14)

(U(N)) _ij＝(U(N-1)) _ij+μ _j(N)v _i(N)(17)

(mathematical formulae 15)

v _i(N)＝v _i(N)+(U(N-1)) _ijv _j(N)(18)

(mathematical formulae 16)

$k\left(N\right)=\frac{1}{{\mathrm{\&alpha;}}_p\left(N\right)}\left[\begin{array}{c}{v}_1\left(N\right)\\ v_2\left(N\right)\\ v_3\left(N\right)\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ v_p\left(N\right)\end{array}\right]---\left(19\right)$

It more than is the detailed content of UD decomposition algorithm.According to this algorithm, the α vector that the recognition operation initial stage that the begins g vector sum of being calculated by (11) formula is calculated by (13) formula all is bigger value.For this reason, if carry out the UD decomposition algorithm with short operational character length and with fixed point arithmetic, then a side or the both sides in the g vector sum α vector will be overflowed (overflow the position), and the problem that parameter vector θ ^ (N) is not converged in true value takes place.Below, detail with regard to these contents.

The value of g vector sum α vector is more greatly with following situation much relations to be arranged at the recognition operation initial stage that begins, and promptly considers toward the convergence of true value and the bigger above-mentioned diagonal angle key element initial value γ that has set diagonal matrix D (N); And D (N) reduces basically gradually.The decrescence property of D (N) can be derived by (15) formula.(15) relation between formula performance D (N) and the D before 1 sampling period (N-1) is if the right alpha _J-1(N)/(λ α _j(N)), just we can say that D (N) reduces gradually always less than 1.Therefore, if pay close attention to the vectorial formula of calculating (13) of α: α just formula can be known _j(N) be α _{J- 1}(N) add the result that f vector and the vectorial j key element product value each other of g obtain.And the j key element of g vector is according to (11) formula, the value that jj key element and the vectorial j key element of f of diagonal matrix D (N) multiplied each other and obtain.According to above situation, can judge: the value that the right of (13) formula is the 2nd, be diagonal matrix D (N) jj key element and f vector the j key element square multiply each other obtain on the occasion of.Therefore, if also consider (12) formula, then (20) formula is generally set up.

α _j(N)≥α _j-1(N)≥λ＞0(20)

If considering the λ of (20) formula is near the value 1, then α _J-1(N)/(λ α _j(N)) be exactly value less than 1.Like this, D (N) just expression is a starting point with very large initial value γ, near a kind of behavior that 0, reduces gradually.Therefore, the α vector that the g vector sum that uses D (N) to calculate uses the g vector to calculate, the danger of overflowing (overflow the position) at the identification beginning initial stage is the highest.Otherwise if can avoid overflowing at the identification beginning initial stage, the danger meeting of overflowing thereafter diminishes with the minimizing of D (N).Therefore think, use regression vector Φ ^*(N) multiply by suitable decay gain and the new regression vector Φ (N) that obtains, can avoid discerning the beginning initial stage, promptly the vector of the g vector sum α on the primary sample time point (N=1) of UD decomposition algorithm overflows.Below, at first will avoid the formula of calculating and its basis of the decay of the overflowing gain Kg of g vector together to represent.

The j key element gj (N) of the g of N sampling time point vector, with D (N) matrix the (j, j) key element is designated as d _j(N) time, be (j, j) the j key element f of the f of key element and N sampling time point vector of D (N) matrix of (N-1) sampling time point _j(N) product can use (21) formula to represent.

(mathematical formulae 17)

g _j(N)＝d _j(N-1)f _j(N)(j＝1，2，····，p)(21)

On the other hand, at the vectorial j key element f of the f of N sampling time point _j(N), can use (22) formula to represent.

(mathematical formulae 18)

$f_j\left(N\right)={\mathrm{\&phi;}}_j\left(N\right)+\underset{i=1}{\overset{j-1}{\mathrm{\&Sigma;}}}{\mathrm{\&phi;}}_i\left(N\right)u_{\mathrm{ij}}(N-1)$ (j＝1，2，····，p)(22)

The j key element of the g vector of primary sample time point through obtaining at the mid-N=1 of (21) formula, becomes (23) formula.

(mathematical formulae 19)

g _j(1)＝d _j(0)f _j(1)(j＝1，2，····，p)(23)

On the other hand, the j key element of the f of primary sample time point vector through obtaining at the mid-N=1 of (23) formula, becomes (24) formula.

(mathematical formulae 20)

$f_j\left(1\right)={\mathrm{\&phi;}}_j\left(1\right)+\underset{i=1}{\overset{j-1}{\mathrm{\&Sigma;}}}{\mathrm{\&phi;}}_i\left(1\right)u_{\mathrm{ij}}\left(0\right)$ (j＝1，2，····，p)(24)

For under the situation of primary sample time point (N=1),, be uij (0)=0 like (25) formula according to the initial value of (8) formula.Therefore, the 2nd of (24) formula is zero later on, can use the performance of (26) formula.

(mathematical formulae 21)

(mathematical formulae 22)

f _j(1)＝φ _j(1)(26)

In addition, under the situation of N=1, owing to, establish D (0)=γ 1 (I is a unit matrix), so d normally as (12) formula _j(0)=γ.Therefore, for (23) formula,, can be deformed into (27) formula with (26) formula substitution.

(mathematical formulae 23)

g _j(1)＝d _j(0)f _j(1)＝γφ _j(1)(j＝1，2，····，p)(27)

Therefore, can be known by (27) formula: the j key element of the g vector of primary sample time point is the j key element Φ of the regression vector Φ (N) of primary sample time point _j(1) multiply by the value that the diagonal angle key element initial value γ of D matrix obtains.

Here, the maximum value key element of the g of primary sample time point vector | g (1) | _MaxDefine by (28) formula.

(mathematical formulae 24)

|g(1)| _max：＝max{|g ₁(1)|，|g ₂(1)|，…，|g _j(1)|}(j＝1，2，····，p)(28)

Next, (27) formula substitution (28) formula is out of shape, just obtains the relation of (29) formula.

(mathematical formulae 25)

|g(1)| _max＝max{|g ₁(1)|，|g ₂(1)|，…，|g _j(1)|}

＝max{γ|φ ₁(1)|，γ|φ ₂(1)|，…，γ|φ _j(1)|}(29)

＝γ·max{|φ ₁(1)|，|φ ₂(1)|，…，|φ _j(1)|}

Below, expression is with the maximum value key element of the g vector of the primary sample time point of (29) formula performance | g (1) | _MaxBe made as the method below the setting (setting for) less than the value that will overflow.The absolute value of all key elements that thus, just can g is vectorial all is suppressed at below the setting.

Here at first, the regression vector Φ (N) in (29) formula as (30) formula, is defined as original regression vector Φ ^*(N) decay the gain K _gThe value that doubly obtains.(wherein, establish 0＜K _g≤1)

(mathematical formulae 26)

φ(N)＝K _gΦ ^*(N)(30)

With (30) formula concern substitution (29) formula, just obtain (31) formula.Wherein, in (31) formula, Φ ^*J (1) carries out decay gain K _gRegression vector Φ doubly ^*(N) j key element.

(mathematical formulae 27)

|g(1)| _max＝γ·K _g·max{|Φ ₁ ^*(1)|，|Φ ₂ ^*(1)|，…，|Φ _j ^*(1)|}(31)

The meaning of above-mentioned (31) formula is: the maximum value key element of the g vector of primary sample time point | g (1) | _MaxRegression vector Φ by the primary sample time point ^*(N) maximum value key element decision.In addition, on (31) formula the right, owing to comprised decay gain K _gSo, can be through suitably setting K _g, will | g (1) | _MaxBe suppressed at below the setting.Therefore, incite somebody to action | g (1) | the regulation that max bore (upper limit) value is changed to ov_lim_g, provides the constraints of (32) formula.

|g(1)|max≤ov_lim_g(32)

With (31) formula substitution (32) formula, to decay gain K _gFind the solution, just obtain (33) formula.

(mathematical formulae 28)

$K_g\&le;\frac{\mathrm{ov}\_\lim \_g}{\mathrm{\&gamma;}\&CenterDot;\max \left\{\right|{\mathrm{\&Phi;}}_1^{*}\left(1\right)|,|{\mathrm{\&Phi;}}_2^{*}\left(1\right)|,\&CenterDot;\&CenterDot;\&CenterDot;,|{\mathrm{\&Phi;}}_p^{*}\left(1\right)\left|\right\}}---\left(33\right)$

Above-mentioned (33) formula is the decay gain K that can avoid the g vector to overflow _gSetting range.Usually, if less setting decay gain K _g, will produce a lot of regression vector information lose the position situation, make accuracy of identification low, so (34) formula is best K _gCalculate formula, the maximum of its (33) formula that is satisfied.

(mathematical formulae 29)

$K_g=\frac{\mathrm{ov}\_\lim \_g}{\mathrm{\&gamma;}\&CenterDot;\max \left\{\right|{\mathrm{\&Phi;}}_1^{*}\left(1\right)|,|{\mathrm{\&Phi;}}_2^{*}\left(1\right)|,\&CenterDot;\&CenterDot;\&CenterDot;,|{\mathrm{\&Phi;}}_p^{*}\left(1\right)\left|\right\}}---\left(34\right)$

In addition, because γ and ov_lim_g are constants, so through coming defining constant A with (35) formula, (34) formula just can use (36) formula to show.

