Ultra-short-term wind speed correction forecasting method and device based on neural network and numerical mode fusion and storage medium

Abstract

The invention provides an ultra-short-term wind speed correction forecasting method and device based on neural network and numerical mode fusion and a storage medium. The method comprises the following steps: training a neural network model by using historical wind speed prediction deviation; inputting the wind speed prediction deviation at the current moment into the trained neural network model, and predicting to obtain the wind speed deviation at the next moment; performing data assimilation operation on the wind speed forecast result at the current moment output by the WRF model and the wind speed observation value at the current moment, and performing prediction by using the WRF model after the data assimilation operation to obtain a wind speed prediction result at the next moment; and correcting a wind speed prediction result at the next moment by using the wind speed deviation at the next moment obtained by prediction to obtain a wind speed prediction result at the next moment. According to the method, the current meteorological effect and real-time observation change which are not fully considered by the model can be captured while historical forecast error characteristics are analyzed, so that the prediction precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorological data and new energy power prediction, and in particular to an ultra-short-term wind speed correction forecast method, device and storage medium based on the fusion of neural network and numerical model. Background Art

[0002] In recent years, with the continuous increase in the scale of renewable energy power generation, the prediction model for renewable energy power is also developing rapidly. The prediction of wind power is inseparable from the prediction of wind speed. At present, few meteorological prediction models can fully and accurately capture all the complex details and real-time changes of atmospheric movement, especially those that are greatly affected by local terrain, sudden changes in meteorological conditions or small-scale effects. The wind speed correction process adjusts the predicted value by combining the latest observation data, field measurement feedback and machine learning algorithms, thereby improving the accuracy and practicality of the prediction, providing more reliable and timely wind speed information support for wind power operation, power grid dispatching and other fields, and ultimately achieving the purpose of optimizing decision-making and resource management.

[0003] At present, meteorological data is mainly predicted through traditional numerical models such as WRF (Weather Research and Forecasting Model). The quality of the initial field data of the WRF model has a great impact on the prediction accuracy and cannot accurately reflect the meteorological changes in small-scale areas. Wind speed forecasts that have not undergone data correction often fail to accurately consider local environmental factors, real-time meteorological changes and inherent model deviations, resulting in a large deviation between the prediction results and the actual situation. Incorrect estimation of power generation causes an imbalance in energy allocation, increases the difficulty of grid dispatch, and directly affects the operating efficiency of wind farms.

[0004] In summary, there is an urgent need for an ultra-short-term wind speed correction forecast method, device and storage medium based on the fusion of neural network and numerical model to solve the problems existing in the prior art. Summary of the invention

[0005] The purpose of the present invention is to provide an ultra-short-term wind speed correction forecast method based on the fusion of neural network and numerical model, aiming to solve the problem that the wind speed prediction result has a large deviation from the actual situation due to local environmental factors, real-time meteorological changes and inherent deviation of the model. The specific technical scheme is as follows: An ultra-short-term wind speed correction forecast method based on the fusion of neural network and numerical model, comprising: The neural network model is trained using historical wind speed forecast deviations; The wind speed prediction deviation at the current moment is input into the trained neural network model to predict the wind speed deviation at the next moment; The wind speed forecast result at the current moment output by the WRF model is assimilated with the wind speed observation value at the current moment, and the wind speed forecast result at the next moment is obtained by using the WRF model after the data assimilation operation; The wind speed deviation at the next moment obtained by prediction is used to correct the wind speed prediction result at the next moment, so as to obtain the wind speed forecast result at the next moment; This also includes periodic updates to the neural network model, specifically: Each interval T The loss between the wind speed deviation predicted by the neural network model and the wind speed prediction deviation at the corresponding moment is calculated, and the parameters of the neural network model are updated using the gradient descent method; the momentum vector is introduced in the gradient descent process and Dynamic Momentum Factor ,in, is the static momentum factor, is the sudden weather noise; assuming that the current parameters of the neural network model are , the learning rate is , the gradient is , update the momentum vector according to formula (6), update the parameters according to formula (7), and calculate the sudden weather noise according to formula (8): (6), (7), (8), in, express Weather data at the moment Similar distance between weather data at each moment; express Weather data at the moment, express Weather data at the moment.

