Wind power prediction method, system and equipment based on cold wave identification and medium

Abstract

The invention discloses a wind power prediction method, system and device based on cold wave identification and a medium, and relates to the technical field of wind power prediction. The method comprises the following steps: acquiring meteorological data of a target area and operation data of a wind turbine generator, and constructing a cold wave discrimination standard; generating a cold wave sample based on a cold wave discrimination standard and a sample enhancement method; constructing an improved U-net-based cold-wave weather identification model and a cold-wave continuity event-based classification optimization method; performing cold-wave identification on the meteorological data to be detected by using the cold-wave weather identification model and a classification optimization method to obtain a cold-wave identification result in a future prediction period; constructing a wind power combined prediction model; and according to a cold wave identification result of a future prediction period, inputting the data into a corresponding prediction model, and combining output results of the two models to obtain a complete wind power prediction result. According to the invention, the identification of future cold-wave weather can be enhanced, and the accuracy of wind power prediction in the cold-wave weather can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power prediction, and particularly to a wind power prediction method, system, device and medium based on cold wave identification. Background Art

[0002] With the depletion of global fossil energy and the increasing prominence of environmental problems, the proportion of new energy represented by wind power in the energy structure has increased. In recent years, under the background of global climate change, extreme cold wave weather occurs frequently. In cold wave weather, the temperature often drops suddenly and the wind speed changes abruptly. When cold wave weather occurs in a concentrated large-scale wind farm area, the volatility and uncertainty of wind power increase, and in severe cases, problems such as shutdown and disconnection of wind turbines on a large scale will occur. The Texas power crisis in 2021 was directly caused by long-term extreme cold weather, which led to insufficient sustainable supply capacity of new energy and caused more than 4.5 million users to lose power supply. Therefore, accurate cold wave weather forecasting and accurate wind power prediction during cold wave weather contribute to the stable operation of the power system under extreme weather conditions.

[0003] Current research on wind power prediction under extreme weather conditions mainly focuses on wind power ramp events, and there is little in-depth research on wind power prediction under the influence of cold wave weather, and research on cold wave event forecasting in the operation scenario of large-scale wind power clusters in large areas is even scarcer. Therefore, there is an urgent need for a new wind power prediction method at present. Summary of the Invention

[0004] The purpose of the present invention is to provide a wind power prediction method, system, device and medium based on cold wave identification, aiming to solve or improve at least one of the above technical problems.

[0005] To achieve the above purpose, the present invention provides the following solutions: A wind power prediction method based on cold wave identification, comprising: Obtaining meteorological data and wind turbine operation data of a target area, and constructing a cold wave discrimination criterion; Generating cold wave samples based on the cold wave discrimination criterion and a sample enhancement method; the cold wave samples include meteorological data with time characteristics and cold wave characteristics; the sample enhancement method is an enhanced classification effect method using generative adversarial network technology; Constructing a cold wave weather recognition model based on an improved U-net and a classification optimization method based on cold wave continuous events; the cold wave weather recognition model is obtained by training the improved U-net with the cold wave samples; Identify cold snaps in the meteorological data to be measured using the cold snap weather identification model and the classification optimization method, and obtain the cold snap identification results for the future prediction period; the cold snap identification results include the cold snap occurrence stage and the non-cold snap occurrence stage; Construct a combined wind power prediction model; the combined wind power prediction model includes a non-cold snap prediction model and a cold snap prediction model; the combined wind power prediction model is constructed based on a long short-term memory network and a non-stationary Transformer combined model; wherein, the non-stationary Transformer combined model includes a non-stationary Transformer data embedding module and a non-stationary Transformer encoder connected in sequence; the output ends of the long short-term memory network and the non-stationary Transformer data embedding module are both connected to the input end of the non-stationary Transformer encoder; According to the cold snap identification results for the future prediction period, input the corresponding data into the corresponding prediction model, and combine the output results of the two models to obtain the complete wind power prediction result.

[0006] Optionally, the cold snap discrimination criterion is specifically expressed as: , wherein, is the daily minimum temperature, is the daily minimum temperature, is the normalized predicted data, is the normalized true power data, is the total installed capacity of the wind power cluster; is the temperature difference; when all conditions are met, it is determined that a cold snap weather occurs at this time.