(mathematical formulae 30)

$A=\frac{\mathrm{ov}\_\lim \_g}{\mathrm{\&gamma;}}---\left(35\right)$

(mathematical formulae 31)

$K_g=\frac{A}{\max \left\{\right|{\mathrm{\&Phi;}}_1^{*}\left(1\right)|,|{\mathrm{\&Phi;}}_2^{*}\left(1\right)|,\&CenterDot;\&CenterDot;\&CenterDot;,|{\mathrm{\&Phi;}}_p^{*}\left(1\right)\left|\right\}}---\left(36\right)$

Below, with the decay gain K that overflows of avoiding of α vector _αThe formula of calculating and its according to together representing.The j key element of the α vector of primary sample time point through putting N=1 and obtain in (13) formula, becomes (37) formula.

(mathematical formulae 32)

α _j(1)＝α _j-1(1)+f _j(1)g _j(1)(j＝1，2，····，p)(37)

With (26) formula, (27) formula substitution (37) formula, just become (38) formula.

(mathematical formulae 33)

α _j(1)＝α _j-1(1)+γφ _j(1) ²(j＝1，2，····，p)(38)

Utilize α ₀(N)=and the relation of λ, the gradually change formula of (38) formula is found the solution, just become (39) formula.

(mathematical formulae 34)

${\mathrm{\&alpha;}}_j\left(1\right)=\mathrm{\&lambda;}+\mathrm{\&gamma;}\underset{k=1}{\overset{j}{\mathrm{\&Sigma;}}}{\mathrm{\&phi;}}_k{\left(1\right)}^2$ (j＝1，2，····，p)(39)

Therefore, if with the maximum key element α (1) of the α vector of (40) formula definition primary sample time point _Max, according to (39) formula, α (1) _MaxJust become α _p(1), can use (41) formula to calculate.

(mathematical formulae 35)

α(1) _max：＝max{α ₁(1)，α ₂(1)，…，α _j(1)}(j＝1，2，····，p)(40)

(mathematical formulae 36)

$\mathrm{\&alpha;}{\left(1\right)}_{\max }={\mathrm{\&alpha;}}_p\left(1\right)=\mathrm{\&lambda;}+\mathrm{\&gamma;}\underset{k=1}{\overset{p}{\mathrm{\&Sigma;}}}{\mathrm{\&phi;}}_k{\left(1\right)}^2---\left(41\right)$

Below, expression is with the maximum key element α (1) of the α vector of the primary sample time point of (41) formula performance _MaxBe suppressed at the method below the setting (setting for) less than the value that will overflow.Thus, just can all key elements of α vector all be suppressed at below the setting.

Therefore, at first,,, be defined as original regression vector Φ as (42) formula with the regression vector Φ (N) in (41) formula ^*(N) value that gain K α doubly obtains that decays.(wherein, establish 0＜K _α≤1)

(mathematical formulae 37)

φ(N)＝K _αΦ ^*(N)(42)

With Φ ^*(N) j key element is made as Φ ^*During j (N), (43) formula derives through (41) formula, (42) formula.

(mathematical formulae 38)

$\mathrm{\&alpha;}{\left(1\right)}_{\max }={\mathrm{\&alpha;}}_p\left(1\right)=\mathrm{\&lambda;}+\mathrm{\&gamma;}{K}_{\mathrm{\&alpha;}}^2\underset{k=1}{\overset{p}{\mathrm{\&Sigma;}}}{\mathrm{\&Phi;}}_k^{*}{\left(1\right)}^2---\left(43\right)$

The meaning of above-mentioned (43) formula is: the maximum key element α (1) of the α vector of primary sample time point _MaxRegression vector Φ by the primary sample time point ^*Square decision of size (N) (vector length).In addition, on (43) formula the right, owing to comprised decay gain K _αSo, can be through suitably setting K _α, with α (1) _MaxBe suppressed at below the setting.Therefore, with α (1) _MaxThe regulation of being born (upper limit) value is changed to ov_lim_ α, provides the constraints of (44) formula.

α(1) _max≤ov_lim_a(44)

With (43) formula substitution (44) formula, to decay gain K _αFind the solution, just obtain the relation of (45) formula.

(mathematical formulae 39)

$K_{\mathrm{\&alpha;}}\&le;\sqrt{\frac{\mathrm{ov}\_\lim \_a-\mathrm{\&lambda;}}{\mathrm{\&gamma;}\&CenterDot;\underset{k=1}{\overset{p}{\mathrm{\&Sigma;}}}{\mathrm{\&Phi;}}_k^{*}{\left(1\right)}^2}}---\left(45\right)$

Above-mentioned (45) formula is the decay gain K that can avoid the α vector to overflow _αSetting range.Usually, if less setting decay gain K _α, will produce the situation that a lot of regression vector information are lost the position, make accuracy of identification low, so (46) formula is best K _αCalculate formula, the maximum of its (45) formula that is satisfied.

(mathematical formulae 40)

$K_{\mathrm{\&alpha;}}=\sqrt{\frac{\mathrm{ov}\_\lim \_a-\mathrm{\&lambda;}}{\mathrm{\&gamma;}\&CenterDot;\underset{k=1}{\overset{p}{\mathrm{\&Sigma;}}}{\mathrm{\&Phi;}}_k^{*}{\left(1\right)}^2}}---\left(46\right)$

In addition, because γ and ov_lim_ α are constants, so through coming defining constant B with (47) formula, (46) formula just can use (48) formula to show.

(mathematical formulae 41)

$B=\sqrt{\frac{\mathrm{ov}\_\lim \_a-\mathrm{\&lambda;}}{\mathrm{\&gamma;}}}---\left(47\right)$

(mathematical formulae 42)

$K_{\mathrm{\&alpha;}}=\frac{B}{\sqrt{\underset{k=1}{\overset{p}{\mathrm{\&Sigma;}}}{\mathrm{\&Phi;}}_k^{*}{\left(1\right)}^2}}---\left(48\right)$

So far, though be K as what can avoid decay gain expression that the g vector overflows _gAs the decay gain expression that can avoid the α vector to overflow is K _α, but in fact also need avoid both sides all to overflow.Therefore, regulation:, calculate the maximum attenuation gain K that takes into account (36) formula and (48) formula with (49) formula or (50) formula _{G α}, the regression vector Φ (N) that (51) formula is obtained is applied to the UD decomposition algorithm.

(mathematical formulae 43)

$K_{\mathrm{g\&alpha;}}=\min \{\frac{\mathrm{ov}\_\lim \_g}{\mathrm{\&gamma;}\&CenterDot;\max \left\{\right|{\mathrm{\&Phi;}}_1^{*}\left(1\right)|,|{\mathrm{\&Phi;}}_2^{*}\left(1\right)|,\&CenterDot;\&CenterDot;\&CenterDot;,|{\mathrm{\&Phi;}}_p^{*}\left(1\right)\left|\right\}},\sqrt{\frac{\mathrm{ov}\_\lim \_a-\mathrm{\&lambda;}}{\mathrm{\&gamma;}\&CenterDot;\underset{k=1}{\overset{p}{\mathrm{\&Sigma;}}}{\mathrm{\&Phi;}}_k^{*}{\left(1\right)}^2}}\}---\left(49\right)$

(mathematical formulae 44)

$K_{\mathrm{g\&alpha;}}=\min \{\frac{A}{\max \left\{\right|{\mathrm{\&Phi;}}_1^{*}\left(1\right)|,|{\mathrm{\&Phi;}}_2^{*}\left(1\right)|,\&CenterDot;\&CenterDot;\&CenterDot;,|{\mathrm{\&Phi;}}_p^{*}\left(1\right)\left|\right\}},\frac{B}{\sqrt{\underset{k=1}{\overset{p}{\mathrm{\&Sigma;}}}{\mathrm{\&Phi;}}_k^{*}{\left(1\right)}^2}}\}---\left(50\right)$

(mathematical formulae 45)

φ(N)＝K _gαΦ ^*(N)(51)

(the 1st execution mode)

Utilize the pie graph of Fig. 1, the motor control system as the 1st execution mode of the present invention is described.

In Fig. 1, motor control system 200 comprises: motor 1; By the load 2 (driven object load) of motor 1 driving; The connection shaft 3 that links motor 1 and load 2; The electric power converter 4 of drive motor 1; Be installed on the rotating shaft of motor 1 the position value of the detecting θ of the rotating shaft of output motor 1 _M Position detector 38; Detect the electric current detector 6 of the electric current of inflow motor 1; With control device of electric motor 100.

Control device of electric motor 100 comprises: velocity calculator 5; Current controller 8; First-order lag filter 36; Torque constant amplifier 20; Automatic tuning portion 37; Discrete system/continuous system parameter converter section 25; Speed control 9; With subtracter 7, their function is through computer and program realizations such as CPU.In addition, control device of electric motor 100 carries out dq vector conversion and controls the motor current that flow to motor 1, makes it form excitation axle composition and electrically go up and the torque axis composition of its quadrature.Below only describe,, record and narrate so omit because the value Id of excitation axle composition is set to zero usually with the value of the detecting Iq of torque current composition.

Motor 1 is a three-phase synchronous motor, has the moment of inertia J _MWith viscous friction coefficient D, torque T takes place.Load 2 is driven by motor 1 through connection shaft 3, has the moment of inertia J _LIn addition, the moment of inertia J of motor 1 _M, the moment of inertia of connection shaft 3 and the moment of inertia J of load 2 _LCombine, be called synthetic the moment of inertia J.Electric power converter 4 converts direct current to alternating current, drive motor 1.Position detector 38 is installed on the rotating shaft of motor 1, the position value of the detecting θ of the rotating shaft of output motor 1 _MElectric current detector 6 detects the torque current value of the detecting Iq that supplies with motor 1.Velocity calculator 5 is according to the position value of detecting θ _MTime change calculations motor rotary speed ω _MElectric current detector 6 detects the torque current value of the detecting Iq that supplies with motor 1.Subtracter 7 calculating torque current instruction value Iq ^*And the current deviation Ie between the torque current value of detecting Iq of supply motor 1.

