CN104899665A - Wind power short-term prediction method - Google Patents

Abstract

The invention relates to the technical field of wind power prediction, and discloses a wind power short-term prediction method. The method uses wind speed as an input, adopts a regression model of a least square support vector machine to predict output power of a wind power plant, and parameters of the regression model of the least square support vector machine are optimized by adoption of a chaotic particle swarm algorithm. The wind power short-term prediction method provided by the invention introduces chaotic motion characteristics into an iterative process, uses ergodicity of chaotic motion to improve a global searching capability of the algorithm in a searching process, overcomes the defects that the particle swarm algorithm is easy to fall into a local extreme point and is slow in convergence and low in precision in a later period of evolution, effectively solves the problem of prematurity of the particle swarm algorithm, can ensure global optimum, and achieves a better prediction effect; the method uses the least square support vector machine to predict, avoids the problem of solving quadratic programming, converts the prediction problem to a process of solving a linear equation set, and the solving process is greatly simplified; and the method adopts single wind speed as input data, and thus a prediction model is simpler.

Description

Short-term wind power prediction method

Technical field

The present invention relates to a kind of wind power forecasting method, particularly, relate to a kind of short-term wind power prediction method.

Background technology

Along with the raising of wind power generating set single-machine capacity and the development of automatic technology, wind generator system also develops from the original user distribution formula energy to centralized large-scale wind power field.Wind-powered electricity generation ratio in electrical network constantly increases, and wind-powered electricity generation grid-connected in a large number brings severe challenge to the management and running of electric system and safety and stability.Effective wind power prediction can reduce electric system margin capacity, reduce system operation cost, alleviate adverse effect that Wind Power Generation on Power System causes, improve the ratio of wind-powered electricity generation in electric system, therefore carries out prediction to wind power and is of great significance.

Current wind power prediction has physical method and statistical method.Physical method is main it is considered that some physical quantitys, the weather data (wind speed, wind direction, air pressure etc.) that such as numerical weather forecast obtains, the information (level line, roughness, barrier etc.) around wind energy turbine set and the technical parameter (taking turns firm height, penetrating coefficient etc.) of Wind turbines.Its objective is the wind speed optimal estimation value finding Wind turbines to take turns firm height to go out, then export with model the error that statistical module reduces to exist, finally calculate the output power of wind energy turbine set according to the powertrace of wind energy turbine set.Because weather forecast only upgrades several times every day, therefore this method is applicable to relatively longer prediction usually.

Statistical prediction methods generally needs a large amount of historical datas to carry out modeling, and statistical method can meet accuracy requirement for the wind power prediction result of several hours in advance, but for shifting to an earlier date predicting the outcome of longer time, precision is inadequate.Statistics Forecasting Methodology conventional at present has Kalman filtering method, time series method, artificial neural network method, support vector machine method, gray model, wavelet analysis etc.These methods deeply expose along with wind power technology the defect being difficult to overcome, and as precision of prediction is poor, speed of convergence is slow, has the shortcomings such as limitation.

Summary of the invention

Object of the present invention is just the shortcoming and defect overcoming above-mentioned prior art, provides that a kind of precision of prediction is high, fast convergence rate, the simple short-term wind power prediction method of calculating.

The present invention's adopted technical scheme that solves the problem is:

Short-term wind power prediction method, using wind speed as input, adopts the output power of regression model to wind energy turbine set of least square method supporting vector machine to predict, the Parameters in Regression Model of least square method supporting vector machine adopts Chaos particle swarm optimization algorithm to be optimized.

As a further improvement on the present invention, described using wind speed as input, adopt the output power of regression model to wind energy turbine set of least square method supporting vector machine to predict, the Parameters in Regression Model of least square method supporting vector machine adopts Chaos particle swarm optimization algorithm to be optimized and specifically comprises the following steps:

Step 1: obtain data, and data are normalized rear as training sample;

Step 2: the parameter of initialization least square method supporting vector machine and Chaos particle swarm optimization algorithm;

Step 3: minimum for objective function with regression error quadratic sum, adopts the Parameters in Regression Model of Chaos particle swarm optimization algorithm to least square method supporting vector machine to be optimized;

Step 4: the parameter generation optimized is returned in the regression model of least square method supporting vector machine, re-starts training;

Step 5: use the output power of regression model to wind energy turbine set training least square method supporting vector machine to predict.

