Wind power prediction method and device based on deep learning

Abstract

The invention discloses a wind power prediction method and device based on deep learning, and belongs to the technical field of artificial intelligence. The method comprises the following steps: acquiring historical data of a target wind power plant related to wind power; the historical data comprises real wind power and a plurality of parameter characteristic data; performing data grouping on the historical data according to a time sequence by using a sliding window; calculating a correlation coefficient between any two parameters based on the multiple pieces of parameter characteristic data; performing feature aggregation on the feature data on the plurality of parameters in each group according to the correlation coefficient so as to perform inter-parameter feature aggregation on the parameter feature data with correlation to obtain an aggregation feature of each group; and training a deep learning model by using the aggregation features of the plurality of groups and the real wind power, and performing wind power prediction by using the trained deep learning model. According to the invention, the deep learning model can accurately learn the aggregation characteristics between the parameters, so that the accuracy of the wind power prediction result can be improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a wind power prediction method and device based on deep learning. Background Art

[0002] With the rapid development of wind power generation, the accuracy of wind power forecasting is crucial for grid dispatching and system stability. Wind power is affected by meteorological conditions and equipment status, and has significant volatility and uncertainty. In related technologies, deep learning models are used for ultra-short-term wind power forecasting. Deep learning models are trained using historical meteorological data and historical operating data. In related technologies, feature learning is performed for each single input parameter during model training, and the accuracy of wind power forecasting results is low. Summary of the invention

[0003] The present invention provides a wind power prediction method and device based on deep learning. The technical solution is as follows: In one aspect, a wind power prediction method based on deep learning is provided, the method comprising: Acquire historical data related to the target wind farm and wind power; the historical data includes real wind power and corresponding multiple parameter characteristic data; the parameters include meteorological parameters and wind turbine operation parameters; The set time length is used as a sliding window, and the historical data is grouped according to the time sequence by using the sliding window to obtain multiple groups; Calculate the correlation coefficient between any two parameters based on the plurality of parameter feature data; Performing feature aggregation on the feature data of multiple parameters in each group according to the correlation coefficient, so that the parameter feature data with correlation is subjected to inter-parameter feature aggregation to obtain the aggregated features of each group; The aggregated features of multiple groups and the actual wind power are used as training samples to train the deep learning model, and the trained deep learning model is used to predict wind power.

[0004] In a possible implementation, the correlation coefficient changes dynamically.

[0005] In a possible implementation, the correlation coefficient is calculated according to a plurality of parameter feature data in the group; The calculating the characteristic correlation coefficient between any two parameters based on the plurality of parameter characteristic data comprises: For any two parameters, execute: Using the characteristic data of the first parameter and the second parameter in the group at each time point, calculate the characteristic means corresponding to the first parameter and the second parameter respectively; The correlation coefficient between the first parameter and the second parameter is calculated according to the Pearson coefficient method using the characteristic data of the first parameter and the second parameter at each time point and the corresponding characteristic means.

[0006] In a possible implementation, according to the characteristic correlation coefficient between the two parameters, if the two parameters are positively correlated, it is determined that the two parameters are correlated; The aggregation features of each group are obtained as follows: The feature data of each parameter in the group are all aggregated with the feature data of the parameters with correlation, and the aggregated data after each parameter is aggregated is obtained; it is determined whether the current aggregation times have reached the set times, if not, the aggregated data after each parameter is aggregated is used as the feature data of the corresponding parameter, and this step is continued to enter the next feature aggregation; if the set times have been reached, the aggregated data after each parameter is aggregated is used as the aggregated feature of the corresponding parameter to obtain the aggregated feature of the group; Each time feature aggregation is performed, the correlation coefficient between two parameters of feature aggregation is used.

[0007] In a possible implementation, the setting number of times is determined as follows: Taking the parameters as nodes, connecting the nodes corresponding to two related parameters to obtain a topological graph; Determine the number of nodes on the longest path in the topology graph; The set number of times is determined according to the number of nodes on the longest path.