[0006] In the above technical solution, preferably, the neural network model is trained using the historical wind speed prediction deviation, specifically: Find historical weather data similar to the current weather conditions based on similar distances; Based on the wind speed observations from historical weather data and the wind speed forecast results output by the WRF model, the historical wind speed forecast deviation is obtained; The historical wind speed prediction deviation is input into the neural network model to complete the training of the neural network model.

[0007] Preferably, in the above technical solution, the similarity distance is calculated as follows: (1), in: For the current moment Weather data, For historical moments Weather data, Similarity Factor In the time window Weather data at the moment, Similarity Factor In the time window Weather data at the moment, is half of the effective impact time window, Similarity Factor The weight of is the number of similar factors, is the similarity factor in historical weather data The standard deviation of the time series.

[0008] After calculating the similarity distance according to formula (1), take the one with the highest similarity to the current weather conditions. K Historical samples are used for neural network model training. K An integer greater than or equal to 1000.

[0009] Preferably, in the above technical solution, the wind speed prediction deviation is calculated according to formula (2): (2), in, is the observed wind speed, is the wind speed forecast result output by the WRF model, is the wind speed forecast deviation.

[0010] In the above technical solution, the neural network model uses the past n The wind speed forecast deviation of time steps is used to predict the future m The wind speed deviation of time steps is calculated by cross-validation method for the historical wind speed prediction deviation data set. The prediction process is expressed as: (3), in: express The wind speed deviation at the time, Represents the model mapping relationship, express The wind speed forecast deviation at the time, k represents the number of time steps for prediction, ,in n is an integer greater than or equal to 12, m is an integer greater than or equal to 4.

[0011] The preferred technical solution above is that the interval timeT = n × M , M is a natural number greater than or equal to 1.

[0012] In the above technical solution, preferably, the wind speed deviation at the next moment obtained by prediction is used to correct the wind speed prediction result at the next moment to obtain the wind speed forecast result at the next moment, specifically: (5), in, is the wind speed prediction result of the WRF model, is the wind speed deviation predicted by the neural network model, It is the corrected wind speed forecast result.

[0013] The present invention also provides an ultra-short-term wind speed correction forecasting device based on the fusion of neural network and numerical model, the device adopts the ultra-short-term wind speed correction forecasting method based on the fusion of neural network and numerical model, and the device comprises: Sample acquisition module: obtains wind speed prediction deviation for training neural network model; Sample training module: used to complete the training of neural network model; Deviation prediction and update module: Use the trained neural network model to predict the wind speed deviation at the next moment; at the same time, periodically update the parameters of the neural network model; Wind speed prediction module: Perform data assimilation on the wind speed forecast result at the current moment output by the WRF model and the wind speed observation value at the current moment, and use the WRF model after the data assimilation operation to predict the wind speed forecast result at the next moment; Wind speed correction module: Use the predicted wind speed deviation at the next moment to correct the wind speed prediction result at the next moment to obtain the wind speed forecast result at the next moment.

[0014] The present invention also provides an ultra-short-term wind speed correction forecasting device based on the fusion of neural network and numerical model, the device includes a memory and a processor, wherein the memory stores the ultra-short-term wind speed correction forecasting method based on the fusion of neural network and numerical model, and the processor is used to load and execute the ultra-short-term wind speed correction forecasting method based on the fusion of neural network and numerical model.

[0015] The present invention also provides a storage medium, in which the ultra-short-term wind speed correction forecast method based on the fusion of neural network and numerical model is stored.

[0016] The application of the technical solution of the present invention has the following beneficial effects: During the wind speed correction process, the present invention continuously adjusts the predicted value by combining the latest wind speed observation data, field measurement feedback and machine learning algorithm parameter updates. It aims to analyze the historical forecast error characteristics while capturing the current meteorological effects and real-time observation changes that are not fully considered by the model, thereby improving the prediction accuracy. It has an objective effect on improving the accuracy of ultra-short-term wind speed forecasts in wind farms.

[0017] The present invention utilizes historical wind speed prediction deviations under similar weather conditions to train a neural network model, and predicts the wind speed deviation at the next moment through the trained neural network model, so as to analyze the characteristics of historical forecast errors; through data assimilation and periodic updating of the parameters of the neural network model, local meteorological effects and instantaneous changes that are not fully considered by the model are regularly captured, thereby realizing the correction and update of the WRF model and the neural network model, ensuring the accuracy of the model prediction data, and providing an accurate data basis for wind speed correction, thereby improving the prediction accuracy and ensuring the effective management of wind energy and the stability of power grid dispatching.