Current controller 8 carries out pulse amplitude control (PWM control) according to current deviation Ie to electric power converter 4.First-order lag filter 36 input torque current instruction value Iq ^*, output torque current presumed value Iq^, its gain crossing angle frequency, the gain crossing angle frequency of setting the current control system that constitutes with current controller 8 for equates.Torque constant amplifier 20, input torque electric current presumed value Iq^ calculates motor torque value τ _M, set torque constant k with employed motor 1 _tThe value that equates.

Speed control 9 has the motor of the making rotary speed value of detecting ω _MApproach electromotor velocity command value ω _M ^*Function.Automatic tuning portion 37 is with motor torque value τ _MWith the motor rotary speed value of detecting ω _MTime series data as input value, the total the moment of inertia presumed value J^ and the viscous friction coefficient presumed value D^ of the mechanical system that output is made up of motor 1 and connection shaft 3 and driven object load 2.

Speed control 9 is according to motor rotary speed deviation ω _eOutput torque current command value Iq ^*The controller of proportional integral structure, it comprises: subtracter 10; Adder 14,16; Proportional gain 11,15,17,18; Storage gain 12; With integrator 13.Proportional gain 11, the ratio characteristic of realization speed control system.Storage gain 12, decision is with 13 pairs of motor rotary speeies of integrator deviation ω _eGain ω s when carrying out integration ²/ N _NAdder 14 is with the ratio characteristic of speed control system and the output addition of integrator 13.Proportional gain 15 is amplified the output of adder 14 with total the moment of inertia J of mechanical system, generates motor torque instruction τ before the viscous friction compensation _S Proportional gain 17 usefulness viscous friction coefficient presumed value D^ and motor rotary speed ω _MMultiply each other, calculate viscous friction compensating torque presumed value τ _d^.Adder 16 is with the preceding motor torque instruction of viscous friction compensation τ _SWith viscous friction compensating torque presumed value τ _dThe ^ addition, calculating motor torque instruction τ _M ^* Gain 18 is with inverse and the motor torque instruction τ of torque constant kt _M ^*Multiply each other, generate torque current command value Iq ^*

First-order lag filter 36 is input torque current instruction value Iq ^*, the filter of output torque current presumed value I^q, its gain crossing angle frequency setting becomes the gain crossing angle frequency of the current control system that constitutes with current controller 8 to equate.

Torque constant amplifier 20 is input torque electric current presumed value I^q, calculating motor torque tau _MGain, set the value equate with the torque constant kt of employed motor 1.

Automatic tuning portion 37 is with motor torque value τ _MWith the motor rotary speed value of detecting ω _MTime series data as input value, total the moment of inertia presumed value J^ of calculating machine system and viscous friction coefficient presumed value D^ are individually set to proportional gain 15,17 with them, comprising: low pass filter (LPF) 21,22; Parameter recognition device working signal generator 23; Automatic gain control part working signal generator (AGI) 24; Discrete system/continuous system parameter converter section 25; With parameter recognition device 19.

Low pass filter 21 is the motor torque τ with 20 outputs of torque constant amplifier _MConvert the motor torque τ after the filtration to _MFLow pass filter, low pass filter 22 is with the motor rotary speed value of detecting ω _MConvert the motor rotary speed value of the detecting ω after the filtration to _MFLow pass filter. Low pass filter 21,22 all mixes repeatedly as anti-that filter uses, its cut frequency be configured to automatic tuning portion 37 sampling frequency about 1/2～1/3.

Parameter recognition device working signal generator 23 (IDA) use the motor torque τ after filtering _MF, generate parameter recognition device working signal Id_act, the motor rotary speed value of the detecting ω after filtering _MFSize surpass setting ω _AThe time, output Id_act=1; Work as ω _MFSize be lower than setting ω _IThe time, output Id_act=0.

Automatic gain control part working signal generator 24 (AGI); Operation parameter identifier working signal Id_act; Generate automatic gain control part working signal Agc_act; Only become 1 sampling period of the next one output Agc_act=1 after 1, export Agc_act=0 afterwards at parameter recognition device working signal Id_act.

Parameter recognition device 19 is under the situation of Id_act=1 at parameter recognition device working signal Id_act only, with the motor torque τ after filtering _MFWith the motor rotary speed value of the detecting ω after the filtration _MFAs input, upgrade output parameter vector presumed value θ ^ (N-1) last time one by one at a distance from the N sampling period every.

The effect of the device of simple declaration parameter recognition here, 19.The identifying object of parameter recognition device 19 is the total the moment of inertia presumed value J^ and the viscous friction coefficient presumed value D^ of mechanical system.At this moment, motor torque τ _MRepresent with following formula.

τ _M＝J(dω _M/dt)+D·ω _M(52)

At this moment, the motor rotary speed value of detecting ω _MTo motor torque τ _MTransfer function G _RD(s) in, s is accorded with as Laplace's operation, can use the steel phantom type of (53) formula to carry out modelling.

(mathematical formulae 46)

$G_{\mathrm{RD}}\left(s\right)=\frac{{\mathrm{\&omega;}}_M}{{\mathrm{\&tau;}}_M}=\frac{1}{\hat{J}\&CenterDot;s+\hat{D}}---\left(53\right)$

Parameter recognition device 19 with the coefficient J^ in the denominator of following formula (53) and D^ as identifying object.But, in order to realize parameter recognition device 19 with the discrete time computing, at first will be through for example integrating the transfer function G that z changes the steel phantom type that makes shown in (52) formula _RD(s) discretization.That is to say that so-called integration z conversion is when the sampling period is made as T, the utmost point s at zero point of continuous system transfer function is mapped as the utmost point z=e at zero point of discrete system transfer function respectively ^STTransformation approach.Integrate the z conversion according to this, with z ^-1Be made as when meaning the operator that postponed for 1 sampling period, can use (56) formula to represent corresponding transfer function G _RD(s) discrete system transfer function G _RD(z).In addition, the coefficient a in (54) formula ₁, b ₀Represent by (55) formula, (56) formula respectively.

(mathematical formulae 47)

$G_{\mathrm{RD}}\left(z\right)=\frac{{\mathrm{\&omega;}}_M}{{\mathrm{\&tau;}}_M}=\frac{{\mathrm{<figure-callout\; id="0"\; label="b"\; filenames="H200810184719XE00021.png,H200810184719XE00072.png"\; state="\{\{state\}\}">b</figure-callout>}}_0}{1+a_1{z}^{-1}}---\left(54\right)$

(mathematical formulae 48)

$a_1=-e^{-\frac{\hat{D}}{\hat{J}}T}---\left(55\right)$

(mathematical formulae 49)

${\mathrm{<figure-callout\; id="0"\; label="b"\; filenames="H200810184719XE00021.png,H200810184719XE00072.png"\; state="\{\{state\}\}">b</figure-callout>}}_0=\frac{1+a_1}{\hat{D}}---\left(56\right)$

Use above relational expression (54)～(56) formula, just can be with the continuous system transfer function G of steel phantom type _RD(S) convert corresponding discrete system transfer function G to _RD(z).In addition, on the contrary, also can be according to G _RD(z) coefficient a ₁, b ₀, calculate G by (57) formula, (58) formula _RD(s) parameter J^, D^.

(mathematical formulae 50)

$\hat{J}=-\frac{(1+a_1)\mathrm{<figure-callout\; id="1"\; label="T\; b"\; filenames="H200810184719XE00011.png,H200810184719XE00021.png"\; state="\{\{state\}\}">T</figure-callout>}}{b_{}}---\left(57\right)$

(mathematical formulae 51)

$\hat{D}=\frac{1+a_1}{b_0}---\left(58\right)$

Therefore, can think in this execution mode, be in parameter recognition device 19, to discern discrete system transfer function G through above-mentioned UD decomposition algorithm _RD(z) unknowm coefficient a ₁, b ₀, then in the parameter converter section 25 of discrete system/continuous system, use (57), (58) formula, convert total the moment of inertia presumed value J^ and viscosity friction presumed value D^ to, and embody in the proportional gain 15,17 of control system.

Here, use Fig. 2, the internal structure of parameter recognition device 19 is described.Parameter recognition device 19 (Fig. 1) comprising: zero- order holder assembly 26,27; 1 sampling Delay Element 28,29; Multiplier 30,31; Automatic gain control part 32 (AGCI); Regression vector generation portion 33; Discrete system parameter estimation portion 34; With working signal generator 35 (PVAI) of discrete system parameter estimation portion.

Zero-order holder assembly 26, in sampling period T to the motor torque τ after filtering _MFCarry out sample circuit and keep, the centrifugal pump τ that output keeps _MS(N).Zero-order holder assembly 27, in sampling period T to the motor torque ω after filtering _MFCarry out sample circuit and keep, the centrifugal pump ω that output keeps _MS(N).1 sampling Delay Element 28, the operation time that is used for guaranteeing automatic gain control part 32 (AGCI), input τ _MS(N), output τ _MS(N-1).1 sampling Delay Element 29, input centrifugal pump ω _MS(N), output ω _MS(N-1).

Multiplier 30, the decay that (59) formula is calculated gain K _{G α}Multiply by centrifugal pump τ _MS(N-1), output centrifugal pump τ _M(N).Multiplier 31 will decay the gain K _{G α}Multiply by centrifugal pump ω _MS(N-1), output centrifugal pump ω _M(N).Wherein, in the computing of (59) formula, regression vector Φ originally ^*(N) be (60) bivector that formula showed.

