CN105093066A - Line fault judgment method based on wavelet analysis and support vector machine - Google Patents

Abstract

The invention belongs to the technical field of power system line fault judgment, and particularly relates to a line fault judgment method based on wavelet analysis and a support vector machine. The method comprises steps: firstly, fault current signals in a wave recording system are extracted, and the wavelet analysis technology is adopted for extracting and analyzing feature information of the fault current signals; then, a wavelet energy entropy theory is used for decomposing the fault current signals, an energy entropy value corresponding to each phase of current is calculated, and zero-sequence current is combined for building a four-dimensional feature vector for transmission line fault classification; and finally, a two-layer classification model is used for specific fault judgment. The judgment method comprises a linear classification module and a nonlinear classification module, wherein linear classification is initial classification on data samples according to the two-layer classification structure of the zero-sequence current and threshold parameters; and on the basis, according to data features for transmission line small sample fault, the support vector machine is used for nonlinear classification on multiple kinds of fault data, and transmission line fault judgment is finally realized.

Description

Based on the line fault determination methods of wavelet analysis and support vector machine

Technical field

The invention belongs to line fault of electrical power system judgment technology field, particularly relate to a kind of line fault determination methods based on wavelet analysis and support vector machine.

Background technology

It is far that transmission line of electricity is crossed on the one hand, be generally tens to several thousand kms, be chronically exposed to the open air of harsh environmental conditions on the other hand, cannot effectively safeguard, compare with other electrical equipments, the conditional decision residing for transmission line of electricity it be the ring the most easily broken down in electric system.On transmission line of electricity, the improper connection between to be also the most dangerous fault while of the most common be phase and phase or phase and ground, i.e. short circuit.These faults are divided into single-line to ground fault, two-phase phase fault, two-phase grounding fault and three-phase ground short circuit in electric system.Wherein common with single-line to ground fault, and three-phase shortcircuit is more rare.Short circuit can produce very large short-circuit current when occurring, and makes voltage in system greatly reduce simultaneously.The thermal effect of short dot short-circuit current and short-circuit current and mechanical effect directly can damage electrical equipment.Voltage drop affects the normal work of user, affects product quality.The consequence that short circuit is more serious, is because voltage drop may cause the stability of paired running between electric power system power plant to wreck, causes system oscillation, until whole system is disintegrated.Therefore the short trouble diagnosis of transmission line of electricity is an emphasis of power system failure diagnostic.

When breaking down in electric system, being attended by the generation of higher hamonic wave, for avoiding the harmful effect of these harmonic waves, being necessary to be analyzed it and suppress.The conversion of this type of signal is projected to the characteristic that different yardsticks can show significantly these high frequencies, unusual higher hamonic wave signal by wavelet analysis.Particularly wavelet packet has the characteristic segmented further frequency space, will well for suppressing higher hamonic wave to provide reliable foundation.Wavelet transformation can characterize the data of the electrical power system transient signal aspect required for analysis that other signal analysis technologies cannot meet.Under normal circumstances, the fast algorithm of the wavelet transformation Multiresolution Decomposition of transient signal is expressed, the signal under utilizing Orthogonal Wavelets signal decomposition to be become different frequency.It equals the high pass of recursive filtering and low-pass filter is analyzed signal.

At present, adopt wavelet transformation to carry out breakdown judge mainly Wavelet Entropy to be used for identifying fault in the heuritic approach such as neural network or fuzzy system.Generate Wavelet Entropy proper vector by wavelet transformation and wavelet time-frequency parameter, then identify fault in conjunction with neural network.This can obtain very complicated model and good recognition effect when doing theoretical research, Fault Identification ability is strong.But for on-the-spot actual, the system complex designed by this method, well can not be suitable for practical application, and different scenes exists each species diversity, the method versatility is poor.Meanwhile, identify that the time that fault needs is long, be unfavorable for power system stability and economical operation.

