CN104951763B - The subsynchronous methods of risk assessment of generating set based on recording big data abnormality detection - Google Patents
The subsynchronous methods of risk assessment of generating set based on recording big data abnormality detection Download PDFInfo
- Publication number
- CN104951763B CN104951763B CN201510332872.2A CN201510332872A CN104951763B CN 104951763 B CN104951763 B CN 104951763B CN 201510332872 A CN201510332872 A CN 201510332872A CN 104951763 B CN104951763 B CN 104951763B
- Authority
- CN
- China
- Prior art keywords
- subsynchronous
- feature
- sample
- eigenmatrix
- dimensionality reduction
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A kind of subsynchronous methods of risk assessment of generating set based on recording big data abnormality detection.Including the following contents：By collecting same unit all torsional oscillation recorded wave files interior for a period of time, Fourier transformation is carried out, and extract power frequency component, subsynchronous and supersynchronous harmonic component coefficient to generator terminal three-phase current signal therein, construction initial characteristic data collection Feature；Initial characteristic data collection is carried out by decorrelation and data cleansing by Principal Component Analysis, obtains dimensionality reduction characteristic data set Feature_comp；Based on obtained dimensionality reduction characteristic data set, the mahalanobis distance distribution between each sample and statistical nature is established, subsynchronous risk size is weighed with the size of mahalanobis distance, the system alarm when value-at-risk reaches 90% or more.The quick security evaluation for realizing SSO risks of the invention provides aid decision means for SSO defence and inhibition in operation of power networks.
Description
Technical field
The invention belongs to power system stability and control technical fields, and in particular to one kind is examined extremely based on recording big data
The subsynchronous methods of risk assessment of generating set of survey, for power system stability and safe operation provides aid decision and safety is pre-
It is alert.
Technical background
Under the background of long-distance and large-capacity power transmission demand, due to largely mending transmission of electricity, alternating current-direct current mixing transmission of electricity using string
And FACTS equipment so that the safety and stability of sub-synchronous oscillation (SSO) serious threat for thus causing power grid and generating set fortune
Row.At present, in terms of laying particular emphasis on torsional oscillation of the unit protection and vibrational control to SSO researchs, oscillation defence is lacked to grid operating monitoring
Means.Therefore, subsynchronous risk assessment and its preventive control technology in machine network operation is furtherd investigate to be of great significance.
At present, shaft system of unit delivering polarization monitoring/control/protective device produces a large amount of recorded wave files, has recorded power grid in real time
Shaft system of unit tach signal and set end voltage/current signal in operation, however, the processing mode to recorded wave file is often artificial
, manually, targetedly analyze after failure, therefore, a large amount of history recorded wave file is accumulated all the year round, in addition to being used for going through
History is put on record outside, and there is no the knowledge that formation can help operations staff to understand operation of power networks state.In addition, about SSO analysis mainly from
Highly sensitive shafting tach signal sets out, and is handled by some cumbersome algorithms and technology (such as filtering, Damping calculating),
And electrical quantity signal is only used as used in assistant analysis, utilization rate very little.If the mass data that these can be made full use of to accumulate
The abundant information that recorded wave file contains behind is excavated, realizes system SSO Fast Identifications and security evaluation early warning, it will carry significantly
The intelligent level of elevator row.
Invention content
The purpose of the present invention is providing a kind of subsynchronous risk identification model by a large amount of torsional oscillation recorded wave files, so as to quick
It realizes the security evaluation of SSO risks, aid decision means is provided for SSO defence and inhibition in operation of power networks.
