CN105184683A - Probability clustering method based on wind electric field operation data - Google Patents

Abstract

The invention discloses a probability clustering method based on wind electric field operation data. The method comprises steps of: acquiring the actually-measured data of a wind electric field; setting the number k of clusters according to the geographic position of the wind electric field and randomly selecting k samples as initial cluster centers; calculating the Euclidean distance between each sample of the actually-measured data and the each cluster center, selecting the minimum Euclidean distance corresponding to each sample, allocating each sample to the cluster to which the cluster center corresponding to the minimum Euclidean distance belongs in order to generate a clustering result; calculating the sum of square errors of the cluster according to the Euclidean distance; cyclically executing the above iterative computation process, comparing the sum of square errors of the cluster in two adjacent times, and stopping iterative computation when E-E' < [epsilon]; classifying the clustering result and computing the probability of each classification; selecting the clustering result of the classification with the maximum probability, evaluating the each clustering result in the clustering results of the classification by using a contour value function and using an evaluated result as a probability clustering final result.

Description

A kind of probability clustering grouping method based on wind power plant operation data

Technical field

The present invention relates to electric system simulation field, espespecially a kind of probability clustering grouping method based on wind power plant operation data.

Background technology

Current, along with the develop rapidly of wind power technology, the scale of integrated wind plant increases sharply.Because wind-power electricity generation has certain randomness, be therefore necessary that the operation characteristic after to large-scale wind power access electrical network carries out deep research, matter of utmost importance is exactly wind energy turbine set modeling.

Wind energy turbine set modeling can be divided into detailed modeling and equivalent modeling.Detailed modeling is the model setting up every typhoon group of motors, slip ring system in wind energy turbine set.Along with the progressively expansion of wind energy turbine set scale, if carry out detailed modeling to every Fans, significantly certainly will increase the complexity of wind energy turbine set model and calculate duration.Therefore, the means of equivalent modeling are adopted to be necessary.

Wind energy turbine set equivalent modeling is divided into unit equivalence and multimachine equivalence, and the key problem of multimachine equivalence is the validity of clustering algorithm.Current wind energy turbine set equivalent modeling research mainly concentrates in the method for unit equivalence.But, for the Large Scale Wind Farm Integration be made up of tens of Fans even up to a hundred, due to the impact of topography and geomorphology, wake effect and time lag, in wind energy turbine set, wind speed profile is uneven, cause that Wind turbines actual motion state is not identical even very big-difference, therefore there is comparatively big error in unit equivalence method usually.

Multimachine equivalence refers to and is rationally hived off by the Wind turbines in wind energy turbine set, and is represented with separate unit machine by all units in each group, and then sets up wind energy turbine set model by a small amount of a few typhoon group of motors.The method of current multimachine equivalence mainly concentrates on K-means clustering algorithm, but traditional K-means algorithm also exists the limitation of self, index of even hiving off (as wind speed) changes, and grouping result will change to some extent, therefore just needs again to hive off with K-means.This adds the complicacy of multimachine equivalence greatly, is unfavorable for engineer applied.

Summary of the invention

The present invention is directed to the problems referred to above, on the basis of traditional K-means clustering algorithm, propose a kind of probability clustering grouping method based on wind power plant operation data, the grouping result that this algorithm obtains can be applicable to the situation of various wind speed, for the use of later wind energy turbine set modeling provides great convenience.

