CN105205736A - Rapid detection method for power load abnormal data based on empirical mode decomposition - Google Patents

Abstract

The invention provides a rapid detection method for power load abnormal data based on empirical mode decomposition for solving the problems that in the prior art, results obtained by developing abnormal data detection and mining are few and are scattered, and no universal algorithm for abnormal data detection exists due to varieties and complexity of abnormal data. Accordingly, the detection method for power load abnormal data is high in accuracy and operability, and rapid detection and mining of the power load abnormal data under a large sample are achieved.

Description

Based on the method for quick of the electric load abnormal data of empirical mode decomposition

Technical field

The invention belongs to electric load anomaly data detection technical field, particularly relate to the method for quick of abnormal data in a kind of electric load of large sample.

Background technology

Under normal circumstances, the load of electrical network presents certain regular trend.But when some is special, load curve there will be huge fluctuation, and the huge fluctuation on this load data is likely on sample and transform caused by error, also may indicate the drastic change of load.Therefore analyzing the drastic change of load data, excavating with prediction is a very important content, it can give management and running department in advance to an estimation, so that the management mode under formulating emergency episode and power cuts to limit consumption tagmeme table, to prevent mains breakdown and to disintegrate.

Carrying out anomaly data detection with excavation is more thorny work, the result of gained is few and loose, and because the diversity of abnormal data and complicacy cause the general-purpose algorithm that there is not anomaly data detection, need to propose effective detection method for dissimilar data.

Summary of the invention

The present invention is directed to prior art Problems existing and make improvement, namely technical matters to be solved by this invention is to provide the abnormal deviation data examination method of the electric load that a kind of accuracy is high, operability is high, realizes quick detection and the excavation of the electric load abnormal data under large sample.

Technical scheme of the present invention is: a kind of method for quick of the electric load abnormal data based on empirical mode decomposition, it is characterized in that, described method comprises the following steps:

Step 1: carry out data prediction to measured data, rejects the data incomplete in measured data, sets up training sample set.

Step 2: to set up sample data collection, structure normal distribution model, determines normal distribution model parameter.

Step 3: the fiducial interval calculating normal distribution in step 3 under given level of confidence, as the decision threshold of electric load abnormal data.

Step 4: empirical mode decomposition is carried out to electric load to be detected dredging, obtains the intrinsic subsequence of different Power system load datas.

Step 5: after the different intrinsic subsequence reconstruct utilizing step 4 to obtain, carry out the anomaly data detection of electric load.

The method for quick of described a kind of electric load abnormal data based on empirical mode decomposition, the mathematic(al) representation of normal distribution model is:

$f\left(y\right)=\frac{1}{\sqrt{2\mathrm{\π}}\mathrm{\σ}}\exp (-\frac{{(y-\mathrm{\μ})}^2}{2{\mathrm{\σ}}^2})$

Wherein f is the probability density function of normal distribution, and μ is location parameter, and σ is scale parameter.

The method for quick of described a kind of electric load abnormal data based on empirical mode decomposition, the method for the determination Parameters of Normal Distribution used is Maximum Likelihood Estimation Method.

The method for quick of described a kind of electric load abnormal data based on empirical mode decomposition, under the given confidence level of the calculating described in step 3, the method for the fiducial interval of normal distribution comprises:

Step 3.1: sample estimates standard deviation sigma, its mathematic(al) representation is:

$\mathrm{\σ}=\frac{S}{\sqrt{n}}$

Wherein S is population standard deviation, and n is number of samples.

Step 3.2: when the sample size extracted is enough large, according to central limit theorem, with thinking sample average approximation Normal Distribution.The computing formula of its Z statistic is:

$Z=\frac{\stackrel{\‾}{y}-\mathrm{\μ}}{\mathrm{\σ}/\sqrt{n}}$

Wherein for sample average, μ is population mean.

Step 3.3: under the confidence level of given 1-α, the fiducial interval of population mean μ is:

$\stackrel{\‾}{y}-Z_{\mathrm{\α}/2}\×\mathrm{\σ}\≤\mathrm{\μ}\≤\stackrel{\‾}{y}+Z_{\mathrm{\α}/2}\×\mathrm{\σ}$

Wherein Z _α/2value obtain by tabling look-up.

