CN104156504A - Parameter identifiability judgment method for generator excitation system - Google Patents

Abstract

The invention discloses a parameter identifiability judgment method for a generator excitation system and relates to judgment of parameter identifiability of the generator excitation system. The parameter identifiability judgment method includes steps of judging relevance parameters by setting up a parameter time domain flexibility matrix, classifying the parameters to be identified into a well-conditioned parameter set and a bad-conditioned parameter set, taking the smallest value of the sum of simulated-frequency-domain flexibility in the bad-conditioned parameter set as a valuation representative, valuing the valuation representative with an empirical value and identifying parameters of the corresponding well-conditioned parameter set. The parameter identifiability judgment method has the advantages that accuracy and stability of parameter judgment results of the generator excitation system are improved, safe and stable operation of the power system is benefited, and the parameter identifiability judgment method has higher engineering practice value.

Description

A kind of generator excited system parameter identifiability decision method

Technical field

The present invention relates to parameters of electric power system identification, more specifically relate to generator excited system parameter identifiability decision method.

Background technology

The accuracy of generator excited system model structure and parameters is electric system simulation computational analysis basis, do a lot of work for Excitation System Modeling both at home and abroad, the international organizations such as IEEE are by IEEE Recommended Practice for Excitation System Models for Power System Stability Studies, IEEE Standard421.5-2005. and China Electric Power Research Institute are by mathematical model of excitation system expert group. calculate mathematical model of excitation system [J] for power system stability. and Proceedings of the CSEE, 1991, 19 (5): 65-71. has proposed multiple standards model structure for excitation system, wherein excitation system model parameter is mainly to be obtained by identification algorithm by field test data.Li Defeng etc. the identifiability research [J] of aggregate power load model. Automation of Electric Systems, 1997,21 (7): 10-14. and Ju Ping. electric system Modeling Theory and method [M]. Science Press, 2010. research and practice shows: although utilize the field test data that model parameter that discrimination method obtains can fine matching system, still potentially unstable of the identification result of some parameter.This phenomenon belongs to the Identifiability Problem of parameter, if the structures shape of model itself parameter can not be by unique identification, it is inaccurate only carrying out identified parameters by measurement data.Therefore, carry out the identifiability of parameter and judge it is the basis of carrying out parameter identification work, caused extensive attention and concern.

Parameter Identifiability Problem is mainly divided into two kinds according to the difference of analytical approach: analytical method and Sensitivity Method.Bring up equality. Approaches To Identifiability Analysis of Electric Load Models [J]. Automation of Electric Systems, 1999,23 (19): first 29-33. has proposed analytical method, analytical method is according to the state equation of model and transport function, parses the relation between parameter.For analytical method, along with the increase of model order and identified parameters number, it is realized difficulty and sharply increases, and greatly reduces using value and the usable range of himself.Therefore, the parameter identifiability research based on sensitivity analysis has obtained broad development.Xie Huiling etc. the parameters of electric power system identification analysis [J] based on Calculation of Sensitivity. Automation of Electric Systems, 2009,33 (7): 17-21. has proposed Sensitivity Method the earliest, the computing method of the time domain sensitivity with identified parameters are provided, and by 5 double happiness etc. the identification of electric system relevance parameter and assessment [J]. Proceedings of the CSEE, 2011,31 (22): 73-79. furthers investigate.Sensitivity Method is according to the relation between the identifiability of parameters of electric power system and time domain sensitivity, judges parametric sensitivity track homophase or anti-phase parameter.Because sensitivity track homophase or anti-phase parameter are to be mutually related, therefore their all identifications.Thereby need to, according to the relation between time domain sensitivity and the easy identification of parameter, only carry out identification to emphasis parameter, can effectively improve the accuracy of parameter identification.But consider that time domain sensitivity can not distinguish important parameter and minor parameter completely, frequency domain sensitivity should be served as leading indicator in parameter of measurement identification complexity.To the research of parameters of excitation system identifiability, bring up equality. frequency domain sensitivity and the application in parameters of electric power system identification [J] thereof. Proceedings of the CSEE, 2010,30 (28): 19-24. proposes excitation system to be converted to linearizing modular system, applying frequency domain method identification generator excited system parameter, but this method is not considered the impact of excitation system nonlinear element.Due to generator excited system structure relative complex, need the parameter of identification more, there is each other correlation, simultaneously due to the existence of the non-linear factors such as saturated, amplitude limit, exceeded the range of application of frequency-domain sensitivity, the frequency-domain sensitivity of nonlinear system parameter needs further to be furtherd investigate.