The probability clustering grouping method based on wind power plant operation data that the present invention proposes comprises: step 1, obtains wind energy turbine set measured data within a certain period of time; Step 2, the geographic position according to wind energy turbine set sets the number k that hives off, and under initial situation, chooses k sample at random as initial cluster center in described measured data; Step 3, calculate each sample in described measured data and divide the Euclidean distance being clipped to each initial cluster center, choose the minimum Eustachian distance that each sample described is corresponding, and each sample described is put under in the group at the initial cluster center place corresponding with described minimum euclidean distance, generate grouping result first; Step 4, according to the Euclidean distance that step 3 calculates, calculates the square error summation obtaining cluster corresponding to grouping result first; Step 5, in non-initial situation, calculates acquisition k non-initial cluster centre according to the grouping result that last time generates; Step 6, calculate each sample in described measured data and divide the Euclidean distance being clipped to each non-initial cluster centre, choose the minimum Eustachian distance that each sample described is corresponding, and each sample described is put under in the group at the non-initial cluster centre place corresponding with described minimum euclidean distance, generate new grouping result; Step 7, according to the Euclidean distance that step 6 calculates, calculates the square error summation obtaining cluster corresponding to new grouping result; Step 8, circulation performs the iterative process of above-mentioned steps 5 to step 7, the relatively square error summation of the cluster of twice adjacent calculation acquisition, when meeting E-E'< ε, stop iterative computation, wherein, E, E' are respectively the cluster square error summation that twice adjacent calculation obtains, being worth large is E, and being worth little is E'; The all grouping result obtained are classified, are set the wind speed probability under fan operation operating mode corresponding to every component group result by step 9, calculate the probability obtaining each class; Step 10, chooses a class grouping result of maximum probability, utilizes profile value function to assess, as the net result that probability clustering hives off the every component group result in such grouping result.

The present invention propose the probability clustering grouping method based on wind power plant operation data with the actual operating data of wind speed for index, according to the principle that Euclidean distance is nearest, the probability clustering algorithm in conjunction with K-means algorithm is adopted to hive off, and test by the rationality of profile value function to grouping result, the most effective selected grouping result, this grouping result can be used for the ruuning situation of various wind speed, substantially increases the efficiency of work of hiving off.Utilize said method to obtain grouping result comparatively accurate, tally with the actual situation, have good effect to the blower fan irregular wind energy turbine set that distributes, for the use of later wind-field model provides great convenience, there is important engineer applied and be worth.

Accompanying drawing explanation

Accompanying drawing described herein is used to provide a further understanding of the present invention, forms a application's part, does not form limitation of the invention.In the accompanying drawings:

Fig. 1 is the probability clustering grouping method process flow diagram based on wind power plant operation data of one embodiment of the invention.

Fig. 2 is the probability distribution schematic diagram of each class of the present invention one specific embodiment.

Fig. 3 is the profile value schematic diagram of the category-A grouping result of the present invention one specific embodiment.

Fig. 4 is that the wind energy turbine set of the present invention one specific embodiment is hived off division schematic diagram.

Fig. 5 is that the multimachine Equivalent Model of the present invention one specific embodiment and detailed model active power contrast schematic diagram.

Fig. 6 is that the multimachine Equivalent Model of the present invention one specific embodiment and detailed model reactive power contrast schematic diagram.

Fig. 7 is multimachine Equivalent Model and the detailed model current vs schematic diagram of the present invention one specific embodiment.

Embodiment

Below coordinating preferred embodiment graphic and of the present invention, setting forth the technological means that the present invention takes for reaching predetermined goal of the invention further.

Fig. 1 is the probability clustering grouping method process flow diagram based on wind power plant operation data of one embodiment of the invention.As shown in Figure 1, the method comprises:

Step 1, obtains wind energy turbine set measured data within a certain period of time.

Such as, obtain the measured data of n typhoon group of motors in time period m can be expressed as:

$V=(V_1,V_2,...,V_j,...,V_m)={\left(v_{i,j}\right)}_{n\&times;m}=\left[\begin{array}{cccc}{v}_{11}& v_{12}& ...& v_{1m}\\ v_{21}& v_{22}& ...& v_{2m}\\ ...& ...& ...& ...\\ v_{n1}& v_{n2}& ...& v_{nm}\end{array}\right];$

Wherein, v _ijrepresent the measured data of the i-th typhoon group of motors in a jth moment wind speed index; Vectorial V _jregard a time samples group of SPACE V as, in V, the quantity of sample group is m, and dimension is n, V _jrepresent the vector of a jth moment all Wind turbines measured datas.

Step 2, the geographic position according to wind energy turbine set sets the number k (generalized case is set to 3) that hives off, and then obtains cluster centre.

Further, the determination of cluster centre is divided into initial situation (step 2-step 4) and non-initial situation (step 5-step 7).

Under initial situation, in described measured data, choose k sample at random as initial cluster center.Initial cluster center be expressed as:

$m_p^{\left(1\right)}=(v_{i1},v_{i2},...,v_{ij});$

Wherein, p=1,2 ..., k, k represent number of hiving off

Step 3, calculate each sample in described measured data and divide the Euclidean distance being clipped to each initial cluster center, choose the minimum Eustachian distance that each sample described is corresponding, and each sample described is put under in the group at the initial cluster center place corresponding with described minimum euclidean distance, generate grouping result first.

