Regional wind power short-term power prediction method and system based on fish swarm algorithm improved prediction large model

Abstract

The invention provides a regional wind power short-term power prediction method and system based on a fish swarm algorithm improved prediction large model, and the method comprises the steps: collecting historical data, and constructing a historical data set; a multi-mode module is added to the front end of an embedded layer of a meteorological large model, a regression module is added to an output result of the meteorological large model, and the output result of the meteorological large model is changed from outputting different meteorological elements to directly outputting regional wind power short-term power prediction; pre-training the meteorological large model added with the multi-modal module and the regression module by using a historical data set, and performing parameter adjustment through a fish swarm algorithm to obtain a power prediction large model; and performing regional wind power short-term power prediction by using the large power prediction model. According to the method, overall prediction is carried out aiming at regional wind power conditions in multiple scenes, and the complex relationship between weather and power data in the region is disclosed in combination with a current relatively advanced large model technology, so that the regional wind power in multiple complex weather scenes is predicted.

Description

Technical Field

[0001] The present invention belongs to the technical field of renewable energy power generation prediction, and specifically relates to a method and system for regional wind power short-term power prediction based on a fish school algorithm to improve a large prediction model. Background Art

[0002] With the vigorous development of renewable energy power generation, the grid-connected capacity of wind farms has increased rapidly. Accurately and effectively predicting the output of wind farms in the region can assist power system dispatchers in making accurate and effective decisions and maintaining the balance of the power grid. In the short-term prediction of wind power regions, the methods used mainly include combining the information of all stations in the region and selecting model stations with strong output correlation in the region. Based on the site numerical weather forecast of the model station or the grid numerical weather forecast in the region, as well as the power generation of each model station, a regional upscaling power prediction method is performed to achieve the result of predicting the short-term wind power prediction of the entire region. There are many methods for regional upscaling power prediction. Currently, the commonly used algorithms include traditional statistical methods, machine learning methods, and deep learning methods. Comparing different upscaling methods, the upper limit of the short-term prediction accuracy (root mean square accuracy) of wind power regions is about 93%-95%.

[0003] Existing technologies include statistical methods, machine learning methods, and deep learning methods. Their technical disadvantages include: 1) Statistical methods: Traditional statistical methods can reveal the relationship between data to a certain extent. Through curve fitting, a relatively accurate power prediction curve can be produced on days with high accuracy of meteorological data forecasts. However, it has a greater impact on the prediction accuracy in complex weather and complex data relationships. Statistical methods are highly dependent on data and can only reveal relatively simple data relationships between data; 2) Machine learning methods: Traditional machine learning methods can learn complex relationships in data and can be applied to some nonlinear problems. For results with less data, the prediction accuracy is still quite impressive. However, machine learning is still highly dependent on data and is not capable of processing complex environments with variable scales and unstructured data. 3) Deep learning methods: Deep learning methods have a high model complexity and can effectively reveal complex data structures and relationships. However, deep learning algorithms take a long time to calculate and require high computing resources. At the same time, deep learning algorithms are relatively poor in interpretability and cannot provide a reasonable explanation for the conclusions drawn.

[0004] Therefore, a more accurate and comprehensive method is needed to predict wind power under various complex weather scenarios in the region. Summary of the invention

[0005] The purpose of the present invention is to provide a regional wind power short-term power forecasting method and system based on the fish school algorithm to improve the prediction large model, to make an overall forecast of the regional wind power situation under multiple scenarios, and to combine the currently more advanced large model technology to generally reveal the complex relationship between meteorological and power data in the region, so as to predict the regional wind power under a variety of complex weather scenarios.

[0006] In order to achieve the above object, the technical solution of the present invention is as follows: A regional wind power short-term power forecasting method based on a fish school algorithm to improve the large forecasting model, comprising: S1. Collect historical data, including weather forecast grid data, satellite cloud image data, radar inversion data and actual data from each wind farm station, and construct a historical data set; S2. Add a multimodal module to the front end of the embedding layer of the meteorological model to analyze multimodal data including meteorological forecast grid data, satellite cloud image data, radar inversion data and actual data from each wind farm station; add a regression module to the output results of the meteorological model to change the output results of the meteorological model from outputting different meteorological elements to directly outputting regional wind power short-term power forecast; S3. Pre-train the meteorological large model with added multimodal module and regression module using historical data sets, adjust parameters through fish school algorithm, and obtain the power prediction large model; S4. Use a large power prediction model to perform short-term regional wind power prediction.

