CN107979086A - Voltage sag reason recognition methods based on EM algorithms and gradient boosted tree - Google Patents
Voltage sag reason recognition methods based on EM algorithms and gradient boosted tree Download PDFInfo
- Publication number
- CN107979086A CN107979086A CN201711119905.0A CN201711119905A CN107979086A CN 107979086 A CN107979086 A CN 107979086A CN 201711119905 A CN201711119905 A CN 201711119905A CN 107979086 A CN107979086 A CN 107979086A
- Authority
- CN
- China
- Prior art keywords
- mrow
- data
- voltage
- temporarily
- voltage dip
- 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.)
- Granted
Links
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
Abstract
The invention discloses a kind of voltage sag reason recognition methods based on EM algorithms and gradient boosted tree, it is intended to solve similar and foreign peoples voltage dip under feature temporarily drops and can not exact classification, the problems such as data characteristics extraction extraction is difficult.The present invention is based on voltage dip Wave data, supplemented by electric network fault, thunder and lightning, meteorological data, the perunit value of electric grid secondary voltage is accurately extracted by EM algorithms, and the feature on Wave data statistical significance is extracted on this basis, gradient is trained by certain training sample data lift tree-model in single-phase and three-phase data waveform, realize voltage sag reason identification, voltage dip caused by voltage dip caused by system side failure has been clearly distinguished and user side large user load start.
Description
Technical field
The present invention relates to a kind of voltage sag reason recognition methods based on EM algorithms and gradient boosted tree, belong to electric power
System power quality field.
Background technology
The event that voltage dip refers to rms voltage rapid decrease and the duration is not grown, its Typical duration are
10ms~1min cycle.Voltage sag conditions can usually be described with voltage temporary decline, duration.Temporary decline
The ratio of voltage effective value and rated value when being defined as temporarily dropping, duration refer to the time that temporarily drop is being undergone from occurring to terminating.
The characteristic quantity of the two characterization voltage dips has considerable influence to load operation.It is temporary that International Electrotechnical Commission (IEC) defines voltage
Range of decrease value is drop to rated value 90%~1%；Institute of Electrical and Electronics Engineers (IEEE) is defined as dropping to volume
The 90%~10% of definite value.
Counted according to actual operating experience, in the complaint of custom power quality problem, 80% above is by voltage dip
It is caused.Why this quality of voltage problem has so big influence, is on the one hand due to the high frequency time hair of voltage dip
It is raw；On the other hand, some sensitive equipment abnormal runnings can also be made even even if the voltage dip for lasting only for about 4~5 cycles
Unordered start and stop, huge economic losses are brought to responsible consumer.
When in electrical power trans mission/distribution system occur short trouble, the startup of large capacity induction machine, lightning stroke, switching manipulation, transformer with
And the switching of capacitor group can cause voltage dip when event.Wherein lightning stroke can cause shaft tower insulation flashover or put over the ground
Electricity can act protective device, so as to cause supply voltage to decline.This temporarily drop coverage is big, and the duration is generally more than
100ms.The startup of large capacity induction machine and high-power impact load can be brought not with voltage dip caused by short trouble
With the temporary drop of feature.Induction machine temporary drawdown degree caused by starting depends on the system short-circuit of induction machine characteristic and connection place
Capacity, amplitude gradually revert to normal value with electric current and rise, and have the characteristics that drop depth is shallow and the duration is long, generally
User will not be caused to seriously affect.And voltage dip amplitude size has with abort situation and type as caused by short trouble
Close, the temporary drop duration is determined by the actuation time protected.Typically voltage falls suddenly, after waiting breaker actuation, electricity
Pressure is recovered immediately.Therefore, short trouble can cause more serious voltage dip, influence the normal work of sensitive equipment.Transformer
When putting into operation, due to core sataration characteristic, it can be produced in sending end and be several times as much as shoving for rated current, its size and transformer
The initial phase angle of sinusoidal voltage and iron core remanent magnetism are related when putting into operation.Shoving for maximum is produced when initial phase angle is 0 °, at this time voltage dip
Degree is also most deep；When phase angle is 90 °, then it will not produce and shove.The initial phase angle of three-phase mutual deviation all the time when being put into operation due to transformer
120 °, therefore, transformer put into operation caused by voltage dip always three-phase imbalance.Coil copper loss causes the recovery of temporarily drop voltage
It is a gradual process, the resistance of miniature transformer is larger, and reactance is smaller, and about several cycles just reach stable state；And large-scale transformation
For device since resistance is smaller, reactance is larger, generally requires tens cycles and can be only achieved stable state.
Voltage dip either caused by voltage dip caused by short trouble or transformer excitation flow, this belongs to
Voltage dip caused by system side, and voltage dip caused by the startup of large capacity induction machine belongs to voltage caused by user side
Temporarily drop, how by both distinguish be voltage dip administer prerequisite.Or the prior art is by substantial amounts of voltage dip waveform
Database is stored in, is matched by way of cumbersome, time-consuming, spends substantial amounts of computing resource and time resource, or by aobvious
The rule of formula come judge, it is necessary to constantly addition rule.In the identification process of voltage sag source, according to voltage effective value and
Duration carry out intuitively feature extraction often have granularity it is big, can not exact classification, data characteristics extraction etc. it is tired
It is difficult.
The content of the invention
To reach above-mentioned purpose, the present invention discloses a kind of based on the knowledge of the voltage sag reason of EM algorithms and gradient boosted tree
Other method, specifically include voltage dip Feature Engineering, temporarily drop event category model training associated with external data, can overcome
Prior art voltage sag reason identification granularity is big, can not the difficulty such as exact classification, data characteristics extraction.
To realize above-mentioned technical purpose and the technique effect, the present invention is achieved through the following technical solutions：
A kind of voltage sag reason recognition methods based on EM algorithms and gradient boosted tree, it is characterised in that including voltage
Temporarily drop Feature Engineering, temporary drop event category model training are associated with external data；Voltage is extracted based on voltage dip Feature Engineering
After the characteristic of temporary drop data, after voltage dip data reasons are classified by temporarily dropping event category model training, with outside
Data correlation.
Voltage dip Feature Engineering extracts characteristic from voltage dip Wave data；
The characteristic specifically includes：Voltage dip is separate, temporary drawdown degree, the voltage dip duration, location
Area, monitoring point voltage class, temporarily drop 10%-180% sections accounting, temporarily drop initial to the most deep period temporarily dropped, most deep temporarily drop to
The period of temporarily drop recovery, temporary drawdown degree, temporarily squareness factor, temporarily drop 20% bottom of concave and head ratio, drop 20% area of concave
Ratio, the waveform degree of bias, waveform set-back and frequency domain cluster 5 groups of accountings.
Temporarily drop event category model training includes the voltage dip Wave data training pattern and EM algorithms of gradient boosted tree
Voltage dip perunit value prediction model, voltage dip Wave data training pattern be used for for sample data carry out parameter instruction
Practice, voltage dip perunit value prediction model is used to identify electric grid secondary voltage perunit value.