Wherein, each sample is to the Euclidean distance of the initial cluster center of each group described for:

$d(V_i,m_p^{\left(1\right)})=\sqrt{\underset{j=1}{\overset{m}{\&Sigma;}}{(v_{ij}-m_p^{\left(1\right)})}^2};$

Wherein, i represents the i-th typhoon group of motors, p=1,2 ..., k, k represent the number that hives off, represent p the initial cluster center hived off.

Step 4, according to the Euclidean distance that step 3 calculates, calculates the square error summation obtaining cluster corresponding to grouping result first, the square error summation E of cluster ⁽¹⁾for:

$E^{\left(1\right)}=\underset{p=1}{\overset{k}{\&Sigma;}}\underset{i=1}{\overset{n}{\&Sigma;}}{d}^2(V_i,m_p^{\left(1\right)});$

Wherein, i represents the i-th typhoon group of motors, p=1,2 ..., k, k represent the number that hives off, for the Euclidean distance that step 3 calculates.

It should be noted that, the square error summation (E of step 4 of cluster ⁽¹⁾and the E of step 7 ^(r)) be the square error summation of cluster corresponding to each grouping result, that is circulation correspondence at every turn can in the hope of a cluster square error summation.Have in the formula of step 4 and step 7 two add with nested, the meaning is exactly be introduced into some hiving off, and then correspond to the Euclidean distance (that is, minimum Eustachian distance) drawing and belong to the sample that this hives off, and then calculate square to add and, finally the value that k is hived off is gathered.

Step 5, in non-initial situation, calculates acquisition k non-initial cluster centre according to the grouping result that last time generates.

Specifically, in non-initial situation, according to the last time grouping result generated, calculate previous grouping result each hive off in the mean value of sample value that comprises, obtain k non-initial cluster centre, non-initial cluster centre be expressed as:

$m_p^{\left(r\right)}=\frac{1}{l_p^{\left(r-1\right)}}(\underset{V_t\&Element;G_D}{\&Sigma;}{V}_{i1},\underset{V_t\&Element;G_D}{\&Sigma;}{V}_{i2},\mathrm{...},\underset{V_t\&Element;G_D}{\&Sigma;}{V}_{im});$

Wherein, p=1,2 ..., k, k represent number of hiving off, and r represents the number of times of iteration, l _prepresent p the unit number hiving off contained, V _trepresent the some sample values in measured data, G _dto represent in grouping result some hives off.

It should be noted that, this formula is exactly the sample value averaged that comprises during last time circulation marked off each hives off, and is the cluster centre that this calculates grouping result.

Step 6, calculate each sample in described measured data and divide the Euclidean distance being clipped to each non-initial cluster centre, choose the minimum Eustachian distance that each sample described is corresponding, and each sample described is put under in the group at the non-initial cluster centre place corresponding with described minimum euclidean distance, generate new grouping result.

Wherein, each sample divides the Euclidean distance being clipped to each non-initial cluster centre for:

$d(V_i,m_p^{\left(r\right)})=\sqrt{\underset{j=1}{\overset{m}{\&Sigma;}}{(v_{ij}-m_p^{\left(r\right)})}^2};$

Wherein, i represents the i-th typhoon group of motors, p=1,2 ..., k, k represent the number that hives off, represent p the non-initial cluster centre hived off.

Step 7, according to the Euclidean distance that step 6 calculates, calculates the square error summation obtaining cluster corresponding to new grouping result, the square error summation E of cluster ^(r)for:

$E^{\left(r\right)}=\underset{p=1}{\overset{k}{\&Sigma;}}\underset{i=1}{\overset{n}{\&Sigma;}}{d}^2(V_i,m_p^{\left(r\right)});$

Wherein, i represents the i-th typhoon group of motors, p=1,2 ..., k, k represent the number that hives off, for the Euclidean distance 1 that step 6 calculates.

