CN103440398B - A kind of grid branch importance appraisal procedure based on pattern recognition - Google Patents

Abstract

The invention discloses a kind of grid branch importance appraisal procedure based on pattern recognition.On the basis of structure branch road moves back three kinds of risk indicators of fortune, use ISODATA algorithm that three-dimensional risk is carried out self-organizing clustering, determine each bar branch road safe class, then PCA methods analyst each bar branch road three-dimensional risk vector is utilized, determine the main constituent of all branch road risks, and using first principal component as the reference of Rate of aggregative risk index Design, it is achieved the design of the comprehensive evaluation index after Data Dimensionality Reduction, the importance ranking of branch road and importance classification thereof.The present invention uses branch road to move back fortune risk to assess the importance of branch road, compares relatively single assessment from the angle of topological structure and has more cogency；And branch road importance stage division based on ISODATA clustering algorithm and the branch road importance ranking of Based PC A are proposed, search and structure for core backbone frame are provided fundamental basis.

Description

A kind of grid branch importance appraisal procedure based on pattern recognition

Technical field

The invention belongs to the differentiation planning technology field of power system, particularly to a kind of electrical network based on pattern recognition Branch road importance appraisal procedure.

Background technology

In recent years, the most extreme frequent natural calamity occurs, and power system may be caused to occur from local to large area Power outage, had a strong impact on the safe operation of electrical network.Core backbone frame constructed by differentiation planning possesses reply certainly So ensure the ability of important load continued power when disaster and catastrophe failure.Build the branch road that it is critical only that of core backbone frame Importance is assessed.Traditional method is concentrated mainly on the analysis to topological structure of electric.Owing to the analysis method of topological structure does not has There is the probability considering topologies change, and seldom relate to Operation of Electric Systems characteristic and himself constraint.But, risk is commented The method of estimating has considered probability and the consequence of generation that accident occurs, and branch road is carried out importance and assesses more overall scientific.

In Study of Risk Evaluation Analysis for Power System field, the integrated approach of multi-risk System index typically uses the assessment side of weighted type Method, the determination that it is critical only that each index weights of various methods of weighting." the power system peace that Sun Fei delivers at thesis for the doctorate 2011 Full risk assessment and vulnerability " utilize 3 scaling laws of analytic hierarchy process (AHP) and 9 scaling laws to realize index power in conjunction with expert survey The distribution of weight；Liu Xindong etc. " push away with fuzzy based on Risk Theory what Electric Power Automation Equipment 2009,29 (2): 15-20 was delivered The Transient Security for Power Systems risk assessment of reason " use the integrated approach of norm weighting to carry out the comprehensive of security risk index. The weight coefficient of these risk integrative methods is affected bigger by subjective factors.The current research for branch road safety classification problem is also The most deep enough, Wang Ning etc. at east china electric power 2008,36 (3): 66-69 " the power system low voltage safeties based on risk delivered Early warning " voltage security is divided into 5 grades, principle of grading be directly according to Risk Calculation desired value come artificial uniformly set each The classification of risks of level is interval.Obviously, this stage division lacks necessary Theoretical Proof, and form is simple, and subjectivity is strong.

Summary of the invention

It is an object of the invention to explore a kind of grid branch importance appraisal procedure based on pattern recognition.The method exists On the basis of structure grid branch moves back fortune risk assessment index, the three-dimensional risk using ISODATA algorithm that branch road moves back fortune is carried out Self-organizing clustering, determines the importance rate of each bar branch road, then utilizes PCA principal component analytical method to determine all branch road risks Main constituent, it is achieved the importance ranking of branch road and importance degree classification thereof after Data Dimensionality Reduction, build for core backbone framework Theoretical foundation is provided.