Although nowadays some new heuritic approaches, as more in the research in electric power system fault judgement such as neural network, Bayesian network, fuzzy set algorithm, these researchs are much all in theoretical research stage, and practical application limitation is very large.In the face of number of nodes is huge, the provincialism power transmission network of real network complexity, these method application difficult.

Summary of the invention

In order to solve the problems referred to above that existing method exists, the present invention proposes a kind of line fault determination methods based on wavelet analysis and support vector machine, comprising:

Step 1: extract three-phase current signal as training dataset from the historical data base of transmission line of electricity record wave system system;

Step 2: adopt wavelet analysis to calculate wavelet energy entropy corresponding to three-phase current;

Step 3, by whether there is zero-sequence current after judging transmission line malfunction, being two classes by transmission line malfunction Preliminary division, if there is zero-sequence current, is ground short circuit fault, if there is not zero-sequence current, is phase fault; Wherein ground short circuit fault comprises single-line to ground fault and two-phase grounding fault; Phase fault comprises two-phase phase fault and three-phase shortcircuit;

Step 4: find out maximal value max1, intermediate value max2 and the minimum value min in wavelet energy entropy corresponding to three-phase current, introduces three-phase current wavelet energy entropy ratio R ₁and R ₂:

$R_1=\frac{max1}{max2},R_2=\frac{max2}{\min }$

Setting threshold value alpha and beta, works as R ₁this fault of > alpha belongs to single-phase grounding fault; Work as R ₁this fault of≤alpha belongs to two-phase short circuit and ground fault; Work as R ₂this fault of > beta belongs to two-phase phase fault; Work as R ₂this fault of≤beta belongs to three phase short circuit fault;

Step 5: trained as training sample input support vector classification by wavelet energy entropy corresponding for three-phase current, sets up 3 two classifiers be connected successively of corresponding three kinds of single-phase grounding faults, 3 two classifiers be connected successively of three kinds of two-phase phase faults and 3 two classifiers be connected successively of three kinds of two-phase short circuit and ground faults respectively; The decision function of each two classifiers carrys out failure judgement type by calculating the final analysis result that exports;

Step 6, repetition step 1 ~ 5, the parameter of the multi-group data sample concentrated by training data to threshold value alpha and beta and support vector classification is trained and optimizes, the judged result of fault type and historical data are verified, after meeting error expected rate, the actual transmission line of electricity three-phase current signal that will judge is gathered, and performs step 2 ~ 5 to judge actual transmission line malfunction type; If do not meet error expected rate, repeated execution of steps 6.

Described step 2 specifically comprises:

During given discrete signal x (k), at moment k and yardstick j rapid conversion, after conversion, obtain high fdrequency component D _j(k) and low frequency component A _j(k); Band information is included in component of signal D _j(k) and A _jin (k), obtain reconstruction in the following manner:

$\left\{\begin{array}{c}{D}_j\left(k\right):\[2^{-\left(j+1\right)}{f}_s,2^{-j}{f}_s\]\\ A_j\left(k\right):\[0,2^{-\left(j+1\right)}{f}_s\]\end{array}\right.,(j=1,2,...,m)---\left(1\right)$

Be expressed as by original signal sequence x (k) after wavelet transform:

$x\left(k\right)=D_1\left(k\right)+A_1\left(k\right)=D_1\left(k\right)+D_2\left(k\right)+A_2\left(k\right)=\underset{j=1}{\overset{J}{\Σ}}{D}_j\left(k\right)+A_j\left(k\right)---\left(2\right)$

Wherein, f _sbe discrete signal samples frequency, m represents the decomposition scale to signal at a time, and J is positive integer;

E _jkbe Wavelet Energy Spectrum under moment k and yardstick j, computing method are as follows:

E _jk＝|D _j(k)| ²(3)

Adopt and develop and next wavelet energy entropy computing method on the basis of information entropy, computing formula is as follows

