CN104077493B - Method for constructing state evaluation index system of electric relaying protection system - Google Patents

Abstract

Disclosed is a method for constructing a state evaluation index system of an electric relaying protection system. The method is characterized by comprising the steps that a hierarchical level model is constructed based on a level analysis method; a candidate evaluation factor set of a relaying protection device and a secondary loop is determined through a group decision-making method based on entropy, then the candidate evaluation factor set is further extracted by applying a kernel principal component analysis method, and an optimal evaluation factor set is obtained; different weight coefficients are given to different factors to reflect the importance degree and influence of the factors, subjective weighting based on a fuzzy evaluation weighting method is carried out on the optimal factor set, then objective weighting based on the kernel principal component analysis method is carried out on the optimal factor set, and the comprehensive weight of the optimal factor set is obtained by combining the two kinds of weighting; the model is hierarchically divided according to the built relaying protection system, the extracted optimal factor set is combined, a weight system is obtained through a subjective and objective comprehensive weighting method, the complete state evaluation index system of the electric relaying protection system is built, and objective, accurate and comprehensive state evaluation is carried out on the electric relaying protection system.

Description

A kind of construction method of system of relay protection State Assessment Index System

Technical field

The present invention relates to POWER SYSTEM STATE evaluation index technical field, it is a kind of system of relay protection state estimation The construction method of index system.

Background technology

System of relay protection as a huge complication system, have comprise that equipment is numerous, between each equipment with And the feature that between equipment and system, relatedness is strong, run time behaviour is complicated, affect its state factor these factors numerous it Between relation intricate, relay protection generating state migration when, often be accompanied by multiple quantity of states change.

The running status of relay protection is carried out with accurate state estimation comprehensively, needs to extract most representative and energy Enough objective quantity of states definitely reflecting working condition, set up relay protection state evaluation index system.

It has been related to the correlational study achievement of system of relay protection State Assessment Index System, one of which at present The method of main flow is to adopt PCA, and this method is passed through to study index system immanent structure relation, thus by multiple indexs It is converted into a few separate aggregative indicator, its advantage is the determination of its weights is based on data, is not subject to subjective impact, There is preferable objectivity, and separate between the aggregative indicator drawing, reduce the intersection of information, this has to analysis and evaluation Profit.

But the shortcoming of this method is to process nonlinear data effect on driving birds is not good, and system of relay protection this is huge It is greatly nonlinear data that the data of complication system has, and does not have system, comprehensively considers each state estimation factor Effect it is necessary to improve so as to can be to nonlinear data substantial amounts of in power protection system to PCA Processed, objective and accurately provide conclusion, but so far, there is not yet relevant document report and practical application.

Content of the invention

The purpose of the present invention is the shortcoming overcoming prior art, provides a kind of objective, accurate and comprehensive electric power relay to protect The construction method of protecting system State Assessment Index System, the method can be to non-linear number substantial amounts of in system of relay protection According to preferably being processed, thus obtaining more objective, accurate and comprehensive conclusion.

The purpose of the present invention to be realized by technical scheme below: a kind of system of relay protection state estimation index The construction method of system, is characterized in that, it comprises the following steps:

(1) build Recurison order hierarchy model using based on analytic hierarchy process (AHP), relay protection system divided according to equipment, Equipment divides according to protective relaying device and secondary circuit, relay protection system state estimation is decomposed into simple, easy behaviour The multi-level fuzzy judgment made is analyzed and assesses；

(2) the candidate's appraisal parameters determining protective relaying device and secondary circuit based on the Group Decision Method of entropy are first used, Then candidate's appraisal parameters are carried out more fully with core principle component analysis method again and objectively extract further, obtain Good appraisal parameters；

(3) its significance level and power of influence are embodied by giving different weight coefficients to different factors, to optimal factor Collection carries out the subjective weights of enabling legislation based on fuzzy evaluation, more optimal set of factors is carried out objective based on core principle component method Assign power, assign the comprehensive weight that power obtains optimal set of factors for comprehensive two kinds；