The method for quick of described a kind of electric load abnormal data based on empirical mode decomposition, treats detection signal use experience mode decomposition, decomposites the eigenmode subsequence C that n bandwidth increases successively from signal to be detected _i(t), i=1,2 ..., a n and residual signals r _n(t), and by the eigenmode subsequence of decomposing, data to be tested can be reconstructed $\hat{y}\left(t\right)={\Σ}_{i=1}^n{C}_i\left(t\right)+r_n\left(t\right),t=1,2,\mathrm{...},M.$

Treat Power system load data y (t) of judgement, t=1,2 ..., M, when the value of the electric load of the reconstruct of t when falling in fiducial interval required by step 4, then the Power system load data in this moment is normal value; Otherwise, when the value of the electric load of the reconstruct of t when exceeding fiducial interval required by step 4, then this moment Power system load data is exceptional value, now by rejecting the highest for the eigenmode subsequence Mid Frequency comprised in reconstruction signal, again detect, until reconstruction signal falls into fiducial interval, and disallowable eigenmode subsequence is the part that exceptional value departs from normal value, still can be used as normal data after correction and use.

Effect of the present invention is: the abnormal deviation data examination method based on empirical mode decomposition electric load data of the present invention, the abnormal deviation data examination method of the electric load that a kind of accuracy is high, operability is high can be provided, realize quick detection and the excavation of electric load abnormal data.

Accompanying drawing explanation

Fig. 1 is the electric load anomaly data detection process flow diagram based on empirical mode decomposition.

Fig. 2 is the intrinsic subsequence to the different scale obtained after embodiment use experience mode decomposition.

Fig. 3 is the result figure of the electric load anomaly data detection of the practical application test adopting the embodiment of this method to carry out.

Embodiment

Below in conjunction with accompanying drawing, preferred embodiment is elaborated.It is emphasized that following explanation is only exemplary, instead of in order to limit the scope of the invention and apply.

Fig. 1 is process flow diagram of the present invention.In Fig. 1, method provided by the invention comprises the following steps:

Step 1: data prediction, sets up training sample set and test sample book collection.

For selected embodiment data, first reject the data incomplete in collection data, by whole sample set according to the arrangement of time series order, set up initial sample set.Initial sample set is divided into training sample set and test sample book collection.Wherein, training sample set is used for foundation and the model parameter estimation of the anomaly data detection model of electric load, and test sample book collection is sample data section to be detected.

In the present embodiment, data acquisition intervals is half an hour, then the data volume of a day is 48 points.Initial sample set is that in system, all image data obtain through rejecting data incomplete, and test sample book collection is 48 × 7 days=336 points.

Step 2: to sample sequence x (t), t=1,2 ..., N sets up normal distribution model.

Step 2.1: the population mean of hypothetical sequence is μ, then former sequence x (t), t=1,2 ..., N can be analyzed to average μ and error e (t) sum:

x(t)＝μ+e(t),t＝1,2,…,N

Step 2.2: use maximum-likelihood method to estimate population average μ, make the maximum probability that x (t) occurs exactly, namely make likelihood function maximum:

$\begin{array}{c}L\left(\mathrm{\μ}\right)={\Π}_{i=1}^{N}f\left(x_1\right)\mathrm{...}f\left(x_N\right)={\Π}_{t=1}^N\frac{1}{\sqrt{2\mathrm{\π}}\mathrm{\σ}}{e}^{-\frac{{\left(x\right(t)-\mathrm{\μ})}^2}{2{\mathrm{\σ}}^2}}\\ ={\left(\frac{1}{\sqrt{2\mathrm{\π}}\mathrm{\σ}}\right)}^n{e}^{-{\Σ}_{t=1}^n\frac{{\left(x\right(t)-\mathrm{\μ})}^2}{2{\mathrm{\σ}}^2}}\end{array}$