(mathematical formulae 52)

$K_{\mathrm{g\&alpha;}}=\min \{\frac{\mathrm{ov}\_\lim \_g}{\mathrm{\&gamma;}\&CenterDot;\max \left\{\right|{\mathrm{\&Phi;}}_1^{*}\left(1\right)|,|{\mathrm{\&Phi;}}_2^{*}\left(1\right)\left|\right\}},\sqrt{\frac{\mathrm{ov}\_\lim \_a-\mathrm{\&lambda;}}{\mathrm{\&gamma;}\&CenterDot;\underset{k=1}{\overset{2}{\mathrm{\&Sigma;}}}{\mathrm{\&Phi;}}_k^{*}{\left(1\right)}^2}}\}---\left(59\right)$

(mathematical formulae 53)

Φ ^*(N)＝[-ω _MS(N-1)，ω _MS(N)] ^T(60)

The 33 input centrifugal pump τ of regression vector generation portion _M(N), ω _M(N), the two-dimentional regression vector Φ (N) of output (61) formula performance.

(mathematical formulae 54)

φ(N)＝[-ω _M(N-1)，τ _M(N)] ^T(61)

(mathematical formulae 55)

$\widehat{\mathrm{\&theta;}}\left(N\right)={[{\hat{a}}_1\left(N\right),{\hat{b}}_0\left(N\right)]}^T---\left(62\right)$

Discrete system parameter estimation portion 34 is under the situation of Pv_actI=1 at the working signal Pv_actI of discrete system parameter estimation portion only, through inferring the discrete system parameter based on the recurrent least square method of UD decomposition algorithm.Particularly be exactly, as input, the output parameter vector is presumed value θ ^ (N-1) last time with regression vector Φ (N), as based on (62) formula with the parameter vector presumed value θ ^ (N) of bivector performance last time till the presumed value of sampled data.Here, the element of vector a of parameter vector presumed value θ ^ (N) ₁^ (N), b ₀^ (N) is respectively (55) formula and a shown in (56) formula ₁, b ⁰Presumed value on the N sampling time point.

Automatic gain control part 32 (AGCI) is with τ _MS(N) and ω _MS(N) as input signal, only under the situation of automatic gain control part working signal Agc_act=1, according to (61) formula and (62) formula, computing and output attenuatoin gain K _{G α}, under the situation of Agc_act=0, the decay gain K before the output during sub-sampling _{G α}

Discrete system parameter estimation portion working signal generator (PVAI) 35 generates the discrete system parameter estimation working signal Pv_actI of portion, and in order to ensure the data time in the regression vector generation portion 33, the Id_act that 2 samplings are postponed exports as Pv_actI.

It more than is the explanation that utilizes first execution mode that Fig. 1 carries out.Below, will be elaborated successively to the functional module of pie graph 1,2.Fig. 3 is the internal structure figure of automatic gain control part working signal generator 24 (AGI).

In Fig. 3, automatic gain control part working signal generator 24 (AGI), as input, automatic gain control part working signal Agc_act76 comprises as output with parameter recognition device working signal Id_act70: 1 sampling Delay Element 71,73,75; OR logic element 72 and xor logic element 74.

1 sampling Delay Element 71 is used for confirming the output of 1 sampling Delay Element 86 in the AGCI pie graph (Fig. 4).OR logic element 72 only exports 0 under 2 inputs all are 0 situation, all the other situation outputs 1.1 sampling Delay Element 73 as input, is exported the output of OR logic element 72 to the input of OR logic element 72.

Because the feedback of OR logic element 72 and 1 sampling Delay Element 73 constitutes, OR logic element 72 continues output 1 always after output one time 1.

Xor logic element 74 only exports 1, all the other situation outputs 0 under any side of the output valve of preceding 1 sub-sampling of the output of OR logic element 72 or OR logic element 72 is 1 situation.The output of 1 sampling Delay Element, 75 input OR logic elements 72, it is exported the input of xor logic element 74.Through the formation of above-mentioned 1 sampling Delay Element 75 and xor logic element 74, xor logic element 74 only between 1 sampling date of the time point that the output of OR logic element 72 changes in, output 1, all the other situation outputs 0.According to above formation, automatic gain control part working signal Ggc_act76, the next one that only becomes after 1 at the parameter recognition device working signal Id_act70 that is transfused to exports 1 in 1 sampling period, just exports 0 later on.

Fig. 4 is the internal structure figure of automatic gain control part AGCI.Automatic gain control part (AGCI) 32, only under the situation of parameter recognition device working signal Id_act=1, according to (61) formula and (62) formula, computing and output attenuatoin gain K _{G α}, under the situation of Id_act=0, output attenuatoin gain K during sub-sampling before being implemented in _{G α}Function.

In Fig. 4, AGCI32 is according to input signal ω _MS(N) 80, τ _MS(N) 81 computing K _{G α}83, comprising: signed magnitude arithmetic(al) device 84,85 (ABS); 1 sampling Delay Element 86 and decay gain K _{G α} Operational part 97.

Signed magnitude arithmetic(al) device 84,85 is calculated ω respectively _MS(N), τ _MS(N) absolute value, the output of signed magnitude arithmetic(al) device 84 through 1 sampling Delay Element 86, is transfused to decay gain K _{G α} Operational part 97, the output of signed magnitude arithmetic(al) portion 85 is by direct input attenuation gain K _{G α}Operational part 97.In addition, automatic gain control part working signal Agc_act82 is also by direct input attenuation gain K _{G α} Operational part 97.

Decay gain K _{G α}Only work under the situation of Agc_act=1 in the inside of operational part 97, to decay gain K _{G α}83 upgrade.

Decay gain K _gCalculate portion 98, calculate the K that is equivalent to (61) formula the right first element _g, decay gain K _αCalculate portion 99 and calculate the K that is equivalent to (61) formula the right second key element _αMinimum value is relatively selected element 90, the decay gain K that output is transfused to _gWith decay gain K _αIn a less side.

Constant 88 is the g as the g element of vector that state in postpone ₁(N) and g ₂When the maximum value that (N) is allowed is ov_lim_g, with the diagonal angle key element initial value γ (γ=d of this absolute value ov_lim_g divided by the D matrix ₁(0)=d ₂(0)) value that obtains.At this moment, as ov_lim_g, the arithmetic processing apparatus of being installed, but select not overflow and below the maximum of the numerical value that tape symbol shows.Especially select the to be fixed greatest measure of computing figure place restriction of decimal point computing just can use the CPU of low price to control.

Maximum is relatively selected element 87, the regression vector Φ that output is transfused to ^*(N) certain bigger side in the absolute value of key element.Divider 89 is selected the output of element 87 with constant 88 divided by high specific, output attenuatoin gain K _gIn addition, what constant 95 was got is, the α as the α element of vector that states in postpone ₁(N) and α ₂When the maximum value that (N) is allowed was ov_lim_ α, the value that from ov_lim_ α, deducts forgetting factor λ (below 1 and be near the positive number 1) was again divided by the diagonal angle key element initial value γ (γ=d of D matrix ₁(0)=d ₂The square root of the value that (0)) obtains.At this moment as ov_lim_ α, the arithmetic processing apparatus of selecting to be installed is selected not overflow and can not had below the maximum of numerical value of symbol performance (because the key element of α vector all be on the occasion of).

Multiplier 91,92 calculates the regression vector Φ that is transfused to respectively ^*(N) absolute value of key element square.Adder 93 is with the output results added of multiplier 91,92.Extracting operation is carried out in the output of 94 pairs of adders 93 of extracting operation device.Divider 96, with the output of constant 95 divided by extracting operation device 94, output attenuatoin gain K _αThe formation of automatic gain control part 32 (AGCI) more than has been described.

Below, utilize the block diagram of Fig. 5, the formation of regression vector generation portion 33 is described.In Fig. 5, regression vector generation portion 33 comprises: 1 sampling Delay Element 113; Sign-inverted device 114; With vectorization mark 115, establish motor rotary speed ω _M(N) and motor torque τ _M(N) be respectively input signal 110,111, establish regression vector Φ (N) and be output signal 112.In addition, each input signal is in multiplier shown in Figure 2 30,31, to multiply by decay gain K _{G α}After signal.

1 sampling Delay Element 113, input ω _M(N), output ω _M(N-1).The output ω of sign-inverted device input 1 sampling Delay Element 113 _M(N-1), output-ω _M(N-1).Vectorization mark 115 is used for right-ω _M(N-1) and τ _M(N) carry out vectorization performance, output is with-ω _M(N-1) and τ _M(N) be the regression vector Φ (N) of (61) formula of key element.What replenish is, for vectorization mark 115, installs and can ignore, and is embodied on the data structure as tectosome.

Below, utilize Fig. 6, the formation of discrete system parameter estimation portion 34 is described.In Fig. 6, discrete system parameter estimation portion 34 is with motor rotary speed centrifugal pump ω _M(N) and regression vector Φ (N) as input signal 120,121, with parameter vector last time presumed value θ ^ (N-1) as output signal 122, with fallout predictor 123, subtracter 124, parameter vector arithmetic unit 125, UD decomposition portion 126 as inscape.In addition, parameter vector is presumed value θ ^ (N-1) last time, mean based on the parameter vector presumed value θ ^ (N) of the bivector of (62) formula performance last time till the presumed value of sampled data.

