CN105760945A - Wind power generation power determining method and device - Google Patents

Abstract

The embodiments of the invention provide a wind power generation power determining method and device. The method includes the following steps that: a Gauss regression training model is constructed, and the hyper-parameters of the Gauss regression training model are determined; the hyper-parameters of the Gauss regression training model are screened by adopting a memetic algorithm, and screened-out hyper-parameters are utilized to construct a new Gauss regression training regression model; the new Gauss regression training regression model is utilized to perform computation processing on a first sample data set, so that wind speed data at a wind power generation power prediction time point is obtained; and the new Gauss regression training regression model is utilized to perform computation on the wind speed data at the wind power generation power prediction time point and a second sample data set, so as to obtain predictive wind electricity power. According to the method provided by the technical schemes of the invention, the memetic algorithm is adopt to screen the hyper-parameters of the Gauss regression training model; the screened-out hyper-parameters are utilized to construct the new Gauss regression training regression model; the new Gauss regression training regression model is utilized to predict the wind electricity power; and therefore, the efficiency of the prediction of the wind electricity power can be improved, and the accuracy of prediction results can be improved.

Description

Wind-power electricity generation power determining method and device

Technical field

The present invention relates to technical field of wind power generation, particularly to a kind of wind-power electricity generation power determining method and device.

Background technology

Wind farm grid-connected generating, it is necessary to participate in market price bidding and submit a tender is therefore more and more big to the dependency of wind power prediction and demand degree.It is predicted Power Output for Wind Power Field being alleviate power system frequency modulation, peak regulation pressure, improves a kind of very useful method receiving wind-powered electricity generation ability.Additionally, be predicted for the reference frame providing important of overhauling of planning wind energy turbine set, and then the utilization rate of wind energy and the economic benefit of wind energy turbine set generation to be improved to the power of wind-power electricity generation.Going through exploratory development for many years and technological innovation, be widely used in the wind power forecasting system with China's independent intellectual property right in the power dispatching station that 12 nets are economized, its capacity has exceeded 12GW, runs to dispatching of power netwoks and brings certain benefit.

In machine learning field, Gaussian process is a kind of new machine learning method grown up in Gaussian random process with Bayesian Learning Theory basis, it is based on strict Statistical Learning Theory, process is high-dimensional, small sample, nonlinear challenge have well adapting to property, have been successfully applied in the fields such as nonlinear regression, classification, Multilayer networks.Compared to neutral net (NN), Gaussian process is easier to realize.Comparing support vector machine (SVM), the parameter of Gaussian process can obtain adaptively in modeling process, and can export and have predicting the outcome of probability meaning.Hyper parameter in Gaussian process is chosen the very important impact that predicted the outcome.Carry out in the process of Wind power forecasting in application Gaussian process, owing to the hyper parameter in Gaussian process adopts conjugate gradient method to obtain, need longer calculating time and degree of accuracy not high when finding hyper parameter, make the Wind power forecasting efficiency that application Gaussian process carries out low, it was predicted that result degree of accuracy is low.

Summary of the invention

Embodiments provide a kind of wind-power electricity generation power determining method, to improve Wind power forecasting efficiency, improve the degree of accuracy that predicts the outcome.The method includes: builds Gauss regression training model and determines the hyper parameter of described Gauss regression training model；Adopt close female algorithm to screen the hyper parameter of described Gauss regression training model, and adopt the hyper parameter screened to be built into new Gauss regression training model；Utilize described new Gauss regression training model, be calculated the first sample data set processing, it is thus achieved that the air speed data in prediction wind-power electricity generation power moment, wherein, described first sample data set includes the air speed data in preset period of time；Utilize described new Gauss regression training model, air speed data and the second sample data set to the described prediction wind-power electricity generation power moment are calculated processing, determining wind-power electricity generation power, wherein, described second sample data set includes the wind-power electricity generation related data in preset period of time.

In one embodiment, described first sample data set includes before the described prediction wind-power electricity generation power moment air speed data of each hour in 24 hours.

In one embodiment, described second sample data set includes before the described prediction wind-power electricity generation power moment air speed data of each hour, wind direction data, temperature data and wind-power electricity generation power data in 24 hours.

