Island wind power short-term prediction method considering multiple meteorological scenes
The K-means clustering algorithm divides meteorological scenes and establishes a targeted wind power prediction sub-model, which solves the accuracy of wind power prediction in complex meteorological environments in the island microgrid, and realizes high-precision wind power prediction.
Patent Information
- Application Number
- CN202411512233.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-28
- Publication Date
- 2025-02-11
AI Technical Summary
The accuracy of wind power prediction of island microgrids reduces in complex meteorological environments, and it is difficult for the prior art to effectively deal with meteorological uncertainty and extreme meteorological events.
The K-means clustering algorithm is used to divide the typical meteorological scenes of wind power prediction into multiple scenarios, and a targeted wind power prediction sub-model is established based on the divided scenarios. The sub-model weight is calculated by calculating the membership function and data-driven method, and the sub-model weight is calculated. Weighted prediction.
The accuracy of wind power prediction is improved, so that the prediction model can adaptively output high-precision results for different meteorological conditions, and enhance the stable operation ability of the isolated microgrid.
Smart Images
Figure CN119419766A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a short-term prediction method for isolated island wind power considering multiple meteorological scenarios. Background Art
[0002] Wind power has the characteristics of randomness, intermittency and volatility. Large-scale incorporation of wind power into isolated microgrids will make it more difficult to balance power generation and consumption and significantly increase the uncertainty of system operation. The contradiction between the stable operation of isolated microgrids and the randomness and volatility of new energy is becoming increasingly prominent.
[0003] In this regard, establishing a high-precision and high-reliability wind power prediction system can weaken these effects and ensure the stable operation of isolated island microgrids. However, the meteorological conditions in isolated island areas vary greatly, and the existing wind power prediction methods are insufficient in dealing with meteorological uncertainties and extreme meteorological events, resulting in reduced accuracy of wind power prediction in complex meteorological environments.
[0004] Therefore, the present invention provides a short-term prediction method for isolated island wind power taking into account multiple meteorological scenarios. Summary of the invention
[0005] The present invention provides a short-term prediction method for isolated island wind power taking into account multiple meteorological scenarios, which can effectively solve the above problems.
[0006] The present invention is achieved in that:
[0007] A short-term prediction method for isolated island wind power considering multiple meteorological scenarios comprises the following steps:
[0008] S1, using K-means clustering algorithm and comprehensive meteorological factors, the typical meteorological scenarios for wind power prediction are divided into multiple scenarios;
[0009] S2, establish a targeted wind power prediction sub-model based on the divided typical meteorological scenarios;
[0010] S3, calculate the membership function of the sample to be predicted and the cluster center, use the data-driven method to calculate the sub-model weight, and use the combination of the two as the weight coefficient of each sub-model;
[0011] S4, by weighting the prediction results of each sub-model, the final wind power prediction result is obtained.
[0012] As a further improvement, step S1 specifically comprises obtaining the meteorological factors of the environment at different times of the day, performing preprocessing, clustering the historical meteorological scenes using the K-means clustering algorithm based on the preprocessed data, classifying the historical sample data according to typical meteorological conditions, constructing a sub-dataset, and dividing the sub-dataset into a training set and a validation set.
[0013] As a further improvement, the step S1 comprises the following steps:
[0014] S11, annotate the wind power data according to its meteorological factors; the i-th sample is shown in formula (1);
[0015] x i =(wind power i , air pressure i ,wind direction i , wind speed i ,temperature i ,illumination i ,humidity i ) (1)
[0016] S12, randomly select k samples with different weather conditions as the initial cluster centers, at least different weather conditions, such as sunny, cloudy, thunderstorm, typhoon, foggy, rainy, and snowy; calculate the remaining samples x in turn i To the jth cluster center μ j The distance D j (i);
[0017] D j (i)=||x i -μ j || 2 (2)
[0018] S13, according to sample x i The minimum distance from other cluster centers is assigned to the nearest cluster center;
[0019] D min (i) = min(D j (i)) (3)
[0020] S14, then use formula (4) to update the cluster center; continue to iterate until the cluster center does not change; assign the samples to the corresponding clusters according to the final cluster center; the jth cluster center is shown in formula (4);
[0021]
[0022] Where N j Represents the number of samples of the jth cluster center.
