CN102799785B - Based on nuclear power generating sets prime mover of simplicial method and the method for governor parameter identification thereof - Google Patents

Abstract

The invention provides a kind of nuclear power generating sets prime mover based on simplicial method and governor parameter discrimination method thereof, its feature is simplicial method to be used for nuclear power generating sets prime mover and governor parameter identification thereof, amount of calculation can be reduced, comparatively rapid convergence, and can avoid, to problems such as objective function differentiates, obtaining the analytical form of objective function.Described nuclear power generating sets prime mover based on simplicial method and governor parameter discrimination method very fast to controling parameters optimizing convergence ratio, amount of calculation is little, can the parameter of the effectively actual nuclear power generating sets prime mover of identification and speed regulator thereof, there is important engineer applied and be worth.

Description

Based on nuclear power generating sets prime mover of simplicial method and the method for governor parameter identification thereof

Technical field

The present invention relates to a kind of method of nuclear power generating sets prime mover and governor parameter identification thereof, especially relate to a kind of based on nuclear power generating sets prime mover of simplicial method and the method for governor parameter identification thereof.

Background technology

The basic tool analyzing electric system is digital simulation software, the power system analysis software of main flow all supports prime mover and the governor model thereof of multiple nuclear power generating sets, but in parameter, all without actual verification " representative value ", when studying dynamo-electric transient stability, may can not bring very large error, but, for the bulk power grid that extra-high voltage grid is set up, the impact of prime mover and speed regulator thereof will progressively display, and therefore it may be necessary measured data and pick out actual parameter value replacement current " representative value ".

Least square method is a kind of classical effective parameter identification method, but when very large, the to be identified number of parameters of measured data amount is more, calculated amount will be very large.Method of steepest descent and method of conjugate gradient are the effective ways of univariate parameter identification, but these two kinds of methods all require the gradient of calculating target function, and in many optimization problem, in fact usually can not get the analytical form of objective function, often speed of convergence is very slow near optimum point for method of steepest descent in addition, for obtaining higher low optimization accuracy, often to search for many times near optimum point.

Summary of the invention

Technical matters to be solved by this invention, just be to provide a kind of based on nuclear power generating sets prime mover of simplicial method and the method for governor parameter identification thereof, the present invention does not need the derivative obtaining objective function, namely when larger, the to be identified number of parameters of convenient measured data amount is more, calculated amount is also less, speed of convergence is very fast, and searching times is less, and the parameter picked out can ensure that prime mover and governing system model thereof can accurately for bulk power grid.

Solve the problem, the present invention adopts following technical scheme:

Based on nuclear power generating sets prime mover of simplicial method and a method for governor parameter identification thereof, comprise the following steps:

The given α of S1 ₀, λ, μ, ε, N, K, K ₁=0,

Wherein: α ₀for needing one group of estimated value of identified parameters, μ is broadening factor, and μ is compressibility factor, and ε is convergence, and N equals to need identified parameters number, and K is maximum iteration time;

S2 determines initial simplex, calculates α _i=α ₀+ h*e _i, i=1,2 ... n (2);

S3 target function type (1), calculates mean deviation corresponding to each group of parameter according to objective function, calculates C _i=Q (α _i) (i=1,2 ..., N);

S4 finds out α _h, α _l, α _g;

S5 iterations adds 1, K ₁=K ₁+ 1;

S6 judges | C _h-C _l| < ε C _lif set up, export successful α _l, C _l, terminate;

S7 judges K ₁if >K sets up, export failed α _l, CL, terminates; Or: return step S1 given α again ₀, λ, μ, ε, N, K, K ₁=0, re-start parameter identification;

S8 computational reflect point ${\mathrm{\&alpha;}}_R=\frac{2}{N}*(\underset{i=0}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_H)-{\mathrm{\&alpha;}}_H$ And C _r=Q (α _r);

S9 judges C _r<C _gs15 is gone to step if be false;