(4) according to the relay protection system distinguishing hierarchy model set up, in conjunction with the optimal set of factors extracted, using subjective and objective The weight system that combination weights method obtains, sets up complete relay protection state evaluation index system,

Wherein in core principle component analysis method, select suitable kernel function and its parameter, according to Polynomial kernel function:

k₁=(＜ x_i,x_j＞+c)^d

With rbf kernel function:

k₂=exp (- | | x_i-x_j||²/2σ²), (i, j=1,2 ... are q)

Create a new kernel function by this two kernel functions, using closure one kernel function of creation of kernel function:

$k=\sqrt{\frac{{(k_1+k_2)}^2}{2}}=\sqrt{\frac{{({(<x_i,x_j>+c)}^d+\exp (-||x_i-x_j{\left|\right|}^2/2{\mathrm{\&sigma;}}^2))}^2}{2}}$

In that step of comprehensive weight, choose formula

$\mathrm{\&alpha;}={\mathrm{\&lambda;}}^{\&prime;}+\mathrm{\&theta;}(r^{\&prime;}-{\mathrm{\&lambda;}}^{\&prime;})={\mathrm{\&lambda;}}^{\&prime;}+(1-\frac{r^{\&prime;}\&centerdot;{\mathrm{\&lambda;}}^{\&prime;}}{\left|\right|r^{\&prime;}\left|\right|\left|\right|{\mathrm{\&lambda;}}^{\&prime;}\left|\right|})/2(r^{\&prime;}-{\mathrm{\&lambda;}}^{\&prime;})$

As the method for comprehensive weight,

α is comprehensive weight, and λ ' is the weights that core principle component analysis obtain, and r' is the weights based on fuzzy evaluation.

System was carried out by the construction method of a kind of system of relay protection State Assessment Index System of the present invention before this Hierachical decomposition, is broken down into multi-level fuzzy judgment.Then again candidate's appraisal parameters are obtained by the group's plan method based on entropy, then By core principle component method, candidate's appraisal parameters are extracted and obtain optimal set of factors, it is to avoid sample data volume is big and non-linear number It is not optimal set of factors according to leading to find out.Assigned by the synthesis based on fuzzy evaluation enabling legislation with based on core principle component enabling legislation Power method, obtains more objective, science assessment factor comprehensive weight.Build system of relay protection state estimation index body System can carry out objective, accurate and comprehensive state estimation to system of relay protection.

Brief description

Fig. 1 is a kind of construction method block diagram of system of relay protection State Assessment Index System；

The membership function figure of the temperature in Fig. 2 running environment；

The membership function figure of Fig. 3 humidity；

The membership function figure of the failure condition of Fig. 4 software and hardware；

The anti-implementation of conditions membership function figure of arranging of Fig. 5；

The membership function figure of Fig. 6 protection device action accuracy；

Fig. 7 works as the membership function figure that loop insulation is more than 10m ω insulation status；

There is not the membership function figure of the insulation status of DC earthing in Fig. 8；

Fig. 9 works as the membership function figure that analog quantity error is less than 5% data sampling；

Figure 10 communication condition membership function figure；

The membership function figure of Figure 11 channel communications situation；

Figure 12 operating ambient temperature scoring figure.

Specific embodiment

Concrete reality with reference to a kind of system of relay protection State Assessment Index System method to the present invention for the Fig. 1 The mode of applying is described in detail.

With reference to Fig. 1, a kind of construction method of system of relay protection State Assessment Index System of the present invention, including with Lower step:

(1) relay protection system is decomposed into multiple relay protection devices: hierachical decomposition is carried out to relay protection system, builds The Recurison order hierarchy model of vertical relay protection system；

(2) will index for selection factor from protective relaying device and secondary circuit as requested, to the index factor chosen Carry out process and obtain optimal set of factors and its weight, finally build with regard to evaluation index system: commented according to relay protection state first Estimate and the regulation of inspection and repair, in conjunction with expert advice, from the quality of equipment itself, preventive trial, operating condition and historical data collection Data, sets up assessment candidate agents collection, and the group's plan method based on entropy that reuses is extracted to assessment candidate agents collection, rejects and does not weigh The factor wanted, obtains candidate's appraisal parameters, then utilizes core principle component analysis method to extract the master of above-mentioned candidate's appraisal parameters Compositional factors, set up its best-evaluated set of factors；