Logarithm is asked to L (μ), has

$\ln L=-\frac{n}{2}\ln \left(2\mathrm{\π}\right)-\frac{n}{2}\ln \left({\mathrm{\σ}}^2\right)-\frac{1}{2{\mathrm{\σ}}^2}{\Σ}_{t=1}^n{\left(x\right(t)-\mathrm{\μ})}^2$

Ask lnL to the partial derivative of μ, and make it equal 0:

$\frac{\∂\ln L}{\∂\mathrm{\μ}}=\frac{1}{{\mathrm{\σ}}^2}\left(\underset{t=1}{\overset{N}{\Σ}}{x}^2\right(t)-n\mathrm{\μ})=0$

The estimator then obtaining population average is

$\widehat{\mathrm{\μ}}=\frac{1}{n}\underset{t=1}{\overset{N}{\Σ}}x\left(t\right)$

Step 2.3: estimate population sample variances sigma ², mathematic(al) representation is as follows:

Similarly, ask lnL to σ ²partial derivative, and make it equal 0:

$\frac{\∂\ln L}{\∂{\mathrm{\σ}}^2}=\frac{n}{2{\mathrm{\σ}}^2}+\frac{1}{{\left(2{\mathrm{\σ}}^2\right)}^2}{\Σ}_{t=1}^n{\left(x\right(t)-\mathrm{\μ})}^2)=0$

Thus, σ ²be estimated as:

${\widehat{\mathrm{\σ}}}^2=\frac{1}{n}\underset{t=1}{\overset{N}{\Σ}}{\left(x\right(t)-\stackrel{\‾}{x})}^2=\frac{n-1}{n}{s}^2$

Wherein, s ²for sample variance.

Step 2.4: initiation sequence x (t), t=1,2,3 ..., N sets up normal distribution model and is

In the present embodiment, the normal distribution model of trying to achieve is N (9115.66,1895.2 ²).

Step 3: given confidence level 1-α, the fiducial interval that in calculation procedure 3, required normal distribution is corresponding.

Step 3.1: by step 2, supposes then have

$\frac{\stackrel{\‾}{x}-\mathrm{\μ}}{\mathrm{\σ}/\sqrt{n}}~N(0,1)$

Step 3.2: according to given confidence level 1-α, table look-up to obtain z _α/2with-z _α/2value.

Step 3.3: by the definition of Normal probability distribution, can obtain following mathematic(al) representation

$P\left(\right|\frac{\stackrel{\‾}{x}-\mathrm{\μ}}{\mathrm{\σ}/\sqrt{n}}|\≥z_{\mathrm{\α}/2})=\mathrm{\α}$

Then have

$P(\stackrel{\‾}{x}-z_{\mathrm{\α}/2}\×\frac{\mathrm{\σ}}{\sqrt{n}}\≤\mathrm{\μ}\≤\stackrel{\‾}{x}+z_{\mathrm{\α}/2}\×\frac{\mathrm{\σ}}{\sqrt{n}})=1-\mathrm{\α}$

Random interval be the fiducial interval that confidence level is 1-α.In the present embodiment, choose 1-α=0.95, try to achieve the fiducial interval of gained normal distribution in step 3 for (5401.1,12830).

Step 4: carry out empirical mode decomposition to test sample book collection, obtains the intrinsic subsequence of different scale.

Suppose that cycle tests is y (t), t=1,2 ..., M, decomposes the intrinsic subsequence C obtaining a different scale _i(t), i=1,2 ..., a n and residual signals r _n(t).

Therefore, the reconstruct to testing data can be obtained

Step 5: the anomaly data detection of test sample book collection.By step 4, after empirical mode decomposition is carried out to test sample book collection, obtain being reconstructed into test sample book collection to be detected when the value of the electric load of the reconstruct of t when falling in fiducial interval required by step 4, then the Power system load data in this moment is normal value; Otherwise, when the value of the electric load of the reconstruct of t when exceeding fiducial interval required by step 4, then this moment Power system load data is exceptional value, now by rejecting the highest for the eigenmode subsequence Mid Frequency comprised in reconstruction signal, again detect, until reconstruction signal falls into fiducial interval, and disallowable eigenmode subsequence is the part that exceptional value departs from normal value, still can be used as normal data after correction and use.