S10 calculates and judges (1-μ) * C _h+ μ * C _r<C _lif set up and go to step S13;

S11 makes $\begin{array}{c}{\mathrm{\&alpha;}}_S={\mathrm{\&alpha;}}_R\\ C_S=C_R\end{array};$

S12 replaces worst point $\begin{array}{c}{\mathrm{\&alpha;}}_H={\mathrm{\&alpha;}}_S\\ C_H=C_S\end{array},$ Return S4;

S13 calculates $\begin{array}{c}{\mathrm{\&alpha;}}_E=(1-\mathrm{\&mu;})*{\mathrm{\&alpha;}}_H+\mathrm{\&mu;}*{\mathrm{\&alpha;}}_R\\ C_E=Q\left({\mathrm{\&alpha;}}_E\right)\end{array};$

S14 judges C _e<C _rif set up, go to step S11, otherwise go to step S12;

S15 R, H point exchanges;

S16 calculates α _s=(1-λ) * α _h+ λ * α _rand CS;

S17 judges C _s<C _gif set up and go to step S12;

S18 calculates α _i=(α _l+ α _i)/2(i=1,2 ..., N), go to step S3.

The mean deviation that described objective function is defined as simulation curve and measured curve divided by the absolute value of the maximum deviation of measured power during emulating and its initial value, that is:

$e=\left(\frac{1}{T}\underset{0}{\overset{T}{\&Integral;}}\right|P_{\mathrm{sim}}-P_{\mathrm{mes}}\left|\mathrm{dt}\right)/\left|{\mathrm{\&Delta;P}}_{\max }\right|---\left(1\right)$

Wherein T is simulation time length, P _simfor simulation value, P _mesfor measured value, | Δ P _max| be the absolute value of the maximum deviation of measured power and its initial value.

Estimated value in described step 1 gets the representative value of nuclear power generating sets prime mover and governor model thereof.

Implication in above-mentioned concrete steps is explained as follows:

One, initial simplex is determined

Suppose that certain nuclear power generating sets prime mover and governor model thereof have n parameter to need identification, then initial simplex should be made up of n+1 point (parameter value of the corresponding one group of prime mover of each point and speed regulator thereof), and this n+1 the necessary Line independent of point, otherwise probably search for less than minimal point.General, first can give one group of estimated value (" representative value " of nuclear power generating sets prime mover and governor model thereof) to this n parameter, then change each parameter respectively to obtain a new point, specifically, if select initial estimate α ₀, then α ₁, α ₂..., α _nfor:

α _i=α ₀+h*e _i,i＝1,2,…n （2）

In above formula, h for a change measures; e _ifor n-dimensional vector, except i-th element is 1, other is zero, i.e. e _i=[0,0 ..., 0,1,0 ..., 0]

Initial estimation point α ₀add n the point obtained like this, just constitute initial simplex.

Two, the determination of sublating and newly putting of old point

If objective function gets certain group parameter alpha _htime its value maximum, then α _hthe point that will remove exactly; New point is generally taken on " opposite " of the point be removed, and becomes reflection spot, for:

${\mathrm{\&alpha;}}_R=\frac{2}{n}*(\underset{i=0}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_H)-{\mathrm{\&alpha;}}_H---\left(3\right)$

The expansion of three, newly putting, compression and simplex are shunk

If α _hfor worst point (objective function C _hvalue maximum), α _lfor the most better (objective function C _lvalue minimum), α _gfor secondary bad point (C _gcompare C _hlittle, but all larger than the target function value of other each point).

If new point (reflection spot) α _rfunctional value C _rbe less than C _g, this illustrates α _rpoint may advance not, and now can readvance along reflection direction, this is called expansion, obtains α _e:

α _E=(1-μ)*α _H+μ*α _R,μ>1 （4）

Otherwise, if C _rbe greater than C _g, α is described _rpoint advances too far away, needs compression, namely retreats along original reflection direction, obtain α _s:

α _S=(1-λ)*α _H+λ*α _R（5）

λ is compressibility factor, and span between 0 to 1, but can not equal 0.5, otherwise can reduce the space dimensionality of simplex, is unfavorable for search.