[0018] In addition to the above-described purposes, features and advantages, the present invention has other purposes, features and advantages. The present invention will be further described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings constituting a part of this application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings: Figure 1 It is a flow chart of the ultra-short-term wind speed correction forecast method based on the fusion of neural network and numerical model of the present invention; Figure 2 It is a structural schematic diagram of the neural network model in the present invention. DETAILED DESCRIPTION

[0020] In order to facilitate the understanding of the present invention, the present invention will be described more fully below, and preferred embodiments of the present invention are provided. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present invention more thorough and comprehensive.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0022] Example: See also Figure 1This embodiment provides an ultra-short-term wind speed correction forecast method based on the fusion of neural network and numerical model, comprising the following steps: S1. Use historical wind speed forecast deviations to train the neural network model; S2, inputting the wind speed prediction deviation at the current moment into the trained neural network model to predict the wind speed deviation at the next moment; S3, performing data assimilation operation on the wind speed forecast result at the current moment output by the WRF model and the wind speed observation value at the current moment, and using the WRF model after the data assimilation operation to predict the wind speed forecast result at the next moment; S4, using the predicted wind speed deviation at the next moment to correct the wind speed prediction result at the next moment, to obtain the wind speed forecast result at the next moment; S5. Periodically update the neural network model, and repeat steps S2 to S4 to continuously output wind speed forecast results.

[0023] Specifically, the WRF model refers to the Weather Research and Forecasting Model, that is, the WRF weather forecast model.

[0024] Preferably, in step S1, the neural network model is trained using historical wind speed prediction deviations, specifically: S1.1. Find historical weather data similar to the current weather conditions based on similar distances; Specifically, the similarity distance refers to the similarity distance between the historical weather conditions and the current weather conditions in the same space within a specific time.

[0025] The weather conditions at a certain time can be expressed by the weighted average of multiple similarity factors. The selection of similarity factors and their weights can be analyzed by season. The correlation between the similarity factors of the full-field forecast of ECMWF (European Centre for Medium-Range Weather Forecasts) and the full-field measured wind speed, the similarity factors include 100-meter wind speed (W_10), 10-meter wind speed (W_10), temperature (T), relative humidity (RH) and air pressure (P), etc.; select the 5 most relevant similarity factors under each seasonal working condition of spring, summer, autumn and winter, and assign the weight of each similarity factor according to the correlation; specifically, the predicted wind speed is absolutely correlated with the measured wind speed, and the weight is 1; the weights of the remaining 4 similarity factors are assigned according to their correlation with the measured wind speed. For example: the remaining 4 similarity factors are sorted from high to low according to the results of the correlation analysis, and the corresponding weight values ​​are 0.3, 0.2, 0.1, and 0.05 respectively.

[0026] Furthermore, the similarity distance is calculated as follows: (1), in: For the current moment Weather data, For historical moments Weather data, Similarity Factor In the time window Weather data at the moment, Similarity Factor In the time window Weather data at the moment, is half of the effective impact time window, Similarity Factor The weight of is the number of similar factors, is the similarity factor in historical weather data The standard deviation of the time series.

[0027] After calculating the similarity distance according to formula (1), take the one with the highest similarity to the current weather conditions. K Historical samples are used for subsequent neural network model training. K is an integer greater than or equal to 1000; specifically, in this embodiment, K =2000, set the time window length to 8 hours.

[0028] Furthermore, since there are significant differences in weather conditions in spring, summer, autumn and winter, when executing the method of this embodiment, historical samples with high similarity should be selected for spring, summer, autumn and winter respectively, and neural network models should be trained for spring, summer, autumn and winter respectively.

[0029] S1.2, based on the wind speed observations of historical weather data and the wind speed forecast results output by the WRF model, obtain the historical wind speed forecast deviation; Specifically, the wind speed prediction deviation is calculated according to formula (2): (2), in, is the observed wind speed, is the wind speed forecast result output by the WRF model, is the wind speed forecast deviation.

[0030] S1.3. Input the historical wind speed prediction deviation into the neural network model to complete the training of the neural network model.