[0023] As a further improvement, the step S2 comprises the following sub-steps:
[0024] Step S21, extracting local features using the CNN layer; inputting the sample data and historical wind power obtained in step S1 into the CNN-Transformer model, and extracting meteorological features through a series of CNN layers;
[0025] Step S22, converting the feature map into a sequence format, adding the extracted meteorological features and historical meteorological features to the fully connected layer, and converting the meteorological features extracted by the CNN layer into a sequence format suitable for the Transformer encoder input;
[0026] Step S23, input embedding: the Transformer encoder receives the meteorological feature vector from the CNN as input;
[0027] Step S24, multi-head self-attention mechanism: a multi-head attention mechanism is used to capture the global dependencies in the meteorological feature vector. By calculating the correlation scores between different features in the input sequence, the self-attention mechanism can effectively model long-distance dependencies.
[0028] Step S25, feed-forward neural network layer; after each self-attention layer, connect to the feed-forward neural network layer to further process these meteorological features;
[0029] Step S25, residual connection and layer normalization; in order to speed up training and improve model performance, Transformer uses residual connection and layer normalization in each sublayer; residual connection allows the model to skip certain layers and directly pass gradient information; layer normalization helps stabilize the training process and accelerate convergence.
[0030] As a further improvement, a fusion stage S5 is further included between the CNN-Transformer processing stage and the output stage; the fusion stage includes the following steps:
[0031] S51, feature fusion: according to the specific fusion mode, such as early layer fusion, horizontal layer fusion, and sequential fusion, the output features of CNN and CNN-Transformer are combined or concatenated;
[0032] S52, the fused features are applied to a fully connected layer to map the high-dimensional features to the output space of the classification or regression task and output the predicted data.
[0033] As a further improvement, in step S3, the membership function is used to calculate the similarity between different meteorological data sets, and then the weather recognition model E is obtained by training the sample data of the training set. s ; E s A set of weight values can be output according to the actual meteorological conditions; specifically, the calculation method of the membership function is shown in formula (5);
[0034]
[0035] Among them, δ(x i ,μ j ) represents the sample xi For the membership degree of wind power prediction sub-model j, μ j It is the core of the wind power prediction sub-model.
[0036] As a further improvement, the weather recognition model E s The training steps are as follows:
[0037] Step S41, it is necessary to construct a mapping relationship between meteorological conditions and sub-model weight values; the output result is shown in formula (6);
[0038]
[0039] α i is the weight of the ith sub-model, P i is the wind power prediction result output by the i-th sub-model;
[0040] Step S42, through the neural network structure of the SVR algorithm, the particle swarm algorithm is used to perform parameter optimization training to obtain the weather recognition model E s .
[0041] As a further improvement, in step S3, the weight coefficient of the j-th sub-model is as follows:
[0042] w i =α i ·δ(x i ,μ j ) (7)
[0043] As a further improvement, the step S4 further comprises: i Weighted, the final wind power prediction result P is obtained. result , the formula is shown in (8);
[0044]
[0045] The beneficial effects of the present invention are as follows: the present invention classifies meteorological conditions, and by training CNN-Transformer models for different weather conditions respectively, it is possible to better capture the impact of meteorological conditions on wind speed, thereby improving the accuracy of the sub-prediction model; finally, the weight coefficient of the sub-model is calculated through the results of the membership function and the data-driven model. This improvement enables the prediction model to adaptively output high-precision results for different meteorological conditions. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.
[0047] Figure 1 The present invention provides a wind power prediction model block diagram of a short-term prediction method for isolated island wind power taking into account multiple meteorological scenarios.