[0008] In a possible implementation, the difference of the number of nodes minus 1 is used as the set number of times.

[0009] On the other hand, a wind power prediction device based on deep learning is provided, the device comprising: An acquisition unit, used to acquire historical data related to the target wind farm and wind power; the historical data includes real wind power and corresponding multiple parameter characteristic data; the parameters include meteorological parameters and wind turbine operation parameters; A grouping unit, used to use a set time length as a sliding window, and use the sliding window to group the historical data according to time sequence to obtain multiple groups; A calculation unit, used for calculating a characteristic correlation coefficient between any two parameters based on the plurality of parameter characteristic data; An aggregation unit, used for performing feature aggregation on the feature data of multiple parameters in each group according to the correlation coefficient, so that the parameter feature data with correlation is subjected to inter-parameter feature aggregation to obtain the aggregated features of each group; The prediction unit is used to train the deep learning model by using the aggregated features of multiple groups and the actual wind power as training samples, and to predict the wind power by using the trained deep learning model.

[0010] On the other hand, a computer device is provided, which includes a memory and a processor, the memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory to implement the steps of the above-mentioned deep learning-based wind power prediction method.

[0011] On the other hand, a computer-readable storage medium is provided, wherein a computer program is stored in the storage medium, and when the computer program is executed by a processor, the steps of the above-mentioned wind power prediction method based on deep learning are implemented.

[0012] On the other hand, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the steps of the above-mentioned deep learning-based wind power prediction method.

[0013] The technical solution provided by the present invention can at least bring the following beneficial effects: Since the wind power of a wind farm is associated with the characteristic data of multiple parameters, and there may be correlation between the multiple parameters, directly using multiple parameters for deep learning cannot accurately learn the correlation characteristics between the parameters. Therefore, when obtaining training samples based on historical data, the characteristic correlation coefficient between any two parameters is calculated, and the characteristic data on multiple parameters are aggregated according to the correlation coefficient, so that the characteristic data of parameters with correlation can be aggregated between parameters, thereby obtaining the aggregated features between multiple parameters. The deep learning model is trained using the aggregated features and the actual wind power, so that the deep learning model can accurately learn the aggregated features between parameters, thereby improving the accuracy of the wind power prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0015] Figure 1 This is a flow chart of a wind power prediction method based on deep learning provided by an embodiment of the present invention; Figure 2 This is a structural diagram of a wind power prediction device based on deep learning provided by an embodiment of the present invention; Figure 3 It is a hardware architecture diagram of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0017] Please refer to Figure 1 , an embodiment of the present invention provides a wind power prediction method based on deep learning, the method comprising: Step 100, obtaining historical data related to the target wind farm and wind power; the historical data includes real wind power and corresponding multiple parameter characteristic data; the parameters include meteorological parameters and wind turbine operation parameters; Step 102, using the set time length as a sliding window, and using the sliding window to group the historical data according to time sequence to obtain multiple groups; Step 104, calculating the correlation coefficient between any two parameters based on the plurality of parameter feature data; Step 106, performing feature aggregation on the feature data of multiple parameters in each group according to the correlation coefficient, so that the parameter feature data with correlation is subjected to inter-parameter feature aggregation to obtain the aggregated features of each group; Step 108: Use the aggregated features of the multiple groups and the actual wind power as training samples to train the deep learning model, and use the trained deep learning model to predict wind power.

[0018] In the embodiment of the present invention, since the wind power of a wind farm is associated with characteristic data of multiple parameters, and there may be correlation between the multiple parameters, directly using multiple parameters for deep learning cannot accurately learn the correlation characteristics between the parameters. Therefore, when obtaining training samples based on historical data, the characteristic correlation coefficient between any two parameters is calculated, and the characteristic data on the multiple parameters are feature aggregated according to the correlation coefficient, so that the parameter characteristic data with correlation can be feature aggregated between the parameters, thereby obtaining the aggregated features between the multiple parameters, and using the aggregated features and the actual wind power to train the deep learning model, so that the deep learning model can accurately learn the aggregated features between the parameters, thereby improving the accuracy of the wind power prediction results.

[0019] Described below Figure 1 How the various steps are performed.

[0020] First, step 100 "obtaining historical data related to the target wind farm and wind power" and step 102 "using the set time length as a sliding window and using the sliding window to group the historical data in time sequence to obtain multiple groups" are explained.

[0021] In the embodiment of the present invention, in order to predict the wind power of the target wind farm at a future time, it is necessary to use the historical data related to the wind power of the target wind farm to obtain training samples. The historical data includes the real wind power and the corresponding multiple parameter feature data. The parameters include meteorological parameters and wind turbine operation parameters. The meteorological parameters may include at least one of wind speed, wind direction, temperature, air pressure and humidity; the wind turbine operation parameters may include at least one of speed, torque, wind axial force, current and voltage.

[0022] After obtaining the historical data, in order to train the deep learning model, it is necessary to obtain training samples from the historical data. By setting the time length as a sliding window, the historical data is grouped according to the time sequence using the sliding window to obtain multiple groups. For example, the historical data includes the period from time point t1 to time point t m data, m is a positive integer. If the time length is set to 10 time points, then the time points t1-t 10 The historical data is grouped as a group, and the time points t2-t 11 The historical data of time points t3-t 12 The historical data is taken as a group, ..., and multiple groups can be formed in this way.

[0023] It should be noted that the historical data may be data obtained from multiple wind turbines in the target wind farm.

[0024] Since wind power has time series characteristics, each group includes multi-parameter feature data at multiple time points. Taking each group as a training sample can more accurately learn the time series characteristics, and combine the multi-parameter characteristics at each time point to learn the spatial characteristics, so that the wind power can be predicted together using the time series characteristics and spatial characteristics to improve the accuracy of wind power prediction.

[0025] Then, for step 104, a characteristic correlation coefficient between any two parameters is calculated based on the plurality of parameter characteristic data.

[0026] Considering that multiple parameter feature data are directly used as input for training samples, the deep learning model cannot accurately learn the correlation between parameters. However, there is a correlation between multiple parameters, such as wind speed and rotation speed, temperature and rotation speed, etc. Therefore, it is possible to pre-calculate which parameters are correlated.

[0027] When calculating the characteristic correlation coefficient between any two parameters, in one implementation, all historical data on the two parameters can be directly used to calculate and obtain it as a fixed correlation coefficient.

[0028] In another implementation, since wind power has time series characteristics, and meteorological data and wind turbine operation data also change frequently, the correlation coefficient also changes dynamically. Dynamically adjusting the correlation coefficient between parameters can better fit the actual operation status and improve the accuracy of the prediction results.

[0029] Since the parameter feature data in each group is dynamically changing, when the correlation coefficient is dynamically changing, the correlation coefficient can be calculated according to the multiple parameter feature data in the group. Specifically, in the embodiment of the present invention, the correlation coefficient between any two parameters can be calculated in the following way: For any two parameters, execute: Using the characteristic data of the first parameter and the second parameter in the group at each time point, calculate the characteristic means corresponding to the first parameter and the second parameter respectively; The correlation coefficient between the first parameter and the second parameter is calculated according to the Pearson coefficient method using the characteristic data of the first parameter and the second parameter at each time point and the corresponding characteristic means.

[0030] The calculation formula of Pearson coefficient method is: Among them, r is the correlation coefficient between the first parameter and the second parameter, X i is the characteristic data at the i-th time point in the first parameter, Y i is the characteristic data at the i-th time point in the second parameter, , are the characteristic means corresponding to the first parameter and the second parameter respectively, and n is the total number of time points corresponding to the group.

[0031] After calculating the correlation coefficient between two parameters, if the correlation coefficient is greater than 0 and less than or equal to 1, it indicates that there is a correlation between the two parameters, and the correlation is positive; if the correlation coefficient is equal to 0, it indicates that there is no correlation between the two parameters; if the correlation coefficient is greater than or equal to -1 and less than 0, it indicates that there is a negative correlation between the two parameters.

[0032] It should be noted that, in addition to using the above-mentioned Pearson coefficient method to calculate the correlation coefficient between two parameters, other methods can also be used for calculation, such as the Spearman rank correlation coefficient method, the Kendall rank correlation coefficient method, and the like.

[0033] Next, for step 106, feature aggregation is performed on the feature data of multiple parameters in each group according to the correlation coefficient, so that the parameter feature data with correlation is subjected to inter-parameter feature aggregation to obtain the aggregated features of each group.

[0034] In order to accurately learn the correlation characteristics between different parameters, the correlation coefficient between parameters can be used to aggregate the feature data on multiple parameters in each group, so that the feature data of correlated parameters can be aggregated between parameters, so that the deep learning model can use the aggregated features to accurately learn the correlation between parameters, thereby improving the accuracy of the prediction results.

[0035] In one embodiment of the present invention, when aggregating features, situations where two parameters are positively correlated, uncorrelated, or negatively correlated can all be considered as correlations for feature aggregation, so that the deep learning model can fully learn the different correlation situations between different parameters.

[0036] In another embodiment of the present invention, only positive correlation is considered, and feature aggregation is performed between positively correlated parameters to ensure the validity of the aggregated parameters. That is, based on the correlation coefficient between two parameters, if the two parameters are positively correlated, it is determined that the two parameters are correlated.

[0037] Then, the method for obtaining the multidimensional aggregation features of each group can be: feature aggregation is performed on the feature data of each parameter in the group with the feature data of the parameters with correlation, and the aggregated data after each parameter is aggregated is obtained; it is determined whether the current aggregation times reaches the set times, if not, the aggregated data after each parameter is aggregated is used as the feature data of the corresponding parameter, and this step is continued to enter the next feature aggregation; if the set times are reached, the aggregated data after each parameter is aggregated is used as the aggregation feature of the corresponding parameter, and the multidimensional aggregation features of the group are obtained; Each time feature aggregation is performed, the correlation coefficient between two parameters of feature aggregation is used.

[0038] Each time the feature data of each parameter in the group is aggregated with the feature data of the related parameters, it may include: Generate a first matrix and a second matrix; the first matrix includes M rows and N columns of elements, M is the total number of parameters, N is the total number of time points, and the element value of the i-th row and j-th column is the characteristic data of the j-th time point corresponding to the i-th parameter; the second matrix includes M rows and M columns of elements, and the element value of the i-th row and j-th column is the correlation coefficient between the i-th parameter and the j-th parameter; The first matrix after the current feature aggregation is determined by multiplying the first matrix by the second matrix, wherein the first matrix after the current feature aggregation includes the aggregated data after each parameter of the current feature aggregation.

[0039] If the set number of times is not reached, the first matrix after this feature aggregation is used to continue to execute the method of determining the first matrix after this feature aggregation by multiplying the first matrix and the second matrix to enter the next feature aggregation; if the set number of times is reached, the first matrix after this feature aggregation is used as the multidimensional aggregation feature.

[0040] In one implementation, when determining the first matrix after the feature aggregation using the product of the first matrix and the second matrix, a convolutional network model may be used. The convolutional network model includes a set number of convolutional layers connected in sequence; the set number is equal to the set number of times.

[0041] The processing process of each convolution layer may include: performing feature aggregation on the first matrix and the second matrix input to the current convolution layer to obtain the first matrix after this feature aggregation, and inputting the first matrix after this feature aggregation to the next convolution layer.

[0042] The feature aggregation method of each convolutional layer may be to directly multiply the first matrix and the second matrix, or to achieve feature aggregation by using a weight matrix and a calculation degree matrix.

[0043] Based on this, the process of the convolutional network model finally outputting the first matrix is: inputting the first matrix and the second matrix into the first convolutional layer, the first convolutional layer outputs the first matrix after the first feature aggregation to the second convolutional layer, the second convolutional layer outputs the first matrix after the second feature aggregation for the input first matrix and second matrix to the third convolutional layer, ..., until the last convolutional layer outputs the first matrix after the last feature aggregation.

[0044] In one embodiment, the set number of times may be determined by: taking parameters as nodes, connecting the nodes corresponding to two correlated parameters to obtain a topological graph; determining the number of nodes on the longest path in the topological graph; and determining the set number of times based on the number of nodes on the longest path.

[0045] Preferably, the difference of the number of nodes minus 1 can be used as the set number of times.

[0046] It should be noted that, in addition to the above-mentioned determination method of setting the number of times, other determination methods may also be used, such as directly determining the number of nodes as the set number of times, or setting the number of times to an integer greater than 3 and not greater than the number of nodes.

[0047] In this way, through multiple feature aggregations, each element of the first matrix, that is, the multi-dimensional aggregated feature, which is finally output, can contain feature data on all parameters, and this method takes into account the correlation of feature data in time series, and further improves the accuracy of feature learning. For example, assuming that parameter V1 is correlated with parameter V2, and parameter V2 is correlated with parameter V3; then in the first feature aggregation, the feature of parameter V1 aggregates the feature of parameter V2 to obtain a new V1' feature, parameter V2 aggregates the feature of parameter V3 to obtain a new V2' feature, and parameter V3 also aggregates the feature of parameter V2 to obtain a new V3' feature; in the second feature aggregation, the new V1' feature is aggregated with the new V2' feature to obtain a new V1'' feature. At this time, the V1'' feature has the features of parameter V2 and parameter V3. It can be seen that through multiple feature aggregations, each element can contain feature data on all parameters, and parameters that are not directly related can also obtain feature relationships between parameters through interval-related parameters.

[0048] Finally, for step 108, the aggregated features of the multiple groups and the actual wind power are used as training samples to train the deep learning model, and the trained deep learning model is used to predict the wind power.

[0049] In the embodiment of the present invention, when a group includes n time points, when training the deep learning model, it is also necessary to use the real wind power corresponding to the n time points as the output of the training sample pair. That is, the training sample includes the multi-dimensional aggregation features of a group as input, and the n real wind power corresponding to the group as output.

[0050] It should be noted that when the multi-dimensional aggregation features in step 106 are implemented through a convolutional network model, the convolutional network model and the deep learning model can be jointly trained to obtain a trained convolutional network model and a deep learning model.

[0051] After obtaining the trained convolutional network model and deep learning model, multiple parameter feature data related to the target wind farm and wind power can be obtained in real time, and then the multiple parameter feature data within the set time length are used as the current group, and the convolutional network model is used to output the aggregated features for the multiple parameter feature data and correlation coefficients of the current group, and the aggregated features are input into the deep learning model to obtain the output predicted wind power. The set time length includes n continuous time points, and the n continuous time points include: tn ,t n-1 ,…,t1, t1 is the current time. The deep learning model outputs t n ,t n-1 , …, t1 respectively correspond to the predicted wind power, and the predicted wind power corresponding to t1 is taken as the predicted wind power at the current moment.

[0052] In an embodiment of the present invention, when obtaining training samples based on historical data, the correlation coefficient between any two parameters is calculated, and feature data on multiple parameters are aggregated according to the correlation coefficient, so that parameter feature data with correlation can be aggregated between parameters, thereby obtaining multi-dimensional aggregation features. The deep learning model is trained using the multi-dimensional aggregation features and the actual wind power, so that the deep learning model can accurately learn the aggregation features between parameters, thereby improving the accuracy of the wind power prediction results.

[0053] Please refer to Figure 2 Based on the same concept as the above-mentioned wind power prediction method based on deep learning, an embodiment of the present invention provides a wind power prediction device based on deep learning, the device comprising: The acquisition unit 200 is used to acquire historical data related to the target wind farm and wind power; the historical data includes real wind power and corresponding multiple parameter characteristic data; the parameters include meteorological parameters and wind turbine operation parameters; A grouping unit 202, configured to use a set time length as a sliding window, and use the sliding window to group the historical data according to time sequence to obtain a plurality of groups; A calculation unit 204, configured to calculate a characteristic correlation coefficient between any two parameters based on the plurality of parameter characteristic data; Aggregation unit 206, used for performing feature aggregation on the feature data of multiple parameters in each group according to the correlation coefficient, so as to perform inter-parameter feature aggregation on the parameter feature data with correlation, and obtain the aggregated features of each group; The prediction unit 208 is used to train the deep learning model by using the aggregated features of the multiple groups and the actual wind power as training samples, and to predict the wind power by using the trained deep learning model.

[0054] In one embodiment of the present invention, the correlation coefficient changes dynamically.

[0055] In one embodiment of the present invention, the correlation coefficient is calculated according to a plurality of parameter feature data in the group; The calculation unit is specifically used to execute, for any two parameters: using the characteristic data of the first parameter and the second parameter in the group at each time point, calculating the characteristic means corresponding to the first parameter and the second parameter respectively; using the characteristic data of the first parameter and the second parameter at each time point and the corresponding characteristic means, calculating the correlation coefficient between the first parameter and the second parameter according to the Pearson coefficient method.

[0056] In one embodiment of the present invention, according to the correlation coefficient between the two parameters, if the two parameters are positively correlated, it is determined that the two parameters are correlated; The multi-dimensional aggregation features of each group are obtained as follows: The characteristic data of each parameter in the group are all subjected to characteristic aggregation with the characteristic data of the parameters with correlation, and the aggregated data after each parameter is aggregated is obtained; it is determined whether the current aggregation times have reached the set times, if not, the aggregated data after each parameter is aggregated is used as the characteristic data of the corresponding parameter, and the present step is continued to enter the next characteristic aggregation; if the set times have been reached, the aggregated data after each parameter is aggregated is used as the aggregated feature of the corresponding parameter, and the multi-dimensional aggregated feature of the group is obtained; Each time feature aggregation is performed, the correlation coefficient between two parameters of feature aggregation is used.

[0057] In one embodiment of the present invention, the setting number of times is determined as follows: Taking the parameters as nodes, connecting the nodes corresponding to two related parameters to obtain a topological graph; Determine the number of nodes on the longest path in the topology graph; The set number of times is determined according to the number of nodes on the longest path.

[0058] In one embodiment of the present invention, the difference of the number of nodes minus 1 is used as the set number of times.

[0059] It should be noted that the wind power prediction device based on deep learning provided in the above embodiment is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the wind power prediction device based on deep learning provided in the above embodiment and the wind power prediction method embodiment based on deep learning belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0060] The embodiment of the present application also provides a computer device, please refer to Figure 3The computer device includes a processor and a memory, in which at least one instruction, at least one program, a code set or an instruction set is stored, and the at least one instruction, at least one program, a code set or an instruction set is loaded and executed by the processor to implement the wind power prediction method based on deep learning provided in the above-mentioned method embodiments.

[0061] An embodiment of the present application also provides a computer-readable storage medium, on which is stored at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement the deep learning-based wind power prediction method provided in the above-mentioned method embodiments.

[0062] An embodiment of the present application also provides a computer program product, which includes a computer program. A processor of a computer device reads the computer program from a computer-readable storage medium, and the processor executes the computer program, so that the computer device executes any of the deep learning-based wind power prediction methods described in the above embodiments.

[0063] For the convenience of description, the above system or device is described by dividing it into various modules or units according to its functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0064] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly contributed to the prior art in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present application or certain parts of the embodiments.

[0065] Finally, it should be noted that, in this article, relational terms such as first, second, third and fourth are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0066] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A wind power prediction method based on deep learning, characterized in that: include: Acquire historical data related to the target wind farm and wind power; the historical data includes real wind power and corresponding multiple parameter characteristic data; The parameters include meteorological parameters and fan operation parameters; The set time length is used as a sliding window, and the historical data is grouped according to the time sequence by using the sliding window to obtain multiple groups; Calculate the characteristic correlation coefficient between any two parameters based on the plurality of parameter characteristic data; Performing feature aggregation on the feature data of multiple parameters in each group according to the correlation coefficient, so that the parameter feature data with correlation is subjected to inter-parameter feature aggregation to obtain the aggregated features of each group; The aggregated features of multiple groups and the actual wind power are used as training samples to train the deep learning model, and the trained deep learning model is used to predict wind power.

2. The method according to claim 1, characterized in that The correlation coefficient changes dynamically.

3. The method according to claim 2, characterized in that The correlation coefficient is calculated according to the multiple parameter feature data in the group; The calculating the characteristic correlation coefficient between any two parameters based on the plurality of parameter characteristic data comprises: For any two parameters, execute: Using the characteristic data of the first parameter and the second parameter in the group at each time point, calculate the characteristic means corresponding to the first parameter and the second parameter respectively; The correlation coefficient between the first parameter and the second parameter is calculated according to the Pearson coefficient method using the characteristic data of the first parameter and the second parameter at each time point and the corresponding characteristic means.

4. The method according to claim 1, characterized in that: According to the characteristic correlation coefficient between the two parameters, if the two parameters are positively correlated, it is determined that the two parameters are correlated; The aggregation features of each group are obtained as follows: The feature data of each parameter in the group are all aggregated with the feature data of the parameters with correlation, and the aggregated data after each parameter is aggregated is obtained; it is determined whether the current aggregation times have reached the set times, if not, the aggregated data after each parameter is aggregated is used as the feature data of the corresponding parameter, and this step is continued to enter the next feature aggregation; if the set times have been reached, the aggregated data after each parameter is aggregated is used as the aggregated feature of the corresponding parameter to obtain the aggregated feature of the group; Each time feature aggregation is performed, the correlation coefficient between two parameters of feature aggregation is used.

5. The method according to claim 4, characterized in that The setting number of times is determined as follows: Taking the parameters as nodes, connecting the nodes corresponding to two related parameters to obtain a topological graph; Determine the number of nodes on the longest path in the topology graph; The set number of times is determined according to the number of nodes on the longest path.

6. The method according to claim 5, characterized in that The difference between the number of nodes and 1 is used as the set number of times.

7. A wind power prediction device based on deep learning, characterized in that: The device comprises: An acquisition unit, used to acquire historical data related to the target wind farm and wind power; the historical data includes real wind power and corresponding multiple parameter characteristic data; the parameters include meteorological parameters and wind turbine operation parameters; A grouping unit, used to use a set time length as a sliding window, and use the sliding window to group the historical data according to time sequence to obtain multiple groups; A calculation unit, used for calculating the correlation coefficient between any two parameters based on the plurality of parameter feature data; An aggregation unit, used for performing feature aggregation on the feature data of multiple parameters in each group according to the correlation coefficient, so that the parameter feature data with correlation is subjected to inter-parameter feature aggregation to obtain the aggregated features of each group; The prediction unit is used to train the deep learning model by using the aggregated features of multiple groups and the actual wind power as training samples, and to predict the wind power by using the trained deep learning model.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the steps of any one of the methods described in claims 1-6.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The method comprises a computer program, wherein when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images