CN108062595B

CN108062595B - WRF/CFD/SAHDE-RVM coupling-based short-time wind energy prediction method for complex landform area

Info

Publication number: CN108062595B
Application number: CN201711218419.4A
Authority: CN
Inventors: 闫渤文; 李大隆
Original assignee: Chongqing University
Current assignee: Chongqing University
Priority date: 2017-11-28
Filing date: 2017-11-28
Publication date: 2021-09-28
Anticipated expiration: 2037-11-28
Also published as: CN108062595A

Abstract

The invention provides a WRF/CFD/SAHDE-RVM coupling-based short-time wind energy prediction method for a complex landform area. The prediction method comprises the steps of WRF wind direction prediction, SAHDE-RVM short-time wind speed prediction, APG preliminary manufacturing, APG perfection, short-time wind energy prediction and the like. The method enables a more precise matching of the information of the corresponding boundary conditions by means of the observation data of the anemometer. The wind direction and wind speed information in the future short time can be provided for the wind power towers in the complex landform area, each wind power tower in the area is convenient to adjust in real time, and the wind biting capacity is improved. Has wide application prospect.

Description

WRF/CFD/SAHDE-RVM coupling-based short-time wind energy prediction method for complex landform area

Technical Field

The invention relates to the field of wind power, in particular to a wind energy prediction method.

Background

In order to predict short-time wind energy of a complex geomorphic region, The existing method is to predict The wind field of The region in WAsP software based on meteorological data provided by The Weather Research and Forecasting Model (WRF), and The basic idea of The WAsP Model is to linearize a momentum equation for controlling The motion of airflow, so that The difficulty in solving a nonlinear partial differential equation can be obviously reduced, and a large amount of time cost is saved. In addition, a method for performing real-time simulation prediction based on meteorological data provided by WRF, and then inputting the simulation result into CFD software for calculation to obtain the prediction wind field information of the whole area is provided. In addition, there is a method of measuring the Correlation with prediction (MCP).

The inventor finds that the prior art has at least the following defects:

firstly, a WAsP model is adopted, the wind speed of the wind is calculated through anemometer observation data, the wind speed of the near-stratum at any point and wind energy parameters are calculated through the wind speed according to the surface characteristics of the surrounding area on the assumption that the wind speed of the wind in a certain range does not change. Because the error increases with the distance from the wind measuring point due to extrapolation from an observation point, it is not feasible to directly use WAsP for wind energy estimation of a larger area; in addition, the WAsP is calculated by adopting a linear model, so that the uncertainty of the calculation result of the complex fluid flowing through the terrain can be caused, and the WAsP has certain limitation; meanwhile, the existing research finds that the linear model can overestimate the acceleration of airflow at the mountain top and cannot solve the problem of gas diversion on the lee side of the mountain.

And secondly, a CFD software real-time numerical simulation method is adopted, short-time prediction meteorological data needs to be extracted from the WRF in time, meanwhile, the calculation resources consumed by operation are huge, certain calculation time is needed, the efficiency is difficult to guarantee, and the wind power tower is inconvenient to perform real-time corresponding adjustment according to a short-time wind energy prediction result.

Thirdly, two necessary conditions are needed by adopting the MCP method: (1) observation data of a complex topographic region in a global and local period of time needs to be obtained, and (2) the global data and the local data have good correlation. However, due to the nonlinearity of fluid characteristics in complex terrain areas, it is difficult to find a specific equation to describe their correlation; meanwhile, a plurality of anemometers are required to be arranged for observing wind speed and wind direction of the whole area for a long time when global field data measurement is carried out, the method firstly has high construction and maintenance cost, and secondly, the expression of wind field information of a complex terrain area can be influenced by improper selection of positions of individual measurement places.

Therefore, the method for predicting the short-term wind energy of the complex terrain area based on WRF/CFD/SAHDE-RVM coupling is provided, and has important significance for wind energy evaluation of the complex terrain area.

Disclosure of Invention

The invention aims to provide a speed-limiting control system for an on-bridge vehicle, which can calculate a reasonable speed-limiting value according to actual conditions, so as to solve the problems in the prior art.