Further, the regression model construction method of described least square method supporting vector machine comprises the following steps:

Adopt a training sample set (x _i, y _i), i=1,2 ..., l, x _iwind speed, y _ibe the measured value of wind power output power, l is the sum that training sample concentrates data point set, utilizes a Nonlinear Mapping sample space is mapped to feature space, and constructs optimal decision function in higher dimensional space:

Utilize structural risk minimization, the optimization aim of least square method supporting vector machine can be expressed as:

Wherein: w is weight vector, || w|| ²it is the complexity of model; C is marginal coefficient, controls the punishment degree to error; ξ _iit is error vector; B is departure;

Utilize Lagrangian method to solve formula (2), can obtain:

In formula: α _i(i=1,2 ..., l) be Lagrange multiplier;

From KKT (Krush-Kuhn-Tucker) condition:

By kernel function, (4) are converted into linear equation to solve, select the good Radial basis kernel function of effect in regression forecasting, definition kernel function is system of linear equations after conversion is:

$\left[\begin{array}{cccc}0& 1& \mathrm{...}& 1\\ 1& K(x_1,x_1)+\frac{1}{C}& \mathrm{...}& K(x_1,x_l)+\frac{1}{C}\\ .& .& & .\\ .& .& & .\\ .& .& & .\\ 1& K(x_l,x_1)+\frac{1}{C}& \mathrm{...}& K(x_l,x_l)+\frac{1}{C}\end{array}\right]\×\left[\begin{array}{c}b\\ {\mathrm{\α}}_i\\ .\\ .\\ .\\ {\mathrm{\α}}_l\end{array}\right]=\left[\begin{array}{c}b\\ y_i\\ .\\ .\\ .\\ y_l\end{array}\right]---\left(5\right);$

Solve above formula by least square method, obtain regression coefficient α _iwith deviation b, just nonlinear regression model can be obtained:

$f\left(x\right)=\underset{i=1}{\overset{1}{\Σ}}{\mathrm{\α}}_{i}K(x,x_i)+b---\left(6\right);$

In formula (6), Radial basis kernel function in formula, σ is core width, || x-x _i|| be two norms;

The regression model of least square method supporting vector machine is c in σ in this formula and formula (2) adopts Chaos particle swarm optimization algorithm to be optimized.

Further, step 3 specifically comprises the following steps:

Step 3.1: adopt chaos system z _n+1=μ z _n(1-z _n) produce chaos sequence initialization is carried out to the speed of particle in algorithm and position, wherein μ is chaos controlling parameter, z _nbe Chaos Variable, n is sequence number;

Step 3.2: traversal search obtains personal best particle, and it is minimum for fitness with regression error quadratic sum to press formula (7), calculates and compares fitness value, iterative search goes out global optimum position;

$\min \underset{i=1}{\overset{1}{\Σ}}{\left(f\right(x_i)-y_i)}^2---\left(7\right);$

In formula: f (x _i) be the predicted value in the i-th moment; x _ifor the input value of the regression model of least square method supporting vector machine, y _iit is the measured value in the i-th moment;

Step 3.3: produce chaos sequence based on global optimum position, using the reposition of a position best in the sequence of generation as one of them particle;

Step 3.4: the speed upgrading other particle by formula (8), upgrades the position of other particles by formula (9);

v _ik(t+1)＝ωv _ik(t)+c ₁δ ₁(p _ik(t)-x _ik(t))+c ₂δ ₂(p _gk(t)-x _ik(t)) (8)

x _ik(t+1)＝x _ik(t)+v _ik(t+1) (9)；

In formula: ω is inertia weight; c ₁, c ₂for Studying factors; δ ₁, δ ₂for the random number between 0 and 1; In addition, the maximal rate of particle should not exceed its maximal rate V _max, x _ikt () is the position of an i moment kth particle, x _ik(t+1) be the position of an i+1 moment kth particle; v _ikt () is the speed of an i moment kth particle, v _ik(t+1) be the speed of an i+1 moment kth particle, p _ikt () is personal best particle, p _gkt () is global optimum position;

Step 3.5: judge whether to meet iterations restriction, be, terminate to optimize and Output rusults, otherwise return step 3.2.

