CN104200484A - Distributed photovoltaic system ultra-short-term power output prediction method based on cloud cluster characteristic analysis - Google Patents

Abstract

The invention provides a distributed photovoltaic system ultra-short-term power output prediction method based on cloud cluster characteristic analysis. The distributed photovoltaic system ultra-short-term power output prediction method comprises the steps of firstly, preprocessing a sparse cloud cluster acquired by a sky cloud cluster detection device, extracting the core region of the cloud cluster as a target region and performing mathematical description on the key dynamic characteristic parameters, affecting the distributed photovoltaic system ultra-short-term power output, of the target region, after image preprocessing, finding a special point of which the motion condition is used for representing the motion condition of the whole cloud cluster from the core region of the cloud cluster, performing thickness description and digital description by use of a pixel deep-shallow differential clustering analysis method, namely classifying gray levels mainly by use of a gray segmentation method and visually and vividly representing the thicknesses of different regions of the cloud cluster by use of pseudo-color map processing, next, performing cloud cluster extraction and output prediction, designing an independent ultra-short-term power output prediction software module based on the dynamic characteristics of the sparse cloud cluster and a database system of the prediction software module, and then realizing the distributed photovoltaic system ultra-short-term power output prediction based on the cloud cluster characteristic analysis.

Description

Distributed photovoltaic system ultra-short term based on Features of Cloud Cluster Causing analysis goes out force prediction method

Technical field:

The present invention relates to a kind of computer vision analysis means that adopt the sparse cloud cluster image sequence of obtained sky carried out to key feature analysis, so to distributed photovoltaic system carry out ultra-short term exert oneself prediction method.

Background technology:

The energy is as the driving force that promotes social development progress, and for guarantee national economy, important supporting role is played in fast development.Sun power is as one of green energy resource, because energy is huge, cleanliness without any pollution, the advantage such as safe and reliable become the important selection of reply energy shortage, climate change and energy-saving and emission-reduction.As the photovoltaic generation of one of main application form of sun power, because of the almost discharge of contamination-free of its power generation process, accomplish real cleanly, green, therefore become one of main selection of new forms of energy and be widely applied.

Distributed photovoltaic power generation is incorporated into the power networks can on-site elimination electric energy, without carrying out remote transmission of electric energy, there are start and stop and respond soon, dispatch the advantages such as convenient, flexible, peaking performance is good, also can adjust photovoltaic system amount of capacity according to site requirements, and can alleviate to a certain extent local shortage of electric power situation, be that the strong of power system operating mode supplemented.Thereby presented in recent years explosive growth, along with the lasting propelling of Development of China's Urbanization, distributed photovoltaic power generation system is incorporated into the power networks and will obtains promoting largelyr.

At present mainly concentrate on short-term forecasting for the forecasting research of photovoltaic generating system both at home and abroad, i.e. time scale <24 hour, temporal resolution is the short-term of the 15 minutes prediction of exerting oneself.The exert oneself time scale of short-term forecasting and resolution of photovoltaic generation is mainly used for reference and the short-term forecasting technical indicator of wind-electricity integration defined, but in fact the fluctuation of the ultra-short term of photovoltaic generation wants complicated and frequent more than wind-powered electricity generation, one of them important reason is exactly that sky is floating indefinite and almost there is no a cloud cluster of inertia, the dynamic change of cloud cluster is real-time, violent on the impact of solar radiation, thereby cause acutely and the rapidly fluctuation of ground photovoltaic system output power, the stable and safe operation of photovoltaic and relevant electric system is brought to threat.Conventional short-term forecasting temporal resolution, about 15 minutes, cannot be predicted because cloud cluster changes the power stage transience big ups and downs that cause at all; Meanwhile, the sun angle often considered in short-term forecasting, temperature, humidity, the factor such as weather rain or shine, due to its intrinsic physical characteristics, to change fluctuation slower for sparse cloud cluster, less to the ultra-short term predicted impact of photovoltaic generation on the contrary.

For the problems referred to above, thereby the present invention proposes a kind of computer vision analysis technological means that adopts, obtained sparse cloud cluster image sequence is carried out to key feature analysis to the exert oneself method of ultra-short term prediction of distributed photovoltaic system.

Summary of the invention:

The present invention will overcome tradition time scale and the temporal resolution predicting the outcome of exerting oneself can not reflect the major effect that sparse cloud cluster dynamic change is exerted oneself to photovoltaic system in real time, the not high shortcoming of precision causes predicting the outcome, for improve exert oneself prediction temporal resolution, thereby the present invention propose a kind of adopt computer vision analysis technological means to obtained sparse cloud cluster image sequence carry out key feature analysis realize to distributed photovoltaic system carry out ultra-short term exert oneself prediction method.

The present invention is that the technical scheme that technical solution problem adopts is:

With reference to Fig. 1: first obtain continuous sparse cloud cluster image sequence by cloud layer pick-up unit, adopt computer vision analysis technological means to carry out key feature analysis to image sequence, the crucial behavioral characteristics that causes photovoltaic system ultra-short term to go out fluctuation to it carries out mathematical description, by applicable image processing algorithm and feature extraction algorithm, key feature is extracted to calculating.History is gone out to force data to application RBF nerve network and Features of Cloud Cluster Causing carries out off-line training, eliminates wind speed, temperature to the impact predicting the outcome by setting up online compensator.Finally obtain accurately, predict the outcome in real time and reliably according to the forecast model of exerting oneself, for energy the flow-direction and flowrate, adjustment electric load and the electric power supply adjusted between electrical network, photovoltaic micro-grid system, energy storage device and load provide decision support data.