The technical scheme adopted for achieving the aim of the invention is that the method for predicting the short-time wind energy of the complex landform area based on WRF/CFD/SAHDE-RVM coupling comprises the following steps:

1) and acquiring topographic data of the target complex landform area, and performing short-time wind direction prediction based on meteorological data provided by WRF. And obtaining the short-time mainstream wind direction of the target complex landform area.

2) SAHDE-RVM short term wind speed prediction.

2.1) carrying out regression prediction by using RVM, and establishing an RVM regression prediction model.

2.2) collecting historical wind speed and direction data at the anemometer of the target complex geomorphic region as a training sample of the SAHDE algorithm. And (3) adaptively acquiring the optimal parameters of the RVM regression prediction model in the step 2.1) by training the sample data set.

And 2.3) returning a short-time wind speed prediction result at the anemometer according to the real-time data of the anemometer.

3) In CFD software, numerical simulation calculation of target complex landforms under a plurality of wind direction angles is uniformly carried out. And making an initial APG according to the CFD result.

4) And (4) perfecting the APG through an algorithm according to the long-term measured data of the anemometer.

5) Matching the wind direction prediction result of the WRF in the step 1) with the APG in the step 4), and extracting corresponding boundary speed information and flow field information. And obtaining a short-time prediction result of the wind energy of the whole area.

Further, the regression prediction model of RVM in step 2.1) is:

in the formula, w_iAs weighting factor, K (x, x)_i) For the kernel function, N is the number of samples.

Further, a kernel function of the RVM regression prediction model is a combined kernel function obtained by combining a Gaussian kernel function and a binomial kernel function. Wherein the combined kernel function is:

K(x_i,x_j)＝λG(x_i,x_j)+(1-λ)P(x_i,x_j) (2)

in the formula, G (x)_i,x_j) Is a Gaussian kernel function, P (x)_i,x_j) Is a binomial kernel function and λ is the weight. Wherein, the lambda is more than or equal to 0 and less than or equal to 1.

Further, the step 3) specifically includes:

3.1) the CFD software obtains the detailed site information of the target complex landform area, corresponding model establishment is carried out by using preprocessing software, a calculation domain is established, and boundary conditions are set.

3.2) dividing the input boundary speed into an x direction and a y direction after non-dimensionalization. Wherein the speed in the x direction is represented by u and the speed in the y direction is represented by v. The wind speed at the boundary is recorded as u under different wind direction angles theta_B＝Vcosθ，v_BVsin θ, B represents a boundary. And calculating the non-dimensionalized flow field information, and expressing the flow field information by f (i, j, k).

3.3) recording the wind speed data (u) of the corresponding wind meter side wind point at the observation height thereof_A,v_A) And wind speed information corresponding to the ground surface of the whole field.

3.4) comparing the non-dimensionalized boundary velocity and flow field information of step 3.2) with the point (u) of the anemometer under the dimensionless condition of step 3.3)_A,v_A) And (4) associating.

3.5) multiple (u) obtained from numerical simulation at different inlet wind speeds_A,v_A) As a result, it is unitized

Is marked as (U)_A,V_A)。

3.6) mixing (U)_A,V_A) And inserting the points and boundary speed information and flow field information contained in the points into the APG, and drawing the initial APG. The APG only displays data measured by the anemometer, but each point contains boundary wind speed information and flow field information corresponding to the point.

The technical effects of the invention are undoubted:

A. the minimization of required computing resources is realized;

B. the information of the corresponding boundary conditions can be more accurately matched through the observation data of the anemometer;

C. the wind direction and wind speed information in a short time in the future can be provided for the wind power towers in the complicated landform area, so that each wind power tower in the area can be conveniently adjusted in real time, and the wind biting capacity is improved; meanwhile, the power generation capacity in a short period can be calculated according to the relation between the wind speed and the power generation power.