Step 8, circulation performs the iterative process of above-mentioned steps 5 to step 7, can calculate the square error summation of multiple cluster.After executing first time step 5-7, circulation performs second time step 5, and now, during execution second time step 5, grouping result used is the grouping result obtained in first time step 6.When performing third time step 5, the grouping result obtained in second time step 6 during used grouping result, calculates with this loop iteration.Step 2-4 is initial situation, and the step 5-7 that circulation performs is non-initial situation.

By comparing the square error summation of the cluster that twice adjacent calculation obtains, when meeting E-E'< ε, stop iterative computation, wherein, E, E' are respectively the cluster square error summation that twice adjacent calculation obtains, being worth large is E, is worth little to represent a very little positive number for E', ε.In this step, because the square error summation of cluster is the method for a judgement error calculated.And to hive off itself be the process of an iteration, therefore, a stable grouping result that what this programme was wanted to obtain is, using it as final result.So, when the value error that each loop iteration obtains enough little (namely there is similarity), can think that grouping result enough stabilizes.

The all grouping result obtained are classified, are set the wind speed probability under fan operation operating mode corresponding to every component group result by step 9, calculate the probability obtaining each class.

Concrete sorting technique is: in all grouping result obtained, and identical for identical for the number of hiving off comprised and the blower fan quantity dividing mode of hiving off is divided into a class grouping result.

Such as, after 15 Fans groupings, two component group results are wherein:

Group 1: group 1 (1#-6#), group 2 (7#-10#), group 3 (11#-15#);

Group 2: group 1 (7#-12#), group 2 (13#-15#, 1#), group 3 (2#-6#);

Wherein, 1#, 2# ..., 15# is the numbering of blower fan.

Can be found out by above-mentioned hiving off, the number of hiving off of these two groupings is identical, be all 3 groups, and the blower fan quantity organizing 1 each group is 6,4,5, the blower fan quantity organizing 2 each groups is 6,4,5, the blower fan quantity dividing mode of hiving off of each group is identical, then these two groups can be divided into same class grouping result.

Further, the wind speed probability P under fan operation operating mode corresponding to every component group result is set _{wind, i}, calculate the probability obtaining each class, the formula of utilization is as follows:

$P^{\left(X\right)}=\underset{i=1}{\overset{n\left(X\right)}{\&Sigma;}}{P}_{wind,i};$

Wherein, P ^(X)for the probability of each class, X is the numbering of class, and n (X) is for comprising the group number of grouping result, P _{wind, i}it is the wind speed probability under the i-th Fans operating condition.

Step 10, chooses a class grouping result of maximum probability, utilizes profile value function to assess, as the net result that probability clustering hives off the every component group result in such grouping result.

Wherein, the formula of utilization is as follows:

$S\left(i\right)=\frac{min\left(b\right)-a}{max\&lsqb;a,min\left(b\right)\&rsqb;},i=1,2,...,n;$

Wherein, a is the mean distance in sample i and place grouping result between same other sample hived off;

B is the mean distance between the sample i sample hived off different from the grouping result of place;

S (i) represents profile value, and span is that the value of-1 to 1, S (i) is larger, and the classification of sample i is more reasonable, as S (i) < 0, illustrates that scheme of hiving off is unreasonable;

In a class grouping result of the described maximum probability selected, choose profile value S (i) be all greater than 0 and closest to 1 a component group result, as the net result that probability clustering hives off.

In order to more clearly explain the above-mentioned probability clustering grouping method based on wind power plant operation data, be described below in conjunction with a specific embodiment, but it should be noted that this embodiment is only to better the present invention is described, do not form and the present invention is limited improperly.

For the actual operating data of China's wind energy turbine set, this wind energy turbine set installs the double-fed wind power generator group of 24 single-machine capacity 2.0MW.Computational analysis was carried out to the measured data on August 28 August 5 for this wind energy turbine set.

Based on the probability clustering grouping method of wind power plant operation data, adopt K-means algorithm, in MATLAB, carry out cluster to the actual measurement air speed data matrix in 24 Fans August 5 to Augusts 28, partial data is as shown in table 1.According to the geographic position figure of wind energy turbine set 24 units, the number that hives off of K-means algorithm is set as 3.Wind turbines grouping result after cluster is added up, always has 3456 component group results.

