CN115545279A

CN115545279A - Wind power plant wind power prediction method

Info

Publication number: CN115545279A
Application number: CN202211137026.1A
Authority: CN
Inventors: 刘阳; 孙鑫; 滕卫军; 李朝晖; 刘善峰; 张亚飞; 谷青发
Original assignee: Electric Power Research Institute of State Grid Henan Electric Power Co Ltd
Current assignee: Electric Power Research Institute of State Grid Henan Electric Power Co Ltd
Priority date: 2022-09-19
Filing date: 2022-09-19
Publication date: 2022-12-30

Abstract

A wind power plant wind power prediction method collects historical wind speed, air temperature and air humidity data of a wind power plant, abnormal data and missing data in a database are processed by adopting a differential algorithm, and accuracy and integrity of the data are improved. And introducing optimized meteorological data such as wind speed, air temperature and air humidity of the wind power plant as input of the combined prediction model, and effectively improving the wind speed prediction precision by combining the combined prediction model with the insect-saving optimization algorithm. The wind power is obtained by introducing the wind speed value and combining with a standard power characteristic curve of the wind turbine generator, the wind power is calculated according to the standard power characteristic curve corresponding to the specific fan model, the wind power values of the wind power field are obtained by accumulating the power of each fan of the wind power field, and the short-term wind power prediction precision is improved.

Description

Wind power prediction method for wind power plant

Technical Field

The invention belongs to the field of wind speed power generation, and particularly relates to a wind power prediction method for a wind power plant.

Background

The large-scale development of new energy power generation modes such as wind energy and the like is important content for strategic adjustment of energy and conversion of power development modes in China. However, wind energy is influenced by uncertain conditions such as climate, temperature and the like, and the power generation characteristics of the wind energy are random, fluctuating and intermittent, so that a large-scale wind power plant power supply cannot meet the requirement of accessing a power grid, the development of wind speed power generation in China is restricted, the installed capacity of wind power is continuously and rapidly increased for many years, and the influence of the wind power prediction accuracy on the operation stability of the power grid is increasingly obvious due to factors such as high proportion and random fluctuation of wind power output.

In the existing research, some scene for wind power prediction is optimized, but the existing research is limited by imperfect description of NWP (numerical weather forecast) on the atmospheric operation process and the limitation of matching a statistical model on a physical random process, and prediction errors are difficult to avoid completely. Therefore, how to effectively carry out multi-dimensional evaluation and deep analysis of errors is a key problem of further improving the pilot power prediction precision, and has important practical application value. Based on insufficient wind power prediction precision and complicated process, an effective comprehensive solution is not provided.

Disclosure of Invention

In order to solve the defects in the prior art, the invention aims to collect the historical wind speed, air temperature and air humidity data of the wind power plant, and adopts a difference algorithm to process abnormal data and missing data in a database, thereby improving the accuracy and integrity of the data. Meteorological data such as wind speed, air temperature and air humidity of the optimized wind power plant are introduced to be input into the combined prediction model, and the wind speed prediction precision is effectively improved by combining the combined prediction model with the festival worm optimization algorithm. The wind power is obtained by introducing the wind speed value and combining with a standard power characteristic curve of the wind turbine generator, the wind power is calculated according to the standard power characteristic curve corresponding to the specific fan model, the wind power value of the wind power field is obtained by accumulating the power of each fan of the wind power field, and the short-term wind power prediction precision is improved.

The invention adopts the following technical scheme.

A wind power prediction method for a wind power plant specifically comprises the following steps:

step 1, collecting historical wind speed, air temperature and air humidity data of a wind power plant, and processing abnormal data and missing data in a database by adopting a differential algorithm;

step 2: constructing a combined prediction-festival worm optimization algorithm model as a combined prediction model;

and step 3: acquiring a wind speed prediction value from the meteorological information of the data of the wind power plant by using a festival combination prediction-festival worm optimization algorithm model, and improving the wind speed prediction precision;

and 4, step 4: and calculating a wind power value based on the obtained wind speed predicted value and the wind generating set standard power curve, and accumulating the power of all the fans in the wind power plant to obtain the wind power value of the wind power plant.

Preferably, the step 1 specifically includes: inputting original wind speed, air temperature and air humidity data of the wind power plant every fifteen minutes, calculating to obtain a smoothing factor l, and solving an equation of the smoothing factor l as follows:

wherein Y is _i Comprising Y _1,i 、Y _2,i And Y _3,i ，Y _1,i 、Y _2,i And Y _3,i Respectively representing the wind speed, air temperature and air humidity variables, Y, input at time i _i Representing the raw wind speed, air temperature and air humidity data input at time i, E _i The observed values of the wind speed, air temperature and air humidity variables input at the time i are represented by E _i And Y _i Form an error model ER ⁱ I and t are integers;

applying difference algorithm to error model ER ⁱ In the data sequence of the noise point (ERN) ^k Removing noise points from each error metric index to obtain a corrected error model { ERF ^k Therein { ERF } ^k }＝{ER ^k }-{ERN ^k }，k＝1,2,3，ERF ^k Is an error metric, ERF ^k Respectively establishing three corresponding error measurement indexes for the wind speed, the air temperature and the air humidity, and then correcting an error model { ERF ^k Calculating corresponding error threshold value th according to formula (1) for each error metric index ^k :

th ^k ＝mean(ERF ^k )+αsd(ERF ^k ),k＝1,2,3 (1)

Where mean is the mean function, sd is the standard deviation function, and α is the tuning coefficient. ERF ^k Is an error metric value, and then the anomaly detection of the error metric is carried out on the meteorological data points by a voting mechanism.

Preferably, the anomaly detection of the error metric consists of three main indicators, which are MAE, MAPE and MASE, respectively;

o’ _i is a meteorological data observation value element transverse vector including o' _1,i ，o’ _2,i And o' _3,i Three elements, respectively wind speed, air temperature and air humidity in the meteorological data observations at time i, e _i Is a predicted value element of meteorological data, er _i The error function is an error model element vector, i is a meteorological data serial number, n represents the number, MAE _i Denotes mean absolute error, MAPE _i Representing mean absolute percentage error, MASE _i Represents the mean absolute scale error and is derived from equations (2) through (5):

if the majority of error metrics in the plurality of error metrics of a data point exceed the corresponding threshold in the corrected error model, the data point is regarded as an abnormal data point; the finally obtained abnormal point result of the time-series meteorological data abnormal detection can be expressed by a data set according to the formula (6) and the formula (7) as follows:

ED _i is an outlier data set, ed outlier data element, representing er ^k _i And th ^k Express er _i And the second element of the th vector, when ^k _i If two of the three index elements in the vector exceed the threshold, the data o 'is represented' _i The data is abnormal data, judge is a judgment function, and x and y are used for numerical comparison, wherein x and y refer to threshold values and error model element values, and the representative values are used for representing errors.