Drawings

FIG. 1 is a flow chart of a prediction method;

FIG. 2 is a diagram illustrating WRF short-term wind direction prediction;

FIG. 3 is a schematic diagram of the manufacture, perfection and application of APG;

FIG. 4 is a schematic diagram of a short-term wind speed prediction of SAHDE-RVM;

FIG. 5 is a schematic diagram of the fabrication of an initial APG based on CFD results;

FIG. 6 is a schematic diagram of refining an APG according to an algorithm;

FIG. 7 is a schematic diagram of a short-term wind speed prediction of SAHDE-RVM;

FIG. 8 is a WRF short-term dominant wind direction prediction result and observation data comparison graph;

FIG. 9 is a graph comparing the short term wind speed prediction results of the SHADE-RVM with the measured data;

FIG. 10 is a mesh partition diagram for a complex terrain area model in CFD software;

FIG. 11 is a diagram of initial APG;

FIG. 12 is a plot of wind energy density for a complex terrain area.

Detailed Description

The present invention is further illustrated by the following examples, but it should not be construed that the scope of the above-described subject matter is limited to the following examples. Various substitutions and alterations can be made without departing from the technical idea of the invention and the scope of the invention is covered by the present invention according to the common technical knowledge and the conventional means in the field.

Example 1:

the embodiment discloses a method for predicting short-term wind energy of a complex landform area based on WRF/CFD/SAHDE-RVM coupling, and the method is shown in figure 1 and comprises the following steps:

1) referring to fig. 2, the topographic data of the target complex topographic region is obtained, and multiple nested grids are divided. And setting a boundary layer parameter scheme. And carrying out short-time wind direction prediction based on meteorological data provided by WRF. Simulating the wind speeds in two mutually perpendicular directions in the horizontal plane in a period of time in the future, outputting data at intervals, taking the average value of the nine adjacent grid points as the output result of the wind speed, and obtaining the short-term main flow direction of the complex geomorphic region in the period of time in the future by synthesizing the speeds in the two perpendicular directions, referring to fig. 8. Wherein the meteorological data comprises assimilation global circulation mode reanalysis data, conventional ground meteorological data and sounding meteorological data.

2) SAHDE-RVM short term wind speed prediction.

2.1) carrying out regression prediction by using RVM, and establishing an RVM regression prediction model. Wherein given an input set X ═ { X) of training samples₁,x₂,x₃,…,x_nAnd with the corresponding output set T ═ T₁,t₂,t₃,…,t_nWhere n is the number of samples, let t_iIs a target value and has t_iE.g. R. Output value t_iThe functional model of (a) can be expressed as:

t_i＝y(x_i,w)+ε_i (1)

in the formula, epsilon_iRepresents Gaussian white noise, and ε_iObey distribution epsilon_i～N(0,σ²) Then p (t)_i|x_i)＝N(t_i|y(x_i,w),σ²). The output of the RVM model can be represented as a combination of non-linear kernel functions, which need not satisfy the mercer condition. The regression prediction model for RVM is:

For independently distributed output sets t_iThe likelihood estimate of (c) is:

wherein phi is (phi)₁,Φ₂,…,Φ_N)，Φ_i＝(1,K(x_i,x₁),...,K(x_i,x_n))^T(i＝1,…,N)。

The gaussian prior distribution with zero weight parameter defined by the sparse bayesian principle is:

in the formula, alpha_iBeing a hyperparameter of the prior Gaussian distribution, α ═ α₀,…,α_N)^T. Each independent hyper-parameter alpha_iControlling the right parameter w_iThe prior distribution of (2) makes the correlation vector machine model sparse.

According to the Bayes principle, the posterior distribution of the weight vector w is calculated by equations (3) and (4):

in the formula (I), the compound is shown in the specification,

as shown in equation (5), the hyper-parameters α and σ are needed to determine the weight vector w²A determination is made.

And (3) calculating the likelihood distribution of the hyperparameters by adopting a Bayesian framework:

p(t|α,σ²)＝∫p(^t|w,σ²)p(w|α)dW＝N(0,C)(6)

wherein C is covariance and C ═ σ -²I+ΦA^-1Φ^T。

For two hyper-parameters alpha and sigma in formula (6)²Partial derivatives are respectively calculated and made equal to zero.