Voltage dip Wave data training pattern J (θ) based on gradient boosted tree is expressed as：
Wherein f (xi, θ) be voltage dip Wave data classification discreet value；L(yi,f(xi, θ)) it is loss function, table
Show the similitude between voltage dip waveform classification discreet value and class label, θ is parameter sets；N is sample data number, Ω
(θ) is regular terms, chooses regular terms of the L2 canonicals as the training pattern of voltage dip Wave data, then：
Loss function is used as using logistic regression loss：
L(yi,f(xi, θ))=- ylog (f (x；θ))-(1-y) log (1-f (x, θ)) (6)
In order to simplify computation complexity and lifting calculating speed, loss function is approached with Newton interpolating method, i.e.,：
xiFor training sample, yiFor the corresponding class scalar of training sample, x is training sample set, and y is corresponding for training sample
Category quantity set, xi∈ x, yi∈y；For Interpolation-Radix-Function；
The corresponding category quantity set y of training sample has K class waveforms, by the corresponding category quantity set of training sample labeled as set
D, the ratio shared by middle kth class waveform sample are pk, then the purity Ent (D) of voltage dip waveform data sample set be：
According to the purity formula (10) of sample set, select a certain feature as decision tree root node Det (D, a), root section
(D, selection principle a) is to compare the minimum value for choosing purity ratio under all features to point Det：
Its discreet value f (x are calculated after each feature of iterationi；θ) and yiBetween difference, using difference as next certainly
The target of plan tree；Decision tree under each feature of cycle calculations, until difference is less than or equal to ε, i.e. algorithmic statement, ε is definition
Convergency value.
Voltage caused by voltage dip, large user's load caused by voltage dip Wave data classification specifically includes failure
Temporarily drop and transformer excitation shove caused voltage dip.
In Wave data Normal Distribution X~N (μ；σ2) on the premise of, the voltage dip perunit value based on EM algorithms
Prediction model includes E steps and M steps；E walks the distributed constant that optimal hidden variable Z is inferred to according to training data；M is walked optimal hidden
On the basis of the distributed constant of variable Z, the distributed constant of Z is corrected according to Wave data；Loop iteration E is walked to be walked with M, until most
The parameter convergence of excellent hidden variable Z.
Voltage dip perunit value prediction model based on EM algorithms specifically includes following steps：
S101, calculates the average value e of virtual value under the first two cycle in voltage dip Wave data0Most latter two week
The average value e of virtual value under ripple1, respectively for e0Retain [0.9e0, 1.1e0] in the range of virtual value and e1Retain [0.9e1,
1.1e1] in the range of virtual value, wherein e0And e1The composition component for the optimal hidden variable Z data group estimated in being walked for E；That is Z by
e0And e1Composition.
S102, M step specifically include following steps, are directed to [0.9e respectively0, 1.1e0] and [0.9e1, 1.1e1] in the range of
Effective Value Data carries out Gauss curve fitting X~N (μ under normal distribution0；σ0 2), X~N (μ1；σ1 2), obtain the ginseng within the scope of two
Number μ0And μ1, μ0And μ1For the parameters revision of M Walk；
S103, calculates respectively | μ0-e0|、|μ0-e1|、|μ1-e0| and | μ1-e1|, if any one value is less than or equal to ε,
Algorithm stops calculating, and takes less than or equal to the μ under ε0Or μ1As secondary voltage perunit value, wherein ε (such as takes for parameter error
0.01 or 0.001)；
S104, if | μ0-e0|、|μ0-e1|、|μ1-e0|、|μ1-e1| both greater than ε, then respectively in μ0Or μ1It is lower to retain [0.9
μ0, 1.1 μ0] and [0.9 μ1, 1.1 μ1] in the range of virtual value, carry out normal distribution under Gauss curve fittingObtain parameterWith
S105, repeats claim steps S103 and step S104 until convergence.
External system data correlation combination topological structure of electric is and outer by the data after voltage dip disaggregated model training
The time of portion's data is associated analysis, specifically includes following steps：
S301, voltage dip event category：Voltage dip Wave data training pattern is divided into two parts, and Part I is
The training pattern of Wave data temporarily drops in single-phase voltage, and Part II is the training pattern of three-phase voltage sag Wave data；It is single-phase
All voltage dip Wave datas are regarded training sample, three-phase voltage sag by the training pattern of voltage dip Wave data
The training pattern of Wave data is temporarily dropped by a tuple sample, wherein single-phase voltage of A, B, C three-phase voltage sag Wave data
Data prediction part of the prediction data of waveform training pattern as three-phase voltage sag waveform training pattern；
S302, it is sample data that single-phase voltage is temporarily dropped to Wave data, sample dataFor N band mark
The data of label, divide K classes；Sample data is divided into training sample data and test sample number in the method for K- cross-trainings
According to wherein training sample data are row data, using the characteristic of extraction as column data；
S3O3：One voltage dip event is divided into 3 sample datas according to separate, K classes are picked out according to wave character
Waveform numerical example data are used for model training and verification；
S3O4：The sample size manually chosen is less, in order to make full use of sample, uses K- cross validation methods
The voltage dip Wave data training pattern of training gradient boosted tree, give over to every time verification for the total sample size of voltage dip
1/K, being used for the sample size of training every time accordingly increases, and chooses the model of validation error minimum；
S3O5：Using the voltage dip Wave data training pattern of trained gradient boosted tree in unfiled data
Test, checks the data of classification error, if the sample of homogenous characteristics is less than 2 in training data, is intersected using leaving-one method and tested
Card, chooses the model of error minimum；
S306：By sorted monophasic waveform data, combined by three-phase, become a new classification, new point
Class is associated with electric network fault data, lightning stroke data and meteorological data, and temporally the correlation method of piece realizes voltage dip occurrence cause
Association analysis；
S307：The phase angle of the sorted data of voltage dip is calculated, for voltage dip caused by failure, according to phase angle
Voltage dip is divided into seven class asymmetrical three-phases temporarily to drop, is single-phase short circuit by voltage dip type identification fault occurrence reason
Failure, two-phase short-circuit fault or three phase short circuit fault.