In one embodiment, close female algorithm is adopted to screen the hyper parameter of described Gauss regression training model, including: for each hyper parameter of described Gauss regression training model, it is determined that the initial value range of this hyper parameter；Circulation performs following steps and this hyper parameter is screened, until it reaches preset termination condition: calculating the fitness value of each numerical value in present initial value scope, this fitness value is for representing that this numerical value meets the degree of hyper parameter requirement；Fitness value according to each numerical value and default selection strategy, select to meet multiple numerical value of described default selection strategy from initial value range；According to the default rule that selects, reselection numerical value from the multiple numerical value selected, the numerical value that reselection is gone out carries out intersection operation and/or mutation operation obtains new numerical value；The numerical value gone out from reselection and new numerical value select the numerical value that fitness value is the highest, forms new initial value range and carry out screening operation.

In one embodiment, reselection numerical value from the multiple numerical value selected, the numerical value that reselection is gone out carries out intersection operation and obtains new numerical value, including: two numerical value of reselection from the multiple numerical value selected, these two numerical value are carried out intersection operation and obtains two new numerical value.

In one embodiment, reselection numerical value from the multiple numerical value selected, the numerical value that reselection is gone out carries out mutation operation and obtains new numerical value, including: one numerical value of reselection from the multiple numerical value selected, this numerical value is carried out mutation operation and obtains a new numerical value.

The embodiment of the present invention additionally provides the determination device of a kind of wind-power electricity generation power, to improve Wind power forecasting efficiency, improves the degree of accuracy that predicts the outcome.This device includes: model construction module, for building Gauss regression training model and determining the hyper parameter of described Gauss regression training model；Screening module, for adopting close female algorithm to screen the hyper parameter of described Gauss regression training model, and adopts the hyper parameter screened to be built into new Gauss regression training model；Forecasting wind speed module, for utilizing described new Gauss regression training model, is calculated the first sample data set processing, it is thus achieved that the air speed data in prediction wind-power electricity generation power moment, wherein, described first sample data set includes the air speed data in preset period of time；Wind-power electricity generation power determination module, for utilizing described new Gauss regression training model, air speed data and the second sample data set to the described prediction wind-power electricity generation power moment are calculated processing, determine wind-power electricity generation power, wherein, described second sample data set includes the wind-power electricity generation related data in preset period of time.

In one embodiment, described screening module, including: initial value range determines unit, for each hyper parameter for described Gauss regression training model, it is determined that the initial value range of this hyper parameter；Screening unit, screens this hyper parameter for circulation, until it reaches preset termination condition, described screening unit, including: computation subunit, for calculating the fitness value of each numerical value in present initial value scope, this fitness value is for representing that this numerical value meets the degree of hyper parameter requirement；First selects subelement, for the fitness value according to each numerical value and default selection strategy, selects to meet multiple numerical value of described default selection strategy from initial value range；Operator unit, is used for selecting rule, reselection numerical value from the multiple numerical value selected according to presetting, and the numerical value that reselection is gone out carries out intersection operation and/or mutation operation obtains new numerical value；Second selects subelement, for selecting the numerical value that fitness value is the highest the numerical value gone out from reselection and new numerical value, forms new initial value range and carries out screening operation.

In one embodiment, described operator unit, specifically for according to described default selection rule, two numerical value of reselection from the multiple numerical value selected, carrying out intersection operation and obtain two new numerical value to these two numerical value.

In one embodiment, described operator unit, also particularly useful for according to described default selection rule, one numerical value of reselection from the multiple numerical value selected, this numerical value is carried out mutation operation and obtains a new numerical value.

In embodiments of the present invention, after building Gauss regression training model, by adopting the hyper parameter of close female algorithm screening Gauss regression training model, and adopt the hyper parameter screened to be built into new Gauss regression training model, and utilize new Gauss regression training model, it is calculated the first sample data set processing, obtain the air speed data in prediction wind-power electricity generation power moment, and utilize new Gauss regression training model, air speed data and the second sample data set to the prediction wind-power electricity generation power moment are calculated processing, it is determined that wind-power electricity generation power.New Gauss regression training model owing to being formed after adopting close female algorithm screening hyper parameter can be found more accurate hyper parameter, calculate the time faster, there is more powerful Nonlinear Processing ability and higher-dimension disposal ability, ask compared with the Gauss regression training model of hyper parameter carries out Wind power forecasting with prior art uses based on conjugate gradient method, the embodiment of the present invention can improve the efficiency of prediction wind-power electricity generation power, improves the degree of accuracy predicted the outcome.

