CN105046383A - Real-time wind power predicting method based on ensemble empirical mode decomposition and relevant vector machine - Google Patents

Abstract

The invention discloses a real-time wind power predicting method based on ensemble empirical mode decomposition and a relevant vector machine. The real-time wind power predicting method is characterized by comprising the steps of obtaining and processing data, building a multi-step rolling prediction mode, building wind power prediction modules and prediction evaluating indexes of the ensemble empirical mode decomposition and the relevant vector machine, and the like. The real-time wind power predicting method has the advantages of being scientific, reasonable and capable of reducing influences brought by nonlinearity and nonstability wind power of wind power, meeting prediction precision requirements, facilitating electric power system dispatching, ensuring power energy quality, reducing the running cost, and the like.

Description

A kind of wind power real-time predicting method based on gathering empirical mode decomposition and Method Using Relevance Vector Machine

Technical field

The present invention relates to technical field of wind power, is a kind of wind power real-time predicting method based on gathering empirical mode decomposition and Method Using Relevance Vector Machine.

Background technology

The change of global energy crisis makes the demand of new forms of energy increase sharply; wind energy is the regenerative resource that current most large-scale develops and utilizes potentiality; wind-power electricity generation is the effective way utilizing wind energy on a large scale, the most realistic choice of Ye Shi China energy and the electric power strategy of sustainable development.But in recent years, along with the quick growth of installed capacity of wind-driven power, the ratio of wind-powered electricity generation in electrical network increases year by year.Due to undulatory property and the randomness of wind energy itself, when wind-powered electricity generation penetrates after power exceedes certain value, severe challenge will be brought to the management and running of electric system and the quality of power supply, this severely limits the development of wind-powered electricity generation.If can effectively predict wind power, not only can reduce the margin capacity of electric system, reduce system operation cost, but also effectively can improve the maximum installation ratio of wind-powered electricity generation in electric system, improve wind-powered electricity generation competitive power.

Wind power real-time estimate refers to be predicted the wind power of following 15 minutes to 4 hours from the prediction moment, and sampling interval is 15 minutes, and one time predicted data is 16, it can thus be appreciated that wind power real-time estimate is ultra-short term multi-step prediction.

Summary of the invention

The object of this invention is to provide a kind of scientific and reasonable, non-linear, the non-stationary impact brought of wind power can be reduced, meet precision of prediction requirement, be beneficial to electric power system dispatching, ensure the quality of power supply, reduce operating cost based on set empirical mode decomposition and the wind power real-time predicting method of Method Using Relevance Vector Machine.

The object of the invention is to be realized by following technical scheme: a kind of wind power real-time predicting method based on set empirical mode decomposition and Method Using Relevance Vector Machine, it is characterized in that, it comprises the following steps:

(1) data acquisition and process

Choose the actual wind power data that certain wind energy turbine set whole field unit was the time interval with 15 minutes, historical data is divided into two parts, and front portion is as training sample, and rear portion is as forecast sample;

(2) multistep rolling forecast pattern is set up

When carrying out wind power prediction, general hypothesis current time is designated as i, and sampling interval is designated as i ^*, known historical data y (i-ni ^*), n=0,1,2 ..., N, the value of prediction is y (i+mi ^*), m=1,2 ..., M, M are the step number of multi-step prediction, then the predicted value of the multi-step prediction that rolls can be expressed as:

y _G(i+mi ^*)＝f(y(i-(N-m+1)i ^*),

(1)

y(i)，y _G(i+i ^*)，…，y _G(i+(m-1)i ^*))

(3) wind power prediction model of set empirical mode decomposition and Method Using Relevance Vector Machine is set up

1. set empirical modal is adopted to decompose original series

Set empirical mode decomposition (EEMD) joins on original series by organizing different white noise sequences more, then respectively empirical mode decomposition (EMD) is carried out to it, regard the average of corresponding natural mode of vibration component (IMF) as true component afterwards, each natural mode of vibration component (IMF) need to meet simultaneously signal zero point number and extreme value to count difference one at most; Any time, the average of the envelope defined by local minizing point and maximum point is zero, and namely go up lower curve about coordinate axis Local Symmetric, particular content is:

A. in original series x (t), white noise sequence is added, white noise sequence is obeyed (0, (α ε) ²) normal distribution, obtain multiplexed sequence X (t),

B. empirical mode decomposition (EMD) is carried out to multiplexed sequence X (t), obtain each rank natural mode of vibration component (IMF), now wherein c _it () is each rank natural mode of vibration component (IMF), r _nt () is residual components, represent the average tendency of original signal,