For solving above-mentioned technical problem, the present invention adopts the following technical scheme that

A kind of grid branch importance appraisal procedure based on pattern recognition, the method includes the steps of:

Step 1, structure grid branch importance evaluation index；

The importance of branch road can move back after fortune other all branch roads or nodes in whole power system according to single branch road The consequence caused judges, considers probability and corresponding consequence thereof that accident occurs；

Branch road moves back fortune risk:

The branch road of definition power system move back fortune risk be branch road move back the probability of fortune with move back fortune after the product of consequence that produces, That is:

$Risk\left(Y\right|E_i)=P\left(E_i\right)\×\underset{j\≠i}{\Σ}\∫f\left(Y\right|E_i,L_j)\×Sev\left(Y\right)dY---\left(1\right)$

In formula, E_iRefer to that i-th branch road of power grid accident electrical network moves back fortune, P (E_i) refer to accident E_iThe probability occurred, typically Obey Poisson distribution；L_jFor j-th strip branch road；f(Y|E_i,L_j) it is that after branch road i moves back fortune, in system, branch road j is in specific run state The probability distribution of Y；Sev (Y) describes the severity of accident when specific run state Y.Specific run state Y includes branch road tide Stream, node voltage and node load.∫f(Y|E_i,L_j) × Sev (Y) dY refers to accident E_iAfter generation produce branch road j is corresponding Consequence.Self-explanatory characters event E_iConsequence summation to other all branch roads after generation, and Risk (Y | E_i) refer to The branch road of power system moves back fortune risk.

Branch road moves back fortune risk indicator:

Move back to other all branch roads in whole power system or the difference of the caused consequence of node after fortune according to branch road, will Branch road moves back fortune risk and is divided into overload risk, low-voltage risk and loses load risk three kinds.

(1) overload risk

After overload risk is power system accident, the active power of other non-fault branches is caused to exceed its rated value Probability and the combination of the order of severity, it may be assumed that

$Risk\left(P_L\right|E_i)=P\left(E_i\right)\underset{j\≠i}{\Σ}\∫f\left(P_L\right|E_i,L_j)\×Sev\left(P_L\right){\mathrm{dP}}_L---\left(2\right)$

Wherein, i=1,2 ..., n, j=1,2 ..., n, n are the circuitry number in system, f (P_L|E_i,L_j) it is that branch road i moves back fortune After, the branch power relative value P of branch road j in system_LProbability distribution；Sev(P_L) to describe branch power relative value be P_LCurrent events Therefore severity.∫f(P_L|E_i,L_j)×Sev(P_L)dP_LRefer to accident E_iThe overladen consequence after generation, branch road j produced.Self-explanatory characters event E_iOverload consequence summation to other all branch roads after generation, Risk (P_L|E_i) Refer to accident E_iOverload risk after generation.Overload severity Sev (P_L) depend on the trend of other all branch roads after accident Distribution, shown in severity function such as formula (3).

$Sev\left(P_L\right)=\left\{\begin{array}{cc}1& P_L>1\\ 10P_L-9& 0.9<P_L\&le;1\\ 0& P_L\&le;0.9\end{array}\right.---\left(3\right)$

In formula, P is branch power, P_eFor the rated power of this branch road, P_L=P/P_eFor this branch power relative value.

(2) low-voltage risk

Low-voltage risk is to cause after power system accident system interior joint voltage less than the probability of rated value and serious The combination of property, it may be assumed that

$Risk\left(V_B\right|E_i)=P\left(E_i\right)\underset{j=1}{\overset{m}{\&Sigma;}}\&Integral;f\left(V_B\right|E_i,L_j)\&times;Sev\left(V_B\right){\mathrm{dV}}_B---\left(4\right)$

Wherein, m is system interior joint number, f (V_B|E_i,L_j) it is after branch road i moves back fortune, the node voltage of system interior joint j Relative value V_BProbability distribution；Sev(V_B) to describe node voltage relative value be V_BTime accident severity.∫f(V_B|E_i,L_j)× Sev(V_B)dV_BRefer to accident E_iThe consequence of low-voltage after generation, node j produced.Self-explanatory characters Therefore E_iLow-voltage consequence summation to all nodes after generation, Risk (V_B|E_i) refer to accident E_iLow-voltage risk after generation. Low-voltage severity Sev (V_B) depend on the voltage of accident posterior nodal point, shown in severity function such as formula (5).