Further, in step 2, described initialization least square method supporting vector machine and the parameter of Chaos particle swarm optimization algorithm comprise the following steps:

Step 2.1, the punishment parameter C determining least square method supporting vector machine and nuclear parameter σ ²scope, the inertia weight ω of initialization least square method supporting vector machine;

Step 2.2, determine the Studying factors c of Chaos particle swarm optimization algorithm ₁, c ₂; Chaos controlling parameter and iterations.

The invention has the beneficial effects as follows:

1, the present invention adopts the parameter of Chaos particle swarm optimization algorithm to support vector machine to be optimized.Particle cluster algorithm is a kind of intelligent optimization algorithm with concurrent operation ability, but this algorithm also existing defects, and algorithm easily falls into local best points in the middle of iteration, makes algorithm Premature Convergence and can not ensure the precision that calculates.Chaos particle swarm optimization algorithm is based upon on chaology, chaos state feature is ergodicity, randomness and regularity, although chaos state is a kind of random state, but can be obtained by the equation determined, and chaotic motion can repeatedly not travel through all states according to a certain rule within the specific limits.Chaotic motion characteristic is incorporated in the middle of iterative process by Chaos particle swarm optimization algorithm, in search procedure, utilize the ergodicity of chaotic motion to improve the ability of searching optimum of algorithm, improve particle cluster algorithm and be easily absorbed in the slow and shortcoming that precision is low of Local Extremum, later stage of evolution convergence, effectively solve " precocity " problem of particle cluster algorithm, can ensure Global Optimality, prediction effect is better.

2, the present invention uses least square method supporting vector machine to predict.When processing the regression problem of non-linear complexity, regression forecasting problem is converted into the problem that solves quadratic programming by support vector machine, can in the hope of globally optimal solution.The forecast model of least square method supporting vector machine then avoids the problem solving quadratic programming, and forecasting problem is converted into the process solving system of linear equations, and its solution procedure simplifies greatly.

3, the present invention adopts single wind speed as input data, and forecast model is simpler.

Accompanying drawing explanation

Fig. 1 is short-term wind power prediction flow process of the present invention;

Fig. 2 is the calculation process of Chaos particle swarm optimization algorithm;

Fig. 3 is the forecast of regression model value comparison diagram of measured value and least square method supporting vector machine.

Embodiment

The present invention is directed to the feature of wind speed intermittence and randomness, adopt a kind of short-term wind power prediction method based on Chaos-Particle Swarm Optimization and least square method supporting vector machine (CPSO-LSSVM) using wind speed as input, with the output power of wind energy turbine set for exporting, the output power of regression model to wind energy turbine set of least square method supporting vector machine is adopted to predict, the Parameters in Regression Model of least square method supporting vector machine adopts Chaos particle swarm optimization algorithm to be optimized, overcome classic method computing method and be absorbed in local minimum, the shortcomings such as the training time is long, there are better Generalization Capability and accuracy.

Below in conjunction with embodiment and accompanying drawing, to the detailed description further of the present invention's do, but embodiments of the present invention are not limited thereto.

As shown in Figure 1, short-term wind power prediction method comprises the following steps:

Step 1: the acquisition of data and normalized.The air speed data x in a certain moment is adopted in the present embodiment _itraining as input data, for predicting the wind speed of subsequent time, being first normalized by formula 10 after getting air speed data, then as training sample T:

$\stackrel{\‾}{x_i}=\frac{x_i-x_{min}}{x_{\max }-x_{min}}---\left(10\right)$

In formula: it is the result after certain the air speed data normalization in training sample T; x _maxand x _minmaximal value and the minimum value of this group variable data in training sample T respectively.In this step, the acquisition of data can directly be extracted from the SCADA system of electric system, also can gather by other means.