Obtaining the optimal solution hyper-parameter

Updating iterative algorithm in the process:

in the formula u_jIs the jth posterior average weight, gamma_j＝1-α_jM_jj(γ_j∈[0,1])，M_jjIs the diagonal element of the weight covariance matrix Σ.

In the RVM learning process, firstly, initializing two parameters in the formulas (7) and (8), continuously and newly calculating two hyper-parameters by updating an iterative formula, stopping updating calculation when the parameters in the model reach the maximum training times, and obtaining alpha and sigma at the moment²Is the optimum value.

If an input value X of the system is given_*Then the output probability distribution is:

y_*＝u^TΦ(X_*)＝u_MPΦ(X_*) (10)

in which, when an input value X is given_*Then there is a corresponding predicted output value y_*。

To predict variance, uncertainty is represented.

2.2) referring to fig. 7, collecting historical wind speed and direction data at the anemometer of the target complex geographical area as a training sample of an adaptive hybrid-based differential evolution (SAHDE) algorithm. And (3) adaptively acquiring the optimal parameters of the RVM regression prediction model in the step 2.1) by training the sample data set. In this embodiment, the SAHDE operation steps are as follows:

a) mutation operation: the DE algorithm is performed by a plurality of mutation operation methods, and the following methods are selected for carrying out mutation operation.

x_m＝x_best+F[(x₁-x₂)+(x₃-x₄)] (12)

Randomly selecting four parent individuals, and respectively marking as x₁,x₂,x₃,x₄，x_bestIs the best individual in the parent, x_mFor the variant individuals, F is the mutation rate (F. epsilon. [0, 1.2.)])。

b) And (3) cross operation: selecting two individuals x_iAnd x_mPerforming cross operation, wherein the new individuals generated after the cross operation are x_cThe specific operation method is as follows:

wherein rand () represents [0,1 ]]Randr (i) e {1,2, …, D } is a randomly generated integer, D is the dimension of the optimization variable, C_RIs the crossing rate (C)_R∈(0,1))。

c) Selecting operation: the filial generation individuals of DE are generated by a greedy selection method, and after the crossing and variation operations, the individuals x are generated_cAnd target individual

The two compete, the better one is selected as the offspring individual, and the selection operation method is shown as formula 11:

in the formula, x_cThe individuals generated after the cross operation and the mutation operation,

is the ith individual in the filial generation.

The SAHDE algorithm performs a global search initially and a local search later, C_RIs determined by the current evolution algebra g_nowAnd maximum evolution algebra g_maxThe specific method for self-adaptive adjustment is as follows:

C_R0as a crossover operator C_RInitial value of (C)_RIs adaptively adjusted according to the above formula, and the initial value C_R0Smaller, and then gradually increased in value.

And (4) carrying out secondary search on the current optimal individuals generated by the SAHDE by using a simulated annealing algorithm. Among SAHDE, the current best individual is selected as the initial individual, y₀＝x_bestThe initial temperature is selected to be T₀The manner in which new individuals are generated is as follows:

y_r+1,j＝y_r,j+ηε(x_jmax-x_jmin),j＝1,2,...,D (16)

in the formula, r is the iteration number of the simulated annealing algorithm. y is_rNew individuals were generated after r iterations. Eta is the control disturbance amplitude. ε is a random variable that follows a mean or normal distribution. x is the number of_jmax、x_jminRespectively, the value ranges of the j-th dimension optimization variables.

After the optimal individuals generated by the SAHDE are subjected to secondary search by a simulated annealing algorithm, the fitness is changed into delta F, and the delta F is F (y)_r+1)-f(y_r). If Δ F < 0, a new individual is accepted and the original best individual is replaced with the new individual. If e^(-ΔF/T)Now also newly generated individuals are accepted and replaced with one non-optimal individual in the population. Otherwise, rejecting. If a new individual is selected, press T_r+1＝aT_r(a is more than 0 and less than 1) cooling. Otherwise, the temperature is not reduced.