C. repeat step (a), (b) r time, the white noise amplitude at every turn added is different,

D. utilize the zero-mean principle of white Gaussian noise frequency spectrum, natural mode of vibration component (IMF) corresponding to original signal is $c_n\left(t\right)=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}{c}_{i,n}\left(t\right);$

2. Method Using Relevance Vector Machine forecast model is set up

Method Using Relevance Vector Machine (RVM) is a kind of sparse probability model, it is on the architecture basics of Study first, sample data is carried out iteration, the Posterior probability distribution of most parameters is gone to zero, by model rarefaction, Method Using Relevance Vector Machine (RVM) forecast model is set up to each component that set empirical mode decomposition (EEMD) obtains;

3. superpose and final predicting the outcome is obtained to each component;

(4) prediction evaluation index

According to the relevant regulations of Bureau of Energy of the People's Republic of China (PRC), adopt root-mean-square error M, accuracy rate zq and qualification rate r as the index evaluated, when root-mean-square error is less, accuracy rate is larger, qualification rate is larger, precision of prediction is higher, and the calculating formula of each index is as follows:

The root-mean-square error M of whole day prediction:

$M=\sqrt{\frac{1}{96\×16}\underset{i=1}{\overset{96}{\Σ}}\underset{k=1}{\overset{16}{\Σ}}\left(\frac{{(p_{Mi}^k-p_{Li}^k)}^2}{p_{Cap}}\right)}\×100\%---\left(2\right)$

In formula, M is root-mean-square error, the predicted value in kth moment when being i-th measured power, the actual value of kth moment power when being i-th measured power, p _capthe installed capacity of wind-powered electricity generation resultant field,

Per day predictablity rate zq is expressed as:

${\mathrm{zq}}_i=\[1-{\sqrt{\frac{1}{16}\underset{k=1}{\overset{16}{\Σ}}\left(\frac{p_{Mi}^k-p_{Li}^k}{p_{Cap}}\right)}}^2\]\×100\%---\left(3\right)$

$zq=\frac{1}{96}\underset{i=1}{\overset{96}{\Σ}}{\mathrm{zq}}_i---\left(4\right)$

In formula, zq _ibe the accuracy rate of i-th prediction,

Per day prediction qualification rate r is:

$r_i=\frac{1}{16}\underset{k=1}{\overset{16}{\Σ}}{B}_i^k\×100\%$

$(1-\frac{|p_{Mi}^k-p_{Li}^k|}{p_{Cap}})\×100\%\≥85\%,B_i^k=1$

$(1-\frac{|p_{Mi}^k-p_{Li}^k|}{p_{Cap}})\&times;100\%<85\%,B_i^k=0$

$r=\frac{1}{96}\underset{i=1}{\overset{96}{\&Sigma;}}{r}_i---\left(5\right)$

In formula, r _iit is the qualification rate of i-th prediction.

The advantage applies of the wind power real-time predicting method based on gathering empirical mode decomposition and Method Using Relevance Vector Machine of the present invention exists:

1. gather empirical mode decomposition and solve the phenomenons such as end effect that empirical mode decomposition algorithm occurs in decomposable process and modal overlap well, effectively can reduce the non-stationary property of signal;

2. Method Using Relevance Vector Machine is a kind of sparse probability model, has good nonlinear fitting ability;

3. precision of prediction is high;

4. be applicable to the real-time estimate of wind power, make adjustment contributing to electric power system dispatching department in time plan according to the change of wind power, ensureing the quality of power supply, reduce operating cost, to realizing, wind-powered electricity generation standardization is grid-connected most important, and it is scientific and reasonable, and practical value is high.

Accompanying drawing explanation

Fig. 1 is set empirical mode decomposition process flow diagram;

Fig. 2 is under Forecasting Methodology of the present invention, on the sunny side the contrast schematic diagram of wind energy turbine set ultrashort-term wind power prediction curve and real power curve.

Embodiment

Below in conjunction with the drawings and specific embodiments, the wind power real-time predicting method based on set empirical mode decomposition and Method Using Relevance Vector Machine of the present invention is described further.

