CN106548270A - A kind of photovoltaic plant power anomalous data identification method and device - Google Patents

Abstract

The present invention relates to a kind of photovoltaic plant power anomalous data identification method and device, the method, is clustered to power curve with fuzzy clustering algorithm with reference to the various factors of photovoltaic plant power data with photovoltaic plant active power data as object of study；Obvious abnormal data is recognized according to photovoltaic plant abnormal data criterion.Using the data set training photovoltaic power forecast model after rejecting abnormalities data, the precision of prediction and predictive efficiency of photovoltaic plant can be effectively improved, with extensive engineering application value.

Description

A kind of photovoltaic plant power anomalous data identification method and device

Technical field

The invention belongs to new forms of energy control technology field, and in particular to a kind of photovoltaic plant power anomalous data identification method And device.

Background technology

Collection in worksite photovoltaic power data are the bases of the work such as photovoltaic power generation quantity analytical calculation, power prediction, but by It is numerous in the reason for producing photovoltaic power abnormal data, such as communication abnormality, equipment fault and artificially ration the power supply etc., cause many light The power data of overhead utility collection in worksite is second-rate, and these abnormal datas can be to extracting photovoltaic power and irradiance, temperature etc. Between factor, true rule causes to have a strong impact on, directly can be substantially reduced using abnormal data the accuracy of photovoltaic power prediction with Effectiveness, also can have a negative impact to photovoltaic plant operational management and dispatching of power netwoks.

The content of the invention

It is an object of the invention to provide a kind of photovoltaic plant power anomalous data identification method and device, to realize to photovoltaic Power abnormal data is recognized, and improves the precision of power prediction.

In order to solve above-mentioned technical problem, the present invention provides a kind of photovoltaic plant power anomalous data identification method, including Six method schemes：

Method scheme one, comprises the steps：

1) day power data is carried out into pretreatment, day power curve is calculated with reference to the influence factor of photovoltaic plant power data The degree of association；

2) according to different affecting factors day power curve the degree of association, day power curve is gathered using clustering algorithm Class process, obtains the data classification results under different clusters numbers, that is, obtains the characteristic curve under different affecting factors；

3) according to the abnormal data criterion of photovoltaic plant power data, to the exception that there is obvious characteristic in day power Data are recognized, and the abnormal data criterion is：

A) photovoltaic power value is higher than characteristic curve value in the lasting setting time, and is not changed with irradiance；

B) photovoltaic power is less than characteristic curve value in the lasting setting time, and does not change with irradiance；

C) solar global irradiance is not substantially 0 in the lasting setting time, and photovoltaic power remains 0 or is close to 0；

If there is any of the above-described situation, it is judged to abnormal data.

Method scheme two, on the basis of method scheme one, also includes the day power data is sentenced according to abnormal data After other criterion judges, to judging not for abnormal data, the step of carry out secondary identification extremely using longitudinal method and horizontal method.

Method scheme three, on the basis of method scheme one, the calculating correlation method is gray prediction degree of association side Method, comprises the steps：

The influence factor of power data is converted into into day character vector, as the sample for calculating grey relational grade, each point Correlation coefficient is：

Wherein, ξ (k) is sequence x₀With x_iIn the grey incidence coefficient of k points, Respectively 2 grades minimum extreme differences and maximum extreme difference, ρ is resolution ratio, between 0 and 1, takes 0.5；

Ask for the meansigma methodss of each coefficient of association：

Method scheme four, on the basis of method scheme one or method scheme three, the shadow of the photovoltaic plant power data The factor of sound includes：

Weather pattern：

Season：

The highest temperature：

Lowest temperature：

2nd, 8,14,20 when temperature：t_k=t_Moment/t_max

The influence factor is converted into into day character vector (d, s, t_max,t_min,t₂,t₈,t₁₄,t₂₀) close as calculating Lycoperdon polymorphum Vitt The sample of connection degree.

