CN112836870A

CN112836870A - Method, device, equipment and computer readable storage medium for predicting wind power

Info

Publication number: CN112836870A
Application number: CN202110097370.1A
Authority: CN
Inventors: 刘永前; 张�浩; 阎洁; 韩爽; 李莉; 孟航
Original assignee: North China Electric Power University
Current assignee: North China Electric Power University
Priority date: 2021-01-25
Filing date: 2021-01-25
Publication date: 2021-05-25
Anticipated expiration: 2041-01-25
Also published as: CN112836870B

Abstract

The invention provides a method, a device, equipment and a computer readable storage medium for predicting wind power, which are characterized in that the first historical data and the first future data of a wind power plant to be predicted are respectively converted into a first historical two-dimensional array and a first future two-dimensional array which can be identified by a wind power prediction model by acquiring actual data of various categories before the prediction time of the wind power plant to be predicted, and the future prediction data which can be acquired after the prediction time, namely the first future data, the first historical two-dimensional array is input into an encoding part of the wind power prediction model by using the wind power prediction model built by a coding and decoding neural network, the first future two-dimensional array is input into a decoding part of the wind power prediction model to obtain the prediction power of the wind power plant to be predicted, and the time sequence dependence of wind power prediction is fully considered, therefore, the wind power prediction precision is higher, and the predicted power is more accurate.

Description

Method, device, equipment and computer readable storage medium for predicting wind power

Technical Field

The present disclosure relates to the field of computer technologies, and in particular, to a method, an apparatus, a device, and a computer-readable storage medium for predicting wind power.

Background

The wind power prediction can provide future wind power generation power to guide the operation of a power system and the electric power market transaction, and the method is widely applied to the aspects of spare capacity setting, unit combination, electricity price setting and the like.

At present, a wind power prediction method usually adopts single point location or single source weather forecast data such as future temperature, wind speed and the like to predict wind power.

However, this makes the wind power prediction accuracy not high.

Disclosure of Invention

To solve the above technical problem or to at least partially solve the above technical problem, the present disclosure provides a method, an apparatus, a device, and a computer-readable storage medium for predicting wind power.

In a first aspect, the present disclosure provides a method for predicting wind power, including:

acquiring first historical data and first future data of a wind power plant to be predicted, wherein the first historical data comprises: actual data corresponding to the N sampling time points in the first historical time period, respectively, where the actual data includes: actual wind speed, actual power and actual weather data, wherein N is an integer greater than or equal to 1; the first future data includes: prediction data corresponding to the M sampling time points in the first future time period, respectively, where the prediction data includes: predicting wind speed and weather data, wherein M is an integer greater than or equal to 1;

converting first historical data of the wind power plant to be predicted into a first historical two-dimensional array, wherein the dimensionality of the historical two-dimensional array comprises a first time dimensionality and a first historical data category dimensionality;

converting first future data of the wind power plant to be predicted into a first future two-dimensional array, wherein the dimensionality of the future two-dimensional array comprises a second time dimensionality and a first future data category dimensionality;

inputting a first historical two-dimensional array and a first future two-dimensional array of the wind power plant to be predicted into a wind power prediction model to obtain the predicted power of the wind power plant to be predicted, wherein the predicted power of the wind power plant to be predicted is a predicted value of power corresponding to M sampling time points in a first future time period, the wind power prediction model is a trained encoding and decoding neural network model, the encoding and decoding neural network comprises an encoding part and a decoding part, the first historical two-dimensional array is input into the encoding part, and the first future two-dimensional array is input into the decoding part.

Optionally, after the obtaining of the first historical data and the first future data of the wind farm to be predicted, the method further includes:

correcting the first historical data to obtain first historical corrected data of the wind power plant to be predicted;

correcting the first future data to obtain first future correction data of the wind power plant to be predicted;

correspondingly, the converting the first historical data of the wind farm to be predicted into a first historical two-dimensional array includes: converting the historical correction data of the wind power plant to be predicted into a historical two-dimensional array;

the step of converting the first future data of the wind power plant to be predicted into a future two-dimensional array comprises the following steps: and converting the future correction data of the wind power plant to be predicted into a future two-dimensional array.

Optionally, the modifying the first historical data includes one or more of the following first historical data modification methods:

correcting abnormal data in the first historical data;

detecting whether missing data exists in the first historical data, and if the number of the missing data in the continuous preset range is less than a preset threshold value, supplementing the missing data in the first historical data; if the number of the missing data in the continuous preset range is larger than or equal to a preset threshold value, deleting the data in the continuous preset range;

correcting the prediction data according to the acquired wind turbine generator operation state data of the wind power plant, wherein the wind turbine generator operation state data comprises: pitch angle data, rotational speed data, torque data and unit states of the wind turbine, wherein the unit states include: a shutdown state, a power limit state and a normal operation state;

the correcting the first future data comprises one or more of the following first future data correcting methods:

correcting anomalous data in the first future data;

detecting whether missing data exists in the first future data, and if the number of the missing data in the continuous preset range is less than a preset threshold value, supplementing the missing data in the first future data; if the number of the missing data in the continuous preset range is larger than or equal to a preset threshold value, deleting the data in the continuous preset range;

correcting the prediction data according to the acquired wind turbine generator operation state data of the wind power plant, wherein the wind turbine generator operation state data comprises: pitch angle data, rotational speed data, torque data and unit states of the wind turbine, wherein the unit states include: a shutdown state, a power limit state, and a normal operation state.

Optionally, the converting the first historical data of the wind farm to be predicted into a first historical two-dimensional array includes:

normalizing the first historical data of the wind power plant to be predicted, and converting the first historical data into a first historical two-dimensional array;

the converting the first future data of the wind power plant to be predicted into a first future two-dimensional array comprises the following steps:

and normalizing the first future data of the wind power plant to be predicted, and converting the first future data into a first future two-dimensional array.

Optionally, the method further includes:

training the wind power prediction model;

the training of the wind power prediction model comprises:

acquiring a plurality of samples and supervisory data corresponding to the plurality of samples respectively, wherein each sample comprises: second historical data and second future data, the second historical data comprising: actual data corresponding to the N sampling time points in the second historical time period respectively; the second future data includes: respectively corresponding prediction data of M sampling time points in a second future time period; the supervision data of the samples are corresponding actual powers of M sampling time points in a second future time period;

converting the second historical data into a second historical two-dimensional array, wherein the dimensions of the second historical two-dimensional array comprise a time dimension and a second historical data category dimension; converting the second future data into a second future two-dimensional array, the dimensions of the second future two-dimensional array including a time dimension and a second future data category dimension;

inputting a second historical two-dimensional array and a second future two-dimensional array corresponding to the plurality of samples into a wind power prediction model for training to obtain the predicted power of the plurality of samples;

determining the loss of the plurality of samples according to the predicted power of the plurality of samples and the supervision data corresponding to the plurality of samples respectively;

if the losses of the samples do not meet the convergence condition, continuing to carry out iterative training on the wind power prediction model;

and if the losses of the plurality of samples meet the convergence condition, stopping training to obtain a trained wind power prediction model.