The wind speed measured data in a moment of table 1 blower fan

Machine group number	Wind speed (m/s)	Machine group number	Wind speed (m/s)
				F001#	8.64	F013#	9.05
F002#	8.01	F014#	8.18
				F003#	8.09	F015#	8.74
F004#	8.09	F016#	9.1
				F005#	9.2	F017#	9.55
F006#	2.93	F018#	9.77
				F007#	10.52	F019#	8.12
F008#	9.8	F020#	10.13
				F009#	4.09	F021#	6.41
F010#	13.93	F022#	8.76
				F011#	9.74	F023#	6.79
F012#	10.43	F024#	8.61

According to step 6,3456 component group results are divided into 316 classes, and calculate the probability of each class, result of calculation as shown in Figure 2.Wherein, ordinate is probability, and horizontal ordinate is the sequence number of every class.

As shown in Figure 2, the probability of category-A is 0.4642, is the situation of maximum probability in 316 classes.Therefore, selected category-A is as sorted grouping result, and the grouping result of category-A is group's number 3, and unit number of units is respectively (15,3,6).

Profile value function is adopted to assess all groups in category-A.Because close to 1, profile value S (i) more represents that grouping result is more reasonable, it is unreasonable that S (i) is less than null representation grouping result.Therefore, select in category-A profile value S (i) be all greater than zero and closest to 1 a component group result, as shown in Figure 3.Final grouping result is as shown in table 2.

The final grouping result of table 2 probability clustering

Correctness and the validity of above-mentioned result of calculation is verified below from actual geographic distribution situation and simulation result two angles.

A. actual geographic distribution

The geographic distribution situation of final grouping result and wind energy turbine set reality is contrasted, as shown in Figure 4, in figure, has irised out the blower fan that each group (group 1, group 2, group 3) comprises.

Can obtain after analysis drawing a conclusion:

1, the terrain residing for this wind energy turbine set is comparatively complicated, and final unit grouping result is substantially identical with the result dividing unit by geographic position, but not quite identical.Difference is mainly: blower fan F011# and blower fan F006# to F010# is distributed on prevailing wind direction according to straight line, but because the mima type microrelief of F011# is larger with other blower fan gaps, and apart from closer to the F018# blower fan of upper left, finally not have and other blower fans with row are divided in a same group of planes.

2, the impact of the factor such as terrain, wake effect is subject to, the nearer blower fan of some geographic distance is not in a group of planes, it is close that blower fan F015# to F017# and peripheral as middle in position place encloses blower fan (in blue region) sea level elevation, but due to wake effect impact on each wind direction, middle blower fan and peripheral blower fan put two group of planes respectively under.

3, comparative analysis illustrates, the geographic distribution situation that this grouping result is substantially realistic, illustrates that probability clustering clustering algorithm of the present invention is rationally effective.For the irregular wind energy turbine set of layout, simple employing by geographic position, wind energy turbine set simply can not be divided, rationally should hive off in conjunction with actual ruuning situation.

B. simulating, verifying

Wind energy turbine set multimachine Equivalent Model and detailed realistic model has been built based on PSCAD.Consider the influence factor of wake effect, topography and geomorphology, in actual wind energy turbine set every platform unit be blown into wind speed and incomplete same, therefore in order to be consistent with actual conditions, it is not identical that the present invention to suppose in realistic model that the different blower fan collected on line is blown into wind speed, but be all defined in maximal power tracing wind speed section, be set to by propeller pitch angle constant, do not consider award setting system.

Wind energy turbine set three wind speed collecting line are set and are followed successively by 10m/s, 8m/s and 6m/s.Line voltage drops to 30% when 8s, continues 0.2s.The simulation result of meritorious, idle and electric current as shown in Figures 5 to 7.

Can be seen by the simulation result of Fig. 5 to Fig. 7, the curve of the active power of output of the multimachine Equivalent Model obtained by probability clustering clustering algorithm, reactive power and electric current is with wind-field model height is consistent in detail, no matter in the steady state or under transient state, multimachine Equivalent Model all can simulate wind energy turbine set overall permanence more accurately.Therefore, when not waiting wind speed, the probability clustering wind energy turbine set grouping method that the present invention proposes is rationally effective.