Fallout predictor 123 is according to regression vector Φ (N) and parameter vector presumed value θ ^ (N-1) calculating motor prediction of speed value ω ^ last time _M(N), be equivalent to the 2nd the computing on (4) formula the right.Subtracter 124 is through from motor rotary speed ω _M(N) deduct motor rotary speed predicted value ω ^ _M(N), calculate predicated error ε (N), be equivalent to the computing of (4) formula the right subtraction.Parameter vector arithmetic unit 125 uses the computing that is equivalent to (5) formula, according to predicated error ε (N) and gain vector k (N), parameter vector presumed value θ ^ (N) is upgraded computing.UD analysis portion 126 is come calculated gains vector k (N) according to regression vector Φ (N) through the UD decomposition algorithm, comprising: f vector operation device 127; G vector operation device 128; α μ arithmetic unit 129; U matrix operation device 130; D matrix operation device 131; With gain vector arithmetic unit 132.

F vector operation device 127 utilizes the computing of (65) formulas, is the previous value U (N-1) of (2 * 2) upper triangular matrix U (N) of 1 according to the diagonal angle composition shown in regression vector Φ (N) and (63) formula, calculating f vector f (N).

(mathematical formulae 56)

$U\left(N\right)=\left[\begin{array}{cc}1& u_{12}\left(N\right)\\ 0& 1\end{array}\right]---\left(63\right)$

(mathematical formulae 57)

$D\left(N\right)=\left[\begin{array}{cc}{d}_1\left(N\right)& 0\\ 0& d_2\left(N\right)\end{array}\right]---\left(64\right)$

(mathematical formulae 58)

$f\left(N\right)=\left[\begin{array}{c}{f}_1\left(N\right)\\ f_2\left(N\right)\end{array}\right]=U^T(N-1)\mathrm{\&phi;}\left(N\right)---\left(65\right)$

$=\left[\begin{array}{cc}1& 0\\ u_{12}\left(N\right)& 1\end{array}\right]\left[\begin{array}{c}{-\mathrm{\&omega;}}_M(N-1)\\ {\mathrm{\&tau;}}_M\left(N\right)\end{array}\right]=\left[\begin{array}{c}-{\mathrm{\&omega;}}_M(N-1)\\ -{\mathrm{\&omega;}}_M(N-1)u_{12}\left(N\right)+{\mathrm{\&tau;}}_M\left(N\right)\end{array}\right]$

G vector operation device 128 utilizes the computing of (66) formula, according to the previous value D (N-1) of (2 * 2) the diagonal matrix D (N) shown in f vector f (N) and (64), calculates g vector g (N).

(mathematical formulae 59)

$g\left(N\right)=\left[\begin{array}{c}{g}_1\left(N\right)\\ g_2\left(N\right)\end{array}\right]=D(N-1)f\left(N\right)---\left(66\right)$

$=\left[\begin{array}{cc}{d}_1\left(N\right)& 0\\ 0& d_2\left(N\right)\end{array}\right]\left[\begin{array}{c}{f}_1\left(N\right)\\ f_2\left(N\right)\end{array}\right]=\left[\begin{array}{c}{d}_1\left(N\right)\&CenterDot;f_1\left(N\right)\\ d_2\left(N\right)\&CenterDot;f_2\left(N\right)\end{array}\right]$

α μ arithmetic unit 129 utilizes the computing of (67) formula and (68) formula, according to g vector g (N) and f vector f (N), calculates α vector α (N) through (69) formula, calculates μ ₂(N).

(mathematical formulae 60)

$\mathrm{\&alpha;}\left(N\right)=\left[\begin{array}{c}{\mathrm{\&alpha;}}_0\left(N\right)\\ {\mathrm{\&alpha;}}_1\left(N\right)\\ {\mathrm{\&alpha;}}_2\left(N\right)\end{array}\right]=\left[\begin{array}{c}\mathrm{\&lambda;}\\ \mathrm{\&lambda;}+f_1\left(N\right)\&CenterDot;g_1\left(N\right)\\ \mathrm{\&lambda;}+f_1\left(N\right)\&CenterDot;g_1\left(N\right)+f_2\left(N\right)\&CenterDot;g_2\left(N\right)\end{array}\right]---\left(67\right)$

(mathematical formulae 61)

${\mathrm{\&mu;}}_2\left(N\right)=-\frac{f_2\left(N\right)}{{\mathrm{\&alpha;}}_1\left(N\right)}---\left(68\right)$

U matrix operation device 130 utilizes the computing of (69) formula, according to g vector g (N) and μ ₂(N) calculate u ₁₂(N), and pass through (70) formula, calculate v vector v (N).In addition, U matrix operation device 130 is with the u of (63) formula use ₁₂(N-1) value feeds back to f vector operation device 127.

(mathematical formulae 62)

u ₁₂(N)＝u ₁₂(N-1)+μ ₂(N)·g ₁(N)(69)

(mathematical formulae 63)

$v\left(N\right)=\left[\begin{array}{c}{v}_1\left(N\right)\\ v_2\left(N\right)\end{array}\right]=\left[\begin{array}{c}{g}_1\left(N\right)+u_{12}(N-1)\&CenterDot;g_2\left(N\right)\\ g_2\left(N\right)\end{array}\right]---\left(70\right)$

D matrix operation device 131 utilizes the computing of (71) formula, calculates d according to α vector α (N) ₁(N), and through (72) formula calculate d ₂(N), thus D (N-1) is fed back to g vector operation device 128.

(mathematical formulae 64)

$d_1\left(N\right)=\frac{{\mathrm{\&alpha;}}_0\left(N\right)\&CenterDot;d_1(N-1)}{\mathrm{\&lambda;}\&CenterDot;{\mathrm{\&alpha;}}_1\left(N\right)}---\left(71\right)$

(mathematical formulae 65)

$d_2\left(N\right)=\frac{{\mathrm{\&alpha;}}_1\left(N\right)\&CenterDot;d_2(N-1)}{\mathrm{\&lambda;}\&CenterDot;{\mathrm{\&alpha;}}_2\left(N\right)}---\left(72\right)$

Gain vector arithmetic unit 132 utilizes the computing of (73) formula, according to v vector v (N) and α ₂(N), calculate gain vector k (N).

(mathematical formulae 66)

$k\left(N\right)=\left[\begin{array}{c}{k}_1\left(N\right)\\ k_2\left(N\right)\end{array}\right]=\frac{1}{{\mathrm{\&alpha;}}_2\left(N\right)}\left[\begin{array}{c}{v}_1\left(N\right)\\ v_2\left(N\right)\end{array}\right]---\left(73\right)$

More than, be the formation of discrete system parameter estimation portion 34.Below, successively the functional module that constitutes discrete system parameter estimation portion 34 is elaborated.

The pie graph of Fig. 7 is that the inside of fallout predictor 123 constitutes sketch map.In Fig. 7, fallout predictor 123 with parameter vector last time presumed value θ ^ (N-1) and regression vector (N) as input signal 140,141, with motor rotary speed predicted value ω ^ _M(N) as output signal 147, comprising: vectorization mark 142,143; Multiplier 144,145; With adder 146.

Vectorization mark 142, carry out from parameter vector last time presumed value θ ^ (N-1) take out key element a ₁^ (N-1), b ₀The operation of ^ (N-1).Vectorization mark 143 carries out taking out key element-ω from regression vector Φ (N) _M(N-1), τ _M(N) operation.Multiplier 144 carries out key element a ₁^ (N-1) and key element-ω _M(N-1) multiplying.Multiplier 145 carries out key element b ₀^ (N-1) and key element τ _M(N) multiplying.Adder 146, with the output signal of multiplier 144 and the output signal plus of multiplier 145, with the result as electromotor velocity predicted value ω ^ _M(N) output.

Fig. 8 is the internal structure figure of parameter vector arithmetic unit 125 that carries out the computing of (5) formula.

In Fig. 8, establishing gain vector k (N) and predicated error ε (N) is input signal 160,161, and setting parameter vector last time presumed value θ ^ (N-1) comprises: vectorization mark 163,170 for output signal 162; Multiplier 164,167; Adder 165,168; Postpone key element 166,169 with 1 sampling.

Vectorization mark 163, expression is taken out key element k from gain vector k (N) ₁(N), k ₂(N) operation.Multiplier 164 carries out key element k ₁(N) and the multiplying of predicated error ε (N), multiplier 167 carries out key element k ₂(N) and the multiplying of predicated error ε (N).Adder 165 is with the output and a of multiplier 164 ₁^ (N-1) addition, the coefficient a that calculates (1) formula ₁^ (N), adder 168 is with the output and the coefficient b of multiplier 167 ₀Coefficient b is calculated in ^ (N-1) addition ₀^ (N).

1 sampling Delay Element 166, input coefficient a ₁^ (N), output factor a ₁^ (N-1), 1 sampling Delay Element, 169 input b ₀^ (N), output factor b ₀^ (N-1).170 couples of coefficient a1^ of vectorization mark (N-1) and b ₀^ (N-1) carries out vectorization, last time presumed value θ ^ (N-1) output of the parameter vector to 162.

Below, utilize Fig. 9, explain that the inside of f vector operation device constitutes.In Fig. 9, f vector operation device 127 utilizes the computing of (67) formula, with the 1st row the 2nd row key element u of regression vector Φ (N) and U matrix previous value U (N-1) ₁₂(N-1) input signal 180,181 converts the output signal 182 of f vector f (N) to, comprising: vectorization mark 183,186; Multiplier 184 and adder 185.