Optionally, after the obtaining of the plurality of samples and the supervisory data corresponding to the plurality of samples, the method further includes:

correcting the second historical data in each sample to respectively obtain second historical corrected data of each sample;

correcting the second future data in each sample to respectively obtain second future corrected data of each sample;

correspondingly, the converting the second history data into a second history two-dimensional array includes: converting the second historical corrected data into a second historical two-dimensional array;

the converting the second future data into a second future two-dimensional array comprises: converting the second future correction data into a second future two-dimensional array.

In a second aspect, the present disclosure provides an apparatus for predicting wind power, comprising:

the obtaining module is used for obtaining first historical data and first future data of a wind power plant to be predicted, wherein the first historical data comprises: actual data corresponding to the N sampling time points in the first historical time period, respectively, where the actual data includes: actual wind speed, actual power and actual weather data, wherein N is an integer greater than or equal to 1; the first future data includes: prediction data corresponding to the M sampling time points in the first future time period, respectively, where the prediction data includes: predicting wind speed and weather data, wherein M is an integer greater than or equal to 1;

the conversion module is used for converting the first historical data of the wind power plant to be predicted into a first historical two-dimensional array, and the dimensionality of the historical two-dimensional array comprises a first time dimensionality and a first historical data category dimensionality;

the conversion module is further configured to: converting first future data of the wind power plant to be predicted into a first future two-dimensional array, wherein the dimensionality of the future two-dimensional array comprises a second time dimensionality and a first future data category dimensionality;

the obtaining module is used for inputting a first historical two-dimensional array and a first future two-dimensional array of the wind power plant to be predicted into a wind power prediction model to obtain the predicted power of the wind power plant to be predicted, the predicted power of the wind power plant to be predicted is a predicted value of power corresponding to M sampling time points in a first future time period, the wind power prediction model is a trained encoding and decoding neural network model, the encoding and decoding neural network comprises an encoding part and a decoding part, the first historical two-dimensional array is input into the encoding part, and the first future two-dimensional array is input into the decoding part.

Optionally, the apparatus further comprises:

the correction module is used for correcting the first historical data to obtain first historical correction data of the wind power plant to be predicted; correcting the first future data to obtain first future correction data of the wind power plant to be predicted;

correspondingly, the conversion module is specifically configured to: converting the historical correction data of the wind power plant to be predicted into a historical two-dimensional array; and converting the future correction data of the wind power plant to be predicted into a future two-dimensional array.

correcting abnormal data in the first historical data;

correcting anomalous data in the first future data;

correcting the prediction data according to the acquired wind turbine generator operation state data of the wind power plant to be predicted, wherein the wind turbine generator operation state data comprises: pitch angle data, rotational speed data, torque data and unit states of the wind turbine, wherein the unit states include: a shutdown state, a power limit state, and a normal operation state.

Optionally, the conversion module is specifically configured to:

normalizing the first historical data of the wind power plant to be predicted, and converting the first historical data into a first historical two-dimensional array; and normalizing the first future data of the wind power plant to be predicted, and converting the first future data into a first future two-dimensional array.

Optionally, the apparatus further comprises:

the training module is used for training the wind power prediction model;

the training module comprises:

the acquisition module is further configured to: acquiring a plurality of samples and supervisory data corresponding to the plurality of samples respectively, wherein each sample comprises: second historical data and second future data, the second historical data comprising: actual data corresponding to the N sampling time points in the second historical time period respectively; the second future data includes: respectively corresponding prediction data of M sampling time points in a second future time period; the supervision data of the samples are corresponding actual powers of M sampling time points in a second future time period;

the conversion module is further configured to: converting the second historical data into a second historical two-dimensional array, wherein the dimensions of the second historical two-dimensional array comprise a time dimension and a second historical data category dimension; converting the second future data into a second future two-dimensional array, the dimensions of the second future two-dimensional array including a time dimension and a second future data category dimension;

the obtaining module is further configured to: inputting a second historical two-dimensional array and a second future two-dimensional array corresponding to the plurality of samples into a wind power prediction model for training to obtain the predicted power of the plurality of samples;

the determining module is used for determining the loss of the plurality of samples according to the predicted power of the plurality of samples and the supervision data corresponding to the plurality of samples respectively;

Optionally, the modification module is further configured to:

correspondingly, the conversion module is specifically configured to: converting the second historical corrected data into a second historical two-dimensional array; converting the second future correction data into a second future two-dimensional array.

In a third aspect, the present disclosure provides an apparatus for predicting wind power, comprising:

a memory for storing processor-executable instructions;

a processor for implementing the method according to the first aspect as described above when the computer program is executed.

In a fourth aspect, the present disclosure provides a computer-readable storage medium having stored therein computer-executable instructions for implementing the method of predicting wind power as described in the first aspect above when the computer-executable instructions are executed by a processor.

Compared with the prior art, the technical scheme provided by the embodiment of the disclosure has the following advantages:

the method comprises the steps of obtaining actual data of various categories before the prediction time of the wind power plant to be predicted, namely first historical data, and future prediction data which can be obtained after the prediction time, namely first future data, namely actual wind speed, actual power and actual weather data which can influence wind power and can be obtained historically in the wind power plant to be predicted, and prediction data which can influence the wind power and can be obtained in the future, and utilizing a wind power prediction model built by a coding and decoding neural network to realize power prediction.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and together with the description, serve to explain the principles of the disclosure.

In order to more clearly illustrate the embodiments or technical solutions in the prior art of the present disclosure, the drawings used in the description of the embodiments or prior art will be briefly described below, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without inventive exercise.

Fig. 1 is a schematic flow chart of a method for predicting wind power according to an embodiment of the present disclosure;

FIG. 2 is a schematic diagram of a data structure of a first historical two-dimensional array and a first future two-dimensional array provided by the present disclosure;

FIG. 3 is a schematic flow chart of another method for predicting wind power according to the embodiment of the present disclosure;

fig. 4 is a schematic structural diagram of a training method of a wind power prediction model according to an embodiment of the present disclosure;

FIG. 5 is a schematic structural diagram of a wind power prediction model provided by the present disclosure;

fig. 6 is a schematic structural diagram of an apparatus for predicting wind power provided in an embodiment of the present disclosure;

fig. 7 is a schematic structural diagram of a device for predicting wind power according to an embodiment of the present disclosure.

Detailed Description

In order that the above objects, features and advantages of the present disclosure may be more clearly understood, aspects of the present disclosure will be further described below. It should be noted that the embodiments and features of the embodiments of the present disclosure may be combined with each other without conflict.