The present invention propose the probability clustering grouping method based on wind power plant operation data with the actual operating data of wind speed for index, according to the principle that Euclidean distance is nearest, the probability clustering algorithm in conjunction with K-means algorithm is adopted to hive off, and test by the rationality of profile value function to grouping result, the most effective selected grouping result, this grouping result can be used for the ruuning situation of various wind speed, substantially increases the efficiency of work of hiving off.Utilize said method to obtain grouping result comparatively accurate, tally with the actual situation, have good effect to the blower fan irregular wind energy turbine set that distributes, for the use of later wind-field model provides great convenience, there is important engineer applied and be worth.

Above-described specific embodiment; object of the present invention, technical scheme and beneficial effect are further described; be understood that; the foregoing is only specific embodiments of the invention; the protection domain be not intended to limit the present invention; within the spirit and principles in the present invention all, any amendment made, equivalent replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (10)

1., based on a probability clustering grouping method for wind power plant operation data, it is characterized in that, the method comprises:

Step 1, obtains wind energy turbine set measured data within a certain period of time;

Step 2, the geographic position according to wind energy turbine set sets the number k that hives off, and under initial situation, chooses k sample at random as initial cluster center in described measured data;

Step 3, calculate each sample in described measured data and divide the Euclidean distance being clipped to each initial cluster center, choose the minimum Eustachian distance that each sample described is corresponding, and each sample described is put under in the group at the initial cluster center place corresponding with described minimum euclidean distance, generate grouping result first;

Step 4, according to the Euclidean distance that step 3 calculates, calculates the square error summation obtaining cluster corresponding to grouping result first;

Step 5, in non-initial situation, calculates acquisition k non-initial cluster centre according to the grouping result that last time generates;

Step 6, calculate each sample in described measured data and divide the Euclidean distance being clipped to each non-initial cluster centre, choose the minimum Eustachian distance that each sample described is corresponding, and each sample described is put under in the group at the non-initial cluster centre place corresponding with described minimum euclidean distance, generate new grouping result;

Step 7, according to the Euclidean distance that step 6 calculates, calculates the square error summation obtaining cluster corresponding to new grouping result;

Step 8, circulation performs the iterative process of above-mentioned steps 5 to step 7, the relatively square error summation of the cluster of twice adjacent calculation acquisition, when meeting E-E'< ε, stop iterative computation, wherein, E, E' are respectively the cluster square error summation that twice adjacent calculation obtains, being worth large is E, and being worth little is E';

The all grouping result obtained are classified, are set the wind speed probability under fan operation operating mode corresponding to every component group result by step 9, calculate the probability obtaining each class;

Step 10, chooses a class grouping result of maximum probability, utilizes profile value function to assess, as the net result that probability clustering hives off the every component group result in such grouping result.

2. the probability clustering grouping method based on wind power plant operation data according to claim 1, is characterized in that, in step 1, obtains the measured data of n typhoon group of motors in time period m and is expressed as:

$V=(V_1,V_2,\mathrm{...},V_j,\mathrm{...},V_m)={\left(v_{i,j}\right)}_{n\&times;m}=\left[\begin{array}{cccc}{v}_{11}& v_{12}& ...& v_{1m}\\ v_{21}& v_{22}& ...& v_{2m}\\ ...& ...& ...& ...\\ v_{n1}& v_{n2}& ...& v_{nm}\end{array}\right];$

Wherein, v _ijrepresent the measured data of the i-th typhoon group of motors in a jth moment wind speed index; Vectorial V _jregard a time samples group of SPACE V as, in V, the quantity of sample group is m, and dimension is n, V _jrepresent the vector of a jth moment all Wind turbines measured datas.

3. the probability clustering grouping method based on wind power plant operation data according to claim 2, it is characterized in that, in step 2, geographic position according to wind energy turbine set sets the number k that hives off, under initial situation, k sample is chosen as initial cluster center, initial cluster center at random in described measured data be expressed as:

$m_p^{\left(1\right)}=(v_{i1},v_{i2},\mathrm{...},v_{ij});$

Wherein, p=1,2 ..., k, k represent number of hiving off.

4. the probability clustering grouping method based on wind power plant operation data according to claim 3, is characterized in that, in step 3, each sample is to the Euclidean distance of each initial cluster center described for:

$d(V_i,m_p^{\left(1\right)})=\sqrt{\underset{j=1}{\overset{m}{\&Sigma;}}{(v_{ij}-m_p^{\left(1\right)})}^2};$

Wherein, i represents the i-th typhoon group of motors, p=1,2 ..., k, k represent the number that hives off, represent p the initial cluster center hived off.