More preferably, temporarily drop 10%-180% sections accounting is：Using perunit value as 100%, 18 areas are divided into up and down
Between, calculate effective Value Data and fall number and the accounting of sum in each section；
Temporarily drop initial is to most deep period for temporarily dropping：Between being temporarily reduced at the beginning of perunit value 90% with it is temporary drop to it is most deep
The time difference tsag1 at place；
Most it is deep temporarily drop to temporarily drop recover period be：Temporarily drop bosom returns to the time difference of perunit value 90% with temporarily drop
tsag2；
Squareness factor is：The temporarily ratio of drop area and temporary decline rectangular area, in order to facilitate reference area, makes perunit
It is 1 to be worth corresponding virtual value, temporarily drops numerical value of the sampling number in region according to corresponding rate conversion between [0 1], temporarily drop
Amplitude rectangular area is SfThe sampling number in region temporarily drops in=perunit value *, and it is S temporarily to drop areaa=temporary decline rectangular area-temporarily
The sampled point summation in region is dropped,
Temporarily drop 20% bottom of concave and head ratio：Temporarily drop most bottom section duration and temporarily drop duration ratio, temporarily drop
20% bottom value range of concave is less than ((perunit value-temporary drawdown degree) the temporary drawdown degree of * 0.2+)；
Temporarily drop 20% area ratio of concave：Temporarily drop most bottom section area accounts for the area ratio in whole temporarily drop region；
Frequency domain clusters 5 groups of accountings：The effective Value Data of waveform switchs to the data of domain space by Fourier transform, utilizes K
Means clustering method gathers data under frequency domain accounts for the ratio of sum for 5 classes, every one kind data.
More preferably, temporarily drop specifically includes seven class asymmetrical three-phases：A phases single-phase earthing fault, B phases single-phase earthing fault, C
Phase single-phase earthing fault, AB two-phase short-circuit faults, BC two-phase short-circuit faults, AC two-phase short-circuit faults and three phase short circuit fault.
Beneficial effects of the present invention include：
The present invention discloses a kind of voltage sag reason recognition methods based on EM algorithms and gradient boosted tree, establishes electricity
Temporary drop data analysis model is pressed, tree algorithm is lifted by gradient, from the angle of Wave data by voltage dip event category, point
Class is accurate；
The present invention establishes Wave data and the closed-Loop Analysis model of external data, passage time piece association analysis method
The external system data such as electric network fault data, lightning stroke data, meteorological data have been merged, have realized voltage dip occurrence cause
Association analysis, improves the utilization rate and accuracy rate of data.
Brief description of the drawings
Fig. 1 is a kind of voltage sag reason recognition methods flowage structure based on EM algorithms and gradient boosted tree of the present invention
Figure；
Fig. 2 is voltage dip perunit value, temporarily temporary drawdown degree, drop duration schematic diagram；
Fig. 3 is voltage dip sampling number 10%-180% sections accounting schematic diagram；
Fig. 4 temporarily drops initial to the most deep period t temporarily droppedsag1, it is most deep temporarily to drop to the period t that temporarily drop is recoveredsag2；
Fig. 5 is voltage dip squareness factor schematic diagram；
Concave bottom and head ratio temporarily drop in Fig. 6；
20% area ratio of concave temporarily drops in Fig. 7；
Schematic diagram after Fig. 8 voltage dip virtual value waveforms and Fourier transformation.
Embodiment
Technical solution of the present invention is described in further detail with specific embodiment below in conjunction with the accompanying drawings, so that this
The technical staff in field can be better understood from the present invention and can be practiced, but illustrated embodiment is not as to the present invention's
Limit.
The present invention provides a kind of pressure equalizing control method of the combined DC/DC converters of ISOS, makes ISOS combined converters each
Power equalization between module, input and output voltage can be divided equally, and shifting is being expanded in the power reflux of single module DC/DC converters
Power reflux is inhibited under phase control.
As shown in Figure 1, a kind of voltage sag reason recognition methods based on EM algorithms and gradient boosted tree, including voltage
Temporarily drop Feature Engineering, temporary drop event category model training are associated with external data；Voltage is extracted based on voltage dip Feature Engineering
After the characteristic of temporary drop data, after voltage dip data reasons are classified by temporarily dropping event category model training, with outside
Data correlation.
Voltage dip Feature Engineering extracts characteristic from voltage dip Wave data；
The characteristic specifically includes：Voltage dip is separate, temporary drawdown degree, the voltage dip duration, location
Area, monitoring point voltage class, temporarily drop 10%-180% sections accounting, temporarily drop initial to the most deep period temporarily dropped, most deep temporarily drop to
The period of temporarily drop recovery, temporary drawdown degree, temporarily squareness factor, temporarily drop 20% bottom of concave and head ratio, drop 20% area of concave
Ratio, the waveform degree of bias, waveform set-back and frequency domain cluster 5 groups of accountings.
The present embodiment, chooses the sample data that learner is used to learn, the temporarily drop event conduct of Qu Mou cities May 671
Sample data, is divided into 3 single-phase sample datas by separate by a voltage dip event, is divided into 2013 sample datas.Will
All sample data virtual value waveforms save as picture, manually choose the obvious sample data of 5 category features, the present embodiment
394 sample datas are chosen altogether is used for learner, learning training gradient lifting tree-model.
Temporary drawdown degree：Difference between being temporarily reduced at the beginning of perunit value 90% between recovery time continues for temporarily drop
Time t, the difference for temporarily dropping bosom to bottom is temporary drawdown degree Usag, as shown in Figure 2；
Voltage dip is separate：A, occur temporarily to drop in B, C three-phase voltage separate.
The voltage dip duration：Continue between being temporarily reduced at the beginning of perunit value 90% with recovery time difference for temporarily drop
Time t.
Location：The location of voltage dip occurs, is identified with substation and districts and cities.
Monitoring point voltage class：The rated voltage rank of electric system and power equipment series.Voltage class is divided into：1、
Safe voltage (usual below 36V)；2nd, low pressure (and dividing 220V and 380V)；3rd, high pressure (10KV-220KV)；4th, super-pressure
330KV-750KV；5th, extra-high voltage 1000KV is exchanged, more than ± 800KV direct currents；
Temporarily drop 10%-180% sections accounting is：Using perunit value as 100%, 18 sections are divided into up and down, calculating has
Valid value data fall the number and the accounting of sum in each section；As shown in Figure 3.
Temporarily drop initial is to most deep period for temporarily dropping：Between being temporarily reduced at the beginning of perunit value 90% with it is temporary drop to it is most deep
The time difference tsag1 at place；As shown in Figure 4.
Most it is deep temporarily drop to temporarily drop recover period be：Temporarily drop bosom returns to the time difference of perunit value 90% with temporarily drop
tsag2；
Squareness factor is：The temporarily ratio of drop area and temporary decline rectangular area, in order to facilitate reference area, makes perunit
It is 1 to be worth corresponding virtual value, temporarily drops numerical value of the sampling number in region according to corresponding rate conversion between [0 1], temporarily drop
Amplitude rectangular area is SfThe sampling number in region temporarily drops in=perunit value *, and it is S temporarily to drop areaa=temporary decline rectangular area-temporarily
The sampled point summation in region is dropped,As shown in Figure 5.
Temporarily drop 20% bottom of concave and head ratio：Temporarily drop most bottom section duration and temporarily drop duration ratio, temporarily drop
20% bottom value range of concave is less than ((perunit value-temporary drawdown degree) the temporary drawdown degree of * 0.2+)；As shown in Figure 6.