In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present disclosure, but the present disclosure may be practiced in other ways than those described herein; it is to be understood that the embodiments disclosed in the specification are only a few embodiments of the present disclosure, and not all embodiments.

The terms to which the present invention relates will be explained first:

a wind farm, i.e. a wind power plant, refers to a power plant that converts wind energy into mechanical energy and then converts the mechanical energy into electrical energy.

The present disclosure provides a wind farm including, but not limited to: the wind measuring tower, the wind turbine generator, the current collecting line, the booster station and the like. The wind measuring tower receives wind energy, the wind energy is finally converted into electric energy to be output through the output end of the booster station, and the wind power of the wind power plant is the total power output by the output end of the booster station. The wind power prediction power of the wind power plant can be used for guiding the operation of a power system and the electric power market transaction, and is widely applied to the aspects of spare capacity setting, unit combination, electricity price setting and the like.

At present, a wind power prediction method usually adopts single point location or single source weather forecast data such as future temperature, wind speed and the like to predict wind power. However, wind power prediction accuracy is not high.

In order to solve the technical problem, the present disclosure provides a method for predicting wind power, wherein first historical data and first future data of a wind farm to be predicted are input into a trained wind power prediction model, so that wind power prediction power of the wind farm to be predicted is obtained.

The following describes the technical solution of the present disclosure and how to solve the above technical problems with reference to specific examples.

Fig. 1 is a schematic flow diagram of a method for predicting wind power provided in an embodiment of the present disclosure, as shown in fig. 1, the method of the present embodiment is executed by a computer or a server, and the method of the present embodiment includes:

s101, obtaining first historical data and first future data of a wind power plant to be predicted.

Wherein the first history data comprises: actual data corresponding to the N sampling time points in the first historical time period respectively include: actual wind speed, actual power and actual weather data, wherein N is an integer greater than or equal to 1; the first future data includes: prediction data corresponding to the M sampling time points in the first future time period respectively, wherein the prediction data comprises: predicting wind speed and weather data, wherein M is an integer greater than or equal to 1;

in this embodiment, predicting the wind power is to predict the future wind power of the wind farm to be predicted at a time, and it is necessary to obtain actual data (i.e., first historical data) of multiple categories that may affect the future wind power in a first historical time period before the time, and then obtain prediction data (i.e., first future data) of multiple categories that can be obtained in the first future time period after the time, so as to predict the wind power. For example, the wind farm needs to predict the wind power from 5 pm to 5 pm, and the wind power can be predicted by acquiring the first historical data 48 hours before 4 pm and the first future data from 5 pm to 5 pm at 4 pm.

Wherein the first historical data includes but is not limited to: and actual data corresponding to the N sampling time points in the first historical time period respectively. The N sampling time points may be sampling time points for uniform sampling, that is, sampling at intervals of a first preset time step, or non-uniform sampling time points. The actual data may be data measured at N sampling time points, where N is an integer equal to or greater than 1. Actual data for various categories that affect future wind power include, but are not limited to: actual wind speed, actual power, and actual weather data, wherein the actual wind speed includes an actual wind speed value and an actual wind direction, and the actual weather data includes, but is not limited to, one or more of the following: actual air pressure data, actual temperature data, actual humidity data.

Optionally, the actual weather data may be actual weather data of at least one location in the wind farm to be predicted or within a preset distance around the wind farm to be predicted, and further, the actual weather data of one source of one location may also be actual weather data of at least one source of the wind farm to be predicted. For example, one unit of measured weather data of a wind farm to be predicted may be used.

Optionally, the actual weather data may include, but is not limited to, one or more of the following: actual air pressure data, actual temperature data, actual humidity data.

Optionally, the first preset time step is a preset time length, and is a duration between two adjacent sampling time points among the N sampling time points, where the first preset time step may also be called a resolution of the first historical data, and the first preset time step is smaller than the first historical time period, for example, the first preset time step may be 15 minutes, 30 minutes, or 1 hour, and the disclosure is not limited thereto.

Wherein the first future data includes but is not limited to: and respectively corresponding to the M sampling time points in the first future time period. The M sampling time points may be sampling time points of uniform sampling, that is, the sampled prediction data is prediction data acquired every second preset time step, and may also be non-uniform sampling time points. The prediction data may be data predicted at M sampling time points, where M is an integer equal to or greater than 1. Prediction data for various categories that affect future wind power include, but are not limited to: predicted wind speed and predicted weather data, wherein the predicted wind speed includes a predicted wind speed value and a predicted wind direction, and the predicted weather data includes, but is not limited to, one or more of the following: predicted air pressure data, predicted temperature data, predicted humidity data.

Optionally, the predicted weather data may be predicted weather data of at least one position in the wind farm to be predicted or within a preset distance around the wind farm to be predicted, and further, the predicted weather data of one source of one position may also be predicted weather data of at least one source of the wind farm to be predicted. For example, one unit of measured weather data of a wind farm to be predicted may be used.

Optionally, the predicted weather data includes, but is not limited to, one or more of: predicted air pressure data, predicted temperature data, predicted humidity data.

Optionally, the second preset time step is a preset time length, and is a time length between two adjacent sampling time points among the M sampling time points, where the second preset time step may also be called a resolution of the first future data, and the second preset time step is smaller than the first future time period, for example, the second preset time step may be 15 minutes, 30 minutes, or 1 hour, which is not limited in this disclosure.

Further, the first preset time step is the same as the second preset time step.

Optionally, the predicted weather data is Numerical Weather Prediction (NWP) data.

S102, converting first historical data of the wind power plant to be predicted into a first historical two-dimensional array.

The dimensions of the first history two-dimensional array comprise a first time dimension and a first history data category dimension.

S103, converting the first future data of the wind power plant to be predicted into a first future two-dimensional array.

Wherein the dimensions of the first future two-dimensional array include a second time dimension and a first future data category dimension.

In this embodiment, the first historical data and the first future data acquired in S101 are respectively converted into a data form, that is, a two-dimensional array, which can be used for inputting a wind power prediction model, and both the first historical data and the first future data are converted into a two-dimensional array including a time dimension and a data category dimension. The data category dimension is a type of the first historical data and the first future data, one type of data for each action. For each type of second historical data, arranging the actual data of the N sampling time points according to the time sequence to obtain a one-dimensional array, wherein the one-dimensional array can also be called a sequence, and arranging the sequence of each type of data in the same time dimension according to the data of the same sampling time point to obtain a first historical two-dimensional array. For each type of first future data, a one-dimensional array (sequence) can be obtained, and the sequence of each type of data is arranged in the same time dimension according to the data of the same sampling time point to obtain a first future two-dimensional array. It is understood that, in the time dimension, the first history two-dimensional array and the first future two-dimensional array may respectively include a plurality of one-dimensional vectors at different time instants, and each one-dimensional vector is a data value at a sampling time point, which is sequentially arranged according to a data category order.