5. the probability clustering grouping method based on wind power plant operation data according to claim 4, is characterized in that, in step 4, and the square error summation E of the cluster that grouping result is corresponding first ⁽¹⁾for:

$E^{\left(1\right)}=\underset{p=1}{\overset{k}{\&Sigma;}}\underset{i=1}{\overset{n}{\&Sigma;}}{d}^2(V_i,m_p^{\left(1\right)});$

Wherein, i represents the i-th typhoon group of motors, p=1,2 ..., k, k represent the number that hives off, for the Euclidean distance that step 3 calculates.

6. the probability clustering grouping method based on wind power plant operation data according to claim 5, it is characterized in that, in steps of 5, in non-initial situation, according to the grouping result that last time generates, calculate the mean value of the sample value comprised in each hiving off, obtain k non-initial cluster centre, non-initial cluster centre be expressed as:

$m_p^{\left(r\right)}=\frac{1}{l_p^{(r-1)}}(\underset{V_t\&Element;G_D}{\&Sigma;}{v}_{i1},\underset{V_t\&Element;G_D}{\&Sigma;}{v}_{i2},\mathrm{...},\underset{V_t\&Element;G_D}{\&Sigma;}{v}_{im});$

Wherein, p=1,2 ..., k, k represent number of hiving off, and r represents the number of times of iteration, l _prepresent p the unit number hiving off contained, V _trepresent the some sample values in measured data, G _dto represent in grouping result some hives off.

7. the probability clustering grouping method based on wind power plant operation data according to claim 6, is characterized in that, in step 6, each sample divides the Euclidean distance being clipped to each non-initial cluster centre for:

$d(V_i,m_p^{\left(r\right)})=\sqrt{\underset{j=1}{\overset{m}{\&Sigma;}}{(v_{ij}-m_p^{\left(r\right)})}^2};$

Wherein, i represents the i-th typhoon group of motors, p=1,2 ..., k, k represent the number that hives off, represent p the non-initial cluster centre hived off.

8. the probability clustering grouping method based on wind power plant operation data according to claim 7, is characterized in that, in step 7, and the square error summation E of the cluster that new grouping result is corresponding ^(r)for:

$E^{\left(r\right)}=\underset{p=1}{\overset{k}{\&Sigma;}}\underset{i=1}{\overset{n}{\&Sigma;}}{d}^2(V_i,m_p^{\left(r\right)});$

Wherein, i represents the i-th typhoon group of motors, p=1,2 ..., k, k represent the number that hives off, for the Euclidean distance 1 that step 6 calculates.

9. the probability clustering grouping method based on wind power plant operation data according to claim 8, it is characterized in that, in step 9, the all grouping result obtained are classified, set the wind speed probability under fan operation operating mode corresponding to every component group result, calculate the probability obtaining each class, comprising:

In all grouping result obtained, identical for identical for the number of hiving off comprised and the blower fan quantity dividing mode of hiving off is divided into a class grouping result;

Calculate the probability obtaining each class, the formula of utilization is as follows:

$P^{\left(X\right)}=\underset{i=1}{\overset{n\left(X\right)}{\&Sigma;}}{P}_{wind,i};$

Wherein, P ^(X)for the probability of each class, X is the numbering of class, and n (X) is for comprising the group number of grouping result, P _{wind, i}it is the wind speed probability under the i-th Fans operating condition.

10. the probability clustering grouping method based on wind power plant operation data according to claim 9, is characterized in that, in step 10, utilize profile value function to assess the every component group result in such grouping result, formula is as follows:

$S\left(i\right)=\frac{min\left(b\right)-a}{max\&lsqb;a,min\left(b\right)\&rsqb;},i=1,2,\mathrm{...},n;$

Wherein, a is the mean distance in sample i and place grouping result between same other sample hived off;

B is the mean distance between the sample i sample hived off different from the grouping result of place;

S (i) represents profile value, and span is that the value of-1 to 1, S (i) is larger, and the classification of sample i is more reasonable, as S (i) <0, illustrates that scheme of hiving off is unreasonable;

In a class grouping result of the described maximum probability selected, choose profile value S (i) be all greater than 0 and closest to 1 a component group result, as the net result that probability clustering hives off.