Distributed photovoltaic system ultra-short term based on Features of Cloud Cluster Causing analysis goes out force prediction method, comprises the following steps:

1) the sparse cloud cluster first sky cloud cluster pick-up unit being collected carries out pre-service, extracts the nucleus of cloud cluster as target area, and then the crucial dynamic feature coefficient that it is affected to photovoltaic system ultra-short term power stage carries out mathematical description; Cloud cluster is adopted to opening operation operation, cloud cluster is first corroded the general character of rear expansion extraction cloud cluster, first adopt a relatively large structural elements to corrode cloud cluster, if the point set that A is cloud cluster, B is structural elements, B is exactly the set z translation that is contained in the some z in A in all B to the corrosion operation of A, and formula is:

F＝AΘB＝{z|(B) _z∩A ^c＝φ}

Wherein, A ^cbe the supplementary set of A, φ is empty set;

Owing to adopting a larger structural elements to corrode operation to cloud cluster, most edge has all been eliminated, remaining cloud cluster nucleus, i.e. the most concentrated region of cloud cluster, then carries out expansive working to the cloud cluster through excessive erosion, and formula is:

$G=F\&CirclePlus;B=\left\{z\right|{\left(B\right)}_z\&cap;A\&NotEqual;\mathrm{\&phi;}\}$

Has the set of the overlapping all some z of element at least with the B of z translation and F;

2) by after image pre-service, the motion conditions of then finding a particular point in cloud cluster nucleus represents the motion conditions of whole cloud cluster, the i.e. barycenter of cloud cluster nucleus; Center-of-mass coordinate formula is:

$\stackrel{\&OverBar;}{x}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{x}_i$

$\stackrel{\&OverBar;}{y}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{y}_i$

Obtain the average coordinates of cloud cluster barycenter by above-mentioned two average formulas; Wherein ( ) be exactly the center-of-mass coordinate calculating, x _iand y _irepresent respectively cloud cluster nucleus each point x and y direction coordinate in cloud atlas, n represents the pixel number of cloud cluster nucleus, and before and after adopting, the barycenter of a few width sequential chart pictures extraction cloud clusters carries out the movement velocity of vectorization tracking calculating cloud cluster; That is:

represent t _nthe horizontal ordinate of moment cloud cluster barycenter, represent t _nthe ordinate of moment cloud cluster barycenter; represent t _n+1the horizontal ordinate of moment cloud cluster barycenter, represent t _n+1the ordinate of moment cloud cluster barycenter; t _nwith t _n+1be separated by the unit interval, t _n+1movement velocity v, the t of moment cloud cluster _n+1moment and t _nline between moment cloud cluster barycenter and the angle β of X-axis are expressed as:

$v=\sqrt{{({\stackrel{\&OverBar;}{x}}_{n+1}-{\stackrel{\&OverBar;}{x}}_n)}^2+{({\stackrel{\&OverBar;}{y}}_{n+1}-{\stackrel{\&OverBar;}{y}}_n)}^2}$

$\mathrm{\&beta;}=\arctan \left(\frac{\stackrel{\&OverBar;}{y_{n+1}}-\stackrel{\&OverBar;}{y_n}}{\stackrel{\&OverBar;}{x_{n+1}}-\stackrel{\&OverBar;}{x_n}}\right)$

Above-mentioned two formula can go out cloud cluster motion conditions within a certain period of time by quantificational expression;

Adopt barycenter angle function to describe the shape facility of cloud cluster, taking the barycenter of cloud cluster as limit, horizontal direction sets up polar coordinate system as pole axis, on cloud cluster profile, arbitrfary point C can be represented by polar coordinates function r (θ) to the distance r of barycenter, and the transfer equation that is tied to polar coordinate system from rectangular coordinate is:

$r_p=\sqrt{{(X_p-\stackrel{\&OverBar;}{x})}^2+{(Y_p-\stackrel{\&OverBar;}{y})}^2}$

Wherein, (X _p, Y _p) represent the cloud cluster nucleus frontier point in rectangular coordinate system, r _pbe illustrated in frontier point (X in polar coordinate system _p, Y _p) to the distance of barycenter, θ _prepresent in polar coordinate system the angle of arbitrfary point C and center of mass point line and pole axis on border;

3) application pixel depth difference clustering methodology carries out thickness and digitized description, mainly utilizes gray scale split plot design that gray shade scale is classified, and visual pattern ground represents the cloud cluster quality of cloud cluster zones of different; In the database of experimental system, design cloud cluster behavioral characteristics vector, above-mentioned each behavioral characteristics is formed to unified data structure and stored and call;

First judge near some empty clouds that cloud cluster border is, taking each frontier point bi of cloud cluster as the center of circle, set taking bi as the center of circle, R is the circle of radius and the borderline region that is cloud cluster with the overlapping region of cloud cluster, it is the region at empty cloud place, circle taking radius as R, home position was along the Boundary Moving of cloud cluster one week, and the region that now formed and the common factor of cloud cluster are territory, empty cloud sector:

I＝{z|(B) _z∩A≠φ}

Wherein A represents the point set of cloud cluster, and (B) z represents the point set in the circle region after a week along cloud cluster Boundary Moving taking r as radius, and I represents the point set of the empty cloud region of cloud cluster;

Then the cloud cluster quality of remaining cloud cluster part is carried out corresponding with the color of cloud cluster, by different gray-scale values, cloud cluster quality is carried out to quantitative analysis thus, obtain the gray-scale map of cloud cluster by pre-service, use cluster analysis that cloud cluster is divided into several different gray-scale value regions, the gray-scale value zoning of rolling into a ball by continuous several Zhang Yuns is analyzed, the basic zoning of predict future moment cloud cluster gray-scale value, draw again the gray-scale value of the effective occlusion area of cloud cluster, by shielded area and corresponding grey scale value are carried out to Integral Processing, finally draw the efficiency of blocking of following a certain moment cloud cluster, by cluster analysis, cloud cluster figure is divided after several different gray areas, utilize gray scale split plot design to carry out representing between cloud cluster quality subarea,