Preferably, when abnormal data in the data set is obtained, the data is supplemented according to a bilinear interpolation algorithm.

Preferably, the step 2 specifically includes: the combined prediction-festival worm optimization algorithm model is a prediction training sample (O) with N wind speeds _i ，J _i ) ^N _i＝1 In which O is _i To be complete as complete meteorological dataTime-series data of time i in time-series data O, O _i O in (1) _1,i ，o _2,i And o _3,i Respectively historical wind speed, air temperature and air humidity at time i in the time series data, J _i Is a wind speed prediction value, N is a positive integer, N wind speed prediction training samples (O) _i ，J _i ) ^N _i＝1 A meteorological data sample set is formed.

Preferably, the step 2 further comprises: carrying out weighted combination on the combined prediction-disinfestations optimization algorithm model through the disinfestations optimization algorithm to obtain a combined model, wherein the predicted value of the combined model is shown in a formula (8) to a formula (11):

j _2,i ＝μ ₂ W ₁ j _3,i-1 +μ ₂ j _2,i-1 +ζb _f (11)

in the formula, mu ₁ And mu ₂ The weight coefficients, T, corresponding to the autoregressive heterovariance model and the classification tree model respectively ₁ And T ₂ The prediction errors of the autoregressive heteroscedasticity model and the classification tree model are respectively, and the model with small error is endowed with larger weight H ₁ And H ₂ And the prediction results of the autoregressive heteroscedastic model and classification tree model, respectively, on the complete time series data, H _t Is the prediction result of the combined model, j _1,i-1 Is the wind speed value of the previous moment, j _1,i Is the predicted wind speed value at time i, j _2,i-1 Is the gas temperature value at the previous moment, j _3,i-1 Is the air humidity value at the previous moment, W ₁ Is normalized proximity, b _f Is a combined model proximityζ is the activation function, e ⁽⁾ Is an exponential function.

Preferably, the step 3 specifically includes: uniformly dividing a meteorological data sample set into n independent subsets, training the n-1 independent subsets as training samples by using an insect-saving optimization algorithm in turn to obtain training values, and taking the rest 1 as verification samples to obtain a classification test error e when each subset is taken as a test set, so that a stable and optimal error index is obtained, and the wind speed predicted value under the error index is the obtained wind speed predicted value.

Preferably, the classification test error e when each subset is obtained as the test set is obtained through a festival worm optimization algorithm, and the festival worm optimization algorithm includes:

firstly, initializing an insect-node optimization algorithm, randomly generating parent insect-nodes U multiplied by V multiplied by rho, starting to generate offspring insect-nodes by weaving mutual insect-nodes, judging whether the insect-nodes have attached leaf-nodes, calculating the fitness value of the unattached insect-nodes, screening out the optimal insect-nodes through iterative comparison, returning the unattached insect-nodes to search for the attached leaf-nodes, screening out the successfully attached offspring insect-nodes to replace the old optimal insect-nodes, and outputting an optimal characteristic extraction matrix corresponding to the optimal insect-nodes when the screened insect-nodes meet adaptive fitness constraint.

Preferably, the step 4 specifically includes: firstly, the air density is set to be a fixed value of 1.225kg/m ³ And selecting the wind turbine generator as a GE1.5s double-fed induction type wind turbine generator or a corresponding generator model, introducing the wind speed predicted value obtained in the step 3, calculating a wind power value by referring to a standard power characteristic curve of the wind turbine generator, and accumulating the power values of the wind turbine generators in the wind power field to obtain the wind power value of the wind power field.

A wind power plant wind power prediction device comprises:

the collection module is used for collecting historical wind speed, air temperature and air humidity data of the wind power plant and processing abnormal data and missing data in the database by adopting a differential algorithm;

a construction module for constructing a combined prediction-arthromy optimization algorithm model as a combined prediction model;

the acquisition module is used for acquiring a wind speed predicted value from the meteorological information of the data of the wind power plant by using a section combination prediction-section worm optimization algorithm model and improving the wind speed prediction precision;

and the calculation module is used for calculating a wind power value based on the obtained wind speed predicted value and the standard power curve of the wind power generation unit, and accumulating the power of all the fans in the wind power plant to obtain the wind power value of the wind power plant.

Preferably, the collection module is further configured to input the original wind speed, air temperature and air humidity data of the wind farm every fifteen minutes and calculate and obtain a smoothing factor l, and an equation of the smoothing factor l is calculated as follows:

wherein Y is _i Comprising Y _1,i 、Y _2,i And Y _3,i， Y _1,i 、Y _2,i And Y _3,i Respectively representing the wind speed, air temperature and air humidity variables input at time i, Y _i Representing the raw wind speed, air temperature and air humidity data input at time i, E _i The observed values of the variables of wind speed, air temperature and air humidity input at the moment i are represented by E _i And Y _i Form an error model ER ⁱ I and t are integers;

applying difference algorithm to error model ER ⁱ In which a noisy data sequence { ERN }is found ^k Removing noise points from each error metric index to obtain a corrected error model { ERF ^k Therein { ERF } ^k }＝{ER ^k }-{ERN ^k }，k＝1,2,3，ERF ^k Is an error metric, ERF ^k Respectively establishing three corresponding error measurement indexes for the wind speed, the air temperature and the air humidity, and then correcting an error model { ERF } ^k Each error metric index of the algorithm calculates a corresponding error threshold th according to formula (1) ^k :

th ^k ＝mean(ERF ^k )+αsd(ERF ^k ),k＝1,2,3 (1)

Wherein mean isThe mean function, sd, is the standard deviation function, α tuning coefficient. ERF ^k Is an error metric value, and then the anomaly detection of the error metric is carried out on the meteorological data points by a voting mechanism.

