CN102478584B - Wind power station wind speed prediction method based on wavelet analysis and system thereof - Google Patents

Abstract

The invention relates to a wind power station wind speed prediction method based on wavelet analysis and a system thereof. The method comprises the following steps: according to a specific prediction time interval, determining an input and an output variable of a prediction model; reading a historical wind speed value and correcting an incomplete point in the historical wind speed value so as to acquire a training sample value sequence of a wind speed prediction model; carrying out rapid wavelet decomposition to the training sample value sequence so as to acquire an approximation detail component value sequence; establishing the wind speed prediction model according to the approximation detail component value sequence so as to carry out the wind speed prediction. According to the wind power station wind speed prediction method based on the wavelet analysis and the system of the invention, through the wavelet decomposition, the training sample value sequence is decomposed into different layers according to a scale so that a trend term, a period term and a random term are separated. Each layer is individually analyzed and predicted and finally the corresponding prediction value can be obtained through reconstruction. By using the method, any prediction interval can be selected according to different demands. The wind speed prediction which is many steps ahead and has high precision can be performed.

Description

Method for forecasting based on wavelet analysis and system

Technical field

The present invention relates to method for forecasting and system, more particularly, relate to a kind of method for forecasting and system based on wavelet analysis.

Background technology

Wind-power electricity generation, as one of the most competitive in a short time generation mode, its advantage be renewable, pollution-free, take up an area less, the construction period is short, investment is flexible, automatization level is high, managerial personnel are few etc.But wind is not to exist all the time, it depends primarily on the variation of Air Flow, so the wind energy little randomness energy that is a kind of density.The energy size of its generation is unstable, be subject to geographic restriction serious, and conversion efficiency is low.

China is vast in territory, and shore line is very long, and wind energy resources is abundanter.Along with the development of China's wind-power electricity generation industry, wind-powered electricity generation total installation of generating capacity increases day by day.Big-and-middle-sized wind-powered electricity generation unit generates electricity by way of merging two or more grid systems, and has become the principal mode of world's Wind Power Utilization.Along with grid-connected unit sustainable growth, single-machine capacity improves, machine group performance optimization, and failure rate reduces, and production cost declines, and wind-powered electricity generation slowly possesses the ability with conventional energy resources competition.But due to the characteristic such as randomness, intermittence that wind-powered electricity generation is exerted oneself, must leave enough unit for subsequent use and peak when operation of power networks, system still can stable operation when ensureing that fluctuating widely appears in wind-powered electricity generation.This is the topmost feature that wind-power electricity generation is different from other generation modes, also becomes the topmost problem of restriction wind-power electricity generation large-scale application simultaneously.The solution of present stage is, dispatching center by and wind power plant grid-connected, reading out data upgrades systematic parameter at any time, thus being controlled in the scope that can tackle by the fluctuation that causes of wind-power electricity generation.But along with the increase of wind energy turbine set scale, wind-power electricity generation is also more and more significant on the impact of electric system, has brought larger pressure to operation of power networks.Therefore, in order to improve the utilization ratio of wind energy, more and more wind-power electricity generation enterprise need to predict service accurately, thereby provides by prediction wind resource the curve that generates electricity more accurately, so that regulation and control point power distributing amount, realizes modern wind-powered electricity generation and the operation of tradition generating combined optimization.Wind-force prediction accurately can also help investor to determine and build wherein wind energy turbine set, and helps the better maintenance and management wind-powered electricity generation of the network operator unit of wind energy turbine set.

At present, the prediction of wind speed is mainly depended on to physical prediction model, its calculated amount is large, error accumulation rate is high, and need professional personage to safeguard, and can not meet the demand of wind-powered electricity generation enterprise to wind energy short-time forecast, more can not make meticulous forecast to the wind speed profile within the scope of wind energy turbine set.In recent years, artificial neural network is slowly widely used in wind-force prediction, and it can, according to input and output data Direct Modeling, have very large advantage aspect solution nonlinearity and serious uncertain recurrence.Neural network is of a great variety, but is to adopt which kind of type neural network to particular problem actually, and which kind of network weight learning algorithm, does not all have clear and definite conclusion at present.The problems such as meanwhile, calculated amount is large, speed of convergence is slow, local optimum are also the main difficulties that neural network faces.Based on above consideration, the present invention utilizes wavelet analysis technology in conjunction with neural network model, sets up wind energy turbine set short-term wind-force forecast model, realizes wind speed is accurately estimated.