Gray scale split plot design is exactly to represent gray level with [0, N-1], makes V ₀represent black [f (x, y)=0], and make V _n-1represent white [f (x, y)]=N-1; Suppose and in [0, N-1], get M point, this M point is defined as to gray level V ₁, V ₂..., V _m.Then, suppose 0<M<N-1, M point is just divided into gray scale M+1 interval S ₁, S ₂... S _m+1; Gray level is carried out according to following relation to colored assignment:

IF:f(x,y)∈S _k

THEN:f(x,y)＝c _k

Wherein c _kbe and k gray area between S _kcorresponding color, S _kby being positioned at x=k-1 and x=k section definition;

256 gray shade scales are divided into 8 intervals, comprise 32 gray shade scales in each interval, each interval is similar to corresponding a kind of cloud cluster quality, 8 interval corresponding 8 cloud cluster quality;

4) Features of Cloud Cluster Causing extracts and the prediction of exerting oneself

Obtain one group of sequential cloud cluster figure by cloud layer monitoring device and picture collection device, respectively at t ₀, t ₁, t ₂... t _n-1moment claps, then in conjunction with the algorithm of above-mentioned steps by t ₀, t ₁, t ₂... t _n-1the information such as cloud cluster position, shape and the cloud cluster mass distribution situation in moment extract, and dope t _nposition, shape and the cloud cluster mass distribution situation at moment cloud cluster place, the effective coverage of causing photovoltaic panel to block in conjunction with corresponding time point again, determine the overlapping area size of corresponding time point cloud cluster and effective coverage and the cloud cluster mass distribution situation of lap, and then by the solar radiation index of photovoltaic panel under the corresponding time point of calculative determination;

Suppose that the photovoltaic panel total area is S _always, be S through calculating following moment cloud cluster with the area that overlaps of effective occlusion area _effectively, in the area of overlapping region, divide n region by n kind color, every kind of color represents between a kind of gray area, and then represents a kind of cloud cluster quality, therefore represented n kind cloud cluster quality, n kind is blocked ability e ₁, e ₂, e ₃..., e _n, e _irepresent e ₁to e _nin any one, and 0≤e _i< 1, corresponding photovoltaic panel shielded area is respectively S ₁, S ₂, S ₃..., S _n;

If unit area receivable maximum solar value in ground is Q _s,, under the blocking of cloud cluster, the corresponding shielded area of photovoltaic panel is the solar radiation quantity Q that Si region receives _ican be expressed as:

Q _i＝Q _s×S _i×(1-e _i)

And the solar radiation index that photovoltaic panel receives can be expressed as:

Wherein represent the solar radiation index that photovoltaic panel receives, scope is [0,1], in the time that fine day is unobstructed be 1, when the strongest black clouds of the ability of blocking causes while blocking completely photovoltaic panel, now be 0, between the partial occlusion between the two for (0,1);

5) independently exert oneself forecasting software module and Database Systems thereof of the ultra-short term based on sparse cloud cluster behavioral characteristics of design, collection comprises the historical output power data of the historical weather datas such as temperature, wind speed, solar radiation quantity and photovoltaic system, and imports the ultra-short term prediction module of exerting oneself; Off-line training and the online compensation of identification by Processing Algorithm realize target cloud cluster in this ultra-short term is exerted oneself forecasting software, feature extraction, historical data, finally realize the prediction of exerting oneself of distributed photovoltaic system ultra-short term based on Features of Cloud Cluster Causing analysis.

First the historical data segmentation to accumulation, uses wherein approximately 70% data to carry out RBF neural network prediction algorithm off-line training.After the precision of prediction that reaches setting requires, with remaining approximately 30% data to training after neural network model carry out simulating, verifying, the indexs such as the consensus forecast precision in the current precision of analysis simulation calculating, precision of prediction and the predicted time range scale of different time points, and further improve the overall quality of improving RBF neural network prediction.

Advantage of the present invention is: by the modeling of the features such as the position to cloud cluster, speed, shape, cloud cluster quality and thickness, carry out neural network off-line training based on historical data and set up the mapping relations that signature analysis and distributed photovoltaic system ultra-short term are exerted oneself, and carry out on-line prediction compensation in conjunction with real-time weather data, the prediction of exerting oneself of the ultra-short term that can realize distributed photovoltaic system, its prediction index can reach temporal resolution <5 second, time scale 0.5～5 minute, predicated error <8%.

Brief description of the drawings

Fig. 1: system model structural drawing of the present invention

Fig. 2: shape one dimension of the present invention is described

Fig. 3: gray scale demixing technology mapping function of the present invention

Fig. 4: Features of Cloud Cluster Causing of the present invention extracts and prediction general flow chart

Fig. 5: cloud cluster occlusion effect figure of the present invention

Fig. 6: ultra-short term of the present invention is exerted oneself and predicted RBF neural network model

Fig. 7 On-line Fuzzy compensation principle of the present invention

Concrete implementing method:

Below in conjunction with accompanying drawing, process of the present invention is further described.

With reference to Fig. 2～Fig. 7, a kind of computer vision analysis means that adopt are carried out key feature analysis to the sparse cloud cluster image sequence of obtained sky, so to distributed photovoltaic system carry out ultra-short term exert oneself prediction method, comprise the following steps:

1) set up target cloud cluster position and rate pattern

First obtain continuous image sequence by camera monitoring device.Because the change of shape of cloud cluster is all generally marginal variation, therefore want to extract accurately a representative point, should eliminate the impact of edge on cloud cluster shape by the pre-service of image, extract the nucleus of cloud cluster, obtain continuous shape approximation figure, then obtain these shape approximations figure center-of-mass coordinate separately.The object of processing is like this to reduce the continuous growth and decline of cloud cluster and cause shape constantly to change caused cloud cluster barycenter error, can reduce the error of subsequent prediction.