Temporarily drop 20% area ratio of concave：Temporarily drop most bottom section area accounts for the area ratio in whole temporarily drop region；
Frequency domain clusters 5 groups of accountings：The effective Value Data of waveform switchs to the data of domain space by Fourier transform, utilizes K
Means clustering method gathers data under frequency domain accounts for the ratio of sum for 5 classes, every one kind data.As shown in Fig. 7.
Section scales：Temporary drop initial is to the most deep period temporarily dropped, the most deep period for temporarily dropping to temporarily drop recovery is contracted using section
Put and zoom to value between [0,1]；
X is initial characteristic value, and x ' is the value after scaling.
Chi-square Test：Or χ2Examine, calculate χ2Value formula is：
Wherein A is actual value, and T is theoretical value.χ2Difference degree for weighing actual value and theoretical value (namely blocks
The core concept just examined), contain following two information：1st, the absolute size of actual value and theoretical value deviation (due to square
Presence, difference is exaggerated) 2, the relative size of difference degree and theoretical value characteristic value is used into Chi-square Test selection K
A best feature learns for learner, and K value optimum valuing ranges are obtained using circulation verification；
Related coefficient：The index of linearly related degree between two stochastic variables of measurement, dependency relation are a kind of non-determined
Property relation, related coefficient is that the amount for studying linearly related degree between variable uses correlation coefficient process, first to calculate each spy
Sign checks that the related coefficient of feature removes the spy that related coefficient is 0 to the related coefficient of desired value and the P values of related coefficient
Sign.Related coefficient represents two groups of data linearly relevant degree (while degree of increase or reduction), measures from another point of view
For point relative to the distribution situation of standard deviation, it does not have unit.The meter of the correlation coefficient r of two groups of data of X, Y comprising n numerical value
Calculation method：
The value of r illustrates that data dependence is stronger closer to positive and negative 1, and closer to 0 explanation, data dependence is smaller for the value of r,
Calculate feature correlation and remove related coefficient close to 0 feature.
Temporarily drop event category model training includes the voltage dip Wave data training pattern and EM algorithms of gradient boosted tree
Voltage.
Temporarily drop perunit value prediction model, voltage dip Wave data training pattern are used to carry out parameter for sample data
Training, voltage dip perunit value prediction model are used to identify electric grid secondary voltage perunit value.
Voltage dip Wave data training pattern J (θ) based on gradient boosted tree is expressed as：
Wherein f (xi, θ) be voltage dip Wave data classification discreet value；L(yi,f(xi, θ)) it is loss function, table
Show the similitude between voltage dip waveform classification discreet value and class label, θ is parameter sets；N is sample data number, Ω
(θ) is regular terms, chooses regular terms of the L2 canonicals as the training pattern of voltage dip Wave data, then：
Loss function is used as using logistic regression loss：
L(yi,f(xi, θ))=- ylog (f (x, θ))-(1-y) log (1-f (x, θ)) (6)
In order to simplify computation complexity and lifting calculating speed, loss function is approached with Newton interpolating method, i.e.,：
xiFor training sample, yiFor the corresponding class scalar of training sample, x is training sample set, and y is corresponding for training sample
Category quantity set, xi∈ x, yi∈y；▽fL (f, y) is Interpolation-Radix-Function；
The corresponding category quantity set y of training sample has K class waveforms, by the corresponding category quantity set of training sample labeled as set
D, the ratio shared by middle kth class waveform sample are pk, then the purity Ent (D) of voltage dip waveform data sample set be：
According to the purity formula (10) of sample set, select a certain feature as decision tree root node Det (D, a), root section
(D, selection principle a) is to compare the minimum value for choosing purity ratio under all features to point Det：
Its discreet value f (x are calculated after each feature of iterationi；θ) and yiBetween difference, using difference as next certainly
The target of plan tree；Decision tree under each feature of cycle calculations, until difference is less than or equal to ε, i.e. algorithmic statement, ε is definition
Convergency value.
Voltage caused by voltage dip, large user's load caused by voltage dip Wave data classification specifically includes failure
Temporarily drop and transformer excitation shove caused voltage dip.
The present embodiment, temporarily drops event category model training and specifically includes following steps：
SO1：The sample size manually chosen is less, in order to make full use of sample, rolls over cross validation method using K-
Training gradient lifting tree-model, give over to every time verification for the 1/10 of total sample size, therefore be used for the sample size phase of training every time
It should add, but K- folding cross validations are required for operation 10 times for each model, last average test result can weigh
The performance of model is measured, chooses the model of validation error minimum.
SO2：Waveform separation is pressed using 671 training datas, 5 two disaggregated models are respectively trained out, 5 models intersect
Verify score, be respectively 0.989,0.979,0.949,0.952,0.995, by remaining 1342 test datas, by model performance
Index is descending to use 5 categories of model successively, and the data for being not belonging to 5 classification are put into other groups of the 6th class.
SO4：By 2013 monophasic waveform data of point good classification, 3 phases are merged, become 34 groups of 3 new phases point
Class, respectively falls in different classifications wherein combining according to three kinds of modes, 1, three monophasic waveform；2nd, two monophasic waveforms fall
In same category, another monophasic waveform falls in different classifications；3rd, three monophasic waveforms fall in same category.
S05：The amplitude and phase angle under three-phase waveform are calculated, will be electric caused by failure according to the phase angle difference between A, B, C
Temporarily drop is divided into seven classes to pressure；
In Wave data Normal Distribution X~N (μ；σ2) on the premise of, the voltage dip perunit value based on EM algorithms
Prediction model includes E steps and M steps；E walks the distributed constant that optimal hidden variable Z is inferred to according to training data；M is walked optimal hidden
On the basis of the distributed constant of variable Z, the distributed constant of Z is corrected according to Wave data；Loop iteration E is walked to be walked with M, until most
The parameter convergence of excellent hidden variable Z.
The voltage dip recorder of monitoring power grid is distributed on the busbar of different voltages grade, and one sub-value is respectively
One sub-value is converted to 57V or so by 500kV, 220kV, 110kV, 35kV, 10kV etc., monitoring and recording instrument often through PT no-load voltage ratios
Two sub-values.However, with reasons such as the magnanimity number of device, measurement errors, the perunit value of monitoring and recording instrument, is distributed in
Between 100V~40V, this judges whether a voltage dip waveform is temporary drop or temporary liter to system, brings great noise.Cause
This, the present invention asks for the secondary voltage perunit value of temporarily drop recorder monitoring with the thought of EM algorithms.
Voltage dip perunit value prediction model based on EM algorithms specifically includes following steps：
S101, as shown in figure 8, calculating the average value e of virtual value under the first two cycle in voltage dip Wave data0With
The average value e of virtual value under most latter two cycle1, respectively for e0Retain [0.9e0, 1.1e0] in the range of virtual value and e1Protect
Stay [0.9e1, 1.1e1] in the range of virtual value, wherein e0And e1The composition for the optimal hidden variable Z data group estimated in being walked for E
Component；I.e. Z is by e0And e1Composition.