Vectorization mark 183, key element-ω of regression vector Φ (N) is taken out in expression _M(N-1), τ _M(N) operation.Multiplier 184 carries out-ω _M(N-1) and u ₁₂(N-1) multiplying.Adder 185 is with the output and the τ of multiplier 184 _M(N) the 2nd key element f is exported in addition ₂(N).Vectorization mark 186, expression general-ω _M(N-1) carry out vectorization and make it to become first element f ₁(N), in addition vectorization is carried out in the output of adder 185 and made it to become the second key element f ₂(N) operation, output f vector f (N) 182.

Utilize Figure 10 below, explain that the inside of g vector operation device constitutes.In Figure 10; G vector operation device 128 utilizes the computing of (66) formulas, with f vector f (N) and the vector that is made up of the diagonal angle composition of D matrix previous value D (N-1), as input signal 200,201; Convert the output signal 202 of g vector g (N) to, comprising: vectorization mark 203,204,209; Multiplier 205,207; With amplitude limiter 206,208.

Vectorization mark 203, the key element f of f vector f (N) is taken out in expression ₁(N), f ₂(N) operation, the diagonal angle key element d of D (N-1) is taken out in 204 expressions of vectorization mark ₁(N-1), d ₂(N-1) operation.Multiplier 205 carries out key element f ₁(N) and diagonal angle key element d ₁(N-1) multiplying, multiplier 207 carries out key element f ₂(N) and diagonal angle key element d ₂(N-1) multiplying.

Amplitude limiter 206,208, the upper lower limit value that use can show under the situation of not overflowing carries out limit processing to the output of multiplier 205,207 respectively.But in the formation of this execution mode that automatic gain control part 32 (Fig. 2) is installed, amplitude limiter 206,208 is not worked usually.Vectorization mark 209 carries out vectorization to the output of amplitude limiter 206 and makes it to become first element g ₁(N), in addition vectorization is carried out in the output of amplitude limiter 208 and made it to become the second key element g ₂(N), output g vector g (N).

Below, utilize Figure 11, explain that the inside of α μ arithmetic unit constitutes.In Figure 11, α μ arithmetic unit 129 utilizes the computing of (67) formulas, (68) formula, as input signal 220,221, and converts g vector g (N) and f vector f (N) to μ ₂(N) and the output signal 222,223 of α vector α (N), comprising: vectorization mark 236,237,235; Multiplier 225,230; Adder 226,231; Amplitude limiter 227,232; Divider 228; With sign-inverted device 229.

The key element g of g vector g (N) is taken out in 236 expressions of vectorization mark ₁(N), g ₂(N) operation, the key element f of f (N) is taken out in 237 expressions of vectorization mark ₁(N), f ₂(N) operation.Constant 224 has been deposited forgetting factor λ.Multiplier 225 carries out g ₁(N) and f ₁(N) multiplying, multiplier 230 carries out key element g ₂(N) and key element f ₂(N) multiplying.Adder 226 is with the output signal plus of forgetting factor λ and multiplier 225.

Amplitude limiter 227 uses the higher limit that can under the situation of not overflowing, show, and amplitude limiting processing is carried out in the output of adder 226.Adder 231 is with the output of amplitude limiter 227 and the output addition of multiplier 230.Amplitude limiter 232 uses the higher limit that can under the situation of not overflowing, show, and amplitude limiting processing is carried out in the output of adder 231.But in the formation of this execution mode that automatic gain control part 32 (Fig. 2) is installed, amplitude limiter 227,232 is not worked usually.

Vectorization mark 235 carries out vectorization to forgetting factor λ and makes it to become first element α ₀(N), vectorization is carried out in the output of amplitude limiter 227 and made it to become the second key element α ₁(N), vectorization is carried out in the output of amplitude limiter 232 and made it to become three elements α ₂(N), output α vector α (N).Divider 228 is with f ₂(N) α that exports divided by amplitude limiter 227 ₁(N).The output valve of 229 pairs of dividers 228 of sign-inverted device is carried out sign-inverted, as μ ₂(N) output.

Below, utilize Figure 12, explain that the inside of U matrix operation device constitutes.U matrix operation device 130 utilizes the computing of (69) formula, (70) formula in Figure 12, with g vector g (N) and μ ₂(N), convert the 1st row the 2nd row key element u of v vector v (N) and U matrix previous value U (N-1) into as input signal 250,251 ₁₂(N-1) output signal 252,253 comprises: vectorization mark 254,260; Multiplier 255,258; Adder 256,259; With 1 sampling Delay Element 257.

Vectorization mark 254, the key element g of g (N) is taken out in expression ₁(N), g ₂(N) operation.Multiplier 255 is with key element g ₁(N) and μ ₂(N) multiply each other.Adder 256 is with the output signal and the u of multiplier 255 ₁₂(N-1) u is carried out in addition ₁₂(N) computing.1 sampling Delay Element 257 is with the u of adder 256 outputs ₁₂(N) as input signal, output u ₁₂(N-1).Multiplier 258 is with u ₁₂(N-1) and g ₂(N) multiply each other.Adder 259 is with g ₁(N) with the output signal plus of multiplier 258.Vectorization mark 260, expression is carried out vectorization with the output of adder 259 makes it to become first element v ₁(N), to g ₂(N) carry out vectorization and make it to become the second key element v ₂(N), output v vector v (N).

Below, utilize Figure 13, explain that the inside of D matrix operation device constitutes.In Figure 12, D matrix operation device 131 utilizes the computing of (71) formula, (72) formula, as input signal 270, converts α vector α (N) the diagonal angle key element d of D matrix previous value D (N-1) to ₁(N-1), d ₂The output signal 271 of the vector that (N-1) constitutes comprises: vectorization mark 272,281; Divider 273,277; Multiplier 274,278; Gain 275,278; With 1 sampling Delay Element 276,280.

Vectorization mark 272, the key element α of α vector α (N) is taken out in expression ₀(N), α ₁(N), α ₂(N) operation.Divider 273 is with key element α ₀(N) divided by α ₁(N).Divider 277 is with key element α ₁(N) divided by key element α ₂(N).Multiplier 274 is with d ₁(N-1) with the output signal multiplication of divider 273.Gain 275 multiply by 1/ λ to the output of multiplier 274.The output of 1 sampling Delay Element, 276 input gains 275, output diagonal angle key element d ₁(N-1).Multiplier 278 is with diagonal angle key element d ₂(N-1) with the output multiplication of divider 277.1/ λ is multiply by in the output of 279 pairs of multipliers 278 of gain.The output of 1 sampling Delay Element, 280 input gains 279, output diagonal angle key element d ₂(N-1).Vectorization mark 281 is to diagonal angle key element d ₁(N-1) and diagonal angle key element d ₂(N-1) carry out vectorization, and export as D matrix previous value D (N-1).

Below, utilize Figure 14, explain that the inside of gain vector arithmetic unit constitutes.In Figure 14, gain vector arithmetic unit 132 utilizes the computing of (73) formula, with v vector v (N) and α ₂(N) as input signal 290,291, convert the output signal 292 of gain vector k (N) to, comprising: vectorization mark 293,296; With divider 294,295.

Vectorization mark 293, the key element v of v (N) is taken out in expression ₁(N) and v ₂(N) operation.Divider 294 is with v ₁(N) divided by α ₂(N), output k ₁(N).Divider 295 is with v ₂(N) divided by α ₂(N) output k ₂(N).Vectorization mark 296, expression is with k ₁(N) and k ₂(N) carry out vectorization, and export as gain vector k (N).

More than, specified the formation of automatic tuning portion 37, below, describe according to flow chart shown in Figure 15 processing sequence automatic tuning portion 37.

In order, initial value (S301) is set in above-mentioned flow processing (S300) beginning.Particularly be exactly setup parameter identifier working signal Id_act=0, identification frequency counter Id_count=0, automatic gain control part working signal Agc_act=0, the working signal Pv_actI=0 of discrete system parameter estimation portion.

Next, parameter recognition device working signal generator 23 is judged the motor rotary speed value of the detecting ω after filtering _MFAbsolute value (size) whether at setting ω _AMore than (S302).If not at setting ω _AMore than (No), just repeat the processing of S302, wait for reaching setting ω _AMore than.In addition, if reach setting ω _AMore than (Yes), parameter recognition device working signal generator 23 just becomes 1 with parameter recognition device working signal Id_act by 0, thus, parameter recognition device 19 begins to handle (S303).

Then, in order to ensure the data filling time (K of the 1 sampling Delay Element 86 of the 1 sampling Delay Element 28,29 of giving Fig. 2 and Fig. 4 _{G α}Operation time), automatic tuning portion 37 carries out 1 sampling period delay (S304).Then, automatic gain control part working signal generator 24, Agc_act is provided with 1 to automatic gain control part working signal.Thus, the computing of automatic gain control part (AGCI) 32 will be performed, and 32 pairs of automatic gain control parts (AGCI) overflow and suppress gain (decay gain) K _{G α}Carry out computing (S305).Then, automatic gain control part working signal generator 24 makes automatic gain control part working signal Agc_act return 0.Thus, this routine plays again by starting from S300 between, decay gain K _{G α}Will fix (S306).

Then, in order to ensure the data filling time (regression vector filling time) of Delay Element 113 that 1 of regression vector generation portion 33 is taken a sample, carry out 1 sampling period delay (S307).Then, through the working signal Pv_actI of discrete system parameter estimation portion being become 1 by 0, the processing (S311) of beginning discrete system parameter estimation portion 34.Then, in order repeatedly to discern to improve accuracy of identification, identification frequency counter Id_count is added 1 (S312) through carrying out.