Cloud cluster is adopted to opening operation operation, cloud cluster is first corroded the general character of rear expansion extraction cloud cluster, first adopt a relatively large disc structure unit to corrode cloud cluster, if the point set that A is cloud cluster, B is structural elements, B is exactly a B with z translation does not share all some z of any common element set with background to the corrosion operation of A, and formula is:

F＝AΘB＝{z|(B) _z∩A ^c＝φ} (1-1)

Wherein, A ^cbe the supplementary set of A, φ is empty set.

Owing to adopting a larger structural elements to corrode operation to cloud cluster, most edge has all been eliminated, remaining cloud cluster nucleus, i.e. the most concentrated region of cloud cluster, then carries out expansive working to the cloud cluster through excessive erosion, and formula is:

$G=F\&CirclePlus;B=\left\{z\right|{\left(B\right)}_z\&cap;A\&NotEqual;\mathrm{\&phi;}\}---(1-2)$

Has the set of the overlapping all some z of element at least with the B of z translation and F.

Extracted by image pre-service after the nucleus of cloud cluster, the motion conditions of then finding a particular point in cloud cluster nucleus represents the motion conditions of whole cloud cluster, the i.e. barycenter of cloud cluster nucleus.Center-of-mass coordinate formula is:

$\stackrel{\&OverBar;}{x}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{x}_i---(1-3)$

$\stackrel{\&OverBar;}{y}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{y}_i---(1-4)$

Obtain the average coordinates of cloud cluster barycenter by above-mentioned formula (1-3) and two average formulas of formula (1-4); Wherein ( ) be exactly the center-of-mass coordinate calculating, x _iand y _irepresent respectively cloud cluster nucleus each point x and y direction coordinate in cloud atlas, n represents the pixel number of cloud cluster nucleus, and before and after adopting, the barycenter of a few width sequential chart pictures extraction cloud clusters carries out the movement velocity of vectorization tracking calculating cloud cluster; That is:

represent t _nthe horizontal ordinate of moment cloud cluster barycenter, represent t _nthe ordinate of moment cloud cluster barycenter; represent t _n+1the horizontal ordinate of moment cloud cluster barycenter, represent t _n+1the ordinate of moment cloud cluster barycenter; t _nwith t _n+1be separated by the unit interval, t _n+1movement velocity v, the t of moment cloud cluster _n+1moment and t _nline between moment cloud cluster barycenter and the angle β of X-axis are expressed as:

$v=\sqrt{{({\stackrel{\&OverBar;}{x}}_{n+1}-{\stackrel{\&OverBar;}{x}}_n)}^2+{({\stackrel{\&OverBar;}{y}}_{n+1}-{\stackrel{\&OverBar;}{y}}_n)}^2}---(1-5)$

Formula (1-5) and formula (1-6) can go out cloud cluster motion conditions within a certain period of time by quantificational expression.

Then the cloud cluster centroid position of a timing series is extracted, then centroid position is carried out to the position of quadratic polynomial matching predict future moment cloud cluster.Extract t ₀, t ₁, t ₂..., t _n-1cloud cluster centroid position in the n width sequential cloud cluster figure that moment gathers, prediction t _nmoment cloud cluster barycenter.Quadratic polynomial model of fit is taking time t as independent variable, the horizontal ordinate x of cloud cluster barycenter, and ordinate y is dependent variable, formula can be expressed as:

$\left[\begin{array}{c}x\\ y\end{array}\right]=\left[\begin{array}{ccc}{a}_1& b_1& c_1\\ a_2& b_2& c_2\end{array}\right]\left[\begin{array}{c}{t}^2\\ t\\ 1\end{array}\right]---(1-7)$

The matrix function of formula (1-7) has represented respectively above-mentioned sequential cloud cluster barycenter x axle, the matched curve of y axial coordinate, in order to obtain optimum fit curve, determines the value of coefficient, adopt the optimum criterion that least square method is optimum fit curve, formula can be expressed as:

In formula (1-8), represent respectively the error of fitting value of x direction and the error of fitting value of y direction, x _i, y _irepresent respectively true x direction coordinate and the y direction coordinate of i-1 moment cloud cluster barycenter.Ask extremum method according to three meta-functions, to a in the axial matched curve of x in formula (1-8) ₁, b ₁, c ₁obtain respectively partial derivative, in the axial matched curve of y to a ₂, b ₂, c ₂obtaining respectively partial derivative obtains:

Make formula (1-9) and formula (1-10) be equal to zero:

$\left.\begin{array}{c}\underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{x}_i{t}_i^2-a_1\underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_i^4-b_1\underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_i^3-c_1\underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_i^2=0\\ \underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{x}_i{t}_i-a_1\underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_i^3-b_1\underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_i^2-c_1\underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_i=0\\ \underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{x}_i-a_1\underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_i^2-b_1\underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_i-n{c}_1=0\end{array}\right\}---(1-11)$

$\left.\begin{array}{c}\underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{y}_i{t}_i^2-a_2\underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_i^4-b_2\underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_i^3-c_2\underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_i^2=0\\ \underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{y}_i{t}_i-a_2\underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_i^3-b_2\underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_i^2-c_2\underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_i=0\\ \underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{y}_i-a_2\underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_i^2-b_2\underset{i=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_i-n{c}_2=0\end{array}\right\}---(1-12)$

By formula (1-11) and (1-12) obtain a ₁, b ₁, c ₁, a ₂, b ₂, c ₂, determined thus the optimum fit curve of this sequential chart x direction and y direction, by this optimum fit curve prediction t _nthe x direction coordinate of moment cloud cluster position and y direction coordinate.