Optionally, the data category dimension may also include temporal coding. For example, the time code of the first sampling time point of the N sampling time points may be defined as 1, the time code of the second sampling time point may be defined as 2, and so on.

For example, fig. 2 is a data structure schematic diagram of a first historical two-dimensional array and a first future two-dimensional array provided by the present disclosure, as shown in fig. 2, when wind power prediction is performed on a wind farm to be predicted, data of the wind farm to be predicted need to be converted into a first historical two-dimensional array 201 and a first future two-dimensional array 202, a solid square shown in fig. 2 represents actual data in the two-dimensional array and is a time dimension in a horizontal direction, where 1, 2, 3, … …, and T0 respectively represent a 1 st time, a 2 nd time, a 3 rd time, … …, a T0 th time, and a difference between two adjacent times is a preset time step. The first period is the time period from the first timing to the T0 th timing and the second period is the time period from the T0+1 th to the T0+ τ. The vertical direction is a data category dimension, in the first two-dimensional array 201, time is encoded into one category, actual air pressure data is one category, actual temperature data is one category, actual wind speed is one category, actual power is one category, and other categories can be included, which is only an example. Similarly, in the first future two-dimensional array 202, time is encoded as a category, predicted pressure data is a category, predicted temperature data is a category, and predicted wind speed is a category, just to name a few examples.

It should be noted that, the steps S102 and S103 are not executed in a sequential order, and S102 and S103 may be executed first, or S103 and S102 may be executed first, or S102 and S103 may be executed simultaneously.

And S104, inputting the first historical two-dimensional array and the first future two-dimensional array of the wind power plant to be predicted into a wind power prediction model to obtain the predicted power of the wind power plant to be predicted.

The prediction power of the wind power plant to be predicted is a prediction value of power corresponding to M sampling time points in a first future time period, wherein the wind power prediction model is a trained encoding and decoding neural network model, the encoding and decoding neural network comprises an encoding part and a decoding part, a first historical two-dimensional array is input into the encoding part, and a first future two-dimensional array is input into the decoding part.

The first future time period may be 0-4 hours (ultra-short term), or 24-48 hours (short term), and the length of the first future time period is not limited in the present disclosure.

Optionally, the codec neural network includes a plurality of linear or nonlinear conversion layers, an attention mechanism layer and a probability output layer connected in sequence.

The first history two-dimensional array is input to a coding part of the multilayer linear or nonlinear conversion layer, and a first history intermediate representation feature matrix is obtained in the process of extracting features of the first history two-dimensional array. The first future two-dimensional array is input to a decoding part of the multilayer linear or nonlinear conversion layer, and a first future intermediate representation characteristic matrix is obtained for the process of extracting characteristics of the first future two-dimensional array. And the time dimension of the first history intermediate representation characteristic matrix corresponds to the time dimension of the first history two-dimensional array. The time dimension of the first future intermediate representation feature matrix corresponds to the time dimension of the first future two-dimensional array.

The first historical intermediate representation feature matrix and the first future intermediate representation feature matrix are input into the attention mechanism layer, resulting in a weight matrix for the attention representation. The attention-expressing weight matrix can acquire information of a plurality of continuous moments near the query variable, and the important moments are focused on the information through weighted summation, so that the wind power prediction model can efficiently learn the time sequence dependency in the wind power generation process.

And inputting the weight matrix A represented by the attention into a probability output layer to obtain the predicted power of the wind power plant to be predicted, wherein the predicted power of the wind power plant to be predicted is a wind power prediction result at each time step in a second time period.

In this embodiment, by acquiring actual data of various categories before a prediction time of a wind farm to be predicted, that is, first historical data, and future prediction data that can be acquired after the prediction time, that is, first future data, the first historical data includes: actual data corresponding to the N sampling time points in the first historical time period respectively include: actual wind speed, actual power, and actual weather data; the first future data includes: prediction data corresponding to the M sampling time points in the first future time period respectively, wherein the prediction data comprises: predicting wind speed and predicting weather data; the method comprises the steps of converting first historical data and first future data of a wind power plant to be predicted into a first historical two-dimensional array and a first future two-dimensional array which can be identified by a wind power prediction model respectively, inputting the first historical two-dimensional array into a coding part of the wind power prediction model, inputting the first future two-dimensional array into a decoding part of the wind power prediction model, and obtaining predicted power of the wind power plant to be predicted, wherein the predicted power of the wind power plant to be predicted is a predicted value of power corresponding to M sampling time points in a first future time period. The method comprises the steps of utilizing a wind power prediction model built by a coding and decoding neural network to realize power prediction by using actual wind speed, actual power and actual weather data which can be acquired historically and affect wind power of a wind power plant to be predicted and prediction data which can be acquired in the future and affect the wind power, fully considering the influence of the actual data and the prediction data on the wind power, acquiring data at a plurality of moments before and after a prediction moment, utilizing the wind power prediction model, and fully considering the time sequence dependence of wind power prediction, so that the wind power prediction precision is higher, and the predicted power is more accurate.

Based on the above embodiment, further, when the first historical data and the first future data of the wind farm to be predicted, which are obtained, may have missing or abnormal (dead pixel) data, etc., the first historical data and/or the first future data of the wind farm to be predicted, which are obtained, may be respectively corrected, and a specific embodiment is described below.

Fig. 3 is a schematic flow diagram of another method for predicting wind power provided by the embodiment of the present disclosure, and fig. 3 is a flowchart of the embodiment shown in fig. 1, and as shown in fig. 3, S101 may further include S101A and/or S101B after S101:

S101A, the first historical data are corrected, and first historical corrected data of the wind power plant to be predicted are obtained.

The first history data may be corrected to correct missing data, abnormal data, or the like in the first history data, thereby obtaining corrected first history corrected data.

Optionally, the first historical data is corrected, including but not limited to one or more of the following first historical data correction methods:

the first historical data correction method corrects abnormal data in the first historical data.

For each type of actual data in the first historical data, abnormal data can be detected by using an abnormal detection algorithm, and then the abnormal data can be corrected by using a data interpolation algorithm. The abnormal data may be data that does not conform to the trend of the change rule of the whole data.

Detecting whether missing data exists in the first historical data, and if the number of the missing data in the continuous preset range is less than a preset threshold value, supplementing the missing data in the first historical data; and if the quantity of the missing data in the continuous preset range is larger than or equal to the preset threshold value, deleting the data in the continuous preset range.