Accompanying drawing explanation

Accompanying drawing described herein is used for providing a further understanding of the present invention, constitutes the part of the application, is not intended that limitation of the invention.In the accompanying drawings:

Fig. 1 is the flow chart of a kind of wind-power electricity generation power determining method that the embodiment of the present invention provides；

Fig. 2 is a kind of flow chart optimizing hyper parameter that the embodiment of the present invention provides；

Fig. 3 is the structured flowchart of the determination device of a kind of wind-power electricity generation power that the embodiment of the present invention provides.

Detailed description of the invention

For making the object, technical solutions and advantages of the present invention clearly understand, below in conjunction with embodiment and accompanying drawing, the present invention is described in further details.At this, the exemplary embodiment of the present invention and explanation thereof are used for explaining the present invention, but not as a limitation of the invention.

In embodiments of the present invention, it is provided that a kind of wind-power electricity generation power determining method, as it is shown in figure 1, the method includes:

Step 101: build Gauss regression training model and determine the hyper parameter of described Gauss regression training model；

Step 102: adopt close female algorithm to screen the hyper parameter of described Gauss regression training model, and adopt the hyper parameter screened to be built into new Gauss regression training model；

Step 103: utilize described new Gauss regression training model, is calculated the first sample data set processing, it is thus achieved that the air speed data in prediction wind-power electricity generation power moment, wherein, described first sample data set includes the air speed data in preset period of time；

Step 104: utilize described new Gauss regression training model, air speed data and the second sample data set to the described prediction wind-power electricity generation power moment are calculated processing, determining wind-power electricity generation power, wherein, described second sample data set includes the wind-power electricity generation related data in preset period of time.

Flow process shown in Fig. 1 is known, in embodiments of the present invention, after building Gauss regression training model, by adopting the hyper parameter of close female algorithm screening Gauss regression training model, and adopt the hyper parameter screened to be built into new Gauss regression training model, and utilize new Gauss regression training model, it is calculated the first sample data set processing, obtain the air speed data in prediction wind-power electricity generation power moment, and utilize new Gauss regression training model, air speed data and the second sample data set to the prediction wind-power electricity generation power moment are calculated processing, determine wind-power electricity generation power.New Gauss regression training model owing to being formed after adopting close female algorithm screening hyper parameter can be found more accurate hyper parameter, calculate the time faster, there is more powerful Nonlinear Processing ability and higher-dimension disposal ability, ask compared with the Gauss regression training model of hyper parameter carries out Wind power forecasting with prior art uses based on conjugate gradient method, the embodiment of the present invention can improve the efficiency of prediction wind-power electricity generation power, improves the degree of accuracy predicted the outcome.

When being embodied as, it is possible to build Gauss regression training model by following steps and determine the hyper parameter of described Gauss regression training model:

(1) definition of Gaussian process

Gaussian process is the stochastic process that any point joint density function belongs to Gauss distribution.Assuming that stochastic process X (t) in one t=0 to t=T interval of observation, it is multiplied by stochastic process X (t) with a certain function g (t), and on this observation interval, product g (t) X (t) is integrated, thus obtain a stochastic variable Y

$Y={\∫}_0^{T}g\left(t\right)X\left(t\right)\mathrm{dt}---(2-1)$

Wherein, Y is called a linear functional of X (t).

In formula (2-1), if weighting function g (t) makes the mean-square value of stochastic variable Y be limited, and it is all Gauss distribution for each g (t), stochastic variable Y, then claiming X (t) is Gaussian process.In other words, if all linear functionals of X (t) are all Gaussian random variables, then X (t) is Gaussian process.

(2) learning process

(X, y), X represents that d × n ties up input matrix, and y represents output vector to assume to have the learning sample collection D=of n observed data.