In the present embodiment, test sample book collection to be detected is chosen for 48 × 7 days=336 points, and the anomaly data detection result obtained as shown in Figure 3.

Claims (6)

1. based on a method for quick for the electric load abnormal data of empirical mode decomposition, it is characterized in that, described method comprises the following steps:

Step 1: carry out data prediction to measured data, rejects the data incomplete in measured data, sets up training sample set;

Step 2: to set up sample data collection, structure normal distribution model, determines normal distribution model parameter;

Step 3: the fiducial interval calculating normal distribution in step 3 under given level of confidence, as the decision threshold of electric load abnormal data;

Step 4: empirical mode decomposition is carried out to electric load to be detected dredging, obtains the intrinsic subsequence of different Power system load datas;

Step 5: after the different intrinsic subsequence reconstruct utilizing step 4 to obtain, carry out the anomaly data detection of electric load.

2. the method for quick of a kind of electric load abnormal data based on empirical mode decomposition according to claim 1, it is characterized in that, the mathematic(al) representation of described normal distribution model is:

$f\left(y\right)=\frac{1}{\sqrt{2\mathrm{\π}}\mathrm{\σ}}\exp (-\frac{{(y-\mathrm{\μ})}^2}{2{\mathrm{\σ}}^2})$

Wherein f is the probability density function of normal distribution, and μ is location parameter, and σ is scale parameter.

3. the method for quick of a kind of electric load abnormal data based on empirical mode decomposition according to claim 1, is characterized in that, describedly determines that the method for Parameters of Normal Distribution is Maximum Likelihood Estimation Method.

4. the method for quick of a kind of electric load abnormal data based on empirical mode decomposition according to claim 1, it is characterized in that, under the given confidence level of the calculating described in step 3, the method for the fiducial interval of normal distribution comprises:

Step 4.1: sample estimates standard deviation sigma, its mathematic(al) representation is:

$\mathrm{\σ}=\frac{S}{\sqrt{n}}$

Wherein S is population standard deviation, and n is number of samples;

Step 4.2: when the sample size extracted is enough large, according to central limit theorem, with thinking sample average approximation Normal Distribution; The computing formula of its Z statistic is:

$Z=\frac{\stackrel{\‾}{x}-\mathrm{\μ}}{\mathrm{\σ}/\sqrt{n}}$

Wherein for sample average, μ is population mean;

Step 4.3: under the confidence level of given 1-α, the fiducial interval of population mean μ is:

$\stackrel{\‾}{x}-Z_{\mathrm{\α}/2}\×\mathrm{\σ}\≤\mathrm{\μ}\≤\stackrel{\‾}{x}+Z_{\mathrm{\α}/2}\×\mathrm{\σ}$

Wherein Z _α/2with-Z _α/2value obtain by tabling look-up.

5. the method for quick of a kind of electric load abnormal data based on empirical mode decomposition according to claim 1, it is characterized in that, treat detection signal use experience mode decomposition, from signal to be detected, decomposite the eigenmode subsequence C that n bandwidth increases successively _i(t), i=1,2 ..., a n and residual signals r _n(t), and by the eigenmode subsequence of decomposing, data to be tested can be reconstructed $\hat{y}\left(t\right)={\mathrm{\Σ}}_{i=1}^n{C}_i\left(t\right)+$ $r_n\left(t\right),t=1,2,...,M.$

6. the method for quick of a kind of electric load abnormal data based on empirical mode decomposition according to claim 1, is characterized in that, treat Power system load data y (t) of judgement, t=1,2 ... M, when the value of the electric load of the reconstruct of t when falling in fiducial interval required by step 4, then the Power system load data in this moment is normal value; Otherwise, when the value of the electric load of the reconstruct of t when exceeding fiducial interval required by step 4, then this moment Power system load data is exceptional value, now by rejecting the highest for the eigenmode subsequence Mid Frequency comprised in reconstruction signal, again detect, until reconstruction signal falls into fiducial interval, and disallowable eigenmode subsequence is the part that exceptional value departs from normal value, still can be used as normal data after correction and use.