Preferably, the collection module is also for anomaly detection of the error metric consisting of three main indicators, MAE, MAPE and MASE respectively;

o’ _i is a meteorological data observation element transverse vector comprising o' _1,i ，o’ _2,i And o' _3,i Three elements, respectively wind speed, air temperature and air humidity in the meteorological data observations at time i, e _i Is a predicted value element of meteorological data, er _i The error function is an error model element vector, i is a meteorological data serial number, n represents the number, MAE _i Denotes mean absolute error, MAPE _i Representing mean absolute percent error, MASE _i Mean absolute scale error is expressed and derived from equations (2) to (5):

if the majority of error metrics in the plurality of error metrics of a data point exceed the corresponding threshold in the modified error model, the data point is regarded as an abnormal data point; the finally obtained abnormal point result of the time-series meteorological data abnormal detection can be expressed by a data set according to the formula (6) and the formula (7) as follows:

Preferably, the collection module is further configured to complete the data according to a bilinear interpolation algorithm when obtaining abnormal data in the data set.

Preferably, the construction module is further used for the combined prediction-festival worm optimization algorithm model to predict training samples (O) for N wind speeds _i ，J _i ) ^N _i＝1 In which O is _i Time series data for time i in complete time series data O as complete meteorological data, O _i O in (1) _1,i ，o _2,i And o _3,i Respectively historical wind speed, air temperature and air humidity at time i in the time series data, J _i Is a wind speed prediction value, N is a positive integer, N wind speed prediction training samples (O) _i ，J _i ) ^N _i＝1 A set of meteorological data samples is constructed.

Preferably, the building module is further configured to perform weighted combination on the combined prediction-festival optimization algorithm model through an festival optimization algorithm to obtain a combined model, and a predicted value of the combined model is shown in formulas (8) to (11):

j _2,i ＝μ ₂ W ₁ j _3,i-1 +μ ₂ j _2,i-1 +ζb _f (11)

in the formula, mu ₁ And mu ₂ The weight coefficients, T, corresponding to the autoregressive heteroscedasticity model and the classification tree model, respectively ₁ And T ₂ The prediction errors of the autoregressive heteroscedasticity model and the classification tree model are respectively determined, and the model with small error is given larger weight H ₁ And H ₂ And the prediction results of the autoregressive heteroscedastic model and classification tree model, respectively, on the complete time series data, H _t Is the prediction result of the combined model, j _1,i-1 Is the wind speed value of the previous moment, j _1,i Is the predicted wind speed value at time i, j _2,i-1 Is the gas temperature value at the previous moment, j _3,i-1 Is the air humidity value at the previous moment, W ₁ Is normalized proximity, b _f Is the combined model proximity, ζ is the activation function, e ⁽⁾ Is an exponential function.

Preferably, the obtaining module is further configured to uniformly divide the meteorological data sample set into n independent subsets, select n-1 independent subsets in turn as training samples by using an arthromy optimization algorithm to perform training to obtain training values, and obtain a classification test error e when each subset is used as a test set by using the rest 1 independent subset as a verification sample, so as to obtain a stable and optimal error index, where the wind speed prediction value under the error index is the obtained wind speed prediction value.

Preferably, the obtaining module is further configured to obtain the classification test error e of each subset as the test set by using a festival worm optimization algorithm, where the festival worm optimization algorithm includes:

firstly, initializing an insect-saving optimization algorithm, randomly generating parent insects of U multiplied by V multiplied by rho, weaving mutual insects to start generating offspring insects, judging whether the insects have leaf adhesion, calculating the fitness value of the insects which are not adhered, screening out the optimal insects through iterative comparison, returning the insects which are not adhered to find the leaf adhesion, screening out the insects which are successfully adhered to the offspring to replace the old optimal insects, and outputting an optimal characteristic extraction matrix corresponding to the optimal insects when the screened insects meet adaptive fitness constraint.

Preferably, the calculation module is also configured to first set the air density to a constant value of 1.225kg/m ³ Selecting a GE1.5s double-fed induction type wind turbine generator or a corresponding generator model, introducing the wind speed predicted value obtained in the step 3, calculating a wind power value by referring to a standard power characteristic curve of the wind turbine generator, and accumulating the power values of the wind turbine generators in the wind power place to obtain the wind power value of the wind power plant.

A terminal comprising a processor and a storage medium;

the storage medium is used for storing instructions;

the processor is used for operating according to the instruction to execute the steps of the wind power prediction method of the wind power plant.

A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the wind farm wind power prediction method.

Compared with the prior art, the method has the advantages that the historical wind speed, air temperature and air humidity data of the wind power plant are collected, abnormal data and missing data in the database are processed by adopting a differential algorithm, and the accuracy and integrity of the data are improved. And introducing optimized meteorological data such as wind speed, air temperature and air humidity of the wind power plant as input of the combined prediction model, and effectively improving the wind speed prediction precision by combining the combined prediction model with the insect-saving optimization algorithm. The wind power is obtained by introducing the wind speed value and combining with a standard power characteristic curve of the wind turbine generator, the wind power is calculated according to the standard power characteristic curve corresponding to the specific fan model, the wind power values of the wind power field are obtained by accumulating the power of each fan of the wind power field, and the short-term wind power prediction precision is improved.

Drawings

FIG. 1 is a diagram of a data structure of an original wind farm optimized by a differential algorithm according to the present invention.

FIG. 2 is a structural diagram of meteorological data extraction by the festival worm optimization model of the present invention.

FIG. 3 is a flowchart of the festival worm optimization algorithm of the present invention.

Fig. 4 is an overall structure diagram of wind power prediction according to the present invention.

Detailed Description

The present application is further described below with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present application is not limited thereby.

The invention provides a wind power prediction method for a wind power plant, aiming at solving the problems that key meteorological factors are difficult to extract for wind power prediction and the matching of the wind speed and the wind power fluctuation process is poor, and the specific flow is shown in FIG. 4.

In order to achieve the above purpose, the method for predicting wind power of a wind power plant specifically comprises the following steps:

step 1, collecting historical wind speed, air temperature and air humidity data of a wind power plant, and processing abnormal data and missing data in a database by adopting a differential algorithm to improve the accuracy and integrity of the data;

specifically, a difference algorithm is adopted to detect and deduce meteorological data with different expected values from a historical meteorological data set of the wind power plant, detect anomalies accurately from a large amount of time sequence meteorological data, and finish the replacement of anomalous data and missing data according to data trends.