[0031] Specifically, Figure 2As shown, the neural network model includes a recurrent neural network (RNN) and a one-dimensional convolutional neural network (CNN). First, the CNN is used to extract features from the input time series, and then the RNN network is used to predict the features extracted by the CNN. n The wind speed forecast deviation of time steps is used to predict the future m The wind speed deviation of a time step is calculated by cross-validation method for the historical wind speed prediction deviation data set under a single seasonal condition. The prediction process is expressed as: (3), in: express The wind speed deviation at the time, Represents the model mapping relationship, express The wind speed forecast deviation at the time, k represents the number of time steps for prediction, ,in n is an integer greater than or equal to 12, m is an integer greater than or equal to 4.

[0032] Preferably, in this embodiment, a single time step is set to 30 minutes.

[0033] Specifically, the training process of the neural network model includes: Data normalization: In order to reduce the impact of singular values ​​in the input data on the prediction results and accelerate the convergence of the loss function, it is necessary to normalize the input wind speed prediction deviation data. The specific formula is: (4), in, and Represents the maximum and minimum values ​​in the wind speed forecast deviation data series, Indicates the wind speed forecast deviation of the current input.

[0034] Model training: reconstruct the normalized data and perform model training. Specifically, first input it into CNN to extract features. The number of input nodes is consistent with the number of input data. n nodes; then input into RNN, the number of input nodes is consistent with the number of CNN output nodes, which is n nodes, the number of nodes in the output layer is consistent with the prediction step size, which is m nodes.

[0035] Specifically, in the neural network model of this embodiment, the CNN input is 12 nodes, the output is 12 nodes, the convolution kernel size is 3, and the activation function is tanh; the RNN input is 12 nodes, the intermediate layer activation function is tanh, the output layer is 4 nodes, the activation function of the output layer is a linear function, the loss function is MSE (root mean square error), the learning rate is 0.001, the maximum number of iterations is 200, and the optimization method is gradient descent.

[0036] Preferably, in step S2, the wind speed prediction deviation at the current moment is input into the trained neural network model to predict the wind speed deviation at the next moment, specifically: will pass n The wind speed prediction deviation of time steps is input into the trained neural network model, and the future wind speed prediction is obtained through the neural network model. m The wind speed deviation of each time step is calculated and the prediction results output by the neural network model are denormalized.

[0037] Furthermore, since there are significant differences in weather conditions in spring, summer, autumn and winter, wind speed deviation prediction needs to be performed using corresponding neural network models in spring, summer, autumn and winter respectively.

[0038] Preferably, in step S3, the wind speed forecast result at the current moment output by the WRF model is subjected to data assimilation operation with the wind speed observation value at the current moment, and the wind speed forecast result at the next moment is obtained by using the WRF model after the data assimilation operation, specifically: After the WRF model is run, the wrfinput file is obtained. According to the current wind speed observation value, the wrfda data assimilation operation is performed on wrfinput, where the time window is set to 2 hours. If the current time is 2024-09-01_12:00:00, the parameters of the data assimilation namelist are set to: time_window_min=2024-09-01_11:00:00 time_window_max=2024-09-01_13:00:00 Get the wrfvar_output (updated wrfinput) of the fused wind speed observation value, and then perform the wrf operation again to get the updated wrfout file (which contains the wind speed prediction results).

[0039] Preferably, in step S4, the wind speed deviation at the next moment obtained by prediction is used to correct the wind speed prediction result at the next moment to obtain the wind speed forecast result at the next moment, specifically: (5), in, is the wind speed prediction result of the WRF model, is the wind speed deviation predicted by the neural network model, It is the corrected wind speed forecast result.

[0040] According to formula (5), the wind speed forecast results output by the WRF model can be corrected. Through correction, the final output wind speed forecast results are closer to the actual wind speed observation values, greatly improving the accuracy of the wind speed forecast results.

[0041] Preferably, in step S5, the neural network model is periodically updated, specifically: Each interval T The loss between the wind speed deviation predicted by the neural network model and the wind speed prediction deviation at the corresponding moment is calculated, and the parameters of the neural network model are updated using the gradient descent method; the momentum vector is introduced in the gradient descent process and Dynamic Momentum Factor , assuming that the current parameters of the neural network model are , the learning rate is , the gradient is , update the momentum vector according to formula (6), update the parameters according to formula (7), and calculate the sudden weather noise according to formula (8): (6), (7), (8), in, is the static momentum factor, For sudden weather noise, express Weather data at the moment The similarity distance between the weather data at the time is calculated by referring to formula (1); express Weather data at the moment, express Weather data at the moment.