Method scheme five, on the basis of method scheme one, the clustering algorithm be Fuzzy Mean Clustering Algorithm, power sample This space is X={ x₁,x₂,…,x_n(numbers of the n for input sample), comprise the steps：

S1, given cluster classification number c (2≤c≤n), set iteration stopping threshold values ε, as needed if initialized cluster Prototype pattern is P⁽⁰⁾, iteration count b=0；

S2, according to the meansigma methodss Matrix dividing U of coefficient of association^(b)：For any k (k=1,2 ..., n), i (i=1, 2 ..., c), ifThen have：

Wherein, m ＞ 1 (typically taking m=2) are referred to as fuzzy coefficient, d_ikRepresent the typical sample of the sample and the i-th class of kth apoplexy due to endogenous wind The degree of association between this center；

If there is i, k so thatThen haveAnd to j ≠ k,

S3, seek clustering prototype mode matrix P^(b+1)：

Wherein, u_ikThe Matrix dividing calculated according to coefficient of association by previous step, d_ikRepresent the sample and i-th of kth apoplexy due to endogenous wind The degree of association between the typical sample center of class；

S4, judgement | | P^(b)-P^(b+1)| | and ε relations：If | | P^(b)-P^(b+1)| | >=ε, then b=b+1, iteration process, Until | | P^(b)-P^(b+1)| | ＜ ε；Wherein, ε represents iteration stopping threshold values.

Method scheme six, on the basis of method scheme two, the horizontal method is for the work(between power curve s point Rate situation of change carries out disorder data recognition, comprises the steps：

Q1, obtain selected i-th point of power variation rate of sample day：

Δl_i=(l_i-l_i-1)/l_i

Wherein, Δ l_iFor the power variation rate between 2 continuous power points, i counts for power, and i=1,2 ..., s；

N days mean power rates of change in the same time before Q2, calculating：

Wherein, Δ l_i,avFor the mean power rate of change of n sample data, i counts for power, and i=1,2 ..., s；

If Q3, | Δ l_i|≥kΔl_i,av, it is judged as abnormal data, wherein, k is chugging coefficient；

The longitudinal method is to choose the power data of n days as sample, by the performance number of synchronization in day power curve Contrasted, comprised the steps：

P1, the day power data of n days s points is considered as the amount of being open into for s points, longitudinal direction amount is the array of n, obtains at per Expect

And variance

P2, the deviation ratio for obtaining each numerical value in two-dimensional array

Threshold values λ of the side-play amount with setting is compared, if deviation ratio is more than λ, is judged to abnormal data, wherein σ_jFor The variance of jth point.

The present invention also provides a kind of photovoltaic plant power anomalous data identification device, including six device schemes：

Device scheme one, including such as lower unit：

1) for day power data is carried out pretreatment, day power is calculated with reference to the influence factor of photovoltaic plant power data The unit of the degree of association of curve；

2) for according to different affecting factors day power curve the degree of association, day power curve is entered using clustering algorithm Row clustering processing, obtains the unit of the data classification results under different clusters numbers, that is, obtains the feature under different affecting factors Curve；

3) for the abnormal data criterion according to photovoltaic plant power data, to having obvious characteristic in day power The unit recognized by abnormal data, the abnormal data criterion is：

A) photovoltaic power value is higher than characteristic curve value in the lasting setting time, and is not changed with irradiance；

B) photovoltaic power is less than characteristic curve value in the lasting setting time, and does not change with irradiance；

C) solar global irradiance is not substantially 0 in the lasting setting time, and photovoltaic power remains 0 or is close to 0；

If there is any of the above-described situation, it is judged to abnormal data.

Device scheme two, on the basis of device scheme one, also includes the day power data is sentenced according to abnormal data After other criterion judges, to judging data not as exception, the unit of secondary abnormal identification is carried out using longitudinal method and horizontal method.

Device scheme three, on the basis of device scheme one, the calculating correlation method is gray prediction degree of association side Method, including such as lower module：

For the influence factor of power data is converted into day character vector, as the mould of the sample for calculating grey relational grade Block, the correlation coefficient of each point is：

Wherein, ξ (k) is sequence x₀With x_iIn the grey incidence coefficient of k points, Respectively 2 grades minimum extreme differences and maximum extreme difference, ρ is resolution ratio, between 0 and 1, takes 0.5；

For asking for the module of the meansigma methodss of each coefficient of association：

Device scheme four, on the basis of device scheme one or device scheme three, the shadow of the photovoltaic plant power data The factor of sound includes：

Weather pattern：

Season：

The highest temperature：

Lowest temperature：

2nd, 8,14,20 when temperature：t_k=t_Moment/t_max

The influence factor is converted into into day character vector (d, s, t_max,t_min,t₂,t₈,t₁₄,t₂₀) close as calculating Lycoperdon polymorphum Vitt The sample of connection degree.