S102, M step specifically include following steps, are directed to [0.9e respectively0, 1.1e0] and [0.9e1, 1.1e1] in the range of
Effective Value Data carries out Gauss curve fitting X~N (μ under normal distribution0；σ0 2), X~N (μ1；σ1 2), obtain the ginseng within the scope of two
Number μ0And μ1, μ0And μ1For the parameters revision of M Walk；
S103, calculates respectively | μ0-e0|、|μ0-e1|、|μ1-e0| and | μ1-e1|, if any one value is less than or equal to ε,
Algorithm stops calculating, and takes less than or equal to the μ under ε0Or μ1As secondary voltage perunit value, wherein ε (such as takes for parameter error
0.01 or 0.001)；
S104, if | μ0-e0|、|μ0-e1|、|μ1-e0|、|μ1-e1| both greater than ε, then respectively in μ0Or μ1It is lower to retain [0.9
μ0, 1.1 μ0] and [0.9 μ1, 1.1 μ1] in the range of virtual value, carry out normal distribution under Gauss curve fittingObtain parameterWith
S105, repeats claim steps S103 and step S104 until convergence.
External system data correlation combination topological structure of electric is and outer by the data after voltage dip disaggregated model training
The time of portion's data is associated analysis, specifically includes following steps：
S301, voltage dip event category：Voltage dip Wave data training pattern is divided into two parts, and Part I is
The training pattern of Wave data temporarily drops in single-phase voltage, and Part II is the training pattern of three-phase voltage sag Wave data；It is single-phase
All voltage dip Wave datas are regarded training sample, three-phase voltage sag by the training pattern of voltage dip Wave data
The training pattern of Wave data is temporarily dropped by a tuple sample, wherein single-phase voltage of A, B, C three-phase voltage sag Wave data
Data prediction part of the prediction data of waveform training pattern as three-phase voltage sag waveform training pattern；
S302, it is sample data that single-phase voltage is temporarily dropped to Wave data, sample dataFor N band mark
The data of label, divide K classes；Sample data is divided into training sample data and test sample number in the method for K- cross-trainings
According to wherein training sample data are row data, using the characteristic of extraction as column data；
S3O3：One voltage dip event is divided into 3 sample datas according to separate, K classes are picked out according to wave character
Waveform numerical example data are used for model training and verification；
S3O4：The sample size manually chosen is less, in order to make full use of sample, uses K- cross validation methods
The voltage dip Wave data training pattern of training gradient boosted tree, give over to every time verification for the total sample size of voltage dip
1/K, being used for the sample size of training every time accordingly increases, but K- folding cross validations are required for operation 10 times for each model,
Last average test result can weigh the performance of model, choose the model of validation error minimum；
S3O5：Using the voltage dip Wave data training pattern of trained gradient boosted tree in unfiled data
Test, checks the data of classification error, if the sample of homogenous characteristics is less than 2 in training data, is intersected using leaving-one method and tested
Card, chooses the model of error minimum；
S306：By sorted monophasic waveform data, combined by three-phase, become a new classification, new point
Class is associated with electric network fault data, lightning stroke data and meteorological data, and temporally the correlation method of piece realizes voltage dip occurrence cause
Association analysis；
S307：The phase angle of the sorted data of voltage dip is calculated, for voltage dip caused by failure, according to phase angle
Voltage dip is divided into seven class asymmetrical three-phases temporarily to drop, is single-phase short circuit by voltage dip type identification fault occurrence reason
Failure, two-phase short-circuit fault or three phase short circuit fault.
More preferably, temporarily drop specifically includes seven class asymmetrical three-phases：A phases single-phase earthing fault, B phases single-phase earthing fault, C
Phase single-phase earthing fault, AB two-phase short-circuit faults, BC two-phase short-circuit faults, AC two-phase short-circuit faults and three phase short circuit fault.
Electric network fault data, customer charge data, power grid lightning stroke data and the main transformer exploitation in synchronous certain city May are thrown
Data are cut, are analyzed with being associated property of timeslice, the sample data there will be relevance temporarily drop reason knowledge with above-mentioned model
Not, the 34 component classes and the degree of consistency of the result of correlation analysis that confirmation Model Identification goes out.There are 75 in 671 samples
Sample matches to electric network fault, 20 sample matches to customer charge, 10 sample matches to power grid be struck by lightning, 3 sample matches
To main transformer switch；The voltage sag reason of Model Identification and the voltage sag reason uniformity of correlation analysis reach 89%, warp
Overmatching, has 15 to match electric network fault in 34 waveform separations, removes 20% noise data, and failure accounting is maximum
Preceding 12 classification in 420 temporarily drop recording marks temporarily to drop caused by electric network fault, calculate the phase angle of the temporary drop data of failure,
Voltage dip is divided into seven class asymmetrical three-phases according to phase angle temporarily to drop, is single-phase by temporarily dropping type identification fault occurrence reason
A certain kind in short trouble, two-phase short-circuit fault, three phase short circuit fault, double earthfault.
It the above is only the preferred embodiment of the present invention, it should be pointed out that：Come for those skilled in the art
Say, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications should also regard
For protection scope of the present invention.
Claims (10)
1. a kind of voltage sag reason recognition methods based on EM algorithms and gradient boosted tree, it is characterised in that temporary including voltage
Drop Feature Engineering, temporarily drop event category model training is associated with external data；It is temporary based on voltage dip Feature Engineering extraction voltage
After the characteristic of drop data, by temporarily drop event category model training by voltage dip data reasons classify after, with external number
According to association.
2. the voltage sag reason recognition methods according to claim 1 based on EM algorithms and gradient boosted tree, its feature
It is,
Voltage dip Feature Engineering extracts characteristic from voltage dip Wave data；
The characteristic specifically includes：Voltage dip is separate, temporary drawdown degree, the voltage dip duration, location, monitoring
Point voltage class, temporarily drop 10%-180% sections accounting, temporarily drop initial are to the most deep period temporarily dropped, the most deep temporarily drop that temporarily drops to is recovered
Period, temporary drawdown degree, squareness factor, temporarily drop 20% bottom of concave and head ratio, 20% area ratio of concave, waveform temporarily drop
The degree of bias, waveform set-back and frequency domain cluster 5 groups of accountings.
3. the voltage sag reason recognition methods according to claim 1 based on EM algorithms and gradient boosted tree, its feature
It is,
Temporarily drop event category model training includes the voltage dip Wave data training pattern of gradient boosted tree and the electricity of EM algorithms
Temporarily drop perunit value prediction model, voltage dip Wave data training pattern are used to carry out parameter training, electricity for sample data pressure
Temporarily drop perunit value prediction model is used to identify electric grid secondary voltage perunit value pressure.