In a preferred but non-limiting embodiment of the invention, said step 1 comprises in particular: inputting original wind speed, air temperature and air humidity data of the wind power plant every fifteen minutes, calculating to obtain a smoothing factor l, and solving an equation of the smoothing factor l as follows:

wherein Y is _i Comprising Y _1,i 、Y _2,i And Y _3,i ，Y _1,i 、Y _2,i And Y _3,i A single-row vector Y respectively representing the wind speed, air temperature and air humidity variables input at the time i _i Representing the raw wind speed, air temperature and air humidity data input at time i, E _i A single-row vector consisting of observed values of wind speed, air temperature and air humidity variables input at time i, and input E _i And Y _i Form an error model ER ⁱ The error model represents the error drawing of the wind speed, air temperature and air humidity data actually measured by the wind power plant and the wind speed, air temperature and air humidity data presumed according to the historical data of the wind speed, air temperature and air humidity, and i and t are integers;

as shown in FIG. 1, a difference algorithm is applied to the error model ER ⁱ In the data sequence of the noise point (ERN) ^k Removing noise points from each error metric index to obtain a corrected error model { ERF ^k Therein { ERF } ^k }＝{ER ^k }-{ERN ^k }，k＝1,2,3，ERF ^k Is an error metric, ERF ^k Respectively establishing three corresponding error measurement indexes for the wind speed, the air temperature and the air humidity, and then correcting an error model { ERF ^k Calculating corresponding error threshold value th according to formula (1) for each error metric index ^k :

th ^k ＝mean(ERF ^k )+αsd(ERF ^k ),k＝1,2,3 (1)

In a preferred but non-limiting embodiment of the invention, the anomaly detection of the Error metric consists of three main indicators, which are MAE (Mean Absolute Error), MAPE (Mean Absolute Percentage Error) and mask (Mean Absolute Scaled Error), respectively;

o’ _i is the meteorological data observed value element transverse vector, meteorological dataThe observed value element transverse vector includes o' _1,i ，o’ _2,i And o' _3,i Three elements, respectively wind speed, air temperature and air humidity in the meteorological data observations at time i, e _i Is a predicted value element of meteorological data, er _i The error function is an error model element vector, i is a meteorological data serial number, n represents the number, MAE _i Denotes mean absolute error, MAPE _i Representing mean absolute percent error, MASE _i Represents the mean absolute scale error and is derived from equations (2) through (5):

if the majority of error measurement indexes in a plurality of error measurement indexes of a data point (one data point is the historical wind speed, air temperature and air humidity data of the wind farm collected in a set time interval) exceed corresponding thresholds (the thresholds are preset according to actual conditions) in the corrected error model, the data point is regarded as an abnormal data point; through a voting mechanism formed by three error measurement indexes, if more than half of the measurement indexes determine that the data point is abnormal, the data point is finally determined to be an abnormal data point, and as described above, the finally obtained abnormal point result of the time-series meteorological data abnormality detection can be represented by a data set according to the formula (6) and the formula (7):

ED _i is an outlier data set, ed outlier data element, representing er ^k _i And th ^k Means er _i And th the second element of the vector, when ^k _i If two of the three index elements in the vector exceed the threshold, the data o 'is represented' _i The data is abnormal data, judge is a judgment function, and x and y are used for numerical comparison, wherein x and y refer to threshold values and error model element values, and the representative values are used for representing errors.

In a preferred but non-limiting embodiment of the present invention, when abnormal data in a data set is obtained, the data is complemented according to a bilinear interpolation algorithm, which performs linear interpolation by using values of 4 adjacent points of the abnormal data and giving different weights according to the distance from the interpolation point. The method has an averaged low-pass filtering effect, and the edge is smoothed to generate a relatively coherent output to obtain complete time sequence data O.

As shown in fig. 2, step 2: constructing a combined prediction-festival worm optimization algorithm model as a combined prediction model;

specifically, the method in step 2 is implemented by taking the historical wind speed, air temperature and air humidity data of the wind power plant in the obtained complete time sequence data O as the input of a combined prediction-festival worm optimization algorithm, constructing a combined prediction model based on an autoregressive variance model and a classification tree model, establishing a meteorological data extraction matrix to obtain meteorological data including wind speed, air temperature and air humidity due to the randomness of weather, training by adopting the festival worm optimization algorithm to obtain optimized meteorological data, inputting the optimized meteorological data into the combined prediction model, and outputting a wind speed prediction result.

In a preferred but non-limiting embodiment of the invention, said step 2 comprises in particular: the combined prediction-festival worm optimization algorithm model is designed to set N wind speed predictionsTraining sample (O) _i ，J _i ) ^N _i＝1 In which O is _i Time series data for time i in complete time series data O as complete meteorological data, O _i O in (1) _1,i ，o _2,i And o _3,i Respectively historical wind speed, air temperature and air humidity at time i in the time series data, J _i Is a wind speed prediction value, N is a positive integer, N wind speed prediction training samples (O) _i ，J _i ) ^N _i＝1 A set of meteorological data samples is constructed. The model facilitates optimizing meteorological features by subsequently adopting an insect-saving optimization algorithm to obtain a wind speed predicted value J _i ，J _i From the predicted wind speed value j at each time i _1,i Constitution j _1,i Contains j _1,i-1 、j _2,i-1 And j _3,i-1 Is j _1,i-1 Is the predicted wind speed value at time i, j _2,i-1 Is the predicted air temperature value at time i, j _3,i-1 The air humidity value predicted at the moment i is input into a combined prediction model to obtain a predicted wind speed value j _1,i 。

In a preferred but non-limiting embodiment of the invention, said step 2 further comprises: carrying out weighted combination on the combined prediction-festival worm optimization algorithm model through an festival worm optimization algorithm to obtain a combined model, wherein the predicted value of the combined model is shown in a formula (8) to a formula (11):

j _2,i ＝μ ₂ W ₁ j _3,i-1 +μ ₂ j _2,i-1 +ζb _f (11)

in the formula, mu ₁ And mu ₂ The weight coefficients, T, corresponding to the autoregressive heterovariance model and the classification tree model respectively ₁ And T ₂ The prediction errors of the autoregressive heteroscedasticity model and the classification tree model are respectively, and the model with small error is endowed with larger weight H ₁ And H ₂ And the prediction results of the autoregressive heterovariance model and the classification tree model on the complete time series data, H _t Is the prediction result of the combined model, j _1,i-1 Is the wind speed value of the previous moment, j _1,i Is the predicted wind speed value at time i, j _2,i-1 Is the gas temperature value at the previous moment, j _3,i-1 Is the air humidity value at the previous moment, W ₁ Is normalized proximity, b _f Is the combined model proximity, ζ is the activation function, e ⁽⁾ Is an exponential function.