(3) to best-evaluated set of factors, first obtain subjective weight using based on the enabling legislation of fuzzy evaluation, then with the main one-tenth of core Point analytic process carries out Objective Weight and obtains a new weight to best-evaluated set of factors, finally comprehensive two weights；

(4) the relay protection system distinguishing hierarchy model that foundation is set up, best-evaluated set of factors and tax power method, obtain every The weight of one factor, obtains corresponding factor scores further according to Fig. 2-Figure 12, further according to taking minimum scoring principle, obtains relay The scoring of protection device, can obtain the scoring of secondary circuit, the scoring of equipment in the same manner, finally give commenting of relay protection system Divide thus building relay protection system State Assessment Index System；

The step that described group's plan method based on entropy extracts best-evaluated factor is:

Assume that assessment candidate agents collection has m factor, have n expert each factor to be carried out assign weight, then, obtain Individual weight matrix f,

$f=\left[\begin{array}{cccc}{r}_{11}& r_{12}& ...& r_{1n}\\ r_{21}& r_{22}& ...& r_{2n}\\ .& .& & .\\ .& .& ...& .\\ .& .& & .\\ r_{m1}& r_{m2}& ...& r_{\mathrm{mn}}\end{array}\right]$

r_ijRepresent j-th expert tax weights to i-th index, wherein i=1,2 ... m, j=1,2 ... n,

First the weights of every string are carried out asking entropy computing, be the entropy asking each expert to m Index Weights value, and $\underset{j=1}{\overset{m}{\mathrm{\&sigma;}}}{r}_{\mathrm{ij}}=1.$

The formula of row entropy is:

$h_j=\underset{i=1}{\overset{m}{\mathrm{\&sigma;}}}-r_{\mathrm{ij}}\ln r_{\mathrm{ij}}---\left(1\right)$

h_jRepresent is j-th expert entropy to m Index Weights value.

When row entropy bigger expression expert j is almost more or less the same for the tax weights of m index, represent j expert for tax Weights consider deeply, authoritative low, give up the tax weights of this expert,

By above-mentioned row entropy according to arranging from small to large, set a threshold value 80%, take its expert corresponding to front q row entropy Assigned weights, that is, allow $\frac{\underset{j=1}{\overset{q}{\mathrm{\&sigma;}}}{h}_j}{\underset{i=1}{\overset{m}{\mathrm{\&sigma;}}}{h}_j}\&greaterequal;80\%;$

Reconfigure and take q expert that m index factor is assigned with the matrix f' of weights:

$f^{\&prime;}=\left[\begin{array}{cccc}{r}_{11}& r_{12}& ...& r_{1q}\\ r_{21}& r_{22}& ...& r_{2q}\\ .& .& & .\\ .& .& ...& .\\ .& .& & .\\ r_{m1}& r_{m2}& ...& r_{\mathrm{mq}}\end{array}\right]$

Seek the entropy of the tax weights to m index for the q expert again, be row entropy, first each to f' before asking for row entropy The weights of row are normalized, as:

$r_{\mathrm{ij}}^{\&prime;}=\frac{r_{\mathrm{ij}}}{\underset{j=1}{\overset{q}{\mathrm{\&sigma;}}}{r}_{\mathrm{ij}}}---\left(2\right)$

r'_ijIt is r_ijValue after normalization.

Wherein i=1,2 ... m j=1,2 ... q.

