A kind of new energy typical scene construction method based on improvement FCM clustering algorithm
Technical field:
The present invention relates to the construction methods of new energy typical case power output scene, are clustered more particularly to one kind based on FCM is improved
The new energy typical scene construction method of algorithm.
Background technique
As the scale of China's new energy develops and uses, generation of electricity by new energy is persistently grown rapidly, and installed capacity increases rule
Mould continuous enlargement.With the sustainable growth of new energy installed capacity, new energy accounts for the continuous promotion of electric network source ratio, new energy
Consumption demand formulation etc. of Economical Operation of Power Systems, the assessment of power grid new energy digestion capability, dispatching of power netwoks plan is proposed
Higher requirement.Accordingly, it is considered to which the seasonality and periodicity of the new energy such as wind-powered electricity generation, photoelectricity power output, go out in force data from history
Extract representative typical power output scene, reflected with these typical case's power output scenes in it is long-term in new energy contribute it is special
Property, it is of great significance to the power source planning of the electric system of the new energy containing high proportion.
In recent years, domestic and international researcher studies the building of new energy typical case power output scene.Currently, using compared with
Choosing method for extensive medium-term and long-term new energy power output scene or part throttle characteristics is generally divided into three kinds: typical Fa, timing are imitative
True method and clustering algorithm.Typical Fa is often referred to using within certain period the one day power producing characteristics closest with average value as allusion quotation
Type daily output scene, or select representative in certain period one day as typical daily output scene.It is obtained by typical Fa
Take the power producing characteristics of new energy simple and fast, but since not abundant enough the variation that cannot embody annual new energy power output of scene is special
Property, error is larger in long -- term generation expansion planning calculating.Time stimulatiom method refers to the practical power output time sequence by history new energy
Column data, is adjusted to obtain and simulates power time series further according to installed capacity and the variation of other factors.Time stimulatiom method
As a result accurately and reliably the practical power producing characteristics of the annual power output time series obtained close to the new energy such as daily wind-powered electricity generation, photovoltaic lack
Point is that computational efficiency is low.Clustering algorithm is then the method power output scene practical to the new energy of long-time timing by clustering
Information extraction, classification and abbreviation are carried out, and then obtains typical scene set.Clustering algorithm both ensure that out the original spy of force data
Property, and taken into account computational efficiency.Currently, most research is by Classic Clustering Algorithms to a large amount of actual load or wind-powered electricity generation
Data are analyzed, and have obtained the customer charge that can more accurately reflect actual characteristic and wind power output scene collection, but not
It is thoroughly discussed between the selection of cluster numbers, with the correlation the multiclass new energy of region.
In consideration of it, it is necessary to be improved and optimizated to traditional FCM clustering algorithm, so that area plentiful to certain new energy is new
The history timing of the energy goes out force data and carries out clustering, generates the new energy typical case power output scene collection on the ground, and is actually asking
The validity of verification method in topic.
Summary of the invention
Goal of the invention: the object of the present invention is to provide it is a kind of be able to solve defect existing in the prior art based on improvement
The new energy typical scene construction method of FCM clustering algorithm.
Technical solution: to reach this purpose, the invention adopts the following technical scheme:
A kind of new energy typical scene construction method based on improvement FCM clustering algorithm, method includes the following steps:
S1: FCM clustering algorithm is improved by establishing Cluster Validity Index function;
S2: clustering is carried out to new energy power output historical data using FCM clustering algorithm is improved;
S3: the new energy power output typical scene after clustering in each classification is chosen.
It is described based on the new energy typical scene construction method for improving FCM clustering algorithm, the cluster in step S1 is effective
Property target function CH(+)Are as follows:
Wherein, c is cluster numbers, and n is sample number, Tc、PcBetween class, sum of squares of deviations in class.
It is described based on the new energy typical scene construction method for improving FCM clustering algorithm, it is poly- by establishing in step S1
The improved FCM clustering algorithm process of class Validity Index function are as follows:
1) variation range of cluster numbers c is set
2) FCM clustering algorithm is called, cluster centre to objective function is updated and restrains;
3) CH is calculated(+)Index;
4) c=c+1 turns to step 2), until
5) CH under more different c values(+)Index size, determines preferable clustering number;
6) cluster result under preferable clustering number is exported.
