CN101162884A - Inverse model control method of generator excitation system - Google Patents

Abstract

The invention relates to an inverse-model control method for a generator excitation system, including the steps: (1) terminal voltage signal Vt is collected from the generator, and sent to A/D converter for sampling, and Vt (k) is collected after sampling and read into the CPU; (2) the deviation e between the sampling value Vt (k) and the expected value V<*>t (k+1) is calculated, when e is greater than or equal to 0.005, the controlled quantity of excitation is calculated according to Step (3); otherwise, the excitation system does not act; (3) the controlled quantity of excitation is calculated as u(k)=u(k-1)+ triangle u(k), therein, triangle u(k)= V<*>t (k+1)-Vt(k)/f<1>[q(k-1),u(k-1)]; and when | triangle u(k)|is more than Delta (k), triangle u(k)= Delta (k)question mark sign(triangle u(k), herein, sign(question mark) is a symbolic function, Delta (k) is a limit to the physical characteristics of the excitation system; (4) the calculation result triangle u(k) is sent into D/A convertor, and the signal acts onto the excitation system directly. According to the expected value and the calculated value of the terminal voltage of he generator, and inverse-model control of the excitation system is achieved through the control law designed off-line. The invention is easy to implement, the computation is rapid, online study or parameter adjustment is not required, and the generator has excellent terminal-voltage adjusting performance.

Description

The inverse model control method of generator excited system

Technical field

The present invention relates to a kind of inverse model control method of generator excited system.

Background technology

The security and stability problem of power system operation, promptly the dynamic security integrity problem is the major issue of power system operation, is subjected to expert's the attention of going together deeply both at home and abroad always, and has carried out a large amount of research and experimental work.Improving with the main means that improve stability of power system is to adopt novel electric power equipment and control mode.Generating set is the visual plant in the power system operation, its effectively control help stable, safety, the economical operation of electric power system, also be a kind of effective link of improving power system transient stability.

The control of the excitation of generator is to improve one of major measure of transmission system fine motion steady-state level, and the excitation control of existing generator exists less stable, technical problem that control precision is lower.

Summary of the invention

Have less stable, technical problem that control precision is lower in the existing generator excitation control procedure in order to solve, the present invention provides a kind of inverse model control method of simple and reliable, real-time good, control precision is high generator excitation.

The technical scheme that the present invention solves the problems of the technologies described above may further comprise the steps:

(1) from generator collection terminal voltage V _tSignal is sent into A/D converter and is sampled, and obtains V after sampling is finished _t(k), read among the CPU;

(2) calculating sampling value V _t(k) with desired value V _t ^*(k+1) error e, error e were more than or equal to 0.005 o'clock, and (3) calculate the excitation controlled quentity controlled variable set by step; Otherwise excitation system is failure to actuate;

(3) calculating the excitation controlled quentity controlled variable is: u (k)=u (k-1)+Δ u (k), wherein $\mathrm{\&Delta;u}\left(k\right)=\frac{V_t^{*}(k+1)-V_t\left(k\right)}{{\hat{f}}^1[q(k-1),u(k-1)]},$ And work as | Δ u (k) |＞δ (k), Δ u (k)=δ (k) sign (Δ u (k)), sign () is a sign function here, δ (k) is an amplitude limit according to the excitation system physical characteristic;

(4) the Δ u (k) that calculates is sent into D/A converter, directly act on excitation system with this signal.

In the excitation inverse model control method of above-mentioned generator, the excitation controlled quentity controlled variable in the described step (3) is: $u_c\left(k\right)=u(k-1)+\mathrm{\&Delta;u}\left(k\right)-\frac{F\left(z\right)\mathrm{\&zeta;}\left(k\right)}{{\hat{f}}^1[q(k-1),u(k-1)]}.$

Inverse model control method is a kind of research feedback linearization method more widely, utilizes the inversion model of controlled device and master mould to be composed in series a pseudo-linear system, realizes the linearisation of non-linear object and the decoupling zero between the passage effectively.The excitation system Inverse Model Control device of this paper is made up of two parts, a part is based on the inversion model of the equivalent linearization of Taylor expansion, with this as feedforward controller, second portion is a feedback element, adopt a kind of robust filter as feedback element, improve the terminal voltage pressure regulation precision and the stability of a system.