4. the voltage sag reason recognition methods according to claim 3 based on EM algorithms and gradient boosted tree, its feature
It is,
Voltage dip Wave data training pattern J (θ) based on gradient boosted tree is expressed as：
<mrow>
<mi>J</mi>
<mrow>
<mo>(</mo>
<mi>&theta;</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>N</mi>
</munderover>
<mi>L</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<mi>f</mi>
<mo>(</mo>
<mrow>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<mi>&theta;</mi>
</mrow>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mo>+</mo>
<mi>&Omega;</mi>
<mrow>
<mo>(</mo>
<mi>&theta;</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>4</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein f (xi, θ) be voltage dip Wave data classification discreet value；L(yi,f(xi, θ)) it is loss function, represent voltage
The temporarily similitude between drop waveform classification discreet value and class label, xiFor training sample, θ is parameter sets；N is sample data number
Mesh, Ω (θ) are regular terms, choose regular terms of the L2 canonicals as the training pattern of voltage dip Wave data, then：
<mrow>
<mi>&Omega;</mi>
<mrow>
<mo>(</mo>
<mi>&theta;</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mn>2</mn>
</mfrac>
<msup>
<mi>&theta;</mi>
<mn>2</mn>
</msup>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>5</mn>
<mo>)</mo>
</mrow>
</mrow>
Loss function is used as using logistic regression loss：
L(yi,f(xi, θ))=- ylog (f (x, θ))-(1-y) log (1-f (x, θ)) (6)
In order to simplify computation complexity and lifting calculating speed, loss function is approached with Newton interpolating method, i.e.,：
<mrow>
<mi>L</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<mi>f</mi>
<mo>(</mo>
<mrow>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<mi>&theta;</mi>
</mrow>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msub>
<mo>&dtri;</mo>
<mi>f</mi>
</msub>
<mi>L</mi>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<mi>&theta;</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
<msubsup>
<mo>&dtri;</mo>
<mi>f</mi>
<mn>2</mn>
</msubsup>
<mi>L</mi>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<msup>
<mi>f</mi>
<mn>2</mn>
</msup>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<mi>&theta;</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>7</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msub>
<mo>&dtri;</mo>
<mi>f</mi>
</msub>
<mi>L</mi>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mrow>
<mo>-</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
</mrow>
<mrow>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<mi>&theta;</mi>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>+</mo>
<mfrac>
<mrow>
<mn>1</mn>
<mo>-</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
</mrow>
<mrow>
<mn>1</mn>
<mo>-</mo>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<mi>&theta;</mi>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>8</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msubsup>
<mo>&dtri;</mo>
<mi>f</mi>
<mn>2</mn>
</msubsup>
<mi>L</mi>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mrow>
<mo>-</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
</mrow>
<msup>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<mi>&theta;</mi>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mfrac>
<mo>+</mo>
<mfrac>
<mrow>
<mn>1</mn>
<mo>-</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
</mrow>
<msup>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>-</mo>
<mi>f</mi>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<mi>&theta;</mi>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mfrac>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>9</mn>
<mo>)</mo>
</mrow>
</mrow>
xiFor training sample, yiFor the corresponding class scalar of training sample, x is training sample set, and y is the corresponding category of training sample
Quantity set, xi∈ x, yi∈y；For Interpolation-Radix-Function；
The corresponding category quantity set y of training sample has K class waveforms, and the corresponding category quantity set of training sample is labeled as set D, in
Ratio shared by k class waveform samples is pk, then the purity Ent (D) of voltage dip waveform data sample set be：
<mrow>
<mi>E</mi>
<mi>n</mi>
<mi>t</mi>
<mrow>
<mo>(</mo>
<mi>D</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mn>1</mn>
<mo>-</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>k</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>K</mi>
</munderover>
<msubsup>
<mi>p</mi>
<mi>k</mi>
<mn>2</mn>
</msubsup>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>10</mn>
<mo>)</mo>
</mrow>
</mrow>
According to the purity formula (10) of sample set, select a certain feature as decision tree root node Det (D, a), root node Det
(D, selection principle a) is to compare the minimum value for choosing purity ratio under all features：
<mrow>
<mi>D</mi>
<mi>e</mi>
<mi>t</mi>
<mrow>
<mo>(</mo>
<mi>D</mi>
<mo>,</mo>
<mi>a</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munder>
<mrow>
<mi>arg</mi>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
</mrow>
<mi>a</mi>
</munder>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>k</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>K</mi>
</munderover>
<mfrac>
<mrow>
<mo>|</mo>
<msup>
<mi>D</mi>
<mi>k</mi>
</msup>
<mo>|</mo>
</mrow>
<mi>D</mi>
</mfrac>
<mi>E</mi>
<mi>n</mi>
<mi>t</mi>
<mrow>
<mo>(</mo>
<mi>D</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>11</mn>
<mo>)</mo>
</mrow>
</mrow>
Its discreet value f (x are calculated after each feature of iterationi；θ) and yiBetween difference, using difference as next decision tree
Target；Decision tree under each feature of cycle calculations, until difference is less than or equal to ε, algorithmic statement, ε is the convergency value of definition.
5. the voltage sag reason recognition methods according to claim 4 based on EM algorithms and gradient boosted tree, its feature
It is,
Voltage dip caused by voltage dip, large user's load caused by voltage dip Wave data classification specifically includes failure and
Transformer excitation shoves caused voltage dip.
6. the voltage sag reason recognition methods according to claim 3 based on EM algorithms and gradient boosted tree, its feature
It is,
In Wave data Normal Distribution X~N (μ；σ2) on the premise of, the voltage dip perunit value based on EM algorithms estimates mould
Type includes E steps and M steps；E walks the distributed constant that optimal hidden variable Z is inferred to according to training data；M is walked optimal hidden variable Z's
On the basis of distributed constant, the distributed constant of Z is corrected according to Wave data；Loop iteration E is walked to be walked with M, until optimal hidden variable Z
Parameter convergence.
7. the voltage sag reason recognition methods according to claim 6 based on EM algorithms and gradient boosted tree, its feature
It is,
Voltage dip perunit value prediction model based on EM algorithms specifically includes following steps：
S101, calculates the average value e of virtual value under the first two cycle in voltage dip Wave data0Have under most latter two cycle
The average value e of valid value1, respectively for e0Retain [0.9e0, 1.1e0] in the range of virtual value and e1Retain [0.9e1, 1.1e1] model
Enclose interior virtual value, wherein e0And e1The composition component for the optimal hidden variable Z data group estimated in being walked for E；
S102, M step specifically include following steps, are directed to [0.9e respectively0, 1.1e0] and [0.9e1, 1.1e1] in the range of virtual value
Data carry out Gauss curve fitting X~N (μ under normal distribution0；σ0 2), X~N (μ1；σ1 2), obtain the parameter μ within the scope of two0With
μ1, μ0And μ1For the parameters revision of M Walk；
S103, calculates respectively | μ0-e0|、|μ0-e1|、|μ1-e0| and | μ1-e1|, if any one value is less than or equal to ε, algorithm stops
Only calculate, take less than or equal to the μ under ε0Or μ1As secondary voltage perunit value, wherein ε for parameter error (such as take 0.01 or
0.001)；
S104, if | μ0-e0|、|μ0-e1|、|μ1-e0|、|μ1-e1| both greater than ε, then respectively in μ0Or μ1Lower reservation [0.9 μ0, 1.1
μ0] and [0.9 μ1, 1.1 μ1] in the range of virtual value, carry out normal distribution under Gauss curve fittingObtain parameterWith
S105, repeats claim steps S103 and step S104 until convergence.
