Summary of the invention
In order to overcome the above-mentioned deficiencies of the prior art, the present invention provide a kind of probability forecast method for wind power and
System.
A kind of probability forecast method for wind power comprising: from the wind speed probability density function mould pre-established
Wind speed sample is extracted in type and inputs the wind speed wind-powered electricity generation transformation model pre-established, obtains the sample data of wind power;To wind-powered electricity generation
The sample data of power carries out Density Estimator, obtains the Density Estimator Fitted probability density function of wind power;Based on core
Density estimation Fitted probability density function extracts probability forecast result from preset confidence interval.
The wind speed probability density estimation pre-established, comprising: be based on Wind observation data, obtain point of initial wind speed
Cloth fitting function;Fitting of distribution function based on initial wind speed, determines wind speed probability density function.
Based on Wind observation data, the fitting of distribution function of initial wind speed is obtained, comprising: be based on wind speed DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM member
Forecast departure is corrected with the value of forecasting of historical data, mean value and variance needed for obtaining fitting distribution;Based on mean value and side
Difference determines the fitting of distribution function of initial wind speed.
Based on mean value and variance, the fitting of distribution function of initial wind speed is determined, comprising: the fitting of distribution letter of initial wind speed
Several calculation formula, as follows:
In formula, y is to carry out different rank power treated wind speed, f according to practical wind field datakFor wind speed DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM
Member, α are form parameter, and β is scale parameter;
In formula, μkFor mean value, σkFor variance.
Fitting of distribution function based on initial wind speed, determines wind speed probability density function, comprising: point based on initial wind speed
Cloth fitting function calculates latent variable;Using latent variable, the weight of latent variable maximum value is calculated;Based on weight, using pair
Number likelihood function, calculates the linear corrected parameter of variance;Iteration knot of the alternating iteration until weight and the linear corrected parameter
Fruit is less than preset minimal tolerance, then calculates stopping;Based on iteration result, wind speed probability density function is obtained.
Calculate+1 latent variable of jthFormula, it is as follows:
Wherein, g(j)(yt|fkt) it is fitting of distribution function,For k-th of weight iteration j as a result, ytFor observation
Value, n are observation number of days;
+ 1 latent variable of jthThe calculation formula of the weight of maximum value, as follows:
The local derviation calculation formula of log-likelihood function is carried out, as follows:
Obtain the first linear corrected parameter c0With the second linear corrected parameter c1。
Wind speed sample is extracted from the wind speed probability density estimation pre-established inputs the wind speed wind-powered electricity generation pre-established
Transformation model obtains the sample data of wind power, comprising: converts BP neural network using wind-powered electricity generation, is based on wind speed DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM
The wind power data of Wind observation data and synchronization in member, fitting obtain wind speed wind-powered electricity generation transformation model;Using with
Machine sampling extracts the wind speed sample from the wind speed probability density estimation pre-established, obtains wind power forecast.
Density Estimator is carried out to the sample data of wind power, the Density Estimator Fitted probability for obtaining wind power is close
The expression formula for spending function is as follows:
In formula, z is intermediate variable, and h is bandwidth, and K indicates that gaussian kernel function, n indicate sample number of days, XiIndicate i-th of wind
The sample data of electrical power.
A kind of probability forecast system for wind power comprising: wind power sample data obtains module, for from
Wind speed sample is extracted in the wind speed probability density estimation pre-established and inputs the wind speed wind-powered electricity generation transformation model pre-established, is obtained
To the sample data of wind power;Density Estimator Fitted probability density function module, for the sample data to wind power
Density Estimator is carried out, the Density Estimator Fitted probability density function of wind power is obtained;Probability forecast result extraction module,
For being based on Density Estimator Fitted probability density function, probability forecast result is extracted from preset confidence interval.
Wind power sample data obtains module, comprising: wind speed probability density estimation setting up submodule;Wind speed probability
Density function model foundation submodule, comprising: fitting of distribution function unit and wind speed probability density function determination unit;Distribution is quasi-
Function unit is closed, for being based on Wind observation data, obtains the fitting of distribution function of initial wind speed;Wind speed probability density function is true
Order member, for the fitting of distribution function based on initial wind speed, determines wind speed probability density function.
Fitting of distribution function unit, comprising: the fitting of distribution function subelement of mean variance subelement and initial wind speed;?
It is worth variance subelement, corrects forecast departure for the value of forecasting based on wind speed DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM member and historical data, intended
Mean value and variance needed for closing distribution;The fitting of distribution function subelement of initial wind speed is determined for being based on mean value and variance
The fitting of distribution function of initial wind speed.