Assuming that observed value y is by noise retrial, its function output valve f differs ε, namely

Y=f (x)+ε (2-2)

Wherein ε is independent stochastic variable, meets Gauss distribution, and average is 0, and variance isNamely

$\mathrm{\ϵ}~N(0,{\mathrm{\σ}}_n^2)---(2-3)$

The prior distribution of object observing value y is

$y~N(0,K+{\mathrm{\σ}}_n^{2}I)---(2-4)$

In formula (2-4), the covariance matrix that K=K (X, X) is n × n rank symmetric positive definite, any one K in matrix_ijMeasure x_iAnd x_jDependency.

Assume there is new input vector x^*Corresponding output valve y^*, then n training sample output vector y and 1 test sample output valve y^*The associating Gaussian prior formed is distributed as:

$\left[\begin{array}{c}y\\ y^{*}\end{array}\right]~N\left(\begin{array}{c}0,\left[\begin{array}{cc}K(X,X)+{\mathrm{\σ}}_n^{2}I& K(X,x^{*})\\ K{(X,x^{*})}^T& k(x^{*},x^{*})\end{array}\right]\end{array}\right)---(2-5)$

In formula (2-5), K (X, x^*) it is test point x^*Rank, n × 1 covariance matrix with all input point X of training set.k(x^*,x^*) it is test point x^*The covariance of self.Gaussian process can need to select different covariance functions according to difference, but covariance function needs to meet: can both ensure arbitrary point set to produce a non-negative positive definite covariance matrix.

The covariance function adopted in this application is:

$k_y(x_p,x_q)={\mathrm{\σ}}_f^2\exp [-\frac{1}{2l^2}{(x_p-x_q)}^2]+{\mathrm{\σ}}_n^2{\mathrm{\δ}}_{\mathrm{pq}}---(2-6)$

In formula (2-6), x_pAnd x_qIt is the input vector of two arbitrary samples, p, q ∈ { 1 ... n}.σ_f、σ_nAll hyper parameter with l, l for relatedness measure, on duty more big time, represent input with output between dependency more little.σ_fIt is used for controlling local degree of relevancy；σ_nRepresent the variance of noise.δ_pqFor Kronecker symbol, as p=q, δ_pq=1, otherwise δ_pq=0.

The selection of hyper parameter is on the very important impact that predicted the outcome.When training Gauss to return (GP) training pattern, it is possible to obtained by the maximization self adaptation of the log-likelihood of learning sample.Namely set up the log-likelihood function of learning sample, then hyper parameter is asked local derviation, then adopts conjugate gradient optimization to search out the optimal solution of hyper parameter.Wherein, the logarithmic form of likelihood function is:

$L=\log p\left(y\right|X)=-\frac{1}{2}{y}^T{(K+{\mathrm{\σ}}_n^{2}I)}^{-1}y-\frac{1}{2}\log |K+{\mathrm{\σ}}_n^{2}I|-\frac{n}{2}\log 2\mathrm{\π}---(2-7)$

(3) prediction process

GP model integrates, with learning sample, the nonlinear mapping relation obtaining X and y as knowledge source according to Bayesian regression principle, it was predicted that go out and x^*Corresponding most probable output valve.I.e. given new input x^*, training set input value X and target observation value y when, infer y^*Posterior distrbutionp p (the y of maximum possible^*|x^*, X, y), its prediction distribution is also Gaussian:

$p\left(y^{*}\right|x^{*},X,y|~N(\stackrel{\‾}{y}\left(x^{*}\right),\mathrm{\σ}\left(x^{*}\right)))---(2-8)$

y^*Average and variance be:

$\stackrel{\‾}{y}\left(x^{*}\right)=k^T\left(x^{*}\right){(K+{\mathrm{\σ}}_n^{2}I)}^{-1}y---(2-9)$

$\mathrm{\σ}\left(x^{*}\right)=k(x^{*},x^{*})-k^T\left(x^{*}\right){(K+{\mathrm{\σ}}_n^{2}I)}^{-1}k\left(x^{*}\right)$

In formula (2-9), k (x^*)=k (X, x^*), for rank, n × 1 covariance matrix.