The new energy typical scene construction method based on improvement FCM clustering algorithm, new energy includes wind energy and light
Can, the process for carrying out clustering to new energy power output historical data using improvement FCM clustering algorithm in step S2 are as follows:
1) wind-powered electricity generation is acquired, the history of two class new energy of photoelectricity goes out force data;
2) go out force data to new energy history to pre-process, including losing data and the amendment of accidental data etc., obtain
Continuous n days, the output of wind electric field data acquisition system of daily m constant duration beBetween the corresponding time
Every photovoltaic plant photovoltaic plant power output data set beWherein,
Indicate n-th day wind power output data,Indicate n-th day photoelectricity power output data;
3) using the improvement FCM clustering algorithm proposed in step S1 to wind-powered electricity generation, photoelectricity power output historical data Xw、XsGathered
Class divides.
It is described based on the new energy typical scene construction method for improving FCM clustering algorithm, it is every after the cluster in step S3
The selection process of new energy power output typical scene in one classification are as follows:
1) force data out of each history day in such is read;
2) average value for all history sunrise activity of force that will belong to such is calculatedJ is to belong to such to go through
History day is total, PiFor the sunrise activity of force of history day i;
3) calculate each history day goes out activity of force and mean power distance di=| Pi-Pavg|;
4) to diAscending sort is carried out, according to the typical field selected apart from nearest history sunrise force curve for the type
Scape.
The utility model has the advantages that the present invention is proposed and is based on using the new energy power producing characteristics in the plentiful region of new energy as research object
New energy typical scene collection construction method and the operating cost of the New-energy power system containing high proportion for improving FCM clustering algorithm are excellent
Change model.It contributes using based on the typical scene collection construction method for improving FCM clustering algorithm to the history timing of the ground new energy
The typical power output scene collection on the ground is generated after data progress clustering.The simulation result of specific embodiment shows based on changing
The wind-powered electricity generation that is generated into the new energy typical scene collection construction method of FCM clustering algorithm, the photoelectricity typical scene year obvious, symbol of feature
Close practical power output situation, and advantage small with computational efficiency height, error in practical engineering applications.
Detailed description of the invention
Fig. 1 is wind power output scene clustering Validity Index in the specific embodiment of the invention;
Fig. 2 is wind-powered electricity generation spring and autumn typical case's power curve in the specific embodiment of the invention;
Fig. 3 is wind-powered electricity generation summer typical case power curve in the specific embodiment of the invention;
Fig. 4 is wind-powered electricity generation winter typical case power curve in the specific embodiment of the invention;
Fig. 5 is photoelectricity spring and autumn typical case's power curve in the specific embodiment of the invention;
Fig. 6 is photoelectricity summer typical case power curve in the specific embodiment of the invention;
Fig. 7 is photoelectricity winter typical case power curve in the specific embodiment of the invention.
Specific embodiment
Technical solution of the present invention is further introduced With reference to embodiment.
New energy typical scene construction method of the present invention based on improvement FCM clustering algorithm, comprising the following steps:
S1: FCM clustering algorithm is improved by establishing Cluster Validity Index function;
S2: clustering is carried out to new energy power output historical data using FCM clustering algorithm is improved;
S3: the new energy power output typical scene after clustering in each classification is chosen.
Further, the FCM clustering algorithm in step S1 is a kind of clustering algorithm based on division, by introducing degree of membership letter
Relational extensions between object and class cluster are described to any number on [0,1] closed interval, pass through judgement by this concept of number
The numerical value of subordinating degree function divides that object is more prone to which class cluster belonged to.
Firstly, the concept definition of fuzzy subset are as follows:
For arbitrary x ∈ U, a several μ can be determinedA(x) ∈ [0,1] describes the degree that x belongs to A, is defined as A's
Subordinating degree function, the degree of membership constant μ of element in U to fuzzy subset AA(x) it describes.μA(x) degree of membership is closer to 0, table
Show x belong to A degree it is smaller;μA(x) closer to 1, the degree that expression x belongs to A is bigger.
For a data set X={ x1,x2,x3,…,xn, using FCM cluster data set X be divided into c class (2≤c≤
N), wherein the set of cluster centre can be expressed as V={ v1,v2,…,vc, in fuzzy division, with certain degree of membership numerical value
Belong to certain one kind to describe each data object, without being strictly divided into certain one kind.
I-th of data sample x in data set XiBelong to the degree of membership μ of jth classijIt is indicated with following numerical relation:
Subordinated-degree matrix U={ μ is initialized with random number of the value in [0,1] sectioni,j, meet it in above formula
The minimum of objective function is solved on the basis of constraint condition:
It is subordinate to angle value μijIt calculates as follows:
In formula, F (X, V) indicates data sample to the Weighted distance quadratic sum of cluster centre, and weight is data sample xiBelong to
In jth class degree of membership μijF power, smoothing factor (FUZZY WEIGHTED parameter) f ∈ [1 ,+∞), the index for adjusting degree of membership can be with
For controlling the smoothness of adjustment degree of membership, in general example, such as without particular/special requirement, smoothing factor f generally takes 2.U=
{μi,jIt is fuzzy membership matrix.dij=| | xi-vj| | indicate i-th of data sample to j-th of cluster centre Euclid away from
From.