The expression formula of the fitting of distribution function of initial wind speed, as follows:
In formula, y is to carry out different rank power treated wind speed, f according to practical wind field datakFor wind speed DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM
Member, α are form parameter, and β is scale parameter;
In formula, μkFor mean value, σkFor variance.
Wind speed probability density function determination unit, comprising: latent variable computation subunit, weight calculation subelement, linear
Corrected parameter computation subunit, iteration subelement and probability density function subelement;Latent variable computation subunit, for being based on
The fitting of distribution function of initial wind speed, calculates latent variable;Weight calculation subelement calculates potential for utilizing latent variable
The weight of variable maximum;Linear corrected parameter computation subunit utilizes log-likelihood function, calculating side for being based on weight
The linear corrected parameter of difference;Iteration subelement, for alternating iteration until the weight and linear corrected parameter iteration result
Less than preset minimal tolerance, then stopping is calculated;Probability density function subelement obtains wind for being based on iteration result
Fast probability density function.
+ 1 latent variable of jthExpression formula, it is as follows:
Wherein, g(j)(yt|fkt) it is fitting of distribution function,For k-th of weight iteration j as a result, ytFor observation
Value, n are observation number of days;+ 1 latent variable of jthThe expression formula of the weight of maximum value, as follows:
First linear corrected parameter c0With the second linear corrected parameter c1It is obtained, is expressed by the local derviation of log-likelihood function
Formula is as follows:
Wind power sample data obtains module, further includes: wind speed wind-powered electricity generation transformation model is fitted submodule and wind power
Forecast obtains submodule;Wind speed wind-powered electricity generation transformation model is fitted submodule, for converting BP neural network using wind-powered electricity generation, is based on wind speed
The wind power data of Wind observation data and synchronization in DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM member, fitting obtain wind speed wind-powered electricity generation modulus of conversion
Type;Wind power forecasts to obtain submodule, for using method of random sampling from the wind speed probability density estimation pre-established
Middle extraction wind speed sample obtains wind power forecast.
The expression formula of Density Estimator Fitted probability density function is as follows:
In formula, z is intermediate variable, and h is bandwidth, and K indicates that gaussian kernel function, n indicate sample number of days, XiIndicate i-th of wind
The sample data of electrical power.
Compared with the immediate prior art, technical solution provided by the invention is had the advantages that
1. the present invention extracts what the input of wind speed sample pre-established from the wind speed probability density estimation pre-established
Wind speed wind-powered electricity generation transformation model takes full advantage of the weather unascertained information of DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM offer, is capable of providing reflection wind-powered electricity generation function
The probabilistic continuous probability density function of rate, and using the confidence interval of setting, effectively increase probability forecast result
Accuracy;
2. the present invention nonlinear fitting ability outstanding using neural network, the accuracy of promotion wind-powered electricity generation conversion process, one
Aspect can form the continuous probability density function for describing complete forecasting wind speed uncertain information, on the other hand, according to reality
The amendment of border observation data can also be improved the performance of weather forecast, so that the quality of subsequent wind power forecast input data mentions
It is high.
Specific embodiment
The present invention is described in further detail below in conjunction with the accompanying drawings.
As shown in Figure 1, forecasting procedure of the invention includes: to extract from the wind speed probability density estimation pre-established
Wind speed sample inputs the wind speed wind-powered electricity generation transformation model pre-established, obtains the sample data of wind power;To the sample of wind power
Notebook data carries out Density Estimator, obtains the Density Estimator Fitted probability density function of wind power;Based on Density Estimator
Fitted probability density function extracts probability forecast result from preset confidence interval.
As shown in Fig. 2, algorithm design cycle of the invention is specific as follows:
1, the fitting of distribution of each set member's wind speed variable
Main weather factor in view of influencing wind power output is wind speed, therefore is used by research object of wind speed
Gamma fitting of distribution wind friction velocity probability distribution of various shapes, wherein wind speed can first carry out not according to practical wind field data
Power with order is handled, and present case is fitted only with original wind speed.