A kind of subsynchronous methods of risk assessment of generating set based on recording big data abnormality detection disclosed in this invention,
The current signal in a large amount of history recorded wave files of torsional oscillation equipment series accumulation is relied on, is compressed by feature extraction, dimensionality reduction, model
Establishing realizes the rapid evaluation and safe early warning of SSO risks with links such as risk identifications.The present invention specifically uses following technology
Scheme：
A kind of subsynchronous methods of risk assessment of generating set based on recording big data abnormality detection, which is characterized in that institute
The method of stating includes the following steps：
(1) the torsional oscillation recorded wave file accumulated in same generating set operation is collected, by therefrom extracting generator terminal three-phase current
Signal ia, ib, ic, establish the database that dimension is N × 3, and N is the number of recorded wave file；
(2) for the corresponding each phase current signal of object, that is, each recorded wave file each in database, pass through Fourier
Transformation extraction pattern feature, the eigenmatrix Feature, P that construction dimension is N × P refer to for the corresponding feature of each recorded wave file
Mark number；
(3) eigenmatrix of acquisition is projected to K mutually orthogonal reference axis, K<P, it is N × K's to form dimension
Eigenmatrix Feature_comp, wherein K mutually orthogonal coordinate directions of dimensionality reduction are the covariance matrix features by Feature
Decomposition result determines；
(4) based on above-mentioned dimensionality reduction eigenmatrix, each row vector is regarded as a sample, is extracted repeatedly by sample, parameter
Reliable estimation and mahalanobis distance statistics establish the subsynchronous risk identification model based on data-driven；
(5) for the new recorded wave file acquired in real time, repeated characteristic extraction, dimensionality reduction compression and apart from calculating, and according to building
The subsynchronous risk assessment of unit under the vertical current recording method of operation of subsynchronous risk identification model realization.
The present invention still further comprises following preferred embodiment：
In step (2), the construction of the eigenmatrix Feature specifically includes：
2.1 by the corresponding three objects, that is, three-phase current signal i of each recorded wave file in the dimensional database of N × 3a、ib、icPoint
Discrete Fourier transform is not carried out, obtains the transform sequence I of three-phase currenta、Ib、Ic；
2.2 extract power frequency, the corresponding amplitude of subsynchronous and supersynchronous frequency of oscillation as pattern from each transform sequence
Characteristic quantity；
When the torsion frequency of certain generating set is respectively f1, f2..., fm, m is torsional oscillation mode exponent number, then from each transformation sequence
Extraction and frequency vector [50-f in rowm..., 50-f2, 50-f1, 50,50+f1, 50+f2..., 50+fm] corresponding amplitude vector,
Obtain the corresponding feature vector of each recorded wave file：
Featurej=[Aj,1 … Aj,2m+1 Bj,1 … Bj,2m+1 Cj,1 … Cj,2m+1] (1)
In formula, j=1,2 ... N, subscript j represent j-th of recorded wave file, and A, B, C are represented respectively and Ia、Ib、IcCorresponding mould
Formula feature, each object corresponds to 2m+1 characteristic index in database；
The corresponding feature vector of 2.3 all recorded wave files of synthesis, forms the eigenmatrix Feature that dimension is N × P,
Middle P=3 × (2m+1)；
In step (3), the specific implementation of dimensionality reduction eigenmatrix is obtained by rectangular projection and is included：
3.1 standardized feature matrix F eature, then calculate its covariance matrix Cov；
3.2 couples of covariance matrix Cov carry out Eigenvalues Decompositions, by characteristic value according to being ranked sequentially from big to small, if λ1
≥λ2≥…≥λP, corresponding normal orthogonal feature vector is denoted as γ respectively1, γ2..., γP；
3.3 seek the condition of satisfactionMinimum K values, by the eigenmatrix after standardization to feature vector
Subspace [γ1,γ2,…,γK] projection, dimensionality reduction eigenmatrix Feaure_comp is obtained, while store the i.e. aforementioned spy of substrate
Levy vector subspace [γ1,γ2,…,γK]。
In step (4), the subsynchronous risk identification model foundation process based on dimensionality reduction eigenmatrix specifically includes：
4.1 randomly select H sample from dimensionality reduction eigenmatrix Feaure_comp, and wherein N/2≤H≤3N/4 calculates it
Sample average T1With covariance matrix S1；
4.2 according to sample average T1With covariance matrix S1, the mahalanobis distance d of all N number of samples of acquisitionj：
Wherein, j=1,2 ... N, Feature_compjJth row vector for dimensionality reduction eigenmatrix.
