CN106096494A - A kind of traveling wave fault based on information fusion framework and interference recorder data discrimination method - Google Patents

Abstract

The present invention relates to a kind of traveling wave fault based on information fusion framework and interference recorder data discrimination method, belong to Relay Protection Technology in Power System field.First, use wavelet energy entropy, average and variance three to estimate and portray the row ripple recorder data that traveling wave ranging device gets, the eigenmatrix of formation；Secondly, use principal component analysis that eigenmatrix is carried out dimension-reduction treatment；Finally, the method using mahalanobis distance builds criterion, when row ripple recorder data distance d away from population distribution center_Mq≥d_M,setTime, then it is judged as traveling wave fault recorder data；Otherwise, when row ripple recorder data distance d away from population distribution center_Mq<d_M,setTime, then it is judged as row wave interference recorder data.A large amount of measured data analyses show, the method is the most effective.

Description

A kind of traveling wave fault based on information fusion framework and interference recorder data discrimination method

Technical field

The present invention relates to a kind of traveling wave fault based on information fusion framework and interference recorder data discrimination method, belong to electricity Force system technical field of relay protection.

Background technology

Informationalized iterative method in power transmission line, is mounted with substantial amounts of measurement apparatus, real time record electricity in transmission line of electricity Event in net, the record that these measurement apparatus produce, contain abundant information, if being made full use of, must energy Promote operations staff's understandability to electrical network behavior.The purpose of current most of wave recording device is provided to postmortem analysis fault Reason, analysis means is typically also to be manually main, lacks means online, extensive, that automatically analyze.Wave recording device opens mostly The condition of dynamic record ripple is often transfiniting of single index, and in order to furnish abundant evidence when analyzing cause of accident, more It is relatively low, for current traveling wave range unit that limit threshold value is typically all the ratio set, it is common to use Sudden Changing Rate Starting mode In order to ensure Weak fault is had higher sensitivity, enabling gate threshold value often arrange ratio is relatively low, its value is the most only circuit Through transformer secondary side current.Therefore, the consequence caused is, substantial amounts of interference causes device to start frequently, have recorded Substantial amounts of clutter, only have recorded the most important failure wave-recording, causes interference record ripple and failure wave-recording ratio serious unbalance, fault The utilization of record ripple is also just to play a role when necessity is analyzed.For row ripple recorder data, due to the most cumulative The row ripple recorder data file of substantial amounts of different faults type, occupies substantial amounts of hard drive space except putting on record for history, does not has Tangible paired row ripple recorder data feature is effectively extracted, and lacks the description to different faults class catalog wave characteristic rule.

When choosing traveling wave fault recorder data from substantial amounts of row ripple recorder data at present, need to compare the letter of trip protection Breath realizes, and labor workload is big and complicated.If able to row ripple recorder data is analyzed effectively, process and Refine the useful knowledge needed for obtaining, and play a role, then work efficiency will be greatly improved.Therefore, if can be first row Wave interference data and failure wave-recording identification are come and are just particularly important.After row wave interference and fault recorder data differentiate, The data utilizing identification to come can carry out the row ripples such as fault type differentiation, Fault Phase Selection, nature of trouble identification and automatically analyze merit The practical enforcement of energy provides safeguard, and the realization for follow-up work provides basis.

Summary of the invention

The technical problem to be solved in the present invention is to provide a kind of fault based on information fusion framework and interference row ripple record ripple Data identification method, in order to solve the problems referred to above.

The technical scheme is that a kind of traveling wave fault based on information fusion framework and interference recorder data identification side Method, first, uses wavelet energy entropy, average and variance three to estimate and portrays the row ripple record wave number that traveling wave ranging device gets According to, the eigenmatrix of formation；Secondly, use principal component analysis that eigenmatrix is carried out dimension-reduction treatment；Finally, mahalanobis distance is used Method build criterion, when row ripple recorder data distance d away from population distribution center_Mq≥d_M,setTime, then it is judged as traveling wave fault Recorder data；Otherwise, when row ripple recorder data distance d away from population distribution center_Mq<d_M,setTime, then it is judged as that row wave interference is recorded Wave datum.