Technique effect of the present invention is: (1) this Inverse Model Control device is made up of feedforward, two links of feedback, feedforward is fed back to robust filter, the combination of the two for the inversion model of excitation system, not only have reliable precision and stability, and do not need on-line study or parameter adjustment; (2) feedforward controller, i.e. the inversion model of excitation system is by a kind of inearized model based on the Taylor expansion, and it is equivalent to a kind of inearized model, calculates easyly, is convenient to actual realization, real-time is more excellent; (3) adopt SVMs to come the input/output relation (excitation controlled quentity controlled variable-terminal voltage relation) of generator excited system, identification precision is higher, and the learning training time of algorithm of support vector machine is short, and the model ability is strong.(4) the SVMs selection of parameter is regarded Combinatorial Optimization as, needn't consider the complexity of model and the dimension of variable and become yardstick chaos optimization algorithm.

Description of drawings

Fig. 1 realizes the structure chart of the inventive method.

Among Fig. 2 the present invention based on the generator system identification structure figure of SVMs.

Generator voltage response curve among Fig. 3 the present invention.

Generator's power and angle response curve among Fig. 4 the present invention.

Generator active power response curve among Fig. 5 the present invention.

Generator angular speed response curve among Fig. 6 the present invention.

Embodiment

Be described further below with reference to the drawings and specific embodiments.

Referring to Fig. 1, realize that the engine exciter control system of the inventive method comprises feedforward Inverse Model Control device and two parts of feedback compensation.The physical significance of each variable is among the figure: V _t ^*(k) expression generator voltage desired value, V _t(k) expression generator voltage measured value,

Expression generator voltage identifier; u _c(k) the synthetic actual excitation controlled quentity controlled variable of expression, u _f(k) expression FEEDBACK CONTROL amount, u (k) expression feedfoward control amount.The Inverse Model Control device obtains the equivalent linearization inversion model of generator excited system, is used to offset the nonlinear characteristic of excitation system, and that the mode of feedfoward control is calculated is easy, be convenient to realize, and can improve quick follow-up control.Feedforward Inverse Model Control device and excitation system are together in series, and constitute a quasi-linear system.Feedback compensation will adopt a robust filter to improve the control precision and the stability of a system, eliminate the influence of inversion model modeling error to a certain extent.

The control law of feedforward Inverse Model Control device.According to three rank dynamic model equations of generator excited system, adopt discretization method obtain-the discrete NARMA model of kind of equivalence is:

V _t(k+1)＝f[V _t(k)，V _t(k-1)，V _t(k-2)，u(k)] (1)

K represents current time in the following formula, and k-1, k+1 represent previous moment and back one respectively constantly, V _t(k+1) expression generator voltage, u (k) expression generator excitation controlled quentity controlled variable.Feedforward Inverse Model Control device is taked a kind of equivalent linearization inversion model based on the Taylor expanded type, and contrary control law only needs the input/output relation of identification excitation system, has avoided complicated Inverse Model Control rule to derive.With the NARMA model representation of the excitation system shown in (1) is V _t(k+1)=f[V _t(k), V _t(k-1), V _t(k-2), u (k)]=f[q (k), u (k)], q (k)=[V wherein _t(k), V _t(k-1), V _t(k-2)], its Taylor expanded type is expressed as:

$V_t(k+1)=f[q\left(k\right),u\left(k\right)]=f[q(k-1),u(k-1)]+\underset{r=1}{\overset{\&infin;}{\mathrm{\&Sigma;}}}\frac{1}{r!}\frac{{\&PartialD;}^{r}f[q(k-1),u(k-1)]}{\&PartialD;u{(k-1)}^r}{[u\left(k\right)-u(k-1)]}^r$