[0007] Optionally, the structure of the improved U-net is that in the first-layer down-convolution operation, by changing the input channels, the feature data and the context features are used as data inputs at the same time; the feature data is the given data; the context features are the data features constructed from relevant information; the feature data and the context features have the same time step.

[0008] Optionally, the classification optimization method based on cold snap continuous events is specifically expressed as: , wherein, is the original cold snap classification result, is the optimized cold snap classification result, is the duration of the current normal weather event, is the duration of the current cold snap event, is the neighborhood merging threshold, is the isolated determination threshold, is the classification result sequence, .

[0009] Optionally, before predicting using the wind power combined prediction model, it further includes: evaluating the trained wind power combined prediction model; the process of the evaluation is to calculate the error between the predicted power calculated by the model and the measured power through an error evaluation criterion; the error evaluation criterion includes root mean square error and mean absolute error.

[0010] The present invention also provides a wind power prediction system based on cold wave identification, including: A cold wave discrimination criterion establishment unit, configured to obtain meteorological data and wind turbine operation data of a target area, and construct a cold wave discrimination criterion; A cold wave sample generation unit, configured to generate cold wave samples based on the cold wave discrimination criterion and a sample enhancement method; the cold wave samples include meteorological data with time characteristics and cold wave characteristics; the sample enhancement method is an enhanced classification effect method using generative adversarial network technology; A cold wave classification method construction unit, configured to construct a cold wave weather recognition model based on an improved U-net and a classification optimization method based on cold wave continuous events; the cold wave weather recognition model is obtained by training the improved U-net using the cold wave samples; A cold wave identification unit, configured to use the cold wave weather recognition model and the classification optimization method to identify cold waves in the to-be-measured meteorological data, and obtain cold wave identification results for a future prediction period; the cold wave identification results include a cold wave occurrence stage and a non-cold wave occurrence stage; A power prediction model construction unit, configured to construct a wind power combined prediction model; the wind power combined prediction model includes a non-cold wave prediction model and a cold wave prediction model; the wind power combined prediction model is constructed based on a long short-term memory network and a non-stationary Transformer combined model; wherein, the non-stationary Transformer combined model includes a non-stationary Transformer data embedding module and a non-stationary Transformer encoder connected in sequence; the output end of the long short-term memory network and the output end of the non-stationary Transformer data embedding module are both connected to the input end of the non-stationary Transformer encoder; A power prediction unit, configured to input corresponding data into corresponding prediction models according to the cold wave identification results for the future prediction period, and combine the output results of the two models to obtain a complete wind power prediction result.

[0011] The present invention also provides an electronic device, including a memory and a processor. The memory is used for storing a computer program, and the processor runs the computer program to enable the electronic device to execute the wind power prediction method based on cold snap identification as described above.

[0012] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the wind power prediction method based on cold snap identification as described above.

[0013] According to the specific embodiments provided by the present invention, the following technical effects are disclosed by the present invention: The present invention discloses a wind power prediction method, system, device and medium based on cold snap identification. The method includes obtaining meteorological data and wind turbine operation data of a target area, and constructing a cold snap discrimination criterion; generating cold snap samples based on the cold snap discrimination criterion and a sample enhancement method; constructing a cold snap weather recognition model based on an improved U-net and a classification optimization method for cold snap continuous events; using the cold snap weather recognition model and the classification optimization method to identify cold snaps in the measured meteorological data to obtain cold snap identification results for future prediction periods; constructing a wind power combined prediction model; according to the cold snap identification results for future prediction periods, inputting the data into the corresponding prediction models, and combining the output results of the two models to obtain a complete wind power prediction result. The present invention can enhance the identification of future cold snap weather and improve the accuracy of wind power prediction under cold snap weather. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0015] Figure 1 is the overall flowchart of the present invention; Figure 2 is the meteorological characteristics of cold snap segments and the wind power segment diagram in this embodiment; Figure 3 is the meteorological data diagram of cold snap weather in this embodiment; Figure 4 is the classification optimization method diagram of cold snap continuous events in this embodiment; Figure 5 is the wind power prediction diagram without cold snap weather in this embodiment; Figure 6 is the wind power prediction diagram of cold snap weather in this embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0017] The purpose of the present invention is to provide a wind power prediction method, system, device and medium based on cold wave identification, aiming to solve or improve at least one of the above technical problems.