4.3 select H sample of corresponding mahalanobis distance minimum from dimensionality reduction eigenmatrix, calculate its sample average T2And association
Variance matrix S2, when meeting det (S2)=det (S1) or det (S2During)=0, wherein, det represents determinant operation, by T1And S1
The reliable estimation of T and variance S it is expected respectively as dimensionality reduction eigenmatrix overall distribution；Otherwise, based on T2And S2It recalculates all
The mahalanobis distance d of samplej, and H sample of corresponding mahalanobis distance minimum is selected, calculate its sample average T3And covariance matrix
S3... so repeatedly, until det (Sn+1)=det (Sn) or det (Sn+1Stop iteration during)=0, and by TnAnd SnRespectively as
Dimensionality reduction eigenmatrix overall distribution it is expected that the reliable estimation of T and variance S is stored；
4.4 are obeyed freely based on the reliable estimator of expectation T and variance S stored in step 4.3, sample mahalanobis distance d
The chi square distribution for K is spent, when the mahalanobis distance d of sample meetsWhen be considered as exceptional sample and there are subsynchronous wind
Danger, α is significance.
In step (5), subsynchronous risk assessment processes specifically include under the current method of operation：
The 5.1 new recorded wave file for acquiring in real time, by carrying out direct computation of DFT to generator terminal three-phase current signal therein
Leaf transformation, extraction and power frequency, subsynchronous and supersynchronous corresponding pattern character vector U；
5.2 substrate [the γ for storing U into step (3)1,γ2,…,γK] projection, it obtains compressing vectorial V；
5.3 based on the mean value T and covariance S stored in step (4), calculate the mahalanobis distance D of compression vector V, when P (x≤
When D) >=95%, it is believed that unit is there are subsynchronous risk under the current method of operation, and the value more Risks are bigger, and wherein x is certainly
By spending the chi square distribution stochastic variable value for K.
The present invention has following advantageous effects：The present invention is a kind of Risk Identification Method of data-driven, is overcome
The shortcomings that modeling of conventional method is complicated, Study first difficulty determines and is only used for ex-post analysis.This method is accumulated based on unit
History recorder data in electrical quantity signal, can torsional oscillation risk evaluation model be established by the analysis to data.It is applying
In, according to real-time recording combination risk evaluation model, it can quickly and effectively identify the subsynchronous wind under the current method of operation of unit
Dangerous situation condition grasps generating set health status in time, to take accident prevention in time, has very important reality
Value.
Description of the drawings
Fig. 1 is the subsynchronous methods of risk assessment flow chart of generating set disclosed in the present application；Fig. 2 is the dress that the present invention realizes
Put recorder data access block diagram (by taking 3 mode as an example)；
Fig. 3 is the eigenmatrix dimensionality reduction flow chart that the present invention realizes；
Fig. 4 is the subsynchronous risk identification model foundation flow chart that the present invention realizes；
Fig. 5 is that (it is that the selection reference of dimensionality reduction eigenmatrix dimension is bent wherein, to scheme (a) to the specific example effect of the invention realized
The SSO risk assessment citing that line, figure (b) are mahalanobis distance distribution map, figure (c) is part sample).
Specific embodiment
Technical scheme of the present invention is further elaborated with embodiment with reference to the accompanying drawings of the specification.
It is as shown in Figure 1 the subsynchronous methods of risk assessment flow chart of generating set disclosed in the present application, of the invention is specific
Realize that technical solution is as follows：
Step 1：Recorded wave file is collected, establishes database
The present embodiment is combined in a certain thermal power generation unit operational process of certain domestic power plant, CSC-812 Turbo-generator Sets
The recorded wave file of shafting torsional oscillation protecting equipment record illustrates.As shown in Fig. 2, device is superfluous by being arranged in shaft system of unit head
Remaining speed probe arranges PT/CT at tail, is that the SSO caused by a variety of causes is studied and provided analysis data, wherein, rotating speed
Signal sampling rate is 1kHz, and voltage/current signals sample rate is 600Hz.Device includes two kinds of recording modes, and one kind is that device opens
Dynamic recording, one kind are manually recordings.