Such that it is able to obtain row entropy h_iFor:

$h_i=\underset{j=1}{\overset{q}{\mathrm{\&sigma;}}}-r_{\mathrm{ij}}^{\&prime;}\ln r_{\mathrm{ij}}^{\&prime;}---\left(3\right)$

Work as h_iBigger explanation q expert have no objection to the tax weights of i-th index, and recognition rate is high, and accuracy is higher, table Show that i-th index is not result in that larger error in assessment result, otherwise then make assessment result error big, so h_iRepresent simultaneously The significance level of index, then, to h_iIt is normalized $h_i^{\&prime;}=\frac{h_i}{\underset{i=1}{\overset{m}{\mathrm{\&sigma;}}}{h}_i},$

(r_i1,r_i2,…r_iq)_iIt is the row of f' matrix, according to (r_i1,r_i2,…r_iq)_iOne judgment matrix x of construction_i,

$x_i=\left[\begin{array}{cccc}1& w_1/w_2& ...& w_1/w_q\\ w_2/w_1& 1& ...& w_2/w_q\\ ...& ...& & \\ w_q/w_1& & ...& 1\end{array}\right],$ Quote digital 1-9 and its inverse and regard matrix x as scale_iUnit Element, x_iThe corresponding judgment matrix being obtained by matrix f' i-th row, then asks for x_iEigenvalue of maximum λⁱ _max,

Judgment matrix scale defines, table 1:

Wherein j, k=1,2 ... q, i=1,2 ... m, w_aWith w_bIt is exactly scale, the as element of matrix xi,

Then in calculating coincident indicator:

$\mathrm{ci}=\frac{{{\mathrm{\&lambda;}}^i}_{\max }-q}{q-1},---\left(4\right)$

Wherein λⁱ _maxFor judgment matrix x_iEigenvalue of maximum.

Ask for coincident indicator ratio:

Cr=ci/ri, (5)

Wherein ri is Aver-age Random Consistency Index, and as cr, < when 0.10, the concordance of matrix can accept, otherwise to sentencing Disconnected matrix x_iIt is modified,

So, the concrete weights of each index factor are finally obtained, as:

r_i'=λⁱ _maxh_i'(r_i1,r_i2,…r_iq)_i(6)

Wherein r_i' represent be i-th index weights, (r_i1,r_i2,…r_iq)_iRepresent that q expert assigns to i-th index Weights block median, as: r_i=median (r_i1+r_i2+…+r_iq), r_iRepresent the weights to i-th index of q expert Weights after stage median is processed,

Then set a threshold value, take 80% as its threshold value, by r_i' according to descending, take its front q value to make r_i' so that80%, finally extract this q r_i' corresponding factor, obtain candidate's appraisal parameters.

Described core principle component analysis method, i.e. the having main steps that of kpca:

As process candidate appraisal parameters x_i, (i=1,2 ..., when q) running into nonlinear data, map that to feature In space, feature space is changed into linear data, introduces nonlinear mapping functionRealize input space rm to feature space The mapping of f, that is,

Select suitable kernel function and its parameter, selection Polynomial kernel function:

k₁=(＜ x_i,x_j＞+c)^d, (7)

Rbf kernel function:

k₂=exp (- | | x_i-x_j||²/2σ²), (i, j=1,2 ... are q). (8)

Choose two kernel functions and create a new kernel function, using closure one kernel function of creation of kernel function:

$k=\sqrt{\frac{{(k_1+k_2)}^2}{2}}.---\left(9\right)$

If data is zero-mean, asThen the data covariance matrix on feature space f is:

Now by solution following equations:

c^fV=λ v (11)

Obtain eigenvalue λ (λ >=0) and corresponding characteristic vector v (v ∈ f/ { 0 })

According to formula (11) build simultaneously peer-to-peer both sides do andInner product equation:

All can be by view of all characteristic vectorsLinear expression, and there is a_i(i=1,2 ... Q) so that:

Formula (13) and (10) are brought into formula (12) obtain:

Define a q rank matrix k', the wherein element of k'So, substitute into (5) formula, obtain:

Q λ ka=k²a (15)

Wherein a=(a₁,a₂…a_q)^t,

Abbreviation (15) formula obtains:

Q λ a=k a (16)

The eigenvalue of the non-zero of (16) formula of solution,

Carry out descending to the value of non-zero λ to obtain, λ₁,λ₂,…λ_p, wherein 1≤p≤q,

Due to formula:

Therefore pass through standardization a^k, carry out the characteristic vector in standardized feature space, i.e. (v_i·v_i)=1,

(17) formula is derived by:

a^k·a^k=1/ λ_k(18)

Wherein λ_k(k=1,2 ... p)