Wind power real-time predicting method based on gathering empirical mode decomposition and Method Using Relevance Vector Machine of the present invention, comprises the following steps:

(1) data acquisition and process

The measured data that the present invention faces south wind energy turbine set in July, 2012 for Jilin Province is analyzed.This wind energy turbine set installed capacity is 400.5MW, and blower fan quantity is 267, and the rated capacity of separate unit blower fan is 1500kW, and data sampling is spaced apart 15 minutes;

(2) multistep rolling forecast pattern is set up

When carrying out wind power prediction, general hypothesis current time is designated as i, and sampling interval is designated as i ^*, known historical data y (i-ni ^*), n=0,1,2 ..., N, the value of prediction is y (i+mi ^*), m=1,2 ..., M, M are the step number of multi-step prediction, then the predicted value of the multi-step prediction that rolls can be expressed as:

y _G(i+mi ^*)＝f(y(i-(N-m+1)i ^*),

(1)

y(i)，y _G(i+i ^*)，…，y _G(i+(m-1)i ^*))

(3) wind power prediction model of set empirical mode decomposition (EEMD) and Method Using Relevance Vector Machine (RVM) is set up

1. set empirical modal is adopted to decompose original series

Set empirical mode decomposition (EEMD) is a kind of adaptive selection method improved phenomenons such as the modal overlap of Conventional wisdom mode decomposition, can reduce the non-stationary property of signal.In order to obtain the actual signal of data, just need joining on original series by organizing different white noise sequences more, then respectively EMD decomposition being carried out to it, regard the average of corresponding natural mode of vibration component (IMF) as true component afterwards.Each natural mode of vibration component (IMF) need to meet simultaneously signal zero point number and extreme value to count difference one at most; Any time, the average of the envelope defined by local minizing point and maximum point is zero, namely goes up lower curve about coordinate axis Local Symmetric.Set empirical mode decomposition (EEMD) process flow diagram is as Fig. 1, and particular content is:

A. in original series x (t), white noise sequence is added, white noise sequence is obeyed (0, (α ε) ²) normal distribution, obtain multiplexed sequence X (t),

B. EMD decomposition is carried out to multiplexed sequence X (t), obtain each rank natural mode of vibration component (IMF), now wherein c _it () is each rank natural mode of vibration component (IMF), r _nt () is residual components, represent the average tendency of original signal,

C. repeat step (a), (b) r time, the white noise amplitude at every turn added is different,

D. utilize the zero-mean principle of white Gaussian noise frequency spectrum, IMF component corresponding to original signal is

2. Method Using Relevance Vector Machine (RVM) forecast model is set up

Method Using Relevance Vector Machine (RVM) is a kind of sparse probability model, it is on the architecture basics of Study first, sample data is carried out iteration, the Posterior probability distribution of most parameters is gone to zero, by model rarefaction, there is good nonlinear fitting ability, Method Using Relevance Vector Machine (RVM) forecast model is set up to each component that set empirical mode decomposition (EEMD) obtains;

3. superpose and final predicting the outcome is obtained to each component;

(4) prediction evaluation index

According to can Bureau of Energy of the People's Republic of China (PRC) analyze error the evaluation index that the administrative provisions of wind power real-time estimate propose.

Table 1 and table 2 for July 10, respectively to the predicted exactitude evaluation index that No. 87 blower fans and 267, whole field unit wind power carry out under each Forecasting Methodology.

Table 187 unit wind power prediction evaluation of result index

Should 20% be less than to the root-mean-square error of whole day prediction according in " wind farm power prediction forecast management tentative method " that Bureau of Energy of the People's Republic of China (PRC) issues, as can be seen from table 1 and table 2, no matter be single unit or whole field unit, eemd-rvm Forecasting Methodology all meets the demands, not only root-mean-square error is less than 20%, accuracy rate and qualification rate are also all very high, describe the validity of eemd-rvm model.

No. 87 units at the predicted value of the 10th prediction on July 10 and actual value comparison diagram as shown in Figure 2.Can find, predicted value and actual value close, the applicability of eemd-rvm model is described.

Specific embodiment of the present invention has made detailed explanation to content of the present invention, but does not limit to the present embodiment, any apparent change that those skilled in the art's enlightenment according to the present invention is done, and all belongs to the scope of rights protection of the present invention.