FCM clustering algorithm process may be summarized as follows:
1) cluster numbers c, the number of iterations k=0, maximum number of iterations T, termination error ε are set, and initializes cluster centre
V0;
2) subordinated-degree matrix U is updated with formula (4)k;
3) next cluster centre V is updated with formula (3), formula (4)k+1;
If 4) | | Uk+1-Uk| | < ε terminates algorithm.Otherwise, k=k+1, return step 2 are enabled).
Further, the Cluster Validity Index function CH in step S1(+)Are as follows:
Wherein, Tc、PcSum of squares of deviations between class and in class.Sum of squares of deviations can reflect between class and class between class
Otherness, the value are the bigger the better;Sum of squares of deviations then reflects the otherness between similar sample in class, and the value is the smaller the better.
So finding preferable clustering number c can be exchanged into the maximum value for seeking the index.
Further, it is summarized in step S1 by establishing the improved FCM clustering algorithm process of Cluster Validity Index function
Are as follows:
1) variation range of cluster numbers c is set
2) FCM clustering algorithm is called, cluster centre to objective function is updated and restrains;
3) CH is calculated(+)Index;
4) c=c+1 turns to step 2), until
5) CH under more different c values(+)Index size, determines preferable clustering number;
6) cluster result under preferable clustering number is exported.
Further, the new energy in step S2 includes wind energy and luminous energy, is contributed using FCM clustering algorithm is improved to new energy
The process of historical data progress clustering are as follows:
1) wind-powered electricity generation is acquired, the history of two class new energy of photoelectricity goes out force data;
2) go out force data to new energy history to pre-process, including losing data and the amendment of accidental data etc., obtain
Continuous n days, the output of wind electric field data acquisition system of daily m constant duration beBetween the corresponding time
Every photovoltaic plant photovoltaic plant power output data set beWherein,
Indicate n-th day wind power output data,Indicate n-th day photoelectricity power output data;
3) using the improvement FCM clustering algorithm proposed in step S1 to wind-powered electricity generation, photoelectricity power output historical data Xw、XsGathered
Class divides.
Further, the selection process of the new energy power output typical scene after the cluster in step S3 in each classification are as follows:
1) force data out of each history day in such is read;
2) average value for all history sunrise activity of force that will belong to such is calculatedJ is to belong to such to go through
History day is total, PiFor the sunrise activity of force of history day i;
3) calculate each history day goes out activity of force and mean power distance di=| Pi-Pavg|;
4) to diAscending sort is carried out, according to the typical field selected apart from nearest history sunrise force curve for the type
Scape.
Below by taking Zhejiang Province's new energy enriching area as an example, with the new energy proposed in this paper based on improved FCM algorithm
The construction method of source power output typical scene collection constructs the ground new energy typical scene collection.This area's wind-powered electricity generation installation 50MW, photovoltaic hair
Denso machine 320MW accounts for about the 25% of full area installation, choose the ground from wind-powered electricity generation on May 31,1 day to 2018 June in 2017 with
The force data that goes out of photovoltaic power generation is sample, it is contemplated that wind-powered electricity generation, photoelectricity power output all have periodicity relevant to season, this is specific real
It applies and carries out typical scene again after wind-powered electricity generation, photoelectricity history power output data sample are divided into spring and autumn, summer and winter three classes in mode
Building, and by the new energy typical scene set constructed by be applied to high proportion New-energy power system medium-term and long-term operation at
The practical problem of this optimization evaluates the practical application value of this method.
Embodiment 1:
Embodiment 1 goes out force data to wind-powered electricity generation, photoelectricity history using improvement FCM clustering algorithm and carries out scene partitioning, clusters
Cluster Validity Index operation is carried out in journey first, carries out scene partitioning again after obtaining preferable clustering number.With the wind-powered electricity generation of this area
For, calculate the wind power output scene clustering validity CH in each season(+)Index, herein using extreme value normalization method to CH(+)Refer to
Mark is handled, and is shown below:
Treated wind power output scene clustering validity CH(+)Index is as shown in Figure 1.