2) set up target cloud cluster shape and deformation descriptive model.

Utilize the cloud cluster feature different from the color of sky to extract the profile of cloud cluster, then reflect to the distance of cloud cluster profile and variation thereof by barycenter.The resemblance of cloud cluster is described with barycenter angle function: borderline point is to the distance r of barycenter, as the function r (θ) of angle theta.If Fig. 2 (a) is irregular figure profile and corresponding parameter, Fig. 2 (b) is distance that irregular figure profile is corresponding and the one dimension funtcional relationship of angle.Specific practice: convert all points that represent with rectangular coordinate system XOY on the profile of gained cloud cluster to point that the polar coordinate system taking barycenter as initial point represents, formula is as follows:

$r_p=\sqrt{{(X_p-\stackrel{\&OverBar;}{x})}^2+{(Y_p-\stackrel{\&OverBar;}{y})}^2}---(2-1)$

(X in formula (2-1), (2-2) _p, Y _p) represent the cloud cluster nucleus frontier point in rectangular coordinate system, r _pbe illustrated in frontier point (X in polar coordinate system _p, Y _p) to the distance of barycenter, θ _prepresent in polar coordinate system the angle of arbitrfary point C and center of mass point line and pole axis on border.

Because the time interval of taking is relatively little, so the change of shape amplitude of cloud cluster is unlikely to excessive, the deformation of cloud cluster has certain Changing Pattern and trend.By by the cloud cluster shape description of a timing series out after, profile during every Zhang Yun in a timing series is cliqued graph is made following value: to pass the horizontal line of barycenter and the right intersection point of profile as starting point, with the point on profile and barycenter line corresponding 120 corresponding point of arithmetic progression of getting from 1 ° to 360 ° of interval angle with 1 °, then taking the time as horizontal ordinate, each corresponding point of profile in the cloud cluster figure of a timing series are carried out to quadratic polynomial matching, the matching reference point on the profile that moment cloud cluster may occur that makes new advances, again these points are connected, just predict the shape that makes new advances and carve cloud cluster for the moment.To the t gathering ₀, t ₁, t ₂..., t _n-1cloud cluster point in moment n width sequential cloud cluster figure is carried out corresponding quadratic polynomial matching, and establishing first point is m ₁, second point is m ₂, the like, the 120th point is m ₁₂₀, quadratic polynomial model of fit is taking time t as independent variable, i point m _ihorizontal ordinate X _i, ordinate Y _ifor dependent variable:

$\left[\begin{array}{c}{X}_i\\ Y_i\end{array}\right]=\left[\begin{array}{ccc}{a}_{i1}& b_{i1}& c_{i1}\\ a_{i2}& b_{i2}& c_{i2}\end{array}\right]\left[\begin{array}{c}{t}^2\\ t\\ 1\end{array}\right]---(2-3)$

The matrix of formula (2-3) has represented respectively the x axle of point, and the matched curve of y axle, in order to obtain optimum fit curve, is determined the value of coefficient, adopts the optimum criterion that least square method is optimum fit curve, can representation formula be:

In formula (2-4), represent respectively the error of fitting value of i point of x direction and the error of fitting value of i point of y direction, X _ij, Y _ijrepresent respectively true x direction coordinate and the y direction coordinate of i the point of j width figure.Ask extremum method according to three meta-functions, to a in the axial matched curve of formula (2-4) x _i1, b _i1, c _i1obtain respectively partial derivative, a in the axial matched curve of y _i2, b _i2, c _i2obtaining respectively partial derivative obtains:

Make formula (2-5) and formula (2-6) be equal to zero:

$\left.\begin{array}{c}\underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{X}_{\mathrm{ij}}{t}_j^2-a_{i1}\underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_j^4-b_{i1}\underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_j^3-c_{i1}\underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_j^2=0\\ \underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{X}_{\mathrm{ij}}{t}_j-a_{i1}\underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_j^3-b_{i1}\underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_j^2-c_{i1}\underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_j=0\\ \underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{X}_{\mathrm{ij}}-a_{i1}\underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_j^2-b_{i1}\underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_j-n{c}_{i1}=0\end{array}\right\}---(2-7)$

$\left.\begin{array}{c}\underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{Y}_{\mathrm{ij}}{t}_j^2-a_{i2}\underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_j^4-b_{i2}\underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_j^3-c_{i2}\underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_j^2=0\\ \underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{Y}_{\mathrm{ij}}{t}_j-a_{i2}\underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_j^3-b_{i2}\underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_j^2-c_{i2}\underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_j=0\\ \underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{Y}_{\mathrm{ij}}-a_{i2}\underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_j^2-b_{i2}\underset{j=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{t}_j-n{c}_{i2}=0\end{array}\right\}---(2-8)$

By formula (2-7) and (2-8) obtain a _i1, b _i1, c _i1, a _i2, b _i2, c _i2, determined thus i point m _ix direction and the optimum fit curve of y direction.