If C after compression _sstill C is greater than _g, illustrate that original simplex obtains too large, all limits all can be reduced to form new simplex, be called contraction.Concrete grammar is:

${\mathrm{\&alpha;}}_i=\frac{{\mathrm{\&alpha;}}_L+{\mathrm{\&alpha;}}_i}{2},i=\mathrm{1,2},...n---\left(6\right)$

Then return step 1 and continue cycle calculations smallest point.

Four, convergence is judged

If ε is convergence, K is maximum search number of times, if | (C _h-C _l)/C _l| < ε, then illustrate and search for successfully, can think α _lfor the most better; If still do not meet above formula through K search, then search for failure; Now can attempt returning step 1 and adjust initial estimation point α ₀, and identified parameters h, μ, λ, re-start parameter identification.

Principle of the present invention: function derivative is its important feature, the opposite direction of such as gradient is exactly the direction of steepest descent of this function near this point.If for a certain reason, can not obtain gradient information, then first can calculate the functional value at several some places, these points constitute initial simplex.Then they are compared, just can infer the approximate trend of function from the magnitude relationship between them, for the descent direction seeking function provides reference.Then in these being put, maximum one of functional value removes, and determines a new point by certain algorithm simultaneously, constitutes a new simplex.So circulation is gone down always, the point that finally functional value just can be found minimum.

Beneficial effect: of the present invention based on nuclear power generating sets prime mover of simplicial method and the method for governor parameter identification thereof, do not need the derivative obtaining objective function, without the need to target, searching times is less, very fast to controling parameters optimizing convergence ratio, amount of calculation is less, and the parameter picked out can ensure that prime mover and governing system model thereof can accurately for bulk power grid.

Accompanying drawing explanation

Below in conjunction with accompanying drawing, the method for simplicial method to nuclear power generating sets prime mover and governor parameter identification thereof that utilize that the present invention proposes is described in detail.

The process flow diagram of Fig. 1 simplex Identification of parameter of the present invention;

Fig. 2 GGOV1 prime mover-governor model transport function figure;

Fig. 3 measured power curve and simulation curve.

Embodiment

Of the present invention based on nuclear power generating sets prime mover of simplicial method and the method for governor parameter identification thereof, comprise the following steps:

The given α of S1 ₀, λ, μ, ε, N, K, K ₁=0,

Wherein: α ₀for needing one group of estimated value of identified parameters, μ is broadening factor, and μ is compressibility factor, and ε is convergence, and N equals to need identified parameters number, and K is maximum iteration time;

S2 determines initial simplex, calculates α _i=α ₀+ h*e _i, i=1,2 ... n (2);

S3 target function type (1), calculates mean deviation corresponding to each group of parameter according to objective function, calculates C _i=Q (α _i) (i=1,2 ..., N);

S4 finds out α _h, α _l, α _g;

S5 iterations adds 1, K ₁=K ₁+ 1;

S6 judges | C _h-C _l| < ε C _lif set up, export successful α _l, CL, terminates;

S7 judges K ₁if >K sets up, export failed α _l, CL, terminates; Or: return step S1 given α again ₀, λ, μ, ε, N, K, K ₁=0, re-start parameter identification;

S8 computational reflect point ${\mathrm{\&alpha;}}_R=\frac{2}{N}*(\underset{i=0}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_i-{\mathrm{\&alpha;}}_H)-{\mathrm{\&alpha;}}_H$ And C _r=Q (α _r);

S9 judges C _r<C _gs15 is gone to step if be false;

S10 calculates and judges (1-μ) * C _h+ μ * C _r<C _lif set up and go to step S13;

S11 makes $\begin{array}{c}{\mathrm{\&alpha;}}_S={\mathrm{\&alpha;}}_R\\ C_S=C_R\end{array};$