Step 2: each parameter of initialization least square method supporting vector machine (LSSVM) and Chaos particle swarm optimization algorithm (CPSO).In this step, each parameter of initialization least square method supporting vector machine and Chaos particle swarm optimization algorithm specifically comprises:

Step 2.1, the punishment parameter C determining LSSVM and nuclear parameter σ ²scope, the inertia weight ω of initialization LSSVM;

Step 2.2, determine the correlation parameter of CPSO, such as but not limited to algorithm parameter, iterations; Algorithm parameter comprises Studying factors c ₁, c ₂; Chaos controlling parameter μ;

Step 2.3, at the punishment parameter C of LSSVM and nuclear parameter σ ²scope in random initializtion population.

Step 3: minimum for objective function with regression error quadratic sum, adopts the parameter of Chaos particle swarm optimization algorithm to least square method supporting vector machine to be optimized.

Step 4: the parameter generation optimized is returned in the regression model of least square method supporting vector machine, obtains forecast model, and training is re-started to forecast model;

Step 5: according to the air speed data obtained, uses the regression model training least square method supporting vector machine to predict as the output power of forecast model to wind energy turbine set.

Wherein, the construction method of the regression model of above-mentioned LSSVM is:

For the training sample set (x obtained in advance _i, y _i), x _iwind speed, y _ibe the measured value of wind power output power, l is the sum that training sample concentrates data point set, utilizes a Nonlinear Mapping sample space is mapped to feature space, and constructs optimal decision function in higher dimensional space:

Utilize structural risk minimization, the optimization aim of least square method supporting vector machine can be expressed as:

Wherein: w is weight vector, for control above wind speed on the impact of current wind speed, || w|| ²it is the complexity of model; C is marginal coefficient, controls the punishment degree to error; ξ _iit is error vector; B is departure.

Utilize Lagrangian method to solve formula (2), can obtain:

In formula: α _i(i=1,2 ..., l) be Lagrange multiplier.

By KKT (Krush-Kuhn-Tucker) condition: can obtain:

$\underset{i=1}{\overset{1}{\Σ}}{\mathrm{\α}}_i=0,$ α _i＝Cξ _i，

Namely formula is obtained:

Finally, need only kernel function be passed through, above-mentioned optimizing process is converted into linear equation, then carry out solving.In the present embodiment, select the good Radial basis kernel function of effect in regression forecasting, definition kernel function is system of linear equations after conversion is:

$\left[\begin{array}{cccc}0& 1& \mathrm{...}& 1\\ 1& K(x_1,x_1)+\frac{1}{C}& \mathrm{...}& K(x_1,x_l)+\frac{1}{C}\\ .& .& & .\\ .& .& & .\\ .& .& & .\\ 1& K(x_l,x_1)+\frac{1}{C}& \mathrm{...}& K(x_l,x_l)+\frac{1}{C}\end{array}\right]\×\left[\begin{array}{c}b\\ {\mathrm{\α}}_i\\ .\\ .\\ .\\ {\mathrm{\α}}_l\end{array}\right]=\left[\begin{array}{c}b\\ y_i\\ .\\ .\\ .\\ y_l\end{array}\right]---\left(5\right)$

Solve above formula by least square method, obtain regression coefficient α _iwith deviation b, just nonlinear regression model can be obtained:

$f\left(x\right)=\underset{i=1}{\overset{l}{\Σ}}{\mathrm{\α}}_{i}K(x,x_i)+b---\left(6\right)$

In formula (6), Radial basis kernel function in formula, σ is core width, || x-x _i|| be two norms;

The regression model of least square method supporting vector machine is c in σ in this formula and formula (2) adopts Chaos particle swarm optimization algorithm to be optimized.

Chaos sequence is produced based on the optimal location that Chaos-Particle Swarm Optimization mainly utilizes the ergodicity of chaotic motion to search by current whole population, again the particle position that the optimal location particle in the chaos sequence produced replaces in current particle group at random, improve that particle cluster algorithm is easily absorbed in Local Extremum, later stage of evolution restrains the low shortcoming of slow precision.