$Sev\left(V_B\right)=\left\{\begin{array}{cc}1& V_B\&GreaterEqual;0.9\\ 10-10V_B& 0.9<V_B\&le;1\\ 0& V_B>1\end{array}\right.---\left(5\right)$

In formula, V is node voltage, V_eFor the rated voltage of this node, V_B=V/V_eFor this node voltage relative value.

(3) load risk is lost

Losing load risk is that system loading node loses the probability of load and the combination of the order of severity after accident, it may be assumed that

$Risk\left(P_q\right|E_i)=P\left(E_i\right)\underset{i=1}{\overset{m_d}{\&Sigma;}}{P}_{qi}\&times;Sev\left(P_q\right)---\left(6\right)$

In formula, m_dFor load bus number, P_qiFor accident E_iThe load that rear i-th load bus loses, P_qLose for system Load, Sev (P_q) it is that system loses load P_qSeverity, Risk (P_q|E_i) it is accident E_iMistake load risk after generation.

Wherein, lose the ratio that load severity depends on losing load, shown in its severity function such as formula (7).

$Sev\left(P_q\right)=\left\{\begin{array}{cc}1& 0.3<P_q/P_{fh}\&le;1\\ \frac{10}{3}{P}_q/P_{fh}& 0<P_q/P_{fh}\&le;0.3\end{array}\right.---\left(7\right)$

In formula, P_fhFor the original loads of load bus in system, this numerical value is known quantity, and each load bus is original minus Lotus is different, P_qFor the mistake loading of load bus in system.

Step 2, structure three-dimensional risk vector

By overload risk, low-voltage risk and mistake load the risk forms three-dimensional risk vector x_i, it is expressed as:

x_i=(Risk (P_L|E_i),Risk(V_B|E_i),Risk(P_q|E_i)) (8)

Step 3, carry out branch road importance classification based on ISODATA clustering algorithm

ISODATA clusters, i.e. iteration self-organizing data analysis algorithm, is that the pattern of a kind of non-supervisory Dynamic Clustering Algorithm is known Other method.ISODATA uses Euclidean distance to carry out the similarity of analytical data itself, and its core concept is that Euclidean distance is the least, phase The biggest like degree, data high for similarity are flocked together automatically.

The thought of branch road importance classification i.e. sets accident set and moves back fortune as single branch road, calculates this respectively and moves back transport line road Overload risk, low-voltage risk and lose load risk, the three of every branch road value-at-risks are expressed as a three-dimensional risk to Amount, uses ISODATA clustering algorithm to be clustered according to data similarity by the three-dimensional risk vector of all branch roads, obtains every one-level Cluster centre and branch number, it is achieved every preliminary automatic classification of branch road importance rate；With cluster centre and initial point European away from From size judge branch road importance rate, cluster centre distance initial point is the most remote, the three-dimensional risk point distance that this rank is comprised Initial point is the most remote, and therefore the risk of these risk points is the biggest, and importance information belonging to it is the most important；Set by this patent Initial cluster center be random, branch road importance classification results is the most incomplete same, but occurs that classification results is inconsistent Place all at boundary.

The basic step of ISODATA clustering algorithm is as follows:

1) parameter is set: sample x to be sorted_i；Intended cluster centre number K；Initial cluster centre number N_c；Each Number of samples θ minimum in Clustering Domain_N；Standard deviation θ of sample range distribution in Clustering Domain_S；Narrow spacing between two cluster centres From θ_C；Judge the number of times I of the interative computation of circulation stopping_P；Distance D between two cluster centres_ij。

2) N is randomly selected_cIndividual sample is as the center of initial clustering；

3) by sample x_iIt is assigned to nearest cluster S_j.Rule is: if D_j=min (| | x_i-C_j| |), i=1,2 ..., p, j =1,2 ..., c, p are total sample number to be sorted, and c is the sum having chosen cluster centre；Then by x_iIt is grouped into cluster S_j.Wherein, C_j For jth cluster centre, D_jFor sample x_iTo the distance of jth cluster centre, this distance is the shortest.