Concretely comprise the following steps:

The first step, the wavelet energy entropy of calculating row ripple recorder data:

Utilize wavelet transformation that row ripple recorder data is carried out m layer decomposition；Wherein, 1～m is high-frequency wavelet coefficient D, and m+1 is Low-frequency wavelet coefficients A；

First, the ENERGY E in k moment under yardstick j is calculated_jkFor:

E_jk=| | D_j(k)||²J=1,2 ... m (1a)

E_jk=| | A_j(k)||²J=m+1 (1b)

In formula (1), D_jK () is the high-frequency wavelet coefficient in k moment, A under yardstick j_jK () is that under yardstick j, k moment low frequency is little Wave system number, | | | |²For the most squared after signal is asked for absolute value；

Secondly, the signal gross energy E of each layer is calculated_jFor:

$E_j=\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{E}_{jk}---\left(2\right)$

In formula (2),For to signal k=1,2 ... n sues for peace, the length of n signal wavelet coefficient；

Again, being calculated wavelet energy entropy WEE according to Energy distribution is:

$WEE=-\underset{k}{\overset{n}{\mathrm{\&Sigma;}}}\frac{E_{jk}}{E_j}log\left(\frac{E_{jk}}{E_j}\right)---\left(3\right)$

Second step, the wavelet energy average calculating row ripple recorder data and variance:

Calculate the wavelet energy average of every layer according to formula (4), (5) and energy variance be:

${\mathrm{EXP}}_j=\frac{E_j}{length\left(j\right)}---\left(4\right)$

${\mathrm{VAR}}_j=\frac{\underset{k}{\mathrm{\&Sigma;}}(E_{jk}-{\mathrm{EXP}}_j)}{length\left(j\right)-1}---\left(5\right)$

In formula, length (j) represents the length of jth layer wavelet coefficient；

3rd step, dimension-reduction treatment:

According to step one and step 2, obtain characterizing 3 (m+1) × N eigenmatrixes of row ripple record ripple, use principal component analysis Carrying out dimension-reduction treatment, obtain 5 × N eigenmatrix S, wherein, N is the length of row ripple recorder data；

4th step, employing mahalanobis distance structure identical criterion:

First, the eigenmatrix after the dimensionality reduction obtained in the 3rd step is utilized to calculate in the distribution of row ripple recorder data population sample Heart μ=[μ₁,μ₂,……μ₅], and μ_iFor:

${\mathrm{\&mu;}}_i=\frac{\underset{q=1}{\overset{N}{\mathrm{\&Sigma;}}}S(i,q)}{n},i=1,2,......5---\left(6\right)$

In formula (6), (i q) represents the i-th row q column data of eigenmatrix S to S；

Secondly, the mahalanobis distance d of sample data and distribution center is calculated_Mq:

$d_{Mq}=\sqrt{{\left(S\right(i,q)-\mathrm{\&mu;})}^T{\mathrm{\&Sigma;}}^{-1}\left(S\right(i,q)-\mathrm{\&mu;})},q=1,2,......N---\left(7\right)$

Wherein, Σ is population sample covariance matrix, and T is device computing, and-1 is inversion operation；

Identical criterion is:

If d_Mq≥d_M,setTime, then it is judged as fault traveling wave recorder data；

Otherwise, when row ripple recorder data distance d away from population distribution center_Mq<d_M,setTime, then it is judged as that row wave interference is recorded Wave datum.

The invention has the beneficial effects as follows:

(1) present invention uses the row ripple recorder data mahalanobis distance away from population sample distribution center to build criterion, and it is examined Consider the average of population sample and covariance has considered the distribution situation of population sample, in the data identification of great amount of samples more Tool advantage.

(2) information fusion under utilizing wavelet energy entropy, average and variance three to estimate in the present invention is to fault and interference Recorder data is analyzed, and i.e. considers complexity and the intensity of variation of data, it is contemplated that the population distribution feelings of data Condition.

Accompanying drawing explanation

Fig. 1 is that in the embodiment of the present invention 1, main constituent number determines figure；

Fig. 2 is calculated mahalanobis distance rectilinear in the embodiment of the present invention 1；

Fig. 3 is calculated mahalanobis distance frequency histogram in the embodiment of the present invention 1.

Detailed description of the invention

Below in conjunction with the accompanying drawings and detailed description of the invention, the invention will be further described.

