Forecasting model training method and device, storage medium and electronic equipment

Abstract

The embodiment of the invention provides a prediction model training method and device, a storage medium and electronic equipment, and the method comprises the steps: traversing a target prediction region based on a sliding window of a preset step length under the condition that the size of a first region of the target prediction region is determined to be greater than a preset threshold value, dividing the target forecasting area into a plurality of first forecasting sub-areas; a target forecasting model is constructed according to a first model and a second model, the first model is used for representing regional features of each first forecasting sub-region, and the second model is used for representing general features of the target forecasting region; and training the target forecasting model by taking the first forecasting data of the plurality of first forecasting sub-regions at the first moment as input data and taking the second forecasting data of the plurality of first forecasting sub-regions at the second moment as output data, the second moment being the next moment of the first moment.

Description

Technical Field

[0001] This application relates to the field of wind energy resource prediction technology, and more specifically, to a method and apparatus for training a prediction model, a storage medium, and an electronic device. Background Technology

[0002] Three-dimensional wind farm forecasting plays a crucial role in wind power generation. Accurate wind speed and direction predictions can improve the power generation efficiency of wind farms. Wind speed and direction are key factors affecting the operating efficiency of wind turbines. Accurate three-dimensional wind farm forecasts allow wind farms to precisely schedule operations before power generation, optimizing turbine operation and avoiding efficiency losses caused by wind speed fluctuations. Furthermore, three-dimensional wind farm forecasting helps wind farms develop reasonable maintenance plans, detect potential equipment failures or abnormal operations in advance, reduce downtime, and extend turbine lifespan. In terms of grid dispatching, the intermittency and volatility of wind power pose challenges to grid stability. Accurate three-dimensional wind farm forecasts allow grid operators to predict changes in wind power output and proactively dispatch other energy resources (such as energy storage systems and thermal power) to balance grid load, ensuring the stability and reliability of the power system. Overall, three-dimensional wind farm forecasting not only improves the operating efficiency of wind farms but also promotes intelligent management of the power system, providing strong support for the efficient application of renewable energy.

[0003] Traditional wind field forecasting relies on physical mechanisms and numerical models, calculating future wind field data by solving a series of physical and atmospheric equations. However, this approach faces challenges such as high computational cost, long processing time, and inherent systemic biases. With the development of artificial intelligence (AI) technology, more and more researchers are exploring AI-based wind field forecasting methods. However, most current wind field forecasting methods treat the forecast area as a whole, which has the following problems: treating the entire area as a whole consumes a lot of computational resources; the model has poor generalization ability and can only forecast selected areas.

[0004] Therefore, traditional wind field forecasting models treat the forecast area as a whole, resulting in poor generalization and insufficient forecast accuracy.

[0005] This addresses the problem that forecasting models in related technologies treat the forecast area as a whole for prediction, resulting in poor generalization and insufficient forecast accuracy.

[0006] Therefore, it is necessary to improve the relevant technology to overcome the aforementioned defects. Summary of the Invention

[0007] This application provides a method and apparatus for training a forecasting model, a storage medium, and an electronic device to at least solve the problem in related technologies where forecasting models treat the forecast area as a whole for prediction, resulting in poor generalization and insufficient forecasting accuracy.

[0008] According to one embodiment of this application, a method for training a forecast model is provided, comprising: when the size of a first region of a target forecast region is determined to be greater than a preset threshold, traversing the target forecast region by a sliding window with a preset step size to divide the target forecast region into multiple first forecast sub-regions; constructing a target forecast model according to a first model and a second model, wherein the first model is used to characterize the regional characteristics of each first forecast sub-region, and the second model is used to characterize the general characteristics of the target forecast region;

[0009] In an exemplary embodiment, before constructing the target prediction model based on the first model and the second model, the method further includes: constructing a positive sample set and a negative sample set corresponding to each of the first prediction data as center sample data; calculating a first similarity between the center sample data and each positive sample in the positive sample set, and calculating a second similarity between the center sample data and each negative sample in the negative sample set; inputting the first similarity and the second similarity into a target loss function to obtain a loss value corresponding to the first model; and performing gradient calculation on the model parameters in the first model based on the loss value to train the first model.

[0010] In an exemplary embodiment, constructing a positive sample set corresponding to the central sample data based on the plurality of first forecast data includes: determining a second forecast sub-region corresponding to the central sample data, and determining each third forecast sub-region corresponding to the other forecast data besides the central sample data in the plurality of first forecast data, wherein the plurality of first forecast sub-regions include: the second forecast sub-region and the plurality of third forecast sub-regions; determining a fourth forecast sub-region in the plurality of third forecast sub-regions whose area of ​​a first overlapping region with the second forecast sub-region is greater than or equal to a preset area, and determining the forecast data corresponding to the fourth forecast sub-region as the positive sample data in the positive sample set, wherein the other forecast data includes: the forecast data corresponding to the fourth forecast sub-region; and constructing the positive sample set based on the positive sample data.