Summary of the invention

The technical problem to be solved in the present invention is, for prior art, the prediction of wind speed is mainly depended on to physical prediction model, its calculated amount is large, error accumulation rate is high, and need professional personage to safeguard, can not meet the demand of wind-powered electricity generation enterprise to wind energy short-time forecast, more can not make the defects such as meticulous forecast to the wind speed profile within the scope of wind energy turbine set, a kind of method for forecasting and system based on wavelet analysis is provided.

The technical solution adopted for the present invention to solve the technical problems is: construct a kind of method for forecasting based on wavelet analysis, it comprises the following steps:

According to a specific predicted time interval, determine the input and output variable of forecast model;

Read historical wind speed value, revise the Incomplete Point in described historical wind speed value, to obtain the training sample value sequence of forecasting wind speed model;

Described training sample value sequence is carried out to Wavelet fast decomposition, to obtain approximate details component value sequence;

According to described approximate details component value sequence, set up described forecasting wind speed model, to carry out forecasting wind speed.

In method for forecasting of the present invention, also comprise:

The Output rusults of described forecasting wind speed is carried out to wavelet reconstruction, to obtain corresponding forecasting wind speed value after weighting.

In method for forecasting of the present invention, also comprise:

The real-time Wind observation value and the described forecasting wind speed value that gather are compared, when a continuous specific times N relatively in, the average relative error between described real-time Wind observation value and forecasting wind speed value all exceedes 10%, adjusts the weighted value of described forecasting wind speed model, wherein, N is natural number.

In method for forecasting of the present invention,

Use calculating formula calculate the average relative error between described real-time Wind observation value and forecasting wind speed value, wherein, v (t) is the real-time Wind observation value in t moment, v _*(t) be the forecasting wind speed value in t moment.

In method for forecasting of the present invention,

Use calculating formula ${\mathrm{\α}}_i=\left(t\right)=\frac{\underset{j=1,j\≠i}{\overset{2}{\mathrm{\Σ}}}\underset{u=0}{\overset{s}{\mathrm{\Σ}}}\left|e_j(t-u-p)\right|}{\underset{j=1}{\overset{2}{\mathrm{\Σ}}}\underset{u=0}{\overset{s}{\mathrm{\Σ}}}\left|e_j(t-u-p)\right|}$ With $\underset{i=1}{\overset{2}{\mathrm{\Σ}}}{\mathrm{\α}}_i\left(t\right)=1,\∀t,i,k_i\left(t\right)\≥0,$ Adjust the weighted value of described forecasting wind speed model, wherein, α _i(t) be the weight of t moment i submodel, e _j(t-u) is t-u moment model predictive error, and s is cumulative errors time interval, and p is predicted time interval.

In method for forecasting of the present invention,

According to calculating formula revise the Incomplete Point in described historical wind speed value, wherein, t is data Incomplete Point, and v (t) is revised historical wind speed value, t ₁and t ₂for nearest former and later two effective observation stations adjacent with Incomplete Point, and t ₁<t<t ₂, v (t ₁) and v (t ₂) be respectively and observation station t ₁and t ₂corresponding historical wind speed value.

In method for forecasting of the present invention, use the western small echo of many shellfishes to described training sample value sequence, carry out three layer depth Wavelet fast decompositions, to obtain three groups of approximate details component value sequences.

According to another aspect of the present invention, provide a kind of predicting wind speed of wind farm system based on wavelet analysis, it comprises:

Variable determination module, for according to a specific predicted time interval, determines the input and output variable of forecast model;

Read module, for reading historical wind speed value, revises the Incomplete Point in described historical wind speed value, to obtain the training sample value sequence of forecasting wind speed model;

Data decomposition module, for described training sample value sequence is carried out to Wavelet fast decomposition, to obtain approximate details component value sequence;

Modeling and forecasting module, for according to described approximate details component value sequence, sets up described forecasting wind speed model, to carry out forecasting wind speed.

In predicting wind speed of wind farm system of the present invention, also comprise:

Reconstruct weighting block, for the Output rusults of described forecasting wind speed is carried out to wavelet reconstruction, to obtain corresponding forecasting wind speed value after weighting.