Particularly, in step 3, minimum for objective function with regression error quadratic sum, as shown in Figure 2, step is as follows for the specific algorithm flow process that the Parameters in Regression Model of employing Chaos particle swarm optimization algorithm to least square method supporting vector machine is optimized:

Step 3.1: adopt chaos system z _n+1=μ z _n(1-z _n) produce chaos sequence initialization is carried out to the speed of particle in algorithm and position, wherein μ is chaos controlling parameter, z _nbe Chaos Variable, n is sequence number;

Step 3.2: according to the ergodicity of chaotic motion, traversal search obtains personal best particle p _ik(t), and with regression error quadratic sum minimum be fitness such as formula (7), calculate and compare fitness value, iterative search goes out global optimum position p _gk(t),

$min\underset{i=1}{\overset{1}{\Σ}}{(f\left(x_i\right)-y_i)}^2---\left(7\right)$

In formula: f (x _i) be the predicted value in the i-th moment; x _ifor the input value of the regression model of least square method supporting vector machine, y _iit is the measured value in the i-th moment.

Step 3.3: produce chaos sequence based on global optimum position, using a reposition as wherein any one particle best in the sequence that produces.

Step 3.4: the speed upgrading other particle by formula (3), upgrades the position of other particles by formula (4);

v _ik(t+1)＝ωv _ik(t)+c ₁δ ₁(p _ik(t)-x _ik(t))+c ₂δ ₂(p _gk(t)-x _ik(t)) (3)

x _ik(t+1)＝x _ik(t)+v _ik(t+1) (4)

In formula: ω is inertia weight; c ₁, c ₂for Studying factors; δ ₁, δ ₂for the random number between 0 and 1; In addition, the maximal rate of particle should not exceed its maximal rate V _max, x _ikt () is the position of an i moment kth particle, x _ik(t+1) be the position of an i+1 moment kth particle; v _ikt () is the speed of an i moment kth particle, v _ik(t+1) be the speed of an i+1 moment kth particle, and upgrade global optimum and personal best particle in particle populations.

Step 3.5: judge whether to meet iterations restriction, terminate to optimize and namely Output rusults exports the Parameters in Regression Model after optimization: punishment parameter C and core width cs, otherwise return step 3.2, the iterations such as reaching setting then terminates and Output rusults, otherwise returns step 3.2 and continue iteration optimization.

The present invention adopts the parameter of Chaos particle swarm optimization algorithm to support vector machine to be optimized.Particle cluster algorithm is a kind of intelligent optimization algorithm with concurrent operation ability, but this algorithm also existing defects, and algorithm easily falls into local best points in the middle of iteration, makes algorithm Premature Convergence and can not ensure the precision that calculates.Chaos particle swarm optimization algorithm is based upon on chaology, chaos state feature is ergodicity, randomness and regularity, although chaos state is a kind of random state, but can be obtained by the equation determined, and chaotic motion can repeatedly not travel through all states according to a certain rule within the specific limits.Chaotic motion characteristic is incorporated in the middle of iterative process by Chaos particle swarm optimization algorithm, utilizes the ergodicity of chaotic motion to improve the ability of searching optimum of algorithm in search procedure, effectively solves " precocity " problem of particle cluster algorithm.

The present invention uses least square method supporting vector machine to predict.When processing the regression problem of non-linear complexity, regression forecasting problem is converted into the problem that solves quadratic programming by support vector machine, can in the hope of globally optimal solution.The forecast model of least square method supporting vector machine then avoids the problem solving quadratic programming, and forecasting problem is converted into the process solving system of linear equations, and its solution procedure simplifies greatly.

The present invention also to choose in actual measurement wind-power electricity generation machine data more complete one group and predicts as training sample data, and using the air speed data in a certain moment as input data, the wind speed of prediction subsequent time, tests.

The operating time of the unit chosen is 96 hours, and sampling interval is 10 minutes, and this unit collected is totally 576 groups of wind speed and temperature data, and 576 groups of generated output power data.