Then, rotary speed ω judges in automatic tuning portion 37 _MFAbsolute value not enough setting ω whether _I(S313).Automatic tuning portion 37 is at rotary speed ω _MFAbsolute value be setting ω _IUnder the above situation (No), carry out S313 once more, wait for becoming not enough setting ω _IOn the other hand, automatic tuning portion 37 is at ω _MFThe not enough setting ω of absolute value _ISituation under (Yes), through Id_act and Pv_actI are become 0 by 1, stop parameter recognition device 19 (S314).Then, automatic tuning portion 37 judges that identification frequency counter Id_count is whether more than setting times N C (S315).

If identification frequency counter Id_count is less than stipulated number NC (No), parameter recognition device working signal generator 23 is just judged the motor rotary speed value of the detecting ω after filtering _MFAbsolute value whether at setting ω _AMore than (S308).If be not setting ω _AMore than (No), parameter recognition device working signal generator 23 just repeats the processing of S308.If at setting ω _AMore than (Yes), parameter recognition device working signal generator 23 just begins the processing (S309) of parameter recognition device 19 through parameter recognition device working signal Id_act is become 1 by 0.Then, automatic tuning portion 37 is in order to ensure K _{G α}Operation time and regression vector filling time, carry out 2 sampling periods delays (S310), carry out the processing of above-mentioned S311 to S315.

If at identification frequency counter Id_count in the judgement of S315 is (Yes) more than the stipulated number NC; Then the parameter recognition device 19, will as the parameter vector of recognition result last time presumed value θ ^ (N-1) on discrete system/continuous system parameter converter section 25, convert total the moment of inertia presumed value J^ and viscous friction coefficient presumed value D^ (S316) to as the continuous system parameter.Then, automatic tuning portion 37 is provided with J^, D^ (S317) respectively to the proportional gain 15,17 of speed control 9, finishes this routine and handles (S318).

It more than is the flow process of a succession of processing in the automatic tuning portion.Below, utilize Figure 16, the simulated action waveform shown in 17, the effect that this execution mode brought is described.

Wherein, as simulated conditions, total the moment of inertia of the mechanical system be made up of motor, connection shaft and driven object load is provided with J=5.71 * 10 ^-5[Kgm ²], the viscous friction coefficient is provided with D=1.0 * 10 ^-3[Nm/ (rad/s)] is provided with T=8.96 [ms] to the sampling period.Thus, can know according to the relation of (55) formula and (56) formula:

is true value.

Figure 16 (a) is the motor rotary speed value of the detecting ω after filtering _MF Waveform 330, Figure 16 (b) is the waveform 331 of automatic gain control part working signal Agc_act, Figure 16 (c) is the waveform 332 of parameter recognition device working signal Id_act, Figure 16 (d) is the waveform 333 of the working signal Pv_actI of discrete system parameter estimation portion.

In addition, as comparative example, Figure 17 is illustrated in decay gain K _{G α}Be fixed as the waveform when discerning under 1 the condition.Figure 17 (a) is a ₁The waveform 334 of ^, Figure 17 (b) is b ₀The waveform 335 of ^.In addition, under situation about discerning, with a with the formation of this execution mode ₁^ is expressed as waveform 336, with b ₀^ is expressed as waveform 337.Can know that according to waveform 330, waveform 332 and waveform 333 in this emulation, expression is until the action of identification frequency counter Id_count=2.

At this moment, can confirm: gain K will decay _{G α}Be fixed as in the comparative example under 1 the situation, as waveform 334 and waveform 335 are certifiable, at the identification beginning initial stage, excessive identification error take place, even become the symbol opposite with respect to true value, the convergence of past true value is also lower thereafter.And in the present embodiment,,, do not have excessive identification error, and toward the true value convergence at the identification beginning initial stage just as waveform 336 and waveform 337 certifiable that kind.

(the 2nd execution mode)

Below, utilize Figure 18, the parameter recognition device of the 2nd execution mode of the present invention is described.Be that gain K will decay with the difference of parameter recognition device 19 shown in Figure 2 _{G α}With the regression vector this point that multiplies each other.In addition, same as Fig. 1 of the formation of motor control system and the 1st execution mode.

Regression vector generation portion 351, according to (62) formula, input τ _MS(N) and ω _MS(N), generate regression vector Φ ^*(N).Multiplier 353 is with the decay gain K of automatic gain control part 352 (AGC2) output _{G α}With regression vector Φ ^*(N) multiply each other, operation result is exported as regression vector Φ (N).Multiplier 354, gain K will decay _{G α}With ω _MS(N) multiply each other, with operation result as ω _M(N) output.

Working signal generator 355 (PVA2) of discrete system parameter estimation portion, in order to ensure the data time in the regression vector generation portion 351, the Id_act that 1 sampling is postponed exports as the working signal Pv_act of discrete system parameter estimation portion.

Automatic gain control part 352 (AGC2) is with regression vector Φ ^*(N) as input signal, only under the situation of automatic gain control part working signal Agc_act=1, according to (61) formula and (62) formula, at computing decay gain K _{G α}Time output operation result, under the situation of Agc_act=0, the decay gain K before the output during sub-sampling _{G α}

Below, utilize Figure 19, explain that the inside of automatic gain control part 352 (AGC2) constitutes.In Figure 17, the difference with automatic gain control part 32 (AGCI) shown in Figure 4 is merely the importation.In the 2nd execution mode, it is characterized in that, owing to imported regression vector Φ ^*(N), generate regression vector so need not 1 sampling Delay Element, 86 that kind of image pattern 4 in inside, but, be input to decay gain K by vectorization mark 373 through signed magnitude arithmetic(al) device 374,375 _gCalculate portion 98.At regression vector Φ ^*(N) key element-ω of 370 _M(N-1), τ _M(N) be input to respectively signed magnitude arithmetic(al) device 374, after 375, respectively to constituting the decay identical K that gains with Fig. 3 _{G α} Operational part 97 is exported.

Below, utilize flow chart shown in Figure 20, the processing sequence of parameter recognition device 350 is described.

In order, carry out the processing (S400) of this flow process, parameter recognition device 350 carries out the initialization (S401) of flag bit and variable.Judge the motor rotary speed value of the detecting ω after filtering then _MFAbsolute value (size) whether at setting ω _AMore than (S402).Be to return S402 under the situation of No in result of determination, wait for becoming setting ω _AMore than.

On the other hand, be under the situation of Yes in result of determination, through parameter recognition device working signal Id_act being become 1 by 0, the processing (S403) of beginning parameter recognition device 350.Then, in order to ensure the data filling time (regression vector filling time) of Delay Element that 1 of regression vector generation portion 351 is taken a sample, carry out 1 sampling period delay (S404).Then automatic gain control part working signal Agc_act is provided with 1, carries out the computing (S405) of automatic gain control part AGC2 thus, calculate decay gain K _{G α}

Then automatic gain control part working signal Agc_act is returned 0.Thus, from handle 400 play restart between, decay gain K _{G α}(S406) will be fixed.Then, through the working signal Pv_act of discrete system parameter estimation portion being become 1 by 0, the processing (S410) of beginning discrete system parameter estimation portion 34.

Then, in order repeatedly to discern to improve accuracy of identification, identification frequency counter Id_count is added 1 (S411) through carrying out.

Then, judge rotary speed ω _MFAbsolute value not enough setting ω whether _I(S412).If at setting ω _IMore than (No), just repeat the processing of S412.And if ω _MFAbsolute value less than setting ω _I(Yes), just through Id_act and Pv_act2 are become 0 by 1, stop the processing (S413) of parameter recognition device 350.

Then, judge whether identification frequency counter Id_count is stipulated number NC above (S414).If be not stipulated number NC above (No), just judge the motor rotary speed value of the detecting ω after filtering _MFSize whether be setting ω _AMore than (S407).The processing of repetition S407 under the situation of No.And be under the situation of Yes in judged result, just through parameter recognition device working signal Id_act being become 1 by 0, the processing (S408) of beginning parameter recognition device 350.Then, in order to ensure the regression vector filling time, carry out 1 sampling period delay (S409).Then, carry out the processing from S410 to S414,, just repeat these processing if be not (No) more than the stipulated number NC at identification frequency counter Id_count on the S414.

On the other hand; If be judged as (Yes) more than the stipulated number NC at S414; Just will in discrete system/continuous system parameter converter section 25, convert total the moment of inertia presumed value J^ and viscous friction coefficient presumed value D^ (S415) to as the parameter vector of recognition result presumed value θ ^ (N-1) last time as the continuous system parameter.Then, the proportional gain 15,17 to speed control 9 is provided with J^, D^ (S416), end process (S417) respectively.

More than be the flow process of a series of processing in the automatic tuning portion, thus, the 2nd execution mode shown in Figure 16 also can be brought into play the effect same with the 1st execution mode.

Based on this embodiment, can be in suppressing the recurrent least square method computing overflow in, keep accuracy of identification than the highland.Thus, can use recognition result to carry out the high-precision automatic setting of Control Parameter, realize high performance control more simply.

(variation)

The invention is not restricted to above-mentioned execution mode, can carry out for example following various distortion.

(1) in above-mentioned each execution mode, though all be to keep assembly 26,27 to obtain N parameter τ through zeroth order _MS(N), ω _MSBut also can pass through FIFO (First In First Out) and realize above-mentioned functions (N).

In (2) the 2nd execution modes, though be the electromotor velocity sampling value ω after will not filtering through multiplier _MS(N) import regression vector generation portion 351, but also can signal be imported other regression vector generation portion through multiplier.

