CN105279315B - A kind of transformer online monitoring information fusion analysis method based on correlation analysis and mahalanobis distance - Google Patents
A kind of transformer online monitoring information fusion analysis method based on correlation analysis and mahalanobis distance Download PDFInfo
- Publication number
- CN105279315B CN105279315B CN201510631520.7A CN201510631520A CN105279315B CN 105279315 B CN105279315 B CN 105279315B CN 201510631520 A CN201510631520 A CN 201510631520A CN 105279315 B CN105279315 B CN 105279315B
- Authority
- CN
- China
- Prior art keywords
- data
- sample
- mahalanobis distance
- distance
- correlation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Testing And Monitoring For Control Systems (AREA)
Abstract
The present invention provides a kind of transformer online monitoring information fusion analysis method based on correlation analysis and mahalanobis distance, belongs to the Condition Monitoring Technology field of electric system high-tension apparatus.The technical scheme is that the multidimensional information obtained using transformer online monitoring, data normalization and normalized number Data preprocess are carried out first, and the calculating that one group of time window length carries out partial auto-correlation secondly is constituted with the preceding M point of current sampling point to pretreated data.Again, the mahalanobis distance of each sample and distribution center is calculated using original multi-dimensional information characteristics matrix obtained, is extracted distance and is greater than apart from setting valve zsetData exception sample, corresponding original anomaly data sample is found based on exceptional sample, the final data exception that carries out verifies explanation.Through emulating the mahalanobis distance shown using distance center, exception can be not only gone out with qualitative detection, but also operation situation can be perceived with quantificational expression intensity of anomaly convenient for operations staff.
Description
Technical field
The present invention provides a kind of transformer online monitoring information fusion analysis method based on correlation analysis and mahalanobis distance,
Belong to the Condition Monitoring Technology field of electric system high-tension apparatus.
Background technique
Substation equipment on-line monitoring system is widely used in intelligent substation due to that can reflect equipment running status in real time
Primary equipment is intelligent, carries out all-the-way tracking to the operating status of transformer equipment, monitors its healthy shape in actual moving process
State and life situations, under the premise of guaranteeing safety and reliability, reasonable arrangement test maintenance improves the use of power equipment
Rate and service life are the pursuit of electric system all the time.On-line monitoring system is come into being under such demand
's.At present since substation's on-line monitoring equipment data precision is low, the continuity of data is difficult to ensure, results in substation equipment
" sign " information island is serious, interconnects between the information such as unrealized equipment operation, monitoring, shortage is caused to run based on equipment,
Each monitoring unit multidimensional information transverse dimensions and the time reliable comprehensive diagnos of longitudinal direction dimension panoramic information.Due in substation
In on-line monitoring, transformer carries out processing with important as the main equipment in substation, to transformer online monitoring data
Meaning, and in power equipment state monitoring research, Transformer's Condition Monitoring development time longest, using also most extensive, number
According to compared with horn of plenty and perfect.So choosing transformer online monitoring as case study.
The problems of since there are above-mentioned factors, at present for current substation equipment on-line monitoring, to utilize change
Depressor monitors multidimensional information, inherent law present in mined information, a kind of change based on correlation analysis and mahalanobis distance on-line
Depressor monitors information fusion analysis method on-line.The multidimensional information obtained using transformer online monitoring, first progress data are returned
One changes with normalized number Data preprocess, window when secondly being constituted one group to pretreated data with the preceding M point of current sampling point
The calculating of length progress partial auto-correlation.Again, using original multi-dimensional information characteristics matrix obtained calculate each sample with
The mahalanobis distance of distribution center extracts distance and is greater than apart from setting valve zsetData exception sample, based on exceptional sample find with
Corresponding original anomaly data sample, the final data exception that carries out verifies explanation.Through emulating the horse shown using distance center
Family name's distance can not only be gone out exception with qualitative detection, but also can perceive operation with quantificational expression intensity of anomaly convenient for operations staff
Situation.
Summary of the invention
The object of the present invention is to provide a kind of transformer online monitoring information fusion based on correlation analysis and mahalanobis distance
Analysis method, to solve the above problems.
The technical scheme is that：A kind of transformer online monitoring information fusion based on correlation analysis and mahalanobis distance
Analysis method, the multidimensional information obtained using transformer online monitoring, first progress data normalization and standardized data are located in advance
Secondly reason is constituted one group of time window length to pretreated data with the preceding M point of current sampling point and carries out partial auto-correlation
Calculating.Again, the mahalanobis distance that each sample and distribution center are calculated using original multi-dimensional information characteristics matrix obtained, is mentioned
Distance is taken to be greater than apart from setting valve zsetData exception sample, corresponding original anomaly data are found based on exceptional sample
Sample, the final data exception that carries out verify explanation.It, not only can be qualitative through emulating the mahalanobis distance shown using distance center
It detects exception, but also operation situation can be perceived with quantificational expression intensity of anomaly convenient for operations staff.