In order to accelerate the training speed of the sample and improve the prediction accuracy of the model, the normalization method is used in this embodiment to preprocess the sample data of the actual measurement of the wind speed:

in the formula (I), the compound is shown in the specification,

normalized values for the data. x is the number of_maxIs the maximum value of the wind speed sample data. x is the number of_minIs the minimum value of the wind speed sample data.

And selecting a combined kernel function obtained by combining the Gaussian kernel function and the binomial kernel function as the kernel function of the RVM model in the step 2.1), and taking advantages and making up for the disadvantages to play respective advantages. The combined kernel functions are as follows:

K(x_i,x_j)＝λG(x_i,x_j)+(1-λ)P(x_i,x_j) (18)

wherein, G (x)_i,x_j) Is a Gaussian kernel function, and the expression is: g (x)_i,x_j)＝exp(-||x_i-x_j||²/σ²)。P(x_i,x_j) Is a binomial kernel function. λ is weight, and λ is more than or equal to 0 and less than or equal to 1.

Collecting wind speed and direction data of the anemometer in the past year as training samples of the SAHDE-RVM algorithm, and adaptively acquiring the optimal parameters alpha and sigma of the model through training the sample data set²。

2.3) returning a short-term wind speed prediction result at the anemometer based on the anemometer real-time data, see FIG. 4. Each pair of instantaneous wind speed data at the location where it is input to an anemometer can then return a short (specified time, e.g., 15min) wind speed prediction at an anemometer. FIG. 9 is a comparison graph of the short term wind speed prediction results of the SHADE-RVM at 100 measuring points and the measured data.

3) Referring to fig. 5, in the CFD software, numerical simulation calculation of a target complex landform under a plurality of wind angles is uniformly performed. An initial Anemometer phase map (APG) is created from the CFD results. The mesh partition diagram of the complex geomorphic region model in the CFD software is shown in fig. 10. The initial APG map is shown in fig. 11.

3.1) CFD software obtains detailed site information of the target complex landform area, corresponding model establishment is carried out by using preprocessing software (such as ICEM), meanwhile, a computing domain matched with the model establishment is established, FLUENT is introduced after grid division, and numerical simulation calculation is carried out after setting how each boundary of boundary conditions is set.

3.2) Inlet wind speed V taking into account the reference height z_refNumerical simulations were performed at 10m/s for 360 wind direction angles at which the field was evenly divided. Recording wind speed data u of a corresponding wind meter side wind point, namely a point A under the observation height of the point A_A,v_AAnd wind speed information corresponding to the ground surface of the whole field. The input boundary velocity is non-dimensionalized into x-direction and y-direction. Wherein the speed in the x direction is represented by u and the speed in the y direction is represented by v. The wind speed at the boundary is recorded as u under different wind direction angles theta_B＝Vcosθ，v_BVsin θ, B represents a boundary. And calculating the non-dimensionalized flow field information, and expressing the flow field information by f (i, j, k).

3.5) 360 (u) obtained by numerical simulation at different inlet wind speeds_A,v_A) As a result, it is unitized

Is marked as (U)_A,V_A)。

3.6) mixing (U)_A,V_A) And inserting the points and boundary speed information and flow field information contained in the points into the APG, and drawing the APG. The APG only displays data measured by the anemometer, but each point contains boundary wind speed information and flow field information corresponding to the point.

Referring to FIG. 6, APG is refined from long-term measured data of the anemometer. Suppose u_*,v_*Unitizing certain data measured by an anemometer to obtain U_*,V_*Defining its point in the APG as point P_*. If in APG, point P_*Close enough to a point T, using point T as PTo approximate, the flow of the next anemometer data begins. If not, the following process is carried out: point P_*Projecting to APG curve nearest to the APG curve, setting the projection point as P, and defining two end points of the curve segment as T₁And T₂Calculate P_*In the relative position of the segments:

can point P according to r_*Boundary wind speed information included in the APG map is calculated: u. of_B’＝(1-r)U_B1+rU_B2，v_B’＝(1-r)V_B1+rV_B2，U_B1,V_B1U_B2,V_B2Are respectively a point T₁、T₂Corresponding boundary velocity information. Then through u_B’,v_B'information f' i, j, k of the flow field is calculated and inserted into the anemometer data u_A’,v_A' of (1). u. of_A’,v_A' is set to | u_AObtaining the amount of each of the above-mentioned components in units

Will point U_A’,V_A' and its associated boundary velocity information U_B’,V_B'and flow field information F' (i, j, k) are inserted into the APG.