[0018] To make the above objects, features and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] As Figure 1 shown, the present invention provides a wind power prediction method based on cold wave identification, including: Obtain the meteorological data and wind turbine operation data of the target area, and construct a cold wave discrimination criterion; Generate cold wave samples based on the cold wave discrimination criterion and the sample enhancement method; the cold wave samples include meteorological data with time characteristics and cold wave characteristics; the sample enhancement method is an enhanced classification effect method using generative adversarial network technology; Construct a cold wave weather recognition model based on the improved U-net and a classification optimization method based on cold wave continuous events; the cold wave weather recognition model is obtained by training the improved U-net with the cold wave samples; Use the cold wave weather recognition model and the classification optimization method to identify cold waves in the measured meteorological data, and obtain the cold wave recognition results for the future prediction period; the cold wave recognition results include the cold wave occurrence stage and the non-cold wave occurrence stage; Construct a wind power combined prediction model; the wind power combined prediction model includes a non-cold wave prediction model and a cold wave prediction model; the wind power combined prediction model is constructed based on a long short-term memory network and a non-stationary Transformer combined model; wherein, the non-stationary Transformer combined model includes a non-stationary Transformer data embedding module and a non-stationary Transformer encoder connected in sequence; the output ends of the long short-term memory network and the non-stationary Transformer data embedding module are both connected to the input end of the non-stationary Transformer encoder; According to the cold wave recognition results for the future prediction period, input the corresponding data into the corresponding prediction model, and combine the output results of the two models to obtain the complete wind power prediction results.

[0020] As a specific implementation, a detailed processing procedure for each of the above steps is provided.

[0021] 1) Proposal of the cold wave discrimination criterion Combining the regional meteorological changes and the operating characteristics of wind turbines, a cold wave discrimination criterion is proposed. The traditional cold wave judgment criterion only considers meteorological characteristics, while the cold wave identification in this solution has an important impact on subsequent wind power prediction. Therefore, an innovative cold wave identification index considering the wind power operation mechanism is proposed. This criterion can accurately identify cold wave weather in the region, especially in terms of the duration of the cold wave, providing accurate discrimination at the hourly level. The cold wave discrimination criterion is as follows: (1) (2) (3) Among them, is the daily minimum temperature, is the daily minimum temperature, is the normalized predicted data, is the normalized actual power data, is the total installed capacity of the wind power cluster.

[0022] 2) Improved U-net classification model Based on the U-Net model, a multi-modal classification method is proposed to optimize the model for the seasonal characteristics of cold wave occurrence. The improved model can not only accurately predict the occurrence time of cold waves, but also improve the recognition accuracy of small-sample cold wave events by generating cold wave weather samples through the adversarial generative network.

[0023] Fusing different types of data to make full use of multi-dimensional information has become an important strategy to improve the performance of the model. This solution proposes a multi-modal U-Net classification model. In the compression channel of the model, time features such as quarter and month are combined with cold wave feature data. Cold wave events have significant seasonal characteristics, and the introduction of multi-modal data enables the U-Net model to effectively identify time information, thus significantly improving the recognition accuracy of cold wave weather. This method not only strengthens the model's ability to capture the occurrence law of cold waves, but also improves the accuracy of cold wave identification.

[0024] The imbalance and insufficiency of sample distribution can lead to the failure of the classification model to fully learn sample features during the training process, thereby reducing the reliability of the prediction results. Especially for small-sample events, this problem is more prominent. Cold wave weather is exactly a typical small-sample event. To solve this problem, this solution proposes a method to enhance the classification effect using generative adversarial network technology based on the existing cold wave sample features. The mechanism of this method is as follows: a large number of cold wave samples are generated using generative adversarial network technology through the initial samples. The conventional method mixes the generated samples into the entire dataset, while the method of this solution is to only add the generated samples to the training set data to avoid the generated samples affecting the initial samples. At the same time, the effectiveness of the generated samples can also be judged according to the classification results.