Adopt Chaos particle swarm optimization algorithm to be optimized punishment parameter C and core width cs, initialization inertia weight ω is 0.4; Studying factors c ₁=c ₂=0.2; Chaos controlling parameter μ=4; Iterations is 200 times.Optimum results C=75.5839, σ ²=0.1267.The wind speed of model after application training to subsequent time is predicted, and contrasts with measured value, and result is as Fig. 3.

As shown in Figure 3, predicted value and measured value change and reach unanimity, and illustrate that to adopt method predicting reliability of the present invention higher.

Meanwhile, the present invention has also carried out error analysis with following several method:

(1) Prediction sum squares (SSE)

$e_{SSE}=\underset{t=1}{\overset{N}{\Σ}}{(A_i-P_i)}^2---\left(11\right)$

(2) square error (MSE)

$e_{MSE}=\frac{1}{N}\sqrt{\underset{t-1}{\overset{N}{\Σ}}{(A_i-P_i)}^2}---\left(12\right)$

(3) mean absolute percentage error (MAPE)

$e_{MAPE}=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}\left|\frac{A_i-P_i}{A_i}\right|---\left(13\right)$

(4) all square percentage error (MSPE)

$e_{MSPE}=\frac{1}{N}\sqrt{\underset{t-1}{\overset{N}{\Σ}}{((A_i-P_i)/A_i)}^2}---\left(14\right)$

Above-mentioned various in: A _ifor the measured value of future position, P _ifor the predicted value of future position, N is the number of future position.

Error analysis calculation result is as shown in table 1.

Table 1 CPSO-LSSVM analysis indexes

Forecast model	SEE	MSE	MAPE	MSPE
					CPSO-LSSVM	62.4669	0.0659	0.1178	0.0544

As seen from table, the present invention proposes CPSO-LSSVM forecast model and has good precision of prediction and adaptability.

Below be only the preferred embodiment of the present invention, protection scope of the present invention be not only confined to above-described embodiment, all technical schemes belonged under thinking of the present invention all belong to protection scope of the present invention.It should be pointed out that for those skilled in the art, some improvements and modifications without departing from the principles of the present invention, should be considered as protection scope of the present invention.

Claims (5)

1. short-term wind power prediction method, it is characterized in that, using wind speed as input, adopt the output power of regression model to wind energy turbine set of least square method supporting vector machine to predict, the Parameters in Regression Model of least square method supporting vector machine adopts Chaos particle swarm optimization algorithm to be optimized.

2. short-term wind power prediction method according to claim 1, it is characterized in that, described using wind speed as input, adopt the output power of regression model to wind energy turbine set of least square method supporting vector machine to predict, the Parameters in Regression Model of least square method supporting vector machine adopts Chaos particle swarm optimization algorithm to be optimized and specifically comprises the following steps:

Step 1: obtain data, and data are normalized rear as training sample;

Step 2: the parameter of initialization least square method supporting vector machine and Chaos particle swarm optimization algorithm;

Step 3: minimum for objective function with regression error quadratic sum, adopts the Parameters in Regression Model of Chaos particle swarm optimization algorithm to least square method supporting vector machine to be optimized;

Step 4: the parameter generation optimized is returned in the regression model of least square method supporting vector machine, re-starts training;

Step 5: use the output power of regression model to wind energy turbine set training least square method supporting vector machine to predict.

3. short-term wind power prediction method according to claim 2, is characterized in that, the regression model construction method of described least square method supporting vector machine comprises the following steps:

Adopt a training sample set (x _i, y _i), i=1,2 ..., l, x _iwind speed, y _ibe the measured value of wind power output power, l is the sum that training sample concentrates data point set, utilizes a Nonlinear Mapping sample space is mapped to feature space, and constructs optimal decision function in higher dimensional space:

Utilize structural risk minimization, the optimization aim of least square method supporting vector machine can be expressed as:

$min\frac{1}{2}{w}^{T}w+C\frac{1}{2}\underset{i=1}{\overset{l}{\Σ}}{\mathrm{\ξ}}_i^2---\left(2\right)$

Wherein: w is weight vector, || w|| ²it is the complexity of model; C is marginal coefficient, controls the punishment degree to error; ξ _iit is error vector; B is departure;