[0011] In an exemplary embodiment, constructing a negative sample set corresponding to the central sample data based on the plurality of first forecast data includes: determining a second forecast sub-region corresponding to the central sample data, and determining each third forecast sub-region corresponding to the other forecast data besides the central sample data in the plurality of first forecast data, wherein the plurality of first forecast sub-regions include: the second forecast sub-region and the plurality of third forecast sub-regions; determining a fifth forecast sub-region among the plurality of third forecast sub-regions whose area of ​​a first overlapping region with the second forecast sub-region is smaller than a preset area, and determining the forecast data corresponding to the fifth forecast sub-region as the negative sample data in the negative sample set, wherein the other forecast data includes: the forecast data corresponding to the fifth forecast sub-region; and constructing the negative sample set based on the negative sample data.

[0012] In an exemplary embodiment, after training the target forecast model, the method further includes: acquiring third forecast data of the plurality of first forecast sub-regions at the current time, and inputting the plurality of third forecast data into the trained target forecast model, so that the target forecast model predicts fourth forecast data of each first forecast sub-region at the third time, wherein the third time is the time after the current time; and concatenating the plurality of fourth forecast data to obtain the predicted forecast data of the target forecast region at the third time.

[0013] In an exemplary embodiment, after training the target prediction model, the method further includes: if it is determined that a fifth prediction data of a region to be predicted in the target prediction region needs to be predicted at a third time, determining whether the size of a second region of the region to be predicted is greater than a preset threshold, wherein the third time is the time after the current time; if it is determined that the size of the second region is less than or equal to the preset threshold, inputting a sixth prediction data of the region to be predicted at the current time into the trained target prediction model to obtain the fifth prediction data.

[0014] In an exemplary embodiment, after determining whether the size of the second region of the region to be predicted is greater than the preset threshold, the method further includes: if the size of the second region is greater than the preset threshold, traversing the region to be predicted using a sliding window based on the preset step size to divide the region to be predicted into multiple second prediction sub-regions; inputting the seventh prediction data of the multiple second prediction sub-regions at the current time into the trained target prediction model to obtain the eighth prediction data of each second prediction sub-region at the third time, and determining whether there is a second overlapping region among the multiple second prediction sub-regions; if the second overlapping region is determined to exist among the multiple second prediction sub-regions, determining the average value of the eighth prediction data of the multiple third prediction sub-regions corresponding to the second overlapping region at the third time as the prediction data of the second overlapping region at the third time, wherein the multiple second prediction sub-regions include: the multiple third prediction sub-regions; and concatenating the multiple eighth prediction data and the prediction data of the second overlapping region at the third time to obtain the fifth prediction data.

[0015] According to another embodiment of this application, a training apparatus for a forecast model is provided, comprising: a segmentation module, configured to traverse the target forecast region by a sliding window with a preset step size, based on a first region size greater than a preset threshold, to segment the target forecast region into multiple first forecast sub-regions; a construction module, configured to construct a target forecast model based on a first model and a second model, wherein the first model is used to characterize the regional characteristics of each first forecast sub-region, and the second model is used to characterize the general characteristics of the target forecast region; and an input module, configured to use first forecast data of the multiple first forecast sub-regions at a first time as input data and second forecast data of the multiple first forecast sub-regions at a second time as output data to train the target forecast model, wherein the second time is the time following the first time.

[0016] According to yet another embodiment of this application, a computer-readable storage medium is also provided, wherein a computer program is stored therein, and the computer program is configured to perform the steps in any of the above method embodiments when it is run.

[0017] According to yet another embodiment of this application, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0018] According to yet another embodiment of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps in any of the above method embodiments.

[0019] In this application embodiment, when the size of the first region of the target forecast area is determined to be greater than a preset threshold, a sliding window with a preset step size is used to traverse the target forecast area to divide it into multiple first forecast sub-regions. A target forecast model is constructed based on a first model representing the regional characteristics of each first forecast sub-region and a second model representing the general characteristics of the target forecast area. The first forecast data of the multiple first forecast sub-regions at a first time step is used as input data, and the second forecast data of the multiple first forecast sub-regions at the next time step is used as output data to train the target forecast model. In other words, this application divides the target forecast area into multiple first forecast sub-regions and constructs a target forecast model based on a first model that can represent the regional characteristics of each first forecast sub-region and a second model that represents the general characteristics of the target forecast area, thereby training the target forecast model. According to this application embodiment, the problem of poor generalization and insufficient forecast accuracy caused by forecast models in related technologies predicting the forecast area as a whole can be solved, thereby improving the generalization and forecast accuracy of the target forecast model. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a hardware structure block diagram of a computer terminal device for a training method of a prediction model according to an embodiment of this application.