At j yardstick up-sampling sequence k=1,2 ..., the signal energy summation E of N _jfor:

$E_j=\underset{k=1}{\overset{N}{\Σ}}{E}_{jk}---\left(4\right)$

In order to be consistent with the time probability in information entropy, suppose p _jk=E _jk/ E _j, then p _jkthe ratio that wavelet energy under expression k moment j yardstick is shared in all moment wavelet energy sums under j yardstick, defines corresponding wavelet energy entropy W _eE:

$W_{EE}=-\underset{k}{\Σ}{p}_{jk}{\mathrm{logp}}_{jk}---\left(5\right)$

Finally, the wavelet energy entropy summation that each phase current is corresponding is calculated:

$S_a=\underset{j}{\Σ}{W}_{aj},S_b=\underset{j}{\Σ}{W}_{bj},S_c=\underset{j}{\Σ}{W}_{cj}---\left(6\right)$

Wherein, W _aj, W _bj, W _cjrepresent the wavelet energy entropy that calculate of three-phase current by formula (5) respectively, S _a, S _b, S _crepresent the wavelet energy entropy summation of three-phase current respectively;

After determining decomposition scale, the wavelet energy entropy on each yardstick is calculated according to formula (5), this definition reflects its energy distribution in frequency space of certain signal, characterizes the characteristic information of respective signal, and then reach the object of feature information extraction according to the distribution of energy.

Beneficial effect of the present invention is: as long as obtain transmission line of electricity current signal, just can judge whether corresponding transmission line of electricity breaks down and fault type, there is good booster action to spot dispatch personnel quick, accurate localizing faults region after fault occurs, contribute to the safety and stability improving transmission line of electricity.

Accompanying drawing explanation

Fig. 1 is transmission line short-circuit fault decision flow chart.

Fig. 2 a ~ 2c is short trouble support vector machine recognition system.

Embodiment

Below in conjunction with accompanying drawing, embodiment is elaborated.

In order to verify the validity of power system transmission line short trouble disaggregated model and the rationality of optimum configurations that propose above, gathering somewhere Utilities Electric Co. recorder data formation test set and verifying.Because the setting of parameter in the present invention is arranged according to expertise, so test set also will be revised parameter according to the accuracy rate of failure modes while certificate parameter, to obtain more reasonably parameter.As shown in Figure 1, concrete grammar step comprises:

Step 1: choose data and carry out Fault Classification parameter adjustment from the historical data base of record wave system system.Every bar transmission line of electricity all can be equipped with corresponding fault oscillograph, and obtained the record ripple signal of corresponding line by sensor collection, record ripple signal is herein discrete data sequence.A recorded wave file can comprise the multinomial data of corresponding transmission line of electricity, and this file is stored in system recorder data storehouse, and comprises a large amount of data messages in this system, therefore will extract fault current data and first will resolve recorder data.According to this data system technical specification, resolve recorder data and only need Analysis for CO MTRADE file.Comprising four files in each COMTRADE record, is header files, configuration file, data file and message file respectively.Configuration file contains the information that computer program needs in order to correct unscrambling data (.DAT) file, the items such as these information comprise sampling rate, port number is put, line frequency, channel information.Algorithm of the present invention only uses record ripple electric current, so be first converted to transmission line of electricity record ripple current sequence according to set data conversion rule from above-mentioned recorded wave file; Then generate current sequence text, this sequence data acquisition interval is 0.3125ms, namely gathers 3200 equally spaced data p.s..When adjusting parameter, need to select the data of proper time period as training dataset and test data set, the data when packet of parameter training collection of the present invention and test set breaks down containing 35kv, 110kv and 220kv transmission line of electricity, the diversity of training sample ensure that the rationality of optimum configurations, added up by the transmission line short-circuit fault occurred year October in January, 2011 to 2012 somewhere grid company, the 164 groups of typical datas chosen wherein carry out training and the checking of disaggregated model as sample.Wherein 90 groups of data samples are as training data; 74 groups of samples, as test data, are used for the validity of verification method and the accuracy of parameter.