For each type of actual data in the first historical data, if the number of missing data in the actual data in the continuous preset range is less than a preset threshold, the missing data in the actual data can be interpolated through a data interpolation algorithm. If the number of missing data in the continuous preset range is greater than or equal to the preset threshold, it indicates that the segment of actual data is missing more and cannot be interpolated, and deletes the actual data of the type in the preset range.

The corresponding S102 includes S1021:

s1021, converting the first history correction data of the wind power plant to be predicted into a first history two-dimensional array.

And converting the corrected first history correction data to obtain a first history two-dimensional array.

S101, correcting the first future data to obtain first future correction data of the wind power plant to be predicted.

Optionally, the first future data is corrected, including but not limited to one or more of the following first future data correction methods:

the first future data correction method corrects abnormal data in the first future data.

For each type of predicted data in the first future data, abnormal data therein may be detected using an abnormality detection algorithm, and then the abnormal data may be corrected using a data interpolation algorithm. The abnormal data may be data that does not conform to the trend of the change rule of the whole data.

The second method for correcting the first future data comprises the steps of detecting whether the first future data have missing data or not, and if the number of the missing data in the continuous preset range is less than a preset threshold value, supplementing the missing data in the first future data; and if the quantity of the missing data in the continuous preset range is larger than or equal to the preset threshold value, deleting the data in the continuous preset range.

For each type of predicted data in the first future data, if the number of missing data in the predicted data in the continuous preset range is less than a preset threshold, the missing data in the predicted data can be interpolated by a data interpolation algorithm. If the number of missing data in the continuous preset range is greater than or equal to a preset threshold, it indicates that the section of prediction data is missing more and cannot be interpolated, and deletes the prediction data of the type in the preset range.

Correcting the predicted data according to the acquired wind turbine running state data of the wind power plant to be predicted, wherein the wind turbine running state data comprises: pitch angle data, rotational speed data, torque data and unit states of the wind turbine, the unit states including: a shutdown state, a power limit state, and a normal operation state.

Accordingly, S103 includes S1031:

and S1031, converting the first future correction data of the wind power plant to be predicted into a first future two-dimensional array.

The corrected first future correction data is converted to obtain a first future two-dimensional array.

According to the wind power prediction method and device, the first historical correction data of the wind power plant to be predicted are obtained by correcting the first historical data, the first future data are corrected, and the first future correction data of the wind power plant to be predicted are obtained, so that the data used for wind power prediction are more accurate, the wind power prediction accuracy is higher, and the predicted power is more accurate.

On the basis of the foregoing embodiment, optionally, S102 may include: the first historical data of the wind power plant to be predicted are normalized and converted into a first historical two-dimensional array.

S103 may include: and normalizing the first future data of the wind power plant to be predicted, and converting the first future data into a first future two-dimensional array.

The normalization can be to convert the data into decimal between 0 and 1, so that the data processing is convenient, and the output result of the model is more favorably realized. It is also possible to convert dimensional data into dimensionless data (scalars) so that different units or magnitudes of data can be unified.

According to the embodiment, the first historical data and the first future data are normalized, so that the normalized data are obtained, conversion into a two-dimensional array is facilitated, and the predicted power is obtained by using a wind power prediction model.

For the wind power prediction model, the training can be performed after the wind power prediction model is built, and the training process of the wind power prediction model is specifically described below with reference to fig. 4.

Fig. 4 is a schematic structural diagram of a training method for a wind power prediction model according to an embodiment of the present disclosure, as shown in fig. 4, the method according to the present embodiment is executed by a computer or a server, and the method according to the present embodiment includes:

s401, obtaining a plurality of samples and supervision data corresponding to the samples.

Wherein, each sample comprises: second historical data and second future data, the second historical data comprising: actual data corresponding to the N sampling time points in the second historical time period respectively; the second future data includes: respectively corresponding prediction data of M sampling time points in a second future time period; the supervision data of the samples are the corresponding actual powers of the M sampling time points in the second future time period.

In this embodiment, the data in the obtained multiple samples are from one wind farm or multiple wind farms, where the data in each sample is from the same wind farm, and the data in different samples may be from the same wind farm or from different wind farms, which is not limited in this disclosure. For example, if second historical data and second future data of 3 wind farms are obtained respectively, 3 samples may be formed, and each sample includes data of one wind farm of the 3 wind farms. For another example, multiple samples may be obtained by obtaining multiple sets of second historical data and second future data of a wind farm, multiple times may be selected for data of the wind farm, actual data corresponding to N sampling time points in a second historical time period before the time is obtained as the second historical data, and predicted data corresponding to M sampling time points in a second future time period after the time is obtained as the second future data, so that the second historical data and the second future data may form one sample, where data of the sample is from the sample wind farm.

Wherein the second historical data includes but is not limited to: and actual data corresponding to the N sampling time points in the second historical time period respectively. The N sampling time points are the same as the sampling intervals of the N sampling time points in the embodiment shown in fig. 1, and the second history time period is the same as the first history time period in duration. The actual data may be data measured at N sampling time points, where N is an integer equal to or greater than 1. Actual data for various categories that affect future wind power include, but are not limited to: actual wind speed, actual power, and actual weather data, wherein the actual wind speed includes an actual wind speed value and an actual wind direction, and the actual weather data includes, but is not limited to, one or more of the following: actual air pressure data, actual temperature data, actual humidity data.

Optionally, the actual weather data may be actual weather data of at least one location in the sample wind farm or within a preset distance around the sample wind farm, and further, the actual weather data of the sample wind farm from at least one source may also be actual weather data of one source at one location, and the actual weather data from one source is of one data category. For example, one unit of measured weather data for a sample wind farm may be used.

Wherein the second future data includes, but is not limited to: and the M sampling time points in the second future time period respectively correspond to the predicted data. Wherein the M sampling time points have the same sampling interval as the M sampling time points in the embodiment shown in fig. 1, and the second future time period has the same duration as the first future time period. The prediction data may be data predicted at M sampling time points, where M is an integer equal to or greater than 1. Prediction data for various categories that affect future wind power include, but are not limited to: predicted wind speed and predicted weather data, wherein the predicted wind speed includes a predicted wind speed value and a predicted wind direction, and the predicted weather data includes, but is not limited to, one or more of the following: predicted air pressure data, predicted temperature data, predicted humidity data.

Optionally, the predicted weather data may be predicted weather data of at least one location in the sample wind farm or within a preset distance around the sample wind farm, and further, the predicted weather data of the sample wind farm from at least one source may also be predicted weather data of one source from one location, and the predicted weather data from one source from one location is of one data category. For example, one unit of measured weather data for a sample wind farm may be used.

Optionally, the predicted weather data is numerical weather forecast (multi-empirical weather prediction, abbreviated as multi-WP) data.

S402, converting the second history data into a second history two-dimensional array.

And the dimension of the second history two-dimensional array comprises a time dimension and a second history data category dimension.