Based on this, the present invention is directed to nonlinear system, realize the judgement of relevance parameter by building time Domain Parameter sensitivity matrix, parameter is divided into good state parameter set and ill parameter set, list all good state parameter sets and corresponding ill parameter set thereof, according in ill parameter set, intend frequency domain sensitivity and choose assignment representative, to assignment, assignment is carried out in representative, and corresponding good state parameter set is carried out to identification.The present invention provides reliable measurement index for the assessment of the easy identification of nonlinear system parameter, can effectively improve the identification precision of generator excited system parameter, improve power system safety and stability computational accuracy, be conducive to safe and stable operation and the control of electric system.

Summary of the invention

The object of the present invention is to provide a kind of generator excited system parameter identifiability decision method.The method realizes the judgement of relevance parameter by building time Domain Parameter sensitivity matrix, parameter to be identified is divided into good state parameter set and ill parameter set, get and in ill parameter set, intend representing as assignment of frequency domain sensitivity sum minimum, to assignment, representative is composed with empirical value, and its corresponding good state parameter set is carried out to parameter identification.Advantage is: effectively improved the Stability and veracity of parameter identification of generator excitation systems result, contributed to improve the operation level of power system safety and stability, had higher engineering practical value.

In order to achieve the above object, the present invention adopts following technical scheme.

A kind of generator excited system parameter identifiability decision method, the method includes the steps of:

(1) determine excitation system types of models;

(2), according to the type of excitation system model, calculate respectively the time domain sensitivity matrix of parameter to be identified and the value of plan frequency domain sensitivity; For definite excitation system types of models, the time domain sensitivity matrix that calculates parameter to be identified is prior art;

(3) obtain the order of time domain sensitivity matrix, the value R of this order is the number R of the maximum cognizable parameters of this excitation system model;

(4) the corresponding parameter of column vector of choosing R linear independence in time domain sensitivity matrix is as good state parameter set, and all the other parameters are as ill parameter set corresponding to this good state parameter set; List all good state parameter sets of time domain sensitivity matrix and corresponding ill parameter set thereof;

(5) calculate the plan frequency domain sensitivity sum of each ill parameter lumped parameter, get and minimum representing as assignment;

(6) to assignment, representative is composed with empirical value (taking the canonical parameter value of recommendation in modeling directive/guide), and its corresponding good state parameter set is carried out to parameter identification;

Wherein, the value of the plan frequency domain sensitivity of above-mentioned steps (2) parameter to be identified is calculated by following method:

(2.1) to the definite excitation system model of step (1), given disturbance and parameter reference value θ _i, wherein i=1,2 ..., n, n is total number of excitation system model parameter, energy emulation obtains the time domain performance graph of model output variable y;

(2.2) autocorrelation function of this time domain output variable y is carried out to Fourier transform, rated output spectral density, get its amplitude from starting point to decaying to times time the interval f of frequency range _efas effective frequency range;

(2.3) get Δ θ _i=0.01 × θ _i, given parameters θ _i+ Δ θ _i, maintaining other parameter and input disturbance constant, the time domain performance graph that energy emulation obtains output variable y, carries out Fourier transform to it, obtains the oscillation amplitude Y of time-domain curve y in different frequency situation _j((θ ₁, θ ₂..., θ _i+ Δ θ _i..., θ _n), f _j), wherein f _jrepresent j the frequency that time-domain curve y comprises;