Specific step is as follows：
The first step, data normalization, the pretreatment such as standardization.Since transformer online monitoring data class is various, unit
Different, there are redundancies for data.Therefore, will have under different sample frequencys the data of not commensurate be normalized and standardization at
Reason.Selected object x (n) and remaining monitoring data data y (n), which is sought its root mean square, is：
In formula (1), (2), N is correlated signal samples points.
It defines digital signal x (n) and y (n) cross-correlation function is as follows：
Wherein, N is correlated signal samples points, and j is time difference, j=0,1 ... between two signals.
Operation is normalized to every group of signal according to (3) formula, obtains correlation coefficient ρxy：
Second step, the time window length constituted to pretreated data with the preceding M point of current sampling point are one group of progress
The calculating of partial auto-correlation.
Third step calculates distance to the obtained related coefficient of second step.Assuming that the feature distribution of obtained data is obeyed
Multiple normal distribution, availability data feature is away from distribution center apart from characterize data intensity of anomaly.Using formula (5) to original spy
Sign matrix calculates the mahalanobis distance of each sample and distribution center, obtains the mahalanobis distance frequency distribution histogram of data sample：
d2(X, G)=(X-u) ' ∑-1(X-u) (5)
In formula (5), G is that m dimension is overall (investigating m index), and mean vector is u=(u1,u2,u3...,um) ', covariance
Battle array is ∑=(σij)。
4th step extracts distance greater than apart from setting valve zsetData exception sample.The data sample obtained using third step
This mahalanobis distance frequency distribution histogram extracts and is greater than setting valve zsetData exception sample.
5th step extracts abnormal primary data sample.According to data exception sample acquired in the 4th step, extract abnormal former
The distributed area of beginning data sample is compared using the extracted abnormal data sample of mahalanobis distance with primary data sample,
It can verify that the validity of method.
The beneficial effects of the invention are as follows：
(1) the method applied in the present invention, it is single effectively to breach existing transformer online monitoring data information, difficult
To realize the problems such as the panorama perception to institute's monitoring device, reducing on-line monitoring judging result, there may be errors.
(2) base mahalanobis distance algorithm according to the present invention gets rid of traditional abnormal judging means, utilizes data characteristics
Distance away from distribution center portrays data exception degree, and distance is remoter, intensity of anomaly is higher.Wherein, mahalanobis distance benefit is singly examined
Consider data dependence, and not by dimension impact, is a kind of the effective of discovery transformer online monitoring data exception potential rule
Means.
Detailed description of the invention
Fig. 1 is 1-6 class monitoring quantity and hydrogen correlation analysis result in embodiment 1.
Fig. 2 is 7-12 class monitoring quantity and hydrogen correlation analysis result in embodiment 1.
Fig. 3 is mahalanobis distance calculated result in embodiment 1.
Fig. 4 is that exceptional sample extracts result in embodiment 1.
Fig. 5 is initial data in embodiment 1.
Specific embodiment
Embodiment 1：Example displaying is carried out for No. 1 main transformer online monitoring data of selection 220kV substation in this example.
The oil dissolved gas monitoring sampling interval of No. 1 main transformer is 24 hours, i.e., daily 1 sampled point.It is 73 days shown in this example
Interior historical data.Data include 12 class online monitoring datas, are respectively：A, B, C three-phase dielectric loss, carbon monoxide, methane, second
Alkane, leakage current, Wei Shui, acetylene, hydrogen, ethylene, total hydrocarbon.
(1) the 12 class online monitoring datas under different sample frequencys, with not commensurate are normalized and are standardized
Place.Since hydrogen is a variety of transformer fault characterization gases, hydrogen is chosen as the object with remaining monitoring quantity correlation analysis.It is right
Hydrogen x (n) and remaining monitoring quantity data y (n) seek its root mean square.After operation is normalized to two groups of signals, correlation is obtained
Coefficient ρxy, as shown in Figure 1, 2.
It (2) is one group of progress part with the time window length that preceding 8 points of current sampling point are constituted to the historical data of acquisition
The calculating of related coefficient, as shown in Figure 1, 2.As shown in Figure 1, related coefficient is in addition to total hydrocarbon between 30-40 sampled point, remaining phase
Relationship number is 0.4 or so, weak correlation；Between 40-70 sampled point, related coefficient is stablized 0.9 or so, strong correlation；80-90 sampling
Between point, related coefficient variation tendency is steady, compared with strong correlation.As shown in Figure 2, related coefficient variation tendency and above situation be substantially
It coincide.