And S403, converting the second future data into a second future two-dimensional array.

Wherein the dimensions of the second future two-dimensional array include a time dimension and a second future data category dimension.

In this embodiment, the second historical data and the second future data acquired in S101 are respectively converted into a data form, that is, a two-dimensional array, which can be used for inputting the wind power prediction model, and both the second historical data and the second future data are converted into a two-dimensional array including a time dimension and a data category dimension. The data category dimension is a type of the second historical data and the second future data, each of the rows being one type of data. For each type of second historical data, arranging the actual data of the N sampling time points according to the time sequence to obtain a one-dimensional array, wherein the one-dimensional array can also be called as a sequence, arranging the sequence of each type of data in the same time dimension according to the data of the same sampling time points to obtain a second historical two-dimensional array, and the arrangement sequence of each data type sequence in the second historical two-dimensional array is the same as that of the first historical two-dimensional array. For each type of second future data, a one-dimensional array (sequence) can also be obtained, and the sequence of each type of data is arranged in the same time dimension according to the data of the same sampling time point to obtain a second future two-dimensional array, wherein the arrangement sequence of each data type sequence in the second future two-dimensional array is the same as that of the first future two-dimensional array. It is to be understood that, in the time dimension, the second historical two-dimensional array and the second future two-dimensional array may respectively include a plurality of one-dimensional vectors at different time instants, and each one-dimensional vector is a data value at a sampling time point, which is sequentially arranged according to a data category order.

Optionally, the data category dimension may also include temporal coding.

It should be noted that, the execution of steps S402 and S403 is not in a sequential order, and S402 and S403 may be executed first, or S403 and S402 may be executed first, or S402 and S403 may be executed simultaneously.

S404, inputting a second historical two-dimensional array and a second future two-dimensional array corresponding to the plurality of samples into a wind power prediction model for training to obtain the predicted power of the plurality of samples.

The wind power prediction model is a coding and decoding neural network and comprises a coding part and a decoding part, and aiming at each sample in a plurality of samples, first two-dimensional array data is input into the coding part, and second two-dimensional array data is input into the decoding part.

In a possible implementation manner, data of each sample in the multiple samples may be input into the wind power prediction model, and each sample includes the second history two-dimensional array and the second future two-dimensional array.

In another possible implementation manner, a plurality of samples may also be simultaneously input into the wind power prediction model. Each sample comprises a second historical two-dimensional array and a second future two-dimensional array, and the plurality of second historical two-dimensional arrays and the plurality of second future two-dimensional arrays are respectively superposed, namely, the sample dimension is added on the basis of the two-dimensional arrays, so that a three-dimensional array is obtained, wherein the three dimensions of the three-dimensional array comprise a sample dimension, a time dimension and a data category dimension.

For the two possible implementations described above, the second historical two-dimensional array and the second future two-dimensional array of one sample are input into the codec neural network for illustration.

And inputting the second history two-dimensional array into the coding part of the multi-layer linear or non-linear conversion layer to obtain a second history intermediate representation feature matrix. The second future two-dimensional array is input to a decoding part of the multilayer linear or nonlinear conversion layer to obtain a second future intermediate representation characteristic matrix. And the time dimension of the second history intermediate representation characteristic matrix corresponds to the time dimension of the second history two-dimensional array. The time dimension of the second future intermediate representation feature matrix corresponds to the time dimension of the second future two-dimensional array.

Optionally, the multilayer linear or nonlinear conversion layer may be a multilayer linear or nonlinear conversion layer, which is built by a series of neural network sequence modeling modules, and may include, but is not limited to, one or more of the following network structures: a fully-connected network for parameter sharing, a residual-connected module, a recurrent neural network, a variant of a recurrent neural network (e.g., an episodic memory network, a gated recurrent neural network, or a bi-directional recurrent network), a one-or two-dimensional convolutional neural network, a variant of a convolutional neural network (e.g., a time-series convolutional network, etc.).

The second historical intermediate representation feature matrix and the second future intermediate representation feature matrix are input into the attention mechanism layer, and a weight matrix of the attention representation is obtained.

And taking the obtained second future intermediate representation feature matrix F as a query variable Q of the attention mechanism layer, and taking a spliced feature matrix S of the second historical intermediate representation feature matrix and the second future intermediate representation feature matrix as a key variable K and a value variable V of the attention mechanism layer.

Calculating the similarity between the query variable Q (the second future intermediate representation feature matrix F) and the key variable K (the splicing feature matrix S) to obtain the weight matrix a corresponding to the splicing feature matrix S, where the similarity calculation method may be implemented by various methods in the prior art, for example: (1) can be changed by querying variable QMatrix multiplication QK with transpose of key variable K^TObtaining the similarity; (2) the weight matrix W may be added between the query variable Q and the key variable K_aIncreasing model expression capability and then carrying out matrix multiplication QW_aK^TObtaining the similarity; (3) or by a weight matrix W_aLinear mapping W of the concatenation values of the query variable Q and the key variable K_a[Q；K]Obtaining the similarity; (4) or by weighted matrix W_a、U_aObtaining a degree of similarity, e.g. tanh (W)_aQ+U_aK]。

The weight matrix a is normalized to obtain a new weight matrix e, for example, the weight matrix a may be normalized using a normalization function softmax.

And carrying out weighted summation on the new weight matrix e and the value variable V (the splicing feature matrix S) to obtain the attention-expressing weight matrix A, wherein each value of the attention-expressing weight matrix A corresponds to each column of the value variable V (the splicing feature matrix S).

And inputting the weight matrix A represented by attention into a probability output layer to obtain the predicted power of the sample wind power plant.

And the predicted power of the sample wind power plant is a wind power prediction result at each time step in the second time period.

Optionally, the probability output layer includes multiple linear or nonlinear conversion layers and a fully-connected layer connected in sequence, and the probability output layer may include, but is not limited to, one or more of the following structures: a fully-connected network for parameter sharing, a residual connection module, a recurrent neural network, a variant of a recurrent neural network (e.g., an episodic memory network, a gated recurrent neural network, a bi-directional recurrent network), a one-or two-dimensional convolutional neural network, a variant of a convolutional neural network (e.g., a time-series convolutional network, etc.).

Fig. 5 is a schematic structural diagram of a wind power prediction model provided by the present disclosure, and as shown in fig. 5, a second historical two-dimensional array and a second future two-dimensional array of a sample are respectively input to a multilayer linear or nonlinear conversion layer, that is, a vector at each time is input to the multilayer linear or nonlinear conversion layer 501, so as to obtain a second historical intermediate representation feature matrix and a second future intermediate representation feature matrix. Inputting the second historical intermediate representation feature matrix and the second future intermediate representation feature matrix into the attention mechanism layer 502 to obtain a weight matrix of attention representation, inputting the weight matrix of attention representation into the multiple linear or non-linear conversion layers 503 in the probability output layer, and inputting the obtained result into the full connection layer 504 to obtain the output predicted power, wherein the output predicted power can comprise deterministic output, quantile output, probability density output, interval output and the like.