[0048] Figure 2 The present invention provides a CNN-Transformer model structure diagram of a short-term prediction method for isolated island wind power taking into account multiple meteorological scenarios. DETAILED DESCRIPTION
[0049] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0050] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.
[0051] Reference Figure 1-Figure 2 As shown, this embodiment provides a specific implementation method of a short-term prediction method for isolated island wind power considering multiple meteorological scenarios, including the following steps:
[0052] S1, using K-means clustering algorithm and comprehensive meteorological factors, the typical meteorological scenarios for wind power prediction are divided into multiple scenarios;
[0053] S2, establish a targeted wind power prediction sub-model based on the divided typical meteorological scenarios;
[0054] S3, calculate the membership function of the sample to be predicted and the cluster center, use the data-driven method to calculate the sub-model weight, and use the combination of the two as the weight coefficient of each sub-model;
[0055] S4, by weighting the prediction results of each sub-model, the final wind power prediction result is obtained.
[0056] Specifically, step S1 is to obtain meteorological factors of the environment at different times of the day, take wind speed as the main consideration, and comprehensively consider meteorological factors such as temperature, wind direction, humidity, and weather, perform preprocessing, cluster historical meteorological scenes using the K-means clustering algorithm based on the preprocessed data, classify historical sample data according to typical meteorological states, construct a sub-dataset, and divide the sub-dataset into a training set and a validation set; step S1 includes the following steps:
[0057] S11, annotate the wind power data according to its meteorological factors such as air pressure, wind direction, wind speed, temperature, light, humidity, etc.; the i-th sample is shown in formula (1);
[0058] x i =(wind power i , air pressure i ,wind direction i , wind speed i ,temperature i ,illumination i ,humidity i ) (1)
[0059] S12, randomly select k samples with different weather conditions as the initial cluster centers, at least different weather conditions, such as sunny, cloudy, thunderstorm, typhoon, foggy, rainy, snowy, etc.; calculate the remaining samples x in turn i To the jth cluster center μ j The distance D j (i);
[0060] D j (i)=||x i -μ j || 2 (2)
[0061] S13, according to sample x i The minimum distance from other cluster centers is assigned to the nearest cluster center;
[0062] D min (i) = min(D j (i)) (3)
[0063] S14, then use formula (4) to update the cluster center; continue to iterate until the cluster center does not change; assign the samples to the corresponding clusters according to the final cluster center; the jth cluster center is shown in formula (4);
[0064]
[0065] Where N j Represents the number of samples of the jth cluster center.
[0066] In some embodiments, step S2 is to establish a data-driven prediction sub-model between meteorological factors such as air pressure, wind direction, wind speed, temperature, light, humidity and wind power based on the CNN-Transformer algorithm for different typical meteorological conditions; the wind power prediction model based on CNN-Transformer is a deep learning architecture that combines convolutional neural network and Transformer model. This model simultaneously utilizes the advantages of CNN in extracting local features of wind speed data and the ability of Transformer in processing global relationships and long-term dependencies; more specifically, step S2 includes the following sub-steps:
[0067] Step S21, extracting local features using the CNN layer; inputting the sample data and historical wind power obtained in step S1 into the CNN-Transformer model, and extracting meteorological features through a series of CNN layers;
[0068] Step S22, converting the feature map into a sequence format, adding the extracted meteorological features and historical meteorological features to the fully connected layer, and converting the meteorological features extracted by the CNN layer into a sequence format suitable for the Transformer encoder input;
[0069] Step S23, input embedding: the Transformer encoder receives the meteorological feature vector from the CNN as input. After being processed by the Transformer encoder, the feature vector is further compressed in spatial dimension through an adaptive average pooling layer;
[0070] Step S24, multi-head self-attention mechanism: a multi-head attention mechanism is used to capture the global dependencies in the meteorological feature vector. By calculating the correlation scores between different features in the input sequence, the self-attention mechanism can effectively model long-distance dependencies.