On theory of statistics, covariance is the function of a symmetric positive definite, therefore covariance function is equal to kernel function.IfThen formula (2-9) can the form of kernel function represent:

$\stackrel{\‾}{y}\left(x^{*}\right)=\underset{i}{\overset{n}{\mathrm{\Σ}}}\mathrm{\αk}(x_i,x^{*})---(2-10)$

Namely the average of predictive value is exactly the linear combination of kernel function k, linear relationship data is converted to owing to kernel function can will be mapped to the non-linear relation data of feature space, so that the nonlinear problem of complexity is converted into simple linear problem, so Gaussian process is a kind of core learning machine with probability meaning.

So that Gauss regression training model can be found more accurate hyper parameter, calculate the time faster, there is higher Nonlinear Processing ability and higher-dimension disposal ability, in the present embodiment, realize adopting the hyper parameter of close female algorithm screening Gauss regression training model by following steps, namely the hyper parameter of close female algorithm optimization Gauss regression training model is adopted: for each hyper parameter of described Gauss regression training model, it is determined that the initial value range of this hyper parameter；Circulation performs following steps and this hyper parameter is screened, until it reaches preset termination condition: calculating the fitness value of each numerical value in present initial value scope, this fitness value is for representing that this numerical value meets the degree of hyper parameter requirement；Fitness value according to each numerical value and default selection strategy, select to meet multiple numerical value of described default selection strategy from initial value range；According to the default rule that selects, reselection numerical value from the multiple numerical value selected, the numerical value that reselection is gone out carries out intersection operation and/or mutation operation obtains new numerical value；The numerical value gone out from reselection and new numerical value select the numerical value that fitness value is the highest, forms new initial value range and carry out screening operation.

The process of the concrete hyper parameter adopting close female algorithm optimization Gauss regression training model, as in figure 2 it is shown, comprise the following steps:

Step 1: obtain the hyper parameter σ of GP in Gauss regression training model_f、σ_nAnd l, at this with σ_fFor example, σ_fAs population, population data base will be initialized.In population, each individuality is expressed as:

x_i,G(i=1,2 ..., NP)

In formula: i represents this individuality sequence in population；G represents this individual evolution algebraically；NP represents population scale.Then all of initialization population is:

$x_{\mathrm{ji},0}=\mathrm{rand}\left[\mathrm{0,1}\right]\·(x_j^{\left(U\right)}-x_j^{\left(L\right)})+x_j^{\left(L\right)}(i=\mathrm{1,2}\·\·\·\mathrm{NP};j=\mathrm{1,3}\·\·\·D)$

WhereinWithThe respectively maximum in initial population and minima, D represents learning sample collection.

Generate new population, population data base.If maximum update times is Tmax, initial value t=1.

Initiation parameter: arrange maximum evolutionary generation and (or) desired fitness value as evolution end condition, it is determined that fitness function F (x), sets the probability P of intersection operation_cProbability P with mutation operation_m。

Step 2: initialize population: randomly generate solution composition one group possible in potential solution space and initialize population (i.e. initial value range) P={x₁,...,x_n, wherein x_i(i=1 ..., n) represent one and be likely to solve, be called body or chromosome (i.e. a numerical value in initial value range) one by one, n is chromosomal number in population.Chromosome comprises the characteristic information of individuality.Arranging current evolutionary generation is t=0.

Step 3: assessment ideal adaptation degree: the t time evolution starts, according to fitness function, calculates the fitness value of each individuality in population.

Step 4: selection operates: selecting operation is that the fitness value according to each individuality selects the intersection being follow-up that operates individual by Local Search meeting default selection strategy to operate offer " copulation pond " from current population.Preset selection strategy and global convergence and the convergence rate of genetic algorithm are had material impact.Conventional selection strategy such as has roulette selection, algorithm of tournament selection selection etc..

Roulette selection method (RouletteWheelSelection) is a kind of system of selection based on probability, and individual selected probability is directly proportional to its fitness value.If individual x_iFitness value be f (x_i), then its selected probability is:

$P_{\mathrm{xi}}=\frac{f\left(x_i\right)}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}f\left(x_i\right)}---(2-11)$

In formula, n is population scale.Obviously, the individuality that fitness value is more big, its selected probability is more big.