Device scheme five, on the basis of device scheme one, the clustering algorithm be Fuzzy Mean Clustering Algorithm, power sample This space is X={ x₁,x₂,…,x_n(numbers of the n for input sample), including such as lower module：

S1, for given cluster classification number c (2≤c≤n), set iteration stopping threshold values ε as needed, if initialized Clustering prototype pattern is P⁽⁰⁾, the module of iteration count b=0；

S2, for the meansigma methodss Matrix dividing U according to coefficient of association^(b)Module：For any k (k=1,2 ..., n), i (i=1,2 ..., c), ifThen have：

Wherein, m ＞ 1 (typically taking m=2) are referred to as fuzzy coefficient, d_ikRepresent the typical sample of the sample and the i-th class of kth apoplexy due to endogenous wind The degree of association between this center；

If there is i, k so thatThen haveAnd to j ≠ k,

S3, for seeking clustering prototype mode matrix P^(b+1)Module：

Wherein, u_ikThe Matrix dividing calculated according to coefficient of association by previous step, d_ikRepresent the sample and i-th of kth apoplexy due to endogenous wind The degree of association between the typical sample center of class；

S4, for judging | | P^(b)-P^(b+1) | | and the module of ε relations：If | | P^(b)-P^(b+1)| | >=ε, then b=b+1, repeats Iterative process, until | | P^(b)-P^(b+1)| | ＜ ε；Wherein, ε represents iteration stopping threshold values.

Device scheme six, on the basis of device scheme two, the horizontal method is for the work(between power curve s point Rate situation of change carries out disorder data recognition, including such as lower module：

Q1, the module for obtaining selected i-th point of power variation rate of sample day：

Δl_i=(l_i-l_i-1)/l_i

Wherein, Δ l_iFor the power variation rate between 2 continuous power points, i counts for power, and i=1,2 ..., s；

Q2, the module for calculating first n days mean power rates of change in the same time：

Wherein, Δ l_i,avFor the mean power rate of change of n sample data, i counts for power, and i=1,2 ..., s；

If Q3, for | Δ l_i|≥kΔl_i,av, it is judged as the module of abnormal data, wherein, k is chugging coefficient；

The longitudinal method is to choose the power data of n days as sample, by the performance number of synchronization in day power curve Contrasted, including such as lower module：

P1, for the day power data of n days s points is considered as the amount of being open into for s points, array of the longitudinal direction amount for n is obtained every The expectation of point

And variance

Module；

P2, the deviation ratio for obtaining each numerical value in two-dimensional array

The module that side-play amount is compared with the threshold values λ for setting, if deviation ratio is more than λ, is judged to abnormal data, its Middle σ_jFor the variance of jth point.

The invention has the beneficial effects as follows：The present invention passes through to consider photovoltaic plant site environment, with reference to photovoltaic plant power number According to influence factor power curve is clustered, improve photovoltaic plant anomalous data identification accuracy and reliability；Consider There is very strong randomness and dispersibility to photovoltaic power data, and this feature will be calculated with power-generation analysis to power prediction and be produced It is raw to affect, thus according to this feature, with reference to the abnormal data criterion that live practical experience is summarized, can be bright to major part Aobvious abnormal data is recognized, and improves accuracy and reliability to power anomalous data identification, improves the essence of power prediction Degree.

Description of the drawings

Fig. 1 is the flow chart of the photovoltaic plant power anomalous data identification method of the present invention.

Specific embodiment

Below in conjunction with the accompanying drawings, the present invention is further described in detail.

The photovoltaic plant power anomalous data identification method flow diagram of the present invention is illustrated in figure 1, specifically：

1) day power data is normalized into pretreatment.