Utilize Lagrangian method to solve formula (2), can obtain:

In formula: α _i(i=1,2 ..., l) be Lagrange multiplier;

From KKT (Krush-Kuhn-Tucker) condition:

By kernel function, (4) are converted into linear equation to solve, select the good Radial basis kernel function of effect in regression forecasting, definition kernel function is system of linear equations after conversion is:

$\left[\begin{array}{cccc}0& 1& ...& 1\\ 1& K\left(x_1,x_l\right)+\frac{1}{C}& ...& K\left(x_1,x_l\right)+\frac{1}{C}\\ .& .& & .\\ .& .& & .\\ .& .& & .\\ 1& K\left(x_1,x_l\right)+\frac{1}{C}& ...& K\left(x_1,x_l\right)+\frac{1}{C}\end{array}\right]\×\left[\begin{array}{c}b\\ {\mathrm{\α}}_i\\ .\\ .\\ .\\ {\mathrm{\α}}_l\end{array}\right]=\left[\begin{array}{c}b\\ y_i\\ .\\ .\\ .\\ y_l\end{array}\right]---\left(5\right);$

Solve above formula by least square method, obtain regression coefficient α _iwith deviation b, just nonlinear regression model can be obtained:

$f\left(x\right)=\underset{i=1}{\overset{1}{\Σ}}{\mathrm{\α}}_{i}K(x,x_i)+b---\left(6\right);$

In formula (6), Radial basis kernel function in formula, σ is core width, || x-x _i|| be two norms;

The regression model of least square method supporting vector machine is c in σ in this formula and formula (2) adopts Chaos particle swarm optimization algorithm to be optimized.

4. the short-term wind power prediction method according to Claims 2 or 3, is characterized in that, described step 3 specifically comprises the following steps:

Step 3.1: adopt chaos system z _n+1=μ z _n(1-z _n) produce chaos sequence initialization is carried out to the speed of particle in algorithm and position, wherein μ is chaos controlling parameter, z _nbe Chaos Variable, n is sequence number;

Step 3.2: traversal search obtains personal best particle, and it is minimum for fitness with regression error quadratic sum to press formula (7), calculates and compares fitness value, iterative search goes out global optimum position;

$min\underset{i=1}{\overset{1}{\Σ}}{\left(f\right(x_i)-y_i)}^2---\left(7\right);$

In formula: f (x _i) be the predicted value in the i-th moment; x _ifor the input value of the regression model of least square method supporting vector machine, y _iit is the measured value in the i-th moment;

Step 3.3: produce chaos sequence based on global optimum position, using the reposition of a position best in the sequence of generation as one of them particle;

Step 3.4: the speed upgrading other particle by formula (8), upgrades the position of other particles by formula (9);

v _ik(t+1)＝ωv _ik(t)+c ₁δ ₁(p _ik(t)-x _ik(t))+c ₂δ ₂(p _gk(t)-x _ik(t)) (8)

x _ik(t+1)＝x _ik(t)+v _ik(t+1) (9)；

In formula: ω is inertia weight; c ₁, c ₂for Studying factors; δ ₁, δ ₂for the random number between 0 and 1; In addition, the maximal rate of particle should not exceed its maximal rate V _max, x _ikt () is the position of an i moment kth particle, x _ik(t+1) be the position of an i+1 moment kth particle; v _ikt () is the speed of an i moment kth particle, v _ik(t+1) be the speed of an i+1 moment kth particle, p _ikt () is personal best particle, p _gkt () is global optimum position;

Step 3.5: judge whether to meet iterations restriction, be, terminate to optimize and Output rusults, otherwise return step 3.2.

5. short-term wind power prediction method according to claim 2, is characterized in that, in step 2, described initialization least square method supporting vector machine and the parameter of Chaos particle swarm optimization algorithm comprise the following steps:

Step 2.1, the punishment parameter C determining least square method supporting vector machine and nuclear parameter σ ²scope, the inertia weight ω of initialization least square method supporting vector machine;

Step 2.2, determine the Studying factors c of Chaos particle swarm optimization algorithm ₁, c ₂; Chaos controlling parameter μ and iterations.