And step 3: acquiring a wind speed predicted value from the meteorological information of the wind power plant data by using a section combination prediction-section worm optimization algorithm model, and improving the wind speed prediction precision; the meteorological data extraction process is predicted based on n classification tests and festival worm optimization-combination as follows:

in a preferred but non-limiting embodiment of the invention, said step 3 comprises in particular: uniformly dividing a meteorological data sample set into n independent subsets, training the n-1 independent subsets as training samples by using a festival worm optimization algorithm in turn to obtain training values, and taking the rest 1 independent subsets as verification samples to obtain a classification inspection error e when each subset is taken as a test set, so as to obtain a stable and optimal error index, wherein a wind speed predicted value under the error index is the obtained wind speed predicted value. And e can directly reflect the advantages and disadvantages of the popularization capability of the prediction model, and the optimization is carried out on e through a node optimization algorithm.

In a preferred but non-limiting embodiment of the present invention, the classification test error e when each subset is obtained as the test set is obtained by a festival worm optimization algorithm, which comprises:

specific festival worm optimization algorithm (the festival worm optimization algorithm of the invention is coral reef algorithm, the festival worm is coral worm) as shown in fig. 3 and fig. 4, specifically as follows:

The insect-saving optimization algorithm specifically comprises the following steps: initializing, and setting a known shape with the node size of U multiplied by V, wherein the U multiplied by V nodes can be attached by the budworms, and the ratio of the attached node to all nodes is rho. Setting the internode ratio sigma of the internodes, the division reproduction ratio gamma, the attachment limit times of the filial generation internodes, the probability of each cycle elimination as epsilon, the elimination quantity ratio as delta and the maximum iteration times as psi;

a number of desmosomes UXVXrho had attached to the nodal leaves, and a proportion of the interbody beetles UXVXrho Xsigma was taken as parent C _1,a And C _2,a And 2 internode offspring c shown in formula (12) are generated according to formula (9) by combining a mode of simulating two-step internode optimality _1,a And c _2,a ：

Where α is the number of iterations, φ is a random variable generated according to equation (13), the maximum number of iterations is ψ,

wherein τ is a random number of (0, 1), n is a cross constant, and the remaining U.times.V.times.p.times. (1- σ) number of the cross noduli C produces a daughter noduli C according to the formula (14) _a ；

The offspring nodulids need to find the node leaves to attach, and at this time, the node leaves with the number of U multiplied by V multiplied by (1-rho) are not attached. c. C _a Is the offspring of the arthropod, C _a Is a function of the range of the internode, rand, and the worst health degree of the parents is C ^min _a The parent population health degree optimal festival worm is C ^max _a The offspring internodes randomly search for the node leaf points, if the node leaf points are empty, the offspring internodes can be successfully attached, and if the node leaf points are attached by other internodes, the respective health degree value (fitness value fit (c) needs to be calculated _a ) Preferably, will preempt the leaf node. Repeatedly searching for the nodulids which are not successfully attached according to the steps, and if the descendant nodulids are not successfully attached within the limit times u, the nodulids die;

the dominant nodulids at a ratio of γ produce progeny by way of division that attempt to attach. And (3) eliminating the healthy festival insects with poor health degree in a ratio of delta by eliminating the possibility of epsilon in each cycle. The eliminated festival insects die automatically, leaves are vacated so that other festival insects compete;

and when the maximum iteration times psi is reached repeatedly for multiple times, the health optimum festival worm c on the node leaf point is the optimum solution. And (4) adopting a roulette algorithm to generate random numbers and comparing the random numbers with epsilon in each cycle of the festival worm optimization algorithm to judge whether the current cycle is eliminated. In the early stage of the algorithm, the convergence can be effectively accelerated and the optimal solution is approached. In the middle and later stages of the algorithm, due to the blindness of the roulette algorithm, the diversity of the festival insects is lost, so that the algorithm is stopped and enters a local optimal solution;

in order to solve the blindness of roulette, an improved elimination mechanism is proposed as shown in formula (15):

in the formula, alpha and alpha 'are iteration times and alpha is not equal to alpha', and the internode generation c of the internode is weaved _a Theta is a fitness valueA fixed value difference; maximum number of iterations psi, fitness value fit (c) _a ) When c is | | _α -c _α ' | | is less than or equal to S, namely when the Euclidean distance between 2 individuals is less than the preset value S, | fit (c) _a )-fit(c _a Theta is less than or equal to') and is stopped by the algorithm; epsilon and delta are the probability and the elimination quantity proportion of each cycle elimination in the prior art, and epsilon 'and delta' are the probability and the elimination quantity proportion of each cycle elimination after improvement.

The elimination probability is continuously improved in the early stage of algorithm evolution, and the elimination is forcibly carried out and the elimination proportion is continuously improved in the middle and later stages of the algorithm, so that the diversity of the pests can be always kept, and the situation that the pests are trapped in local optimum is avoided.

Therefore, the specific process of acquiring the wind speed predicted value from the wind power plant data meteorological information by using the section combination prediction-section worm optimization algorithm model is as follows:

(1) Initializing a festival worm optimization-combined prediction model, and setting parameters of an festival worm optimization algorithm and the number of nodes of a combined prediction hidden layer.

(2) Parent beetles generate offspring (meteorological data extraction matrix) F meteorological data extraction matrix is updated along with the generation of offspring of the beetles, and the optimization and updating of the original air temperature, the original air speed and the original air humidity are completed through the updating of the parent beetles and the offspring beetles.

(3) Calculating the meteorological data matrix J = F (o) _i )，o _i And (3) for a randomly input meteorological data matrix, finishing the optimization updating of the original wind speed, air temperature and air humidity through an optimized meteorological data extraction function F (), wherein J is the optimized meteorological data wind speed, air temperature and air humidity.

J _1,i 、J _2,i And J _3,i For optimized meteorological data, wind speed, air temperature and air humidity o _1,i 、o _2,i And o _3,i For the meteorological data before optimization, wind speed, air temperature and air humidity, the fitness value fit (c) _a ) Theta ii is respectively corresponding to the fixed value difference of meteorological data and fitness values at different times, and delta elimination quantity proportionBetween 0 and 1.

(4) And setting the current iteration elimination probability and elimination proportion, eliminating the section e of relatively poor bugs, and eliminating meteorological elements which are useless for wind speed prediction in meteorological data.

(5) Dividing the sample into n parts, selecting n-1 parts as training set in turn, using the rest 1 part as test set, using O data matrix including wind speed, air temperature and air humidity at last moment, substituting optimized meteorological data into combined prediction model to obtain wind speed J _1,i As an output item.

(6) And (3) judging whether the maximum iteration times or the wind speed root mean square error meets the convergence requirement, if not, returning to the step (2), and if so, outputting a predicted wind speed value.

And 4, step 4: and calculating the wind power value based on the obtained wind speed predicted value and the standard power curve of the wind generation set, and accumulating the power of all the fans in the wind power field to obtain the wind power value of the wind power field.