2) influence factor of photovoltaic plant power data, including weather pattern, temperature and season are considered：

Weather pattern：

Season：

The highest temperature：

Lowest temperature：

2nd, 8,14,20 when temperature：t_k=t_Moment/t_max

With reference to the influence factor of above-mentioned photovoltaic plant power data, day power curve is calculated according to Grey Correlation Method The degree of association：

Wherein, ξ (k) is sequence x₀With x_iIn the grey incidence coefficient of k points, Respectively 2 grades minimum extreme differences and maximum extreme difference, ρ is resolution ratio, between 0 and 1, takes 0.5.

The influence factor of power data is converted into into day character vector (d, s, t_max,t_min,t₂,t₈,t₁₄,t₂₀), as meter Calculate the sample of grey relational grade.

Ask for the meansigma methodss of each coefficient of association：

3) according to different affecting factors day power curve the degree of association, it is bent to day power using Fuzzy Mean Clustering Algorithm Line carries out clustering processing, obtains the data classification results under different clusters numbers, and power sample space is X={ x₁,x₂,…,x_n} (numbers of the n for input sample), specifically：

S1, given cluster classification number c (2≤c≤n), set iteration stopping threshold values ε, as needed if initialized cluster Prototype pattern is P⁽⁰⁾, iteration count b=0；

S2, according to the meansigma methodss Matrix dividing U of coefficient of association^(b)：For any k (k=1,2 ..., n), i (i=1, 2 ..., c), ifThen have：

Wherein, m ＞ 1 (typically taking m=2) are referred to as fuzzy coefficient, d_ikRepresent the typical sample of the sample and the i-th class of kth apoplexy due to endogenous wind The degree of association between this center；

If there is i, k so thatThen haveAnd to j ≠ k,

S3, seek clustering prototype mode matrix P^(b+1)：

Wherein, u_ikThe Matrix dividing calculated according to coefficient of association by previous step, d_ikRepresent the sample and i-th of kth apoplexy due to endogenous wind The degree of association between the typical sample center of class；

S4, judgement | | P^(b)-P^(b+1)| | and ε relations：If | | P^(b)-P^(b+1)| | >=ε, then b=b+1, iteration process, Until | | P^(b)-P^(b+1)| | ＜ ε；Wherein, ε represents iteration stopping threshold values.

4) according to the abnormal data criterion for photovoltaic plant power data, to having obvious characteristic in day power Abnormal data is recognized, and the abnormal data criterion is：

A, photovoltaic power value is higher than characteristic curve value in the lasting setting time, and is not changed with irradiance, mainly Producing cause is communication or measuring apparatus failure, causes photovoltaic power data record to remain at larger under high irradiance Value；

B, photovoltaic power is less than characteristic curve value in the lasting setting time, and does not change with irradiance, produces former Because rationing the power supply including photovoltaic, communicating or measuring apparatus failure, the record value of photovoltaic power is caused to be always held at compared with low irradiance Magnitude of power on；

C, solar global irradiance is not substantially 0 in the lasting setting time, and photovoltaic power remains 0 or is close to 0, and producing cause is Communication, measuring apparatus failure or photovoltaic module failure, cause data record to remain 0 or close to 0.

If there is any of the above-described situation, it is judged to abnormal data.

In the present embodiment, clustering processing is carried out to day power curve using Fuzzy Mean Clustering Algorithm, obtains different poly- Data classification results under class number.As other embodiment, other clustering algorithms can also be adopted, to realize obtaining difference The purpose of the data classification results under clusters number.

In the present embodiment, the influence factor of photovoltaic plant power data have selected weather pattern, temperature and season.As Other embodiment, can select photovoltaic plant power influence factor according to practical situation, can also accordingly increase or decrease impact Factor, can also increase atmospheric pressure, wind speed etc..

In the present embodiment, power curve is clustered with reference to the influence factor of photovoltaic plant power data, it is contemplated that Photovoltaic power data have very strong randomness and dispersibility, and this feature will be calculated with power-generation analysis to power prediction and be produced Affect, according to this feature, with reference to the abnormal data criterion that live practical experience is summarized, can be to most of significantly different Regular data is recognized.As other embodiment, in the above-mentioned abnormal data criterion summarized according to live practical experience After recognizing to most of obvious abnormal data, while two are carried out to abnormal data according to the feature of photovoltaic power curve Secondary identification, preferably to improve accuracy.