$+\underset{r=1}{\overset{\&infin;}{\mathrm{\&Sigma;}}}\frac{1}{r!}\frac{{\&PartialD;}^{r}f[q(k-1),u(k-1)]}{\&PartialD;q{(k-1)}^r}{[q\left(k\right)-q(k-1)]}^r---\left(2\right)$

Theoretical according to Taylor expansion, can omit the residual item and obtain the input/output relation of excitation system equivalence, promptly formula (2) can be reduced to: V _t(k+1) ≈ V _t(k)+f ¹[q (k-1), u (k-1)] Δ u (k) (3)

Wherein $f^1[q(k-1),u(k-1)]=\frac{\&PartialD;f[q(k-1),u(k-1)]}{\&PartialD;u(k-1)}---\left(4\right)$

F in the following formula ¹[q (k-1), u (k-1)] is a nonlinear function, and according to the input/output signal of excitation system, the present invention estimates f by SVMs (SVM) ¹[q (k-1), u (k-1)], the model representation that SVM is estimated is ${\hat{f}}^1[q(k-1),u(k-1)].$ Like this, excitation control increment Δ u (k) just can directly calculate according to formula (3).

u(k)＝u(k-1)+Δu(k) (5)

When | Δ u (k) |≤δ (k), $\mathrm{\&Delta;u}\left(k\right)=\frac{V_t^{*}(k+1)-V_t\left(k\right)}{{\hat{f}}^1[q(k-1),u(k-1)]};$ When | Δ u (k) |＞δ (k), Δ u (k)=δ (k) sign (Δ u (k)).

Here V _t ^*(k+1) be the terminal voltage desired value, sign () is a sign function, and δ (k) is an amplitude limit according to excitation system reality.Why to be because consider the physical property of excitation system, and avoid significantly shaking as far as possible with Δ u (k) amplitude limit in δ (k).

Above-mentioned Inverse Model Control device is open loop, and the stability of a system is not strong, and parameter uncertainty in the working control and external interference are inevitable.Thereby take a feedback compensation structure to improve the stability of a system and control precision, feedback compensation adopts robust filter shown in Figure 1 to realize.By system configuration shown in Figure 1 as can be known:

V _t(k+1)＝V _t(k)+f ¹[q(k-1)，u(k-1)]·(Δu(k)+Δu _f(k))+R+γ (6)

Δ u wherein _f(k) be the increment output of robust filter, R represents that the residual item of Taylor expansion and γ are interference and indeterminate.Utilize ζ (k)=z ^-1(R+ γ+ε (k)), ${\mathrm{\&Delta;u}}_f\left(k\right)=-F\left(z\right)\mathrm{\&zeta;}\left(k\right)/{\hat{f}}^1[q(k-1),u(k-1)],$ Then

V _t(k+1)＝V _t(k)+f ¹[q(k-1)，u(k-1)]·Δu(k)-β(k)F(z)ζ(k)+R+γ

＝V _t(k)+f ¹[q(k-1)，u(k-1)]·Δu(k)-F ₁(R+γ) (7)

F wherein ₁=1-β (k) z ^-1F (z).

Therefore, adopt suitable filters F (z), residual item R and interference γ can offset to a certain extent, thereby improve control precision.At this moment, by the synthetic as can be known control law of control system structure shown in Figure 1 be:

$u_c\left(k\right)=u(k-1)+\mathrm{\&Delta;u}\left(k\right)-\frac{F\left(z\right)\mathrm{\&zeta;}\left(k\right)}{{\hat{f}}^1[q(k-1),u(k-1)]}---\left(8\right)$

Comprised Inverse Model Control device and feedback compensation two parts in the control law this moment, thereby had the advantage of Inverse Model Control and FEEDBACK CONTROL concurrently.Inverse Model Control has very good Nonlinear Processing ability, and the feedfoward control mode has improved the response speed and the quick follow-up control of system.FEEDBACK CONTROL can compensate the error in the Inverse Model Control, improves control precision to a certain extent, and the link of filter form, and on-line calculation is less.