(3) mahalanobis distance of each sample and distribution center is calculated primitive character matrix again.The geneva of data sample away from
From frequency distribution histogram, as shown in figure 3, total sample number 2485, center of a sample's point is 3- it is found that data are in normal distribution
7。
(4) it extracts distance and is greater than zsetSample.It verifies, is taken apart from setting valve z through multi-group datasetIt is 8.78.Sample number
It is 10, probability 0.4%.As shown in figure 4, the exceptional sample of data concentrates on 1745~1945；2051~2128；2457~
2470.Know that distance is remoter, intensity of anomaly is higher.
(5) abnormal primary data sample is extracted；It is greater than 8.78 data exception sample according to distance, extracts abnormal original number
According to；Primary data sample concentrates on 31-37；41~43；63~65.As shown in Figure 5.To illustrate, the horse of distance center is used
Family name's distance can not only be gone out exception with qualitative detection, but also can perceive operation with quantificational expression intensity of anomaly convenient for operations staff
Situation.
Claims (1)
1. a kind of transformer online monitoring information fusion analysis method based on correlation analysis and mahalanobis distance, it is characterised in that：
The multidimensional information obtained using transformer online monitoring, first progress data normalization and normalized number Data preprocess are secondly right
Pretreated data are constituted the calculating that one group of time window length carries out partial auto-correlation with the preceding M point of current sampling point；Again
It is secondary, the mahalanobis distance of each sample and distribution center is calculated using original multi-dimensional information characteristics matrix obtained, and it is big to extract distance
In apart from setting valve zsetData exception sample, corresponding original anomaly data sample is found based on exceptional sample, finally
It carries out data exception and verifies explanation；
The specific steps are：
The first step, data normalization, standardization pretreatment；To there are the data of not commensurate to carry out normalizing under different sample frequencys
Change and standardization, selected object x (n) and remaining monitoring data y (n), which is sought its root mean square, is：
In formula (1), (2), N is correlated signal samples points；
It defines digital signal x (n) and y (n) cross-correlation function is as follows：
Wherein, N is that correlated signal samples are counted, j time difference between two signals, j=0,1 ... seconds；
Operation is normalized to every group of signal according to (3) formula, obtains correlation coefficient ρxy：
Second step, the time window length constituted to pretreated data with the preceding M point of current sampling point are one group of progress part
The calculating of related coefficient；
Third step calculates distance to the obtained related coefficient of second step, it is assumed that the feature distribution of obtained data obeys multidimensional
Normal distribution, availability data feature distribution center apart from characterize data intensity of anomaly, using formula (5) to primitive character matrix
The mahalanobis distance for calculating each sample and distribution center obtains the mahalanobis distance frequency distribution histogram of data sample：
d2(X, G)=(X-u) ' ∑-1(X-u) (5)
In formula (5), G is that m dimension is overall wherein, and m is the number of inspection target, and mean vector is u=(u1,u2,u3...,um) ', association
Variance matrix is ∑=(σij), i is the rower of matrix；
4th step extracts distance greater than apart from setting valve zsetData exception sample, the data sample obtained using third step
Mahalanobis distance frequency distribution histogram extracts and is greater than setting valve zsetData exception sample；
5th step extracts abnormal primary data sample, according to data exception sample acquired in the 4th step, extracts abnormal original number
According to the distributed area of sample, it is compared, can be tested with primary data sample using the extracted abnormal data sample of mahalanobis distance
The validity of card method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510631520.7A CN105279315B (en) | 2015-09-29 | 2015-09-29 | A kind of transformer online monitoring information fusion analysis method based on correlation analysis and mahalanobis distance |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510631520.7A CN105279315B (en) | 2015-09-29 | 2015-09-29 | A kind of transformer online monitoring information fusion analysis method based on correlation analysis and mahalanobis distance |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105279315A CN105279315A (en) | 2016-01-27 |
| CN105279315B true CN105279315B (en) | 2018-11-27 |
Family
ID=55148325
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510631520.7A Active CN105279315B (en) | 2015-09-29 | 2015-09-29 | A kind of transformer online monitoring information fusion analysis method based on correlation analysis and mahalanobis distance |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105279315B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106950446A (en) * | 2017-03-22 | 2017-07-14 | 国网上海市电力公司 | The abnormal quick detection of power network based on PCA and localization method |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107436277B (en) * | 2017-07-12 | 2019-07-09 | 中山大学 | The single index data quality control method differentiated based on similarity distance |