Specifically, secondary identification is carried out using longitudinal method and horizontal method to the abnormal data of day power curve.

The horizontal method is to carry out disorder data recognition for the changed power situation between 96 points of power curve, according to Between each point, power variation rate is identified to abnormal data, is comprised the steps：

Q1, obtain selected i-th point of power variation rate of sample day：

Δl_i=(l_i-l_i-1)/l_i

Wherein, Δ l_iFor the power variation rate between 2 continuous power points, i counts for power, and i=1, and 2 ..., 96；

N days mean power rates of change in the same time before Q2, calculating：

Wherein, Δ l_i,avFor the mean power rate of change of n sample data, i counts for power, and i=1, and 2 ..., 96；

If Q3, | Δ l_i|≥kΔl_i,av, it is judged as abnormal data, wherein, k is chugging coefficient.

The longitudinal method is to choose the power data of n days as sample, by the performance number of synchronization in day power curve Contrasted, comprised the steps：

P1, the day power data of n days 96 points is considered as the amount of being open into for 96 points, longitudinal direction amount is the array of n, is obtained per point Expectation

And variance

P2, the deviation ratio for obtaining each numerical value in two-dimensional array

It is compared with the threshold values λ of setting, if deviation ratio is more than λ, is judged to abnormal data, wherein σ_jFor jth point Variance.

The present invention also provides a kind of photovoltaic plant power anomalous data identification device, including such as lower unit：

1) for day power data is carried out pretreatment, day power is calculated with reference to the influence factor of photovoltaic plant power data The unit of the degree of association of curve；

2) for according to different affecting factors day power curve the degree of association, day power curve is entered using clustering algorithm Row clustering processing, obtains the unit of the data classification results under different clusters numbers, that is, obtains the feature under different affecting factors Curve；

3) for the abnormal data criterion according to photovoltaic plant power data, to having obvious characteristic in day power The unit recognized by abnormal data, the abnormal data criterion is：

A) photovoltaic power value is higher than characteristic curve value in the lasting setting time, and is not changed with irradiance；

B) photovoltaic power is less than characteristic curve value in the lasting setting time, and does not change with irradiance；

C in the lasting setting time it is not substantially) 0 in solar global irradiance, photovoltaic power remains 0 or is close to 0；

If there is any of the above-described situation, it is judged to abnormal data.

The device is actually based on a kind of computer solution of the inventive method flow process, i.e., a kind of software architecture, Above-mentioned each unit is each treatment progress corresponding with method flow or program.Due to the introduction to said method it is enough It is clear complete, therefore the device is no longer described in detail.

Claims (10)

1. a kind of photovoltaic plant power anomalous data identification method, it is characterised in that comprise the steps：

1) day power data is carried out into pretreatment, the pass of day power curve is calculated with reference to the influence factor of photovoltaic plant power data Connection degree；

2) according to different affecting factors day power curve the degree of association, day power curve is carried out at cluster using clustering algorithm Reason, obtains the data classification results under different clusters numbers, that is, obtains the characteristic curve under different affecting factors；

3) according to the abnormal data criterion of photovoltaic plant power data, to the abnormal data that there is obvious characteristic in day power Recognized, the abnormal data criterion is：

A) photovoltaic power value is higher than characteristic curve value in the lasting setting time, and is not changed with irradiance；

B) photovoltaic power is less than characteristic curve value in the lasting setting time, and does not change with irradiance；

C) solar global irradiance is not substantially 0 in the lasting setting time, and photovoltaic power remains 0 or is close to 0；

If there is any of the above-described situation, it is judged to abnormal data.

2. photovoltaic plant power anomalous data identification method according to claim 1, it is characterised in that also include to described After day power data foundation abnormal data criterion judgement, to judging data not as exception, using longitudinal method and horizontal method The step of carrying out secondary abnormal identification.