Step 2: data prediction

Data prediction is carried out for the record ripple current data in training set.

Step 2.1: by wavelet function " db4 ", 5 layers of wavelet decomposition are carried out to fault-current signal, obtain the high fdrequency component under different scale and low frequency component.And the low frequency coefficient a obtained under each yardstick _jwith high frequency coefficient d _j.

Step 2.2: calculate Wavelet Energy Spectrum.The wavelet energy of yardstick j moment k equal high frequency coefficient absolute value square, i.e. E _jk=| D _j(k) | ².Thus the Wavelet Energy Spectrum on yardstick j is

Step 2.3: calculate wavelet energy entropy.If E _jkfor the wavelet energy that signal x (k) is inscribed when j yardstick k.Then represent at j yardstick up-sampling sequence k=1,2 ..., the signal energy summation of N.Each component energy E _jsummation be signal gross energy.In order to be consistent with the time probability in information entropy, suppose p _jk=E _jk/ E _j, then so just, corresponding wavelet energy entropy can be defined: $W_{EE}=-\underset{k}{\Σ}{p}_{jk}\log p_{jk}.$

Step 2.4: calculate wavelet energy entropy summation.That is, wherein, W _aj, W _bj, W _cjrepresent the wavelet energy entropy that three-phase current is calculated by step 2.3.

Step 2.5: wavelet energy entropy corresponding to each phase current can be obtained by above four steps.I.e. S _a, S _b, S _c.

Step 3: parameter initialization

The present invention has used two parameter alpha and beta in failure modes linear block, relates to the correlation parameter of 9 classifiers in Nonlinear Classification module.According to estimating and the theoretical analysis of transmission line of electricity current characteristics failure modes model, above-mentioned linear classification module parameter is initialized as: alpha=5.0 and beta=12.0; Nonlinear Classification module parameter is initialized as support vector machine default value.

Step 4: fault diagnosis flow scheme

By carrying out the transient state process of transmission line short-circuit fault deep researching and analysing discovery, for single-phase earthing fault, be short-circuited for A phase, high fdrequency component in A phase fault after-current signal can significantly increase, the complexity that after utilizing wavelet decomposition, it distributes on time-frequency domain increases greatly, according to the known wavelet energy entropy S in this case calculated of the feature of wavelet energy entropy characterization signal _acompare the wavelet energy entropy S that healthy phases is corresponding _b, S _cmuch bigger; For two-phase phase fault, for A, B phase phase fault, the wavelet energy entropy S that fault phase A, B calculate _a, S _bthe wavelet energy entropy S of relative healthy phases C _cmuch bigger; If when there is A, B, C three phase short circuit fault, its corresponding wavelet energy entropy S _a, S _band S _ccapital very greatly and relatively.Therefore, the short trouble of transmission line of electricity 10 type can pass through the analysis of failure phase wavelet energy entropy S corresponding with healthy phases _a, S _band S _cbetween magnitude relationship and ratio carry out failure modes.

If directly adopt support vector machine to classify to fault in 10, classification accuracy is relatively low.By analysing in depth the characteristic signal of ground short circuit and phase fault, find that ground short circuit exists zero-sequence current and (is denoted as I ₀, and phase fault does not have zero-sequence current (to be denoted as I=1) ₀=0), because three-phase ground short trouble belongs to symmetric fault, thus its zero-sequence current is also 0.Thus the present invention introduces zero-sequence current and tentatively 10 kinds of faults is divided into ground short circuit fault and phase fault.The wavelet energy entropy value tag that each phase current of dissimilar fault is corresponding and zero-sequence current value thereof are as listed in table 1.On this basis, concrete failure modes comprises the following steps:

The wavelet energy entropy feature of the dissimilar fault of table 1 and zero-sequence current

Step 4.1: carry out first step linear classification according to zero-sequence current