Claims (11)

1. control device of electric motor possesses: speed control, and according to rotary speed value of detecting of motor and the deviation between the motor rotary speed command value, output torque current command value; Current controller according to said torque current command value and as the deviation between torque current value of detecting of the torque current composition of the drive current that flows in the said motor, generates the control signal of the said drive current of control; With automatic tuning portion; Use motor torque value and said these both sides of the rotary speed value of detecting of said motor, adjust the Control Parameter of said speed control automatically, said control device of electric motor; Control said motor; Make said rotary speed value of detecting near said motor rotary speed command value, it is characterized in that

Said automatic tuning portion comprises:

Import the 1st low pass filter of said motor rotary speed value of detecting; Import the 2nd low pass filter of said motor torque value;

Use the output signal of said the 1st low pass filter and the output signal of said the 2nd low pass filter, the parameter recognition device of discrete ground identification parameter vector; With

Said parameter vector is converted to the parameter converter section of discrete system/continuous system of the Control Parameter of said speed control,

Said parameter recognition device comprises:

The 1st zeroth order keeps assembly, with some cycles the output signal of said the 1st low pass filter is taken a sample;

The 2nd zeroth order keeps assembly, with some cycles the output signal of said the 2nd low pass filter is taken a sample;

Regression vector generation portion uses said the 1st zeroth order to keep the output signal of assembly and the output signal that said the 2nd zeroth order keeps assembly, generates regression vector;

2 multipliers respectively through Delay Element, keep the output signal of assembly and said the 2nd zeroth order to keep the output signal of assembly to decay to said the 1st zeroth order;

Discrete system parameter estimation portion uses said regression vector and said the 1st zeroth order is kept the inhibit signal that 1 sampling period of output signal delay of assembly obtains, and said parameter vector is carried out computing; With

The automatic gain control part uses the 1st zeroth order to keep the output signal of assembly and the output signal that said the 2nd zeroth order keeps assembly, the gain of said 2 multipliers of computing.

2. control device of electric motor according to claim 1 is characterized in that,

Said regression vector generation portion uses the output signal of said 2 multipliers, and said regression vector is carried out computing.

3. control device of electric motor according to claim 1 is characterized in that,

Said regression vector generation portion uses the 1st zeroth order to keep the output signal of assembly and the output signal that said the 2nd zeroth order keeps assembly, generates said regression vector,

Said 2 multipliers keep the output signal of assembly and the output signal of said regression vector generation portion to decay to said the 1st zeroth order respectively,

Said discrete system parameter estimation portion uses the output signal of said 2 multipliers, infers said parameter vector.

4. control device of electric motor according to claim 1 is characterized in that,

Said automatic gain control part only carries out work at the initial stage that the work of said parameter recognition device begins, in the process of the recognition operation of carrying out said discrete system parameter estimation portion, and the yield value of fixing said 2 multipliers.

5. control device of electric motor according to claim 1 is characterized in that,

The gain of said 2 multipliers, the dynamic range restriction of the decimal point computing that is fixed.

6. control device of electric motor according to claim 1 is characterized in that,

Said regression vector is 1 in the gain of establishing said 2 multipliers, with N be made as natural number, when A is made as constant, by the p dimensional vector Φ of mathematical formulae 1 ^*(N) definition,

Mathematical formulae 1: ${\mathrm{\&Phi;}}^{*}\left(N\right)=[{\mathrm{\&Phi;}}_1^{*}\left(N\right),{\mathrm{\&Phi;}}_2^{*}\left(N\right),...,{\mathrm{\&Phi;}}_p^{*}\left(N\right)],$

The gain Kg of said 2 multipliers is set to

Mathematical formulae 2: $K_g=\frac{A}{\mathrm{Max}\left\{\right|{\mathrm{\&Phi;}}_1^{*}\left(1\right)|,|{\mathrm{\&Phi;}}_2^{*}\left(1\right)|,...,|{\mathrm{\&Phi;}}_p^{*}\left(1\right)\left|\right\}}$ Value.

7. control device of electric motor according to claim 1 is characterized in that,

Said regression vector is 1 in the gain of establishing said 2 multipliers, N is made as natural number, when B is made as constant, by the p dimensional vector Φ of mathematical formulae 3 ^*(N) definition,

Mathematical formulae 3: ${\mathrm{\&Phi;}}^{*}\left(N\right)=[{\mathrm{\&Phi;}}_1^{*}\left(N\right),{\mathrm{\&Phi;}}_2^{*}\left(N\right),...,{\mathrm{\&Phi;}}_p^{*}\left(N\right)]$

The gain K α of said 2 multipliers is set to

Mathematical formulae 4: $K_{\mathrm{\&alpha;}}=\frac{B}{\sqrt{\underset{k=1}{\overset{p}{\mathrm{\&Sigma;}}}{\mathrm{\&Phi;}}_k^{*}{\left(1\right)}^2}}$ Value.

8. control device of electric motor according to claim 1 is characterized in that,

Said regression vector is 1 in the gain of said 2 multipliers, with N be made as natural number, when A is made as constant, by the such p dimensional vector Φ of mathematical formulae 5 ^*(N) definition,

Mathematical formulae 5: ${\mathrm{\&Phi;}}^{*}\left(N\right)=[{\mathrm{\&Phi;}}_1^{*}\left(N\right),{\mathrm{\&Phi;}}_2^{*}\left(N\right),...,{\mathrm{\&Phi;}}_p^{*}\left(N\right)]$

The gain of said 2 multipliers is set to

Mathematical formulae 6: $K_g=\frac{A}{\mathrm{Max}\left\{\right|{\mathrm{\&Phi;}}_1^{*}\left(1\right)|,|{\mathrm{\&Phi;}}_2^{*}\left(1\right)|,...,|{\mathrm{\&Phi;}}_p^{*}\left(1\right)\left|\right\}}$

With

Mathematical formulae 7: $K_{\mathrm{\&alpha;}}=\frac{B}{\sqrt{\underset{k=1}{\overset{p}{\mathrm{\&Sigma;}}}{\mathrm{\&Phi;}}_k^{*}{\left(1\right)}^2}}$

In a less side's value.

9. a motor control system possesses: the described control device of electric motor of claim 1; And be inserted in the electric power converter between said current controller and the said motor,

Said electric power converter carries out PWM control by said current controller, and said motor is carried out PWM control.

10. motor control system, possess: motor links with the driven object load; Electric power converter drives said motor; Speed control, according to the deviation between rotary speed value of detecting of rotary speed command value and motor, output torque current command value; Current controller according to said torque current command value and supply to the deviation between torque current value of detecting of said motor, is controlled the output current of said electric power converter; With automatic tuning portion, rotary speed value of detecting of the motor torque value of said motor and said motor as input signal, is adjusted the Control Parameter of said speed control automatically, it is characterized in that,

Said automatic tuning portion comprises:

Import the 1st low pass filter of speed value of detecting of said motor;

Import the 2nd low pass filter of the motor torque of said motor;

The output signal of said the 1st low pass filter is carried out the 1st zeroth order maintenance assembly of periodic sampling and maintenance;

The output signal of said the 2nd low pass filter is carried out the 2nd zeroth order maintenance assembly of periodic sampling and maintenance;

Make said the 1st zeroth order keep the 1st Delay Element of 1 sampling period of output signal delay of assembly; Make said the 2nd zeroth order keep the 2nd Delay Element of 1 sampling period of output signal delay of assembly;

Make the 1st multiplier of the attenuated output signal of said the 1st Delay Element; Make the 2nd multiplier of the attenuated output signal of said the 2nd Delay Element;

Import the output signal of said the 1st multiplier and the output signal of said the 2nd multiplier, generate the regression vector generation portion of regression vector;

Import said regression vector, infer the discrete system parameter estimation portion of parameter vector discretely;

Said parameter vector is converted to the parameter converter section of discrete system/continuous system of the Control Parameter of said speed control; With

Keep the output signal of assembly and output signal that said the 2nd zeroth order keeps assembly as input signal said the 1st zeroth order, export the automatic gain control part of set point of the gain of said the 1st multiplier and said the 2nd multiplier.

11. a motor control system possesses: motor links with the driven object load; Electric power converter drives said motor; Speed control, according to the deviation between speed value of detecting of speed value and said motor, output torque current command value; Current controller according to said torque current command value and supply to the deviation between torque current value of detecting of said motor, is controlled the output current of said electric power converter; With automatic tuning portion, speed value of detecting of the motor torque of said motor and said motor as input signal, is adjusted the Control Parameter of said speed control automatically, it is characterized in that,

Said automatic tuning portion comprises:

Import the 1st low pass filter of speed value of detecting of said motor; Import the 2nd low pass filter of the motor torque of said motor;

The 1st zeroth order that makes said the 1st low-pass filter output signal carry out periodic sampling and keep keeps assembly;

The 2nd zeroth order that makes said the 2nd low-pass filter output signal carry out periodic sampling and keep keeps assembly;

Import said the 1st zeroth order and keep the output signal of assembly and the output signal that said the 2nd zeroth order keeps assembly, generate the regression vector generation portion of regression vector;

Make said the 1st zeroth order keep the 1st multiplier of the attenuated output signal of assembly;

Make the 2nd multiplier of the size decay of the regression vector that said regression vector generation portion generates;

The output signal of the output signal of said the 1st multiplier and said the 2nd multiplier as input, is carried out the discrete system parameter estimation portion of computing to parameter vector;

The parameter vector of said discrete system parameter estimation portion computing is converted to the parameter converter section of discrete system/continuous system of the Control Parameter of speed control; With

Said regression vector as input signal, is exported the automatic gain control part of set point of the gain of said the 1st multiplier and said the 2nd multiplier.

Images