(2.4) given parameters θ again _i-Δ θ _i, maintaining other parameters and input disturbance constant, repeating step (2.3), obtains the Y in this kind of situation _j((θ ₁, θ ₂..., θ _i-Δ θ _i..., θ _n), f _j);

(2.5) calculate according to the following formula the plan frequency domain sensitivity curve of parameter to be identified

$\begin{array}{c}h({\mathrm{\θ}}_i,f_j)=\\ \left[\right|Y_j(({\mathrm{\θ}}_1,{\mathrm{\θ}}_2,...,{\mathrm{\θ}}_i+\mathrm{\Δ}{\mathrm{\θ}}_i,...,{\mathrm{\θ}}_n),f_j)|-\\ \left|Y_j(({\mathrm{\θ}}_1,{\mathrm{\θ}}_2,...,{\mathrm{\θ}}_i-\mathrm{\Δ}{\mathrm{\θ}}_i,...,{\mathrm{\θ}}_n),f_j)\right|]/2\mathrm{\Δ}{\mathrm{\θ}}_i\end{array}$

(2.6) calculating parameter is intended frequency domain sensitivity curve at effective frequency range f according to the following formula _efthe mean value of interior absolute value

$H_{{\mathrm{\θ}}_i}=\frac{1}{K_{\mathrm{ef}}}\underset{k=1}{\overset{K_{\mathrm{ef}}}{\mathrm{\Σ}}}\left|h({\mathrm{\θ}}_i,f_k)\right|$

Wherein, f _kfor f _jin k be less than f _effrequency sampling point, K _effor f _ktotal sampling number.

Compared with prior art, the present invention has the following advantages and technique effect:

1, mechanism is clear and definite: in same system, intend the little easier identification of parameter of parameter analogy frequency domain sensitivity that frequency domain sensitivity is large, the precision of identification is higher, the parametric sensitivity of nonlinear system can be effectively weighed in the sensitivity of parameter plan frequency domain, for the easy identification of nonlinear system parameter provides evaluation foundation.

2, effect is clear and definite: the present invention proposes to intend frequency domain sensitivity concept first, considers the impact of excitation system nonlinear element, has improved frequency domain sensitivity.Optimum parameter and the ill parametric classification chosen according to time domain sensitivity matrix singular value size in conjunction with the size adjustment of intending frequency domain sensitivity have effectively improved parameter identification precision simultaneously.

3, practical: the parameter identification method that the present invention proposes is all suitable for the generator excited system master pattern of present stage IEEE and China Electric Power Research Institute's announcement, can improve the Stability and veracity of Excitation System Parameter Identification of Synchronous result, contribute to improve power system safety and stability operation level, there is higher engineering practical value.

Brief description of the drawings

Fig. 1 is the process flow diagram of a kind of generator excited system parameter of the present invention identifiability decision method.

Fig. 2 is the process flow diagram that calculates the value of the plan frequency domain sensitivity of parameter to be identified in step of the present invention (2).

Fig. 3 is IEEEST2A type excitation system illustraton of model.

Embodiment

Embodiment mono-

The invention provides a kind of generator excited system parameter identifiability decision method, the method realizes the judgement of relevance parameter by building time Domain Parameter sensitivity matrix, parameter to be identified is divided into good state parameter set and ill parameter set, get in ill parameter set, intend frequency domain sensitivity sum minimum as assignment representative, and the good state parameter set of its correspondence is carried out to parameter identification.