4) center of each cluster is calculated:

$Z_j=\frac{1}{N_j}\underset{x_i\&Element;S_j}{\&Sigma;}{x}_i---\left(9\right)$

By new central value Z_jIt is set to the center C of cluster_j=Z_j, N_jFor cluster S_jIn sample number.

5) division.If the number of current cluster is less than intended clusters number K, then proceed by cluster division.

6) merge.When the centre distance of two clusters is less than minimum range θ at the two center_CTime, two Cluster mergings are One new cluster.If all distances between cluster centreBegin to merge.New cluster centre is:

$C_n=\frac{1}{N_i+N_j}\&lsqb;N_i{C}_i+N_j{C}_j\&rsqb;---\left(10\right)$

Wherein, C_iAnd C_jIt is respectively the i-th class and the cluster centre of jth class, N_iAnd N_jRespectively cluster S_iAnd S_jClass number.

7) if iterations reaches maximum iteration time I_P, or process convergence, then iterative process terminates, otherwise I_P=I_P+ 1, return to step 3.

Step 4, Based PC A method carry out branch road importance ranking

Principal component analysis (PCA) method is that a kind of area of pattern recognition that is widely used in finds multidimensional data key property Statistical analysis technique, its main purpose is that high dimensional data is projected to lower dimensional space, parses main component from multidimensional data, Disclose multidimensional data effective information, Simplified analysis challenge.

The main thought of PCA is that former N-dimensional column vector is used linear transformation, obtains sorting from high to low according to importance New N column vector.In the new column vector obtained, choose the maximum M (M ＜ N) of importance tie up subvector, as former column vector Main constituent.

Note x₁,…,x_pFor p component of original column vector, if the component ξ listd after Bian Huan_i, i=1,2 ..., p, is former The linear combination of column vector subcomponent, sets the mould of linear combination coefficient as 1, i.e.

${\mathrm{\&alpha;}}_i^T{\mathrm{\&alpha;}}_i=1---\left(11\right)$

α_iVectorial, for column vector for linear combination coefficient.This p α_iConstitutive characteristic transformation matrix A.Optimum orthogonal transformation A Each component α_iMake corresponding ξ_iVariance reach extreme value, data will more discrete be divided, and similarity is lower, the most just represents more Many information.Require each column vector pairwise orthogonal of composition A, it is ensured that two pairwise uncorrelateds between the new component obtained simultaneously.Additionally, If the variance of certain dimension component is the biggest, this component is the most important, has more information.

Using three-dimensional risk vector as source data, utilize PCA method that three-dimensional risk vector is carried out dimensionality reduction, retain source data In main information, obtain the maximum principal direction of all three-dimensional risk data, then by three-dimensional risk vector projection to principal direction The one-dimensional data that can clearly differentiate different classes of (i.e. the automatic classification results of ISODATA), Jin Ercan just can be obtained on axis Integrated risk index calculating is carried out according to PCA dimensionality reduction result.First principal component ξ₁It is integrated risk, computing formula such as formula (12) Shown in.

${\mathrm{\&xi;}}_1={\mathrm{\&alpha;}}_1^{T}x---\left(12\right)$

Wherein,For the weight vectors of three-dimensional risk indicator, representing the normalization importance proportion of three-dimensional risk, x is for propping up The three-dimensional risk vector on road.ξ₁For integrated risk, the i.e. space risk point distance projecting to initial point on principal direction axle.

Obtain the integrated risk of all branch roads according to formula (12), carry out branch road importance ranking according to integrated risk value, combine Closing risk the biggest, after this branch trouble, the impact on whole power system is the biggest, and therefore this branch road is the most important.

Step 5, the judgement of branch road importance ranking and classification adjust

According to branch road importance ranking result, the up-and-down boundary branch road of mark spatial scalability projection, if adjacent two important There is not intersection in the projection of rank branch road, then the classification point that can set this two-stage is average as the risk of the border branch road of this two-stage Value；If there is the intersection in allowed band, be then as the criterion with the border branch road risk that important level is high, fall border branch road away from The subpoint in initial point direction is all automatically made inclined important level, according to the risk average of new two-stage border branch road as this The classification point of two-stage.Set classification and intersect percentage ratio as intersecting circuitry number N_jcAccount for all circuitry number N_zPercentage ratio, such as formula (13) Shown in.