[0042] In this embodiment, the possibility of sudden weather changes is taken into account during the update of the neural network model parameters. The dynamic momentum factor is the static momentum factor. Subtract sudden weather noise , sudden weather noise The larger the value, the more severe the weather mutation is, and vice versa. When the weather mutation is severe, the larger the similarity distance between the weather data before and after (i.e., the mutation weather noise The larger the dynamic momentum factor The smaller the distance, the more dependent the parameter update is on the current weather, that is, the current weather has a greater impact on the future weather. When the weather does not change suddenly or the change is not obvious, the similarity distance between the weather data before and after is smaller (that is, the sudden change weather noise The smaller the dynamic momentum factor The larger the value is, the more dependent the parameter update is on past weather. This embodiment introduces a dynamic momentum factor to adjust the parameter update scheme according to sudden weather changes, making short-term weather forecasts more accurate.

[0043] Preferably, the interval time T = n × M , M is a natural number greater than or equal to 1; that is, each interval M × n The parameters of the neural network model are updated once every time step (when M When the value is 1, it means that the parameters of the neural network model are updated for each prediction), so that the parameters of the neural network model can be corrected during the prediction process to ensure the accuracy of the neural network model's prediction of wind speed deviation.

[0044] Preferably, the static momentum factor is a value close to 1. In this embodiment, the static momentum factor Set to 0.9.

[0045] This embodiment also provides an ultra-short-term wind speed correction forecast device based on the fusion of neural network and numerical model, the device comprising: Sample acquisition module: obtains wind speed prediction deviation for training neural network model; Sample training module: used to complete the training of neural network model; Deviation prediction and update module: Use the trained neural network model to predict the wind speed deviation at the next moment; at the same time, periodically update the parameters of the neural network model; Wind speed prediction module: The wind speed forecast result at the current moment output by the WRF model is assimilated with the wind speed observation value at the current moment, and the wind speed forecast result at the next moment is obtained by using the WRF model after the data assimilation operation; Wind speed correction module: Use the predicted wind speed deviation at the next moment to correct the wind speed prediction result at the next moment to obtain the wind speed forecast result at the next moment.

[0046] Preferably, in some embodiments, the ultra-short-term wind speed correction forecasting device based on the fusion of neural network and numerical model may include a memory and a processor, wherein the memory stores the ultra-short-term wind speed correction forecasting method based on the fusion of neural network and numerical model, and the processor is used to load and execute the ultra-short-term wind speed correction forecasting method based on the fusion of neural network and numerical model. Further, the memory is a storage medium.

[0047] Applying the technical solution of the present invention has the following effects: The present invention utilizes historical wind speed prediction deviations under similar weather conditions to train a neural network model, and predicts the wind speed deviation at the next moment through the trained neural network model, so as to analyze the characteristics of historical forecast errors; through data assimilation and periodic updating of the parameters of the neural network model, local meteorological effects and instantaneous changes that are not fully considered by the model are regularly captured, thereby realizing the correction and update of the WRF model and the neural network model, ensuring the accuracy of the model prediction data, and providing an accurate data basis for wind speed correction, thereby improving the prediction accuracy and ensuring the effective management of wind energy and the stability of power grid dispatching.

[0048] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for correcting and forecasting ultra-short-term wind speed based on the fusion of neural network and numerical model, characterized in that: include: The neural network model is trained using historical wind speed forecast deviations; The wind speed prediction deviation at the current moment is input into the trained neural network model to predict the wind speed deviation at the next moment; The wind speed forecast result at the current moment output by the WRF model is assimilated with the wind speed observation value at the current moment, and the wind speed forecast result at the next moment is obtained by using the WRF model after the data assimilation operation; The wind speed deviation at the next moment obtained by prediction is used to correct the wind speed prediction result at the next moment, so as to obtain the wind speed forecast result at the next moment; This also includes periodic updates to the neural network model, specifically: Each interval T The loss between the wind speed deviation predicted by the neural network model and the wind speed prediction deviation at the corresponding moment is calculated, and the parameters of the neural network model are updated using the gradient descent method; the momentum vector is introduced in the gradient descent process and Dynamic Momentum Factor ,in, is the static momentum factor, is the sudden weather noise; assuming that the current parameters of the neural network model are , the learning rate is , the gradient is , update the momentum vector according to formula (6), update the parameters according to formula (7), and calculate the sudden weather noise according to formula (8): （6）， （7）， （8）， in, express Weather data at the moment Similar distance between weather data at each moment; express Weather data at the moment, express Weather data at the moment.