Wherein, it is indicated using the probability density function (PDF) that Gamma function obtains are as follows:
In formula, y is to carry out different rank power treated wind speed according to practical wind field data, and α is form parameter, and β is
Scale parameter, α and β can be by mean μskAnd variances sigmakIt acquires, i.e. μk=α β,
To wind speed DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM member { fk, k=1 ..., K } and it is repaired by linear regression according to the value of forecasting of historical data
Positive forecast departure, mean value needed for obtaining fitting distribution;
μk=b0k+b1kfk (1-2)
In formula, k=1 ..., K indicate k-th of set member, b0kIt indicates between DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM member and actual observation wind speed
The fitting parameter and b of linear regression1Indicate DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM member coefficient;
It is analyzed through data, variances sigmakWith fkThere is also certain linear relationships, in tradeoff model complexity and model accuracy
On the basis of, each set member uses unified regression parameter c0And c1, number of parameters is on the one hand significantly reduced, it is on the other hand pre-
It is almost unchanged to work for fruit:
σk=c0+c1fk, k=1 ..., K (1-3)
2, the parameter Estimation of DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM wind speed BMA model
Since what the parameter of the mean value of each set member can be convenient is obtained by linear regression, the remaining needs of model are estimated
The parameter of meter is exactly weight coefficient wi, regression parameter c0And c1, these parameters need according to the principle of maximal possibility estimation from training
Estimate to get in collection data.Assuming that the prediction error of the identical forecasting period in each day is independent from each other, then the logarithm of BMA model
Likelihood function can be written as:
Since log-likelihood function is difficult to be maximized by way of parsing, then the present invention is using ECME (expectation item
Part maximize estimation) algorithm carry out numeralization calculate realize maximize.
Similar with the principle of expectation maximization (EM) algorithm, ECME algorithm is alternately in expectation (E) step and a series of item
Part is iterated calculating between maximizing (CM) step, and calculating desired max log likelihood value every time is all according under current state
Weight calculated.
If observation ytIt is from DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM member fkConditional probability distribution gkt(yt|fkt), then hidden variable
zkt=1, otherwise value is 0.
1) E step: zktEstimated value obtained by following formula:
Wherein, subscript j indicates the iteration result of jth time ECME algorithm, g(j)(yt|fkt) in variance also using jth time
The c estimated0And c1It determines, estimatesValue between 0,1, K estimator and be equal to 1.
2) step 1 of CM: regard log-likelihood function (2-1) as weight w1,…,wKFunction, seek so that (2-1)
The maximum weighted value of value, this can pass through the z of j+1 iterationktEstimated value obtain:
The step 2 of CM: the result with (2-1) is it is known that numeralization estimates so that the maximum c of log-likelihood function0With
c1Log-likelihood function (2-2) is regarded as the linear corrected parameter c of variance by estimated value0And c1Function.
E and CM step alternating iteration is less than preset minimal tolerance until iteration result twice, then calculates stopping, this
Kind iterative algorithm can guarantee that each iterated logarithmic likelihood value is gradually increased, and be finally reached a locally optimal solution, but can not
Guarantee globally optimal solution, and this algorithm is very sensitive for selecting for initial value, for empirically, initial value chooses equal weight,
It is preferable that distribution parameter uses the estimated value of original a certain member to start calculating effect.
3, wind speed probability forecast is distributed the conversion being distributed to wind power probability forecast
Using Monte Carlo method of random sampling from the conditional probability density function of wind speed
It is middle to extract sufficient amount of sample, the knot of wind power forecast is generated using trained wind-powered electricity generation conversion BP neural network model
Fruit.
BP neural network model obtains (wind speed, wind power) training using the data in historical data base, wherein defeated
Enter for wind speed, exports as wind power, the function of wind-powered electricity generation transformation curve may be implemented.
Density Estimator is carried out to the sample point for the wind power forecast that sampling conversion obtains, Density Estimator is fitted general
Rate density expression formula is as follows:
Wherein, z is intermediate variable, and h is bandwidth, and K indicates that gaussian kernel function, n indicate sample data volume, XiIt indicates i-th
The sample data of wind power.
Finally, preset confidence interval is extracted according to the wind power probability density function that fitting obtains, is forecast
As a result.
Preset confidence interval can arbitrarily be set according to requirement of engineering, i.e., the arbitrary value in 0~100%.
Based on the same inventive concept, the present invention also provides a kind of forecast systems for wind power probability, below into
Row explanation.
System provided by the invention includes: that wind power sample data obtains module, for general from the wind speed pre-established
Wind speed sample is extracted in rate density function model and inputs the wind speed wind-powered electricity generation transformation model pre-established, obtains the sample of wind power
Data;Density Estimator Fitted probability density function module carries out Density Estimator for the sample data to wind power, obtains
To the Density Estimator Fitted probability density function of wind power;Probability forecast result extraction module, for being estimated based on cuclear density
Fitted probability density function is counted, probability forecast result is extracted from preset confidence interval.