Step 5: intersect and operate: be i.e. two numerical value of reselection from the multiple numerical value selected, carry out these two numerical value intersecting operating and obtain two new numerical value, concrete, intersect operation from by " the copulation pond " that select operation to select, according to certain two individualities of default selection rules selection as male parent, and according to probability P_cJudging whether to perform to intersect to operate, if it is not, then jump to step 6, if so, then two male parent's exchange features information each other carry out intersection operation, generate two new offspring individuals.Four individualities of parent and filial generation participate in competition jointly, and the individuality that fitness value is high is retained, and the individuality that fitness value is low is eliminated.The operation that intersects has uniform crossover (UniformCrossover), the single-point intersection multiple interleaved mode such as (SinglePointCrossover), multiple-spot detection (MultiPointCrossover).

Step 6: mutation operation: i.e. one numerical value of reselection from the multiple numerical value selected, this numerical value being carried out mutation operation and obtains a new numerical value, concrete, mutation operation is from " copulation pond ", according to certain default selection rules selection body one by one as male parent, according to probability P_mDecide whether mutation operation, if it is not, then jump to step 7, if so, then carry out mutation operation according to the Partial Feature information presetting Mutation Strategy amendment male parent, produce a new offspring individual.If the fitness value of offspring individual is individual higher than parent, filial generation retains, and parent is eliminated.Otherwise filial generation is eliminated, and parent retains.Mutation Strategy has Gaussian mutation (GaussianMutation), boundary mutation (BoundaryMutation), uniformly variation (UniformMutation), non-uniform mutation (Non-uniformMutation) etc..Mutation operation maintains the multiformity of population.Choosing of the probability of variation wants suitable, if too big, causes that All Policies trends towards original random search, if too little, it is easy to it would be absorbed in local optimum.

Step 7: judge end condition: judge whether reach set maximum evolutionary generation or obtain the individuality of the fitness value meeting setting.Without meeting end condition, t=t+1, start to repeat from step 3 step, carrying out evolution of future generation, if meeting end condition, then terminating.

When being embodied as, after the hyper parameter adopting close female algorithm optimization Gauss regression training model, namely new Gauss regression training model can be utilized to be calculated the first sample data set processing, predict the air speed data obtaining the prediction wind-power electricity generation power moment, this first sample data set can be then the air speed data of each hour in 24 hours before the prediction wind-power electricity generation power moment, and often group sample set amounts to 24 data.

When being embodied as, after the hyper parameter adopting close female algorithm optimization Gauss regression training model, just can utilize new Gauss regression training model, air speed data and the second sample data set to the prediction wind-power electricity generation power moment are calculated processing, determine wind-power electricity generation power, this second sample data set may include that the air speed data of each hour, wind direction data, temperature data and wind-power electricity generation power data in 24 hours before the prediction wind-power electricity generation power moment, and often group sample set amounts to 24 × 5=120 data.

Based on same inventive concept, the embodiment of the present invention additionally provides the determination device of a kind of wind-power electricity generation power, as described in the following examples.Owing to the principle of the determination device solution problem of wind-power electricity generation power is similar to wind-power electricity generation power determining method, therefore the enforcement of the determination device of wind-power electricity generation power may refer to the enforcement of wind-power electricity generation power determining method, repeats part and repeats no more.Used below, term " unit " or " module " can realize the software of predetermined function and/or the combination of hardware.Although the device described by following example preferably realizes with software, but hardware, or the realization of the combination of software and hardware is also likely to and is contemplated.

Fig. 3 is a kind of structured flowchart of the determination device of the wind-power electricity generation power of the embodiment of the present invention, as shown in Figure 3, including: model construction module 301, screening module 302, forecasting wind speed module 303 and wind-power electricity generation power determination module 304, below this structure is illustrated.

Model construction module 301, for building Gauss regression training model and determining the hyper parameter of described Gauss regression training model；

Screening module 302, is connected with model construction module 301, for adopting close female algorithm to screen the hyper parameter of described Gauss regression training model, and adopts the hyper parameter screened to be built into new Gauss regression training model；

Forecasting wind speed module 303, is connected with screening module 302, for utilizing described new Gauss regression training model, it is calculated the first sample data set processing, obtaining the air speed data in prediction wind-power electricity generation power moment, wherein, described first sample data set includes the air speed data in preset period of time；

Wind-power electricity generation power determination module 304, it is connected with forecasting wind speed module 303, for utilizing described new Gauss regression training model, air speed data and the second sample data set according to the described prediction wind-power electricity generation power moment determine wind-power electricity generation power, wherein, described second sample data set includes the wind-power electricity generation related data in preset period of time.