As seen from Figure 1, each season Cluster Validity Index CH of wind-powered electricity generation(+)It is maximized at 2, i.e. wind-powered electricity generation each season
The preferable clustering number of section power output scene is 2.Likewise it is possible to which the preferable clustering number for acquiring each season power output scene of photoelectricity is also equal
It is 2.Fig. 2 to 4 gives the typical power curve being polymerized to after two classes in wind-powered electricity generation spring and autumn, summer and winter, and Fig. 5 to 7 gives light
Electric spring and autumn, summer and winter are polymerized to the typical power curve after two classes.The cluster result it can be seen from above-mentioned power curve figure
Substantially divided by the height of power output.The general of each season power output typical scene of wind-powered electricity generation, photoelectricity is set forth in Tables 1 and 2
Rate distribution situation.
Each season power output typical scene probability distribution of 1 wind-powered electricity generation of table
Each season power output typical scene probability distribution of 2 photoelectricity of table
The power curve of Fig. 2 to Fig. 7 intuitively characterizes the typical power output size and variation for finding out wind-powered electricity generation, photoelectricity each season
Trend.Wind-powered electricity generation has stronger fluctuation, and peak, paddy change procedure and anti-tune peak character are obvious.Photoelectricity power output also has a standing wave
Dynamic property, power output size are closely connected with intensity of illumination, are maximized at daily 13 or so.It can be with by above-mentioned power curve figure
Find out, cluster result is substantially divided by the height of power output, and each season power output of wind-powered electricity generation, photoelectricity is set forth in Tables 1 and 2
The probability distribution of typical scene.
Embodiment 2:
The operating cost that wind-powered electricity generation/photoelectricity power output typical scene is applied to high proportion New-energy power system by embodiment 2 is excellent
Change field, for reduce region " abandonment abandoning light " phenomenon, by the operation of the punishment cost and all units of " abandonment abandoning light " electricity at
The sum of this is minimum to be used as optimization aim, and the maximum consumption of new energy, objective function are taken into account while economization operation are as follows:
Wherein, T indicates emulation period sum, cwFor wind-powered electricity generation penalty coefficient, csFor photoelectricity penalty coefficient,Indicate that wind-powered electricity generation exists
The prediction of t period is contributed,Indicate practical power output of the wind-powered electricity generation in the t period,Indicate that photoelectricity is contributed in the prediction of t period,
Indicate practical power output of the photoelectricity in the t period,Indicate practical power output of the thermoelectricity in the t period, ath, bth, cthFor thermoelectricity power generation at
This coefficient, as, bs, csFor wind-powered electricity generation cost of electricity-generating coefficient, aw, bw, cwFor photovoltaic power generation cost coefficient.
The Optimized model consider constraint condition include:
1) electric quantity balancing constrains: region thermoelectricity, the total power generation of wind-powered electricity generation and photoelectricity and customer charge balance.
Wherein, u indicates the zone user load.
2) generated output constrains: the generated energy of all types of units meets respectively maximum, minimum generated energy constraint.
Wherein, qthminIndicate the minimum generated energy of thermal power station, qthmaxIndicate the maximum generating watt of thermal power station, similarly,
qwmin、qwmax、qwmin、qwmaxIndicate maximum, the minimum generated energy of wind-powered electricity generation and photoelectricity.
3) fired power generating unit Climing constant:
Wherein, γdownAnd γupThe downward creep speed and upward creep speed of fired power generating unit are respectively indicated,WithPoint
Not Biao Shi t-1 and t period thermal power plant generated energy.
Using above-mentioned Operation of Electric Systems cost optimization model to annual timing method, improved FCM algorithm and typical case's Fa structure
The scene built out is verified compared with, wherein the power output that typical case Fa chooses spring and autumn, summer and winter wind-powered electricity generation, photoelectricity most connects
One daily output of nearly average value is as typical scene.The annual power output scene constructed using annual timing method is obtained as benchmark
Performance comparison situation to three kinds of scenario building methods is as shown in table 3.
3 scenario building method performance comparison of table
As can be seen that a large amount of because of number of scenes based on the scenario building method of FCM clustering algorithm and typical case Fa is improved
Reduction is so that operation efficiency is obviously improved.But the field of force out of the wind-powered electricity generation, photoelectricity whole year constructed with annual timing method
Scape is compared, and the scene that constructs for improving FCM clustering algorithm and typical case Fa has certain error.Wherein, based on improvement FCM cluster
Algorithm compared with the scenario building method based on annual timing method wind-powered electricity generation, photoelectricity prediction result error rate be respectively 33.4% and
27.1%, based on typical Fa compared with the scene constructed based on annual timing method wind-powered electricity generation, photoelectricity prediction result error rate point
Not Wei 49.8% and 51.9%, in contrast, operation efficiency is being guaranteed based on the scenario building method for improving FCM clustering algorithm
It can also accurately reflect the annual operating condition of wind-powered electricity generation, photoelectricity simultaneously, there is practical application value.