A kind of traveling wave fault based on information fusion framework and interference recorder data discrimination method, first, use little wave energy Amount entropy, average and variance three are estimated and are portrayed the row ripple recorder data that traveling wave ranging device gets, the eigenmatrix of formation；Its Secondary, use principal component analysis that eigenmatrix is carried out dimension-reduction treatment；Finally, use the method for mahalanobis distance to build criterion, work as row Ripple recorder data distance d away from population distribution center_Mq≥d_M,setTime, then it is judged as traveling wave fault recorder data；Otherwise, work as row Ripple recorder data distance d away from population distribution center_Mq<d_M,setTime, then it is judged as row wave interference recorder data.

Concretely comprise the following steps:

The first step, the wavelet energy entropy of calculating row ripple recorder data:

Utilize wavelet transformation that row ripple recorder data is carried out m layer decomposition；Wherein, 1～m is high-frequency wavelet coefficient D, and m+1 is Low-frequency wavelet coefficients A；

First, the ENERGY E in k moment under yardstick j is calculated_jkFor:

E_jk=| | D_j(k)||²J=1,2 ... m (1a)

E_jk=| | A_j(k)||²J=m+1 (1b)

In formula (1), D_jK () is the high-frequency wavelet coefficient in k moment, A under yardstick j_jK () is that under yardstick j, k moment low frequency is little Wave system number, | | | |²For the most squared after signal is asked for absolute value；

Secondly, the signal gross energy E of each layer is calculated_jFor:

$E_j=\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{E}_{jk}---\left(2\right)$

In formula (2),For to signal k=1,2 ... n sues for peace, the length of n signal wavelet coefficient；

Again, being calculated wavelet energy entropy WEE according to Energy distribution is:

$WEE=-\underset{k}{\overset{n}{\mathrm{\&Sigma;}}}\frac{E_{jk}}{E_j}log\left(\frac{E_{jk}}{E_j}\right)---\left(3\right)$

Second step, the wavelet energy average calculating row ripple recorder data and variance:

Calculate the wavelet energy average of every layer according to formula (4), (5) and energy variance be:

${\mathrm{EXP}}_j=\frac{E_j}{length\left(j\right)}---\left(4\right)$

${\mathrm{VAR}}_j=\frac{\underset{k}{\mathrm{\&Sigma;}}(E_{jk}-{\mathrm{EXP}}_j)}{length\left(j\right)-1}---\left(5\right)$

In formula, length (j) represents the length of jth layer wavelet coefficient；

3rd step, dimension-reduction treatment:

According to step one and step 2, obtain characterizing 3 (m+1) × N eigenmatrixes of row ripple record ripple, use principal component analysis Carrying out dimension-reduction treatment, obtain 5 × N eigenmatrix S, wherein, N is the length of row ripple recorder data；

4th step, employing mahalanobis distance structure identical criterion:

First, the eigenmatrix after the dimensionality reduction obtained in the 3rd step is utilized to calculate in the distribution of row ripple recorder data population sample Heart μ=[μ₁,μ₂,……μ₅], and μ_iFor:

${\mathrm{\&mu;}}_i=\frac{\underset{q=1}{\overset{N}{\mathrm{\&Sigma;}}}S(i,q)}{n},i=1,2,......5---\left(6\right)$

In formula (6), (i q) represents the i-th row q column data of eigenmatrix S to S；

Secondly, the mahalanobis distance d of sample data and distribution center is calculated_Mq:

$d_{Mq}=\sqrt{{\left(S\right(i,q)-\mathrm{\&mu;})}^T{\mathrm{\&Sigma;}}^{-1}\left(S\right(i,q)-\mathrm{\&mu;})},q=1,2,......N---\left(7\right)$

Wherein, Σ is population sample covariance matrix, and T is device computing, and-1 is inversion operation；

Identical criterion is:

If d_Mq≥d_M,setTime, then it is judged as fault traveling wave recorder data；

Otherwise, when row ripple recorder data distance d away from population distribution center_Mq<d_M,setTime, then it is judged as that row wave interference is recorded Wave datum.

Embodiment 1:

Input 313 row wave interference recorder datas through data prediction and 71 traveling wave fault recorder datas, sample Data are 384 × 16120 matrixes.

According to step one, db4 wavelet transform is used to complete little wave decomposition.Wherein, discrete wavelet transformation is chosen Number of stories m is 7 layers, has obtained coefficient of wavelet decomposition and has been respectively D₁, D₂, D₃, D₄, D₅, D₆, D₇, A₁.Row is calculated according to formula (1)～(3) Wavelet energy entropy under each yardstick of ripple recorder data

According to step 2, use the wavelet energy average under formula (4)～(5) each yardstick and variance, obtain

The eigenmatrix of 24 × 384.