Then above-mentioned λ₁,λ₂,…λ_pCorresponding index factor is exactly to finally obtain best-evaluated set of factors, then by this p Factor reduces from feature space returns to adopt formula:

WhereinIt is the value that Non-linear Principal Component is mapped to feature space,

The average if data is not zero, reply k matrix center turns toWithGo replace k:

$\stackrel{\&overbar;}{k}=(i-\frac{1}{q}{1}_q{1}_q^t)k(i-\frac{1}{q}{1}_q{1}_q^t)---\left(20\right)$

Reduced by formula (20), thus finally giving best-evaluated set of factors x={ x₁,x₂,…x_p}.

Described based in the enabling legislation step of fuzzy evaluation:

Formula (20) is obtained with smallest evaluation set of factors is x={ x₁,x₂,…x_p, according to the weight to each factor weight for the expert The property wanted, gives each factor one value using method of expertise, such that it is able to construct the fuzzy subset a=[a of an x_i] i= 1,2 ... p, and $\underset{i=1}{\overset{p}{\mathrm{\&sigma;}}}{a}_i=1.$

Then expert weighs, using basic tax, the weights set b that method obtains optimal set of factors, and this weights set b is entered Row is divided based on k mean cluster, marks off 5 intervals, { u₁, u₂, u₃, u₄, u₅}.For each factor, arrange one rationally Membership function f (x), calculate its each interval degree of membership, take maximum the interval corresponding to degree of membership, above-mentioned base The step dividing in k mean cluster:

(1) to the weights set b being given, 5 initial cluster center points are chosen, for { u₁, u₂, u₃, u₄, u₅}.

(2) define a distance metric d (u_i)=| u_i-r_j|, j=1,2 ... p, i=1,2 ... 5.s_jIt is jth in data b Weights corresponding to individual factor, it is to weigh s_jTo each cluster centre apart from length.

(3) calculate s successively_jTo each u_iDistance be d (u_i) value, compare each d (u_i) size, d (u_i) minimum Corresponding s_j, its corresponding factor j is put into corresponding u_iCluster interval, eventually form initial five cluster area Between.

(4) interval for five above-mentioned clusters, again ask for their cluster centre, using formulaS is should The interval weights of i number, s_iIt is the sum of all weights in place the i-th interval.

According to said method thus obtaining 5 new cluster centres.

Then repeat above-mentioned (2), (3), the step of (4), until cluster centre no longer changes.

Five are finally correctly divided with u_iFor the interval of cluster centre, thus meeting is to be carried based on the method for fuzzy evaluation Supplied five correctly objectively interval.

According to Fig. 1, can learn that appraisal parameters finally have 8 classes, running environment, the failure condition of software and hardware, counter arrange practicable Situation, performance factor, insulation status, data sampling, communication condition, lane testing situation.

Then each factor is analyzed.

For the temperature in running environment, the description of its membership function figure is as shown in Figure 2.

Its abscissa is corresponding to be ambient temperature, and vertical coordinate is corresponding to be its degree of membership, its membership function f (x₁) such as Under:

By membership function above, can obtain, when given temperature set of factors, calculating according to this membership function It is subordinate to angle value.By f (x₁) value interval interval to should determine which it is located with the cluster that divided, then take its place again Interval cluster centre value replaces f (x₁).

Then f (the x after institute's replacement values₁) be multiplied with its corresponding fuzzy subset, all finally obtain result, returned After one changes, the result being drawn is exactly its weights.

Different using simply the given membership function of identical method for other factors in the same manner, we only list here The membership function figure of each factor and function.

The membership function figure of corresponding humidity is as shown in Figure 3.

Its membership function f (x₂) as follows:

The membership function figure of the failure condition of software and hardware is as shown in Figure 4.

Its membership function is f (x₃) as follows:

Counter arrange implementation of conditions membership function figure as shown in Figure 5.

The anti-implementation of conditions membership function f (x that arranges₄) as follows:

The membership function figure of protection device action accuracy is as shown in Figure 6.