8. the voltage sag reason recognition methods according to claim 1 based on EM algorithms and gradient boosted tree, its feature
It is,
External system data correlation combination topological structure of electric is by the data after voltage dip disaggregated model training, with external data
Time be associated analysis, specifically include following steps：
S301, voltage dip event category：Voltage dip Wave data training pattern is divided into two parts, and Part I is single-phase electricity
The training pattern of Wave data temporarily drops in pressure, and Part II is the training pattern of three-phase voltage sag Wave data；Single-phase voltage is temporary
All voltage dip Wave datas are regarded training sample, three-phase voltage sag Wave data by the training pattern of drop Wave data
Training pattern using A, B, C three-phase voltage sag Wave data as a tuple sample, wherein single-phase voltage temporarily drop waveform training
Data prediction part of the prediction data of model as three-phase voltage sag waveform training pattern；
S302, it is sample data that single-phase voltage is temporarily dropped to Wave data, sample dataFor the number of N number of tape label
According to division K classes；Sample data is divided into training sample data and test sample data in the method for K- cross-trainings, wherein
Training sample data are row data, using the characteristic of extraction as column data；
S3O3：One voltage dip event is divided into 3 sample datas according to separate, K class waveforms are picked out according to wave character
Numerical example data are used for model training and verification；
S3O4：Using the voltage dip Wave data training pattern of K- cross validation methods training gradient boosted tree, give over to every time
The 1/K for the total sample size of voltage dip of verification, accordingly increases for trained sample size, chooses the mould of validation error minimum
Type；
S3O5：Tested using the voltage dip Wave data training pattern of trained gradient boosted tree in unfiled data,
Check the data of classification error, if the sample of homogenous characteristics is less than 2 in training data, using leave one cross validation, choose
The model of error minimum；
S306：By sorted monophasic waveform data, combined by three-phase, become a new classification, new classification with
Electric network fault data, lightning stroke data are associated with meteorological data, and temporally the correlation method of piece realizes the pass of voltage dip occurrence cause
Connection analysis；
S307：The phase angle of the sorted data of voltage dip is calculated, for voltage dip caused by failure, according to phase angle by voltage
Temporarily drop is divided into seven class asymmetrical three-phases and temporarily drops, and is single-phase earthing fault, two by voltage dip type identification fault occurrence reason
Phase short trouble or three phase short circuit fault.
9. the voltage sag reason recognition methods according to claim 2 based on EM algorithms and gradient boosted tree, its feature
It is,
Temporarily drop 10%-180% sections accounting is：Using perunit value as 100%, 18 sections are divided into up and down, calculate virtual value
Data fall the number and the accounting of sum in each section；
Temporarily drop initial is to most deep period for temporarily dropping：Between being temporarily reduced at the beginning of perunit value 90% with when temporarily dropping to innermost
Between difference tsag1；
Most it is deep temporarily drop to temporarily drop recover period be：Temporarily drop bosom returns to the time difference tsag2 of perunit value 90% with temporarily drop；
Squareness factor is：The temporarily ratio of drop area and temporary decline rectangular area, in order to facilitate reference area, makes perunit value correspond to
Virtual value be 1, the numerical value of the sampling number according to corresponding rate conversion between [0 1] in region, temporary decline square temporarily drop
Shape area is SfThe sampling number in region temporarily drops in=perunit value *, and it is S temporarily to drop areaaRegion temporarily drops in=temporary decline rectangular area-
Sampled point summation,
Temporarily drop 20% bottom of concave and head ratio：Concave, temporarily drops in temporarily drop most bottom section duration and temporarily drop duration ratio
20% bottom value range is less than ((perunit value-temporary drawdown degree) the temporary drawdown degree of * 0.2+)；
Temporarily drop 20% area ratio of concave：Temporarily drop most bottom section area accounts for the area ratio in whole temporarily drop region；
Frequency domain clusters 5 groups of accountings：The effective Value Data of waveform switchs to the data of domain space by Fourier transform, utilizes K averages
Clustering method gathers data under frequency domain accounts for the ratio of sum for 5 classes, every one kind data.
10. the voltage sag reason recognition methods according to claim 8 based on EM algorithms and gradient boosted tree, its feature
It is,
Temporarily drop specifically includes seven class asymmetrical three-phases：A phases single-phase earthing fault, B phases single-phase earthing fault, the event of C phases single-phase short circuit
Barrier, AB two-phase short-circuit faults, BC two-phase short-circuit faults, AC two-phase short-circuit faults and three phase short circuit fault.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711119905.0A CN107979086B (en) | 2017-11-14 | 2017-11-14 | Voltage sag reason identification method based on EM algorithm and gradient lifting tree |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711119905.0A CN107979086B (en) | 2017-11-14 | 2017-11-14 | Voltage sag reason identification method based on EM algorithm and gradient lifting tree |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107979086A true CN107979086A (en) | 2018-05-01 |
| CN107979086B CN107979086B (en) | 2019-12-27 |
Family
ID=62013409
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711119905.0A Active CN107979086B (en) | 2017-11-14 | 2017-11-14 | Voltage sag reason identification method based on EM algorithm and gradient lifting tree |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107979086B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109361263A (en) * | 2018-09-29 | 2019-02-19 | 安徽科派自动化技术有限公司 | Long-distance monitoring for electric power quality system and method based on voltage dip monitor |
| CN109507530A (en) * | 2018-11-16 | 2019-03-22 | 国网江苏省电力有限公司电力科学研究院 | Voltage dip source of trouble retroactive method, system and storage medium |
| CN109787219A (en) * | 2018-12-24 | 2019-05-21 | 河海大学 | A kind of intelligent identification Method of voltage dip |
| CN109800660A (en) * | 2018-12-27 | 2019-05-24 | 国网江苏省电力有限公司电力科学研究院 | A kind of voltage sag source identification method and system based on big data cluster |
| CN110687344A (en) * | 2019-10-24 | 2020-01-14 | 南京南瑞继保电气有限公司 | Single-phase voltage sag detection method and device, voltage restorer, equipment and medium |
| CN110817636A (en) * | 2019-11-20 | 2020-02-21 | 上海电气集团股份有限公司 | Elevator door system fault diagnosis method, device, medium and equipment |