[0071] Step S25, feed-forward neural network layer; after each self-attention layer, connect to the feed-forward neural network layer to further process these meteorological features;
[0072] Step S25, residual connection and layer normalization; in order to speed up training and improve model performance, Transformer uses residual connection and layer normalization in each sublayer; residual connection allows the model to skip certain layers and directly pass gradient information; layer normalization helps stabilize the training process and accelerate convergence.
[0073] In some embodiments, a fusion stage S5 is further included between the CNN-Transformer processing stage and the output stage; the fusion stage includes the following steps:
[0074] S51, feature fusion: according to the specific fusion mode, such as early layer fusion, horizontal layer fusion, and sequential fusion, the output features of CNN and CNN-Transformer are combined or concatenated;
[0075] S52, the fused features are applied to a fully connected layer to map the high-dimensional features to the output space of the classification or regression task and output the predicted data.
[0076] Historical wind power data and meteorological information such as air pressure, wind direction, wind speed, temperature, light, humidity, etc. are selected as input features of the CNN-Transformer model, and the current wind power data is used as label data; the sub-dataset is divided into a training set and a validation set, and the loss function and optimizer are defined. The CNN-Transformer model is trained using the training set; and the performance of the model is verified using the validation set.
[0077] In step S3, the membership function is used to calculate the similarity between different meteorological data sets, and then the weather recognition model E is obtained by training the sample data of the training set. s ; E s A set of weight values can be output according to the actual meteorological conditions; specifically, the calculation method of the membership function is shown in formula (5);
[0078]
[0079] Among them, δ(x i ,μ j ) represents the sample x i For the membership degree of wind power prediction sub-model j, μ j It is the core of the wind power prediction sub-model.
[0080] Weather Identification Model E s The training steps are as follows:
[0081] Step S41, it is necessary to construct a mapping relationship between meteorological conditions and sub-model weight values; the output result is shown in formula (6);
[0082]
[0083] α i is the weight of the ith sub-model, P i It is the wind power prediction result output by the ith sub-model.
[0084] Step S42, through the neural network structure of the SVR algorithm, the particle swarm algorithm is used to perform parameter optimization training to obtain the weather recognition model E s ;
[0085] Finally, the weight coefficient of the jth sub-model is as follows:
[0086] w i =α i ·δ(x i ,μ j ) (7)
[0087] In some embodiments, step S4 further comprises weighting each sub-model result P i Weighted, the final wind power prediction result P is obtained. result , the formula is shown in (8);
[0088]
[0089] Test Trial:
[0090] The hourly wind speed data set of a certain area from 2006 to 2016 was processed to separate the wind speed under specific conditions of sunny days, cloudy days, thunderstorm days, typhoon days, foggy days, rainy days, and snowy days. These separated data were used to train the wind speed prediction sub-model based on CNN-Transformer, and finally the sub-model results were weighted by the weight coefficient.
[0091] The similarity of each sub-model is calculated by the membership function as shown in Table 1 below;
[0092] Table 1. Membership function to calculate the similarity of sub-models
[0093] Model sunny rain Foggy Day Snowy Days Thunderstorm Sunny Day Model 0.887 0.245 0.154 0.0125 0.295 Rainy Day Model 0.268 0.958 0.125 0.2544 0.542 Fog model 0.274 0.244 0.854 0.141 0.125 Snow model 0.241 0.141 0.154 0.981 0.215 Thunderstorm model 0.215 0.598 0.154 0.114 0.841
[0094] The weight coefficients of each sub-model calculated by the data algorithm are shown in Table 2 below;
[0095] Table 2 Data algorithm to calculate the weight coefficient of each sub-model
[0096] Model sunny rain Foggy Day Snowy Days Thunderstorm Sunny Day Model 0.951 0.211 0.221 0.241 0.321 Rainy Day Model 0.189 0.896 0.166 0.154 0.433 Fog model 0.123 0.194 0.954 0.241 0.228 Snow model 0.194 0.211 0.204 0.885 0.105 Thunderstorm model 0.128 0.463 0.208 0.281 0.910
[0097] Combining Table 1 and Table 2, it can be concluded that the similarity of the sub-models calculated by the membership function is higher than the weight coefficient of the sub-models calculated by the data algorithm. The use of the membership function to calculate the sub-models is more conducive to calculating the similarity of the meteorology and improving the short-term forecast of wind power.