In predicting wind speed of wind farm system of the present invention, also comprise:

Weight adjusting module, for the real-time Wind observation value and the described forecasting wind speed value that gather are compared, when a continuous specific times N relatively in, average relative error between described real-time Wind observation value and forecasting wind speed value all exceedes 10%, adjust the weighted value of described forecasting wind speed model, wherein, N is natural number.

Implement method for forecasting and the system based on wavelet analysis of the present invention, there is following beneficial effect: (1) passes through wavelet decomposition, the training sample value sequence that has historical wind speed value to form is become to different levels according to Scale Decomposition, trend term, periodic term and random entry are separated; (2) every one deck is carried out analyzing separately and prediction, can improve the accuracy of prediction; (3) can, according to different application demand, select the predicting interval arbitrarily, carry out leading multistep high precision forecasting wind speed; (4) according to actual observed value and model checking index, the validity of real-time inspection forecast model, on-line study Renewal model weight; (5) provide reliably for yardman makes Optimized Operation decision-making accurately, effective technical support.

Brief description of the drawings

Below in conjunction with drawings and Examples, the invention will be further described, in accompanying drawing:

Fig. 1 is the block diagram that the present invention is based on the predicting wind speed of wind farm system of wavelet analysis;

Fig. 2 is the process flow diagram that the present invention is based on method for forecasting first embodiment of wavelet analysis;

Fig. 3 is the process flow diagram that the present invention is based on method for forecasting second embodiment of wavelet analysis;

Fig. 4 is the process flow diagram that the present invention is based on method for forecasting the 3rd embodiment of wavelet analysis;

Fig. 5 is the schematic diagram of training sample value sequence and approximate details component value sequence in the present invention;

Fig. 6 the present invention is based on the method for forecasting practical application one hour in advance of wavelet analysis and shifts to an earlier date three hours forecasting wind speed result figure.

Embodiment

In order to make object of the present invention, technical scheme and advantage clearer, below in conjunction with drawings and Examples, the present invention is further elaborated.Should be appreciated that specific embodiment described herein, only in order to explain the present invention, is not intended to limit the present invention.

Fig. 1 illustrates the block diagram of the predicting wind speed of wind farm system 1 that the present invention is based on wavelet analysis, should the predicting wind speed of wind farm system 1 based on wavelet analysis comprise modeling and forecasting module 11, variable determination module 12, read module 13, data decomposition module 14, reconstruct weighting block 15 and weight adjusting module 16, wherein, variable determination module 12, data decomposition module 14 and reconstruct weighting block 15 are all connected with modeling and forecasting module 11, read module 13 is connected with data decomposition module 14, and weight adjusting module 16 is connected with reconstruct weighting block 15.It should be noted that in all diagrams of the present invention, the annexation between each equipment is for the needs of clear its information interaction of explaination and control procedure, therefore should be considered as annexation in logic, and should not only limit to physical connection.

At work, variable determination module 12, according to this desired predicted time of predicting wind speed of wind farm system 1 interval based on wavelet analysis, is determined the input and output variable of forecasting wind speed model; Meanwhile, read module 13 will read historical wind speed value from the historical data base of wind energy turbine set data acquisition and supervisor control, and revises the Incomplete Point in historical wind speed value, thereby obtains the training sample value sequence that forecasting wind speed model needs.Then, training sample value sequence is outputed to data decomposition module 13 by read module 13, and data decomposition module 13 is then used discrete wavelet analysis to carry out Wavelet fast decomposition to training sample value sequence, obtains approximate details component value sequence.Thus, modeling and forecasting module 11 just can receive the input and output variable of definite forecast model that variable determination module 12 exports, and the approximate details component value sequence exported of data decomposition module 14, thereby can set up forecasting wind speed model, wind speed is predicted, and output predicts the outcome accordingly.

In order further to improve the work of this predicting wind speed of wind farm system 1 based on wavelet analysis, wavelet reconstruction is carried out in predicting the outcome that modeling and forecasting module 11 is exported by reconstruct weighting block 15, to obtain corresponding forecasting wind speed value after weighted calculation, thereby forecasting wind speed value can be returned to modeling and forecasting module 11, what thus, modeling and forecasting module 11 was exported predict the outcome also can comprise forecasting wind speed value.