S405, determining the loss of the plurality of samples according to the predicted power of the plurality of samples and the supervision data corresponding to the plurality of samples.

The loss of the plurality of samples can be calculated according to a preset loss function, and the selection of the loss function can be: the deterministic prediction can use an absolute error loss function or a squared error loss function, the quantile prediction can use a quantile loss function, the probability density prediction can use a maximum likelihood loss function, and the interval prediction can use an optimal interval loss function.

And S406, judging whether the loss of the plurality of samples meets a convergence condition or not.

And performing iterative training on the wind power prediction model, judging whether the wind power prediction model is converged according to whether the loss meets the convergence condition, and stopping training when the wind power prediction model is converged. The convergence condition may be that the loss is not changed any more, or the loss reaches a set threshold or less.

And if the losses of the multiple samples do not meet the convergence condition, adjusting parameters in the wind power prediction model, continuing to carry out iterative training on the wind power prediction model, and executing the step 404.

If the loss of the plurality of samples satisfies the convergence condition, S407 is performed.

And S407, stopping training to obtain a trained wind power prediction model.

It will be appreciated that the computer or server that performs the method of the present embodiment may be the same as or different from the computer or server that performs the embodiments shown in fig. 1 or 3. If the computer or server performing the method of the present embodiment may be different from the computer or server performing the embodiment shown in fig. 1 or 3, the computer or server performing the method of the present embodiment may transmit the obtained trained wind power prediction model to the computer or server performing the embodiment shown in fig. 1 or 3.

It is understood that the method of the present embodiment may be executed alone or before step S104, and the present disclosure is not limited thereto.

In the embodiment, by using the actual second historical data of the historical measurement related to the wind power prediction and the second future data of the future prediction of the known sample wind power plant, the actual wind speed, the actual power and the actual weather data which affect the wind power and can be acquired historically by the sample wind power plant and the prediction data which affect the wind power and can be acquired in the future are trained on the wind power prediction model built by using the encoding and decoding neural network, the model can fully learn the influence of the actual data and the prediction data on the wind power, and the data at a plurality of moments before and after the prediction moment are acquired, so that the model can fully learn the time sequence dependence of the wind power prediction, the wind power prediction precision of the trained wind power prediction model is higher, and the prediction power is more accurate.

Based on the embodiment shown in fig. 4, further, when the second historical data and the second future data of the obtained sample wind farm may be missing or have abnormal (dead pixel) data, etc., the second historical data and/or the second future data of the obtained sample wind farm may be respectively corrected, and a specific embodiment will be described below.

S401 may also be followed by S401A and/or S401B:

S401A, correcting the second historical data to obtain second historical corrected data of the sample wind power plant.

The second history data may be corrected to correct the missing data or abnormal data in the second history data, thereby obtaining corrected second history corrected data.

Optionally, the second historical data is corrected, including but not limited to one or more of the following second historical data correction methods:

the first method for correcting the second historical data is used for correcting abnormal data in the second historical data.

For each type of actual data in the second historical data, abnormal data can be detected by using an abnormal detection algorithm, and then the abnormal data can be corrected by using a data interpolation algorithm. The abnormal data may be data that does not conform to the trend of the change rule of the whole data.

Detecting whether the second historical data has missing data or not, and if the number of the missing data in the continuous data in the preset range is less than a preset threshold value, supplementing the missing data in the second historical data; and if the quantity of the missing data in the continuous preset range is larger than or equal to the preset threshold value, deleting the data in the continuous preset range.

For each type of actual data in the second historical data, if the number of missing data in the actual data in the continuous preset range is less than a preset threshold, the missing data in the actual data can be interpolated through a data interpolation algorithm. If the number of missing data in the continuous preset range is greater than or equal to the preset threshold, it indicates that the segment of actual data is missing more and cannot be interpolated, and deletes the actual data of the type in the preset range.

Accordingly, S402 may include: and converting the second historical corrected data of the sample wind power plant into a second historical two-dimensional array.

S401B, correcting the second future data to obtain second future correction data of the sample wind power plant.

Optionally, the second future data is corrected, including but not limited to one or more of the following second future data correction methods:

and correcting abnormal data in the second future data.

For each type of predicted data in the second future data, abnormal data therein may be detected using an abnormality detection algorithm, and then the abnormal data may be corrected using a data interpolation algorithm. The abnormal data may be data that does not conform to the trend of the change rule of the whole data.

The second future data correction method comprises the steps of detecting whether the second future data have missing data or not, and if the number of the missing data in the continuous preset range is less than a preset threshold value, supplementing the missing data in the second future data; and if the quantity of the missing data in the continuous preset range is larger than or equal to the preset threshold value, deleting the data in the continuous preset range.

For each type of predicted data in the second future data, if the number of missing data in the predicted data in the continuous preset range is less than a preset threshold, the missing data in the predicted data may be interpolated by a data interpolation algorithm. If the number of missing data in the continuous preset range is greater than or equal to a preset threshold, it indicates that the section of prediction data is missing more and cannot be interpolated, and deletes the prediction data of the type in the preset range.

And correcting the predicted data according to the acquired running state data of the wind generation sets of the sample wind power plant, wherein the running state data of the wind generation sets comprises the following steps: pitch angle data, rotational speed data, torque data and unit states of the wind turbine, the unit states including: a shutdown state, a power limit state, and a normal operation state.

Accordingly, S403 may include: and converting the second future correction data of the sample wind power plant into a second future two-dimensional array.

In the embodiment, the second historical correction data of the sample wind power plant is obtained by correcting the second historical data, and the second future data is corrected to obtain the second future correction data of the sample wind power plant, so that the data used for predicting the wind power is more accurate, the wind power prediction precision is higher, and the predicted power is more accurate.

On the basis of the foregoing embodiment, optionally, S402 may include: and normalizing the second historical data of the sample wind power plant, and converting the second historical data into a second historical two-dimensional array.

S403 may include: and normalizing the second future data of the sample wind power plant, and converting the second future data into a second future two-dimensional array.

According to the embodiment, the second historical data and the second future data are normalized, so that the normalized data are obtained, the conversion into the two-dimensional array is facilitated, and the prediction power is obtained by using the wind power prediction model.

Fig. 6 is a schematic structural diagram of an apparatus for predicting wind power provided in an embodiment of the present disclosure, and as shown in fig. 6, the apparatus provided in the present disclosure includes:

the obtaining module 601 is configured to obtain first historical data and first future data of a wind farm to be predicted, where the first historical data includes: actual data corresponding to the N sampling time points in the first historical time period respectively include: actual wind speed, actual power and actual weather data, wherein N is an integer greater than or equal to 1; the first future data includes: prediction data corresponding to the M sampling time points in the first future time period respectively, wherein the prediction data comprises: predicted wind speed and predicted weather data, M being an integer greater than or equal to 1.