[0098] The error results of the CNN-Transformer prediction model and other algorithms are shown in Table 3 below;
[0099] Table 3
[0100]
[0101] It can be seen from Table 3 that the CNN-Transformer prediction model has the smallest error.
[0102] In summary, the present invention classifies meteorological conditions and trains CNN-Transformer models for different weather conditions respectively, which can better capture the impact of meteorological conditions on wind speed, thereby improving the accuracy of the sub-prediction model; finally, the weight coefficient of the sub-model is calculated through the results of the membership function and the data-driven model. This improvement enables the prediction model to adaptively output high-precision results for different meteorological conditions.
[0103] Using a wind speed dataset under specific circumstances to train a model to predict wind speed under specific circumstances has significant advantages over using a dataset for all circumstances for training.
[0104] By training on data from a specific situation, the model is able to more deeply learn and capture the unique patterns and regularities of wind speed in that specific situation, thereby achieving more accurate predictions of specific wind speeds.
[0105] Data sets in specific situations are often more focused, which can reduce the interference of irrelevant factors and improve the prediction accuracy of wind speed prediction models.
[0106] In addition, data sets for specific situations are usually easier to obtain and annotate, making the training process of the wind speed prediction model more efficient and convenient. At the same time, training for specific situations can also improve the efficiency of the wind speed prediction model.
[0107] Since the size of the wind speed forecast dataset used for training is relatively small, the wind speed forecast model will be trained faster and require less computing resources.
[0108] This allows for more flexible adjustment and optimization of the model in practical applications to meet wind speed prediction requirements in specific scenarios.
[0109] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0110] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0111] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0112] It should be noted that in the claims, any reference signs placed between brackets shall not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claim. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention may be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third etc. does not indicate any order. These words may be interpreted as names.
[0113] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.
[0114] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.
[0115] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0116] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms should not be understood as necessarily referring to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.
Claims
1. A short-term prediction method for isolated island wind power considering multiple meteorological scenarios, characterized in that: The steps include: S1, using K-means clustering algorithm and comprehensive meteorological factors, the typical meteorological scenarios for wind power prediction are divided into multiple scenarios; S2, establish a targeted wind power prediction sub-model based on the divided typical meteorological scenarios; S3, calculate the membership function of the sample to be predicted and the cluster center, use the data-driven method to calculate the sub-model weight, and use the combination of the two as the weight coefficient of each sub-model; S4, by weighting the prediction results of each sub-model, the final wind power prediction result is obtained.
2. The method for short-term prediction of isolated island wind power considering multiple meteorological scenarios according to claim 1 is characterized in that: Specifically, step S1 is to obtain the meteorological factors of the environment at different times of the day, perform preprocessing, cluster the historical meteorological scenes using the K-means clustering algorithm based on the preprocessed data, classify the historical sample data according to typical meteorological conditions, construct a sub-dataset, and divide the sub-dataset into a training set and a validation set.