The device a certain period has recorded 2917 recorded wave files altogether, starts recorded wave file including 2869 devices
With 48 manually recorded wave files.By extracting generator terminal three-phase current signal i from each recorded wave filea、ib、ic, establish dimension
For 2917 × 3 database：
| Recorded wave file name | Electric current ia | Electric current ib | Electric current ic |
| 20131016023828TSRLB-0.dat | |||
| 20131228135153TSRLB-0.dat | |||
| HMI20140413174313.dat | |||
| HMI20140412145423.dat | |||
| …… |
Step 2：Extract feature, construction feature matrix
Shown according to fig. 2, the present embodiment corresponds to shaft system of unit and includes high intermediate pressure cylinder, low pressure (LP) cylinder A, low pressure (LP) cylinder B and generator 4
A lumped mass block corresponds to 3 shafting torsional oscillation modes (frequency is respectively 12.5Hz, 21.8Hz and 25.5Hz).According to electromechanics
Torsional oscillation interaction mechanism, there are corresponding subsynchronous and supersynchronous component, this realization feature extraction tools in stator current
Body step is as follows：
1. by taking j-th of recorded wave file in above-mentioned database as an example, to its three-phase current ia、ib、icDirect computation of DFT is carried out respectively
Leaf transformation obtains transform sequence Ia、Ib、Ic；
2. from Ia、Ib、IcMiddle extraction respectively and frequency vector [24.5,28.2,37.5,50,62.5,71.8,75.5] are corresponding
Amplitude vector, merge all extraction of values and form the corresponding feature vector Feature of the recorded wave filej
Featurej=[Aj,1 … Aj,7 Bj,1 … Bj,7 Cj,1 … Cj,7] in (4) formula, j=1,2 ..., 2917,
A, B, C are represented and i respectivelya、ib、icCorresponding pattern feature；
3. the corresponding feature vector of 2917 recorded wave files of whole in integrated database, construction feature matrix F eature,
Dimension is 2917 × 21.
Step 3：Dimensionality reduction eigenmatrix is obtained, stores major bases
Realize that the basic principle of data compression is as follows by orthogonal transformation：Equipped with N number of sample, each sample has P surveys
Index is tried, obtains raw data matrix
For one group of Complete Orthogonal base { ω of P dimension spaces1, ω2..., ωPMeet
Then raw data set X is in P orthogonal dimension bases { ω1, ω2..., ωPUnder be projected as
The widely used variance of statistics or standard deviation represent uncertain, and variance or standard deviation are bigger, and uncertainty is bigger,
Information content is bigger.Therefore, raw data set X is along a direction ωiVariance after projection is represented by
In formula,Represent expectation and the covariance matrix of raw data set respectively with S
In order to find direction vector ωiSo that formula (9) maximizes, Lagrange multiplier λ is introducedi, construct unconstrained optimization and ask
Topic
The optimization problem obtain extreme point condition be
Sωi=λiωi (12)
At this point, raw data set is along ωiVariance (i.e. information content) after projection is
Vari=λi (13)
Therefore, in order to enable the information content of data set is larger after projection, projecting direction vector should take more larger with covariance
The corresponding feature vector of characteristic value.In addition, it is contemplated that initial data concentrates the measurement unit and dimension impact of P index, dropping
Raw data matrix is usually made into standardization before dimension.
Finally, to provide the Data Dimensionality Reduction flow based on orthogonal transformation with reference to Fig. 3 as follows：
Then 1. standardized feature matrix F eature calculates its covariance matrix Cov；
2. Eigenvalues Decomposition is carried out to covariance matrix Cov, by characteristic value according to being ranked sequentially from big to small, if λ1≥
λ2≥…≥λP, corresponding normal orthogonal feature vector is denoted as γ respectively1, γ2..., γP；
3. seek the condition of satisfactionMinimum K values, by the eigenmatrix after standardization to feature vector
Subspace [γ1,γ2,…,γK] projection, dimensionality reduction eigenmatrix Feaure_comp is obtained, while store substrate [γ1,γ2,…,
γK]。
Based on 2917 × 21 dimensional feature matrix F eature that this realization step 2 obtains, Fig. 5 (a) give preceding k (k=1,
2 ..., 21) tie up the cumulative information amount of data for projection and the ratio curve figure of gross information content.(K=5) is tieed up 5 before result is shown in figure
The information content of data for projection is up to more than 95%, therefore, by Feature to [γ1,γ2,…,γ5] projection obtain dimension be 2917
× 5 dimensionality reduction eigenmatrix Feature_comp.