$\mathrm{\&psi;}=\frac{N_{jc}}{N_z}\&times;100\%---\left(13\right)$

When classification intersection percentage ratio meets ψ ＜ 10%, it is believed that, use first principal component can comprehensively embody p finger Target three dimensions rating information, in i.e. classification intersects at allowed band.If being unsatisfactory for ψ ＜ 10%, then jump to step 3, weight Parameter is newly set.

The invention have the advantages that

1, the present invention use branch road move back fortune risk to assess the importance of branch road, more can from electrical network characteristic and self constraint Angle reacts the importance of branch road, compares relatively single assessment from the angle of topological structure and has more cogency.

2, the present invention proposes branch road importance stage division based on ISODATA clustering algorithm, constructs vector form Risk indicator, three-dimensional risk vector higher for similarity is collected as a class, this method compares existing stage division more There is theoretical basis.

3, the present invention proposes the branch road importance ranking of Based PC A, and fortune Rate of aggregative risk function is moved back in design, with all The three-dimensional risk vector of electrical network, as integrated risk value, is dropped into one-dimensional by risk vector first principal component, can relatively be as the criterion simultaneously The classification results that true reduction is three-dimensional.

4, the present invention has carried out sequence and classification to branch road importance, and the search for bulk transmission grid provides precondition.

Accompanying drawing explanation

Fig. 1 is the schematic flow sheet of grid branch importance based on pattern recognition assessment；

Fig. 2 is grid branch importance classification figure based on pattern recognition；

Fig. 3 is branch road importance ranking figure based on principal component analysis.

Detailed description of the invention

46 branch roads of IEEE39 node system are entered by a kind of grid branch importance appraisal procedure based on pattern recognition Row prominence score level and sequence.The method comprises the steps of

A, structure grid branch importance evaluation index；In the present embodiment, the fluctuation of assumed load allocation factor is 5%, And in this, as the change of system operational parameters.Disconnect successively if contingency set is each branch road in system, these The year on road cut-offs rate λ_yIt is all 0.3.

A1, calculating branch road move back the probability of fortune；

A2, according to three kinds of risk severity functions, ask for considering the single branch road of operational factor change move back fortune after three kinds The order of severity of risk；

A3, utilization are moved back fortune methods of risk assessment based on branch road and are calculated three kinds of value-at-risks of each branch road, its result such as table 1 Shown in, accompanying drawing 2 gives three kinds of Risk Calculation results of all branch roads.

A4, three risk indicators that branch road moves back fortune constitute three-dimensional risks vector, obtain 46 branch roads of 39 node systems Three-dimensional risk vector.

Table 1

B, use ISODATA algorithm, the three-dimensional risk of 46 branch roads is carried out importance classification, be divided into one-level, two grades and Three grades, respectively with * ,+and represent, classification results is shown in Table shown in 1 and Fig. 2.

C, employing PCA methods analyst branch road move back the three-dimensional risk vector of fortune, obtain the first master of all three-dimensional risk vectors All three dimensions risk points are projected, as shown in Figure 2 by composition to the principal direction axle corresponding to first principal component.Three-dimensional wind The normalized weight coefficient vector in danger is:

D, according to α₁ ^TWith the three-dimensional risk vector of all branch roads, the comprehensive wind after all branch road risk dimensionality reductions can be calculated Danger value.Branch road integrated risk value is shown in Table 1.

E, branch road integrated risk is ranked up, obtains branch road importance ranking result, as shown in figure 3 and table 2.

F, preliminary classification being analyzed with importance ranking result, the branch road that obtains intersecting is L30, the intersection existed Branch road percentage ratio ψ=2.2 ＜ 10%, in allowed band, demonstrates effectiveness and the feasibility of this risk integrative method.Table The branch road importance classification point of 3 value-at-risks giving border branch road and 39 node systems.