The wavelet energy entropy feature that short trouble three-phase current dissimilar is as can be seen from Table 1 corresponding is different, and in order to improve accuracy and the validity of failure modes, first the present invention adopts the method for decision tree to carry out preliminary classification to fault.By introducing the zero-sequence current after transmission line malfunction, tentatively transmission line malfunction can be divided into two classes, i.e. earth fault and phase-to phase fault.Wherein earth fault comprises single-line to ground fault and two-phase grounding fault; Phase-to phase fault comprises two-phase phase fault and three-phase shortcircuit (because three-phase shortcircuit belongs to symmetric fault, zero-sequence current is 0, thus can put phase-to phase fault under according to zero-sequence current division).

Step 4.2: carry out second step linear classification according to threshold parameter

By carefully analyzing of his-and-hers watches 1, find that the wavelet energy entropy difference of fault phase and healthy phases is obvious, and no matter S under which kind of failure condition _a, S _band S _call can not be completely equal, dividing, so can compare by size maximal value, intermediate value and the minimum value determined wherein of size must be had.

If maximal value is max1, intermediate value is max2, minimum value is min, introduce three-phase current wavelet energy entropy ratio

$R_1=\frac{max1}{max2},R_2=\frac{max2}{\min }---\left(7\right)$

For earth fault, if single phase ground fault fault, then the maximal value max1 in the wavelet energy entropy that three-phase current is corresponding corresponding fault phase, intermediate value max2 and minimum value are min corresponding healthy phases, now R ₁represent the ratio of fault phase and the wavelet energy entropy of healthy phases, this value is larger; If generation two-phase short circuit and ground fault, then the corresponding fault phase of maximal value max1 and intermediate value max2, minimum value min corresponding healthy phases, now R ₁represent the ratio of the wavelet energy entropy of two fault phases, from the known R of theoretical analysis ₁approximate 1.Therefore can divide earth fault by arranging threshold value alpha, namely working as R ₁during >alpha, this fault belongs to single-phase grounding fault; Work as R ₁during <alpha, this fault is two-phase short circuit and ground fault.By three-phase current wavelet energy entropy ratio R ₁just earth fault can be divided into single-phase grounding fault and two-phase short circuit and ground fault again with threshold value alpha.

For phase-to phase fault, if there is two-phase phase-to phase fault, then the corresponding fault phase of the maximal value max1 in the wavelet energy entropy that three-phase current is corresponding and intermediate value max2, minimum value min corresponding healthy phases, now R ₂represent the ratio of fault phase and the wavelet energy entropy of healthy phases, this value is larger; If generation three phase short circuit fault, then three-phase is fault phase, R ₂represent the ratio of the wavelet energy entropy of two fault phases, from the known R of theoretical analysis ₁approximate 1.Therefore can divide phase-to phase fault by arranging threshold value beta, namely working as R ₂during >beta, this fault belongs to two-phase phase fault; Work as R ₂during <beta, this fault is three phase short circuit fault.By three-phase current wavelet energy entropy ratio R ₂just phase-to phase fault can be divided into two-phase phase fault and three phase short circuit fault again with threshold value beta.

By the introducing of three-phase current wavelet energy entropy ratio and two threshold values, earth fault can be divided into single-phase grounding fault and line to line fault earth fault, phase-to phase fault is divided into two-phase phase fault and three phase short circuit fault.So far, 10 kinds of short troubles of transmission line of electricity are divided into above-mentioned four class faults.