Its membership function is f (x₅) as follows:

$f\left(x_5\right)=\left\{\begin{array}{cc}0& 0\&le;x_5\&le;98.5\\ 10x_5-85& x_5\&greaterequal;98.5\end{array}\right.$

When loop insulation is as shown in Figure 7 more than the membership function figure of 10m ω insulation status.

Its membership function f (x₆) as follows:

$f\left(x_6\right)=\left\{\begin{array}{cc}0& 0=<x_6<10\\ 0.05x_6& 10=<x_6<=20\\ 1& x_6>=20\end{array}\right.$

Do not occur the membership function figure of the insulation status of DC earthing as shown in Figure 8.

Its membership function f (x₇) as follows:

When analog quantity error is as shown in Figure 9 less than the membership function figure of 5% data sampling.

Electric current, the sampled value difference of voltage channel each group measured value For cti, pti.

When analog quantity error is less than 5% its membership function f (x₈):

$f\left(x_8\right)=\left\{\begin{array}{cc}1& 0=<x_8<3\\ -0.5x_8+2.5& 3>x_8<=5\end{array}\right.$

Switching value sampling is correct, no opens into abnormal data sampling membership function:

Communication condition membership function figure is as shown in Figure 10.

Communication condition membership function f (x₁₀) as follows:

The membership function figure of channel communications situation is as shown in figure 11.

Membership function f (the x of channel communications situation₁₁) as follows:

Determine that each factor is located by above-mentioned membership function interval, take the cluster centre value in interval to substitute f (x_i) Value, then f (x_i) being multiplied with its corresponding fuzzy subset is worth, that is, using formula:

r_i=a_i·f(x_i) (21)

Wherein r_iIt is the weight of factor i, then it is normalized and obtains:

$r_i^{\&prime;}=\frac{r_i}{\underset{i=1}{\overset{p}{\mathrm{\&sigma;}}}{r}_i}---\left(22\right)$

Wherein i=1,2 ..., p, r_i' it is r_iThe last weights of i factor after normalization, weight vectors r'=(r₁', r'₂,…,r'_p).

Re-use the operation that core principle component carries out Objective Weight to best-evaluated set of factors again, step is as follows:

It is entrained former that the pivot being extracted by core principle component analysis method illustrates it to the contribution of feature set The size of beginning Feature change information, contribution rate is bigger, then characteristic information is explained stronger.

The weight so asking for optimal set of factors is equal to and asks for eigenvalue collection.

According to above-mentioned core principle component analysis method it is known that passing through formula:

Q λ a=k a and a^k·a^k=1/ λ_k

Ask for λ_kValue, wherein k=1,2 ..., p.

Again to λ_kIt is normalized, that is, with formula:

${\mathrm{\&lambda;}}_k^{\&prime;}=\frac{{\mathrm{\&lambda;}}_k}{\underset{k=1}{\overset{p}{\mathrm{\&sigma;}}}{\mathrm{\&lambda;}}_k}---\left(23\right)$

Obtained λ '=(λ '₁,λ'₂,…λ'_p) value just correspond to the (x inside best-evaluated set of factors₁,x₂,…, x_p) weights.

Then the result of two kinds of weights of summary, using aggregative weighted processing method, comprehensive weight vector α is:

$\mathrm{\&alpha;}={\mathrm{\&lambda;}}^{\&prime;}+\mathrm{\&theta;}(r^{\&prime;}-{\mathrm{\&lambda;}}^{\&prime;})={\mathrm{\&lambda;}}^{\&prime;}+(1-\frac{r^{\&prime;}\&centerdot;{\mathrm{\&lambda;}}^{\&prime;}}{\left|\right|r^{\&prime;}\left|\right|\left|\right|{\mathrm{\&lambda;}}^{\&prime;}\left|\right|})/2(r^{\&prime;}-{\mathrm{\&lambda;}}^{\&prime;})---\left(24\right)$

θ is coefficient of colligation, and ordinary circumstance value is 0.5, andI=1,2 ..., p, α_iIt is the comprehensive of i-th factor Close weights.