In a preferred but non-limiting embodiment of the invention, said step 4 comprises in particular: firstly, the air density is set to be a fixed value of 1.225kg/m ³ Selecting a GE1.5s double-fed induction type wind turbine generator or a corresponding generator model, introducing the wind speed predicted value obtained in the step 3, calculating a wind power value by referring to a wind turbine generator standard power characteristic curve provided by a manufacturer, accumulating the power values of the wind turbine generators in a wind power place to obtain the wind power value of the wind power plant, and improving the wind power calculation efficiency and calculation precision.

Compared with the prior art, the method has the advantages that the difference algorithm is adopted to process abnormal and missing data of the historical wind speed, the wind direction, the temperature, the humidity and other data, the historical meteorological data set of the wind power plant is optimized, the processed data is introduced, the wind speed of the wind power plant is effectively predicted based on the combined prediction model of the festival worm optimization algorithm and the wind power standard power characteristic curve of the wind speed and the wind power plant, and the wind power predicted value of the wind power plant is obtained according to the wind speed and the wind speed-wind power standard power characteristic curve, so that the calculation efficiency of wind power prediction is effectively improved, and meanwhile the calculation accuracy of the wind power is improved.

The invention relates to a wind power prediction device of a wind power plant, which comprises:

and the calculation module is used for calculating a wind power value based on the acquired wind speed predicted value and the standard power curve of the wind generation set, and accumulating the power of all the fans in the wind power plant to acquire the wind power value of the wind power plant.

In a preferred but non-limiting embodiment of the present invention, the collection module is further configured to input the raw wind speed, air temperature and air humidity data of the wind farm every fifteen minutes and calculate and obtain a smoothing factor l, wherein an equation of the smoothing factor l is as follows:

th ^k ＝mean(ERF ^k )+αsd(ERF ^k ),k＝1,2,3 (1)

In a preferred but non-limiting embodiment of the invention, the collection module is also used for anomaly detection of the Error metric and consists of three main indicators, which are MAE (Mean Absolute Error), MAPE (Mean Absolute Percentage Error) and mask (Mean Absolute Scaled Error), respectively;

o’ _i is a meteorological data observation value element transverse vector including o' _1,i ，o’ _2,i And o' _3,i Three elements, respectively wind speed, air temperature and air humidity in the meteorological data observations at time i, e _i Is a predicted value element of meteorological data, er _i The error function is an error model element vector, i is a meteorological data serial number, n represents the number, MAE _i Denotes mean absolute error, MAPE _i Representing mean absolute percentage error, MASE _i Mean absolute scale error is expressed and derived from equations (2) to (5):

if the majority of error measurement indexes in a plurality of error measurement indexes of a data point (one data point is the historical wind speed, air temperature and air humidity data of the wind farm collected in a set time interval) exceed corresponding thresholds (the thresholds are preset according to actual conditions) in the corrected error model, the data point is regarded as an abnormal data point; through a voting mechanism formed by three error metrics, if more than half of the metrics determine that the data point is abnormal, the data point is finally determined to be an abnormal data point, and as described above, the finally obtained abnormal point result of the time-series meteorological data abnormality detection can be represented by a data set according to the formula (6) and the formula (7):

ED _i is an outlier data set, ed outlier data element, representing er ^k _i And th ^k Express er _i And the second element of the th vector, when ^k _i If two of the three index elements in the vector exceed the threshold, the data o 'is represented' _i The abnormal data is judge, the judge is a judgment function, and the values are compared by using x and y, wherein the x and y refer to threshold values and error model element values, and the representative values are used for representing errors.

In a preferred but non-limiting embodiment of the present invention, the collection module is further configured to, when obtaining abnormal data in the data set, complement the data according to a bilinear interpolation algorithm, where the bilinear interpolation algorithm performs linear interpolation by using values of 4 adjacent points of the abnormal data and assigning different weights according to distances from the values to an interpolation point.

In a preferred but non-limiting embodiment of the invention, the building module is further configured to apply the combined prediction-budworm optimization algorithm model to predict training samples (O) for N wind speeds _i ，J _i ) ^N _i＝1 In which O is _i Time series data for time i in complete time series data O as complete meteorological data, O _i O in _1,i ，o _2,i And o _3,i Respectively historical wind speed, air temperature and air humidity at time i in the time series data, J _i Is a wind speed prediction value, N is a positive integer, N wind speed prediction training samples (O) _i ，J _i ) ^N _i＝1 A set of meteorological data samples is constructed. The model facilitates optimizing meteorological features by subsequently adopting an insect-saving optimization algorithm to obtain a wind speed predicted value J _i ，J _i From the predicted wind speed value j at each time i _1,i Constitution j _1,i Contains j _1,i-1 、j _2,i-1 And j _3,i-1 Is j _1,i-1 Is the predicted wind speed value at time i, j _2,i-1 Is the predicted air temperature value at time i, j _3,i-1 The air humidity value predicted at the moment i is input into a combined prediction model to obtain a predicted wind speed value j _1,i 。

In a preferred but non-limiting embodiment of the present invention, the building module is further configured to perform weighted combination on the combined prediction-festival optimization algorithm model through an festival optimization algorithm to obtain a combined model, where the predicted value of the combined model is shown in formulas (8) to (11):

j _2,i ＝μ ₂ W ₁ j _3,i-1 +μ ₂ j _2,i-1 +ζb _f (11)

in the formula, mu ₁ And mu ₂ The weight coefficients, T, corresponding to the autoregressive heterovariance model and the classification tree model respectively ₁ And T ₂ The prediction errors of the autoregressive heteroscedasticity model and the classification tree model are respectively, and the model with small error is endowed with larger weight H ₁ And H ₂ And the prediction results of the autoregressive heteroscedastic model and classification tree model, respectively, on the complete time series data, H _t Is the prediction result of the combined model, j _1,i-1 Is the wind speed value of the previous moment, j _1,i Is the predicted wind speed value at time i, j _2,i-1 Is the gas temperature value at the previous moment, j _3,i-1 Is the air humidity value, W, at the previous moment ₁ Is normalized proximity, b _f Is the combined model proximity, ζ is the activation function, e ⁽⁾ Is an exponential function.