[0007] Furthermore, in step S2, the multimodal module adopts a processing model of an artificial intelligence visual algorithm; and the regression module adopts a linear regression layer.

[0008] Furthermore, the method for adjusting parameters by using the fish school algorithm in step S3 includes: S31, optimizing parameters of the regression module by using a fish swarm algorithm; S32. Based on the regional wind power short-term power prediction results output by the regression module after parameter optimization, the parameters of the multimodal module are optimized through the fish school algorithm.

[0009] Furthermore, the parameter optimization of the regression module in step S31 specifically includes: S311, using the number of linear connection parameters of the regression module as the fish school size N, using the initial value of each linear connection parameter as the initial position of each individual fish, using the neighborhood search radius of each linear connection parameter as the visual field of each individual fish, using the adjustment value of each linear connection parameter as the step size of each individual fish step, setting the control parameter as the crowding factor δ, and setting the number of repetitions Trynumber; S312, evaluating the impact of each behavior on the result through various behaviors of the fish school, and judging the difference between the optimized regional wind power short-term power forecast result obtained through the historical data set and the actual power generation; the behaviors of the fish school include foraging behavior, grouping behavior and tail-chasing behavior; S313, updating the behavior of each individual fish in the school of fish; S314, evaluating the behaviors of all individual fish in the school of fish, and updating the optimal individual fish; S315, returning to step S312 to perform the next fish school behavior until the number of repetitions is reached or the difference between the optimized regional wind power short-term power forecast result and the actual power generation is less than a preset value.

[0010] Furthermore, in step S32, the multimodal module performs parameter optimization including: S321, using the number of connections between the multimodal module and the hidden layer of the meteorological large model as the fish school size N, using the initial value of the weight of each connection as the initial position of each individual fish, using the neighborhood search radius of the weight of each connection as the visual field of each individual fish Visual, using the change amplitude of the weight of each connection as the step size of each individual fish step, setting the control parameter as the crowding factor δ, and setting the number of repetitions Trynumber; S322, evaluating the impact of each behavior on the result through various behaviors of the fish school, and judging the difference between the optimized regional wind power short-term power forecast result obtained through the historical data set and the actual power generation; the behaviors of the fish school include foraging behavior, grouping behavior and tail-chasing behavior; S323, updating the behavior of each individual fish in the school of fish; S324, evaluating the behaviors of all individual fish in the school of fish, and updating the optimal individual fish; S325, returning to step S322 to perform the next fish school behavior until the number of repetitions is reached or the difference between the optimized regional wind power short-term power forecast result and the actual power generation is less than a preset value.

[0011] On the other hand, the present invention also proposes a regional wind power short-term power prediction system based on a fish school algorithm to improve the large prediction model, comprising: Dataset module: collects historical data, including weather forecast grid data, satellite cloud image data, radar inversion data, and actual data from each wind farm station to build a historical data set; Module Adding Module: Add a multimodal module to the front end of the embedding layer of the meteorological model to analyze multimodal data including meteorological forecast grid data, satellite cloud image data, radar inversion data and actual data from each wind farm station; Add a regression module to the output results of the meteorological model to change the output results of the meteorological model from outputting different meteorological elements to directly outputting regional wind power short-term power forecast; Pre-training and parameter adjustment module: Use historical data sets to pre-train the meteorological large model with added multimodal modules and regression modules, adjust parameters through the fish school algorithm, and obtain the power prediction large model; Prediction module: Use the power prediction model to perform short-term regional wind power forecasting.

[0012] Furthermore, the multimodal module adopts a processing model of an artificial intelligence visual algorithm; and the regression module adopts a linear regression layer.

[0013] Furthermore, the pre-training and parameter adjustment module includes: Regression module optimization unit: optimizes the parameters of the regression module by using a fish swarm algorithm; Meteorological large model optimization unit: Based on the regional wind power short-term power forecast results output by the regression module after parameter optimization, the parameters of the multimodal module are optimized through the fish school algorithm.