Table 2

Table 3

Claims (2)

1. a grid branch importance appraisal procedure based on pattern recognition, it is characterised in that: comprise the following steps,

Step 1, structure grid branch importance evaluation index；

Other all branch roads or node in whole power system are caused after moving back fortune according to single branch road by the importance of branch road Consequence judge, consider probability and corresponding consequence thereof that accident occurs；The branch road of definition power system moves back fortune risk Risk(Y|E_i) be branch road move back the probability of fortune with move back fortune after the product of consequence that produces, it may be assumed that

Wherein, E_iRefer to that i-th branch road of electrical network moves back fortune, P (E_i) refer to accident E_iThe probability occurred, obeys Poisson distribution；L_jFor jth Bar branch road；f(Y|E_i,L_j) it is after branch road i moves back fortune, in system, branch road j is in the probability distribution of specific run state Y, Sev (Y) table Showing the severity of accident when specific run state Y, specific run state Y includes Branch Power Flow, node voltage and node load, ∫f(Y|E_i,L_j) × Sev (Y) dY refers to accident E_iThe corresponding consequence after generation, branch road j produced,Self-explanatory characters event E_iConsequence summation to other all branch roads after generation；

Moving back after fortune other all branch roads in whole power system or the difference of the caused consequence of node according to branch road, branch road moves back Fortune risk includes overload risk, low-voltage risk, loses load risk, and circular is as follows:

Overload risk Risk (P_L|E_i) calculating:

Wherein, i=1,2 ..., n, j=1,2 ..., n, n are the circuitry number in system, f (P_L|E_i,L_j) it is, after branch road i moves back fortune, to be The branch power relative value P of branch road j in system_LProbability distribution；Sev(P_L) to describe branch power relative value be P_LTime accident Severity；∫f(P_L|E_i,L_j)×Sev(P_L)dP_LRefer to accident E_iThe overladen consequence after generation, branch road j produced；Self-explanatory characters event E_iOverload consequence summation to other all branch roads after generation；Overload severity Sev(P_L) depend on the trend distribution of other all branch roads after accident, it is embodied as:

In formula, P_L=P/P_eFor this branch power relative value, P is branch power, P_eRated power for this branch road；

Low-voltage risk Risk (V_B|E_i) calculating:

Wherein, m is system interior joint number, f (V_B|E_i,L_j) it is after branch road i moves back fortune, the node voltage of system interior joint j is relative Value V_BProbability distribution；Sev(V_B) to describe node voltage relative value be V_BTime accident severity, ∫ f (V_B|E_i,L_j)×Sev (V_B)dV_BRefer to accident E_iThe consequence of low-voltage after generation, node j produced,Self-explanatory characters' event E_iLow-voltage consequence summation to all nodes after generation, low-voltage severity Sev (V_B) depend on the voltage of accident posterior nodal point, It is embodied as:

In formula, V_B=V/V_eFor this node voltage relative value, V is node voltage, V_eRated voltage for this node；

Lose load risk Risk (P_q|E_i) calculating:

In formula, m_dFor load bus number, P_qiFor accident E_iThe load that rear i-th load bus loses, P_qFor bearing that system loses Lotus, Sev (P_q) it is that system loses load P_qSeverity, be embodied as:

In formula, P_fhFor the original loads of load bus, P in system_qFor the mistake loading of load bus in system；

Step 2, by overload risk, low-voltage risk and lose load the risk forms three-dimensional risk vector x_i, it is embodied as:

x_i=(Risk (P_L|E_i),Risk(V_B|E_i),Risk(P_q|E_i))

x_iIt it is the three-dimensional risk vector of i-th branch road；

Step 3, carry out branch road importance classification based on ISODATA clustering algorithm；

The process of branch road importance classification is carried out particularly as follows: use ISODATA clustering algorithm by institute based on ISODATA clustering algorithm The three-dimensional risk vector having branch road clusters according to data similarity, obtains cluster centre and the branch number of every one-level, it is achieved every The preliminary automatic classification of bar branch road importance rate；Branch road importance etc. is judged with the size of cluster centre Yu initial point Euclidean distance Level, cluster centre distance initial point is the most remote, three-dimensional risk point that this rank is comprised distance initial point is the most remote, therefore these risk points Risk is the biggest, and importance information belonging to it is the most important；