2. The ultra-short-term wind speed correction forecast method based on the fusion of neural network and numerical model according to claim 1 is characterized in that: The neural network model is trained using historical wind speed forecast deviations, specifically: Find historical weather data similar to the current weather conditions based on similar distances; Based on the wind speed observations from historical weather data and the wind speed forecast results output by the WRF model, the historical wind speed forecast deviation is obtained; The historical wind speed prediction deviation is input into the neural network model to complete the training of the neural network model.

3. The ultra-short-term wind speed correction forecast method based on the fusion of neural network and numerical model according to claim 2 is characterized in that: The similarity distance is calculated as follows: （1）， in: For the current moment Weather data, For historical moments Weather data, Similarity Factor In the time window Weather data at the moment, Similarity Factor In the time window Weather data at the moment, is half of the effective impact time window, Similarity Factor The weight of is the number of similar factors, is the similarity factor in historical weather data The time series standard deviation of After calculating the similarity distance according to formula (1), take the one with the highest similarity to the current weather conditions. K Historical samples are used for neural network model training. K An integer greater than or equal to 1000.

4. The ultra-short-term wind speed correction forecast method based on the fusion of neural network and numerical model according to claim 2 is characterized in that: The wind speed prediction deviation is calculated according to formula (2): （2）， in, is the observed wind speed, is the wind speed forecast result output by the WRF model, is the wind speed prediction bias.

5. The ultra-short-term wind speed correction forecast method based on the fusion of neural network and numerical model according to claim 1 is characterized in that: The neural network model utilizes the past n The wind speed forecast deviation of time steps is used to predict the future m The wind speed deviation of time steps is calculated by cross-validation method for the historical wind speed prediction deviation data set. The prediction process is expressed as: （3）， in: express The wind speed deviation at the time, Represents the model mapping relationship, express The wind speed forecast deviation at the time, k represents the number of time steps for prediction, ,in n is an integer greater than or equal to 12, m is an integer greater than or equal to 4.

6. The ultra-short-term wind speed correction forecast method based on the fusion of neural network and numerical model according to claim 5 is characterized in that: Interval time T = n × M , M is a natural number greater than or equal to 1.

7. The ultra-short-term wind speed correction forecast method based on the fusion of neural network and numerical model according to claim 1 is characterized in that: The wind speed deviation at the next moment is used to correct the wind speed forecast result at the next moment, and the wind speed forecast result at the next moment is obtained, specifically: （5）， in, is the wind speed prediction result of the WRF model, is the wind speed deviation predicted by the neural network model, It is the corrected wind speed forecast result.

8. An ultra-short-term wind speed correction forecasting device based on the fusion of neural network and numerical model, characterized in that: The device adopts the ultra-short-term wind speed correction forecast method based on the fusion of neural network and numerical model as described in any one of claims 1 to 7, and the device comprises: Sample acquisition module: obtains wind speed prediction deviation for training neural network model; Sample training module: used to complete the training of neural network model; Deviation prediction and update module: Use the trained neural network model to predict the wind speed deviation at the next moment; at the same time, periodically update the parameters of the neural network model; Wind speed prediction module: Perform data assimilation on the wind speed forecast result at the current moment output by the WRF model and the wind speed observation value at the current moment, and use the WRF model after the data assimilation operation to predict the wind speed forecast result at the next moment; Wind speed correction module: Use the predicted wind speed deviation at the next moment to correct the wind speed prediction result at the next moment to obtain the wind speed forecast result at the next moment.

9. An ultra-short-term wind speed correction forecasting device based on the fusion of neural network and numerical model, characterized in that: The device includes a memory and a processor, wherein the memory stores an ultra-short-term wind speed correction forecast method based on the fusion of a neural network and a numerical model as described in any one of claims 1-7, and the processor is used to load and execute the ultra-short-term wind speed correction forecast method based on the fusion of a neural network and a numerical model as described in any one of claims 1-7.

10. A storage medium, characterized in that: The storage medium stores the ultra-short-term wind speed correction forecast method based on the fusion of neural network and numerical model as described in any one of claims 1-7.

Images