In one embodiment, described screening module 302, including: initial value range determines unit, for each hyper parameter for described Gauss regression training model, it is determined that the initial value range of this hyper parameter；Screening unit, determine that unit is connected with initial value range, for circulation, this hyper parameter is screened, until reaching preset termination condition, described screening unit, including: computation subunit, for calculating the fitness value of each numerical value in present initial value scope, this fitness value is for representing that this numerical value meets the degree of hyper parameter requirement；First selects subelement, is connected with computation subunit, for the fitness value according to each numerical value and default selection strategy, selects to meet multiple numerical value of described default selection strategy from initial value range；Operator unit, selects subelement to be connected with first, is used for selecting rule, reselection numerical value from the multiple numerical value selected according to presetting, and the numerical value that reselection is gone out carries out intersection operation and/or mutation operation obtains new numerical value；Second selects subelement, is connected with operator unit, for selecting the numerical value that fitness value is the highest the numerical value gone out from reselection and new numerical value, forms new initial value range and carries out screening operation.

In one embodiment, described operator unit, specifically for according to described default selection rule, two numerical value of reselection from the multiple numerical value selected, carrying out intersection operation and obtain two new numerical value to these two numerical value.

In one embodiment, described operator unit, also particularly useful for according to described default selection rule, one numerical value of reselection from the multiple numerical value selected, this numerical value is carried out mutation operation and obtains a new numerical value.

In embodiments of the present invention, after building Gauss regression training model, by adopting the hyper parameter of close female algorithm screening Gauss regression training model, and adopt the hyper parameter screened to be built into new Gauss regression training model, and utilize new Gauss regression training model, it is calculated the first sample data set processing, obtain the air speed data in prediction wind-power electricity generation power moment, and utilize new Gauss regression training model, air speed data and the second sample data set to the prediction wind-power electricity generation power moment are calculated processing, it is determined that wind-power electricity generation power.New Gauss regression training model owing to being formed after adopting close female algorithm screening hyper parameter can be found more accurate hyper parameter, calculate the time faster, there is more powerful Nonlinear Processing ability and higher-dimension disposal ability, ask compared with the Gauss regression training model of hyper parameter carries out Wind power forecasting with prior art uses based on conjugate gradient method, the embodiment of the present invention can improve the efficiency of prediction wind-power electricity generation power, improves the degree of accuracy predicted the outcome.

Obviously, those skilled in the art should be understood that, each module of the above-mentioned embodiment of the present invention or each step can realize with general calculation element, they can concentrate on single calculation element, or it is distributed on the network that multiple calculation element forms, alternatively, they can realize with the executable program code of calculation element, thus, can be stored in storage device is performed by calculation element, and in some cases, shown or described step can be performed with the order being different from herein, or they are fabricated to respectively each integrated circuit modules, or the multiple modules in them or step are fabricated to single integrated circuit module realize.So, the embodiment of the present invention is not restricted to the combination of any specific hardware and software.

The foregoing is only the preferred embodiments of the present invention, be not limited to the present invention, for a person skilled in the art, the embodiment of the present invention can have various modifications and variations.All within the spirit and principles in the present invention, any amendment of making, equivalent replacement, improvement etc., should be included within protection scope of the present invention.

Claims (10)

1. a wind-power electricity generation power determining method, it is characterised in that including:

Build Gauss regression training model and determine the hyper parameter of described Gauss regression training model；

Adopt close female algorithm to screen the hyper parameter of described Gauss regression training model, and adopt the hyper parameter screened to be built into new Gauss regression training model；

Utilize described new Gauss regression training model, be calculated the first sample data set processing, it is thus achieved that the air speed data in prediction wind-power electricity generation power moment, wherein, described first sample data set includes the air speed data in preset period of time；

Utilize described new Gauss regression training model, air speed data and the second sample data set to the described prediction wind-power electricity generation power moment are calculated processing, determining wind-power electricity generation power, wherein, described second sample data set includes the wind-power electricity generation related data in preset period of time.

2. wind-power electricity generation power determining method as claimed in claim 1, it is characterised in that described first sample data set includes before the described prediction wind-power electricity generation power moment air speed data of each hour in 24 hours.

