Training method, device and equipment of wind power ultra-short-term power prediction model

Abstract

The application relates to a wind power ultra-short-term power prediction model training method, device and equipment. The method comprises the following steps: obtaining the first derivatives of historical wind speed curves, obtaining the shape similarity degrees between the historical wind speed curves according to the first derivatives, clustering the historical wind speed curves according to the shape similarity degrees, obtaining weather state clusters corresponding to the historical wind speed curves, taking the weather state clusters corresponding to the historical wind speed curves as weather state clusters corresponding to meteorological feature sequences of the historical wind speed curves, and finally performing model training on a wind power ultra-short-term power prediction model to be trained based on the meteorological feature sequences corresponding to the weather state clusters, so as to obtain optimal wind power ultra-short-term power prediction models for performing power prediction under different weather states. The method improves the prediction accuracy of the optimal prediction models corresponding to different weather states.

Description

Technical Field

[0001] This application relates to the field of wind power ultra-short-term power prediction technology, and in particular to a training method, apparatus, computer equipment, computer-readable storage medium and computer program product for a wind power ultra-short-term power prediction model. Background Technology

[0002] Wind power forecasting technology, as a key means of mitigating the impact of wind power grid connection, predicts the output power of wind farms over a future period, providing a basis for power grid dispatching departments to formulate power generation plans and allocate reserve capacity. Based on different forecasting time scales, wind power forecasting can be divided into ultra-short-term forecasting, short-term forecasting, medium-term forecasting, and long-term forecasting. Among these, ultra-short-term and short-term forecasting are directly related to the real-time dispatching and security control of the power system, and represent the current focus and challenge of technological research.

[0003] However, the traditional DTW (Dynamic Time Warping) model still suffers from insufficient prediction accuracy in wind power forecasting applications. Summary of the Invention

[0004] Therefore, it is necessary to provide a training method, apparatus, computer equipment, computer-readable storage medium, and computer program product for wind power ultra-short-term power prediction models that can improve the prediction accuracy of the aforementioned technical problems.

[0005] Firstly, this application provides a training method for a wind power ultra-short-term power prediction model, including:

[0006] Obtain the first derivative of the historical wind speed curve;

[0007] The shape similarity between each historical wind speed curve is obtained based on the first derivative of each curve. The historical wind speed curves are then clustered according to the shape similarity to obtain the weather state clusters corresponding to each historical wind speed curve.

[0008] The weather conditions corresponding to each historical wind speed curve are clustered as the weather condition clusters corresponding to the meteorological feature sequences to which each historical wind speed curve is located.

[0009] Based on the meteorological feature sequences corresponding to the clusters of each weather condition, the wind power ultra-short-term power prediction model to be trained is trained separately to obtain the optimal wind power ultra-short-term power prediction model for power prediction under different weather conditions.

[0010] In one embodiment, historical wind speed curves are clustered according to their shape similarity to obtain weather state clusters corresponding to each historical wind speed curve, including:

[0011] From the historical wind speed curves, determine the central historical wind speed curves of different weather conditions in the current clustering round;

[0012] Based on the shape similarity between the other historical wind speed curves (excluding the historical wind speed curves at the center) and the historical wind speed curves at the center, the other historical wind speed curves are clustered.

[0013] Based on the historical wind speed curves of different weather conditions, update the central historical wind speed curves of different weather condition clusters, and return to execute the steps of determining the central historical wind speed curves of different weather conditions in the current clustering round from each historical wind speed curve, and clustering other historical wind speed curves based on the shape similarity between them and the central historical wind speed curves, until the central historical wind speed curves of different weather conditions are no longer updated, thus obtaining the weather condition clusters corresponding to each historical wind speed curve.

[0014] In one embodiment, updating the central historical wind speed curves of different weather condition clusters according to their historical wind speed curves includes:

[0015] For any current weather state cluster, obtain the shape similarity between any historical wind speed curve in the current weather state cluster and the other historical wind speed curves;

[0016] The sum of the similarity of each shape is obtained by summing the similarity of the shapes of each historical wind speed curve;

[0017] The historical wind speed curve with the highest sum of shape similarity is used as the central historical wind speed curve for the current weather state cluster.

[0018] In an exemplary embodiment, the meteorological feature sequence is constructed through the following steps:

[0019] Obtain the current meteorological feature sequence, the original meteorological features to be screened, and the historical wind power ultra-short-term power corresponding to the current screening round;

[0020] Obtain the average redundancy between the original meteorological features to be screened and the meteorological features contained in the current meteorological feature sequence;

[0021] Based on the mutual information between the original meteorological features to be screened and the historical wind power ultra-short-term power, as well as the average redundancy, target meteorological features are selected from the original meteorological features to be screened, and the current meteorological feature sequence is updated using the target meteorological features. Then, the process returns to the step of obtaining the current meteorological feature sequence corresponding to the current screening round and the original meteorological features to be screened, until the number of meteorological features contained in the current meteorological feature sequence corresponding to the current screening round reaches the preset target number.

[0022] In one embodiment, when the current screening round is the first screening round, obtaining the current meteorological feature sequence corresponding to the current screening round includes:

[0023] The original meteorological features with the highest mutual information are added to the original meteorological feature sequence to obtain the current meteorological feature sequence corresponding to the current screening round.

[0024] In one embodiment, when the current screening round is not the first screening round, the current meteorological feature sequence corresponding to the current screening round is obtained, including:

[0025] Obtain the updated current meteorological feature sequence corresponding to the previous screening round; the previous screening round is the screening round before the current screening round;

[0026] The updated current meteorological feature sequence will be used as the current meteorological feature sequence corresponding to the current screening round.