| CN110049000B (en) * | 2019-01-24 | 2021-11-23 | 浙江工商大学 | Internet of things communication method and system for plaintext aggregation and difference security in large and small spaces |
| CN110346666A (en) * | 2019-07-15 | 2019-10-18 | 南京邮电大学盐城大数据研究院有限公司 | A kind of network transformer state analysis method differentiated based on weighted Mahalanobis distance method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2453259A1 (en) * | 2010-11-10 | 2012-05-16 | Fujitsu Ten Limited | Radar device |
| CN102866321A (en) * | 2012-08-13 | 2013-01-09 | 广东电网公司电力科学研究院 | Self-adaptive stealing-leakage prevention diagnosis method |
| CN103198175A (en) * | 2013-03-04 | 2013-07-10 | 辽宁省电力有限公司鞍山供电公司 | Transformer fault diagnosis method based on fuzzy cluster |
| CN103913663A (en) * | 2014-04-21 | 2014-07-09 | 南京航空航天大学 | Online detection method and protection device for direct current system arc faults |
-
2015
- 2015-09-29 CN CN201510631520.7A patent/CN105279315B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2453259A1 (en) * | 2010-11-10 | 2012-05-16 | Fujitsu Ten Limited | Radar device |
| CN102866321A (en) * | 2012-08-13 | 2013-01-09 | 广东电网公司电力科学研究院 | Self-adaptive stealing-leakage prevention diagnosis method |
| CN103198175A (en) * | 2013-03-04 | 2013-07-10 | 辽宁省电力有限公司鞍山供电公司 | Transformer fault diagnosis method based on fuzzy cluster |
| CN103913663A (en) * | 2014-04-21 | 2014-07-09 | 南京航空航天大学 | Online detection method and protection device for direct current system arc faults |
Non-Patent Citations (1)
| Title |
|---|
| Study on state evalution of high voltage switchgear based on sudden-change componet of fuzzy clustering algorithm;Jing Xie等;《Power System Technology(POWERCON 2014)》;20141222;第1238-1244页 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106950446A (en) * | 2017-03-22 | 2017-07-14 | 国网上海市电力公司 | The abnormal quick detection of power network based on PCA and localization method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105279315A (en) | 2016-01-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107220906B (en) | Multiple Time Scales multiplexing electric abnormality analysis method based on electricity consumption acquisition system | |
| CN105279315B (en) | A kind of transformer online monitoring information fusion analysis method based on correlation analysis and mahalanobis distance | |
| US7961112B2 (en) | Continuous condition monitoring of transformers | |
| CN104007336B (en) | A kind of transformer online monitoring information aggregation method based on Internet of Things | |
| CN107589341B (en) | Single-phase grounding online fault positioning method based on distribution automation main station | |
| CN106841949B (en) | Method and device for monitoring stator insulation of three-phase asynchronous alternating current motor on line | |
| Kim et al. | PMU-based event localization technique for wide-area power system | |
| CN102997838A (en) | Transformer winding deformation fault diagnosis method based on frequency sweep short circuit characteristics | |
| CN107132310A (en) | Transformer equipment health status method of discrimination based on gauss hybrid models | |
| CN110133410A (en) | Diagnosis Method of Transformer Faults and system based on Fuzzy C-Means Cluster Algorithm | |
| CN112098889B (en) | Single-phase earth fault positioning method based on neural network and feature matrix | |
| CN107656161A (en) | A kind of diagnostic method and system of natural esters Insulation Oil Transformer internal fault | |
| CN106295240A (en) | A kind of Condition Assessment for Power Transformer method based on fuzzy mathematics | |
| CN111856209A (en) | Power transmission line fault classification method and device | |
| Meier et al. | Power system data management and analysis using synchrophasor data | |
| CN105137324A (en) | Multi-detecting-point fault component positioning method based on simulation classification model | |
| KR101290295B1 (en) | Fault diagnosis method of oil filled transformer using proportion ratio combination of dissolved gases | |
| CN110824299A (en) | Fault line selection method based on two-dimensional plane judgment of zero-sequence current curve cluster | |
| CN113051763B (en) | Ship system-level electronic equipment reliability evaluation method based on data fusion technology | |
| CN114460519A (en) | On-site and terminal fusion management system and method based on power transformer inspection | |
| CN106646138B (en) | Distribution net work earthing fault localization method based on the conversion of more sample frequency wavelet character energy | |
| CN112801135A (en) | Fault line selection method and device for power plant service power system based on characteristic quantity correlation | |
| CN113066540B (en) | Method for preprocessing non-equilibrium fault sample of oil-immersed transformer | |
| Su | A fuzzy logic tool for transformer fault diagnosis | |
| CN114838923A (en) | Fault diagnosis model establishing method and fault diagnosis method for on-load tap-changer |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |