Method and device for predicting diurnal wind power

Abstract

The application discloses a kind of day wind power prediction method and prediction device, the method comprises: obtaining multiple initial atmospheric prediction data, each initial atmospheric prediction data is obtained from different data source;Multiple initial atmospheric prediction data is executed fusion operation, and obtain after fusion data;After fusion data is executed empirical mode decomposition, and obtain after decomposition data;Obtain composite neural network model;Based on composite neural network model, after decomposition data is analyzed, corresponding day wind power prediction value is generated.Accuracy processing is carried out to the weather data of multiple data sources, the accuracy of weather forecast data is improved, the wind farm data in weather data is decomposed by experience mode, the original weather data based on time sequence is converted into stationary data that can be input into intelligent learning model for processing, the above data is automatically learned by intelligent learning model, and accurate ultra-short-term wind power prediction value is output, meet the actual demand of technical personnel.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of weather data prediction, in particular to a method for predicting intraday wind power, a device for predicting intraday wind power and a computer readable storage medium. BACKGROUND

[0002] With the vigorous development of low-carbon economy, wind energy, as a clean and environmentally friendly renewable energy, is abundant in resources and does not need to be mined and transported, becoming one of the most promising new energy sources, so wind power generation technology is widely used and large-scale promoted.

[0003] With the rapid development of wind power, the installed capacity of wind power has also developed rapidly, and the proportion of wind power in the power grid has increased continuously, but the inherent intermittency and volatility of wind energy have brought a very serious threat to the power quality and safety and stability performance of the power system. Therefore, in order to solve the problem of the impact of large-scale wind power connected to the power grid on the power system, it is necessary to effectively predict wind power.

[0004] There are many technologies for predicting weather in the prior art, such as weather prediction methods based on physical models, weather prediction methods based on statistics, and weather prediction methods based on learning algorithms. However, in actual application, the prediction method based on physical model cannot consider many complex factors that actually affect the weather, so the accuracy is low; the prediction method based on statistics can predict the weather to a certain extent, but the long-term prediction accuracy is still low; the weather prediction method based on learning algorithm, specifically, the weather prediction method based on traditional neural network model has weak processing ability for long time series and problems such as gradient disappearance and overfitting, so the prediction accuracy is low and still cannot meet the actual needs of technical personnel SUMMARY

[0005] In order to overcome the above technical problems existing in the prior art, the embodiments of the present application provide a method for predicting intraday wind power, which accurately processes data obtained from multiple data sources, and analyzes the data using an intelligent analysis model after empirical mode decomposition, thereby realizing accurate wind power prediction and improving prediction accuracy.

[0006] In order to achieve the above purpose, the embodiments of the present application provide a method for predicting intraday wind power, the prediction method comprising: obtaining a plurality of initial atmospheric prediction data, each initial atmospheric prediction data being obtained from different data sources; performing a fusion operation on the plurality of initial atmospheric prediction data to obtain fused data; performing empirical mode decomposition on the fused data to obtain decomposed data; obtaining a composite neural network model; analyzing the decomposed data based on the composite neural network model to generate corresponding intraday wind power prediction values.

[0007] Preferably, the fusion operation is performed on the plurality of initial atmospheric prediction data to obtain fused data, comprising: determining real-time weight information of each initial atmospheric prediction data; performing time sequence fusion processing on the plurality of initial atmospheric prediction data based on the real-time weight information to generate the fused data.

[0008] Preferably, the real-time weight information of each initial atmospheric prediction data is determined, comprising: obtaining real-time forecast quality evaluation information of each initial atmospheric prediction data; processing the real-time forecast quality evaluation information based on a preset neural network to generate the real-time weight information of each initial atmospheric prediction data.

[0009] Preferably, the empirical mode decomposition is performed on the fused data to obtain decomposed data, comprising: obtaining a first random time sequence signal in the fused data; determining a first local extremum of the first random time sequence signal; generating corresponding first envelope information based on the first local extremum; performing empirical mode decomposition on the fused data based on the first envelope information to obtain a plurality of first intrinsic mode functions and first residual component data corresponding to the fused data being a sum of the plurality of first intrinsic mode functions and the first residual component data; and taking the plurality of first intrinsic mode functions and the first residual component data as the decomposed data.

[0010] Preferably, the empirical mode decomposition is performed on the fused data to obtain decomposed data, further comprising: obtaining a second random time sequence signal in the fused data; obtaining a preset average number value and a preset standard white noise; performing noise adding processing on the second random time sequence signal based on the preset average number value and the preset standard white noise to obtain an added signal; determining a second local extremum of the added signal; generating corresponding second envelope information based on the second local extremum; performing empirical mode decomposition on the added signal based on the second envelope information to obtain a plurality of second intrinsic mode functions and second residual component data corresponding to the fused data being a sum of the plurality of second intrinsic mode functions and the second residual component data; and taking the plurality of second intrinsic mode functions and the second residual component data as the decomposed data.

[0011] Preferably, the composite neural network model comprises a convolutional neural network model and a gated recurrent neural network model, and the decomposed data is analyzed based on the composite neural network model to generate corresponding intraday wind power prediction values, comprising: performing first data processing on the decomposed data based on the convolutional neural network model to obtain first processed data; performing second data processing on the first processed data based on the gated recurrent neural network model to generate the intraday wind power prediction values.

[0012] Preferably, the convolutional neural network model comprises a convolutional layer and a pooling layer, the first data processing is performed on the decomposed data based on the convolutional neural network model to obtain first processed data, comprising: performing feature extraction on the decomposed data based on the convolutional layer to obtain corresponding spatial features; and performing dimension reduction processing on the spatial features based on the pooling layer to obtain the first processed data.

[0013] Preferably, the second data processing is performed on the first processed data based on the gated recurrent neural network model to generate an intraday wind power prediction value, comprising: performing data analysis on the first processed data based on the gated recurrent neural network model to obtain time-dependent information between each data in the first processed data; and performing wind power prediction based on the time-dependent information to generate the intraday wind power prediction value.

[0014] Correspondingly, the embodiment of the present application also provides an intraday wind power prediction device, comprising: an initial data acquisition unit, configured to acquire a plurality of initial atmospheric prediction data, each initial atmospheric prediction data being acquired from different data sources; a fusion unit, configured to perform a fusion operation on the plurality of initial atmospheric prediction data to obtain fused data; a decomposition unit, configured to perform empirical mode decomposition on the fused data to obtain decomposed data; a model acquisition unit, configured to acquire a composite neural network model; and a prediction unit, configured to analyze the decomposed data based on the composite neural network model to generate a corresponding intraday wind power prediction value.

[0015] Preferably, the fusion unit comprises: a weight determination module, configured to determine real-time weight information of each initial atmospheric prediction data; and a fusion module, configured to perform time sequence fusion processing on the plurality of initial atmospheric prediction data based on the real-time weight information to generate the fused data.

[0016] Preferably, the weight determination module is specifically configured to: acquire real-time forecast quality evaluation information of each initial atmospheric prediction data; and process the real-time forecast quality evaluation information based on a preset neural network to generate the real-time weight information of each initial atmospheric prediction data.

[0017] Preferably, the decomposition unit comprises a first decomposition module, and the first decomposition module is specifically configured to: acquire a first random time sequence signal in the fused data; determine a first local extreme value of the first random time sequence signal; generate corresponding first envelope information based on the first local extreme value; perform empirical mode decomposition on the fused data based on the first envelope information to obtain a plurality of first intrinsic mode functions and first residual component data, the fused data being a sum of the plurality of first intrinsic mode functions and the first residual component data; and take the plurality of first intrinsic mode functions and the first residual component data as the decomposed data.

[0018] Preferably, the decomposition unit further comprises a second decomposition module, and the second decomposition module is specifically configured to: acquire a second random time sequence signal in the fused data; acquire a preset average number value and a preset standard white noise; perform noise addition processing on the second random time sequence signal based on the preset average number value and the preset standard white noise to obtain an added signal; determine a second local extreme value of the added signal; generate corresponding second envelope information based on the second local extreme value; perform empirical mode decomposition on the added signal based on the second envelope information to obtain a plurality of second intrinsic mode functions and second residual component data, and the fused data is a sum of the plurality of second intrinsic mode functions and the second residual component data; and take the plurality of second intrinsic mode functions and the second residual component data as decomposed data.

[0019] Preferably, the composite neural network model comprises a convolutional neural network model and a gated recurrent neural network model, and the prediction unit comprises: a first processing module configured to perform first data processing on the decomposed data based on the convolutional neural network model to obtain first processed data; and a prediction module configured to perform second data processing on the first processed data based on the gated recurrent neural network model to generate an intraday wind power prediction value.

[0020] Preferably, the convolutional neural network model comprises a convolutional layer and a pooling layer, and the first processing module is specifically configured to: perform feature extraction on the decomposed data based on the convolutional layer to obtain corresponding spatial features; and perform dimension reduction processing on the spatial features based on the pooling layer to obtain the first processed data.

[0021] Preferably, the prediction module is specifically configured to: perform data analysis on the first processed data based on the gated recurrent neural network model to acquire time-dependent information between each data in the first processed data; and perform wind power prediction based on the time-dependent information to generate the intraday wind power prediction value.

[0022] In another aspect, an embodiment of the present application further provides a computer readable storage medium having a computer program stored thereon, and the program is executed by a processor to implement the prediction method provided by the embodiment of the present application.

[0023] Through the technical solutions provided by the present application, the present application has at least the following technical effects:

[0024] The accuracy of weather forecast data is effectively improved by acquiring weather data from multiple data sources and performing accuracy processing on the weather data, and on this basis, the wind farm data in the weather data is subjected to empirical mode decomposition to convert the original time series-based weather data into stationary data that can be input into an intelligent learning model for processing, at this time, the intelligent learning model is used to automatically learn the above data, thereby outputting accurate ultra-short-term wind power prediction values, meeting the actual needs of technical personnel.

[0025] Other features and advantages of the embodiments of the present application will be described in detail in the following detailed description. BRIEF DESCRIPTION OF DRAWINGS

[0026] The accompanying drawings are included to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used together with the following detailed description to explain the embodiments of the present application, but do not constitute a limitation on the embodiments of the present application. In the drawings:

[0027] Figure 1 is a specific implementation flowchart of the prediction method for intra-day wind power provided by the embodiments of the present application;

[0028] Figure 2 is a specific implementation flowchart of obtaining fused data in the prediction method for intra-day wind power provided by the embodiments of the present application;

[0029] Figure 3 is a specific implementation flowchart of obtaining decomposed data in the prediction method for intra-day wind power provided by the embodiments of the present application;

[0030] Figure 4 is a specific implementation flowchart of obtaining decomposed data in the prediction method for intra-day wind power provided by another embodiment of the present application;

[0031] Figure 5 is a schematic diagram of performing empirical mode decomposition on the fused data to obtain decomposed data in the prediction method for intra-day wind power provided by the embodiments of the present application;

[0032] Figure 6 is a structural schematic diagram of the prediction device for intra-day wind power provided by the embodiments of the present application. DETAILED DESCRIPTION

[0033] The specific embodiments of the embodiments of the present application will be described in detail below in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the embodiments of the present application, and are not used to limit the embodiments of the present application.

[0034] The terms "system" and "network" in the embodiments of the present application can be used interchangeably. "Multiple" means two or more, and in view of this, "multiple" in the embodiments of the present application can also be understood as "at least two". "And / or" describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent the following three cases: A exists alone, A and B exist together, and B exists alone. In addition, the character " / ", if not specially stated, generally represents that the associated objects before and after it are in an "or" relationship. In addition, it should be understood that in the description of the embodiments of the present application, "first", "second", etc. are only used for the purpose of distinguishing the description and cannot be understood as indicating or implying relative importance. Also cannot be understood as indicating or implying order.

[0035] Please refer to Figure 1 The embodiments of the present application provide a method for predicting intraday wind power, the prediction method comprising:

[0036] S10) Obtain a plurality of initial atmospheric prediction data, each initial atmospheric prediction data is obtained from different data sources;

[0037] S20) Perform a fusion operation on the plurality of initial atmospheric prediction data to obtain fused data;

[0038] S30) Perform empirical mode decomposition on the fused data to obtain decomposed data;

[0039] S40) Obtain a composite neural network model;

[0040] S50) Based on the composite neural network model, analyze the decomposed data to generate corresponding intraday wind power prediction value.

[0041] In a possible implementation, a plurality of initial atmospheric prediction data are acquired first. In the existing field of weather prediction, there are weather prediction model data of a plurality of data sources, such as the European Centre for Medium-Range Weather Forecasts (EC model), the Shanghai Mesoscale Model (SMS-WARMS, Shanghai model) and the rapid update assimilation model. Correspondingly, EC model data, Shanghai model data and rapid update assimilation model data can be acquired. The data of the plurality of data sources are acquired from different geographical positions or by different acquisition methods, and thus have respective advantages and disadvantages. For example, the EC model and the Shanghai model mainly describe the large-scale average motion process of the atmosphere, and thus the data acquired by the EC model and the Shanghai model have low resolution in time and space. The data of the rapid update assimilation model are more time-effective, but the data accuracy is insufficient. Therefore, in the embodiment of the present application, the data of a plurality of data sources are acquired and comprehensively processed, so as to comprehensively utilize the advantages of the data and eliminate the disadvantages of the data, thereby realizing more accurate atmospheric data processing.

[0042] After the plurality of initial atmospheric prediction data are acquired, a fusion operation is further performed on the plurality of initial atmospheric prediction data. Please refer to Figure 2 In the embodiment of the present application, the fusion operation is performed on the plurality of initial atmospheric prediction data to obtain fused data, and the fusion operation includes:

[0043] S21) determining real-time weight information of each initial atmospheric prediction data;

[0044] S22) performing time sequence fusion processing on the plurality of initial atmospheric prediction data based on the real-time weight information to generate fused data.

[0045] In the embodiment of the present application, the plurality of initial atmospheric prediction data are fused by determining the real-time weight information of each initial atmospheric prediction data. However, in the conventional data fusion method, the weight information of each data to be fused is mainly determined by past experience or the working experience of technical personnel, and thus the actual accuracy requirement cannot be met.

[0046] In the embodiment of the present application, the real-time weight information of each initial atmospheric prediction data is determined by acquiring real-time prediction quality evaluation information of each initial atmospheric prediction data, and processing the real-time prediction quality evaluation information based on a preset neural network to generate the real-time weight information of each initial atmospheric prediction data.

[0047] In a possible implementation, real-time prediction quality evaluation information of each initial atmospheric prediction data is first acquired, for example, the EC mode data, the Shanghai mode data and the rapid update assimilation mode data can be scored respectively according to historical real-time weather prediction data to evaluate the prediction quality of each mode data at different time efficiencies, and then the convolutional neural network is used to analyze the prediction quality to dynamically generate the weight information of each mode data, that is, to acquire the real-time weight information of each initial atmospheric prediction data.

[0048] For example, in the embodiment of the application, the convolutional neural network includes an input layer, a hidden layer and an output layer, wherein the hidden layer includes a convolutional layer, a pooling layer and an activation layer. In the process of analyzing the prediction quality by using the convolutional neural network, the feature information of the prediction quality is first extracted by the convolutional layer, which can effectively reduce the connection between the layers of the network and reduce the risk of overfitting. Then, the activation layer is used to perform nonlinear mapping on the linear calculation result of the convolutional layer to increase the nonlinear segmentation capability and improve the expression capability of the model. At this time, the pooling layer is further used to subsample the data after nonlinear mapping of the activation layer to compress the amount of data and parameters, prevent overfitting and reduce the complexity of the network. For example, mean pooling and max pooling can be used for subsampling. At this time, accurate real-time weight information can be obtained. At this time, the acquired multiple initial atmospheric prediction data are subjected to time sequence fusion processing according to the real-time weight information to generate fused data, for example, the fused data is wind farm data.

[0049] In the embodiment of the application, the real-time weight information of each mode data is intelligently analyzed and determined to realize intelligent fusion of the multiple mode data, and the accuracy of the generated atmospheric prediction data is improved and the prediction time efficiency is improved on the basis of the advantages of compatible data.

[0050] At this time, the data can be analyzed. However, in actual application, since the wind speed and power data have a certain continuity, in addition to obtaining accurate wind farm weather prediction data, the real-time wind speed and power data of the wind farm in the near future are also needed as the input of the model. However, the actual wind speed and wind power curve is time series data with fluctuations, and directly using the time series data as the input is not conducive to the learning and feature extraction of the intelligent model. Therefore, the fused data needs to be subjected to data decomposition processing.

[0051] Please refer to Figure 3 In the embodiment of the application, the experience mode decomposition is performed on the fused data to obtain decomposed data, including:

[0052] S311) Obtain the first random time-series signal from the fused data;

[0053] S312) Determine the local extrema of the first random time-series signal;

[0054] S313) Generate the corresponding first envelope information based on the local extrema;

[0055] S314) Based on the first envelope information, perform empirical mode decomposition on the fused data to obtain a plurality of first intrinsic mode functions and first residual component data, wherein the fused data is the sum of the plurality of first intrinsic mode functions and the first residual component data;

[0056] S315) The plurality of first intrinsic mode functions and the first residual component data are used as decomposed data.

[0057] In one possible implementation, a first random time-series signal x(t) is first determined from the fused data. Then, the first local extrema of the first random time-series signal x(t) are determined, for example, the local maxima and local minima of the first random time-series signal x(t). Then, the upper and lower envelopes of the first random time-series signal x(t) are formed by fitting with cubic spline interpolation as the first envelope information of the first random time-series signal x(t). At this time, empirical mode decomposition is performed on the fused data. Specifically, the mean of the upper and lower envelopes is first set to m(t). Then, h(t) = (t) - m(t) is calculated, and it is determined whether the calculated h(t) satisfies the preset decomposition conditions. If it does not satisfy the conditions, the above calculation is performed again using h(t) as the new x(t) until the obtained h(t) is obtained. k If the preset decomposition condition is met, then the first intrinsic pattern function c1(t) and the residual component r1(t) of the first random time-series signal x(t) are determined, for example, r1(t) = x(t) - c1(t). At this time, the above decomposition steps are continued for the residual component r1(t) until the final residual component r is obtained. n (t) satisfies the preset termination condition. For example, if the above decomposition steps are executed n times, then n intrinsic mode functions c are obtained after decomposing the first random time series signal x(t). i (t) and the final residual component r n In this embodiment of the invention, the first random time-series signal x(t) is one of the aforementioned multiple intrinsic mode functions c. i (t) and residual function r n The sum of (t), for example, can be expressed as: At this point, the aforementioned multiple first intrinsic pattern functions c i (t) and residual function r n(t) as the decomposed data.

[0058] In the embodiment of the present application, by adopting the data decomposition method based on empirical mode decomposition, the non-stationary signal can be effectively decomposed into multiple stationary intrinsic mode functions according to different scale fluctuations or trends, so that the stable decomposition of unstable and dynamic weather forecast data is realized, which facilitates the subsequent data analysis of the intelligent model, and the empirical mode decomposition method is relatively simple, thereby reducing the calculation complexity and improving the calculation accuracy and efficiency.

[0059] However, in the actual application process, the simple empirical mode decomposition method has certain influence on the final decomposition effect due to the mode aliasing linearity and end effect, and therefore, in order to further improve the decomposition accuracy of the weather forecast data, the embodiment of the present application introduces white noise to suppress the aliasing phenomenon, so as to realize more accurate decomposition effect.

[0060] Please refer to Figure 4 In the embodiment of the present application, the performing empirical mode decomposition on the fused data to obtain decomposed data further comprises:

[0061] S321) obtaining a second random time sequence signal in the fused data;

[0062] S322) obtaining a preset average number value and a preset standard white noise;

[0063] S323) performing noise adding processing on the second random time sequence signal based on the preset average number value and the preset standard white noise to obtain an added signal;

[0064] S324) determining a second local extreme value of the added signal;

[0065] S325) generating corresponding second envelope information based on the second local extreme value;

[0066] S326) performing empirical mode decomposition on the noise superimposed data based on the second envelope information to obtain a plurality of second intrinsic mode functions and second residual component data, and the fused data is the sum of the plurality of second intrinsic mode functions and the second residual component data;

[0067] S327) taking the plurality of second intrinsic mode functions and the second residual component data as decomposed data.

[0068] In a possible implementation, first, a second random time sequence signal is determined in the fused data, for example, the second random time sequence signal is also represented as x(t), at this time, a preset average number value M and a preset standard white noise n i(t), and then the preset standard white noise n i (t) is added to the second random timing signal x(t) to generate a new signal (i.e., a post-addition signal), which is represented as: x i (t) = x(t) + n i (t), i = 1, 2, …, M. At this time, the obtained signal containing white noise is subjected to empirical mode decomposition. Based on the same principle, first, the local maximum value and the local limit value of the post-addition signal are determined, then a cubic spline interpolation fitting is performed to form the corresponding upper and lower envelope lines, then the mean values of the upper and lower envelope lines are taken, and step-by-step signal decomposition is performed to obtain multiple intrinsic mode functions of the post-addition signal and a residual component data. At this time, the multiple intrinsic mode functions can be averaged to obtain the final decomposed intrinsic mode function based on the principle that the statistical mean value of an unrelated sequence is 0. Please refer to Figure 5 The schematic diagram of performing empirical mode decomposition on the fused data to obtain decomposed data is provided for the embodiments of the present application.

[0069] In the embodiments of the present application, by using the improved empirical mode decomposition method, the modal aliasing phenomenon and the end effect existing in the traditional empirical mode decomposition method can be effectively suppressed, thereby further improving the accuracy of the final decomposed data, guaranteeing the accuracy of subsequent data analysis and the generated wind power prediction result, and meeting the actual needs of technical personnel.

[0070] At this time, the above decomposed data can be directly input into an intelligent analysis model for wind power prediction. There are various algorithms or models for intelligent analysis of data in the prior art. However, due to the high uncertainty and noise interference of wind farm data, the accuracy is low and modeling is difficult, so the traditional single intelligent model cannot meet the actual needs. In the embodiments of the present application, a composite neural network model is used for analysis of wind farm data.

[0071] In the embodiments of the present application, the composite neural network model includes a convolutional neural network model and a gated recurrent neural network model. The decomposed data is analyzed based on the composite neural network model to generate corresponding intraday wind power prediction values, including: performing first data processing on the decomposed data based on the convolutional neural network model to obtain first processed data; performing second data processing on the first processed data based on the gated recurrent neural network model to generate intraday wind power prediction values.

[0072] In a possible implementation, the first data processing is performed on the decomposed data based on a convolutional neural network to obtain first processed data. For example, in the embodiment of the present application, the convolutional neural network model comprises a convolutional layer and a pooling layer, and the first data processing is performed on the decomposed data based on the convolutional neural network model to obtain the first processed data, which comprises: performing feature extraction on the decomposed data based on the convolutional layer to obtain corresponding spatial features; and performing dimension reduction processing on the spatial features based on the pooling layer to obtain the first processed data.

[0073] In the embodiment of the present application, the convolutional layer in the convolutional neural network can automatically extract data features, and the high-dimensional data can be processed quickly and efficiently by sharing the convolution kernel; and the pooling layer can reduce the dimension of the data, which can effectively solve the technical problem of overfitting, thereby realizing the technical effect of quickly and accurately extracting the spatial features in the decomposed data, and the accuracy of the convolutional neural network model can be effectively improved through continuous iterative training.

[0074] Further, in the embodiment of the present application, the second data processing is performed on the first processed data based on the gated recurrent neural network model to generate the intraday wind power prediction value, which comprises: performing data analysis on the first processed data based on the gated recurrent neural network model to obtain time-dependent information between each data in the first processed data; and performing wind power prediction based on the time-dependent information to generate the intraday wind power prediction value.

[0075] For example, after obtaining the first processed data, the first processed data is further analyzed by the gated recurrent neural network model to further mine the time sequence features in the first processed data, for example, to determine the time-dependent information between each data in the first processed data, and to predict the wind power according to the time-dependent information to obtain an accurate intraday wind power prediction value.

[0076] In the embodiment of the present application, the wind farm data with time sequence features is analyzed by using the gated recurrent neural network model, which can effectively offset the gradient in the prediction process, thereby realizing higher information processing capability, meeting the prediction demand of technical personnel for ultra-short-term (intraday) wind power information, effectively ensuring the accuracy of the processing result, and improving the prediction accuracy of the wind power value.

[0077] The intraday wind power prediction device provided by the embodiment of the present application will be described below with reference to the accompanying drawings.

[0078] Please refer to Figure 6Based on the same inventive concept, the embodiment of the present application provides a device for predicting the wind power in a day, which comprises: an initial data acquisition unit, configured to acquire a plurality of initial atmospheric prediction data, each of which is acquired from different data sources; a fusion unit, configured to perform a fusion operation on the plurality of initial atmospheric prediction data to obtain fused data; a decomposition unit, configured to perform empirical mode decomposition on the fused data to obtain decomposed data; a model acquisition unit, configured to acquire a composite neural network model; and a prediction unit, configured to analyze the decomposed data based on the composite neural network model to generate corresponding predicted wind power in a day.

[0079] In the embodiment of the present application, the fusion unit comprises: a weight determination module, configured to determine real-time weight information of each initial atmospheric prediction data; and a fusion module, configured to perform time sequence fusion processing on the plurality of initial atmospheric prediction data based on the real-time weight information to generate fused data.

[0080] In the embodiment of the present application, the weight determination module is specifically configured to: acquire real-time forecast quality evaluation information of each initial atmospheric prediction data; and process the real-time forecast quality evaluation information based on a preset neural network to generate real-time weight information of each initial atmospheric prediction data.

[0081] In the embodiment of the present application, the decomposition unit comprises a first decomposition module, which is specifically configured to: acquire a first random time sequence signal in the fused data; determine a first local extreme value of the first random time sequence signal; generate corresponding first envelope information based on the first local extreme value; perform empirical mode decomposition on the fused data based on the first envelope information to obtain a plurality of first intrinsic mode functions and first residual component data, wherein the fused data is the sum of the plurality of first intrinsic mode functions and the first residual component data; and take the plurality of first intrinsic mode functions and the first residual component data as decomposed data.

[0082] In the embodiment of the present application, the decomposition unit further comprises a second decomposition module, which is specifically configured to: obtain a second random time sequence signal in the fused data; obtain a preset average number value and a preset standard white noise; perform noise addition processing on the second random time sequence signal based on the preset average number value and the preset standard white noise to obtain an added signal; determine a second local extreme value of the added signal; generate corresponding second envelope information based on the second local extreme value; perform empirical mode decomposition on the added signal based on the second envelope information to obtain a plurality of second intrinsic mode functions and second residual component data, the fused data being a sum of the plurality of second intrinsic mode functions and the second residual component data; and take the plurality of second intrinsic mode functions and the second residual component data as the decomposed data.

[0083] In the embodiment of the present application, the composite neural network model comprises a convolutional neural network model and a gated recurrent neural network model, and the prediction unit comprises: a first processing module configured to perform first data processing on the decomposed data based on the convolutional neural network model to obtain first processed data; and a prediction module configured to perform second data processing on the first processed data based on the gated recurrent neural network model to generate an intraday wind power prediction value.

[0084] In the embodiment of the present application, the convolutional neural network model comprises a convolutional layer and a pooling layer, and the first processing module is specifically configured to: perform feature extraction on the decomposed data based on the convolutional layer to obtain corresponding spatial features; and perform dimension reduction processing on the spatial features based on the pooling layer to obtain the first processed data.

[0085] In the embodiment of the present application, the prediction module is specifically configured to: perform data analysis on the first processed data based on the gated recurrent neural network model to obtain time dependence information between each data in the first processed data; perform wind power prediction based on the time dependence information to generate the intraday wind power prediction value.

[0086] Further, the embodiment of the present application further provides a computer readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the method of the embodiment of the present application.

[0087] The above describes optional embodiments of the embodiment of the present application in detail in combination with the drawings, but the embodiment of the present application is not limited to the specific details in the above embodiments, and various simple modifications can be made to the technical solution of the embodiment of the present application within the technical concept range of the embodiment of the present application, and these simple modifications all belong to the protection range of the embodiment of the present application.

[0088] In addition, it should be noted that various technical features described in the above embodiments can be combined in any suitable manner, as long as there is no contradiction. In order to avoid unnecessary repetition, various possible combinations are not described again in the embodiments of the present application.

[0089] Those skilled in the art can understand that all or part of the steps of the methods in the above embodiments can be completed by a program instructing related hardware. The program is stored in a storage medium, and includes a plurality of instructions for enabling a single-chip microcomputer, a chip or a processor to perform all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media capable of storing program codes.

[0090] In addition, various different embodiments of the embodiments of the present application can also be combined in any suitable manner, as long as they do not deviate from the idea of the embodiments of the present application, and they should also be considered as disclosed in the embodiments of the present application.

Claims

1. A method for predicting intraday wind power, characterized in that, The prediction method includes: Multiple initial atmospheric prediction data are acquired, each from a different data source; Obtain real-time forecast quality assessment information for each initial atmospheric forecast data; The real-time forecast quality assessment information is processed based on a preset neural network to generate real-time weight information for each initial atmospheric prediction data. Based on the real-time weight information, time-series fusion processing is performed on multiple initial atmospheric prediction data to generate fused data; Empirical mode decomposition is performed on the fused data to obtain decomposed data; wherein, The step of performing empirical mode decomposition on the fused data to obtain decomposed data includes: A first random time-series signal is obtained from the fused data; Determine the first local extremum of the first random time-series signal; Generate the corresponding first envelope information based on the first local extremum; Based on the first envelope information, empirical mode decomposition is performed on the fused data to obtain a plurality of corresponding first intrinsic mode functions and first residual component data. The fused data is the sum of the plurality of first intrinsic mode functions and the first residual component data. The plurality of first intrinsic mode functions and the first residual component data are used as the decomposed data; Obtain a composite neural network model; The decomposed data is analyzed based on the composite neural network model to generate corresponding intraday wind power prediction values; wherein... The composite neural network model includes a convolutional neural network model and a gated recurrent neural network model.

2. The prediction method according to claim 1, characterized in that, The step of performing empirical mode decomposition on the fused data to obtain decomposed data further includes: A second random time-series signal is obtained from the fused data; Obtain the preset average frequency value and the preset standard white noise; Based on the preset average frequency value and the preset standard white noise, noise addition processing is performed on the second random time sequence signal to obtain the added signal; Determine the second local extremum of the signal after the addition; The second envelope information is generated based on the second local extremum; Based on the second envelope information, empirical mode decomposition is performed on the added signal to obtain a plurality of corresponding second intrinsic mode functions and second residual component data. The fused data is the sum of the plurality of second intrinsic mode functions and the second residual component data. The plurality of second intrinsic mode functions and the second residual component data are used as the decomposed data.

3. The prediction method according to claim 1, characterized in that, The step of analyzing the decomposed data based on the composite neural network model to generate corresponding intraday wind power prediction values ​​includes: Based on the convolutional neural network model, the decomposed data is subjected to first data processing to obtain first processed data; Based on the gated recurrent neural network model, the first processed data is processed again to generate intraday wind power prediction values.

4. The prediction method according to claim 3, characterized in that, The convolutional neural network model includes convolutional layers and pooling layers. The first data processing, performed on the decomposed data based on the convolutional neural network model to obtain first processed data, includes: Based on the convolutional layer, feature extraction is performed on the decomposed data to obtain the corresponding spatial features; The spatial features are reduced in dimensionality based on the pooling layer to obtain the first processed data.

5. The prediction method according to claim 3, characterized in that, The step of performing a second data processing on the first processed data based on the gated recurrent neural network model to generate intraday wind power prediction values ​​includes: Based on the gated recurrent neural network model, data analysis is performed on the first processed data to obtain the temporal dependency information between each data point in the first processed data. Wind power prediction is performed based on the time-dependent information to generate the intraday wind power prediction value.

6. A device for predicting intraday wind power, characterized in that, The prediction device includes: The initial data acquisition unit is used to acquire multiple initial atmospheric prediction data, each of which is acquired from a different data source. The fusion unit is used to perform a fusion operation on the multiple initial atmospheric prediction data to obtain fused data; wherein, The fusion unit includes: The weight determination module is used to determine the real-time weight information for each initial atmospheric prediction data. The fusion module is used to perform time-series fusion processing on multiple initial atmospheric prediction data based on the real-time weight information to generate fused data; The weight determination module is specifically used for: Obtain real-time forecast quality assessment information for each initial atmospheric forecast data; The real-time forecast quality assessment information is processed based on a preset neural network to generate real-time weight information for each initial atmospheric forecast data. A decomposition unit is used to perform empirical mode decomposition on the fused data to obtain decomposed data; The decomposition unit includes a first decomposition module, which is specifically used for: A first random time-series signal is obtained from the fused data; Determine the first local extremum of the first random time-series signal; Generate the corresponding first envelope information based on the first local extremum; Based on the first envelope information, empirical mode decomposition is performed on the fused data to obtain a plurality of corresponding first intrinsic mode functions and first residual component data. The fused data is the sum of the plurality of first intrinsic mode functions and the first residual component data. The plurality of first intrinsic mode functions and the first residual component data are used as the decomposed data; The model acquisition unit is used to acquire composite neural network models. The prediction unit is used to analyze the decomposed data based on the composite neural network model to generate corresponding intraday wind power prediction values; wherein, The composite neural network model includes a convolutional neural network model and a gated recurrent neural network model.

7. The prediction device according to claim 6, characterized in that, The decomposition unit further includes a second decomposition module, which is specifically used for: A second random time-series signal is obtained from the fused data; Obtain the preset average frequency value and the preset standard white noise; Based on the preset average frequency value and the preset standard white noise, noise addition processing is performed on the second random time sequence signal to obtain the added signal; Determine the second local extremum of the signal after the addition; The second envelope information is generated based on the second local extremum; Based on the second envelope information, empirical mode decomposition is performed on the added signal to obtain a plurality of corresponding second intrinsic mode functions and second residual component data. The fused data is the sum of the plurality of second intrinsic mode functions and the second residual component data. The plurality of second intrinsic mode functions and the second residual component data are used as the decomposed data.

8. The prediction device according to claim 6, characterized in that, The prediction unit includes: The first processing module is used to perform first data processing on the decomposed data based on the convolutional neural network model to obtain first processed data. The prediction module is used to perform a second data processing on the first processed data based on the gated recurrent neural network model to generate intraday wind power prediction values.

9. The prediction device according to claim 8, characterized in that, The convolutional neural network model includes convolutional layers and pooling layers, and the first processing module is specifically used for: Based on the convolutional layer, feature extraction is performed on the decomposed data to obtain the corresponding spatial features; The spatial features are reduced in dimensionality based on the pooling layer to obtain the first processed data.

10. The prediction device according to claim 8, characterized in that, The prediction module is specifically used for: Based on the gated recurrent neural network model, data analysis is performed on the first processed data to obtain the temporal dependency information between each data point in the first processed data. Wind power prediction is performed based on the time-dependent information to generate the intraday wind power prediction value.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the prediction method as described in any one of claims 1-5.

Images