[0027] In an exemplary embodiment, based on the meteorological feature sequences corresponding to each weather state cluster, the wind power ultra-short-term power prediction model to be trained is trained separately to obtain the optimal wind power ultra-short-term power prediction model for power prediction under different weather conditions, including:

[0028] The meteorological feature sequences corresponding to each weather state cluster are input into the wind power ultra-short-term power prediction model to be trained. The LSTM unit in the wind power ultra-short-term power prediction model obtains the hidden state sequence corresponding to each meteorological feature sequence based on the meteorological feature sequence corresponding to each weather state.

[0029] The hidden state sequences corresponding to each meteorological feature sequence are input into the weighting unit in the wind power ultra-short-term power prediction model. Based on the prediction weights corresponding to each hidden state sequence, the hidden state sequences are weighted and summed to obtain the context vector corresponding to each meteorological feature sequence.

[0030] The context vectors corresponding to each meteorological feature sequence are input into the fully connected unit in the wind power ultra-short-term power prediction model. The context vectors corresponding to each meteorological feature sequence are mapped to obtain the predicted wind power ultra-short-term power corresponding to each meteorological feature sequence.

[0031] Based on the historical and predicted wind power ultra-short-term power corresponding to the meteorological feature sequences of each weather state cluster, the model parameters of the wind power ultra-short-term power prediction model to be trained are updated to obtain the optimal wind power ultra-short-term power prediction model for power prediction under different weather conditions.

[0032] Secondly, this application also provides a training device for a wind power ultra-short-term power prediction model, comprising:

[0033] The acquisition module is used to obtain the first derivative of historical wind speed curves;

[0034] The clustering module is used to obtain the shape similarity between historical wind speed curves based on the first derivatives, and to cluster the historical wind speed curves according to the shape similarity to obtain the weather state clusters corresponding to each historical wind speed curve.

[0035] The determination module is used to cluster the weather states corresponding to each historical wind speed curve, and use this cluster as the weather state cluster corresponding to the meteorological feature sequence to which each historical wind speed curve is located.

[0036] The training module is used to train the wind power ultra-short-term power prediction model to be trained based on the meteorological feature sequences corresponding to each weather state cluster, so as to obtain the optimal wind power ultra-short-term power prediction model for power prediction under different weather conditions.

[0037] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0038] Obtain the first derivative of the historical wind speed curve;

[0039] The shape similarity between each historical wind speed curve is obtained based on the first derivative of each curve. The historical wind speed curves are then clustered according to the shape similarity to obtain the weather state clusters corresponding to each historical wind speed curve.

[0040] The weather conditions corresponding to each historical wind speed curve are clustered as the weather condition clusters corresponding to the meteorological feature sequences to which each historical wind speed curve is located.

[0041] Based on the meteorological feature sequences corresponding to the clusters of each weather condition, the wind power ultra-short-term power prediction model to be trained is trained separately to obtain the optimal wind power ultra-short-term power prediction model for power prediction under different weather conditions.

[0042] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0043] Obtain the first derivative of the historical wind speed curve;

[0044] The shape similarity between each historical wind speed curve is obtained based on the first derivative of each curve. The historical wind speed curves are then clustered according to the shape similarity to obtain the weather state clusters corresponding to each historical wind speed curve.

[0045] The weather conditions corresponding to each historical wind speed curve are clustered as the weather condition clusters corresponding to the meteorological feature sequences to which each historical wind speed curve is located.

[0046] Based on the meteorological feature sequences corresponding to the clusters of each weather condition, the wind power ultra-short-term power prediction model to be trained is trained separately to obtain the optimal wind power ultra-short-term power prediction model for power prediction under different weather conditions.

[0047] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0048] Obtain the first derivative of the historical wind speed curve;

[0049] The shape similarity between each historical wind speed curve is obtained based on the first derivative of each curve. The historical wind speed curves are then clustered according to the shape similarity to obtain the weather state clusters corresponding to each historical wind speed curve.

[0050] The weather conditions corresponding to each historical wind speed curve are clustered as the weather condition clusters corresponding to the meteorological feature sequences to which each historical wind speed curve is located.

[0051] Based on the meteorological feature sequences corresponding to the clusters of each weather condition, the wind power ultra-short-term power prediction model to be trained is trained separately to obtain the optimal wind power ultra-short-term power prediction model for power prediction under different weather conditions.

[0052] The training method, apparatus, computer equipment, computer-readable storage medium, and computer program product for the aforementioned wind power ultra-short-term power prediction model obtains the first derivative of historical wind speed curves. Based on these first derivatives, the shape similarity between historical wind speed curves is determined. These curves are then clustered according to their shape similarity to obtain the corresponding weather state clusters. These weather state clusters are then used as the weather state clusters corresponding to the meteorological feature sequences of each historical wind speed curve. Finally, based on the meteorological feature sequences corresponding to each weather state cluster, the wind power ultra-short-term power prediction model is trained separately to obtain the optimal wind power ultra-short-term power prediction model for power prediction under different weather conditions. By characterizing the sequence shape features using the first derivative, singularities that may occur during clustering are reduced, thereby improving the alignment quality and clustering accuracy of similar wind speed curves, and ultimately improving the prediction accuracy of the trained optimal prediction model for different weather conditions. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is an application environment diagram of the training method for the wind power ultra-short-term power prediction model in one embodiment;

[0055] Figure 2 This is a flowchart illustrating the training method for a wind power ultra-short-term power prediction model in one embodiment.

[0056] Figure 3 This is a flowchart illustrating the training method for a wind power ultra-short-term power prediction model in another embodiment.

[0057] Figure 4 This is an architecture diagram of a wind power ultra-short-term power prediction model in another embodiment;

[0058] Figure 5 This is a structural block diagram of a training device for a wind power ultra-short-term power prediction model in one embodiment.

[0059] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0061] The training method for the wind power ultra-short-term power prediction model provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, the wind power generation equipment communicates with server 102 via a network. A data storage system can store the data that server 102 needs to process. The data storage system can be integrated onto server 102 or placed in the cloud or on other network servers. Server 102 obtains the first derivative of historical wind speed curves, determines the shape similarity between each historical wind speed curve based on the first derivative, clusters each historical wind speed curve according to the shape similarity, obtains the weather state clusters corresponding to each historical wind speed curve, and then uses these weather state clusters as the weather state clusters corresponding to the meteorological feature sequences of each historical wind speed curve. Finally, based on the meteorological feature sequences corresponding to each weather state cluster, the wind power ultra-short-term power prediction model to be trained is trained to obtain the optimal wind power ultra-short-term power prediction model for power prediction under different weather conditions. Server 102 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0062] In one exemplary embodiment, such as Figure 2 As shown, a training method for a wind power ultra-short-term power prediction model is provided, and this method is applied to... Figure 1 Taking server 102 as an example, the explanation includes the following steps S201 to S204. Wherein:

[0063] Step S201: Obtain the first derivative of the historical wind speed curve.

[0064] The historical wind speed curve can be understood as a wind speed-time variation curve, and the first derivative can be understood as the instantaneous rate of change of the curve at the corresponding time point.

[0065] For example, server 102 obtains the original meteorological feature sequence, uses mRMR to filter the original meteorological features in the original meteorological feature sequence to obtain the meteorological feature sequence, extracts the historical wind speed curve in the meteorological feature sequence, and calculates the first derivative of the historical wind speed curve.

[0066] Step S202: Obtain the shape similarity between each historical wind speed curve based on the first derivative of each curve, and cluster each historical wind speed curve according to the shape similarity to obtain the weather state cluster corresponding to each historical wind speed curve.

[0067] Among them, the degree of shape similarity can be understood as the quantitative similarity information of the curve trend of historical wind speed curves.

[0068] Optionally, server 102 calculates the shape similarity between historical wind speed curves based on their first derivatives, and clusters the historical wind speed curves according to their shape similarity. That is, it determines the central historical wind speed curve of the current clustering round for different weather conditions from the historical wind speed curves, and clusters the other historical wind speed curves according to their shape similarity with the central historical wind speed curves. It updates the central historical wind speed curves of different weather conditions according to the historical wind speed curves of different weather conditions, and returns to execute the steps of determining the central historical wind speed curve of the current clustering round for different weather conditions from the historical wind speed curves, and clustering the other historical wind speed curves according to their shape similarity with the central historical wind speed curves. This process continues until the central historical wind speed curves of different weather conditions are no longer updated, thus obtaining the weather condition clusters corresponding to each historical wind speed curve.

[0069] Step S203: Cluster the weather states corresponding to each historical wind speed curve as the weather state cluster corresponding to the meteorological feature sequence to which each historical wind speed curve is located.

[0070] Among them, the meteorological feature sequence can be understood as a set of meteorological features that are highly correlated with the ultra-short-term power of wind power.

[0071] For example, server 102 determines the meteorological feature sequence to which each historical wind speed curve belongs based on the group identifier to which each historical wind speed curve belongs, and clusters the weather states corresponding to each historical wind speed curve as the weather state clusters corresponding to the meteorological feature sequences to which each historical wind speed curve belongs.

[0072] Step S204: Based on the meteorological feature sequences corresponding to each weather state cluster, train the wind power ultra-short-term power prediction model to be trained to obtain the optimal wind power ultra-short-term power prediction model for power prediction under different weather conditions.

[0073] Optionally, server 102 performs targeted model training on the wind power ultra-short-term power prediction model to be trained based on the meteorological feature sequences corresponding to each weather state cluster, so as to obtain the optimal wind power ultra-short-term power prediction model for power prediction under different weather conditions.

[0074] In the training method of the aforementioned wind power ultra-short-term power prediction model, the first derivative of historical wind speed curves is obtained. Based on these first derivatives, the shape similarity between historical wind speed curves is determined. These curves are then clustered according to their shape similarity to obtain the corresponding weather state clusters. These weather state clusters are then used as the weather state clusters corresponding to the meteorological feature sequences of each historical wind speed curve. Finally, based on the meteorological feature sequences corresponding to each weather state cluster, the wind power ultra-short-term power prediction model is trained separately, resulting in the optimal wind power ultra-short-term power prediction model for power prediction under different weather conditions. By using the first derivative to characterize the sequence shape features, singularities that may occur during clustering are reduced, thereby improving the alignment quality and clustering accuracy of similar wind speed curves, and ultimately improving the prediction accuracy of the trained optimal prediction model for different weather conditions.

[0075] In one embodiment, clustering historical wind speed curves according to their shape similarity to obtain weather state clusters corresponding to each historical wind speed curve includes: determining the central historical wind speed curve of different weather state clusters in the current clustering round from each historical wind speed curve; clustering other historical wind speed curves (excluding the central historical wind speed curves) according to their shape similarity with the central historical wind speed curves; updating the central historical wind speed curves of different weather state clusters according to their historical wind speed curves, and returning to execute the steps of determining the central historical wind speed curve of different weather state clusters in the current clustering round from each historical wind speed curve, and clustering other historical wind speed curves (excluding the central historical wind speed curves) according to their shape similarity with the central historical wind speed curves, until the central historical wind speed curves of different weather state clusters are no longer updated, thus obtaining the weather state clusters corresponding to each historical wind speed curve.

[0076] Among them, the central historical wind speed curve can be understood as the curve that best represents the characteristics of the current weather state cluster.

[0077] Optionally, server 102 determines the central historical wind speed curves of different weather state clusters in the current clustering round from each historical wind speed curve. Based on the shape similarity between other historical wind speed curves (excluding the central historical wind speed curves) and the central historical wind speed curves, it clusters the other historical wind speed curves. According to the historical wind speed curves of different weather state clusters, it updates the central historical wind speed curves of different weather state clusters and returns to execute the steps of determining the central historical wind speed curves of different weather state clusters in the current clustering round from each historical wind speed curve, and clustering the other historical wind speed curves based on the shape similarity between other historical wind speed curves (excluding the central historical wind speed curves) and the central historical wind speed curves, until the central historical wind speed curves of different weather state clusters are no longer updated, thus obtaining the weather state clusters corresponding to each historical wind speed curve.

[0078] According to the above implementation method, by designing iterative updates for clustering until the historical wind speed curves of the center of each cluster no longer need to be updated, the accuracy and reliability of the final clustering process are guaranteed.

[0079] In one embodiment, updating the central historical wind speed curves of different weather state clusters according to the historical wind speed curves of different weather state clusters includes: for any current weather state cluster, obtaining the shape similarity between any historical wind speed curve of the current weather state cluster and the other historical wind speed curves; summing the shape similarity of each historical wind speed curve to obtain the sum of the shape similarity of each historical wind speed curve; and taking the historical wind speed curve with the largest sum of shape similarity as the central historical wind speed curve of the current weather state cluster.

[0080] For example, server 102 obtains the shape similarity between any historical wind speed curve in the current weather state cluster and the other historical wind speed curves for any current weather state cluster, sums the shape similarity of each historical wind speed curve to obtain the sum of the shape similarity of each historical wind speed curve, and takes the historical wind speed curve with the largest sum of shape similarity, that is, the historical wind speed curve with the largest similarity to the other historical wind speed curves in the current weather state cluster, as the new central historical wind speed curve of the current weather state cluster.

[0081] Based on the aforementioned implementation method, the historical wind speed curve with the highest sum of shape similarity to the other historical wind speed curves in the cluster is updated as the center historical wind speed curve of the cluster, thus ensuring the accuracy and reliability of the clustering.

[0082] In an exemplary embodiment, the meteorological feature sequence is constructed through the following steps: obtaining the current meteorological feature sequence corresponding to the current screening round, the original meteorological features to be screened, and the historical wind power ultra-short-term power; obtaining the average redundancy between the original meteorological features to be screened and the meteorological features contained in the current meteorological feature sequence; based on the mutual information between the original meteorological features to be screened and the historical wind power ultra-short-term power, and each average redundancy, selecting target meteorological features from the original meteorological features to be screened, updating the current meteorological feature sequence using the target meteorological features, and returning to execute the steps of obtaining the current meteorological feature sequence corresponding to the current screening round and the original meteorological features to be screened, until the number of meteorological features contained in the current meteorological feature sequence corresponding to the current screening round reaches the preset target number.

[0083] Optionally, server 102 obtains the current meteorological feature sequence, the original meteorological feature sequence to be screened, and the historical wind power ultra-short-term power corresponding to the current screening round. It obtains the average redundancy between the original meteorological features to be screened and the meteorological features contained in the current meteorological feature sequence. Based on the mutual information between the original meteorological features to be screened and the historical wind power ultra-short-term power and each average redundancy, it obtains the mRMR score corresponding to the original meteorological features to be screened. According to the mRMR score, it determines the target meteorological feature from the original meteorological features to be screened, updates the current meteorological feature sequence using the target meteorological feature, and repeats the above steps until the number of meteorological features contained in the current meteorological feature sequence corresponding to the current screening round reaches the preset target number, and ends the feature screening process.

[0084] According to the above implementation method, mRMR is used to achieve feature screening of the original meteorological features, thereby ensuring a high correlation between the final meteorological feature sequence and the ultra-short-term power of wind power, and ensuring low redundancy among meteorological features in the meteorological feature sequence.

[0085] In one embodiment, when the current screening round is the first screening round, obtaining the current meteorological feature sequence corresponding to the current screening round includes: adding the original meteorological feature to be screened with the largest mutual information to the original meteorological feature sequence to obtain the current meteorological feature sequence corresponding to the current screening round.

[0086] For example, when the current screening round is the first screening round, server 102 adds the original meteorological feature with the highest mutual information to the empty original meteorological feature sequence to obtain the current meteorological feature sequence corresponding to the current screening round. By using the original meteorological feature with the highest mutual information as the first selected meteorological feature in the meteorological feature sequence, a high correlation between the final obtained meteorological feature sequence and the ultra-short-term power of wind power is guaranteed.

[0087] In one embodiment, when the current screening round is not the first screening round, obtaining the current meteorological feature sequence corresponding to the current screening round includes: obtaining the updated current meteorological feature sequence corresponding to the previous screening round; the previous screening round is the previous screening round of the current screening round; and using the updated current meteorological feature sequence as the current meteorological feature sequence corresponding to the current screening round.

[0088] Optionally, if the current screening round is not the first screening round, server 102 obtains the updated current meteorological feature sequence corresponding to the previous screening round and uses the updated current meteorological feature sequence corresponding to the previous screening round as the current meteorological feature sequence corresponding to the current screening round. By using the updated current meteorological feature sequence corresponding to the previous screening round as the current meteorological feature sequence corresponding to the current screening round, the high relevance of the feature screening iteration process is ensured, thereby improving the accuracy of the selected meteorological feature sequences.

[0089] In one embodiment, based on the meteorological feature sequences corresponding to each weather state cluster, the wind power ultra-short-term power prediction model to be trained is trained separately to obtain the optimal wind power ultra-short-term power prediction model for power prediction under different weather conditions. This includes: inputting the meteorological feature sequences corresponding to each weather state cluster into the wind power ultra-short-term power prediction model to be trained; obtaining the hidden state sequences corresponding to each meteorological feature sequence through the LSTM unit in the wind power ultra-short-term power prediction model based on the meteorological feature sequences corresponding to each weather state; inputting the hidden state sequences corresponding to each meteorological feature sequence into the weighting unit in the wind power ultra-short-term power prediction model; and obtaining the hidden state sequences according to each hidden state. The prediction weights corresponding to the sequences are used to sum the hidden state sequences in a weighted manner to obtain the context vectors corresponding to each meteorological feature sequence. The context vectors corresponding to each meteorological feature sequence are then input into the fully connected unit in the wind power ultra-short-term power prediction model to map the context vectors corresponding to each meteorological feature sequence, thereby obtaining the predicted wind power ultra-short-term power corresponding to each meteorological feature sequence. Based on the historical wind power ultra-short-term power and predicted wind power ultra-short-term power corresponding to the meteorological feature sequences corresponding to each weather state cluster, the model parameters of the wind power ultra-short-term power prediction model to be trained are updated respectively, thereby obtaining the optimal wind power ultra-short-term power prediction model for power prediction under different weather conditions.

[0090] Among them, LSTM (Long Short-Term Memory) units can be understood as units that perform hidden state analysis.

[0091] For example, server 102 inputs the meteorological feature sequences corresponding to each weather state cluster into the wind power ultra-short-term power prediction model to be trained. The forward LSTM unit in the LSTM unit of the wind power ultra-short-term power prediction model obtains the first hidden state sequence corresponding to each meteorological feature sequence based on the meteorological feature sequences corresponding to each weather state. The backward LSTM unit in the LSTM unit obtains the second hidden state sequence corresponding to each meteorological feature sequence based on the meteorological feature sequences corresponding to each weather state. Based on the first and second hidden state sequences, the hidden state sequence corresponding to each meteorological feature sequence is obtained, and the hidden state sequence corresponding to each meteorological feature sequence is input into the wind power ultra-short-term power prediction model. The weighted unit in the model performs a weighted summation of each hidden state sequence based on the prediction weights corresponding to each hidden state sequence, obtaining the context vector corresponding to each meteorological feature sequence. The context vector corresponding to each meteorological feature sequence is then input into the fully connected unit in the wind power ultra-short-term power prediction model to map the context vector corresponding to each meteorological feature sequence, obtaining the predicted wind power ultra-short-term power corresponding to each meteorological feature sequence. Based on the historical wind power ultra-short-term power and predicted wind power ultra-short-term power corresponding to the meteorological feature sequences corresponding to each weather state cluster, the model parameters of the wind power ultra-short-term power prediction model to be trained are updated respectively, resulting in the optimal wind power ultra-short-term power prediction model for power prediction under different weather conditions.

[0092] Based on the aforementioned implementation method, the architecture of the ultra-short-term power prediction model is used to provide a more complete context for the attention layer through forward and backward time-series information. The attention mechanism assigns higher weights to power-related information to enhance the extraction of key features, thereby improving prediction accuracy and reducing the impact of irrelevant disturbances.

[0093] In one exemplary embodiment, such as Figure 3 As shown, a specific implementation of a training method for a wind power ultra-short-term power prediction model is provided. Based on DDTW-k-medoids, historical wind speed curves are clustered to identify consistent weather conditions. A Bi-LSTM-Attention prediction model corresponding to the weather condition is selected to train and optimize samples under different weather conditions, resulting in the optimal power prediction model for each weather condition to predict the ultra-short-term power of wind power generation. Specifically:

[0094] The specific processing flow for each module is as follows:

[0095] 1.1mRMR Feature Filtering:

[0096] Input various meteorological data Feature selection is performed using mRMR to maximize the correlation between meteorological features and power. The correlation between meteorological features is analyzed, while redundancy between meteorological features is minimized, and a subset of features with high information content and complementarity is selected. .

[0097] This method, based on mutual information theory, is implemented through a two-stage optimization: first, the feature most strongly correlated with the prediction target is selected; then, new features with the lowest redundancy among the selected features are gradually added. Compared to traditional methods, mRMR avoids redundant contributions between features while preserving key information, thus improving the model's interpretability and generalization ability.

[0098] Maximum correlation through features and target variable The average mutual information between them is expressed as:

[0099]

[0100] In the formula, Representation of features and target variable The mutual information between them is expressed as follows:

[0101]

[0102] In the formula, , , for , the marginal and joint probability density functions of y.

[0103] Minimum redundancy requires each feature The minimum dependency between them can be represented by the following formula:

[0104]

[0105] Then mRMR can be expressed as:

[0106]

[0107] 1.2 Clustering of historical wind speed curves based on DDTW-k-medoids:

[0108] Enter historical wind speed The DDTW-k-medoids clustering algorithm was used to analyze historical wind speeds. Time correlation analysis was performed to divide historical wind speeds into multiple subsets with higher similarity, and different weather conditions were classified based on the clustering results.

[0109] k-medoids clustering is a method based on "representative objects" that performs well in attenuating the influence of outliers. Each time series within a cluster is homogeneous to each other and heterogeneous to time series outside the cluster. The algorithm operates similarly to k-means, but instead of updating the centroids by taking the mean of the sequences within the cluster, it uses the median sequence. This means that the centroids are existing sequences in the dataset, not generated sequences.

[0110] The historical wind speed curves are clustered using the k-medoids clustering algorithm. The calculation formula is as follows:

[0111]

[0112] In the formula: The sum of squared errors of the clustering results; For each cluster The samples in; These are the cluster centers for each cluster.

[0113] The calculation steps of the k-medoids algorithm are as follows:

[0114] Step 1: Randomly select from the dataset Each sample is used as a cluster center.

[0115] Step 2: Calculate the distance between the remaining sample points in the dataset and each currently selected cluster center in turn, and classify each point into the category with the closest distance.

[0116] Step 3: Update the centroid of each cluster according to the principle of reducing squared deviation. That is, traverse the data points of each cluster and select the point in the cluster with the smallest sum of distances to all other points as the new centroid.

[0117] Step 4: Iterate through steps 2 and 3 until the cluster centers no longer change, and determine the final clustering results for the C categories.

[0118] The choice of the number of clusters in k-medoids directly affects the clustering results. According to the elbow principle, as the number of clusters increases... The increase in value, Gradually decrease. When After the value exceeds the equilibrium point, The rate of decrease slows significantly, forming a bend point similar to the elbow, i.e., the elbow point. Further increasing the number of clusters thereafter has a significant effect on reducing... The contribution becomes negligible. To maintain high similarity within clusters while preserving inter-cluster differences to the greatest extent possible, the elbow point is... The value is used as the optimal number of clusters.

[0119] In k-medoids time series clustering, Euclidean distance is typically used, where each point in the first time series sequentially corresponds to a corresponding point in the second time series. However, in reality, wind speeds in different spaces may exhibit a certain degree of time delay, resulting in various time series with similar overall trends but slight offsets along the time axis. The point-based approach of Euclidean distance is insufficient. To address the problem of excessively large distances due to distortion along the time dimension, the DDTW algorithm is chosen for distance calculation.

[0120] While the DTW algorithm mitigates time distortion by rearranging (distorting) sequences to optimize their mutual matching, it may introduce unnatural distortions or warps during the rearrangement process, known as "singularities." To avoid singularities, the DDTW algorithm is employed. The DDTW algorithm considers a higher-level feature, namely the "shape" of the time series data. The first derivative estimate of DDTW is as follows:

[0121] D t [ x ] = ( x i + x i − 1 ) + ( ( x i + 1 + x i − 1 ) / 2 ) 2

[0122] The DDTW algorithm obtains information about the "shape" of time series data by calculating the first derivative of the data. In this case, the elements in the "distance matrix" represent the squared difference of the first derivatives of the time series data at two corresponding points, rather than the distance between the two points.

[0123] The cost function for evaluating the quality of clustering results based on the DDTW-k-medoids method is: Its definition is as follows:

[0124]

[0125] 1.3 Bi-LSTM-Attention model prediction power:

[0126] For the model architecture of the Bi-LSTM-Attention model, please refer to... Figure 4 Based on the different weather conditions obtained from the clustering results, the samples are classified. The Bi-LSTM-Attention model is used to train and optimize the parameters of the samples under different weather conditions to obtain the optimal photovoltaic power generation prediction model under different weather conditions, and output the wind power ultra-short-term prediction power.

[0127] Bi-LSTM neural networks consist of two LSTM neural network units, one for forward and one for backward propagation. LSTM is a special structure of recurrent neural networks, designed to overcome the limitations of general recurrent neural networks in that they cannot remember long-term information and require high memory and long computation time. Each LSTM unit includes an input gate, an output gate, and a forget gate, which respectively control the filtering of input, output, and past information.

[0128] Input data exist Entering the LSTM cell at any time, forget gate The calculation formula is:

[0129]

[0130] in, and It means Hidden state of time and the forget gate The weights; and They are respectively in the Gate of Oblivion The weights and the Sigmoid activation function, The bias in the forget gate determines the amount of information to be forgotten from the previous LSTM cell, i.e., the information from the previous time step. Hidden state and input data The amount of information entering a cell is determined by the input gate:

[0131]

[0132] in, and Indicating the input gate and The weights; This is due to the deviation of the input gate. Used to control temporary information of LSTM cells The forget gate and input gate determine the current state of the LSTM cell. The two variable expressions are shown below:

[0133]

[0134]

[0135] in, and For the corresponding weights; For the corresponding deviation; This represents the current cell state value. This represents the cell state value from the previous moment. This is the activation function.

[0136] Finally, the output gate controls the current LSTM cell state. The output and the hidden state of the output at the current moment. The expression is as follows:

[0137] h t = σ ( W o [ h t − 1 , X t ] + b o ) ⋅ T a n h ( C t )

[0138] in, The weights of the output gates; This represents the deviation of the output gate; This is the output value of the current cell.

[0139] Bi-LSTM combines the information features of both forward and backward LSTMs. The calculation formula for the input time is as follows:

[0140] H t = [ h ← , h → ]

[0141] in, This represents the output of the Bi-LSTM at time t. and These represent the forward and backward LSTMs, respectively. While Bi-LSTM overcomes the limitation of unidirectional LSTM in fully utilizing historical and future information of a given input for time series data and possesses stronger feature extraction capabilities, improving prediction accuracy by better extracting the internal correlations of other data related to power data remains paramount. Therefore, combining existing artificial intelligence techniques, we consider adding an attention mechanism to the Bi-LSTM network. The core idea of ​​the attention mechanism is to allocate more attention to important information, i.e., assign it larger weights, thereby improving the quality of hidden layer feature extraction.

[0142]

[0143]

[0144]

[0145] In the formula, This is the output vector of the attention mechanism. These are the weighting coefficients. This is the original hidden layer state. for The weight, and This is the weight matrix. This is a deviation.

[0146] Bi-LSTM has a special gate structure and memory function, and has good temporal data processing capabilities. The encoded input sequence is trained twice and connected to the same attention layer to provide each point of the input sequence of the attention layer with complete past and future contextual information, which further improves the network's ability to recognize and related information. At the same time, the attention layer can focus on power-related information from this information. Finally, it is connected to the output layer to output the result.

[0147] Compared with existing public disclosures, this application has the following technical advantages:

[0148] 1. The historical wind speed curves are clustered according to DDTW-k-medoids to obtain different weather state classifications. For each weather state, the Bi-LSTM-Attention prediction model is trained and optimized to obtain the optimal prediction model for the corresponding weather state, which is used to output the ultra-short-term wind power prediction.

[0149] 2. DDTW is introduced as a distance metric to address the issue of Euclidean distance failure caused by time delays in wind speed sequences. The first derivative is used to characterize the sequence shape features, reducing the singularity that DDTW may exhibit during matching, thereby improving the alignment quality and clustering accuracy of similar wind speed curves. Bi-LSTM utilizes forward and backward temporal information to provide a more complete context for the attention layer. The attention mechanism assigns higher weights to power-related information to strengthen key feature extraction, thereby improving prediction accuracy and reducing the impact of irrelevant perturbations.

[0150] 3. Modeling different weather conditions separately can overcome the limitation that a single prediction model cannot adapt to all weather conditions, enhance the adaptability to different weather conditions, and improve the prediction stability and robustness under variable weather processes.

[0151] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0152] Based on the same inventive concept, this application also provides a training device for a wind power ultra-short-term power prediction model, which is used to implement the training method for the wind power ultra-short-term power prediction model described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more wind power ultra-short-term power prediction model training device embodiments provided below can be found in the limitations of the wind power ultra-short-term power prediction model training method described above, and will not be repeated here.

[0153] In one exemplary embodiment, such as Figure 5 As shown, a training device for a wind power ultra-short-term power prediction model is provided, comprising: an acquisition module 501, a clustering module 502, a determination module 503, and a training module 504, wherein:

[0154] Module 501 is used to obtain the first derivative of historical wind speed curves;

[0155] Clustering module 502 is used to obtain the shape similarity between each historical wind speed curve based on each first derivative, and to cluster each historical wind speed curve according to each shape similarity to obtain the weather state cluster corresponding to each historical wind speed curve.

[0156] The determination module 503 is used to cluster the weather states corresponding to each historical wind speed curve as the weather state cluster corresponding to the meteorological feature sequence to which each historical wind speed curve is located.

[0157] Training module 504 is used to train the wind power ultra-short-term power prediction model to be trained based on the meteorological feature sequences corresponding to each weather state cluster, so as to obtain the optimal wind power ultra-short-term power prediction model for power prediction under different weather conditions.

[0158] In one embodiment, the clustering module 502 is further configured to: determine the central historical wind speed curves of different weather state clusters in the current clustering round from each historical wind speed curve; cluster other historical wind speed curves based on the shape similarity between them and the central historical wind speed curves; update the central historical wind speed curves of different weather state clusters according to the historical wind speed curves of different weather state clusters; and return to execute the steps of determining the central historical wind speed curves of different weather state clusters in the current clustering round from each historical wind speed curve, and clustering other historical wind speed curves based on the shape similarity between them and the central historical wind speed curves, until the central historical wind speed curves of different weather state clusters are no longer updated, thus obtaining the weather state clusters corresponding to each historical wind speed curve.

[0159] In one embodiment, the clustering module 502 is further configured to, for any current weather state cluster, obtain the shape similarity between any historical wind speed curve of the current weather state cluster and the other historical wind speed curves; sum the shape similarity of each curve to obtain the sum of the shape similarity of each historical wind speed curve; and take the historical wind speed curve with the largest sum of shape similarity as the central historical wind speed curve of the current weather state cluster.

[0160] In an exemplary embodiment, the training device for the wind power ultra-short-term power prediction model is further configured to acquire the current meteorological feature sequence, the original meteorological features to be screened, and the historical wind power ultra-short-term power corresponding to the current screening round; acquire the average redundancy between the original meteorological features to be screened and the meteorological features contained in the current meteorological feature sequence; based on the mutual information between the original meteorological features to be screened and the historical wind power ultra-short-term power, and each average redundancy, select target meteorological features from the original meteorological features to be screened, update the current meteorological feature sequence using the target meteorological features, and return to execute the steps of acquiring the current meteorological feature sequence and the original meteorological features to be screened corresponding to the current screening round, until the number of meteorological features contained in the current meteorological feature sequence corresponding to the current screening round reaches a preset target number.

[0161] In one embodiment, when the current screening round is the first screening round, the training device for the wind power ultra-short-term power prediction model is further used to add the original meteorological features to be screened with the largest mutual information to the original meteorological feature sequence to obtain the current meteorological feature sequence corresponding to the current screening round.

[0162] In one embodiment, when the current screening round is not the first screening round, the training device of the wind power ultra-short-term power prediction model is further used to obtain the updated current meteorological feature sequence corresponding to the previous screening round; the previous screening round is the previous screening round of the current screening round; and the updated current meteorological feature sequence is used as the current meteorological feature sequence corresponding to the current screening round.

[0163] In an exemplary embodiment, the training module 504 is further configured to input the meteorological feature sequences corresponding to each weather state cluster into the wind power ultra-short-term power prediction model to be trained; obtain the hidden state sequences corresponding to each meteorological feature sequence through the LSTM unit in the wind power ultra-short-term power prediction model based on the meteorological feature sequences corresponding to each weather state; input the hidden state sequences corresponding to each meteorological feature sequence into the weighting unit in the wind power ultra-short-term power prediction model; perform weighted summation on each hidden state sequence according to the prediction weights corresponding to each hidden state sequence to obtain the context vector corresponding to each meteorological feature sequence; input the context vector corresponding to each meteorological feature sequence into the fully connected unit in the wind power ultra-short-term power prediction model; map the context vector corresponding to each meteorological feature sequence to obtain the predicted wind power ultra-short-term power corresponding to each meteorological feature sequence; and update the model parameters of the wind power ultra-short-term power prediction model to be trained according to the historical wind power ultra-short-term power and predicted wind power ultra-short-term power corresponding to the meteorological feature sequences corresponding to each weather state cluster, thereby obtaining the optimal wind power ultra-short-term power prediction model for power prediction under different weather conditions.

[0164] The modules in the training device for the aforementioned wind power ultra-short-term power prediction model can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0165] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores historical wind speed curves, the first derivatives of historical wind speed curves, the shape similarity between historical wind speed curves, the weather state clusters corresponding to each historical wind speed curve, and meteorological feature sequences. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a training method for a wind power ultra-short-term power prediction model.

[0166] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0167] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the training method of the wind power ultra-short-term power prediction model of the above embodiment.

[0168] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the training method for the wind power ultra-short-term power prediction model of the above embodiment.

[0169] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the training method for the wind power ultra-short-term power prediction model of the above embodiment.

[0170] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0171] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0172] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0173] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A training method for a wind power ultra-short-term power prediction model, characterized in that, The method includes: Obtain the first derivative of the historical wind speed curve; The shape similarity between the historical wind speed curves is obtained based on the first derivative of each curve. The historical wind speed curves are then clustered according to the shape similarity to obtain the weather state clusters corresponding to each historical wind speed curve. The weather states corresponding to each of the historical wind speed curves are clustered as the weather state clusters corresponding to the meteorological feature sequences to which each of the historical wind speed curves are located. Based on the meteorological feature sequences corresponding to the clusters of the aforementioned weather states, the wind power ultra-short-term power prediction models to be trained are trained respectively to obtain the optimal wind power ultra-short-term power prediction models for power prediction under different weather conditions.

2. The method according to claim 1, characterized in that, The step of clustering the historical wind speed curves according to their shape similarity to obtain the weather state clusters corresponding to each historical wind speed curve includes: From the historical wind speed curves mentioned above, determine the central historical wind speed curves of different weather conditions in the current clustering round; Based on the shape similarity between the other historical wind speed curves (excluding the central historical wind speed curves) and the central historical wind speed curves, the other historical wind speed curves are clustered and divided. Based on the historical wind speed curves of different weather conditions, update the center historical wind speed curves of different weather condition clusters, and return to execute the step of determining the center historical wind speed curves of different weather conditions in the current clustering round from each of the historical wind speed curves, and clustering the other historical wind speed curves according to the shape similarity between them and the center historical wind speed curves, until the center historical wind speed curves of different weather conditions are no longer updated, thus obtaining the weather condition clusters corresponding to each historical wind speed curve.

3. The method according to claim 2, characterized in that, The step of updating the central historical wind speed curves of different weather condition clusters according to the historical wind speed curves of different weather conditions includes: For any current weather state cluster, obtain the shape similarity between any historical wind speed curve in the current weather state cluster and the other historical wind speed curves; The sum of the shape similarity of each of the above is obtained by summing the shape similarity of each of the historical wind speed curves; The historical wind speed curve with the highest sum of shape similarity is used as the central historical wind speed curve for the current weather state cluster.

4. The method according to claim 1, characterized in that, The meteorological feature sequence is constructed through the following steps: Obtain the current meteorological feature sequence, the original meteorological features to be screened, and the historical wind power ultra-short-term power corresponding to the current screening round; Obtain the average redundancy between the original meteorological features to be screened and the meteorological features contained in the current meteorological feature sequence; Based on the mutual information between the original meteorological features to be screened and the historical wind power ultra-short-term power, and the average redundancy of each, target meteorological features are screened from the original meteorological features to be screened, and the current meteorological feature sequence is updated using the target meteorological features. Then, the process returns to the step of obtaining the current meteorological feature sequence corresponding to the current screening round and the original meteorological features to be screened, until the number of meteorological features contained in the current meteorological feature sequence corresponding to the current screening round reaches the preset target number.

5. The method according to claim 4, characterized in that, When the current screening round is the first screening round, obtaining the current meteorological feature sequence corresponding to the current screening round includes: The original meteorological features with the highest mutual information are added to the original meteorological feature sequence to obtain the current meteorological feature sequence corresponding to the current screening round.

6. The method according to claim 4, characterized in that, When the current screening round is not the first screening round, obtaining the current meteorological feature sequence corresponding to the current screening round includes: Obtain the updated current meteorological feature sequence corresponding to the previous screening round; the previous screening round is the screening round before the current screening round. The updated current meteorological feature sequence is used as the current meteorological feature sequence corresponding to the current screening round.

7. The method according to claim 4, characterized in that, The method involves training the wind power ultra-short-term power prediction model based on the meteorological feature sequences corresponding to each weather state cluster, thereby obtaining the optimal wind power ultra-short-term power prediction model for power prediction under different weather conditions, including: The meteorological feature sequences corresponding to each weather state cluster are input into the wind power ultra-short-term power prediction model to be trained. The LSTM unit in the wind power ultra-short-term power prediction model obtains the hidden state sequence corresponding to each meteorological feature sequence based on the meteorological feature sequences corresponding to each weather state. The hidden state sequence corresponding to each meteorological feature sequence is input into the weighting unit in the wind power ultra-short-term power prediction model. Based on the prediction weights corresponding to each hidden state sequence, the hidden state sequences are weighted and summed to obtain the context vector corresponding to each meteorological feature sequence. The context vectors corresponding to each meteorological feature sequence are input into the fully connected unit in the wind power ultra-short-term power prediction model to map the context vectors corresponding to each meteorological feature sequence, thereby obtaining the predicted wind power ultra-short-term power corresponding to each meteorological feature sequence. Based on the historical and predicted wind power ultra-short-term power corresponding to the meteorological feature sequences of each weather state cluster, the model parameters of the wind power ultra-short-term power prediction model to be trained are updated to obtain the optimal wind power ultra-short-term power prediction model for power prediction under different weather conditions.

8. A training device for a wind power ultra-short-term power prediction model, characterized in that, The device includes: The acquisition module is used to obtain the first derivative of historical wind speed curves; The clustering module is used to obtain the shape similarity between the historical wind speed curves based on the first derivative of each curve, and to cluster the historical wind speed curves according to the shape similarity of each curve to obtain the weather state clusters corresponding to each historical wind speed curve. The determination module is used to cluster the weather states corresponding to each of the historical wind speed curves as the weather state clusters corresponding to the meteorological feature sequences to which each of the historical wind speed curves is located. The training module is used to train the wind power ultra-short-term power prediction model to be trained based on the meteorological feature sequences corresponding to each weather state cluster, so as to obtain the optimal wind power ultra-short-term power prediction model for power prediction under different weather conditions.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images