In a preferred but non-limiting embodiment of the present invention, the obtaining module is further configured to uniformly divide the meteorological data sample set into n independent subsets, select n-1 independent subsets in turn as training samples to be trained by using an arthromyodynia optimization algorithm to obtain training values, and obtain a classification test error e when each subset is used as a test set by using the remaining 1 independent subset as a verification sample, so as to obtain a stable and optimal error index, where the wind speed prediction value under the error index is the obtained wind speed prediction value. And e can directly reflect the advantages and disadvantages of the popularization capability of the prediction model, and the optimization is carried out on e through a node optimization algorithm.

In a preferred but non-limiting embodiment of the present invention, the obtaining module is further configured to obtain the classification test error e of each subset as the test set by using a festival worm optimization algorithm, where the festival worm optimization algorithm includes:

In a preferred but non-limiting embodiment of the invention, the calculation module is also adapted to first set the air density to a constant value of 1.225kg/m ³ Selecting a GE1.5s double-fed induction type wind turbine generator or a corresponding generator model, introducing the wind speed predicted value obtained in the step 3, calculating a wind power value by referring to a wind turbine generator standard power characteristic curve provided by a manufacturer, accumulating the power values of the wind turbine generators in a wind power place to obtain the wind power value of the wind power plant, and improving the wind power calculation efficiency and calculation precision.

The invention discloses a terminal, which comprises a processor and a storage medium;

the storage medium is used for storing instructions;

The invention relates to a computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the wind farm wind power prediction method.

The present applicant has described and illustrated embodiments of the present invention in detail with reference to the accompanying drawings, but it should be understood by those skilled in the art that the above embodiments are only preferred embodiments of the present invention, and the detailed description is only for the purpose of helping the reader to better understand the spirit of the present invention, and not for the purpose of limiting the scope of the present invention, and on the contrary, any modifications or modifications based on the spirit of the present invention should fall within the scope of the present invention.

Claims

1. A wind power prediction method for a wind power plant is characterized by comprising the following steps:

step 2: constructing a combined prediction-festival optimization algorithm model as a combined prediction model;

and step 3: acquiring a wind speed predicted value from the meteorological information of the wind power plant data by using a section combination prediction-section worm optimization algorithm model, and improving the wind speed prediction precision;

2. The wind farm wind power prediction method according to claim 1, characterized in that the step 1 specifically comprises: inputting original wind speed, air temperature and air humidity data of the wind power plant every fifteen minutes, calculating to obtain a smoothing factor l, and solving an equation of the smoothing factor l as follows:

wherein Y is _i Comprising Y _1,i 、Y _2,i And Y _3,i， Y _1,i 、Y _2,i And Y _3,i Respectively representing the wind speed, air temperature and air humidity variables input at time i, Y _i Representing the raw wind speed, air temperature and air humidity data input at time i, E _i The observed values of the wind speed, air temperature and air humidity variables input at the time i are represented by E _i And Y _i Form an error model ER ⁱ I and t are integers;

application differenceSub-algorithm in error model ER ⁱ In which a noisy data sequence { ERN }is found ^k Removing noise from each error metric index to obtain a modified error model { ERF } ^k Therein { ERF } ^k }＝{ER ^k }-{ERN ^k }，k＝1,2,3，ERF ^k Is an error metric, ERF ^k Respectively establishing three corresponding error measurement indexes for the wind speed, the air temperature and the air humidity, and then correcting an error model { ERF ^k Each error metric index of the algorithm calculates a corresponding error threshold th according to formula (1) ^k :

th ^k ＝mean(ERF ^k )+αsd(ERF ^k ),k＝1,2,3 (1)

3. The wind farm wind power prediction method according to claim 2, characterized in that the anomaly detection of the error metric consists of three main indicators, which are MAE, MAPE and MASE, respectively;

o’ _i is a meteorological data observation element transverse vector comprising o' _1,i ，o’ _2,i And o' _3,i Three elements, respectively wind speed, air temperature and air humidity in the meteorological data observations at time i, e _i Is a predicted value element of meteorological data, er _i The error function is an error model element vector, i is a meteorological data serial number, n represents the number, MAE _i Denotes mean absolute error, MAPE _i Representing mean absolute percentage error, MASE _i Mean absolute scale error is expressed and derived from equations (2) to (5):

4. The wind power plant wind power prediction method according to claim 3, characterized in that when abnormal data in the data set are obtained, the data are supplemented according to a bilinear interpolation algorithm.

5. The wind farm wind power prediction method according to claim 1, characterized in that the step 2 specifically comprises: the combined prediction-festival worm optimization algorithm model is a prediction training sample (O) with N wind speeds _i ，J _i ) ^N _i＝1 In which O is _i Time series data for time i in complete time series data O as complete meteorological data, O _i O in _1,i ，o _2,i And o _3,i Respectively historical wind speed, air temperature and air humidity at time i in the time series data, J _i Is a wind speed prediction value, N is a positive integer, N wind speed prediction training samples (O) _i ，J _i ) ^N _i＝1 A set of meteorological data samples is constructed.

6. The wind farm wind power prediction method according to claim 5, characterized in that the step 2 further comprises: carrying out weighted combination on the combined prediction-disinfestations optimization algorithm model through the disinfestations optimization algorithm to obtain a combined model, wherein the predicted value of the combined model is shown in a formula (8) to a formula (11):

j _2,i ＝μ ₂ W ₁ j _3,i-1 +μ ₂ j _2,i-1 +ζb _f (11)

in the formula, mu ₁ And mu ₂ The weight coefficients, T, corresponding to the autoregressive heteroscedasticity model and the classification tree model, respectively ₁ And T ₂ Are respectively autoregressive heterosisThe prediction error of the difference model and the classification tree model is given a large weight H to the model with a small error ₁ And H ₂ And the prediction results of the autoregressive heteroscedastic model and classification tree model, respectively, on the complete time series data, H _t Is the prediction result of the combined model, j _1,i-1 Is the wind speed value of the previous moment, j _1,i Is the predicted wind speed value at time i, j _2,i-1 Is the gas temperature value at the previous moment, j _3,i-1 Is the air humidity value at the previous moment, W ₁ Is normalized proximity, b _f Is the combined model proximity, ζ is the activation function, e ⁽⁾ Is an exponential function.

7. The wind farm wind power prediction method according to claim 1, characterized in that the step 3 specifically comprises: uniformly dividing a meteorological data sample set into n independent subsets, training the n-1 independent subsets as training samples by using a festival worm optimization algorithm in turn to obtain training values, and taking the rest 1 independent subsets as verification samples to obtain a classification inspection error e when each subset is taken as a test set, so as to obtain a stable and optimal error index, wherein a wind speed predicted value under the error index is the obtained wind speed predicted value.