Step 4, Based PC A method carry out branch road importance ranking；Detailed process is as follows:

Using three-dimensional risk vector as source data, utilize PCA method that three-dimensional risk vector is carried out dimensionality reduction, retain in source data Main information, obtains the maximum principal direction of all three-dimensional risk data, then by three-dimensional risk vector projection to principal direction axis On can clearly differentiate different classes of one-dimensional data to obtain, and then carry out integrated risk index with reference to PCA dimensionality reduction result Calculate；First principal component ξ₁Being integrated risk, computing formula is as follows:

Wherein,For the weight vectors of three-dimensional risk indicator, representing the normalization importance proportion of three-dimensional risk, x is branch road Three-dimensional risk vector, ξ₁For integrated risk, the i.e. space risk point distance projecting to initial point on principal direction axle；

Obtain the integrated risk of all branch roads according to above formula, carry out branch road importance ranking, integrated risk according to integrated risk value The biggest, after this branch trouble, the impact on whole power system is the biggest, and therefore this branch road is the most important；

Step 5, judge that branch road importance ranking and classification adjust；

According to branch road importance ranking result, mark the up-and-down boundary branch road of spatial scalability projection, if adjacent two severity levels There is not intersection in the projection of branch road, then can set the classification point risk average as the border branch road of this two-stage of this two-stage； If there is the intersection in allowed band, then it is as the criterion with the border branch road risk that important level is high, falls at border branch road away from former Point direction subpoint be all automatically made inclined important level, according to new two-stage border branch road risk average as this two The classification point of level, sets classification and intersects percentage ratio as intersecting circuitry number N_jcAccount for all circuitry number N_zPercentage ratio, as follows,

When classification intersection percentage ratio meets ψ ＜ 10%, first principal component is used can comprehensively to embody the three dimensions of p index Rating information, in i.e. classification intersects at allowed band；If being unsatisfactory for ψ ＜ 10%, then jump to step 3, Reparametrization.

A kind of grid branch importance appraisal procedure based on pattern recognition the most according to claim 1, it is characterised in that: ISODATA clustering algorithm in described step 3 comprises the following steps；

Step 3.1, parameter is set: sample x to be sorted_i；Intended cluster centre number K；Initial cluster centre number N_c；Often Number of samples θ minimum in one Clustering Domain_N；Standard deviation θ of sample range distribution in Clustering Domain_S；Minimum between two cluster centres Distance theta_C；Judge the number of times I of the interative computation of circulation stopping_P；Distance D between two cluster centres_ij；

Step 3.2, randomly select N_cIndividual sample is as the center of initial clustering；

Step 3.3, by sample x_iIt is assigned to nearest cluster S_j；Rule is: if D_j=min (| | x_i-C_j| |), i=1,2 ..., p, J=1,2 ..., c, p are total sample number to be sorted, and c is the sum having chosen cluster centre；Then by x_iIt is grouped into cluster S_j；Wherein, C_j For jth cluster centre, D_jFor sample x_iTo the distance of jth cluster centre, this distance is the shortest；

Step 3.4, calculate the center of each cluster:

By new central value Z_jIt is set to the center C of cluster_j=Z_j, N_jFor cluster S_jIn sample number；

Step 3.5, division；If the number of current cluster is less than intended clusters number K, then proceed by cluster division；

Step 3.6, merging；When the centre distance of two clusters is less than minimum range θ at the two center_CTime, two Cluster mergings are One new cluster；If all distances between cluster centreBeginning to merge, new cluster centre is:

Wherein, C_iAnd C_jIt is respectively the i-th class and the cluster centre of jth class, N_iAnd N_jRespectively cluster S_iAnd S_jClass number；

If step 3.7 iterations reaches maximum iteration time I_P, or process convergence, then iterative process terminates, otherwise I_P= I_P+ 1, return to step 3.