Step 4.3: carry out Nonlinear Classification according to support vector machine

From step 4.2, the fault type linearly inseparable comprised in 4 class faults is now (except three phase short circuit fault, only a kind of fault, without the need to dividing again), wherein single-phase grounding fault comprises A, B, C single-line to ground fault, two-phase phase fault comprises AB, BC, AC line to line fault, and two-phase short circuit and ground fault comprises AB, BC, AC two-phase grounding fault.In order to determine the particular type of fault, the present invention introduces three support vector machine recognition systems and carries out Nonlinear Classification to each class fault, and the detailed process of classification is as follows:

For single-phase grounding fault, first, the wavelet energy entropy of fault phase and healthy phases is inputted support vector classification as training sample train, set up 3 two classification SVM of corresponding three kinds of single-phase grounding faults, can be denoted by as (SVM1-SVM3).Secondly, these 3 SVM are combined according to Fig. 2 a.Can find out, 3 SVM, after series connection, are turned into the three class SVM recognition systems that has separation 3 kinds of fault type abilities.When carrying out class test, first by test sample book x _iproper vector input No. 1 classifier (SVM1).If its decision function f ₁x () output valve is+1, then confirm as A phase ground short circuit fault; If output valve is-1, jump out No. 1 classifier, the proper vector of input is passed to No. 2 classifiers (SVM2).By that analogy, No. 3 classifiers (SVM3) are delivered to.By calculating the output valve of decision function f (x) that each classifier exports respectively, just can judge this input signal belongs to which kind of type in single-phase grounding fault exactly.In like manner, build two-phase phase fault SVM recognition system, two-phase short circuit and ground fault SVM recognition system, as shown in Fig. 2 b, 2c, its Fault Identification process and single-phase grounding fault similar.

Step 5: parameter training

Trained by the parameter of training set 90 groups of data samples to two modules, make disaggregated model accurately can determine the fault type that transmission line of electricity occurs.Final linear classification module parameter is: alpha=4.0 and beta=11.0; Nonlinear Classification module parameter is as shown in table 2.

Table 2SVM mesh parameter optimizing result

In table 2, parameter C represents punishment parameter, and parameter g represents kernel function width, pre presentation class discrimination, d _eexpress support for the mean distance of vectorial place plane to lineoid.

Step 6: the validity of verification method and accuracy

In order to verify the validity of put forward the methods of the present invention further, record ripple current data when test set can select somewhere that typical fault occurs in 22 months.The test set herein chosen comprises 74 groups of data altogether.By the analysis one by one to data fault in test set, final correct classification 72 groups, mis-classification 2 groups; Accuracy rate of diagnosis is 97.3%, and wrong diagnosis rate is 2.7%, meets Engineering Error rate.So far, whole power system transmission line fault diagnosis and method validation flow process terminate.The partial data of table 3 test set.

This embodiment is only the present invention's preferably embodiment; but protection scope of the present invention is not limited thereto; anyly be familiar with those skilled in the art in the technical scope that the present invention discloses, the change that can expect easily or replacement, all should be encompassed within protection scope of the present invention.Therefore, protection scope of the present invention should be as the criterion with the protection domain of claim.

Table 3 test set partial data

Claims (2)

1., based on a line fault determination methods for wavelet analysis and support vector machine, it is characterized in that, comprising:

Step 1: extract three-phase current signal as training dataset from the historical data base of transmission line of electricity record wave system system;

Step 2: adopt wavelet analysis to calculate wavelet energy entropy corresponding to three-phase current;

Step 3, by whether there is zero-sequence current after judging transmission line malfunction, being two classes by transmission line malfunction Preliminary division, if there is zero-sequence current, is ground short circuit fault, if there is not zero-sequence current, is phase fault; Wherein ground short circuit fault comprises single-line to ground fault and two-phase grounding fault; Phase fault comprises two-phase phase fault and three-phase shortcircuit;

Step 4: find out maximal value max1, intermediate value max2 and the minimum value min in wavelet energy entropy corresponding to three-phase current, introduces three-phase current wavelet energy entropy ratio R ₁and R ₂:

$R_1=\frac{max1}{max2},R_2=\frac{max2}{\min }$

Setting threshold value alpha and beta, works as R ₁this fault of > alpha belongs to single-phase grounding fault; Work as R ₁this fault of≤alpha belongs to two-phase short circuit and ground fault; Work as R ₂this fault of > beta belongs to two-phase phase fault; Work as R ₂this fault of≤beta belongs to three phase short circuit fault;