The conversion module 602 is configured to convert first historical data of a wind farm to be predicted into a first historical two-dimensional array, where dimensions of the historical two-dimensional array include a first time dimension and a first historical data category dimension.

The conversion module 602 is further configured to: and converting the first future data of the wind power plant to be predicted into a first future two-dimensional array, wherein the dimensionality of the future two-dimensional array comprises a second time dimensionality and a first future data category dimensionality.

The obtaining module 603 is configured to input a first historical two-dimensional array and a first future two-dimensional array of the wind farm to be predicted into the wind power prediction model to obtain the predicted power of the wind farm to be predicted, where the predicted power of the wind farm to be predicted is a predicted value of power corresponding to M sampling time points in a first future time period, the wind power prediction model is a trained coding and decoding neural network model, the coding and decoding neural network includes a coding portion and a decoding portion, the first historical two-dimensional array is input into the coding portion, and the first future two-dimensional array is input into the decoding portion.

Optionally, the apparatus further comprises:

correspondingly, the conversion module is specifically configured to: converting historical correction data of a wind power plant to be predicted into a historical two-dimensional array; and converting the future correction data of the wind power plant to be predicted into a future two-dimensional array.

correcting abnormal data in the first historical data;

detecting whether missing data exists in the first historical data, and if the number of the missing data in the continuous preset range is less than a preset threshold value, supplementing the missing data in the first historical data; if the number of missing data in the continuous preset range is larger than or equal to a preset threshold value, deleting the data in the continuous preset range;

correcting the first future data, including one or more of the following first future data correction methods:

correcting the anomalous data in the first future data;

detecting whether the first future data has missing data, and if the number of the missing data in the continuous preset range is less than a preset threshold value, supplementing the missing data in the first future data; if the number of missing data in the continuous preset range is larger than or equal to a preset threshold value, deleting the data in the continuous preset range;

correcting the prediction data according to the acquired wind turbine generator running state data of the wind power plant to be predicted, wherein the wind turbine generator running state data comprises: pitch angle data, rotational speed data, torque data and unit states of the wind turbine, the unit states including: a shutdown state, a power limit state, and a normal operation state.

Optionally, the conversion module is specifically configured to:

normalizing first historical data of a wind power plant to be predicted, and converting the first historical data into a first historical two-dimensional array; and normalizing the first future data of the wind power plant to be predicted, and converting the first future data into a first future two-dimensional array.

Optionally, the apparatus further comprises: the training module is used for training a wind power prediction model;

a training module comprising:

the acquisition module is further configured to: acquiring a plurality of samples and supervisory data corresponding to the plurality of samples respectively, wherein each sample comprises: second historical data and second future data, the second historical data comprising: actual data corresponding to the N sampling time points in the second historical time period respectively; the second future data includes: respectively corresponding prediction data of M sampling time points in a second future time period; the supervision data of the sample is corresponding actual power of M sampling time points in a second future time period;

the determining module is used for determining the loss of the plurality of samples according to the predicted power of the plurality of samples and the supervision data corresponding to the plurality of samples; if the losses of the multiple samples do not meet the convergence condition, continuing to carry out iterative training on the wind power prediction model; and if the loss of the plurality of samples meets the convergence condition, stopping training to obtain a trained wind power prediction model.

Optionally, the modification module is further configured to:

correspondingly, the conversion module is specifically configured to: converting the second historical corrected data into a second historical two-dimensional array; the second future correction data is converted into a second future two-dimensional array.

The apparatus of the foregoing embodiment may be configured to implement the technical solution of the foregoing method embodiment, and the implementation principle and the technical effect are similar, which are not described herein again.

Fig. 7 is a schematic structural diagram of a device for predicting wind power according to an embodiment of the present disclosure, and as shown in fig. 7, the device according to this embodiment includes:

a memory 701, a memory for storing processor-executable instructions;

a processor 702 for implementing the method as described in fig. 1, fig. 3 or fig. 4 above when the computer program is executed.

The embodiment of the present disclosure provides a computer-readable storage medium, in which computer-executable instructions are stored, and when the computer-executable instructions are executed by a processor, the method for predicting wind power is implemented as shown in fig. 1, fig. 3 or fig. 4.

It is noted that, in this document, relational terms such as "first" and "second," and the like, may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.

The foregoing are merely exemplary embodiments of the present disclosure, which enable those skilled in the art to understand or practice the present disclosure. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the disclosure. Thus, the present disclosure is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting wind power, comprising:

2. The method of claim 1, wherein after obtaining the first historical data and the first future data of the wind farm to be predicted, further comprising:

correspondingly, the converting the first historical data of the wind farm to be predicted into a first historical two-dimensional array includes:

converting the first historical correction data of the wind power plant to be predicted into a first historical two-dimensional array;

the step of converting the first future data of the wind power plant to be predicted into a future two-dimensional array comprises the following steps: and converting the first future correction data of the wind power plant to be predicted into a first future two-dimensional array.

3. The method of claim 2, wherein the modifying the first historical data comprises one or more of:

correcting abnormal data in the first historical data;

correcting anomalous data in the first future data;

4. The method of claim 1, wherein converting the first historical data of the wind farm to be predicted into a first historical two-dimensional array comprises:

5. The method according to any one of claims 1-4, further comprising:

training the wind power prediction model;

the training of the wind power prediction model comprises:

6. The method of claim 5, wherein after obtaining the plurality of samples and the supervisory data corresponding to the plurality of samples, further comprising:

7. An apparatus for predicting wind power, comprising:

the conversion module is used for converting the first historical data of the wind power plant to be predicted into a first historical two-dimensional array, and the dimensionality of the historical two-dimensional array comprises a first time dimensionality and a first historical data category dimensionality; converting first future data of the wind power plant to be predicted into a first future two-dimensional array, wherein the dimensionality of the future two-dimensional array comprises a second time dimensionality and a first future data category dimensionality;

8. The apparatus of claim 7, further comprising:

correspondingly, the conversion module is specifically configured to: converting the first historical correction data of the wind power plant to be predicted into a first historical two-dimensional array; and converting the first future correction data of the wind power plant to be predicted into a first future two-dimensional array.

9. An apparatus for predicting wind power, comprising:

a memory for storing processor-executable instructions;

a processor for implementing the method of any one of claims 1 to 6 when the computer program is executed.

10. A computer-readable storage medium having stored thereon computer-executable instructions for implementing the method of predicting wind power of any one of claims 1 to 6 when executed by a processor.