Wind power sample data obtains module, comprising: wind speed probability density estimation setting up submodule;Wind speed probability
Density function model foundation submodule, comprising: fitting of distribution function unit and wind speed probability density function determination unit;Distribution is quasi-
Function unit is closed, for being based on Wind observation data, obtains the fitting of distribution function of initial wind speed;Wind speed probability density function is true
Order member, for the fitting of distribution function based on initial wind speed, determines wind speed probability density function.
Fitting of distribution function unit, comprising: the fitting of distribution function subelement of mean variance subelement and initial wind speed;?
It is worth variance subelement, corrects forecast departure for the value of forecasting based on wind speed DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM member and historical data, intended
Mean value and variance needed for closing distribution;The fitting of distribution function subelement of initial wind speed is determined for being based on mean value and variance
The fitting of distribution function of initial wind speed.
The expression formula of the fitting of distribution function of initial wind speed, as follows:
In formula, y is to carry out different rank power treated wind speed, f according to practical wind field datakFor wind speed DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM
Member, α are form parameter, and β is scale parameter;
In formula, μkFor mean value, σkFor variance.
Wind speed probability density function determination unit, comprising: latent variable computation subunit, weight calculation subelement, linear
Corrected parameter computation subunit, iteration subelement and probability density function subelement;Latent variable computation subunit, for being based on
The fitting of distribution function of initial wind speed, calculates latent variable;Weight calculation subelement calculates potential for utilizing latent variable
The weight of variable maximum;Linear corrected parameter computation subunit utilizes log-likelihood function, calculating side for being based on weight
The linear corrected parameter of difference;Iteration subelement, for alternating iteration until the weight and linear corrected parameter iteration result
Less than preset minimal tolerance, then stopping is calculated;Probability density function subelement obtains wind for being based on iteration result
Fast probability density function.
+ 1 latent variable of jthExpression formula, it is as follows:
Wherein, g(j)(yt|fkt) it is fitting of distribution function,For k-th of weight iteration j as a result, ytFor observation
Value, n are observation number of days;+ 1 latent variable of jthThe expression formula of the weight of maximum value, as follows:
First linear corrected parameter c0With the second linear corrected parameter c1It is obtained, is expressed by the local derviation of log-likelihood function
Formula is as follows:
Wind power sample data obtains module, further includes: wind speed wind-powered electricity generation transformation model is fitted submodule and wind power
Forecast obtains submodule;Wind speed wind-powered electricity generation transformation model is fitted submodule, for converting BP neural network using wind-powered electricity generation, is based on wind speed
The wind power data of Wind observation data and synchronization in DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM member, fitting obtain wind speed wind-powered electricity generation modulus of conversion
Type;Wind power forecasts to obtain submodule, for using method of random sampling from the wind speed probability density estimation pre-established
Middle extraction wind speed sample obtains wind power forecast.
The expression formula of Density Estimator Fitted probability density function is as follows:
In formula, z is intermediate variable, and h is bandwidth, and K indicates that gaussian kernel function, n indicate sample number of days, XiIndicate i-th of wind
The sample data of electrical power.
It should be understood by those skilled in the art that, embodiments herein can provide as method, system or computer program
Product.Therefore, complete hardware embodiment, complete software embodiment or reality combining software and hardware aspects can be used in the application
Apply the form of example.Moreover, it wherein includes the computer of computer usable program code that the application, which can be used in one or more,
The computer program implemented in usable storage medium (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.) produces
The form of product.
The application is referring to method, the process of equipment (system) and computer program product according to the embodiment of the present application
Figure and/or block diagram describe.It should be understood that every one stream in flowchart and/or the block diagram can be realized by computer program instructions
The combination of process and/or box in journey and/or box and flowchart and/or the block diagram.It can provide these computer programs
Instruct the processor of general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to produce
A raw machine, so that being generated by the instruction that computer or the processor of other programmable data processing devices execute for real
The device for the function of being specified in present one or more flows of the flowchart and/or one or more blocks of the block diagram.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices.
Finally it should be noted that: the above embodiments are merely illustrative of the technical scheme of the present invention and are not intended to be limiting thereof, institute
The those of ordinary skill in category field can still modify to a specific embodiment of the invention referring to above-described embodiment or
Equivalent replacement, these are applying for this pending hair without departing from any modification of spirit and scope of the invention or equivalent replacement
Within bright claims.