Step 4：Establish abnormality detection model, storage and distribution center and variance
By sample drawn repeatedly and the reliable METHOD FOR ESTIMATING POPULATION DISTRIBUTION parameter of interative computation, established according to mahalanobis distance distribution different
The basic principle of normal detection model is as follows：For the sample matrix Z of a dimension observation containing K
Assuming that sample obeys K dimension normal distributions, enableAnd | H1|=H, If det (S1) ≠ 0 calculates mahalanobis distance
Wherein i=1,2 ..., N.Enable H now2For H subsample of mahalanobis distance minimum in the first step, i.e.,
According still further to H2The mean value T calculated2With covariance S2It repeats the above process, each iteration can lead to det (S)
Minimization.Due to the number H of subset be it is limited, certainly exist n so that operation restrains, i.e. det (Sn+1)=det
(Sn), it is believed that SnAnd TnThe reliable estimation of normal distribution center T and variance S are tieed up for K.Since K ties up stochastic variable under normal distribution
Mahalanobis distance d approximations obey the chi square distribution that degree of freedom is K, therefore, the mahalanobis distance of each sample point to distribution center
Corresponding card side's quantileSo thatWherein α is the significance of corresponding sample.If
α very littles, such as less than 0.05 it may be considered that the sample peels off, are considered as abnormal point, and may infer that the sample point is due to structure system
A certain special reason for fluctuation causes in system.
Finally, the dimensionality reduction eigenmatrix Feature_comp obtained based on step 3, abnormality detection model is provided with reference to Fig. 4
The specific steps of foundation：
1. randomly selecting H sample from dimensionality reduction eigenmatrix Feaure_comp, wherein N/2≤H≤3N/4 calculates it
Sample average T1With covariance matrix S1；
2. according to sample average T1With covariance matrix S1, the mahalanobis distance d of all N number of samples of acquisitionj：
Wherein, j=1,2 ... N.
3. selecting H sample of corresponding mahalanobis distance minimum from dimensionality reduction eigenmatrix, its sample average T is calculated2And association
Variance matrix S2, when meeting det (S2)=det (S1) or det (S2During)=0, by T1And S1It is total respectively as dimensionality reduction eigenmatrix
The reliable estimation of T and variance S it is expected in body distribution.Otherwise, based on T2And S2Recalculate the mahalanobis distance d of all samplesj, and select
H sample of corresponding mahalanobis distance minimum is selected, calculates its sample average T3With covariance matrix S3... so repeatedly, until det
(Sn+1)=det (Sn) or det (Sn+1Stop iteration during)=0, and by TnAnd SnRespectively as the dimensionality reduction eigenmatrix overall distribution phase
Hope that the reliable estimation of T and variance S is stored；
4. based on step 3. middle storage expectation T and variance S reliable estimator, sample mahalanobis distance d obey degree of freedom
For the chi square distribution of K, work as satisfactionWhen be considered as exceptional sample and there are subsynchronous risk, α is significance.
The dimensionality reduction eigenmatrix Feature_comp for being 2917 × 5 for the dimension that the present embodiment step 3 obtains, therefrom instead
Extract that 2187 samples reliably estimate mean value T that dimension is 1 × 5 through interative computation and dimension is 5 × 5 variance S, and deposit again
Store up the two estimators.
T=[- 0.7568 0.1210 0.2137 0.0805-0.0507]
Fig. 5 (b) gives the mahalanobis distance distribution map of all 2917 samples of the present embodiment.
Step 5：SSO risk assessment, for the new recorded wave file acquired in real time, repeated characteristic extraction, dimensionality reduction compression and away from
It is commented from calculating, and according to the subsynchronous risk of unit under the current recording method of operation of subsynchronous risk identification model realization of foundation
Estimate.
For the present embodiment, mahalanobis distance obeys the chi square distribution that degree of freedom is 5, compares chi-square distribution table, works as conspicuousness
Corresponding mahalanobis distance is 11.07 when level is 0.05, therefore, safety report is provided when new samples mahalanobis distance is more than 11.07
Alert and aid decision.Fig. 5 (c) gives the SSO risk evaluation results of part sample.
Above example is only used for the core concept for helping to understand the present invention, it is impossible to the present invention is limited with this, for ability
The technical staff in domain, every thought according to the present invention, any change done in specific embodiments and applications,
It should be included within protection scope of the present invention.