3. wind-power electricity generation power determining method as claimed in claim 1, it is characterized in that, described second sample data set includes before the described prediction wind-power electricity generation power moment air speed data of each hour, wind direction data, temperature data and wind-power electricity generation power data in 24 hours.

4. wind-power electricity generation power determining method as claimed any one in claims 1 to 3, it is characterised in that adopt close female algorithm to screen the hyper parameter of described Gauss regression training model, including:

Each hyper parameter for described Gauss regression training model, it is determined that the initial value range of this hyper parameter；

Circulation performs following steps and this hyper parameter is screened, until it reaches preset termination condition:

Calculating the fitness value of each numerical value in present initial value scope, this fitness value is for representing that this numerical value meets the degree of hyper parameter requirement；

Fitness value according to each numerical value and default selection strategy, select to meet multiple numerical value of described default selection strategy from initial value range；

According to the default rule that selects, reselection numerical value from the multiple numerical value selected, the numerical value that reselection is gone out carries out intersection operation and/or mutation operation obtains new numerical value；

The numerical value gone out from reselection and new numerical value select the numerical value that fitness value is the highest, forms new initial value range and carry out screening operation.

5. wind-power electricity generation power determining method as claimed in claim 4, it is characterised in that reselection numerical value from the multiple numerical value selected, the numerical value that reselection is gone out carries out intersection operation and obtains new numerical value, including:

These two numerical value are carried out intersection operation and obtain two new numerical value by two numerical value of reselection from the multiple numerical value selected.

6. wind-power electricity generation power determining method as claimed in claim 4, it is characterised in that reselection numerical value from the multiple numerical value selected, the numerical value that reselection is gone out carries out mutation operation and obtains new numerical value, including:

One numerical value of reselection from the multiple numerical value selected, carries out mutation operation to this numerical value and obtains a new numerical value.

7. the determination device of a wind-power electricity generation power, it is characterised in that including:

Model construction module, for building Gauss regression training model and determining the hyper parameter of described Gauss regression training model；

Screening module, for adopting close female algorithm to screen the hyper parameter of described Gauss regression training model, and adopts the hyper parameter screened to be built into new Gauss regression training model；

Forecasting wind speed module, for utilizing described new Gauss regression training model, is calculated the first sample data set processing, it is thus achieved that the air speed data in prediction wind-power electricity generation power moment, wherein, described first sample data set includes the air speed data in preset period of time；

Wind-power electricity generation power determination module, for utilizing described new Gauss regression training model, air speed data and the second sample data set according to the described prediction wind-power electricity generation power moment determine wind-power electricity generation power, wherein, described second sample data set includes the wind-power electricity generation related data in preset period of time.

8. the determination device of wind-power electricity generation power as claimed in claim 7, it is characterised in that described screening module, including:

Initial value range determines unit, for each hyper parameter for described Gauss regression training model, it is determined that the initial value range of this hyper parameter；

Screening unit, screens this hyper parameter for circulation, until it reaches preset termination condition, described screening unit, including:

Computation subunit, for calculating the fitness value of each numerical value in present initial value scope, this fitness value is for representing that this numerical value meets the degree of hyper parameter requirement；

First selects subelement, for the fitness value according to each numerical value and default selection strategy, selects to meet multiple numerical value of described default selection strategy from initial value range；

Operator unit, is used for selecting rule, reselection numerical value from the multiple numerical value selected according to presetting, and the numerical value that reselection is gone out carries out intersection operation and/or mutation operation obtains new numerical value；

Second selects subelement, for selecting the numerical value that fitness value is the highest the numerical value gone out from reselection and new numerical value, forms new initial value range and carries out screening operation.

9. the determination device of wind-power electricity generation power as claimed in claim 8, it is characterized in that, described operator unit, specifically for according to described default selection rule, these two numerical value are carried out intersection operation and obtain two new numerical value by two numerical value of reselection from the multiple numerical value selected.

10. the determination device of wind-power electricity generation power as claimed in claim 8, it is characterized in that, described operator unit, also particularly useful for according to described default selection rule, one numerical value of reselection from the multiple numerical value selected, carries out mutation operation to this numerical value and obtains a new numerical value.