[0014] Furthermore, the regression module optimization unit includes: Set up subunits: use the number of linear connection parameters of the regression module as the school size N, the initial value of each linear connection parameter as the initial position of each individual fish, the neighborhood search radius of each linear connection parameter as the visual field of each individual fish, the adjustment value of each linear connection parameter as the step size of each individual fish, set the control parameter as the crowding factor δ, and set the number of repetitions Trynumber; Behavior subunit: evaluate the impact of each behavior on the results through the various behaviors of the fish school, and judge the difference between the optimized regional wind power short-term power forecast results obtained through the historical data set and the actual power generation; the behaviors of the fish school include foraging behavior, grouping behavior and tail-chasing behavior; Update subunit: updates the behavior of each individual fish in the school; Evaluation subunit: evaluates the behavior of all individual fish in the school and updates the optimal individual fish; Repeat subunit: Return to the behavior subunit to perform the next school behavior until the number of repetitions is reached or the difference between the optimized regional wind power short-term power forecast result and the actual power generation is less than the preset value.

[0015] Furthermore, the meteorological large model optimization unit includes: Set up subunits: use the number of connections between the multimodal module and the hidden layer of the meteorological model as the fish school size N, use the initial value of the weight of each connection as the initial position of each individual fish, use the neighborhood search radius of the weight of each connection as the visual field of each individual fish, use the change amplitude of the weight of each connection as the step size of each individual fish, set the control parameter as the crowding factor δ, and set the number of repetitions Trynumber; Behavior subunit: evaluate the impact of each behavior on the results through the various behaviors of the fish school, and judge the difference between the optimized regional wind power short-term power forecast results obtained through the historical data set and the actual power generation; the behaviors of the fish school include foraging behavior, grouping behavior and tail-chasing behavior; Update subunit: updates the behavior of each individual fish in the school; Evaluation subunit: evaluates the behavior of all individual fish in the school and updates the optimal individual fish; Repeat subunit: Return to the behavior subunit to perform the next school behavior until the number of repetitions is reached or the difference between the optimized regional wind power short-term power forecast result and the actual power generation is less than the preset value.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention combines multi-modal data such as meteorological forecast grid data, satellite cloud image data, radar inversion data and actual data of each wind farm station, and applies large model technology to regional wind power prediction, which can effectively predict regional wind power output under multiple scenarios such as normal weather, icing, cold waves, and strong winds, avoiding the situation in conventional algorithms where the same model is difficult to predict regional wind power in multiple scenarios; 2. The present invention uses the fish school optimization algorithm to adjust the parameters of the pre-trained large model, which reduces the computing cost of the large model and improves the prediction efficiency and accuracy of the regional wind power prediction large model. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the process of Example 1 of the present invention. DETAILED DESCRIPTION

[0018] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0019] The design idea of ​​the present invention is to combine the big model technology, use multimodal data, find the implicit and complex data relationships between various types of data, and thus establish a big model for power prediction; the attention mechanism used by the big model technology can effectively focus on the regional wind power conditions under extreme weather conditions such as icing, cold waves, and strong winds, overcome the disadvantage of few relevant data samples, and accurately predict the regional power under extreme weather conditions; at the same time, in order to reduce the calculation time and resource requirements, the present invention uses the fish school algorithm to adjust and improve the parameters of the power prediction big model, reduce the overall computing cost of the big model, and improve the efficiency and accuracy of regional wind power prediction in multiple scenarios.

[0020] The present invention will be further described below in conjunction with specific embodiments and accompanying drawings.

[0021] Embodiment 1: The process of regional wind power short-term power prediction based on the fish school algorithm to improve the prediction model in this embodiment is as follows: Figure 1 As shown, including: 1. Organize the national meteorological forecast grid data, satellite cloud map data, radar inversion data and actual power data of each wind farm into multimodal data in the format of text, pictures, videos, etc., build a historical data set, and prepare for large model training.

[0022] 2. Build a large power prediction model: Numerical weather forecast is the basis of wind power forecast. When numerical weather forecast is available, regression algorithm can be used to forecast regional wind power. Therefore, the power forecast model to be implemented in this embodiment adds multimodal module (which can parse image data, text data, video data, etc.) and regression module (which directly converts the results of meteorological forecast into the results of power forecast) technology to the meteorological model.

[0023] Regarding the multimodal module, it should be noted that the meteorological model uses pre-trained large models from major manufacturers, such as Google DeepMind's GraphCast, Nvidia's FourCastNet, Microsoft's ClimaAI, etc., and can also use large models that can be applied in the meteorological field, such as the time series prediction model TimeGPT, etc. The meteorological model is an open source model on the Internet, and its architecture usually includes an embedding layer, a hidden layer, an attention mechanism, and a decoder. In this embodiment, the architecture of the meteorological model has not been changed, just because the meteorological model needs to use text data in a fixed format and does not have multimodal capabilities. Therefore, the power prediction model constructed in this embodiment is based on the meteorological model, and a processing module for artificial intelligence visual directions such as pictures and videos is added. As a multimodal module, visual related algorithms such as convolutional neural networks are used, and pictures and video data in historical data sets are used to adjust the meteorological model. The architecture is not adjusted, and only the above data is used to adjust the parameters. The parameter adjustment uses the fish school algorithm, also known as the artificial fish school algorithm, and the specific process is introduced in the following steps.

[0024] What needs to be explained about the regression module is that the purpose of adding the regression module is to change the result of the large model from obtaining different meteorological elements to directly outputting the result of the regional wind power short-term power forecast. The implementation of the regression module is relatively intuitive in the neural network, and it can be obtained by directly adding a linear layer (the parameters of the linear layer need to be adjusted), that is, a linear regression layer (fully connected layer) is added to the result of the original meteorological large model outputting each meteorological element, and the regional wind power short-term power forecast result at that moment is obtained (for example, if the meteorological large model outputs N meteorological elements, then an N*1 linear fully connected layer is added, and finally regressed to the power forecast value); and the parameter adjustment of the linear layer also uses the fish school algorithm.

[0025] 3. Use historical data sets to pre-train the large meteorological model with added multimodal modules and regression modules, adjust parameters through the fish school algorithm, and obtain the large power prediction model.

[0026] The fish swarm algorithm is used to adjust and optimize the parameters. Since the parameter adjustment and optimization process of the multimodal module needs to be fixed on the result obtained after the regression module is adjusted, the parameter adjustment and optimization should be performed again. Therefore, the parameters of the regression module need to be optimized first, and then the multimodal module should be optimized.

[0027] The regression module only optimizes the parameters of the added linear regression layer (fully connected layer). The optimization process includes: 1. The number of linear connection parameters of the regression module is used as the fish school size N, the initial value of each linear connection parameter is used as the initial position of each individual fish, the neighborhood search radius of each linear connection parameter is used as the visual field of each individual fish, the adjustment value of each linear connection parameter is used as the step size of each individual fish, the control parameter is set as the crowding factor δ, and the number of repetitions Trynumber is set; 2. Evaluate the impact of various behaviors on the results through the behaviors of the fish school, and judge the difference between the optimized regional wind power short-term power forecast results obtained through the historical data set and the actual power generation; the behaviors of the fish school include foraging behavior, grouping behavior and tail-chasing behavior; 3. Update the behavior of each individual fish in the school; 4. Evaluate the behavior of all individual fish in the school and update the optimal individual fish; 5. Return to step 2 to perform the next fish school behavior until the number of repetitions is reached or the difference between the optimized regional wind power short-term power forecast result and the actual power generation is less than the preset value.

[0028] The multimodal module needs to optimize the hidden layer connection part and the corresponding parameters. By using the adaptive optimization and fast convergence characteristics of the fish swarm algorithm, the network can quickly find nodes that connect the hidden layers in various ways, as well as the corresponding weights of each node, and finally achieve a more accurate prediction of the power prediction results of the wind power area. The optimization process includes: 1. The number of connections between the multimodal module and the hidden layer of the meteorological model is used as the fish school size N, the initial value of the weight of each connection is used as the initial position of each individual fish, the neighborhood search radius of the weight of each connection is used as the visual field of each individual fish, the change amplitude of the weight of each connection is used as the step size of each individual fish, the control parameter is set as the crowding factor δ, and the number of repetitions Trynumber is set; 2. Evaluate the impact of various behaviors on the results through the behaviors of the fish school, and judge the difference between the optimized regional wind power short-term power forecast results obtained through the historical data set and the actual power generation; the behaviors of the fish school include foraging behavior, grouping behavior and tail-chasing behavior; 3. Update the behavior of each individual fish in the school; 4. Evaluate the behavior of all individual fish in the school and update the optimal individual fish; 5. Return to step 2 to perform the next fish school behavior until the number of repetitions is reached or the difference between the optimized regional wind power short-term power forecast result and the actual power generation is less than the preset value.

[0029] After pre-training and optimization, a large power prediction model is obtained.

[0030] 4. Use the power prediction model to predict future wind power output in the region under multiple scenarios such as normal weather, icing, cold waves, and strong winds.

[0031] In this embodiment, the fish swarm algorithm can also be replaced by an optimization algorithm such as a particle swarm algorithm, an ant colony algorithm, etc., but the parameter optimization adjustment effect will be reduced to a certain extent, and the adjustment speed will also be slower.

[0032] This embodiment can combine multimodal data and apply big model technology to regional wind power prediction, thereby realizing wind power prediction in multiple scenarios; the fish school optimization algorithm is used to adjust the parameters of the pre-trained big model, thereby reducing the computing cost of the big model and improving the prediction accuracy of the regional wind power prediction big model.

[0033] Embodiment 2: This embodiment proposes a regional wind power short-term power forecasting system based on a fish school algorithm to improve the large forecasting model, including: Dataset module: collects historical data, including weather forecast grid data, satellite cloud image data, radar inversion data, and actual data from each wind farm station to build a historical data set; Module Adding Module: Add a multimodal module to the front end of the embedding layer of the meteorological model to analyze multimodal data including meteorological forecast grid data, satellite cloud image data, radar inversion data and actual data from each wind farm station; Add a regression module to the output results of the meteorological model to change the output results of the meteorological model from outputting different meteorological elements to directly outputting regional wind power short-term power forecast; Pre-training and parameter adjustment module: Use historical data sets to pre-train the meteorological large model with added multimodal modules and regression modules, adjust parameters through the fish school algorithm, and obtain the power prediction large model; Prediction module: Use the power prediction model to perform short-term regional wind power forecasting.

[0034] The multimodal module adopts a processing model of an artificial intelligence visual algorithm; the regression module adopts a linear regression layer.

[0035] The pre-training and parameter adjustment modules include: Regression module optimization unit: optimizes the parameters of the regression module by using a fish swarm algorithm; Meteorological large model optimization unit: Based on the regional wind power short-term power forecast results output by the regression module after parameter optimization, the parameters of the multimodal module are optimized through the fish school algorithm.

[0036] The regression module optimization unit includes: Set up subunits: use the number of linear connection parameters of the regression module as the school size N, the initial value of each linear connection parameter as the initial position of each individual fish, the neighborhood search radius of each linear connection parameter as the visual field of each individual fish, the adjustment value of each linear connection parameter as the step size of each individual fish, set the control parameter as the crowding factor δ, and set the number of repetitions Trynumber; Behavior subunit: evaluate the impact of each behavior on the results through the various behaviors of the fish school, and judge the difference between the optimized regional wind power short-term power forecast results obtained through the historical data set and the actual power generation; the behaviors of the fish school include foraging behavior, grouping behavior and tail-chasing behavior; Update subunit: updates the behavior of each individual fish in the school; Evaluation subunit: evaluates the behavior of all individual fish in the school and updates the optimal individual fish; Repeat subunit: Return to the behavior subunit to perform the next school behavior until the number of repetitions is reached or the difference between the optimized regional wind power short-term power forecast result and the actual power generation is less than the preset value.

[0037] The meteorological large model optimization unit includes: Set up subunits: use the number of connections between the multimodal module and the hidden layer of the meteorological model as the fish school size N, use the initial value of the weight of each connection as the initial position of each individual fish, use the neighborhood search radius of the weight of each connection as the visual field of each individual fish, use the change amplitude of the weight of each connection as the step size of each individual fish, set the control parameter as the crowding factor δ, and set the number of repetitions Trynumber; Behavior subunit: evaluate the impact of each behavior on the results through the various behaviors of the fish school, and judge the difference between the optimized regional wind power short-term power forecast results obtained through the historical data set and the actual power generation; the behaviors of the fish school include foraging behavior, grouping behavior and tail-chasing behavior; Update subunit: updates the behavior of each individual fish in the school; Evaluation subunit: evaluates the behavior of all individual fish in the school and updates the optimal individual fish; Repeat subunit: Return to the behavior subunit to perform the next school behavior until the number of repetitions is reached or the difference between the optimized regional wind power short-term power forecast result and the actual power generation is less than the preset value.