As shown in Figure 1, the inventive method comprises the following steps:

(1) determine excitation system types of models;

(2), according to the type of excitation system model, calculate respectively the time domain sensitivity matrix of parameter to be identified and the value of plan frequency domain sensitivity;

(3) obtain the order of time domain sensitivity matrix, the value R of this order is the number R of the maximum cognizable parameters of this excitation system model;

(4) the corresponding parameter of column vector of choosing R linear independence in time domain sensitivity matrix is as good state parameter set, and all the other parameters are as ill parameter set corresponding to this good state parameter set; List all good state parameter sets of time domain sensitivity matrix and corresponding ill parameter set thereof;

(5) calculate the plan frequency domain sensitivity sum of each ill parameter lumped parameter, get and minimum representing as assignment;

(6) to assignment, representative is composed with empirical value (taking the canonical parameter value of recommendation in modeling directive/guide), and its corresponding good state parameter set is carried out to parameter identification;

Wherein, the value of the plan frequency domain sensitivity of above-mentioned steps (2) parameter to be identified is calculated by following method:

(2.1) to the definite excitation system model of step (1), given disturbance and parameter reference value θ _i, wherein i=1,2 ..., n, n is total number of excitation system model parameter, energy emulation obtains the time domain performance graph of model output variable y;

(2.2) autocorrelation function of this time domain output variable y is carried out to Fourier transform, rated output spectral density, get its amplitude from starting point to decaying to times time the interval f of frequency range _efas effective frequency range;

(2.3) get Δ θ _i=0.01 × θ _i, given parameters θ _i+ Δ θ _i, maintaining other parameter and input disturbance constant, the time domain performance graph that energy emulation obtains output variable y, carries out Fourier transform to it, obtains the oscillation amplitude Y of time-domain curve y in different frequency situation _j((θ ₁, θ ₂..., θ _i+ Δ θ _i..., θ _n), f _j), wherein f _jrepresent j the frequency that time-domain curve y comprises;

(2.4) given parameters θ again _i-Δ θ _i, maintaining other parameters and input disturbance constant, repeating step (2.3), obtains the Y in this kind of situation _j((θ ₁, θ ₂..., θ _i-Δ θ _i..., θ _n), f _j);

(2.5) calculate according to the following formula the plan frequency domain sensitivity curve of parameter to be identified

$\begin{array}{c}h({\mathrm{\θ}}_i,f_j)=\\ \left[\right|Y_j(({\mathrm{\θ}}_1,{\mathrm{\θ}}_2,...,{\mathrm{\θ}}_i+\mathrm{\Δ}{\mathrm{\θ}}_i,...,{\mathrm{\θ}}_n),f_j)|-\\ \left|Y_j(({\mathrm{\θ}}_1,{\mathrm{\θ}}_2,...,{\mathrm{\θ}}_i-\mathrm{\Δ}{\mathrm{\θ}}_i,...,{\mathrm{\θ}}_n),f_j)\right|]/2\mathrm{\Δ}{\mathrm{\θ}}_i\end{array}$

(2.6) calculating parameter is intended frequency domain sensitivity curve at effective frequency range f according to the following formula _efthe mean value of interior absolute value

$H_{{\mathrm{\θ}}_i}=\frac{1}{K_{\mathrm{ef}}}\underset{k=1}{\overset{K_{\mathrm{ef}}}{\mathrm{\Σ}}}\left|h({\mathrm{\θ}}_i,f_k)\right|$

Wherein, f _kfor f _jin k be less than f _effrequency sampling point, K _effor f _ktotal sampling number.

Embodiment bis-

(1) taking IEEEST2A type excitation system as example, this type excitation system belongs to from compound excitation static excitation system, synthesizes to form power power-supply by the phasor of generator terminal voltage and armature supply, and as shown in Figure 3, model parameter is as shown in table 1 for its illustraton of model.

Consider generator no-load running, armature supply I _tbe 0 o'clock, can not be to parameter K _icarry out identification, therefore, in the process of parameter identification, take from and encourage COEFFICIENT K _e=1, this value does not participate in the identification of systematic parameter.

Table 1IEEEST2A type excitation system model parameter

(2) form the time Domain Parameter sensitivity matrix of this model.

(3) obtain rank of matrix R=4, the number that shows the maximum cognizable parameters of this model is 4.

(4) obtain whole good state parameter sets and corresponding ill parameter set thereof, as shown in table 2.