3. photovoltaic plant power anomalous data identification method according to claim 1, it is characterised in that the calculating association Degree method is gray prediction degree of association method, is comprised the steps：

The influence factor of power data is converted into into day character vector, as the sample for calculating grey relational grade, the correlation of each point Coefficient is：

$\mathrm{\ξ}\left(k\right)=\frac{\underset{i}{min}\underset{k}{min}|x_0\left(k\right)-x_i\left(k\right)|+\mathrm{\ρ}\underset{i}{max}\underset{k}{max}|x_0\left(k\right)-x_i\left(k\right)|}{|x_0\left(k\right)-x_i\left(k\right)|+\mathrm{\ρ}\underset{i}{max}\underset{k}{max}|x_0\left(k\right)-x_i\left(k\right)|}$

Wherein, ξ (k) is sequence x₀With x_iIn the grey incidence coefficient of k points, Respectively 2 grades minimum extreme differences and maximum extreme difference, ρ is resolution ratio, between 0 and 1, takes 0.5；

Ask for the meansigma methodss of each coefficient of association：

$R_i=\frac{1}{N}\underset{k=1}{\overset{N}{\Σ}}{\mathrm{\ξ}}_i\left(k\right)$

4. the photovoltaic plant power anomalous data identification method according to claim 1 or 3, it is characterised in that the photovoltaic The influence factor of power station power data includes：

Weather pattern：

Season：

The highest temperature：

Lowest temperature：

2nd, 8,14,20 when temperature：t_k=t_Moment/t_max

The influence factor is converted into into day character vector (d, s, t_max,t_min,t₂,t₈,t₁₄,t₂₀) as calculating grey relational grade Sample.

5. photovoltaic plant power anomalous data identification method according to claim 1, it is characterised in that the clustering algorithm For Fuzzy Mean Clustering Algorithm, power sample space is X={ x₁,x₂,…,x_n(numbers of the n for input sample), including as follows Step：

S1, given cluster classification number c (2≤c≤n), set iteration stopping threshold values ε, as needed if initialized clustering prototype Pattern is P⁽⁰⁾, iteration count b=0；

S2, according to the meansigma methodss Matrix dividing U of coefficient of association^(b)：For any k (k=1,2 ..., n), i (i=1,2 ..., c), IfThen have：

$u_{ik}^{\left(b\right)}={\left\{\underset{j=1}{\overset{c}{\Σ}}\[{\[\frac{d_{ik}^{\left(b\right)}}{d_{jk}^{\left(b\right)}}\]}^{\frac{2}{m-1}}\]\right\}}^{-1}$

Wherein, m ＞ 1 (typically taking m=2) are referred to as fuzzy coefficient, d_ikIn representing the typical sample of sample and the i-th class of kth apoplexy due to endogenous wind The degree of association between the heart；

If there is i, k so thatThen haveAnd to j ≠ k,

S3, seek clustering prototype mode matrix P^(b+1)：

$P_i^{b+1}=\frac{\underset{k=1}{\overset{n}{\Σ}}{\left(u_{ik}{d}_{ik}^{\left(b\right)}\right)}^{m}xk}{\underset{k=1}{\overset{n}{\Σ}}{\left(u_{ik}{d}_{ik}^{\left(b\right)}\right)}^m},i=1,2,...,c$

Wherein, u_ikThe Matrix dividing calculated according to coefficient of association by previous step, d_ikRepresent the sample and the i-th class of kth apoplexy due to endogenous wind The degree of association between typical sample center；

S4, judgement | | P^(b)-P^(b+1)| | and ε relations：If | | P^(b)-P^(b+1)| | >=ε, then b=b+1, iteration process, until | | P^(b)-P^(b+1)| | ＜ ε；Wherein, ε represents iteration stopping threshold values.