Claims (5)
1. a kind of subsynchronous methods of risk assessment of generating set based on recording big data abnormality detection, which is characterized in that described
Method includes the following steps：
(1) the torsional oscillation recorded wave file accumulated in same generating set operation is collected, by therefrom extracting generator terminal three-phase current signal
Ia, ib, ic, establish the database that dimension is N × 3, and N is the number of recorded wave file；
(2) for object, that is, each phase current signal of the corresponding generator terminal of each recorded wave file each in database, pass through Fourier
Transformation extraction pattern feature, the eigenmatrix Feature, P that construction dimension is N × P refer to for the corresponding feature of each recorded wave file
Mark number；
(3) eigenmatrix of acquisition is projected to K mutually orthogonal reference axis, K<P forms the dimensionality reduction that dimension is N × K
Eigenmatrix Feature_comp, wherein K mutually orthogonal coordinate directions are the covariance matrix feature decompositions by Feature
As a result it determines；
(4) based on above-mentioned dimensionality reduction eigenmatrix, each row vector is regarded as a sample, is extracted repeatedly by sample, reliable in parameters
Estimation and mahalanobis distance statistics establish the subsynchronous risk identification model based on data-driven；
(5) for the new recorded wave file acquired in real time, repeated characteristic extraction, dimensionality reduction compression and apart from calculating, and according to foundation
The subsynchronous risk assessment of unit under the subsynchronous current recording method of operation of risk identification model realization.
2. a kind of subsynchronous risk assessment side of generating set based on recording big data abnormality detection according to claim 1
Method, which is characterized in that in step (2), the construction of the eigenmatrix Feature specifically includes：
2.1 by the corresponding three objects, that is, three-phase current signal i of each recorded wave file in the dimensional database of N × 3a、ib、icRespectively into
Row discrete Fourier transform obtains the transform sequence I of three-phase currenta、Ib、Ic；
2.2 extract power frequency, the corresponding amplitude of subsynchronous and supersynchronous frequency of oscillation as pattern feature from each transform sequence
Amount；
When the torsion frequency of certain generating set is respectively f1, f2..., fm, m is torsional oscillation mode exponent number, then from each transform sequence
Extraction and frequency vector [50-fm..., 50-f2, 50-f1, 50,50+f1, 50+f2..., 50+fm] corresponding amplitude vector, it obtains
The corresponding feature vector of each recorded wave file：
Featurej=[Aj,1 … Aj,2m+1 Bj,1 … Bj,2m+1 Cj,1 … Cj,2m+1] (1)
In formula, j=1,2 ... N, subscript j represent j-th of recorded wave file, and A, B, C are represented respectively and Ia、Ib、IcCorresponding pattern is special
Sign；
The corresponding feature vector of 2.3 all recorded wave files of synthesis, forms eigenmatrix Feature, wherein P=that dimension is N × P
3×(2m+1)；
3. a kind of subsynchronous risk assessment side of generating set based on recording big data abnormality detection according to claim 1
Method, which is characterized in that in step (3), the specific implementation of dimensionality reduction eigenmatrix is obtained by rectangular projection and is included：
3.1 standardized feature matrix F eature, then calculate its covariance matrix Cov；
3.2 Eigenvalues Decomposition is carried out to covariance matrix Cov, by characteristic value according to being ranked sequentially from big to small, if λ1≥λ2
≥…≥λP, corresponding normal orthogonal feature vector is denoted as γ respectively1, γ2..., γP；
3.3 seek the condition of satisfactionMinimum K values, the eigenmatrix after standardization is empty to feature vector
Between [γ1,γ2,…,γK] projection, obtain dimensionality reduction eigenmatrix Feaure_comp, while store the i.e. aforementioned feature of substrate to
Vector subspace [γ1,γ2,…,γK]。
4. a kind of subsynchronous risk assessment side of generating set based on recording big data abnormality detection according to claim 1
Method, which is characterized in that in step (4), the subsynchronous risk identification model foundation process based on dimensionality reduction eigenmatrix is specifically wrapped
It includes：
4.1 randomly select H sample from dimensionality reduction eigenmatrix Feaure_comp, and wherein N/2≤H≤3N/4 calculates its sample
Mean value T1With covariance matrix S1；
4.2 according to sample average T1With covariance matrix S1, the mahalanobis distance d of all N number of samples of acquisitionj：
Wherein, Feaure_compjRepresent the jth row vector of dimensionality reduction eigenmatrix, j=1,2 ... N；
4.3 select H sample of corresponding mahalanobis distance minimum from dimensionality reduction eigenmatrix, calculate its sample average T2And covariance
Matrix S2, when meeting det (S2)=det (S1) or det (S2During)=0, wherein, det represents determinant operation, by T1And S1Respectively
The reliable estimation of T and variance S it is expected as dimensionality reduction eigenmatrix overall distribution；Otherwise, based on T2And S2It recalculates all N number of
The mahalanobis distance d of samplej, and H sample of corresponding mahalanobis distance minimum is selected, calculate its sample average T3And covariance matrix
S3... so repeatedly, until det (Sn+1)=det (Sn) or det (Sn+1Stop iteration during)=0, and by TnAnd SnRespectively as
Dimensionality reduction eigenmatrix overall distribution it is expected that the reliable estimation of T and variance S is stored；
4.4 based on the reliable estimator of expectation T and variance S stored in step 4.3, and it is K that sample mahalanobis distance d, which obeys degree of freedom,
That is the chi square distribution of dimensionality reduction eigenmatrix columns, when the mahalanobis distance d of sample meetsWhen be considered as exceptional sample and
There are subsynchronous risk, α is significance.