3) cloud cluster qualitative character and mathematical description thereof

Cloud cluster quality is exactly the block ability of unit area cloud cluster to solar radiation.The factor that affects cloud cluster quality is mainly the dense degree of thickness and the cloud cluster of cloud cluster.Sparse cloud cluster is generally frequent cumulus humilis and the fracto-cumulus occurring of sunny weather in summer.When these two kinds of clouds occur, general low air layer is more stable, fine.Therefore the relation of the quality of cloud cluster and the bright-dark degree of cloud cluster will be discussed in two kinds of situation.Generally can determine cloud cluster thickness and dense degree by the bright-dark degree of cloud cluster.That is to say that cloud cluster quality can reflect by the color of cloud cluster.Cloud cluster quality is larger, and it is gloomy that cloud cluster color more seems.But because some cloud mass marginal portion also becomes canescence, this is just to determining that by the color of cloud cluster cloud cluster quality causes certain interference, whether the grey dark areas that therefore first should judge cloud cluster is the borderline region of cloud cluster, if near the border of gloomy region in cloud cluster, just at this moment can not use, cloud cluster quality is above larger, the more aobvious gloomy method of cloud cluster color is judged.And the cloud of borderline region is substantially all " empty cloud ", often, as one deck tulle, at this moment cloud cluster quality is relatively little, gets a definite value that represents cloud cluster quality therefore can near region, cloud cluster border.Therefore before the color by cloud cluster judges cloud cluster quality, first will obtain cloud cluster subregion relative position of living in, whether cloud cluster subregion is near border.

First judge near some empty clouds that cloud cluster border is, with each frontier point b of cloud cluster _ifor the center of circle, setting border near zone is with b _ifor the center of circle, the circle that r is radius and the borderline region that is cloud cluster with the overlapping region of cloud cluster, i.e. the region at empty cloud place, the circle taking radius as r, home position was along the Boundary Moving of cloud cluster one week, and the region that now formed and the common factor of cloud cluster are territory, empty cloud sector:

I＝{z|(B) _z∩A≠φ} (3-1)

Wherein A represents the point set of cloud cluster, (B) _zthe point set in the region after the circle of representative taking r as radius streaks along cloud cluster border, I represents the point set of the empty cloud region of cloud cluster.

Then the cloud cluster quality of remaining cloud cluster part and the color of cloud cluster are carried out correspondingly, by different gray-scale values, cloud cluster quality is carried out to quantitative analysis thus.Obtain the gray-scale map of cloud cluster by pre-service, use cluster analysis that cloud cluster is divided into several different gray-scale value regions, the gray-scale value zoning of rolling into a ball by continuous several Zhang Yuns is analyzed, the basic zoning of predict future moment cloud cluster gray-scale value, draw again the gray-scale value of the effective occlusion area of cloud cluster, by shielded area and corresponding grey scale value are carried out to Integral Processing, finally draw the efficiency of blocking of following a certain moment cloud cluster.By cluster analysis, cloud cluster figure is divided after several different gray areas, utilize gray scale split plot design to carry out pseudo-colours processing.

Gray scale split plot design is exactly to represent gray level with [0, N-1], makes x ₀represent black [f (x, y)=0], and make x _n-1represent white [f (x, y)]=N-1.Suppose and in [0, N-1], get M point, this M point is defined as to gray level x ₁, x ₂..., x _m.Then, suppose 0<M<N-1, M point is just divided into gray scale M+1 interval S ₁, S ₂... S _m+1.Gray level is carried out according to following relation to colored assignment:

IF:f(x,y)∈S _k (3-2)

THEN:f(x,y)＝c _k

Wherein c _kbe and k gray area between S _kcorresponding color, S _kby being positioned at x=k-1 and x=k section definition, in Fig. 3, the mapping function of stepped-style has reacted the mapping of gray level to colored assignment visually.

256 gray shade scales are divided into 8 intervals, in each interval, comprise 32 gray shade scales, each interval is similar to corresponding a kind of cloud cluster quality, 8 interval corresponding 8 cloud cluster quality grades, use different colors to represent between different gray areas by pseudocolor image processing, and then represent with different colors the thickness that cloud cluster is different, seem and be very intuitively convenient to again analyze.

Then by every kind of color in the cloud cluster sequential chart of pseudocolour picture processing is extracted respectively to the processing of carrying out predicting shape in similar step 2, obtain position, cloud cluster subregion and the size of following moment every kind of color by processing, by on cloud cluster part inverting to the width figure of every kind of color of prediction gained, just obtain thus the cloud cluster mass distribution situation of following moment cloud cluster again.

4) Features of Cloud Cluster Causing extracts and the prediction of exerting oneself

Obtain one group of sequential cloud cluster figure by cloud layer monitoring device and picture collection device, respectively at t ₀, t ₁, t ₂... t _n-1moment claps, then in conjunction with the algorithm of above-mentioned steps by t ₀, t ₁, t ₂... t _n-1the information such as the cloud cluster position in moment, area, shape, cloud cluster mass distribution situation extract, and dope t _nposition, area, shape and the cloud cluster mass distribution situation at moment cloud cluster place, the effective coverage of causing photovoltaic panel to block in conjunction with corresponding time point again, determine the overlapping area size of corresponding time point cloud cluster and effective coverage and the cloud cluster mass distribution situation of lap, and then by the solar radiation index of photovoltaic panel under the corresponding time point of calculative determination, basic step as shown in Figure 4.

Suppose that the photovoltaic panel total area is S _always, be S through calculating following moment cloud cluster with the area that overlaps of effective occlusion area _effectively, in the area of overlapping region, divide n region by n kind color, every kind of color represents between a kind of gray area, and then represents a kind of cloud cluster quality, therefore represented n kind cloud cluster quality, n kind is blocked ability e ₁, e ₂, e ₃..., e _n, e _irepresent e ₁to e _nin any one, and 0≤e _i< 1, corresponding photovoltaic panel shielded area is respectively S ₁, S ₂, S ₃..., S _n;

If unit area receivable maximum solar value in ground is Q _s,, under the blocking of cloud cluster, the corresponding shielded area of photovoltaic panel is S _ithe solar radiation quantity Q that region receives _ican be expressed as:

Q _i＝Q _s×S _i×(1-e _i)

And the solar radiation index that photovoltaic panel receives can be expressed as:

Wherein represent the solar radiation index that photovoltaic panel receives, scope is [0,1], in the time that fine day is unobstructed be 1, when the strongest black clouds of the ability of blocking causes while blocking completely photovoltaic panel, now be 0, between the partial occlusion between the two for (0,1).