The table 2IEEEST2A good state parameter set of type excitation system and corresponding ill parameter set thereof

(5) calculate the result of plan frequency domain sensitivity of each parameter as shown in table 3; In morbid state parameter set, intend frequency domain sensitivity sum as shown in table 4.

Table 3IEEEST2A type parameters of excitation system is intended frequency domain sensitivity

In table 4IEEEST2A type excitation system morbid state parameter set, intend frequency domain sensitivity sum

(6), according to table 4, choosing good state parameter set is K _f, T _f, K _p, T _e, assignment is represented as T _a, K _a, K _c.

(7) following three kinds of assignment situations are set respectively, IEEEST2A type parameters of excitation system is carried out to identification.

Situation 1: true value is composed in the representative of relevance parameter

Situation 2: the representative of relevance parameter departs from actual value 5%

Situation 3: the representative of relevance parameter departs from actual value 10%

Choosing at random good state parameter set is K _f, T _f, K _a, T _e, ill parameter set is T _a, K _c, K _p, in above-mentioned three kinds of situations, parameter being carried out to 20 identifications, its average result is as shown in table 5.

Table 5 is chosen assignment representation parameter identification result at random

Meanwhile, carry out assignment according to after intending frequency domain sensitivity and adjusting as ill assignment representative, carry out 20 average results after identification as shown in table 6.

Table 6 is according to the parameter identification result of intending after frequency domain sensitivity adjustment

Identification result from table 5 and table 6: parameter K _pidentification precision more much higher than the identification precision of other all parameters, the identification irrelevance of its parameter, in 0.4% left and right, is far smaller than the identification deviation of other parameter.Show the large parameter of parameter plan frequency domain sensitivity, its identification precision is higher, has verified the correctness of parameter that the present invention puies forward plan frequency domain sensitivity computing method.The overall identification precision of comparative study systematic parameter can be found: incorporating parametric is intended frequency domain sensitivity, by good state parameter by K _f, T _f, K _a, T _echange to K _f, T _f, K _p, T _eafter, although parameter K _f, T _fidentification precision decline to some extent, but, due to parameter K _pidentification result compare parameter K _aidentification precision much higher.Compose in true value situation in the representative of relevance parameter, dropped to 5.6570% by the total departure of identified parameters by 5.9150%; Depart from actual value 5% situation in the representative of relevance parameter, total departure drops to 6.9635% by 7.1259%; Depart from true 10% situation in the representative of relevance parameter, total departure drops to 7.6336% by 9.6384%.Although depart from the ratio increase of actual value along with relevance parameter represents assignment, the identification precision of all parameters to be identified presents downtrending.But in identical relevant parameter assignment precision situation, after the representative of frequency domain sensitivity adjustment call by value parameter, all presented downtrending by the total departure of identified parameters in conjunction with intending, the overall identification precision of systematic parameter is improved.

Embodiment tri-

Taking IEEEDC1A type excitation system as example, carry out the judgement of parameter identifiability according to the present invention, good state parameter set is K _a, K _f, T _a, T _f, ill parameter set is T _b, T _c, T _e.Choosing at random good state parameter set is K _a, K _f, T _e, T _f, ill parameter set is T _b, T _c, T _a.Following three kinds of assignment situations are set respectively, IEEEDC1A type parameters of excitation system is carried out to identification.

Situation 1: true value is composed in the representative of relevance parameter

Situation 2: the representative of relevance parameter departs from actual value 5%

Situation 3: the representative of relevance parameter departs from actual value 10%

For three kinds of assignment situations of two kinds of parameter sets, carry out respectively 20 identifications, contrast identification result is known, and the parameter that sensitivity is chosen according to plan frequency domain is carried out to parameter identification, and the identification precision of most parameters is improved.Contrast the identification result of parameter in three kinds of situations, can find out in the time that relevant parameter assignment represents that value and actual value exist deviation, the identification precision entirety of parameter is on a declining curve.But the Identification Errors of most parameters is still in tolerance interval.