8. The wind power plant wind power prediction method according to claim 7, characterized in that the classification test error e obtained when each subset is used as a test set is obtained through a festival worm optimization algorithm, and the festival worm optimization algorithm comprises:

9. The method of claim 1The wind power prediction method for the wind power plant is characterized in that the step 4 specifically comprises the following steps: firstly, the air density is set to be a fixed value of 1.225kg/m ³ Selecting a GE1.5s double-fed induction type wind turbine generator or a corresponding generator model, introducing the wind speed predicted value obtained in the step 3, calculating a wind power value by referring to a standard power characteristic curve of the wind turbine generator, and accumulating the power values of the wind turbine generators in the wind power place to obtain the wind power value of the wind power plant.

10. A wind power prediction device for a wind power plant is characterized by comprising:

the collecting module is used for collecting historical wind speed, air temperature and air humidity data of the wind power plant and processing abnormal data and missing data in the database by adopting a differential algorithm;

the acquiring module is used for acquiring a wind speed predicted value from the meteorological information of the wind power plant data by using the festival combination prediction-festival worm optimization algorithm model and improving the wind speed prediction precision;

11. The wind farm wind power prediction device of claim 10, wherein the collection module is further configured to input raw wind speed, air temperature and air humidity data of the wind farm every fifteen minutes and calculate and obtain a smoothing factor/, and an equation of the smoothing factor/, which is obtained is as follows:

wherein Y is _i Comprising Y _1,i 、Y _2,i And Y _3,i ，Y _1,i 、Y _2,i And Y _3,i Respectively representing the wind speed, air temperature and air humidity variables input at time i, Y _i Representing the raw wind speed, air temperature and air humidity data input at time i, E _i The observed values of the wind speed, air temperature and air humidity variables input at the time i are represented by E _i And Y _i Form an error model ER ⁱ I and t are integers;

th ^k ＝mean(ERF ^k )+αsd(ERF ^k ),k＝1,2,3 (1)

Wherein mean is a mean function, sd is a standard deviation function, and α tuning coefficient. ERF ^k Is an error metric value, and then the anomaly detection of the error metric is carried out on the meteorological data points by a voting mechanism.

12. The wind farm wind power prediction device of claim 11, wherein the collection module is further configured to detect an anomaly in the error metric consisting of three primary indicators, the three primary indicators being MAE, MAPE, and MASE, respectively;

o’ _i is a meteorological data observation value element transverse vector including o' _1,i ，o’ _2,i And o' _3,i Three elements, respectively wind speed, air temperature and air humidity in the meteorological data observations at time i, e _i Is a predicted value element of meteorological data, er _i The error function is an error model element vector, i is a meteorological data serial number, n represents the number, MAE _i Representing mean absolute error，MAPE _i Representing mean absolute percent error, MASE _i Mean absolute scale error is expressed and derived from equations (2) to (5):

ED _i is an outlier data set, ed outlier data element, representing er ^k _i And th ^k Means er _i And th the second element of the vector, when ^k _i If two of the three index elements in the vector exceed the threshold, the data o 'is represented' _i Is the data that is the exception data,judge is a judgment function that compares values using x and y, which refer to threshold values and error model element values used by a vector to represent errors.

13. The wind farm wind power prediction device according to claim 12, wherein the collection module is further configured to, when abnormal data in the data set is obtained, complement the data according to a bilinear interpolation algorithm.

14. The wind farm wind power prediction device according to claim 10, characterized in that the construction module is further configured to apply the combined prediction-arthromy optimization algorithm model to predict training samples (O) for N wind speeds _i ，J _i ) ^N _i＝1 In which O is _i Time series data for time i in complete time series data O as complete meteorological data, O _i O in _1,i ，o _2,i And o _3,i Respectively historical wind speed, air temperature and air humidity at time i in the time series data, J _i Is a wind speed prediction value, N is a positive integer, N wind speed prediction training samples (O) _i ，J _i ) ^N _i＝1 A set of meteorological data samples is constructed.

15. The wind farm wind power prediction device according to claim 14, wherein the construction module is further configured to perform weighted combination on the combined prediction-festival optimization algorithm model through an festival optimization algorithm to obtain a combined model, and a predicted value of the combined model is as shown in formulas (8) to (11):

j _2,i ＝μ ₂ W ₁ j _3,i-1 +μ ₂ j _2,i-1 +ζb _f (11)

in the formula, mu ₁ And mu ₂ The weight coefficients, T, corresponding to the autoregressive heteroscedasticity model and the classification tree model, respectively ₁ And T ₂ The prediction errors of the autoregressive heteroscedasticity model and the classification tree model are respectively, and the model with small error is endowed with larger weight H ₁ And H ₂ And the prediction results of the autoregressive heterovariance model and the classification tree model on the complete time series data, H _t Is the prediction result of the combined model, j _1,i-1 Is the wind speed value j at the previous moment _1,i Is the predicted wind speed value at time i, j _2,i-1 Is the gas temperature value at the previous moment, j _3,i-1 Is the air humidity value, W, at the previous moment ₁ Is normalized proximity, b _f Is the combined model proximity, ζ is the activation function, e ⁽⁾ Is an exponential function.

16. The wind power plant wind power prediction device according to claim 10, characterized in that the obtaining module is further configured to divide the meteorological data sample set into n independent subsets uniformly, select n-1 independent subsets in turn as training samples to be trained by using an arthromy optimization algorithm to obtain training values, and obtain a classification test error e when each subset is used as a test set by using the remaining 1 independent subset as a verification sample, thereby obtaining a stable and optimal error index, where the wind speed prediction value under the error index is the obtained wind speed prediction value.

17. The wind farm wind power prediction device of claim 16, wherein the obtaining module is further configured to obtain the classification test error e of each subset as the test set by using a bug-saving optimization algorithm, and the bug-saving optimization algorithm comprises:

18. A wind farm wind power prediction device according to claim 10, characterised in that the calculation module is further adapted to first set the air density to a constant value of 1.225kg/m ³ And selecting the wind turbine generator as a GE1.5s double-fed induction type wind turbine generator or a corresponding generator model, introducing the wind speed predicted value obtained in the step 3, calculating a wind power value by referring to a standard power characteristic curve of the wind turbine generator, and accumulating the power values of the wind turbine generators in the wind power field to obtain the wind power value of the wind power field.

19. A terminal comprising a processor and a storage medium;

the storage medium is to store instructions;

the processor is configured to operate according to the instructions of any one of claims 1-9 to perform the steps of the wind farm wind power prediction method.

20. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the wind farm wind power prediction method according to any one of the claims 1 to 9.