5) the ultra-short term prediction module of exerting oneself

Cloud cluster behavioral characteristics is many and changeable, and the features such as photovoltaic generation output power has again intermittently, random and fluctuation, need regression analysis and input-output mapping network training algorithm effectively and rapidly.Set up photovoltaic generation ultra-short term based on the RBF neural network forecast model of exerting oneself, as shown in Figure 6.Target Features of Cloud Cluster Causing vector ground photovoltaic panel generation solar radiation being blocked taking sparse cloud cluster as training sample, carries out off-line training to this model with the historical data of photovoltaic generating system power stage, forms basic forecast model and network parameter.

Historical data training patterns: the first historical data segmentation to accumulation, use wherein approximately 70% data to carry out RBF neural network prediction algorithm off-line training.After the precision of prediction that reaches setting requires, with remaining approximately 30% data to training after neural network model carry out simulating, verifying, the indexs such as the consensus forecast precision in the current precision of analysis simulation calculating, precision of prediction and the predicted time range scale of different time points, and further improve the overall quality of improving RBF neural network prediction.

Consider the impact of the factor such as temperature, wind speed, online design fuzzy compensation device, taking temperature, wind speed, the output of photovoltaic generating system real power and off-line prognoses system output power etc. as input parameter, adjust in real time the output control parameter of RBF neural network off-line forecast model, improve precision of prediction, as shown in Figure 7.

Fuzzy compensation particular content: the input using meteorological sensor data (as temperature, wind speed etc.), current predicted power, current real output etc. as fuzzy compensation device, judge according to fuzzy logic, adjust in real time the output control parameter of RBF neural network prediction model, to improve the precision of prediction; Upgrade the photovoltaic generating system power stage prediction curve that can obtain in ultra-short term according to obtained target cloud cluster queue simultaneously.

Claims (1)

1. the distributed photovoltaic system ultra-short term based on Features of Cloud Cluster Causing analysis goes out force prediction method, comprises the following steps:

1) the sparse cloud cluster first sky cloud cluster pick-up unit being collected carries out pre-service, extracts the nucleus of cloud cluster as target area, and then the crucial dynamic feature coefficient that it is affected to photovoltaic system ultra-short term power stage carries out mathematical description; Cloud cluster is adopted to opening operation operation, cloud cluster is first corroded the general character of rear expansion extraction cloud cluster, first adopt a relatively large structural elements to corrode cloud cluster, if the point set that A is cloud cluster, B is structural elements, B is exactly the set z translation that is contained in the some z in A in all B to the corrosion operation of A, and formula is:

F＝AΘB＝{z|(B) _z∩A ^c＝φ}

Wherein, A ^cbe the supplementary set of A, φ is empty set;

Owing to adopting a larger structural elements to corrode operation to cloud cluster, most edge has all been eliminated, remaining cloud cluster nucleus, i.e. the most concentrated region of cloud cluster, then carries out expansive working to the cloud cluster through excessive erosion, and formula is:

$G=F\&CirclePlus;B=\left\{z\right|{\left(B\right)}_z\&cap;A\&NotEqual;\mathrm{\&phi;}\}$

Has the set of the overlapping all some z of element at least with the B of z translation and F;

2) by after image pre-service, the motion conditions of then finding a particular point in cloud cluster nucleus represents the motion conditions of whole cloud cluster, the i.e. barycenter of cloud cluster nucleus; Center-of-mass coordinate formula is:

$\stackrel{\&OverBar;}{x}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{x}_i$

$\stackrel{\&OverBar;}{y}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{y}_i$

Obtain the average coordinates of cloud cluster barycenter by above-mentioned two average formulas; Wherein ( ) be exactly the center-of-mass coordinate calculating, x _iand y _irepresent respectively cloud cluster nucleus each point x and y direction coordinate in cloud atlas, n represents the pixel number of cloud cluster nucleus, and before and after adopting, the barycenter of a few width sequential chart pictures extraction cloud clusters carries out the movement velocity of vectorization tracking calculating cloud cluster; That is:

represent t _nthe horizontal ordinate of moment cloud cluster barycenter, represent t _nthe ordinate of moment cloud cluster barycenter; represent t _n+1the horizontal ordinate of moment cloud cluster barycenter, represent t _n+1the ordinate of moment cloud cluster barycenter; t _nwith t _n+1be separated by the unit interval, t _n+1movement velocity v, the t of moment cloud cluster _n+1moment and t _nline between moment cloud cluster barycenter and the angle β of X-axis are expressed as:

$v=\sqrt{{({\stackrel{\&OverBar;}{x}}_{n+1}-{\stackrel{\&OverBar;}{x}}_n)}^2+{({\stackrel{\&OverBar;}{y}}_{n+1}-{\stackrel{\&OverBar;}{y}}_n)}^2}$

$\mathrm{\&beta;}=\arctan \left(\frac{\stackrel{\&OverBar;}{y_{n+1}}-\stackrel{\&OverBar;}{y_n}}{\stackrel{\&OverBar;}{x_{n+1}}-\stackrel{\&OverBar;}{x_n}}\right)$

Above-mentioned two formula can go out cloud cluster motion conditions within a certain period of time by quantificational expression;

Adopt barycenter angle function to describe the shape facility of cloud cluster, taking the barycenter of cloud cluster as limit, horizontal direction sets up polar coordinate system as pole axis, on cloud cluster profile, arbitrfary point C can be represented by polar coordinates function r (θ) to the distance r of barycenter, and the transfer equation that is tied to polar coordinate system from rectangular coordinate is:

$r_p=\sqrt{{(X_p-\stackrel{\&OverBar;}{x})}^2+{(Y_p-\stackrel{\&OverBar;}{y})}^2}$

Wherein, (X _p, Y _p) represent the cloud cluster nucleus frontier point in rectangular coordinate system, r _pbe illustrated in frontier point (X in polar coordinate system _p, Y _p) to the distance of barycenter, θ _prepresent in polar coordinate system the angle of arbitrfary point C and center of mass point line and pole axis on border;

3) application pixel depth difference clustering methodology carries out thickness and digitized description, mainly utilizes gray scale split plot design that gray shade scale is classified, and visual pattern ground represents the cloud cluster quality of cloud cluster zones of different; In the database of experimental system, design cloud cluster behavioral characteristics vector, above-mentioned each behavioral characteristics is formed to unified data structure and stored and call;

First judge near some empty clouds that cloud cluster border is, taking each frontier point bi of cloud cluster as the center of circle, set taking bi as the center of circle, R is the circle of radius and the borderline region that is cloud cluster with the overlapping region of cloud cluster, it is the region at empty cloud place, circle taking radius as R, home position was along the Boundary Moving of cloud cluster one week, and the region that now formed and the common factor of cloud cluster are territory, empty cloud sector:

I＝{z|(B) _z∩A≠φ}

Wherein A represents the point set of cloud cluster, and (B) z represents the point set in the circle region after a week along cloud cluster Boundary Moving taking r as radius, and I represents the point set of the empty cloud region of cloud cluster;

Then the cloud cluster quality of remaining cloud cluster part is carried out corresponding with the color of cloud cluster, by different gray-scale values, cloud cluster quality is carried out to quantitative analysis thus, obtain the gray-scale map of cloud cluster by pre-service, use cluster analysis that cloud cluster is divided into several different gray-scale value regions, the gray-scale value zoning of rolling into a ball by continuous several Zhang Yuns is analyzed, the basic zoning of predict future moment cloud cluster gray-scale value, draw again the gray-scale value of the effective occlusion area of cloud cluster, by shielded area and corresponding grey scale value are carried out to Integral Processing, finally draw the efficiency of blocking of following a certain moment cloud cluster, by cluster analysis, cloud cluster figure is divided after several different gray areas, utilize gray scale split plot design to carry out representing between cloud cluster quality subarea,

Gray scale split plot design is exactly to represent gray level with [0, N-1], makes V ₀represent black [f (x, y)=0], and make V _n-1represent white [f (x, y)]=N-1; Suppose and in [0, N-1], get M point, this M point is defined as to gray level V ₁, V ₂..., V _m.Then, suppose 0<M<N-1, M point is just divided into gray scale M+1 interval S ₁, S ₂... S _m+1; Gray level is carried out according to following relation to colored assignment:

IF:f(x,y)∈S _k

THEN:f(x,y)＝c _k

Wherein c _kbe and k gray area between S _kcorresponding color, S _kby being positioned at x=k-1 and x=k section definition;

256 gray shade scales are divided into 8 intervals, comprise 32 gray shade scales in each interval, each interval is similar to corresponding a kind of cloud cluster quality, 8 interval corresponding 8 cloud cluster quality;

4) Features of Cloud Cluster Causing extracts and the prediction of exerting oneself

Obtain one group of sequential cloud cluster figure by cloud layer monitoring device and picture collection device, respectively at t ₀, t ₁, t ₂... t _n-1moment claps, then in conjunction with the algorithm of above-mentioned steps by t ₀, t ₁, t ₂... t _n-1the information such as cloud cluster position, shape and the cloud cluster mass distribution situation in moment extract, and dope t _nposition, shape and the cloud cluster mass distribution situation at moment cloud cluster place, the effective coverage of causing photovoltaic panel to block in conjunction with corresponding time point again, determine the overlapping area size of corresponding time point cloud cluster and effective coverage and the cloud cluster mass distribution situation of lap, and then by the solar radiation index of photovoltaic panel under the corresponding time point of calculative determination;

Suppose that the photovoltaic panel total area is S _always, be S through calculating following moment cloud cluster with the area that overlaps of effective occlusion area _effectively, in the area of overlapping region, divide n region by n kind color, every kind of color represents between a kind of gray area, and then represents a kind of cloud cluster quality, therefore represented n kind cloud cluster quality, n kind is blocked ability e ₁, e ₂, e ₃..., e _n, e _irepresent e ₁to e _nin any one, and 0≤e _i< 1, corresponding photovoltaic panel shielded area is respectively S ₁, S ₂, S ₃..., S _n;

If unit area receivable maximum solar value in ground is Q _s,, under the blocking of cloud cluster, the corresponding shielded area of photovoltaic panel is S _ithe solar radiation quantity Q that region receives _ican be expressed as:

Q _i＝Q _s×S _i×(1-e _i)

And the solar radiation index that photovoltaic panel receives can be expressed as:

Wherein represent the solar radiation index that photovoltaic panel receives, scope is [0,1], in the time that fine day is unobstructed be 1, when the strongest black clouds of the ability of blocking causes while blocking completely photovoltaic panel, now be 0, between the partial occlusion between the two for (0,1);

5) independently exert oneself forecasting software module and Database Systems thereof of the ultra-short term based on sparse cloud cluster behavioral characteristics of design, collection comprises the historical output power data of the historical weather datas such as temperature, wind speed, solar radiation quantity and photovoltaic system, and imports the ultra-short term prediction module of exerting oneself; Off-line training and the online compensation of identification by Processing Algorithm realize target cloud cluster in this ultra-short term is exerted oneself forecasting software, feature extraction, historical data, finally realize the prediction of exerting oneself of distributed photovoltaic system ultra-short term based on Features of Cloud Cluster Causing analysis.