S12 replaces worst point $\begin{array}{c}{\mathrm{\&alpha;}}_H={\mathrm{\&alpha;}}_S\\ C_H=C_S\end{array},$ Return S4;

S13 calculates $\begin{array}{c}{\mathrm{\&alpha;}}_E=(1-\mathrm{\&mu;})*{\mathrm{\&alpha;}}_H+\mathrm{\&mu;}*{\mathrm{\&alpha;}}_R\\ C_E=Q\left({\mathrm{\&alpha;}}_E\right)\end{array};$

S14 judges C _e<C _rif set up, go to step S11, otherwise go to step S12;

S15 R, H point exchanges;

S16 calculates α _s=(1-λ) * α _h+ λ * α _rand CS;

S17 judges C _s<C _gif set up and go to step S12;

S18 calculates α _i=(α _l+ α _i)/2(i=1,2 ..., N), go to step S3.

The mean deviation that described objective function is defined as simulation curve and measured curve divided by the absolute value of the maximum deviation of measured power during emulating and its initial value, that is:

$e=\left(\frac{1}{T}\underset{0}{\overset{T}{\&Integral;}}\right|P_{\mathrm{sim}}-P_{\mathrm{mes}}\left|\mathrm{dt}\right)/\left|{\mathrm{\&Delta;P}}_{\max }\right|$

Wherein T is simulation time length, P _simfor simulation value, P _mesfor measured value, | Δ P _max| be the absolute value of the maximum deviation of measured power and its initial value.

Estimated value in described step 1 gets the representative value of nuclear power generating sets prime mover and governor model thereof.

Below to illustrate the parameter identification of nuclear power generating sets GGOV1 prime mover and governor model thereof, in this model, need the parameter of identification to be Kpgov(speed regulator scale-up factor in permanent speed regulation r, PID controller) and Kigov(speed regulator integral coefficient), Tb(steam turbine lag time constant in prime mover) and load governor in Kimw(load governor gain coefficient)

Prime mover and governor parameter thereof of newly calculating are organized for each, needs the bound checking each parameter.In GGOV1 model, if R<0.04, then R=0.04; If Kpgov<0.1, then Kpgov=0.1; If Kigov<0.02, then Kigov=0.02; If Tb>30, then Tb=30; If kimw<0, then kimw=0.

For the GGOV1 model of certain power plant and power measured value, the initial value of above-mentioned 5 parameters respectively:

R	Kpgov	Kigov	Tb	Kimw
					0.082	19.64	2.55	3.2	0.0019

Maximum iteration time K=100 is set, broadening factor μ=1.5, compressibility factor λ=0.7, convergence criterion epsilon=0.005.

Step 1: initial simplex.

	R	Kpgov	Kigov	Tb	Kimw
						α 0	0.082	19.64	2.55	3.2	0.0019
α 1	0.092	19.64	2.55	3.2	0.0019
						α 2	0.082	20.14	2.55	3.2	0.0019
α 3	0.082	19.64	2.75	3.2	0.0019
						α 4	0.082	19.64	2.55	3.7	0.0019
α 5	0.082	19.64	2.55	3.2	0.0039

Step 2: the determination of sublating and newly putting of old point.According to objective function (1), utilize equivalent two machine systems, can calculate in the mean deviation often organizing simulation curve and measured curve under prime mover governor parameter.

	α 1	α 2	α 3	α 4	α 5
						Mean deviation	0.039579	0.064168	0.039807	0.040654	0.038724

According to mean deviation, α ₂should be the parameter that will sublate, according to formula (3), computational reflect point:

	R	Kpgov	Kigov	Tb	Kimw
						α R	0.086	19.84	2.63	3.4	0

Step 3: the expansion of newly putting, compression and simplex are shunk

Reflection spot error is 0.301678, is greater than the mean deviation of worst point, therefore needs compression, calculates compression point according to formula (5):