Step 5: trained as training sample input support vector classification by wavelet energy entropy corresponding for three-phase current, sets up 3 two classifiers be connected successively of corresponding three kinds of single-phase grounding faults, 3 two classifiers be connected successively of three kinds of two-phase phase faults and 3 two classifiers be connected successively of three kinds of two-phase short circuit and ground faults respectively; The decision function of each two classifiers carrys out failure judgement type by calculating the final analysis result that exports;

Step 6, repetition step 1 ~ 5, the parameter of the multi-group data sample concentrated by training data to threshold value alpha and beta and support vector classification is trained and optimizes, the judged result of fault type and historical data are verified, after meeting error expected rate, the actual transmission line of electricity three-phase current signal that will judge is gathered, and performs step 2 ~ 5 to judge actual transmission line malfunction type; If do not meet error expected rate, repeated execution of steps 6.

2. method according to claim 2, it is characterized in that, described step 2 specifically comprises:

During given discrete signal x (k), at moment k and yardstick j rapid conversion, after conversion, obtain high fdrequency component D _j(k) and low frequency component A _j(k); Band information is included in component of signal D _j(k) and A _jin (k), obtain reconstruction in the following manner:

$\left\{\begin{array}{c}{D}_j\left(k\right):\&lsqb;2^{-\left(j+1\right)}{f}_s,2^{-j}{f}_s\&rsqb;\\ A_j\left(k\right):\&lsqb;0,2^{-\left(j+1\right)}{f}_s\&rsqb;\end{array}\right.,(j=1,2,...,m)---\left(1\right)$

Be expressed as by original signal sequence x (k) after wavelet transform:

$x\left(k\right)=D_1\left(k\right)+A_1\left(k\right)=D_1\left(k\right)+D_2\left(k\right)+A_2\left(k\right)=\underset{j=1}{\overset{J}{\&Sigma;}}{D}_j\left(k\right)+A_j\left(k\right)---\left(2\right)$

Wherein, f _sbe discrete signal samples frequency, m represents the decomposition scale to signal at a time, and J is positive integer;

E _jkbe Wavelet Energy Spectrum under moment k and yardstick j, computing method are as follows:

E _jk＝|D _j(k)| ²(3)

Adopt and develop and next wavelet energy entropy computing method on the basis of information entropy, computing formula is as follows

At j yardstick up-sampling sequence k=1,2 ..., the signal energy summation E of N _jfor:

$E_j=\underset{k=1}{\overset{N}{\&Sigma;}}{E}_{jk}---\left(4\right)$

In order to be consistent with the time probability in information entropy, suppose p _jk=E _jk/ E _j, then p _jkthe ratio that wavelet energy under expression k moment j yardstick is shared in all moment wavelet energy sums under j yardstick, defines corresponding wavelet energy entropy W _eE:

$W_{EE}=-\underset{k}{\&Sigma;}{p}_{jk}{\mathrm{logp}}_{jk}---\left(5\right)$

Finally, the wavelet energy entropy summation that each phase current is corresponding is calculated:

$S_a=\underset{j}{\&Sigma;}{W}_{aj},S_b=\underset{j}{\&Sigma;}{W}_{bj},S_c=\underset{j}{\&Sigma;}{W}_{cj}---\left(6\right)$

Wherein, W _aj, W _bj, W _cjrepresent the wavelet energy entropy that calculate of three-phase current by formula (5) respectively, S _a, S _b, S _crepresent the wavelet energy entropy summation of three-phase current respectively;

After determining decomposition scale, the wavelet energy entropy on each yardstick is calculated according to formula (5), this definition reflects its energy distribution in frequency space of certain signal, characterizes the characteristic information of respective signal, and then reach the object of feature information extraction according to the distribution of energy.