6. photovoltaic plant power anomalous data identification method according to claim 2, it is characterised in that the horizontal method is Disorder data recognition is carried out for the changed power situation between power curve s point, is comprised the steps：

Q1, obtain selected i-th point of power variation rate of sample day：

Δl_i=(l_i-l_i-1)/l_i

Wherein, Δ l_iFor the power variation rate between 2 continuous power points, i counts for power, and i=1,2 ..., s；

N days mean power rates of change in the same time before Q2, calculating：

${\mathrm{\Δl}}_{i,av}=\frac{1}{n}\underset{j=1}{\overset{n}{\Σ}}{\mathrm{\Δl}}_{ij}$

Wherein, Δ l_i,avFor the mean power rate of change of n sample data, i counts for power, and i=1,2 ..., s；

If Q3, | Δ l_i|≥kΔl_i,av, it is judged as abnormal data, wherein, k is chugging coefficient；

The longitudinal method is to choose the power data of n days as sample, and the performance number of synchronization in day power curve is carried out Contrast, comprises the steps：

P1, the day power data of n days s points is considered as the amount of being open into for s points, longitudinal direction amount is the array of n, obtains the expectation of per

$E\left(j\right)=\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}x(i,j)$

And variance

$\mathrm{\σ}(i,j)=\underset{i=1}{\overset{n}{\Σ}}\[x(i,j)-E\left(j\right){\]}^2$

P2, the deviation ratio for obtaining each numerical value in two-dimensional array

$\mathrm{\ρ}(i,j)=\frac{|x(i,j)-E\left(j\right)|}{3{\mathrm{\σ}}_j}$

Threshold values λ of the side-play amount with setting is compared, if deviation ratio is more than λ, is judged to abnormal data, wherein σ_jFor jth point Variance.

7. a kind of photovoltaic plant power anomalous data identification device, it is characterised in that include such as lower unit：

1) for day power data is carried out pretreatment, day power curve is calculated with reference to the influence factor of photovoltaic plant power data The degree of association unit；

2) for according to different affecting factors day power curve the degree of association, day power curve is gathered using clustering algorithm Class process, obtains the unit of the data classification results under different clusters numbers, that is, obtains the characteristic curve under different affecting factors；

3) for the abnormal data criterion according to photovoltaic plant power data, to the exception that there is obvious characteristic in day power The unit recognized by data, the abnormal data criterion is：

A) photovoltaic power value is higher than characteristic curve value in the lasting setting time, and is not changed with irradiance；

B) photovoltaic power is less than characteristic curve value in the lasting setting time, and does not change with irradiance；

C) solar global irradiance is not substantially 0 in the lasting setting time, and photovoltaic power remains 0 or is close to 0；

If there is any of the above-described situation, it is judged to abnormal data.

8. photovoltaic plant power anomalous data identification device according to claim 7, it is characterised in that also include to described After day power data foundation abnormal data criterion judgement, to judging data not as exception, using longitudinal method and horizontal method Carry out the unit of secondary abnormal identification.

9. photovoltaic plant power anomalous data identification device according to claim 7, it is characterised in that the calculating association Degree method is gray prediction degree of association method, including such as lower module：

For the influence factor of power data is converted into day character vector, as the module of the sample for calculating grey relational grade, The correlation coefficient of each point is：

$\mathrm{\ξ}\left(k\right)=\frac{\underset{i}{min}\underset{k}{min}|x_0\left(k\right)-x_i\left(k\right)|+\mathrm{\ρ}\underset{i}{max}\underset{k}{max}|x_0\left(k\right)-x_i\left(k\right)|}{|x_0\left(k\right)-x_i\left(k\right)|+\mathrm{\ρ}\underset{i}{max}\underset{k}{max}|x_0\left(k\right)-x_i\left(k\right)|}$

Wherein, ξ (k) is sequence x₀With x_iIn the grey incidence coefficient of k points, Point Not Wei 2 grades of minimum extreme differences and maximum extreme difference, ρ is resolution ratio, between 0 and 1, takes 0.5；

For asking for the module of the meansigma methodss of each coefficient of association：

$R_i=\frac{1}{N}\underset{k=1}{\overset{N}{\Σ}}{\mathrm{\ξ}}_i\left(k\right)$

10. the photovoltaic plant power anomalous data identification device according to claim 7 or 9, it is characterised in that the photovoltaic The influence factor of power station power data includes：

Weather pattern：

Season：

The highest temperature：

Lowest temperature：

2nd, 8,14,20 when temperature：t_k=t_Moment/t_max

The influence factor is converted into into day character vector (d, s, t_max,t_min,t₂,t₈,t₁₄,t₂₀) as calculating grey relational grade Sample.