| CN111337791A (en) * | 2020-03-25 | 2020-06-26 | 国网河南省电力公司电力科学研究院 | Power distribution network single-phase earth fault line selection method based on gradient lifting tree algorithm |
| CN113030616A (en) * | 2021-03-03 | 2021-06-25 | 国网福建省电力有限公司 | Sensitive load identification method based on voltage sag monitoring data |
| CN114050613A (en) * | 2021-11-29 | 2022-02-15 | 国网湖南省电力有限公司 | Online identification and tracing method and system for power grid voltage transient event |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040051387A1 (en) * | 2002-09-17 | 2004-03-18 | Lasseter Robert H. | Control of small distributed energy resources |
| CN101553738A (en) * | 2006-10-13 | 2009-10-07 | Tnb研究调查有限公司 | Flashover analysis tool |
| CN103793853A (en) * | 2014-01-21 | 2014-05-14 | 中国南方电网有限责任公司超高压输电公司检修试验中心 | Overhead power transmission line running state assessment method based on bidirectional Bayesian network |
| CN104966161A (en) * | 2015-06-16 | 2015-10-07 | 北京四方继保自动化股份有限公司 | Electric energy quality recording data calculating analysis method based on Gaussian mixture model |
| CN105160598A (en) * | 2015-08-28 | 2015-12-16 | 国网智能电网研究院 | Power grid service classification method based on improved EM algorithm |
| CN105205571A (en) * | 2015-10-20 | 2015-12-30 | 河海大学 | Risk-considered urban power network operation security assessment method |
-
2017
- 2017-11-14 CN CN201711119905.0A patent/CN107979086B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040051387A1 (en) * | 2002-09-17 | 2004-03-18 | Lasseter Robert H. | Control of small distributed energy resources |
| CN101553738A (en) * | 2006-10-13 | 2009-10-07 | Tnb研究调查有限公司 | Flashover analysis tool |
| CN103793853A (en) * | 2014-01-21 | 2014-05-14 | 中国南方电网有限责任公司超高压输电公司检修试验中心 | Overhead power transmission line running state assessment method based on bidirectional Bayesian network |
| CN104966161A (en) * | 2015-06-16 | 2015-10-07 | 北京四方继保自动化股份有限公司 | Electric energy quality recording data calculating analysis method based on Gaussian mixture model |
| CN105160598A (en) * | 2015-08-28 | 2015-12-16 | 国网智能电网研究院 | Power grid service classification method based on improved EM algorithm |
| CN105205571A (en) * | 2015-10-20 | 2015-12-30 | 河海大学 | Risk-considered urban power network operation security assessment method |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109361263A (en) * | 2018-09-29 | 2019-02-19 | 安徽科派自动化技术有限公司 | Long-distance monitoring for electric power quality system and method based on voltage dip monitor |
| CN109507530A (en) * | 2018-11-16 | 2019-03-22 | 国网江苏省电力有限公司电力科学研究院 | Voltage dip source of trouble retroactive method, system and storage medium |
| CN109787219A (en) * | 2018-12-24 | 2019-05-21 | 河海大学 | A kind of intelligent identification Method of voltage dip |
| CN109787219B (en) * | 2018-12-24 | 2022-09-02 | 河海大学 | Intelligent identification method for voltage sag |
| CN109800660A (en) * | 2018-12-27 | 2019-05-24 | 国网江苏省电力有限公司电力科学研究院 | A kind of voltage sag source identification method and system based on big data cluster |
| CN109800660B (en) * | 2018-12-27 | 2020-11-10 | 国网江苏省电力有限公司电力科学研究院 | Voltage sag source identification method and system based on big data clustering |
| CN110687344A (en) * | 2019-10-24 | 2020-01-14 | 南京南瑞继保电气有限公司 | Single-phase voltage sag detection method and device, voltage restorer, equipment and medium |
| CN110817636A (en) * | 2019-11-20 | 2020-02-21 | 上海电气集团股份有限公司 | Elevator door system fault diagnosis method, device, medium and equipment |
| CN111337791A (en) * | 2020-03-25 | 2020-06-26 | 国网河南省电力公司电力科学研究院 | Power distribution network single-phase earth fault line selection method based on gradient lifting tree algorithm |
| CN113030616A (en) * | 2021-03-03 | 2021-06-25 | 国网福建省电力有限公司 | Sensitive load identification method based on voltage sag monitoring data |
| CN114050613A (en) * | 2021-11-29 | 2022-02-15 | 国网湖南省电力有限公司 | Online identification and tracing method and system for power grid voltage transient event |
| CN114050613B (en) * | 2021-11-29 | 2023-10-27 | 国网湖南省电力有限公司 | Online identification and tracing method and system for power grid voltage transient event |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107979086B (en) | 2019-12-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107979086A (en) | Voltage sag reason recognition methods based on EM algorithms and gradient boosted tree | |
| CN102437573B (en) | Evaluation and control method and system for reliability of electric distribution network based on fuzzy modeling | |
| CN103023023B (en) | Comprehensive evaluation method based on multi-stress for electric energy quality of monitoring points of electrified railway | |
| CN104865499B (en) | A kind of extra high voltage direct current transmission line internal fault external fault recognition methods | |
| CN110488154B (en) | Low-current grounding line selection method for dispatching master station end | |
| CN108306284A (en) | A kind of online load modeling method measured based on local intelligence | |
| CN109842122A (en) | A kind of low-voltage platform area low-voltage administering method | |
| CN107765139A (en) | A kind of resonant earthed system fault line selection method for single-phase-to-ground fault of high-accuracy | |
| CN107167726A (en) | A kind of circuit breaker internal puncture electric arc modeling method | |
| CN102253296A (en) | Method for testing comprehensive device of transformer | |
| CN104036434A (en) | Evaluation method for load supply capacity of power distribution network | |
| CN111044828B (en) | Three-phase transformer winding parameter online monitoring method based on positive and negative sequence equations | |
| CN111614066A (en) | Automatic setting method and system for relay protection setting value of power distribution network | |
| CN106558883A (en) | A kind of electric network fault control system for reactive power compensator | |
| Qian et al. | Probabilistic short-circuit current in active distribution networks considering low voltage ride-through of photovoltaic generation | |
| CN104977488B (en) | A kind of transformer excitation flow recognition method based on difference current gradient angle approximate entropy | |
| CN110531195A (en) | A method of identification transformer excitation flow and internal fault | |
| Elsamahy et al. | Enhancement of the coordination between generator phase backup protection and generator capability curves in the presence of a midpoint STATCOM using support vector machines | |
| CN114461982B (en) | Power transmission line protection characteristic identification and voltage sag duration estimation method | |
| CN104036433A (en) | Method for evaluating running management level of power distribution network | |
| CN114755526A (en) | Fault positioning method introducing mutation quantity ratio and correlation coefficient | |
| CN107565547A (en) | A kind of power distribution network operation reliability evaluation and optimization system | |
| CN115048760A (en) | Load power quality tracing method based on typical power quality feature library | |
| CN104749453A (en) | Method for reducing influences imposed on user voltage sag by external grid single-phase grounding fault | |
| CN114358564A (en) | Intelligent analysis method for low-voltage fault of distribution transformer based on electric power big data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |