CN106127614A - A kind of electricity exception data identification method based on three parameter Weir distributions - Google Patents

Abstract

The invention discloses a kind of electricity exception data identification method based on three parameter Weir distributions, comprise the following steps: gather electric power data, and electric power data is ranked up according to amplitude size, obtain electric power data sequence；Electric power data sequence uses three parameter Weir distributed models be fitted, obtains the estimated value of Weir three parameters of distribution；According to the estimated value of Weir three parameters of distribution, determine electricity exception data marginal value；Electric power data is normal data in critical value range, if it exceeds critical value range is then judged to abnormal data；The present invention has the widely suitability to electric power data distributional pattern, and electric power data when being particularly suited for nonnormal distribution processes, it is ensured that the accuracy of electric power data exception identification and effectiveness.

Description

A kind of electricity exception data identification method based on three parameter Weir distributions

Technical field

The present invention relates to a kind of electricity exception data identification method, be specifically related to a kind of electricity based on three parameter Weir distributions Power anomalous data identification method.

Background technology

For grasping network load situation, analyzing power equipment running status and make Rational Decision, power system acquires Substantial amounts of electric power data, such as Power system load data, power quality data and wind power data etc.；In electric power data acquisition process In, owing to equipment fault, instrument gather the factors such as mistake and outer signals interference, the data of collection there will be abnormal existing As, i.e. produce abnormal data；Electricity exception data can affect the judgement to electrical network or equipment running status, affects aid decision Reasonability；Power monitoring system is also required to identification and the ANOMALOUS VARIATIONS of record Monitoring Data simultaneously, works with research and application equipment State, ensures power system security；Therefore, according to certain criterion, the abnormal data in identification electric power data is power system An important engineering problem；In recent years, certain methods is introduced in anomalous data identification, as used fuzzy clustering and being subordinate to Degree is analyzed electric power data and is to what extent belonged to abnormal data；Use on-line study method identification abnormal data etc.；These sides Method is computationally intensive, process is loaded down with trivial details, still has a certain distance from reality application；Engineering carries out abnormal identification to data, uses more The thinking of Outlier Data identification in probability statistics；Under normal circumstances, there is certain fluctuation range in all kinds of electric power datas, the most greatly Part data be distributed in certain interval in, when data occur in extremely low interval of probability, it is believed that these Monitoring Data are one Data, need to alleviate weight in state estimation, need to draw attention or make early warning etc. and process in aid decision；Based on normal state The normal sample method of distribution is the most conventional method, when determining critical values of anomalies, it is assumed that electric power data meets normal distribution, The data that will deviate from sample main body are considered as exceptional value, use three times of standard deviation policy setting critical values of anomalies, and this method calculates Easy；But normal distribution is a kind of preferably distribution, and electricity exception Data Representation is complicated, be likely to be of randomness, impact and The characteristic that asymmetry etc. are complicated, when data deviation normal distribution, the critical values of anomalies that such method determines is the most applicable.

Summary of the invention

The present invention provide a kind of can the method for accurate and effective identification electricity exception data.

The technical solution used in the present invention is: a kind of electricity exception data identification methods based on three parameter Weir distributions, Comprise the following steps:

Gather electric power data, and electric power data is ranked up according to amplitude size, obtain electric power data sequence；

Electric power data sequence uses three parameter Weir distributed models be fitted, obtains estimating of Weir three parameters of distribution Evaluation；

According to the estimated value of Weir three parameters of distribution, determine electricity exception data marginal value；

Electric power data is normal data in critical value range, if it exceeds critical value range is then judged to abnormal data.

Further, electric power data sequence is carried out pretreatment according to Median rank formula.

Further, described electric power data includes the data such as load prediction data, power quality data and wind power.

A kind of electricity exception data identification method based on three parameter Weir distributions, comprises the following steps:

A, collection electric power data t_i(1≤i≤n), is ranked up according to the amplitude size of electric power data, obtains amplitude from little Electric power data sequence d to longer spread_i(1≤i≤n)；Wherein n is sampling number；

B, to electric power data sequence d_i(1≤i≤n), according to Median rank formulaIt is calculated data set (d_i,F_i) i=1,2 ..., n；

C, electric power data sequence use three parameter Weir fittings of distribution, three Weir distributed constants the most to be estimated are Form parameter a, scale parameter b and location parameter c；The estimated value of three Weir distributed constants is respectivelyWith

C1, initialized location parameter c, c₀=0, initialize iteration precision λ；

C2, according to Y_i=lnln (1/ (1-F_i)),X_i=ln (d_i-c₀) calculate linear function variable X_i, Y_i；Both meet Y_i =BX_i+ A, wherein, B=a；A=alnb；

C3, to data set (d_i,F_i) i=1,2 ..., n and the vectorial X=[X of linear function variable composition₁,X₂,...,X_n] With Y=[Y₁,Y₂,...,Y_n], estimate three parameters of Weir distributionWith

$\hat{a}=\frac{\mathrm{cov}(X,Y)}{D\left(X\right)}$

$\hat{b}=\exp (\stackrel{\‾}{X}-\stackrel{\‾}{Y}/\hat{a})$

$\hat{c}=\frac{1}{n}{\mathrm{\Σ}}_{i=1}^n\{d_i-\hat{b}{\[-ln(1-F_i)\]}^{(1/\hat{a})}\}$

$\stackrel{\‾}{X}=\frac{1}{n}\left({\mathrm{\Σ}}_{i=1}^n{X}_i\right)$

$\stackrel{\‾}{Y}=\frac{1}{n}\left({\mathrm{\Σ}}_{i=1}^n{Y}_i\right)$

Wherein, cov (X, Y) is the covariance of vector X and Y, and D (X) is the variance of vector X；

C4, whenTime, obtain the final estimated result of three parameters, otherwise update location parameter c, order Repeat step C2-C3, until iteration is to meeting the condition of convergence

D, according to Weir distribution three parameters estimated value, determine electricity exception data marginal value；

Abnormal data marginal value t on the upside of it⁺For:

Abnormal data marginal value t on the downside of it_-For:Wherein θ is abnormal data probability；

E, as electric power data t_jMeet t_-<t_j<t⁺Time, it is determined that electric power data t_jFor normal data, when meeting t_j<t_-Or t_j>t⁺ Time, it is determined that electric power data t_jFor abnormal data.

The invention has the beneficial effects as follows:

The present invention uses three parameter Weir distributed models to be fitted electric power data, has electric power data distributional pattern The widely suitability, electric power data when being particularly suited for nonnormal distribution processes；Ensure that electric power data exception identification Accuracy and effectiveness.

Accompanying drawing explanation

Fig. 1 is that Weir is distributed three parameter estimation calculation flow charts.

Fig. 2 is identification result on the upside of electric harmonic current anomaly data.

Fig. 3 is identification result on the downside of electric harmonic current anomaly data.

Detailed description of the invention

The invention will be further described with specific embodiment below in conjunction with the accompanying drawings.

A kind of electricity exception data identification method based on three parameter Weir distributions, comprises the following steps:

Gather electric power data, and electric power data is ranked up according to amplitude size, obtain electric power data sequence；

Electric power data sequence uses three parameter Weir distributed models be fitted, obtains estimating of Weir three parameters of distribution Evaluation；

According to the estimated value of Weir three parameters of distribution, determine electricity exception data marginal value；

Electric power data is normal data in critical value range, if it exceeds critical value range is then judged to abnormal data.

Further, electric power data sequence is carried out pretreatment according to Median rank formula.

Further, described electric power data includes the data such as load prediction data, power quality data and wind power.