Claims (1)

1., based on the wind power real-time predicting method gathering empirical mode decomposition and Method Using Relevance Vector Machine, it is characterized in that, it comprises the following steps:

(1) data acquisition and process

Choose the actual wind power data that certain wind energy turbine set whole field unit was the time interval with 15 minutes, historical data is divided into two parts, and front portion is as training sample, and rear portion is as forecast sample;

(2) multistep rolling forecast pattern is set up

When carrying out wind power prediction, general hypothesis current time is designated as i, and sampling interval is designated as i ^*, known historical data y (i-ni ^*), n=0,1,2 ..., N, the value of prediction is y (i+mi ^*), m=1,2 ..., M, M are the step number of multi-step prediction, then the predicted value of the multi-step prediction that rolls can be expressed as:

$\begin{array}{c}{y}_G(i+{\mathrm{mi}}^{*})=f\left(y(i-(N-m+1)i^{*}\right),\\ ...y\left(i\right),y_G(i+i^{*}),...,y_G(i+(m-1)i^{*}))\end{array}---\left(1\right)$

(3) wind power prediction model of set empirical mode decomposition and Method Using Relevance Vector Machine is set up

1. set empirical modal is adopted to decompose original series

Set empirical mode decomposition (EEMD) joins on original series by organizing different white noise sequences more, then respectively empirical mode decomposition (EMD) is carried out to it, regard the average of corresponding natural mode of vibration component (IMF) as true component afterwards, each natural mode of vibration component (IMF) need to meet simultaneously signal zero point number and extreme value to count difference one at most; Any time, the average of the envelope defined by local minizing point and maximum point is zero, and namely go up lower curve about coordinate axis Local Symmetric, particular content is:

A. in original series x (t), white noise sequence is added, white noise sequence is obeyed (0, (α ε) ²) normal distribution, obtain multiplexed sequence X (t),

B. empirical mode decomposition (EMD) is carried out to multiplexed sequence X (t), obtain each rank natural mode of vibration component (IMF), now wherein c _it () is each rank natural mode of vibration component (IMF), r _nt () is residual components, represent the average tendency of original signal,

C. repeat step (a), (b) r time, the white noise amplitude at every turn added is different,

D. utilize the zero-mean principle of white Gaussian noise frequency spectrum, natural mode of vibration component (IMF) corresponding to original signal is $c_n\left(t\right)=\frac{1}{N}\underset{i=1}{\overset{N}{\&Sigma;}}{c}_{i,n}\left(t\right);$

2. Method Using Relevance Vector Machine forecast model is set up

Method Using Relevance Vector Machine (RVM) is a kind of sparse probability model, it is on the architecture basics of Study first, sample data is carried out iteration, the Posterior probability distribution of most parameters is gone to zero, by model rarefaction, Method Using Relevance Vector Machine (RVM) forecast model is set up to each component that set empirical mode decomposition (EEMD) obtains;

3. superpose and final predicting the outcome is obtained to each component;

(4) prediction evaluation index

According to the relevant regulations of Bureau of Energy of the People's Republic of China (PRC), adopt root-mean-square error M, accuracy rate zq and qualification rate r as the index evaluated, when root-mean-square error is less, accuracy rate is larger, qualification rate is larger, precision of prediction is higher, and the calculating formula of each index is as follows:

The root-mean-square error M of whole day prediction:

$M=\sqrt{\frac{1}{96\&times;16}\underset{i=1}{\overset{96}{\&Sigma;}}\underset{k=1}{\overset{16}{\&Sigma;}}\left(\frac{{\left(p_{Mi}^k-p_{Li}^k\right)}^2}{p_{Cap}}\right)}\&times;100\%---\left(2\right)$

In formula, M is root-mean-square error, the predicted value in kth moment when being i-th measured power, the actual value of kth moment power when being i-th measured power, p _capthe installed capacity of wind-powered electricity generation resultant field,

Per day predictablity rate zq is expressed as:

${\mathrm{zq}}_i=\&lsqb;1-{\sqrt{\frac{1}{16}\underset{k=1}{\overset{16}{\&Sigma;}}\left(\frac{p_{Mi}^k-p_{Li}^k}{p_{Cap}}\right)}}^2\&rsqb;\&times;100\%---\left(3\right)$

$zq=\frac{1}{96}\underset{i=1}{\overset{96}{\&Sigma;}}{\mathrm{zq}}_i---\left(4\right)$

In formula, zq _ibe the accuracy rate of i-th prediction,

Per day prediction qualification rate r is:

$r_i=\frac{1}{16}\underset{k=1}{\overset{16}{\&Sigma;}}{B}_i^k\&times;100\%$

$(1-\frac{\left|p_{Mi}^k-p_{Li}^k\right|}{p_{Cap}})\&times;100\%\&GreaterEqual;85\%,B_i^k=1$

$(1-\frac{\left|p_{Mi}^k-p_{Li}^k\right|}{p_{Cap}})\&times;100\%<85\%,B_i^k=0$

$r=\frac{1}{96}\underset{i=1}{\overset{96}{\&Sigma;}}{r}_i---\left(5\right)$

In formula, r _iit is the qualification rate of i-th prediction.