Again further, weight adjusting module 16 also can compare the real-time Wind observation value and the forecasting wind speed value that gather, when a continuous specific times N relatively in, average relative error between described real-time Wind observation value and forecasting wind speed value all exceedes 10%, adjust the weighted value of described forecasting wind speed model, thereby make the forecasting wind speed model in modeling and forecasting module more accurate, realize the forecasting wind speed of precision, wherein, N is natural number.

Fig. 2 shows the flow process of method first embodiment of the predicting wind speed of wind farm that the present invention is based on wavelet analysis, the system architecture of the method flow process based on shown in Fig. 1, and detailed process is as follows:

S11: according to a specific predicted time interval, determine the input and output variable of forecast model, understandable, for this specific predicted time interval flexible design according to actual needs, at this, this predicted time interval is limited accordingly;

S12: read historical wind speed value from the historical data base of wind energy turbine set data acquisition and supervisor control, revise the Incomplete Point in described historical wind speed value, to obtain the training sample value sequence of forecasting wind speed model;

In real work, can be according to calculating formula revise the Incomplete Point in described historical wind speed value, wherein, t is data Incomplete Point, and v (t) is revised historical wind speed value, t ₁and t ₂for nearest former and later two effective observation stations adjacent with Incomplete Point, and t ₁<t<t ₂, v (t ₁) and v (t ₂) be respectively and observation station t ₁and t ₂corresponding historical wind speed value.

S13: select discrete wavelet analysis to carry out Wavelet fast decomposition to training sample value sequence, to obtain approximate details component value sequence;

In real work, can use the western small echo of many shellfishes to stating training sample value sequence, carry out three layer depth Wavelet fast decompositions, to obtain three groups of approximate details component value sequences.The public affairs that discrete wavelet wherein decomposes

Formula: $\left\{\begin{array}{c}{c}_k^j=\underset{n}{\mathrm{\&Sigma;}}{h}_{n-2k}{c}_n^{j-1}\\ d_k^j=\underset{n}{\mathrm{\&Sigma;}}{g}_{n-2k}{c}_n^{j-1}\end{array}\right.$

Wherein, n represents the number of list entries; it is the low frequency component after decomposing; it is the high fdrequency component after decomposing; J represents j level wavelet decomposition, in the time of j=0, it is exactly the discrete series of original input signal; h _n-2kit is the scale coefficient of multiresolution analysis; g _n-2kit is the wavelet coefficient of multiresolution analysis.

If Fig. 5 is training sample value sequence and the schematic diagram of carrying out three groups of approximate details component value sequences after three layers of decomposition of the western small echo of many shellfishes, wherein, a is training sample value sequence; B is the approximate details component value sequence after one deck wavelet decomposition; C is the approximate details component value sequence after two layers of wavelet decomposition; D is the approximate details component value sequence after three layers of wavelet decomposition.

S14: according to approximate details component value sequence, set up described forecasting wind speed model, to carry out forecasting wind speed, and Output rusults.

Fig. 3 shows the flow process of method second embodiment of the predicting wind speed of wind farm that the present invention is based on wavelet analysis, the system architecture of the method flow process based on shown in Fig. 1, and detailed process is as follows:

Step S21, S22, S23, S24 in the second embodiment are identical with S11, S12, S13, S14 in the first embodiment respectively; Wherein the difference of the second embodiment and the first embodiment is, has increased step S25, in step S25, the Output rusults of forecasting wind speed is carried out to wavelet reconstruction, to obtain corresponding forecasting wind speed value after weighting.

Fig. 4 shows the flow process of method second embodiment of the predicting wind speed of wind farm that the present invention is based on wavelet analysis, the system architecture of the method flow process based on shown in Fig. 1, and detailed process is as follows:

Step S31, S32 in the 3rd embodiment, S33, S34, S35 are identical with S21, S22, S23, S24, S25 in the second embodiment respectively; Wherein the difference of the 3rd embodiment and the second embodiment is, increase step S36, in step S36, the real-time Wind observation value and the described forecasting wind speed value that gather are compared, when a continuous specific times N relatively in, the average relative error between described real-time Wind observation value and forecasting wind speed value all exceedes 10%, adjusts the weighted value of described forecasting wind speed model, wherein, N is natural number.