5. a kind of subsynchronous risk assessment side of generating set based on recording big data abnormality detection according to claim 4
Method, which is characterized in that in step (5), subsynchronous risk assessment processes specifically include under the current recording method of operation：
The 5.1 new recorded wave file for acquiring in real time, by carrying out discrete fourier change to generator terminal three-phase current signal therein
It changes, extraction and power frequency, subsynchronous and supersynchronous corresponding pattern character vector U；
5.2 substrate [the γ for storing U into step (3)1,γ2,…,γK] projection, it obtains compressing vectorial V；
5.3 it is expected T and variance S based on the overall distribution stored in step (4), the mahalanobis distance D of compression vector V are calculated, as P (x
≤ D) >=95% when, it is believed that unit is there are subsynchronous risk under the current recording method of operation, and P (x≤D) more Risks are bigger,
Wherein x is the chi square distribution stochastic variable value that degree of freedom is K.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510332872.2A CN104951763B (en) | 2015-06-16 | 2015-06-16 | The subsynchronous methods of risk assessment of generating set based on recording big data abnormality detection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510332872.2A CN104951763B (en) | 2015-06-16 | 2015-06-16 | The subsynchronous methods of risk assessment of generating set based on recording big data abnormality detection |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104951763A CN104951763A (en) | 2015-09-30 |
| CN104951763B true CN104951763B (en) | 2018-06-26 |
Family
ID=54166405
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510332872.2A Active CN104951763B (en) | 2015-06-16 | 2015-06-16 | The subsynchronous methods of risk assessment of generating set based on recording big data abnormality detection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104951763B (en) |
Families Citing this family (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105320126B (en) * | 2015-10-21 | 2018-05-01 | 中国南方电网有限责任公司 | A kind of secondary device hidden danger method for digging based on big data technology |
| CN105717419B (en) * | 2016-04-21 | 2019-03-08 | 江苏省电力公司苏州供电公司 | The Multiple correspondence analysis method that power cable fault traveling is |
| CN106022719A (en) * | 2016-05-16 | 2016-10-12 | 华北电力科学研究院有限责任公司 | Large-scale generator insulation data management system |
| CN108268489B (en) * | 2016-12-30 | 2020-12-01 | 中国移动通信集团湖北有限公司 | Method and device for evaluating data platform |
| CN107086599A (en) * | 2017-04-13 | 2017-08-22 | 林浩博 | The appraisal procedure that converting plant influences on generator shafting in the case of a kind of three-phase equilibrium |
| CN107086588B (en) * | 2017-05-27 | 2024-01-16 | 清华大学 | Subsynchronous/supersynchronous oscillation analysis method and system for power electronic system |
| CN107966611B (en) * | 2017-11-24 | 2019-10-22 | 广东电网有限责任公司电力调度控制中心 | A kind of supersynchronous harmonic detecting method of electric system based on vector matching time |
| CN109241903B (en) * | 2018-08-30 | 2023-08-29 | 平安科技（深圳）有限公司 | Sample data cleaning method, device, computer equipment and storage medium |
| CN109498037B (en) * | 2018-12-21 | 2020-06-16 | 中国科学院自动化研究所 | Brain cognition measurement method based on deep learning extraction features and multiple dimension reduction algorithm |
| CN110362612B (en) * | 2019-07-19 | 2022-02-22 | 中国工商银行股份有限公司 | Abnormal data detection method and device executed by electronic equipment and electronic equipment |
| CN110488121A (en) * | 2019-08-22 | 2019-11-22 | 广东工业大学 | A kind of fault detection method of MMC, system, device and readable storage medium storing program for executing |
| CN110455517B (en) * | 2019-08-22 | 2021-05-18 | 苏州旋械感知信息科技有限公司 | Tower barrel health monitoring method of wind generating set |