A kind of electricity exception data identification method based on three parameter Weir distributions, comprises the following steps:

A, collection electric power data t_i(1≤i≤n), is ranked up according to the amplitude size of electric power data, obtains amplitude from little Electric power data sequence d to longer spread_i(1≤i≤n)；Wherein n is sampling number；

B, to electric power data sequence d_i(1≤i≤n), according to Median rank formulaIt is calculated data set (d_i,F_i) i=1,2 ..., n；

C, electric power data sequence use three parameter Weir fittings of distribution, three Weir distributed constants the most to be estimated are Form parameter a, scale parameter b and location parameter c；The estimated value of three Weir distributed constants is respectivelyWith

C1, initialized location parameter c, c₀=0, initialize iteration precision λ；

C2, according to Y_i=lnln (1/ (1-F_i)),X_i=ln (d_i-c₀) calculate linear function variable X_i, Y_i；Both meet Y_i =BX_i+ A, wherein, B=a；A=alnb；

C3, to data set (d_i,F_i) i=1,2 ..., n and the vectorial X=[X of linear function variable composition₁,X₂,...,X_n] With Y=[Y₁,Y₂,...,Y_n], estimate three parameters of Weir distributionWith

$\hat{a}=\frac{\mathrm{cov}(X,Y)}{D\left(X\right)}$

$\hat{b}=\exp (\stackrel{\&OverBar;}{X}-\stackrel{\&OverBar;}{Y}/\hat{a})$

$\hat{c}=\frac{1}{n}{\mathrm{\&Sigma;}}_{i=1}^n\{d_i-\hat{b}{\&lsqb;-ln(1-F_i)\&rsqb;}^{(1/\hat{a})}\}$

$\stackrel{\&OverBar;}{X}=\frac{1}{n}\left({\mathrm{\&Sigma;}}_{i=1}^n{X}_i\right)$

$\stackrel{\&OverBar;}{Y}=\frac{1}{n}\left({\mathrm{\&Sigma;}}_{i=1}^n{Y}_i\right)$

Wherein, cov (X, Y) is the covariance of vector X and Y, and D (X) is the variance of vector X；

C4, whenTime, obtain the final estimated result of three parameters, otherwise update location parameter c, order Repeat step C2-C3, until iteration is to meeting the condition of convergence

D, according to Weir distribution three parameters estimated value, determine electricity exception data marginal value；

Abnormal data marginal value t on the upside of it⁺For:

Abnormal data marginal value t on the downside of it_-For:Wherein θ is abnormal data probability； Being set in the distribution of electric power data three parameter Weir, the abnormal data probability of occurrence of upper and lower both sides is equal, the upper side and lower side herein Abnormal data distribution probability is θ/2, determines the upper side and lower side abnormal data marginal value according to equal probability principle；

E, as electric power data t_jMeet t_-＜ t_j＜ t⁺Time, it is determined that electric power data t_jFor normal data, when meeting t_j＜ t_-Or t_j ＞ t⁺Time, it is determined that electric power data t_jFor abnormal data.

For proving effectiveness of the invention and accuracy, with the electric harmonic current monitoring data instance on certain airport, carry out The simulation analysis of anomalous data identification, this electric harmonic current data sampling interval is 1s, uses the electric harmonic of first 30 minutes Current monitoring data carry out parameter estimation, and the data of latter 10 minutes are as data to be identified.

Take iteration precision λ=0.003, abnormal probability θ=0.3%；

This method is used to estimate the form parameter obtainedScale parameterLocation parameterCritical with downside abnormal data Value t, upside abnormal data marginal value t⁺, as shown in table 1；Use normal sample method to obtain mean μ, variances sigma and downside extremely to face Dividing value t, upside abnormal data marginal value t⁺, as shown in table 2；According to upside abnormal data marginal value t determined⁺Abnormal with downside Marginal value t, retains interval interior normal data, rejecting abnormalities data；After distribution uses two kinds of method rejecting abnormalities data, computer Distribution standard deviation, as shown in table 3 with the comparing result of raw power data estimation standard deviation.