	R	Kpgov	Kigov	Tb	Kimw
						α S	0.0832	19.7	2.574	3.26	0.00273

Compression point error is 0.137663, is greater than the mean deviation of worst point, and therefore simplex needs to shrink, and calculates the simplex after shrinking according to formula (6):

	R	Kpgov	Kigov	Tb	Kimw
						α 0	0.082	19.64	2.55	3.45	0.0019
α 1	0.087	19.64	2.55	3.45	0.0019
						α 2	0.082	19.89	2.55	3.45	0.0019
α 3	0.082	19.64	2.65	3.45	0.0019
						α 4	0.082	19.64	2.55	3.7	0.0019
α 5	0.084	19.74	2.59	3.55	0.00095

Return step 2, cycle calculations like this, eventually pass 9 loop iteration convergences, nuclear power generating sets prime mover and governor parameter identification result thereof:

R	Kpgov	Kigov	Tb	Kimw
					0.082046	19.648594	2.640938	3.458594	0.001879

The mean deviation of the final argument that identification obtains is 0.03722713726.

Measured power curve and simulation curve are shown in accompanying drawing 3.

Claims (2)

1., based on nuclear power generating sets prime mover of simplicial method and a method for governor parameter identification thereof, comprise the following steps:

The given α of S1 ₀, λ, μ, ε, N, K, K ₁=0,

Wherein: α ₀for needing one group of estimated value of identified parameters, μ is broadening factor, and λ is compressibility factor, and ε is convergence, and N equals to need identified parameters number, and K is maximum iteration time, K ₁for current iteration number of times;

S2 determines initial simplex, calculates α _i=α ₀+ h*e _i, i=1,2 ..., N, α _ifor the point in simplex, h for a change measures; e _ifor N dimensional vector, except i-th element is 1, other is zero,

I.e. e _i=[0,0 ..., 0,1,0 ..., 0]; (2);

S3 calculates mean deviation corresponding to each group of parameter according to target function type (1), calculates C _i=Q (α _i), i=1,2 ..., N; C _ifor α _icorresponding functional value, Q (α _i) be objective function;

S4 finds out α _h, α _l, α _g, α _hfor worst point, α _lfor the most better, α _gfor secondary bad point;

S5 iterations adds 1, K ₁=K ₁+ 1;

S6 judges | C _h-C _l| < ε C _l, wherein C _hfor the maximal value of objective function, C _lfor the minimum value of objective function; If set up, export successful α _l, C _l, terminate;

S7 judges K ₁> K, if set up, exports failed α _l, C _l, terminate; Or return step S1 given α again ₀, λ, μ, ε, N, K, K ₁=0, re-start parameter identification;

S8 computational reflect point and C _r=Q (α _r), wherein α _rrepresent reflection spot, C _rrepresent reflection spot target function value;

S9 judges C _r< C _gif be false and go to step S15;

S10 calculates and judges (1-μ) * C _h+ μ * C _r< C _lif set up and go to step S13;

S11 makes

S12 replaces worst point return S4;

S13 calculates wherein α ^efor inflexion point, C _efor inflexion point target function value;

S14 judges C _e< C _rif set up, go to step S11, otherwise go to step S12;

S15R, H point exchanges;

S16 calculates α _s=(1-λ) * α _h+ λ * α _rand C _s, α _sfor compression point, C _sfor compression point target function value;

S17 judges C _s< C _gif set up and go to step S12;

S18 shrinks simplex, now α _i=(α _l+ α _i)/2, i=1,2 ..., N, goes to step S3;

The mean deviation that described objective function is defined as simulation curve and measured curve divided by the absolute value of the maximum deviation of measured power during emulating and its initial value, that is:

Wherein T is simulation time length, P _simfor simulation value, P _mesfor measured value, | Δ P _max| be the absolute value of the maximum deviation of measured power and its initial value.

2. according to claim 1 based on nuclear power generating sets prime mover of simplicial method and the method for governor parameter identification thereof, it is characterized in that: the estimated value in described step S1 gets the representative value of nuclear power generating sets prime mover and governor model thereof.