Then judge point P again_*If it is close enough to a certain point T, the above process is repeated.

As more and more information is collected and entered by the anemometer, the APG becomes more and more sophisticated, providing more comprehensive anemometer data and information about the boundary speed.

5) Referring to fig. 3, the wind direction prediction result of the WRF in step 1) is matched with the APG in step 4), and corresponding boundary speed information and flow field information are extracted. And obtaining a short-time prediction result of the wind energy of the whole area. The wind energy density map of the complex topographic region obtained from the prediction results is shown asAs shown in fig. 12. And matching the data at the wind direction prediction result anemometer of the WRF with the drawn APG, extracting corresponding non-dimensionalized boundary speed information and flow field information, and multiplying the wind speed prediction result of the SAHDE-RVM by the non-dimensionalized boundary speed information and the flow field information to obtain a short-time prediction result of the wind energy of the whole area according to the point that the shapes of the flow field are linearly related under the conditions of the same wind direction angle and different reference wind speeds. Predicting a short-term wind speed of u as SAHDE-RVM_C,v_CBoundary speed information U in APG corresponding to wind direction prediction result provided by WRF_BC,V_BCAnd flow field information F_C(i, j, k) calculating the corresponding boundary velocity

At the moment, the short-time wind energy prediction result of any point in the whole area can be obtained.

It is worth to be noted that, in this embodiment, a brand new combination idea of a mesoscale numerical weather prediction mode and a small-scale numerical mode is presented, and by using a method of completing an APG by an algorithm after an initial APG is made by CFD, minimization of required computing resources is achieved, and it is not necessary to perform numerical simulation computation on an entire area by CFD in real time after WRF prediction data is imported. Meanwhile, the dot calculation value in the mesoscale mode is an average value of an area in a grid size within a period of time, and cannot specifically represent the wind speed condition of a certain place, so that the SAHDE-RVM is introduced. By training the SAHDE-RVM using historical data of individual anemometers in the region, wind speed predictions in a shorter time relative to the mesoscale WRF mode are achieved.

Unlike the linear WAsP model, which is a regression result based on previous data and then extrapolates, the present embodiment includes numerical simulation calculation based on a physical model of computational fluid dynamics, and can be more widely applied to complex terrain areas. Meanwhile, historical wind speed and wind direction data of a short-time wind energy prediction site are not needed, only elevation data of the whole complex geomorphic area are needed to be obtained, and the WRF and CFD technologies are combined, so that the wind speed and wind direction state of the whole complex geomorphic area after a period of time can be predicted through observation data obtained by setting an anemometer in the area. Compared with the traditional DE algorithm, the simulated annealing algorithm is introduced in the short-time wind speed prediction process, the optimal individual generated by the SAHDE is searched for the second time, and the defect of premature convergence of the traditional DE algorithm is overcome. Furthermore, when the above method is actually applied to wind energy estimation, the APG will gradually become a large data set with the increase of observation time, so that it can be more accurately matched to the information of the corresponding boundary condition by the observation data of the anemometer than the conventional MCP method provides a fixed correlation equation.

The method is applied to the related field of wind power, the magnitude of the normally blown wind direction and the wind speed of each site in the whole complex landform area can be summarized in the long term, and the site which is most suitable for setting the wind power generation tower in the area and the orientation of the power generation tower can be obtained by integrating the data of each site and analyzing the data. From the short-term, can provide wind direction and wind speed information in the future short time for the wind power tower in complicated landform region, each wind power tower in the region of being convenient for adjusts in real time, improves and stings the wind ability. Meanwhile, the power generation capacity in a short period can be calculated according to the relation between the wind speed and the power generation power.