In real work, can use calculating formula calculate the average relative error between described real-time Wind observation value and forecasting wind speed value, wherein, v (t) is the real-time Wind observation value in t moment, v _*(t) be the forecasting wind speed value in t moment.Preferably, N can be 10, when continuous ten times relatively in, in real time the average relative error between Wind observation value and forecasting wind speed value all exceedes 10%, adjusts the weighted value of forecasting wind speed model, now computing formula is

In addition, can use calculating formula:

${\mathrm{\&alpha;}}_i=\left(t\right)=\frac{\underset{j=1,j\&NotEqual;i}{\overset{2}{\mathrm{\&Sigma;}}}\underset{u=0}{\overset{s}{\mathrm{\&Sigma;}}}\left|e_j(t-u-p)\right|}{\underset{j=1}{\overset{2}{\mathrm{\&Sigma;}}}\underset{u=0}{\overset{s}{\mathrm{\&Sigma;}}}\left|e_j(t-u-p)\right|}$ With $\underset{i=1}{\overset{2}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_i\left(t\right)=1,\&ForAll;t,i,k_i\left(t\right)\&GreaterEqual;0,$ Adjust the weighted value of described forecasting wind speed model, wherein, α _i(t) be the weight of t moment i submodel, e _j(t-u) is t-u moment model predictive error, and s is cumulative errors time interval, and p is predicted time interval.

Thus, realize the optimization to forecasting wind speed model, thereby can carry out forecasting wind speed more accurately.

As shown in Figure 6, taking certain large-scale wind power field as example, adopt the historical wind speed Value Data of this wind energy turbine set, one hour in advance and prediction in three hours in advance, the validity of the method for forecasting of checking based on wavelet analysis.Specific implementation process is as follows:

1), systematic perspective measured value is spaced apart one hour, within desired one hour in advance and three hours, predicts according to system, determines the input and output variable of forecasting wind speed model;

2), adopt a certain large-scale wind power field historical wind speed Value Data of 6 years, revise the Incomplete Point in historical wind speed Value Data, obtain the training sample value sequence that forecasting wind speed model needs;

3), select discrete wavelet analysis training sample value sequence to be carried out to Wavelet fast decomposition, the approximate details component value sequence obtaining;

4), the approximate details component value sequence of utilizing multilayer perceptron neural network to return training sample value sequence and wavelet decomposition is set up respectively forecast model, and is carried out forecasting wind speed;

5), the result of forecasting wind speed model output is carried out to wavelet reconstruction, after weighting, obtain corresponding forecasting wind speed value;

6), in order to test the robustness of method of this predicting wind speed of wind farm based on wavelet analysis, adopt mean absolute error (MAE) to evaluate prediction effect, computing formula is as follows:

$\mathrm{MAE}=\frac{1}{l}\underset{k=1}{\overset{l}{\mathrm{\&Sigma;}}}|v\left(t\right)-v_{*}\left(t\right)|$

In formula, v (t) is the observed reading in t moment, v _*(t) be the predicted value in t moment, the number that l is future position, the statistical value obtaining is less, illustrates that prediction effect is better, and precision of prediction is higher.In this example, l=24, the statistics obtaining is as following table 1

Table 1 test data Performance Ratio

Forecast model

Neural network model (MAE)

The inventive method (MAE)

One hour in advance	0.4813	0.3229
			Three hours in advance	0.5536	0.4800

As shown above, adopt short-term wind speed forecasting method proposed by the invention, precision of prediction is greatly improved, and has illustrated that this method for forecasting based on wavelet analysis has higher accuracy and reliability.

The foregoing is only preferred embodiment of the present invention, not in order to limit the present invention, all any amendments of doing within the spirit and principles in the present invention, be equal to and replace and improvement etc., within all should being included in protection scope of the present invention.

Claims (4)

1. the method for forecasting based on wavelet analysis, is characterized in that: comprise the following steps:

According to a specific predicted time interval, determine the input and output variable of forecast model;

Read historical wind speed value, revise the Incomplete Point in described historical wind speed value, to obtain the training sample value sequence of forecasting wind speed model; According to calculating formula revise the Incomplete Point in described historical wind speed value, wherein, t is data Incomplete Point, and v (t) is revised historical wind speed value, t ₁and t ₂for nearest former and later two effective observation stations adjacent with Incomplete Point, and t ₁<t<t ₂, v (t ₁) and v (t ₂) be respectively and observation station t ₁and t ₂corresponding historical wind speed value;

Described training sample value sequence is carried out to Wavelet fast decomposition, to obtain approximate details component value sequence;

According to described approximate details component value sequence, set up described forecasting wind speed model, to carry out forecasting wind speed;