[0038] The regional wind power short-term power prediction system based on the improved prediction large model based on the fish school algorithm proposed in this embodiment can implement the regional wind power short-term power prediction method based on the improved prediction large model based on the fish school algorithm proposed in Example 1, and has the same technical effect as Example 1.

[0039] The above-mentioned embodiments are only preferred implementations of the present invention, and are only used to help understand the method and core ideas of the present application. The protection scope of the present invention is not limited to the above-mentioned embodiments. All technical solutions under the idea of ​​the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A regional wind power short-term power forecasting method based on a fish school algorithm to improve the large prediction model, characterized in that: include: S1. Collect historical data, including weather forecast grid data, satellite cloud image data, radar inversion data and actual data from each wind farm station, and construct a historical data set; S2. Add a multimodal module to the front end of the embedding layer of the meteorological model to analyze multimodal data including meteorological forecast grid data, satellite cloud image data, radar inversion data and actual data from each wind farm station; add a regression module to the output results of the meteorological model to change the output results of the meteorological model from outputting different meteorological elements to directly outputting regional wind power short-term power forecast; S3. Pre-train the meteorological large model with added multimodal module and regression module using historical data sets, adjust parameters through fish school algorithm, and obtain the power prediction large model; S4. Use a large power prediction model to perform short-term regional wind power prediction.

2. The regional wind power short-term power forecasting method based on the improved prediction model of the fish school algorithm according to claim 1 is characterized in that: In step S2, the multimodal module adopts a processing model of an artificial intelligence visual algorithm; the regression module adopts a linear regression layer.

3. The regional wind power short-term power forecasting method based on the improved prediction model of the fish school algorithm according to claim 1 is characterized in that: The method for adjusting parameters by using the fish school algorithm in step S3 includes: S31, optimizing parameters of the regression module by using a fish swarm algorithm; S32. Based on the regional wind power short-term power prediction results output by the regression module after parameter optimization, the parameters of the multimodal module are optimized through the fish school algorithm.

4. The regional wind power short-term power forecasting method based on the improved prediction model of the fish school algorithm according to claim 3 is characterized in that: The parameter optimization of the regression module in step S31 specifically includes: S311, using the number of linear connection parameters of the regression module as the fish school size N, using the initial value of each linear connection parameter as the initial position of each individual fish, using the neighborhood search radius of each linear connection parameter as the visual field of each individual fish, using the adjustment value of each linear connection parameter as the step size of each individual fish step, setting the control parameter as the crowding factor δ, and setting the number of repetitions Trynumber; S312, evaluating the impact of each behavior on the result through various behaviors of the fish school, and judging the difference between the optimized regional wind power short-term power forecast result obtained through the historical data set and the actual power generation; the behaviors of the fish school include foraging behavior, grouping behavior and tail-chasing behavior; S313, updating the behavior of each individual fish in the school of fish; S314, evaluating the behaviors of all individual fish in the school of fish, and updating the optimal individual fish; S315, returning to step S312 to perform the next fish school behavior until the number of repetitions is reached or the difference between the optimized regional wind power short-term power forecast result and the actual power generation is less than a preset value.

5. The regional wind power short-term power forecasting method based on the improved prediction model of the fish school algorithm according to claim 3 is characterized in that: In step S32, the multimodal module performs parameter optimization including: S321, using the number of connections between the multimodal module and the hidden layer of the meteorological large model as the fish school size N, using the initial value of the weight of each connection as the initial position of each individual fish, using the neighborhood search radius of the weight of each connection as the visual field of each individual fish Visual, using the change amplitude of the weight of each connection as the step size of each individual fish step, setting the control parameter as the crowding factor δ, and setting the number of repetitions Trynumber; S322, evaluating the impact of each behavior on the result through various behaviors of the fish school, and judging the difference between the optimized regional wind power short-term power forecast result obtained through the historical data set and the actual power generation; the behaviors of the fish school include foraging behavior, grouping behavior and tail-chasing behavior; S323, updating the behavior of each individual fish in the school of fish; S324, evaluating the behaviors of all individual fish in the school of fish, and updating the optimal individual fish; S325, returning to step S322 to perform the next fish school behavior until the number of repetitions is reached or the difference between the optimized regional wind power short-term power forecast result and the actual power generation is less than a preset value.