Claims

1. The method for predicting the short-time wind energy of the complex landform area based on WRF/CFD/SAHDE-RVM coupling is characterized by comprising the following steps of:

1) acquiring topographic data of a target complex landform area, and performing short-time wind direction prediction based on meteorological data provided by WRF; obtaining the short-time mainstream wind direction of the target complex landform area;

2) SAHDE-RVM short-term wind speed prediction;

2.1) carrying out regression prediction by using RVM, and establishing an RVM regression prediction model;

2.2) collecting historical wind speed and direction data of the anemometer in the target complex geomorphic region as a training sample of the SAHDE algorithm; adaptively acquiring the optimal parameters of the RVM regression prediction model in the step 2.1) by training a sample data set;

2.3) returning a short-time wind speed prediction result at the anemometer according to the real-time data of the anemometer;

3) in CFD software, uniformly performing numerical simulation calculation of target complex landforms under a plurality of wind direction angles; making an initial APG according to the CFD result;

4) according to the long-term measured data of the anemometer, the APG is perfected through an algorithm;

5) matching the wind direction prediction result of the WRF in the step 1) with the APG in the step 4), and extracting corresponding boundary speed information and flow field information; and obtaining a short-time prediction result of the wind energy of the whole area.

2. The WRF/CFD/SAHDE-RVM coupling-based short-time wind energy prediction method for complex landform areas according to claim 1, characterized in that: given the input set X ═ X of training samples in step 2.1) { X ═ X₁,x₂,x₃,…,x_nAnd with the corresponding output set T ═ T₁,t₂,t₃,…,t_n}; output value t_iThe function model of (a) is expressed by equation (1); the regression prediction model of RVM is formula (2);

t_i＝y(x_i,w)+ε_i (1)

in which x represents the input data, ε_iRepresenting Gaussian white noise, w_iAs weighting factor, K (x, x)_i) For the kernel function, N is the number of samples.

3. The WRF/CFD/SAHDE-RVM coupling-based short-time wind energy prediction method for complex landform areas according to claim 2, characterized in that: the kernel function of the RVM regression prediction model is a combined kernel function obtained by combining a Gaussian kernel function and a binomial kernel function;

wherein the combined kernel function is:

K(x_i,x_j)＝λG(x_i,x_j)+(1-λ)P(x_i,x_j) (3)

in the formula, G (x)_i,x_j) Is a Gaussian kernel function, P (x)_i,x_j) Is a binomial kernel function, and λ is a weight; wherein, the lambda is more than or equal to 0 and less than or equal to 1.

4. The method for predicting the short-term wind energy of the complex landform area based on WRF/CFD/SAHDE-RVM coupling according to claim 1, wherein the step 3) specifically comprises:

3.1) the CFD software obtains detailed site information of the target complex landform area, corresponding model establishment is carried out by using preprocessing software, a calculation domain is established, and boundary conditions are set;

3.2) dividing the input boundary speed into an x direction and a y direction after non-dimensionalization; wherein, the speed in the x direction is represented by u, and the speed in the y direction is represented by v; the wind speed at the boundary is recorded as u under different wind direction angles theta_B＝Vcosθ，v_BVsin θ, B denotes a boundary; calculating the non-dimensionalized flow field information, and expressing the flow field information by f (i, j, k);

3.3) recording the wind speed data (u) of the corresponding wind meter side wind point at the observation height thereof_A,v_A) And wind speed information corresponding to the ground surface of the whole field;

3.4) comparing the non-dimensionalized boundary velocity and flow field information of step 3.2) with the point (u) of the anemometer under the dimensionless condition of step 3.3)_A,v_A) Associating;

Is marked as (U)_A,V_A)；

3.6) mixing (U)_A,V_A) Inserting the points and boundary speed information and flow field information contained in the points into an APG (advanced process group) and drawing an initial APG; where the APG only displays anemometer measured data, but where each point contains its corresponding boundaryWind speed information and flow field information.