The Output rusults of described forecasting wind speed is carried out to wavelet reconstruction, to obtain corresponding forecasting wind speed value after weighting;

The real-time Wind observation value and the described forecasting wind speed value that gather are compared, when a continuous specific times N relatively in, the average relative error between described real-time Wind observation value and forecasting wind speed value all exceedes 10%, adjusts the weighted value of described forecasting wind speed model, wherein, N is natural number;

Use calculating formula ${\mathrm{\&alpha;}}_i=\left(t\right)=\frac{\underset{j=1,j\&NotEqual;i}{\overset{2}{\mathrm{\&Sigma;}}}\underset{u=0}{\overset{s}{\mathrm{\&Sigma;}}}\left|e_j(t-u-p)\right|}{\underset{j=1}{\overset{2}{\mathrm{\&Sigma;}}}\underset{u=0}{\overset{s}{\mathrm{\&Sigma;}}}\left|e_j(t-u-p)\right|}$ With $\underset{i=1}{\overset{2}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_i\left(t\right)=1,\&ForAll;t,i,{\mathrm{\&alpha;}}_i\left(t\right)\&GreaterEqual;0,$ Adjust the weighted value of described forecasting wind speed model, wherein, α _i(t) be the weight of t moment i submodel, e _j(t-u) be the predicated error of t-u moment j submodel, s is cumulative errors time interval, and p is predicted time interval.

2. method for forecasting according to claim 1, is characterized in that,

Use calculating formula calculate the average relative error between described real-time Wind observation value and forecasting wind speed value, wherein, v (t) is the real-time Wind observation value in t moment, v _*(t) be the forecasting wind speed value in t moment.

3. method for forecasting according to claim 1, is characterized in that,

Use the western small echo of many shellfishes to described training sample value sequence, carry out three layer depth Wavelet fast decompositions, to obtain three groups of approximate details component value sequences.

4. the predicting wind speed of wind farm system based on wavelet analysis, is characterized in that: comprising:

Variable determination module, for according to a specific predicted time interval, determines the input and output variable of forecast model;

Read module, for reading historical wind speed value, revises the Incomplete Point in described historical wind speed value, to obtain the training sample value sequence of forecasting wind speed model;

Data decomposition module, for described training sample value sequence is carried out to Wavelet fast decomposition, to obtain approximate details component value sequence; Modeling and forecasting module, for according to described approximate details component value sequence, sets up described forecasting wind speed model, to carry out forecasting wind speed;

Reconstruct weighting block, for the Output rusults of described forecasting wind speed is carried out to wavelet reconstruction, to obtain corresponding forecasting wind speed value after weighting;

And weight adjusting module, for the real-time Wind observation value and the described forecasting wind speed value that gather are compared, when a continuous specific times N relatively in, average relative error between described real-time Wind observation value and forecasting wind speed value all exceedes 10%, adjust the weighted value of described forecasting wind speed model, wherein, N is natural number;

Wherein,

Described read module is according to calculating formula revise the Incomplete Point in described historical wind speed value, wherein, t is data Incomplete Point, and v (t) is revised historical wind speed value, t ₁and t ₂for nearest former and later two effective observation stations adjacent with Incomplete Point, and t ₁<t<t ₂, v (t ₁) and v (t ₂) be respectively and observation station t ₁and t ₂corresponding historical wind speed value;

Described weight adjusting module

Use calculating formula ${\mathrm{\&alpha;}}_i=\left(t\right)=\frac{\underset{j=1,j\&NotEqual;i}{\overset{2}{\mathrm{\&Sigma;}}}\underset{u=0}{\overset{s}{\mathrm{\&Sigma;}}}\left|e_j(t-u-p)\right|}{\underset{j=1}{\overset{2}{\mathrm{\&Sigma;}}}\underset{u=0}{\overset{s}{\mathrm{\&Sigma;}}}\left|e_j(t-u-p)\right|}$ With $\underset{i=1}{\overset{2}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_i\left(t\right)=1,\&ForAll;t,i,{\mathrm{\&alpha;}}_i\left(t\right)\&GreaterEqual;0,$ Adjust the weighted value of described forecasting wind speed model, wherein, α _i(t) be the weight of t moment i submodel, e _j(t-u) be the predicated error of t-u moment j submodel, s is cumulative errors time interval, and p is predicted time interval.