According to step 3, using principal component analytical method dimensionality reduction, extract and obtain front 5 main constituents, its contribution rate is 83.28% as shown in figure 1 and table 1.After principal component analysis processes, form feature 5 × 384 matrix.

Choose traveling wave fault and the distance threshold d of interference recorder data identification_M,set=2.If sample row ripple recorder data with Population distribution centre distance d_Mq>=2, then it is judged as traveling wave fault recorder data；Otherwise, if sample row ripple recorder data is with overall Distribution center's distance d_Mq< 2, then it is judged as row wave interference recorder data.Analysis result is as shown in table 2, traveling wave fault therefore interference record The accuracy of wave datum identification is 92.97%.

Table 1: the contribution rate of accumulative total of front 5 main constituents

Table 2: mahalanobis distance result of calculation is analyzed

Above in association with accompanying drawing, the detailed description of the invention of the present invention is explained in detail, but the present invention is not limited to above-mentioned Embodiment, in the ken that those of ordinary skill in the art are possessed, it is also possible to before without departing from present inventive concept Put that various changes can be made.

Claims (2)

1. a traveling wave fault based on information fusion framework and interference recorder data discrimination method, it is characterised in that: first, adopt Estimate by wavelet energy entropy, average and variance three and portray the row ripple recorder data that traveling wave ranging device gets, the spy of formation Levy matrix；Secondly, use principal component analysis that eigenmatrix is carried out dimension-reduction treatment；Finally, the method using mahalanobis distance builds Criterion, when row ripple recorder data distance d away from population distribution center_Mq≥d_M,setTime, then it is judged as traveling wave fault recorder data； Otherwise, when row ripple recorder data distance d away from population distribution center_Mq<d_M,setTime, then it is judged as row wave interference recorder data.

Traveling wave fault based on information fusion framework the most according to claim 1 and interference recorder data discrimination method, its It is characterised by concretely comprising the following steps:

The first step, the wavelet energy entropy of calculating row ripple recorder data:

Utilize wavelet transformation that row ripple recorder data is carried out m layer decomposition；Wherein, 1～m is high-frequency wavelet coefficient D, and m+1 is low frequency Wavelet coefficient A；

First, the ENERGY E in k moment under yardstick j is calculated_jkFor:

E_jk=| | D_j(k)||²J=1,2 ... m (1a)

E_jk=| | A_j(k)||²J=m+1 (1b)

In formula (1), D_jK () is the high-frequency wavelet coefficient in k moment, A under yardstick j_jK () is k moment low frequency wavelet system under yardstick j Number, | | | |²For the most squared after signal is asked for absolute value；

Secondly, the signal gross energy E of each layer is calculated_jFor:

In formula (2),For to signal k=1,2 ... n sues for peace, the length of n signal wavelet coefficient；

Again, being calculated wavelet energy entropy WEE according to Energy distribution is:

Second step, the wavelet energy average calculating row ripple recorder data and variance:

Calculate the wavelet energy average of every layer according to formula (4), (5) and energy variance be:

In formula, length (j) represents the length of jth layer wavelet coefficient；

3rd step, dimension-reduction treatment:

According to step one and step 2, obtain characterizing 3 (m+1) × N eigenmatrixes of row ripple record ripple, use principal component analysis to carry out Dimension-reduction treatment, obtains 5 × N eigenmatrix S, and wherein, N is the length of row ripple recorder data；

4th step, employing mahalanobis distance structure identical criterion:

First, the eigenmatrix after the dimensionality reduction obtained in the 3rd step is utilized to calculate row ripple recorder data population sample distribution center μ =[μ₁,μ₂,……μ₅], and μ_iFor:

In formula (6), (i q) represents the i-th row q column data of eigenmatrix S to S；

Secondly, the mahalanobis distance d of sample data and distribution center is calculated_Mq:

Wherein, ∑ is population sample covariance matrix, and T is device computing, and-1 is inversion operation；

Identical criterion is:

If d_Mq≥d_M,setTime, then it is judged as fault traveling wave recorder data；

Otherwise, when row ripple recorder data distance d away from population distribution center_Mq<d_M,setTime, then it is judged as that row wave interference records wave number According to.