To relay protection device state estimation optimal set of factors x={ x₁,x₂,…x_pCarry out similar division, will be related or same The each factor of class carries out dividing to be sorted out, and in abstract same division, each factor is high-level factor class.State estimation set of factors comprises Some assessment factor classes, each factor class is by some state estimation factors composition.The factor class set m={ ac setting up₁, ac₂..., ac_t, wherein factor class ac_kComprise | ac_k| individual factor, it is expressed as with factorWherein k=1, 2,…,t.

Calculating factor class ac_kWeighted value in the factor class class of each factor interiorAccording to each in state estimation factor class class Factor weight ratio calculates weight in the class of each factor, assigns power computing formula as follows in the factor class class of factor:

$w_{k_j}=w_{k_j}/\underset{i=1}{\overset{\left|{\mathrm{ac}}_k\right|}{\mathrm{\&sigma;}}}{w}_{k_i}---\left(25\right)$

And $\underset{j=1}{\overset{\left|{\mathrm{ac}}_k\right|}{\mathrm{\&sigma;}}}{w}_{k_j}=1,k=\mathrm{1,2},...,t.$

According to subjective and objective combination weights result of calculation, carry out the tax of relay protection state evaluation index system factor class hierarchy Power calculates, and it is as follows that assessment factor class weight assigns power computing formula:

$w_k=\underset{i=1}{\overset{\left|{\mathrm{ac}}_k\right|}{\mathrm{\&sigma;}}}{w}_{k_i}/\underset{k=1}{\overset{t}{\mathrm{\&sigma;}}}\underset{i=1}{\overset{\left|{\mathrm{ac}}_k\right|}{\mathrm{\&sigma;}}}{w}_{k_i}---\left(26\right)$

Wherein w_kIt is k-th factor class weights, and

The step finally building relay protection system evaluation index system:

Relay protection system is carried out setting up hierarchical model using based on analytic hierarchy process (AHP), relay protection system state is commented Estimate and be decomposed into simply maneuverable multi-level fuzzy judgment and be analyzed and assess, form evaluation system.

Relay protection system is divided according to equipment, relay protection system can be divided into several relay protection devices.

Then proceed to carry out step analysis to each relay protection device, relay protection device is divided into several relays Protection device and several secondary circuit two classes.

Finally again protective relaying device is decomposed, using above-mentioned best-evaluated set of factors extracting method, relay protection Device includes multiple assessment factor classes, assigns power method according to factor class, and each factor class corresponds to factor class weighted value, forms relay The factor class assessment level of protection device；Each assessment factor class includes similar subset of factors again, with the factor of assessment factor In class class, weight assigns power method, obtains weighted value in the class of each factor in factor class, forms the factors assessment of protective relaying device Level.Ultimately form the assessment hierarchical system of protective relaying device.

Obtain the assessment hierarchical system of secondary circuit in the same manner.

Finally built the evaluation index system of a comprehensive, objective and complete relay protection system.

Evaluation index system with relay protection system is estimated, according to by low to upper level be estimated with comprehensive Close, finally give the scoring of relay protection system.It is first depending on factor scores standard and carries out factor scores, in conjunction with the factor of factor In class class, weighted value carries out scoring comprehensively, is calculated factor class comprehensive grading；On this basis, according to factor class weighted value, COMPREHENSIVE CALCULATING obtains the overall score of protection device.Obtain the overall score of secondary circuit in the same manner.Take relay protection device corresponding each Minima in secondary circuit and the scoring of each protective relaying device, as the final scoring of relay protection device.Take relay protection Devices in system scores minima as the final scoring of relay protection system.

Carry out the process of relay protection system assessment with reference to example introduction with evaluation index system.

Figure according to Figure 12 scores to operating ambient temperature in protective relaying device:

The scoring gk of this factor as shown in Figure 12 when obtaining the temperature of running environment, can be obtained_i, then root again According to weights in above-mentioned best-evaluated factor and tax power method, and factor class class, obtain in corresponding running environment factor class class The weights of middle temperatureScoring for middle humidity in running environment factor class class in the same manner isWeightsUsing formula

$g_k=g_{k_i}\&centerdot;w_{k_1}+g_{k_2}\&centerdot;w_{k_2}---\left(27\right)$

Wherein g_kIt is the scoring of running environment.