Among the present invention ${\hat{f}}^1[q(k-1),u(k-1)]$ (support vector machines SVM) estimates, SVM is a kind of novel machine learning algorithm, and learning training speed is fast, and the model ability is strong, is considered a kind of perfect and raising into artificial neural net to adopt SVMs.Its structure is seen Fig. 2.The result of SVM identification sets up the generator system model of estimation with regression function as the formula (9):

${\hat{V}}_t(k+1)=\hat{f}[q\left(k\right),u\left(k\right)]=\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_{t}K[q\left(k\right),u\left(k\right),(q\left(t\right),u\left(t\right))]+b---\left(9\right)$

So just, can obtain in the contrary control law calculating according to System Discrimination ${\hat{f}}^1[q(k-1),u(k-1)]:$

${\hat{f}}^1[q(k-1),u(k-1)]=\frac{\&PartialD;\hat{f}[q(k-1),u(k-1)]}{\&PartialD;u(k-1)}=-\frac{u(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_{t}K\{[q(k-1),u(k-1)],[q\left(t\right),u\left(t\right)]\}---\left(10\right)$

SVMs selection of parameter method based on the chaos optimization algorithm.The present invention regards the selection of parameter of SVM the Combinatorial Optimization of parameter as, and combinatorial optimization problem is set up target function, adopts a kind of change yardstick chaos optimization algorithm to search for the optimal target functional value, thereby finds the suitable parameters value.

Here adopt the Logistic mapping shown in (11) formula: t _K+1=μ t _k(1-t _k) (11)

Get μ=4, system is in chaos state fully.With the equal error (MSE) in side $\frac{1}{n}{\mathrm{\&Sigma;}}_{l=1}^n{(y-f\left(x\right))}^2$ Describe the deviation between SVM recurrence and the reference model, wherein n is a number of samples, and y is a reference model, and f (x) is that SVM returns.

The target function of optimized Algorithm is: choose best parameter combinations, make SVM return with reference model and have the minimum equal error in side, that is:

$\left\{\begin{array}{c}\min f(z_l,\&CenterDot;\&CenterDot;\&CenterDot;{,z}_l)=\min \mathrm{MSE}\\ a_l\&le;z_l\&le;b_l,i=\mathrm{1,2,3}\end{array}\right.---\left(12\right)$

Here optimize variable z _lBe total up to 3, corresponding to parameter c, σ and ε; [a _l, b _l] be each variable z _lThe domain of definition, for different reference models and sample data, the domain of definition space of variable is also different.

Become yardstick chaos optimization algorithm steps and be described below (in the algorithm steps, i=1,2,3):

Step 1 initialization.K=0, fine searching sign r=0, $t_l^k=t_l\left(0\right),$ $t_l^{*}=t_l\left(0\right),$ $a_i^r=a_i,$ $b_i^r=b_i,$ Current optimal objective function value f ^*Be initialized as big positive number;

Step 2 is t _l ^kBe mapped to optimization variable-value interval and become z _l ^k

$z_l^k=a_l^r+(b_l^r-a_l^r)\&CenterDot;t_l^k---\left(13\right)$

Step 3 optimization searching.If $f\left(z_l^k\right)\&le;f^{*},$ Then $f^{*}=f\left(z_l^k\right),$ $t_l^{*}=t_l^k;$ Otherwise continue.

Step 4k:=k+1, $t_l^k:=\mathrm{\&mu;}\&CenterDot;t_l^k(1-t_l^k)$

Step 5 repeating step 2-4, f in certain step number ^*Till remaining unchanged, carry out following steps then.