Table 1 the inventive method parameter estimation result and abnormal data marginal value

Table 2 normal sample method parameter estimation result and abnormal data marginal value

Variable	μ	σ	t-	t+
					Estimated result	0.2770	0.0638	0.0856	0.4684

Table 3 standard deviation results contrast

Data	Raw power data	Normal sample method	The inventive method
				Estimated standard deviation	0.0132	0.0041	0.0031

By table 3 result of calculation it can be seen that the abnormal data marginal value that the method for the present invention determines ensure that electric power is different The accuracy of regular data identification, and this method still have in the case of electric power data portion Normal Distribution good identification effect Really, calibration aspect this law has wider adaptability.

Claims (4)

1. an electricity exception data identification method based on three parameter Weir distributions, it is characterised in that comprise the following steps:

Gather electric power data, and electric power data is ranked up according to amplitude size, obtain electric power data sequence；

Electric power data sequence uses three parameter Weir distributed models be fitted, obtains the estimation of Weir three parameters of distribution Value；

According to the estimated value of Weir three parameters of distribution, determine electricity exception data marginal value；

Electric power data is normal data in critical value range, if it exceeds critical value range is then judged to abnormal data.

A kind of electricity exception data identification methods based on three parameter Weir distributions the most according to claim 1, its feature It is, electric power data sequence is carried out pretreatment according to Median rank formula.

A kind of electricity exception data identification methods based on three parameter Weir distributions the most according to claim 1, its feature Being, described electric power data includes load prediction data, power quality data and wind power data.

4. an electricity exception data identification method based on three parameter Weir distributions, it is characterised in that comprise the following steps:

A, collection electric power data t_i(1≤i≤n), is ranked up according to the amplitude size of electric power data, obtains amplitude from small to large Electric power data sequence d of arrangement_i(1≤i≤n)；Wherein n is sampling number；

B, to electric power data sequence d_i(1≤i≤n), according to Median rank formulaIt is calculated data set (d_i, F_i) i=1,2 ..., n；

C, electric power data sequence using three parameter Weir fittings of distribution, three Weir distributed constants the most to be estimated are shape Parameter a, scale parameter b and location parameter c；The estimated value of three Weir distributed constants is respectivelyWith

C1, initialized location parameter c, c₀=0, initialize iteration precision λ；

C2, according to Y_i=lnln (1/ (1-F_i)),X_i=ln (d_i-c₀) calculate linear function variable X_i, Y_i；Both meet Y_i=BX_i + A, wherein, B=a；A=alnb；

C3, to data set (d_i,F_i) i=1,2 ..., n and the vectorial X=[X of linear function variable composition₁,X₂,...,X_n] and Y =[Y₁,Y₂,...,Y_n], estimate three parameters of Weir distributionWith

$\hat{a}=\frac{\mathrm{cov}(X,Y)}{D\left(X\right)}$

$\hat{b}=\exp (\stackrel{\&OverBar;}{X}-\stackrel{\&OverBar;}{Y}/\hat{a})$

$\hat{c}=\frac{1}{n}{\mathrm{\&Sigma;}}_{i=1}^n\{d_i-\hat{b}{\&lsqb;-ln(1-F_i)\&rsqb;}^{(1/\hat{a})}\}$

$\stackrel{\&OverBar;}{X}=\frac{1}{n}\left({\mathrm{\&Sigma;}}_{i=1}^n{X}_i\right)$

$\stackrel{\&OverBar;}{Y}=\frac{1}{n}\left({\mathrm{\&Sigma;}}_{i=1}^n{Y}_i\right)$

Wherein, cov (X, Y) is the covariance of vector X and Y, and D (X) is the variance of vector X；

C4, whenTime, obtain the final estimated result of three parameters, otherwise update location parameter c, orderRepeat Step C2-C3, until iteration is to meeting the condition of convergence

D, according to Weir distribution three parameters estimated value, determine electricity exception data marginal value；

Abnormal data marginal value t on the upside of it⁺For:

Abnormal data marginal value t on the downside of it_-For:Wherein θ is abnormal data probability；

E, as electric power data t_jMeet t_-<t_j<t⁺Time, it is determined that electric power data t_jFor normal data, when meeting t_j<t_-Or t_j>t⁺Time, Judge electric power data t_jFor abnormal data.