[0025] The specific steps are as follows ① Use the initial cold wave samples to generate the same number of cold wave samples as the normal samples through the generative adversarial network.

[0026] ② Mix the generated cold wave samples with the training set data to form a new sample training dataset, and do not add the generated cold wave samples to the test set.

[0027] ③ Given the feature data, expand it to the specified step length on the time axis through operations to generate extended sequence features.

[0028] ④ Construct context features from other forms of information such as time and images, and expand the context features to the same time step as the sequence features.

[0029] ⑤ Convert the label data into a format suitable for the classification task using the one-hot encoding method.

[0030] ⑥ Improve the U-Net model structure. In the first-layer down-convolution operation, change the input channels, and at the same time use the feature data and the context features as data inputs at the same time to achieve the fusion of the feature data and the context features within the U-Net model, and enhance the classification effect of other steps of the U-Net model.

[0031] ⑦ The classification model outputs the probabilities of normal weather scenarios and cold wave weather scenarios, and selects the event with the highest probability as the prediction result.

[0032] 3) Classification optimization method based on cold wave continuity events Cold snap weather processes are characterized by persistence, and their impact on wind turbines usually lasts for several hours. For short-term cold snaps with a duration less than or equal to two hours, their occurrence locations can be divided into two cases: 1 Short fluctuations during long cold snaps: During a long-lasting cold snap, the temperature may temporarily rise, but the wind turbines are still under the influence of the cold snap. 2 Isolated short-term cold snaps: Appearing in a time series where no cold snap has occurred for a long time, occasional short-term cold snap events. Such cold snaps have a short duration and relatively little impact on wind power, and can be regarded as a situation where no cold snap has occurred. Given the particularity of short-term cold snap events, to improve the accuracy of future cold snap predictions, it is necessary to continuously optimize the judgment of cold snap occurrences. Through the first optimization, short fluctuations that may cause misjudgment are merged into the continuous cold snap period, and the second optimization reasonably merges isolated short-term cold snap events with normal weather to avoid misclassification of cold snap events and thus optimize the classification results.

[0033] According to the continuous optimization requirements for cold snap event classification, the proposed continuous optimization formula is as follows: (4) Where, is the original cold snap classification result, is the optimized cold snap classification result, is the duration of the current cold snap event (unit: hours), is the duration of the current cold snap event (unit: hours), is the neighborhood merging threshold (value: 2 hours), is the isolated determination threshold (value: 6 hours), is the classification result sequence, .

[0034] 4) Wind power combined prediction method based on cold snap weather identification Traditional time series models often struggle to effectively handle non-stationarity and long-term and short-term dependencies in data. To overcome these challenges, a long short-term memory network + non-stationary Transformer combined model is proposed. Transformer is a deep learning model architecture based on self-attention mechanism.

[0035] The long short-term memory network is a variant of the recurrent neural network, specifically designed to address the vanishing gradient and exploding gradient problems of traditional recurrent neural networks. It can effectively capture short-term and long-term dependencies in time series by introducing a gating mechanism of input gate, forget gate, and output gate.

[0036] The proposed model innovation mechanism: The normalized input sequence is input into the long short-term memory network module, and the output of the long short-term memory network is a three-dimensional tensor. The normalized input sequence is input into the non-stationary Transformer data embedding module to enhance the feature representation ability. The output features of the long short-term memory network and the embedding features of the non-stationary Transformer are weighted and fused, and the fused features are input into the non-stationary Transformer encoder. Finally, the output result is obtained through the non-stationary Transformer model.

[0037] The specific steps for combined wind power prediction are as follows: ① Using the constructed U-net-based cold wave identification and classification model above, taking the numerical weather prediction data in the future prediction part as input, and obtaining the cold wave identification results in the future prediction period.

[0038] ② According to the cold wave identification results, divide the future weather types into the cold wave occurrence stage and the non-cold wave occurrence stage.