Step 6 is dwindled the hunting zone of variable

$a_l^{r+1}=z_l^{*}-\mathrm{\&phi;}\&CenterDot;(b_l^r-a_l^r),$ $b_i^{r+1}=z_l^{*}+\mathrm{\&phi;}\&CenterDot;(b_l^r-a_l^r)---\left(14\right)$

φ ∈ (0,0.5) wherein, $z_l^{*}=a_l^r+t_l^{*}(b_l^r-a_l^r)$ Be current optimal solution.In order to guarantee that new range does not cross the border, do following the processing:

If $a_l^{r+1}<a_l^r,$ Then $a_l^{r+1}=a_l^r;$ If $b_l^{r+1}>b_l^r,$ Then $b_l^{r+1}=b_l^r.$ Simultaneously, according to new range with z _l ^*Be mapped to t _l ^*

The new Chaos Variable mt that step 7 is determined with (15) formula _l ^k

${\mathrm{mt}}_l^k=(1-\mathrm{\&alpha;})t_l^{*}+\mathrm{\&alpha;}\&CenterDot;t_l^k---\left(15\right)$

Wherein α is a less positive number.Repeating step 2-4 operation f in certain step number ^*Till remaining unchanged, carry out following steps then.

Step 8r:=r+l reduces α value repeating step 6-7 operation.

Behind repeating step 8 several times, finish optimizing and calculate z _l ^*The optimized parameter value that searches exactly.

The related parameter values of generating set is in the emulation experiment here: x _d=2.156; x _q=2.101, x _d'=0.265; x _T=0.1; x _L=1.46; D=5; H=8; T _DO'=10; P _m=0.6.According to the characteristics of excitation system, determine the input signal of enough abundant pumping signal as system.Control signal u (k) between 0-3 is input in the object with amplitude range, according to these control inputs, thereby produces corresponding generator output V _t(k), set up sample data to { q (k), u (k) } with this.In emulation, SVM training sample data are selected the Gaussian kernel function for use to being 600, and the LS-SVM parameter is: σ=1.5, λ=200.Robust filter is chosen as F (z)=1-r ₁/ 1-r ₁z ¹, and r ₁=0.72.Inverse Model Control system and other two kinds of controllers of being studied have been given contrast simulation, and one is traditional AVR adjuster, another one neuron excitation controller.In analogous diagram, traditional controller (CONVC) is described with dashed lines, and neuron control device (NNC) will represent that the Inverse Model Control device (SVMMC) that approaches based on SVMs of the present invention is then described with heavy line with fine line.

In experiment, generating set is in steady operation, P _t=1.0 (perunit values), Q _t=0.18 (perunit value).When t=2s, give to rise with reference to 5% step of one of exciting voltage, when t=10s, give again to descend with reference to 5% step of one of exciting voltage.Each state of generator when Fig. 3-Fig. 6 has described different controller becomes Vt, δ, P, ω transient response curve, compares with other controller, and the performance of Inverse Model Control of the present invention system is more excellent.

Claims (2)

1. the excitation inverse model control method of a generator may further comprise the steps:

(1) from generator collection terminal voltage V _tSignal is sent into A/D converter and is sampled, and obtains V after sampling is finished _t(k), read among the CPU;

(2) calculating sampling value V _t(k) with desired value V _t ^*(k+1) error e, error e were more than or equal to 0.005 o'clock, and (3) calculate the excitation controlled quentity controlled variable set by step; Otherwise excitation system is failure to actuate;

(3) calculating the excitation controlled quentity controlled variable is: u (k)=u (k-1)+Δ u (k), wherein $\mathrm{\&Delta;u}\left(k\right)=\frac{V_t^{*}(k+1)-V_t\left(k\right)}{{\hat{f}}^1[q(k-1),u(k-1)]},$ And work as | Δ u (k) |＞δ (k), Δ u (k)=δ (k) sign (Δ u (k)), sign () is a sign function here, δ (k) is an amplitude limit according to the excitation system physical characteristic;

(4) the Δ u (k) that calculates is sent into D/A converter, directly act on excitation system with this signal.

2. the excitation inverse model control method of generator according to claim 1, the excitation controlled quentity controlled variable in the described step (3) is: $u_c\left(k\right)=u(k-1)+\mathrm{\&Delta;u}\left(k\right)-\frac{F\left(z\right)\mathrm{\&zeta;}\left(k\right)}{{\hat{f}}^1[q(k-1),u(k-1)]}.$

Images