| CN112905958B (en) * | 2021-01-27 | 2024-04-19 | 南京国电南自电网自动化有限公司 | Short-time data window telemetry data state identification method and system based on measurement and control device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102109554A (en) * | 2009-12-25 | 2011-06-29 | 北京四方继保自动化股份有限公司 | Adaptive real-time detection method for subsynchronous oscillation mode of power grid |
| CN102435811A (en) * | 2011-09-20 | 2012-05-02 | 南京南瑞继保电气有限公司 | Method for distinguishing subsynchronous modal signal diffusion |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6774614B2 (en) * | 2002-02-27 | 2004-08-10 | General Electric Company | Method and apparatus for detecting current |
-
2015
- 2015-06-16 CN CN201510332872.2A patent/CN104951763B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102109554A (en) * | 2009-12-25 | 2011-06-29 | 北京四方继保自动化股份有限公司 | Adaptive real-time detection method for subsynchronous oscillation mode of power grid |
| CN102435811A (en) * | 2011-09-20 | 2012-05-02 | 南京南瑞继保电气有限公司 | Method for distinguishing subsynchronous modal signal diffusion |
Non-Patent Citations (1)
| Title |
|---|
| 贵广Ⅱ回直流次同步振荡抑制措施的录波数据分析;杨煜 等;《南方电网技术》;20130630;第7卷(第3期);第21-24页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104951763A (en) | 2015-09-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104951763B (en) | The subsynchronous methods of risk assessment of generating set based on recording big data abnormality detection | |
| CN106124935B (en) | Middle and low voltage network Fault Locating Method | |
| CN103713237B (en) | A kind of power system transmission line short trouble diagnostic method | |
| Jiang et al. | A novel dominant mode estimation method for analyzing inter-area oscillation in China southern power grid | |
| US11656263B2 (en) | Effective feature set-based high impedance fault detection | |
| Yuan et al. | Outage detection in partially observable distribution systems using smart meters and generative adversarial networks | |
| CN103455855B (en) | A kind of intermittent electricity stealing prevention detection method based on power information data analysis | |
| CN106443244B (en) | Electric equipment type identification method and system | |
| CN104698343A (en) | Method and system for judging power grid faults based on historical recording data | |
| CN105205531A (en) | Anti-electric-larceny prediction method based on machine learning and apparatus thereof | |
| CN105092917B (en) | A kind of electricity anti-theft system | |
| CN114881808B (en) | Big data-based accurate identification method for electric power larceny and electric power larceny prevention system | |
| CN107328974B (en) | Electricity stealing identification method and device | |
| CN109752629B (en) | Intelligent diagnosis method and system for power grid measurement problems | |
| CN105844846B (en) | A kind of method for safety monitoring based on Ibeacon technologies, apparatus and system | |
| CN105406461B (en) | A kind of load self-adapting dynamic monitoring method of power distribution network power-off event | |
| CN107515339A (en) | A kind of Risk Identification Method and system based on DC current distribution situation | |
| CN105023090A (en) | Power generator unit coherence grouping scheme based on wide area information | |
| CN110489783A (en) | A kind of low-voltage platform area line loss per unit zone of reasonableness estimation method based on QRNN | |
| CN107944721A (en) | A kind of general machine learning method based on data mining, device and system | |
| CN109766912A (en) | A kind of powerline ice-covering appraisal procedure and system based on Kalman filtering and support vector machines | |
| CN109521304A (en) | Description and appraisal procedure and the device of resistance characteristics temporarily drop in a kind of node voltage | |
| CN105406491B (en) | It is a kind of towards subsynchronous resonance or the remote monitoring system of vibration | |
| Bank et al. | Generator trip identification using wide-area measurements and historical data analysis | |
| CN107147143A (en) | A kind of chain off-grid failure early warning models method for building up of blower fan |
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 |