[0039] ③ Through continuous optimization, merge the short-term fluctuations that may cause misjudgment into the continuous cold wave period, and reasonably merge the isolated short-term cold wave events with normal weather.

[0040] ④ Use the proposed wind power prediction model to establish a wind power prediction model for the cold wave period and a wind power prediction model for the non-cold wave period respectively.

[0041] ⑤ Input the divided time periods into the two trained prediction models respectively to obtain the wind power prediction results for different time periods.

[0042] ⑥ Combine the prediction results to obtain the final wind power prediction result.

[0043] 5) Simulation calculation Simulation input variables: Analyze based on the measured data of the wind farm to determine the total installed capacity of the power station; the rated capacity of each wind farm; the numerical weather prediction of 100m wind speed, 10m wind speed, temperature, and dew point temperature; the data sampling interval is 1h; according to steps 1) to 4), obtain the real-time prediction result of the whole-field wind power.

[0044] 6) Error analysis Let be i the actual average power in the time period, i be the predicted average power in the time period, (5) The normalized mean absolute error is defined as Equation (6): (6) The predicted power calculated by the model and the measured power are subjected to error calculation through the error evaluation criterion to obtain the root mean square error and mean absolute error of the prediction.

[0045] It can be seen that the present invention not only focuses on the identification method for cold wave weather scenarios, but also particularly forecasts the influence time of cold waves on wind power, accurate to the hour level. By establishing different wind power prediction models for the cold wave season and the non-cold wave season respectively, it shows good wind power prediction accuracy under cold wave weather.

[0046] Based on the above technical solutions, the following embodiments are provided.

[0047] The main steps include: 1) By combining regional meteorological changes and the operating characteristics of wind turbines, a cold wave discrimination criterion is constructed; 2) Using the cold wave discrimination criterion, identify the cold wave occurrence period and analyze the correlation between cold wave meteorological characteristics and wind power; 3) Using existing cold wave samples, generate cold wave samples to expand the number of cold wave samples; combined with an improved U-net classification model, achieve accurate identification of future cold wave weather and effective prediction of the cold wave occurrence time; 4) Classification optimization method for cold wave continuous events; 5) Based on numerical weather prediction data and cold wave identification results, construct a cold wave identification and wind power prediction model considering the characteristics of sample scarcity and weather cycle characteristics.

[0048] The specific embodiment content is as follows: Apply a wind farm cluster in a certain region of China, with 67 wind farms, a total installed capacity of 5388.9 MW, and the data length from January 1, 2021 to December 30, 2022. The data includes: measured power, 100m wind speed, 10m wind speed, temperature, dew point temperature, etc. of NWP data, with a time interval of 1 hour. The training sets for the cold wave occurrence season are from January to February and from November to December in 2021, and the test sets are from November to December in 2022; the training sets for the non-cold wave occurrence season are from March to October in 2021, and the test sets are from September to October in 2022.

[0049] Step 1: Construct a cold wave discrimination criterion. As Figure 2 shown, cold wave weather is usually accompanied by a sharp drop in temperature, but the change in wind speed is more complex, and it is difficult to accurately predict wind power through wind speed. When a cold wave occurs, when the wind speed increases, the wind speed shows an upward trend, but the power shows a downward trend. This wind speed change process will have a significant impact on the prediction accuracy of wind power. According to meteorological principles and the operating characteristics of wind turbines, construct a cold wave judgment criterion; Step 2: Analysis of the correlation between cold wave meteorological characteristics and wind power. Figure 3The power and wind speed, temperature, and dew point temperature curves for all cold wave periods are shown in Figure 2. In order to deeply explore the key factors affecting the occurrence of cold waves, this solution conducted a Pearson correlation coefficient analysis on the relevant meteorological data during the cold wave weather period to reveal the relationship between each meteorological variable and wind power during the cold wave. When a cold wave occurs, the power has the highest correlation with 100m wind speed, 10m wind speed, daily minimum temperature difference, and hourly temperature difference.

[0050] Step 3: Cold wave sample expansion. Cold wave weather is a small sample event, accounting for only about 3.8% of the total data in two years. This sample imbalance has a significant impact on the training process and classification performance of the model, especially in the classification accuracy of cold wave data, which is prone to high overall classification accuracy due to insufficient samples. The number of cold wave samples in the test set is 137, and the recall rate results of cold wave samples calculated by different models are shown in Table 1. The recall rate reflects the model's ability to recognize real cold wave weather. After using generative adversarial networks (GANs) to generate 5,000 cold wave samples and mixing the generated cold wave samples with the original training set data, the total number of samples in the training set was expanded to 22,520. At the same time, the test set data still remains composed of original data, and the number of test set samples is 3,504. Using data generation technology to expand cold wave samples not only enhances the classification model's ability to learn cold wave characteristics, but also significantly improves its recognition effect on cold wave weather. Table 1 Recall rates of cold wave identification without and with data generation , Step 4: Classification optimization method for cold wave continuity events. The specific content of the optimization method is as follows: Figure 4 As shown in the figure, in view of the particularity of short-term cold wave events, in order to improve the accuracy of future cold wave predictions, it is necessary to continuously optimize the judgment of cold wave occurrence. Through the first optimization, short-term fluctuations that may cause misjudgment are merged into the continuous cold wave period, and the second optimization reasonably merges isolated short-term cold wave events with normal weather to avoid misclassification of cold wave events and optimize the classification results.

[0051] Step 5: Construct a cold wave identification and wind power prediction model that considers the characteristics of sample scarcity and weather cycle. First, obtain the corrected cold wave weather identification results, divide the prediction period into cold wave seasons and non-cold wave seasons, and establish prediction models for the two season types. In the season without cold waves, Non-Stationary Transformer is used to directly predict. In the season with cold waves, the identified cold wave forecast information needs to be input into the prediction model. The prediction model can effectively predict the power when the cold wave occurs.

[0052] Use LSTM, random forest, BP, Transformer and the method proposed in this solution to predict two types of seasonal data respectively. The prediction results are as Figure 5 、 Figure 6 shown, and the prediction result metrics are shown in Table 2. In the season without cold snaps, the Ns-Transformer model adopted in this solution achieved a reduction of 3.82%, 4.60%, 7.37%, and 0.39% respectively in terms of the RMSE (root mean square error) metric compared with the LSTM, BP, random forest, and Transformer prediction models. In the non-cold snap season, due to relatively small weather fluctuations, both Transformer and Ns-Transformer can effectively predict the changing trend of power based on the input features. For the prediction results in the season with cold snaps, the performance of the LSTM, random forest, Transformer, and the prediction model of the method proposed in this solution all decreased in the cold snap season. The RMSE in the cold snap season increased by 3.98%, 3.33%, 2.40%, and 2.35% respectively compared with the season without cold snaps. The occurrence of cold snap weather not only brings drastic fluctuations to the wind power, but also reduces the correlation between features such as wind speed and power, thus reducing the prediction accuracy. Through comparative analysis, it can be found that the method proposed in this solution can still maintain good prediction accuracy under cold snap weather conditions, and the increase in its RMSE is significantly lower than that of other prediction methods, making it more suitable for wind power prediction under cold snap weather.

[0053] Table 2 Evaluation metrics of prediction results of different models , The various embodiments in this specification are described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the various embodiments can be referred to each other.

[0054] Specific examples are used in this article to elaborate on the principles and implementation manners of the present invention. The descriptions of the above embodiments are only used to help understand the core idea of the present invention; at the same time, for those of ordinary skill in the art, based on the idea of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A wind power prediction method based on cold wave identification, characterized in that, Including: Obtain the meteorological data and wind turbine operation data of the target area, and construct a cold wave discrimination criterion; Generate cold wave samples based on the cold wave discrimination criterion and the sample enhancement method; The cold wave samples include meteorological data with time characteristics and cold wave characteristics; The sample enhancement method is an enhanced classification effect method using generative adversarial network technology; Construct a cold wave weather recognition model based on the improved U-net and a classification optimization method based on cold wave continuous events; The cold wave weather recognition model is obtained by training the improved U-net with the cold wave samples; Use the cold wave weather recognition model and the classification optimization method to identify cold waves in the meteorological data to be measured, and obtain the cold wave recognition results for the future prediction period; the cold wave recognition results include the cold wave occurrence stage and the non-cold wave occurrence stage; Construct a wind power combined prediction model; the wind power combined prediction model includes a non-cold wave prediction model and a cold wave prediction model; the wind power combined prediction model is constructed based on a combination model of long short-term memory network and non-stationary Transformer; wherein, the non-stationary Transformer combination model includes a non-stationary Transformer data embedding module and a non-stationary Transformer encoder connected in sequence; the output ends of the long short-term memory network and the non-stationary Transformer data embedding module are both connected to the input end of the non-stationary Transformer encoder; According to the cold wave recognition results of the future prediction period, input the corresponding data into the corresponding prediction model, and combine the output results of the two models to obtain the complete wind power prediction result.