3. The method for short-term prediction of isolated island wind power considering multiple meteorological scenarios according to claim 2 is characterized in that: The step S1 comprises the following steps: S11, annotate the wind power data according to its meteorological factors; the i-th sample is shown in formula (1); x i =(wind power i , air pressure i ,wind direction i , wind speed i ,temperature i ,illumination i ,humidity i )(1) S12, randomly select k samples of different meteorological conditions as initial cluster centers, at least different meteorological conditions, such as sunny day, cloudy day, thunderstorm day, typhoon day, foggy day, rainy day, snowy day; Calculate the remaining samples x in turn i To the jth cluster center μ j The distance D j (i); D j (i)=||x i -m j || 2 (2) S13, according to sample x i The minimum distance from other cluster centers is assigned to the nearest cluster center; D min (i)=min(D j (i)) (3) S14, then use formula (4) to update the cluster center; continue to iterate until the cluster center does not change; assign the samples to the corresponding clusters according to the final cluster center; the jth cluster center is shown in formula (4); Where N j Represents the number of samples of the jth cluster center.
4. The method for short-term prediction of isolated island wind power considering multiple meteorological scenarios according to claim 1 is characterized in that: The step S2 comprises the following sub-steps: Step S21, extracting local features using the CNN layer; inputting the sample data and historical wind power obtained in step S1 into the CNN-Transformer model, and extracting meteorological features through a series of CNN layers; Step S22, converting the feature map into a sequence format, adding the extracted meteorological features and historical meteorological features to the fully connected layer, and converting the meteorological features extracted by the CNN layer into a sequence format suitable for the Transformer encoder input; Step S23, input embedding: the Transformer encoder receives the meteorological feature vector from the CNN as input; Step S24, multi-head self-attention mechanism: a multi-head attention mechanism is used to capture the global dependencies in the meteorological feature vector. By calculating the correlation scores between different features in the input sequence, the self-attention mechanism can effectively model long-distance dependencies. Step S25, feed-forward neural network layer; after each self-attention layer, connect to the feed-forward neural network layer to further process these meteorological features; Step S25, residual connection and layer normalization; in order to speed up training and improve model performance, Transformer uses residual connection and layer normalization in each sublayer; residual connection allows the model to skip certain layers and directly pass gradient information; layer normalization helps stabilize the training process and accelerate convergence.
5. The method for short-term prediction of isolated island wind power considering multiple meteorological scenarios according to claim 4 is characterized in that: A fusion stage S5 is also included between the CNN-Transformer processing stage and the output stage; the fusion stage includes the following steps: S51, feature fusion: according to the specific fusion mode, such as early layer fusion, horizontal layer fusion, and sequential fusion, the output features of CNN and CNN-Transformer are combined or concatenated; S52, the fused features are applied to a fully connected layer to map the high-dimensional features to the output space of the classification or regression task and output the predicted data.
6. The method for short-term prediction of isolated island wind power considering multiple meteorological scenarios according to claim 1 is characterized in that: In step S3, the membership function is used to calculate the similarity between different meteorological data sets, and then the weather recognition model E is obtained by training the sample data of the training set. s ; E s A set of weight values can be output according to the actual meteorological conditions; specifically, the calculation method of the membership function is shown in formula (5); Among them, δ(x i ,μ j ) represents the sample x i For the membership degree of wind power prediction sub-model j, μ j It is the core of the wind power prediction sub-model.
7. The method for short-term prediction of isolated island wind power considering multiple meteorological scenarios according to claim 1, characterized in that: Weather Identification Model E s The training steps are as follows: Step S41, it is necessary to construct a mapping relationship between meteorological conditions and sub-model weight values; the output result is shown in formula (6); α i is the weight of the ith sub-model, P i is the wind power prediction result output by the i-th sub-model; Step S42, through the neural network structure of the SVR algorithm, the particle swarm algorithm is used to perform parameter optimization training to obtain the weather recognition model E s .
8. The method for short-term prediction of isolated island wind power considering multiple meteorological scenarios according to claim 7 is characterized in that: In step S3, the weight coefficient of the j-th sub-model is as follows: w i =a i ·δ(x i ,m j ) (7) 9. The method for short-term prediction of isolated island wind power considering multiple meteorological scenarios according to claim 1, characterized in that: The step S4 also includes dividing each sub-model result P according to the weight i Weighted, the final wind power prediction result P is obtained. result , the formula is shown in (8); 。