Method step described herein and data are the specific embodiment of patent of the present invention, be the overall explaination that patent of the present invention spirit is done and illustrate, be not limited to concrete excitation system model, patent those skilled in the art of the present invention also can recognize the multiple possibility of modification or optional embodiment, inspired by the spirit and principles of the present invention, do various amendments, supplement, improve or similar substituting, be understandable that these amendments, supplement, improve or substitute and will be considered as included in the present invention, and can't depart from spirit of the present invention or surmount the defined scope of appended claims.

Claims (1)

1. a generator excited system parameter identifiability decision method, is characterized in that, the method comprises the following step:

(1) determine excitation system types of models;

(2), according to the type of excitation system model, calculate respectively the time domain sensitivity matrix of parameter to be identified and the value of plan frequency domain sensitivity;

(3) obtain the order of time domain sensitivity matrix, the value R of this order is the number R of the maximum cognizable parameters of this excitation system model;

(4) the corresponding parameter of column vector of choosing R linear independence in time domain sensitivity matrix is as good state parameter set, and all the other parameters are as ill parameter set corresponding to this good state parameter set; List all good state parameter sets of time domain sensitivity matrix and corresponding ill parameter set thereof;

(5) calculate the plan frequency domain sensitivity sum of each ill parameter lumped parameter, get and minimum representing as assignment;

(6) to assignment, representative is composed with empirical value, and its corresponding good state parameter set is carried out to parameter identification;

Wherein, the value of the plan frequency domain sensitivity of above-mentioned steps (2) parameter to be identified is calculated by following method:

(2.1) to the definite excitation system model of step (1), given disturbance and parameter reference value θ _i, wherein i=1,2 ..., n, n is total number of excitation system model parameter, energy emulation obtains the time domain performance graph of model output variable y;

(2.2) autocorrelation function of this time domain output variable y is carried out to Fourier transform, rated output spectral density, get its amplitude from starting point to decaying to times time the interval f of frequency range _efas effective frequency range;

(2.3) get Δ θ _i=0.01 × θ _i, given parameters θ _i+ Δ θ _i, maintaining other parameter and input disturbance constant, the time domain performance graph that energy emulation obtains output variable y, carries out Fourier transform to it, obtains the oscillation amplitude Y of time-domain curve y in different frequency situation _j((θ ₁, θ ₂..., θ _i+ Δ θ _i..., θ _n), f _j), wherein f _jrepresent j the frequency that time-domain curve y comprises;

(2.4) given parameters θ again _i-Δ θ _i, maintaining other parameters and input disturbance constant, repeating step (2.3), obtains the Y in this kind of situation _j((θ ₁, θ ₂..., θ _i-Δ θ _i..., θ _n), f _j);

(2.5) calculate according to the following formula the plan frequency domain sensitivity curve of parameter to be identified

$\begin{array}{c}h({\mathrm{\θ}}_i,f_j)=\\ \left[\right|Y_j(({\mathrm{\θ}}_1,{\mathrm{\θ}}_2,...,{\mathrm{\θ}}_i+\mathrm{\Δ}{\mathrm{\θ}}_i,...,{\mathrm{\θ}}_n),f_j)|-\\ \left|Y_j(({\mathrm{\θ}}_1,{\mathrm{\θ}}_2,...,{\mathrm{\θ}}_i-\mathrm{\Δ}{\mathrm{\θ}}_i,...,{\mathrm{\θ}}_n),f_j)\right|]/2\mathrm{\Δ}{\mathrm{\θ}}_i\end{array}$

(2.6) calculating parameter is intended frequency domain sensitivity curve at effective frequency range f according to the following formula _efthe mean value of interior absolute value

$H_{{\mathrm{\θ}}_i}=\frac{1}{K_{\mathrm{ef}}}\underset{k=1}{\overset{K_{\mathrm{ef}}}{\mathrm{\Σ}}}\left|h({\mathrm{\θ}}_i,f_k)\right|$

Wherein, f _kfor f _jin k be less than f _effrequency sampling point, K _effor f _ktotal sampling number.