2. The wind power prediction method based on cold snap identification according to claim 1, characterized in that The cold wave discrimination criterion is specifically expressed as: ， Among them, is the daily minimum temperature, is the daily minimum temperature, is the normalized predicted data, is the normalized actual power data, is the total installed capacity of the wind power cluster; is the temperature difference; when all conditions are met, it is determined that a cold wave weather occurs at this time.

3. The wind power prediction method based on cold snap identification according to claim 1, characterized in that The structure of the improved U-net is that in the first-layer down-convolution operation, by changing the input channels, the feature data and the context features are used as data inputs at the same time; the feature data is the given data; the context features are data features constructed from relevant information; the feature data and the context features have the same time step.

4. The wind power prediction method based on cold snap identification according to claim 1, wherein The classification optimization method based on cold wave continuous events is specifically expressed as: ， Among them, is the original cold snap classification result, is the optimized cold snap classification result, is the duration of the current normal weather event, is the duration of the current cold snap event, is the neighborhood merging threshold, is the isolated determination threshold, is the classification result sequence, .

5. The wind power prediction method based on cold snap identification according to claim 1, wherein Before using the wind power combined prediction model for prediction, it also includes: evaluating the trained wind power combined prediction model; the evaluation process is to calculate the error between the predicted power calculated by the model and the measured power through the error evaluation criteria; the error evaluation criteria include root mean square error and mean absolute error.

6. A wind power prediction system based on cold snap identification, characterized in that Including: A cold wave discrimination criterion establishment unit, configured to obtain the meteorological data and wind turbine operation data of the target area, and construct a cold wave discrimination criterion; A cold wave sample generation unit, configured to generate cold wave samples based on the cold wave discrimination criterion and the sample enhancement method; the cold wave samples include meteorological data with time characteristics and cold wave characteristics; The sample enhancement method is an enhanced classification effect method using generative adversarial network technology; A cold wave classification method construction unit, configured to construct a cold wave weather recognition model based on the improved U-net and a classification optimization method based on cold wave continuous events; The cold snap weather recognition model is obtained by training an improved U-net using the cold snap samples; A cold snap identification unit, configured to use the cold snap weather recognition model and the classification optimization method to identify cold snaps in the measured meteorological data, so as to obtain a cold snap recognition result for a future prediction period; the cold snap recognition result includes a cold snap occurrence stage and a non-cold snap occurrence stage; A wind power prediction model construction unit, configured to construct a combined wind power prediction model; the combined wind power prediction model includes a non-cold snap prediction model and a cold snap prediction model; the combined wind power prediction model is constructed based on a combination model of a long short-term memory network and a non-stationary Transformer; wherein, the non-stationary Transformer combination model includes a non-stationary Transformer data embedding module and a non-stationary Transformer encoder connected in sequence; the output ends of the long short-term memory network and the non-stationary Transformer data embedding module are both connected to the input end of the non-stationary Transformer encoder; A power prediction unit, configured to input corresponding data into the corresponding prediction model according to the cold snap recognition result for the future prediction period, and combine the output results of the two models to obtain a complete wind power prediction result.

7. An electronic device, characterized in that, It includes a memory and a processor, the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the wind power prediction method based on cold snap identification according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, It stores a computer program, and when the computer program is executed by the processor, it implements the wind power prediction method based on cold snap identification according to any one of claims 1-5.

Images