6. A regional wind power short-term power forecasting system based on a fish school algorithm to improve the large prediction model, characterized in that: include: Dataset module: collects historical data, including weather forecast grid data, satellite cloud image data, radar inversion data, and actual data from each wind farm station to build a historical data set; Module Adding Module: Add a multimodal module to the front end of the embedding layer of the meteorological model to analyze multimodal data including meteorological forecast grid data, satellite cloud image data, radar inversion data and actual data from each wind farm station; Add a regression module to the output results of the meteorological model to change the output results of the meteorological model from outputting different meteorological elements to directly outputting regional wind power short-term power forecast; Pre-training and parameter adjustment module: Use historical data sets to pre-train the meteorological large model with added multimodal modules and regression modules, adjust parameters through the fish school algorithm, and obtain the power prediction large model; Prediction module: Use the power prediction meteorological model to perform short-term regional wind power forecasting.

7. The regional wind power short-term power forecasting system based on the fish school algorithm improved forecasting large model according to claim 6 is characterized in that: The multimodal module adopts a processing model of an artificial intelligence visual algorithm; the regression module adopts a linear regression layer.

8. The regional wind power short-term power forecasting system based on the improved prediction model of the fish school algorithm according to claim 6 is characterized in that: The pre-training and parameter adjustment modules include: Regression module optimization unit: optimizes the parameters of the regression module by using a fish swarm algorithm; Meteorological large model optimization unit: Based on the regional wind power short-term power forecast results output by the regression module after parameter optimization, the parameters of the multimodal module are optimized through the fish school algorithm.

9. The regional wind power short-term power forecasting system based on the fish school algorithm improved forecasting large model according to claim 8 is characterized in that: The regression module optimization unit includes: Set up subunits: use the number of linear connection parameters of the regression module as the school size N, the initial value of each linear connection parameter as the initial position of each individual fish, the neighborhood search radius of each linear connection parameter as the visual field of each individual fish, the adjustment value of each linear connection parameter as the step size of each individual fish, set the control parameter as the crowding factor δ, and set the number of repetitions Trynumber; Behavior subunit: evaluate the impact of each behavior on the results through the various behaviors of the fish school, and judge the difference between the optimized regional wind power short-term power forecast results obtained through the historical data set and the actual power generation; the behaviors of the fish school include foraging behavior, grouping behavior and tail-chasing behavior; Update subunit: updates the behavior of each individual fish in the school; Evaluation subunit: evaluates the behavior of all individual fish in the school and updates the optimal individual fish; Repeat subunit: Return to the behavior subunit to perform the next school behavior until the number of repetitions is reached or the difference between the optimized regional wind power short-term power forecast result and the actual power generation is less than the preset value.

10. The regional wind power short-term power forecasting system based on the fish school algorithm improved forecasting large model according to claim 8, characterized in that: The meteorological large model optimization unit includes: Set up subunits: use the number of connections between the multimodal module and the hidden layer of the meteorological model as the fish school size N, use the initial value of the weight of each connection as the initial position of each individual fish, use the neighborhood search radius of the weight of each connection as the visual field of each individual fish, use the change amplitude of the weight of each connection as the step size of each individual fish, set the control parameter as the crowding factor δ, and set the number of repetitions Trynumber; Behavior subunit: evaluate the impact of each behavior on the results through the various behaviors of the fish school, and judge the difference between the optimized regional wind power short-term power forecast results obtained through the historical data set and the actual power generation; the behaviors of the fish school include foraging behavior, grouping behavior and tail-chasing behavior; Update subunit: updates the behavior of each individual fish in the school; Evaluation subunit: evaluates the behavior of all individual fish in the school and updates the optimal individual fish; Repeat subunit: Return to the behavior subunit to perform the next school behavior until the number of repetitions is reached or the difference between the optimized regional wind power short-term power forecast result and the actual power generation is less than the preset value.

Images