Can obtain the failure condition of software and hardware in the same manner successively, counter arrange implementation of conditions, performance factor (rco), insulate shape Condition, data sampling, communication condition, lane testing situation factor class finally scores.

Overall score g for i-th protective relaying device_i, by the above-mentioned g obtaining the scoring of each factor class_k, in conjunction with because Plain class weight w_k, using formula:

$g_i=\underset{k=1}{\overset{t}{\mathrm{\&sigma;}}}{w}_k\&centerdot;g_k---\left(28\right)$

The overall score of each secondary circuit can be obtained in the same manner.

For the final scoring of j-th relay protection device, the corresponding each secondary circuit of taking equipment is filled with each relay protection Minima min (g in commenting on point_i), as the final scoring g of relay protection device^j.So final commenting of relay protection system Divide g, take scoring minima min (g in system equipment^j) as relay protection system scoring.By above-mentioned relay protection system The system of this bulky complex of electric relay protection can be passed through to decompose by different level, such that it is able to comprehensive by evaluation index system Each level of careful analysis, carries out more objective comprehensive analysis to each level, obtains relay protection system correct With objective condition evaluation results, it is convenient for repair based on condition of component.

Claims (1)

1. a kind of construction method of system of relay protection State Assessment Index System, is characterized in that, it comprises the following steps:

(1) build Recurison order hierarchy model using based on analytic hierarchy process (AHP), relay protection system is divided according to equipment, equipment Divide according to protective relaying device and secondary circuit, relay protection system state estimation is decomposed into simple, maneuverable many Hierarchical model is analyzed and assesses；

(2) the candidate's appraisal parameters determining protective relaying device and secondary circuit based on the Group Decision Method of entropy are first used, then Candidate's appraisal parameters are carried out more fully with core principle component analysis method again and objectively extract further, most preferably commented Estimate set of factors；

(3) embody its significance level and power of influence by giving different weight coefficients to different factors, optimal set of factors is entered The subjective weights of the enabling legislation based on fuzzy evaluation for the row, more optimal set of factors is carried out with the objective tax based on core principle component method Power, assigns the comprehensive weight that power obtains optimal set of factors for comprehensive two kinds；

(4) according to the relay protection system distinguishing hierarchy model set up, in conjunction with the optimal factor of evaluation collection extracting, using subjective and objective The weight system that combination weights method obtains, sets up complete relay protection state evaluation index system,

Wherein in core principle component analysis method, Polynomial kernel function and rbf kernel function is selected to set up a new kernel function, according to many Item formula kernel function:

k₁=(< x_i,x_j>+c)^d

x_iAnd x_jIt is respectively the sample value of candidate assessment factor i and candidate assessment factor j,

With rbf kernel function:

k²=exp (- | | x_i-x_j||²/2σ²), (i, j=1,2 ... are q)

Q is that the factor of evaluation extracting concentrates factor quantity,

Create a new kernel function by this two kernel functions, using closure one kernel function of creation of kernel function:

$k=\sqrt{\frac{{(k_1+k_2)}^2}{2}}=\sqrt{\frac{{({(<x_i,x_j>+c)}^d+\exp (-\left|\right|x_i-x_j|{|}^2/2{\mathrm{\&sigma;}}^2))}^2}{2}}$

In that step of comprehensive weight, choose formula

$\mathrm{\&alpha;}={\mathrm{\&lambda;}}^{\&prime;}+\mathrm{\&theta;}(r^{\&prime;}-{\mathrm{\&lambda;}}^{\&prime;})={\mathrm{\&lambda;}}^{\&prime;}+(1-\frac{r^{\&prime;}\&centerdot;{\mathrm{\&lambda;}}^{\&prime;}}{\left|\right|r^{\&prime;}\left|\right|\left|\right|{\mathrm{\&lambda;}}^{\&prime;}\left|\right|})/2(r^{\&prime;}-{\mathrm{\&lambda;}}^{\&prime;})$

As the method for comprehensive weight,

α is comprehensive weight, and λ ' is the weights that core principle component analysis obtain, and r' is the weights based on fuzzy evaluation.