Wind power generation power prediction method and device in extreme weather, equipment and medium
By constructing a mapping relationship of multiple power influence factors and dividing the power influence interval, and combining the LSTM model with dynamic adjustment of training parameters, the problem of accuracy in predicting wind power generation under extreme weather conditions was solved, achieving higher accuracy in prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-31
- Publication Date
- 2026-03-31
AI Technical Summary
Existing technologies struggle to accurately predict wind power output under extreme weather conditions, especially during sudden strong winds or abrupt changes in wind direction. Traditional models lack the ability to model the coupling effect between unit operating status and extreme weather, leading to inaccurate or ineffective predictions.
By constructing a mapping relationship of multiple power influencing factors, dividing the power influence interval, and building a wind power prediction model based on a long short-term memory network (LSTM), the training parameters of the neural network are dynamically adjusted, and predictions are made in combination with real-time meteorological and unit parameters.
It significantly improves the accuracy and robustness of wind power generation prediction under extreme weather conditions, enabling more accurate predictions in complex and non-stationary environments.
Smart Images
Figure CN121769859A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wind power generation prediction, and more particularly to methods, apparatus, equipment and media for predicting wind power generation under extreme weather conditions. Background Technology
[0002] With the increasing penetration rate of wind power, sudden changes in wind speed and drastic fluctuations in power caused by extreme weather have become core challenges restricting the accuracy of wind power forecasting and the safety of grid dispatch. Traditional forecasting methods mostly rely on single models and fixed parameters, using only historical power or basic meteorological data, making it difficult to characterize the nonlinear and non-stationary response characteristics of wind turbines under extreme conditions. Especially when encountering sudden strong winds or abrupt changes in wind direction, existing models often lack the ability to model the coupling effect between unit operating status and extreme weather, leading to severely inaccurate predictions or even failure.
[0003] Deep learning methods (especially Long Short-Term Memory networks, LSTM) are widely used in wind power prediction due to their powerful temporal modeling capabilities. However, existing LSTM models typically employ uniform hyperparameter training, making it difficult to balance stability and sensitivity when faced with diverse operating conditions (especially a mix of extreme and normal scenarios). Therefore, how to achieve more accurate wind power prediction in extreme weather is an urgent problem to be solved. Summary of the Invention
[0004] This invention provides a method, apparatus, equipment, and medium for predicting wind power generation under extreme weather conditions, which solves the technical problem of low accuracy in predicting wind power generation under extreme weather conditions in the prior art, and achieves the technical effect of more accurate prediction of wind power generation under extreme weather conditions.
[0005] In a first aspect, the present invention provides a method for predicting wind power generation under extreme weather conditions, including:
[0006] The historical multivariate data of the target wind turbine were obtained after preprocessing, and a mapping relationship of multivariate power influence factors was constructed based on the historical multivariate data of wind turbine. The historical multivariate data of wind turbine includes meteorological data, unit parameters and power data.
[0007] Based on the mapping relationship of multiple power influence factors, the power influence range of each type of power influence factor is divided;
[0008] Based on long short-term memory networks, a wind power prediction model is constructed, and the neural network training parameters corresponding to the wind power prediction model are determined for each power influence interval. The neural network training parameters include learning rate, network structure, batch size, and regularization rate.
[0009] Based on real-time meteorological data and real-time unit parameters, the training parameters of the target neural network are determined, and the wind power generation prediction model is configured according to the training parameters of the target neural network. After the configuration is completed, the wind power generation prediction model is used for real-time wind power generation prediction.
[0010] Furthermore, a mapping relationship of multiple power influencing factors is constructed based on historical wind power multivariate data, including:
[0011] Based on the Pearson correlation coefficient, the correlation between meteorological data and corresponding power data is determined, as well as the correlation between unit parameters and corresponding power data, wherein the meteorological data, power data, and unit parameters are in one-to-one correspondence.
[0012] Based on partial regression coefficients, the influence weights of meteorological data and corresponding unit parameters are determined.
[0013] Based on the partial regression coefficient and Pearson correlation coefficient, the initial mapping relationship data set of historical wind power multivariate data is obtained, which includes several initial mapping relationships.
[0014] After normalizing the initial mapping relationship, construct the covariance matrix;
[0015] The covariance matrix is solved, and target eigenvectors with values greater than the preset eigenvalues are selected, where each eigenvector corresponds one-to-one with the initial mapping relationship.
[0016] The mapping relationship of multivariate power influence factors is obtained based on the eigenvectors, including: ,in, In the first Type of meteorological data, the first Unit parameters and the first Power influence factors over a given time period In the first Type of meteorological data, the first Unit parameters and the first Maximum power fluctuation amplitude over a given time period In the first Type of meteorological data, the first Unit parameters and the first The average power over a given time period In the first Type of meteorological data, the first Unit parameters and the first Average power fluctuation amplitude over a given time period.
[0017] Furthermore, based on the mapping relationship of multiple power influence factors, the power influence range of each type of power influence factor is divided, including:
[0018]
[0019] in, In the first Type of meteorological data, the first Unit parameters and the first Interval evaluation indicators for a given time period , as well as All are weights. As the first standard value, This is the second standard value. This is the third standard value;
[0020] Based on the values of the interval evaluation indicators, the corresponding multivariate power influence factors are divided into their respective power influence intervals.
[0021] Furthermore, the neural network training parameters for the wind power generation prediction model are determined for each power influence interval, including:
[0022] For each power impact range, the wind power prediction model is trained using historical multivariate data of wind power generation corresponding to all power impact factors in the power impact range.
[0023] Once the preset training requirements are met, the neural network training parameters corresponding to the power influence range are obtained.
[0024] Furthermore, based on real-time meteorological data and real-time unit parameters, the training parameters for the target neural network are determined, including:
[0025] Based on real-time meteorological data and real-time unit parameters, the corresponding target power influence range is determined;
[0026] Based on the target power influence range, determine the corresponding target neural network training parameters.
[0027] Furthermore, the loss training function for the wind power generation prediction model includes:
[0028]
[0029] in, For the first The loss in each power-affected region, For the first The number of power influence factors in each power influence interval. For the first In the power influence interval, the first Preset weights for each power influence factor, For the first In the power influence interval, the first The true power of each power influence factor For the first In the power influence interval, the first Predicted power of each power influencing factor.
[0030] Furthermore, the historical multivariate wind power generation data of the target wind turbine are preprocessed, including:
[0031] After acquiring meteorological data, unit parameters, and power data, align the meteorological data, unit parameters, and power data according to the timestamp;
[0032] Outliers in meteorological data, unit parameters, and power data are removed, and missing values in these data are supplemented using interpolation.
[0033] Secondly, the present invention provides a wind power generation prediction device under extreme weather conditions, comprising:
[0034] The output acquisition module is used to acquire the preprocessed historical multivariate data of the target wind turbine, and to construct a multivariate power influence factor mapping relationship based on the historical multivariate data, which includes meteorological data, unit parameters and power data.
[0035] The interval division module is used to divide the power influence interval of each type of power influence factor based on the mapping relationship of multiple power influence factors;
[0036] The neural network training module is used to build a wind power prediction model based on a long short-term memory network, and to determine the neural network training parameters corresponding to the wind power prediction model under each power influence range. The neural network training parameters include learning rate, network structure, batch size, and regularization rate.
[0037] The power prediction module is used to determine the training parameters of the target neural network based on real-time meteorological data and real-time unit parameters, and configure the wind power prediction model according to the training parameters of the target neural network. After the configuration is completed, the wind power prediction model is used for real-time wind power prediction.
[0038] Thirdly, the present invention provides an electronic device, comprising:
[0039] processor;
[0040] Memory used to store processor-executable instructions;
[0041] The processor is configured to execute a wind power generation prediction method under extreme weather conditions, as provided in the first aspect.
[0042] Fourthly, the present invention provides a non-transitory computer-readable storage medium, wherein when the instructions in the non-transitory computer-readable storage medium are executed by a processor of an electronic device, the electronic device is able to execute the wind power generation prediction method under extreme weather conditions as provided in the first aspect.
[0043] One or more technical solutions provided in this invention have at least the following technical effects or advantages:
[0044] This invention significantly improves the accuracy and robustness of prediction models in complex and non-stationary operating environments by constructing a dynamic mapping mechanism that connects data, influencing factors, operating condition ranges, and model parameters.
[0045] This invention integrates meteorological data, unit parameters, and power data, uses Pearson correlation coefficient and partial regression coefficient to quantify the impact of multiple variables on power, and constructs a multivariate power influence factor mapping relationship based on principal component analysis. Then, it divides the power influence interval based on comprehensive evaluation indicators to achieve a refined characterization of different operating states. On this basis, it independently trains and saves the optimal LSTM model parameters for each interval, forming an adaptive modeling paradigm of one parameter per interval. Attached Figure Description
[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0047] Figure 1 A schematic flowchart illustrating the wind power generation prediction method under extreme weather conditions provided by this invention. Detailed Implementation
[0048] This invention provides a method for predicting wind power generation under extreme weather conditions, which solves the technical problem of low accuracy in predicting wind power generation under extreme weather conditions in the prior art.
[0049] The technical solution of this invention is to solve the above-mentioned technical problems, and the overall idea is as follows:
[0050] A method for predicting wind power generation under extreme weather conditions includes: acquiring preprocessed historical multivariate data of the target wind turbine, and constructing a multivariate power influencing factor mapping relationship based on the historical multivariate data, which includes meteorological data, turbine parameters, and power data; dividing the power influence interval of each type of power influencing factor based on the multivariate power influencing factor mapping relationship; constructing a wind power generation prediction model based on a long short-term memory network, and determining the neural network training parameters corresponding to the wind power generation prediction model under each power influence interval, where the neural network training parameters include learning rate, network structure, batch size, and regularization rate; determining the target neural network training parameters based on real-time meteorological data and real-time turbine parameters, configuring the wind power generation prediction model according to the target neural network training parameters, and using the wind power generation prediction model for real-time wind power generation prediction after configuration.
[0051] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.
[0052] First, it should be clarified that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.
[0053] This invention provides, for example Figure 1 The wind power generation prediction method shown includes steps S11-S14:
[0054] Step S11: Obtain the preprocessed historical multivariate data of the target wind turbine, and construct a multivariate power influence factor mapping relationship based on the historical multivariate data, which includes meteorological data, unit parameters and power data.
[0055] Preprocessing the historical multivariate data of the target wind turbine includes: acquiring meteorological data, turbine parameters, and power data, aligning the meteorological data, turbine parameters, and power data according to timestamps; removing outliers in the meteorological data, turbine parameters, and power data, and supplementing missing values in the meteorological data, turbine parameters, and power data based on interpolation.
[0056] Historical wind power multivariate data refers to the data of the target wind turbine over a historical period.
[0057] Specifically:
[0058] Meteorological data (such as wind speed, wind direction, temperature, etc.), generator parameters (such as pitch angle, speed, yaw angle, etc.), and power data (actual generated power) usually come from different acquisition systems, which may result in inconsistent sampling frequencies or time recording deviations. The meteorological data in this invention can be meteorological data under extreme weather conditions.
[0059] By aligning with a unified timestamp, we ensure that the three types of data at each moment correspond to the same physical state, thus avoiding modeling distortion caused by time misalignment.
[0060] Due to sensor malfunctions, communication errors, or extreme interference, the unprocessed data may contain values that are obviously unreasonable or deviate from the normal range (such as negative wind speeds, power generation values that exceed the rated power by several times, etc.).
[0061] These outliers can be identified and removed using statistical methods (such as the 3σ principle or box plots) or domain knowledge to prevent them from misleading model training.
[0062] During data collection, data may be missing at some points in time due to equipment downtime, network interruption, or data being discarded during the process. To maintain the continuity of the time series, interpolation methods (such as linear interpolation, spline interpolation, or imputation based on the mean of neighboring time periods) are used to reasonably estimate and supplement the missing values, ensuring that the subsequent model can be stably trained and predicted.
[0063] A mapping relationship of multiple power influencing factors was constructed based on historical multivariate data of wind power generation, including:
[0064] Based on the Pearson correlation coefficient, the correlation between meteorological data and corresponding power data, as well as the correlation between unit parameters and corresponding power data, are determined, with a one-to-one correspondence between meteorological data, power data, and unit parameters.
[0065] The Pearson correlation coefficient is an indicator that measures the degree of linear correlation between two variables. Preprocessed meteorological data and unit parameters were paired with their corresponding power data.
[0066] Calculate the Pearson correlation coefficient between each type of meteorological data and power data, as well as the Pearson correlation coefficient between each type of unit parameters and power data, and retain only variables with correlations higher than a preset threshold to focus on key influencing factors.
[0067] The influence weights of meteorological data and corresponding unit parameters are determined based on partial regression coefficients.
[0068] The core function of partial regression coefficients is to accurately reflect the independent influence weight of a variable by showing the change in power data for each unit change in a given variable, while keeping all other variables constant.
[0069] Using power data as the dependent variable and selected highly correlated meteorological data and unit parameters as independent variables, a multiple linear regression model was constructed. The partial regression coefficients for each independent variable were obtained by solving the model; the larger the absolute value of the coefficient, the stronger the independent influence of that variable on power.
[0070] Based on the partial regression coefficient and Pearson correlation coefficient, an initial mapping relationship data set of historical wind power multivariate data is obtained, which includes several initial mapping relationships.
[0071] The filtered relevance variables have all completed independent Pearson correlation coefficient calculations and partial regression coefficient solutions, and the two types of coefficients for each variable are precisely bound to the variable through information such as timestamps and variable identifiers;
[0072] A separate coefficient association carrier is created for each relevant variable. This carrier contains fields such as variable name, variable type (meteorological data, unit parameters), coefficient type, and coefficient value.
[0073] Enter the Pearson correlation coefficient and its physical meaning, as well as the corresponding partial regression coefficient and its physical meaning, into the coefficient association carrier of the variable to complete one-to-one field matching.
[0074] Then, coefficient association annotations were added to the carrier to clarify the complementary relationship between the two sets of coefficients and avoid coefficient isolation;
[0075] A uniformity check was performed on the coefficient association carriers of all variables to confirm that the variable attribution of the two sets of coefficients in each carrier is consistent, the values are accurate, and there are no omissions or duplicate pairings.
[0076] After normalizing the initial mapping relationship, the covariance matrix is constructed.
[0077] Because meteorological data and generator parameters have different dimensions, direct calculations can lead to distorted results. Normalization is used to convert all initial mapping values to the 0-1 range, thus unifying the dimensions.
[0078] The covariance matrix is used to measure the degree of co-variance between two initial mapping relationships (positive covariance indicates co-variance, negative covariance indicates inverse covariance, and 0 indicates no correlation).
[0079] Using all the normalized initial mapping relationships as the rows and columns of a matrix, calculate the covariance between each pair to form a covariance matrix.
[0080] The covariance matrix is solved, and target eigenvectors with values greater than the preset eigenvalues are selected. The eigenvectors correspond one-to-one with the initial mapping relationship.
[0081] Its purpose is to reduce dimensionality, remove redundant information from the initial mapping relationship, and retain the key mapping relationship that best reflects the power change.
[0082] The covariance matrix is decomposed into eigenvalues to obtain several eigenvalues and corresponding eigenvectors (the eigenvectors are linear combinations of the initial mapping relationships).
[0083] The physical meaning of eigenvalues: They represent the amount of information contributed by the corresponding eigenvector. The larger the eigenvalue, the more core information the eigenvector reflects.
[0084] A preset feature value threshold is set to filter out target feature vectors whose feature values are greater than the threshold. Each target feature vector corresponds to a set of key mapping relationships, and different target feature vectors are independent of each other.
[0085] It should be noted that: the feature vector is a linear combination of the initial mapping relationship, which is the way the feature vector is generated; the independence between different target feature vectors is the result of this generation method.
[0086] The mapping relationship of multivariate power influence factors is obtained based on the eigenvectors, including: ,in, In the first Type of meteorological data, the first Unit parameters and the first Power influence factors over a given time period In the first Type of meteorological data, the first Unit parameters and the first Maximum power fluctuation amplitude over a given time period In the first Type of meteorological data, the first Unit parameters and the first The average power over a given time period In the first Type of meteorological data, the first Unit parameters and the first Average power fluctuation amplitude over a given time period.
[0087] This invention uses Pearson correlation coefficient to accurately screen core variables strongly correlated with power and partial regression coefficient to quantify the independent influence weights of variables. The initial mapping relationship formed by the combination of the two has both strong correlation and targeted influence. Then, normalization is used to eliminate dimensional differences, and covariance matrix and eigenvalue decomposition are used to achieve dimensionality reduction and redundancy removal while ensuring that the target feature vectors are independent of each other. Finally, a multivariate influencing factor containing the maximum power fluctuation amplitude, average value, and average fluctuation amplitude is generated. This not only comprehensively captures the complex characteristics of the effect of meteorological data and unit parameters on power under extreme weather conditions, but also optimizes and ensures data quality and information validity, providing high-quality and highly adaptable core inputs for accurate prediction of wind power generation under extreme weather conditions.
[0088] Step S12: Based on the mapping relationship of multiple power influence factors, divide the power influence range of each type of power influence factor.
[0089] Specifically, it includes:
[0090]
[0091] in, In the first Type of meteorological data, the first Unit parameters and the first Interval evaluation indicators for a given time period , as well as All are weights. As the first standard value, This is the second standard value. This is the third standard value;
[0092] Based on the values of the interval evaluation indicators, the corresponding multivariate power influence factors are divided into their respective power influence intervals.
[0093] For example, If the value is 1, and the range of the power influence interval is {1-2}, then... The corresponding power influence factor belongs to the range {1-2}.
[0094] Dividing the power influence factor into intervals realizes the mapping from multidimensional statistical characteristics to operable operating condition intervals, effectively connecting the multivariate influence factors constructed above with the LSTM-based adaptive prediction model in the following section.
[0095] This invention eliminates interference from different dimensions and magnitudes by introducing standardization and weighted fusion, while retaining the key characteristics of power fluctuations under extreme weather conditions. Furthermore, dividing the influence intervals allows the model to dynamically select the optimal training parameters for different operating states, significantly improving prediction accuracy and robustness, especially in extreme or non-steady-state scenarios that traditional single models struggle to handle.
[0096] Step S13: Based on the Long Short-Term Memory network, construct a wind power generation prediction model and determine the neural network training parameters corresponding to the wind power generation prediction model under each power influence interval. The neural network training parameters include learning rate, network structure, batch size, and regularization rate.
[0097] The wind power prediction model based on Long Short-Term Memory (LSTM) network utilizes the powerful capabilities of LSTM in processing time series data to model the dynamic nonlinear characteristics of the wind power generation process.
[0098] LSTM, as a special type of recurrent neural network (RNN), can effectively capture dependencies over long periods of time by introducing input gates, forget gates, and output gates, thus avoiding the gradient vanishing or exploding problems that are prone to occur during the training of traditional RNNs.
[0099] Determine the neural network training parameters for the wind power generation prediction model for each power impact range, including:
[0100] For each power impact range, the wind power prediction model is trained using historical multivariate data of wind power generation corresponding to all power impact factors in the power impact range.
[0101] Once the preset training requirements are met, the neural network training parameters corresponding to the power influence range are obtained.
[0102] For each power impact interval defined above, all historical sample data belonging to that interval (i.e., multivariate wind power generation data corresponding to specific meteorological conditions, unit parameter combinations and time periods) are extracted, and this subset is used as the training set to independently train a wind power generation prediction model based on LSTM.
[0103] During training, hyperparameters such as the model's learning rate, number of network layers and neurons (network structure), batch size, and regularization rate (such as L2 weight decay or Dropout ratio) can be automatically tuned or grid searched until the preset training termination conditions are met (such as validation loss convergence, prediction error below a threshold, or reaching the maximum number of training rounds).
[0104] Finally, the set of neural network training parameters that performs best in this range is saved as a configuration scheme specific to this working condition.
[0105] The loss training function for the wind power generation prediction model includes:
[0106]
[0107] in, For the first The loss in each power-affected region, For the first The number of power influence factors in each power influence interval. For the first In the power influence interval, the first Preset weights for each power influence factor, For the first In the power influence interval, the first The true power of each power influence factor For the first In the power influence interval, the first Predicted power of each power influencing factor.
[0108] Step S14: Based on real-time meteorological data and real-time unit parameters, determine the training parameters of the target neural network, and configure the wind power generation prediction model according to the training parameters of the target neural network. After the configuration is completed, the wind power generation prediction model is used for real-time wind power generation prediction.
[0109] Based on real-time meteorological data and real-time unit parameters, the training parameters for the target neural network are determined, including: determining the corresponding target power influence range based on real-time meteorological data and real-time unit parameters; and determining the corresponding target neural network training parameters based on the target power influence range.
[0110] Upon receiving real-time meteorological data (such as wind speed, wind direction, turbulence intensity, etc.) and real-time unit operating parameters (such as pitch angle, generator speed, yaw status, etc.), the corresponding interval evaluation index is first calculated based on the multi-element power influence factor mapping relationship constructed above, thereby determining the target power influence interval to which the current operating state belongs.
[0111] From multiple sets of neural network hyperparameters pre-trained and saved for each power influence interval, retrieve the target neural network training parameters (including learning rate, network structure, batch size, regularization rate, etc.) that match the target interval.
[0112] The basic LSTM prediction model is dynamically loaded or reconfigured according to this set of parameters (such as adjusting the number of network layers, initializing weights, setting optimizer parameters, etc.) to adapt it to the dynamic characteristics of the current working conditions.
[0113] Once the model is configured, it uses the current and recent real-time input sequences as model input, performs forward inference, and outputs a predicted value for future wind power generation.
[0114] In summary, this invention significantly improves the accuracy and robustness of the prediction model in complex and non-stationary operating environments by constructing a dynamic mapping mechanism that connects data, influencing factors, operating condition intervals, and model parameters.
[0115] This invention integrates meteorological data, unit parameters, and power data, uses Pearson correlation coefficient and partial regression coefficient to quantify the impact of multiple variables on power, and constructs a multivariate power influence factor mapping relationship based on principal component analysis. Then, it divides the power influence interval based on comprehensive evaluation indicators to achieve a refined characterization of different operating states. On this basis, it independently trains and saves the optimal LSTM model parameters for each interval, forming an adaptive modeling paradigm of one parameter per interval.
[0116] Based on the same inventive concept, this invention provides a wind power generation prediction device under extreme weather conditions, comprising:
[0117] The output acquisition module is used to acquire the preprocessed historical multivariate data of the target wind turbine, and to construct a multivariate power influence factor mapping relationship based on the historical multivariate data, which includes meteorological data, unit parameters and power data.
[0118] The interval division module is used to divide the power influence interval of each type of power influence factor based on the mapping relationship of multiple power influence factors;
[0119] The neural network training module is used to build a wind power prediction model based on a long short-term memory network, and to determine the neural network training parameters corresponding to the wind power prediction model under each power influence range. The neural network training parameters include learning rate, network structure, batch size, and regularization rate.
[0120] The power prediction module is used to determine the training parameters of the target neural network based on real-time meteorological data and real-time unit parameters, and configure the wind power prediction model according to the training parameters of the target neural network. After the configuration is completed, the wind power prediction model is used for real-time wind power prediction.
[0121] Based on the same inventive concept, the present invention also provides an electronic device, comprising:
[0122] processor;
[0123] Memory used to store processor-executable instructions;
[0124] The processor is configured to execute a wind power generation prediction method under extreme weather conditions, as described above.
[0125] Based on the same inventive concept, the present invention also provides a non-transitory computer-readable storage medium, which, when the instructions in the storage medium are executed by the processor of an electronic device, enables the electronic device to execute the wind power generation prediction method under extreme weather conditions as described above.
[0126] Since the electronic device described in this embodiment is an electronic device used to implement the information processing method in the embodiments of the present invention, those skilled in the art can understand the specific implementation methods and various variations of the electronic device in this embodiment based on the information processing method described in the embodiments of the present invention. Therefore, how the electronic device implements the method in the embodiments of the present invention will not be described in detail here. Any electronic device used by those skilled in the art to implement the information processing method in the embodiments of the present invention falls within the scope of protection of the present invention.
[0127] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0128] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0129] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0130] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1The steps of the function specified in one or more boxes.
[0131] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.
[0132] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.
Claims
1. A method for wind power production forecasting under extreme weather conditions, characterized in that, The method comprises the following steps: acquiring historical wind power multi-element data of a target wind turbine after preprocessing, and constructing a multi-element power influence factor mapping relationship according to the historical wind power multi-element data, wherein the historical wind power multi-element data comprises meteorological data, unit parameters and power data; dividing a power influence interval of each type of power influence factor based on the multi-element power influence factor mapping relationship; constructing a wind power prediction model based on a long short-term memory network, and determining neural network training parameters corresponding to the wind power prediction model in each power influence interval, wherein the neural network training parameters comprise a learning rate, a network structure, a batch and a regularization rate; determining target neural network training parameters according to real-time meteorological data and real-time unit parameters, and configuring the wind power prediction model according to the target neural network training parameters, and after the configuration is completed, the wind power prediction model is used for real-time wind power prediction.
2. The wind power generation power prediction method in extreme weather according to claim 1, characterized by, The method for constructing a multi-element power influence factor mapping relationship according to historical wind power multi-element data comprises the following steps: determining the correlation between meteorological data and corresponding power data and the correlation between unit parameters and corresponding power data based on a Pearson correlation coefficient, wherein the meteorological data, the power data and the unit parameters are in one-to-one correspondence; determining the influence weight of the meteorological data and the corresponding unit parameters based on a partial regression coefficient; obtaining an initial mapping relationship data set of the historical wind power multi-element data based on the partial regression coefficient and the Pearson correlation coefficient, wherein the initial mapping relationship data set comprises a plurality of initial mapping relationships; constructing a covariance matrix after normalizing the initial mapping relationships; solving the covariance matrix to filter target eigenvectors greater than a preset eigenvalue, wherein the eigenvectors are in one-to-one correspondence with the initial mapping relationships; The mapping relationship of multivariate power influence factors is obtained based on the eigenvectors, including: ,in, In the first Type of meteorological data, the first Unit parameters and the first Power influence factors over a given time period In the first Type of meteorological data, the first Unit parameters and the first Maximum power fluctuation amplitude over a given time period In the first Type of meteorological data, the first Unit parameters and the first The average power over a given time period In the first Type of meteorological data, the first Unit parameters and the first Average power fluctuation amplitude over a given time period.
3. The wind power generation power prediction method in the extreme weather according to claim 2, characterized by, dividing a power influence interval of each type of power influence factor based on the multi-element power influence factor mapping relationship comprises the following steps: ； wherein, is a first standard value, is a second standard value, is a third standard value, is a fourth standard value, is a fifth standard value, is a sixth standard value, is a seventh standard value, is an eighth standard value, is a ninth standard value, is a tenth standard value. dividing the corresponding multi-element power influence factor to the power influence interval according to the value of the interval evaluation index.
4. The wind power generation power prediction method in extreme weather according to claim 1, characterized by, determining neural network training parameters corresponding to the wind power prediction model in each power influence interval comprises the following steps: training the wind power prediction model with respect to each power influence interval and all power influence factors in the power influence interval and the historical wind power multi-element data corresponding to the power influence factors; obtaining the neural network training parameters corresponding to the power influence interval when a preset training requirement is met.
5. The wind power generation power prediction method in extreme weather according to claim 1, characterized by, determining target neural network training parameters according to real-time meteorological data and real-time unit parameters comprises the following steps: determining a target power influence interval corresponding to the real-time meteorological data and the real-time unit parameters; determining the target neural network training parameters corresponding to the target power influence interval.
6. The wind power generation power prediction method in extreme weather according to claim 1, characterized by, The loss training function of the wind power prediction model comprises the following steps: ； wherein, is the loss of the jth power impact interval, is the number of power impact factors of the jth power impact interval, is the preset weight of the ith power impact factor in the jth power impact interval, is the real power of the ith power impact factor in the jth power impact interval, is the predicted power of the ith power impact factor in the jth power impact interval, 7. The wind power generation power prediction method in extreme weather according to claim 1, characterized by, preprocessing historical wind power multi-element data of a target wind turbine comprises the following steps: aligning the meteorological data, the unit parameters and the power data according to the time stamp after acquiring the meteorological data, the unit parameters and the power data; The abnormal values in the meteorological data, the unit parameter and the power data are removed, and the missing values in the meteorological data, the unit parameter and the power data are supplemented based on an interpolation method.
8. A wind power production forecasting device under extreme weather conditions, characterized by, The method comprises the steps of: an output acquisition module configured to acquire historical wind power multi-element data of a target wind turbine after preprocessing, and to construct a multi-element power influence factor mapping relationship according to the historical wind power multi-element data, wherein the historical wind power multi-element data comprises meteorological data, unit parameters and power data; an interval division module configured to divide a power influence interval of each type of power influence factor based on the multi-element power influence factor mapping relationship; a neural network training module configured to construct a wind power prediction model based on a long short-term memory network, and to determine neural network training parameters corresponding to the wind power prediction model in each power influence interval, wherein the neural network training parameters comprise a learning rate, a network structure, a batch and a regularization rate; a power prediction module configured to determine target neural network training parameters according to real-time meteorological data and real-time unit parameters, to configure the wind power prediction model according to the target neural network training parameters, and to use the wind power prediction model for real-time wind power prediction after the configuration is completed.
9. An electronic device, comprising: The method comprises the steps of: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to execute to implement the wind power prediction method in extreme weather as claimed in any one of claims 1 to 7.
10. A non-transitory computer-readable storage medium, comprising: When the instructions in the non-transitory computer readable storage medium are executed by the processor of the electronic device, the electronic device can execute the wind power prediction method in extreme weather as claimed in any one of claims 1 to 7.