[0023] Figure 2 This is a flowchart of a training method for a prediction model according to an embodiment of this application;

[0024] Figure 3 This is a flowchart of a wind field forecasting method based on contrastive learning to mine regional characteristics of wind fields in different areas, according to an optional embodiment of this application;

[0025] Figure 4 This is a schematic diagram illustrating the segmentation of region E according to an optional embodiment of this application;

[0026] Figure 5 This is a schematic diagram of the inference and prediction stage segmentation according to an optional embodiment of this application;

[0027] Figure 6 This is a schematic diagram of the wind field forecasting model architecture according to an optional embodiment of this application;

[0028] Figure 7 This is a structural block diagram of a training device for a prediction model according to an embodiment of this application. Detailed Implementation

[0029] The embodiments of this application will be described in detail below with reference to the accompanying drawings and examples.

[0030] It should be noted that the terms "first," "second," etc., in the specification, claims, and drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0031] The methods and embodiments provided in this application can be executed on a computer terminal device or a similar computing device. Taking running on a computer terminal device as an example, Figure 1 This is a hardware structure block diagram of a computer terminal device for a training method of a prediction model according to an embodiment of this application. For example... Figure 1 As shown, a computer terminal device may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 (which may include, but is not limited to, a microprocessor (MPU) or a programmable logic device (PLD)) and a memory 104 for storing data are also shown. The computer terminal device may further include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the computer terminal device described above. For example, the computer terminal device may also include components that are more complex than those described above. Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0032] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the training method of the prediction model in this embodiment. The processor 102 executes various functional applications and data processing by running the computer programs stored in the memory 104, thus implementing the above-described method. The memory 104 may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to computer terminal devices via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0033] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by a communication provider for the computer terminal equipment. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0034] This embodiment provides a method for training a prediction model. Figure 2 This is a flowchart of a training method for a prediction model according to an embodiment of this application, such as... Figure 2 As shown, the process includes the following steps:

[0035] Step S202: If the size of the first region of the target forecast area is greater than a preset threshold, the target forecast area is traversed by a sliding window with a preset step size to divide the target forecast area into multiple first forecast sub-regions.

[0036] Step S204: Construct a target forecast model based on the first model and the second model, wherein the first model is used to characterize the regional characteristics of each first forecast sub-region, and the second model is used to characterize the general characteristics of the target forecast region;

[0037] Step S206: The target forecast model is trained by taking the first forecast data of the plurality of first forecast sub-regions at the first time as input data and the second forecast data of the plurality of first forecast sub-regions at the second time as output data, wherein the second time is the time after the first time.

[0038] Through the above steps, when the size of the first region of the target forecast area is determined to be greater than a preset threshold, a sliding window with a preset step size is used to traverse the target forecast area to divide it into multiple first forecast sub-regions. A target forecast model is constructed based on a first model representing the regional characteristics of each first forecast sub-region and a second model representing the general characteristics of the target forecast area. The first forecast data of the multiple first forecast sub-regions at the first time step is used as input data, and the second forecast data of the multiple first forecast sub-regions at the next time step is used as output data to train the target forecast model. In other words, this application divides the target forecast area into multiple first forecast sub-regions and constructs a target forecast model based on a first model that can represent the regional characteristics of each first forecast sub-region and a second model that represents the general characteristics of the target forecast area, thereby training the target forecast model. According to the embodiments of this application, the problem of poor generalization and insufficient forecast accuracy caused by forecast models in related technologies predicting the forecast area as a whole can be solved, thereby improving the generalization and forecast accuracy of the target forecast model.

[0039] Optionally, before constructing the target prediction model based on the first model and the second model in step S206 above, the method further includes:

[0040] (1) Using each first forecast data point as the central sample data, construct a positive sample set and a negative sample set corresponding to the central sample data based on the plurality of first forecast data points; specifically:

[0041] 1) Construct a positive sample set:

[0042] A second forecast sub-region corresponding to the central sample data is determined, and a third forecast sub-region corresponding to each of the other forecast data (excluding the central sample data) among the plurality of first forecast data is determined, wherein the plurality of first forecast sub-regions include: the second forecast sub-region and the plurality of third forecast sub-regions; a fourth forecast sub-region is determined among the plurality of third forecast sub-regions whose area of ​​a first overlapping region with the second forecast sub-region is greater than or equal to a preset area, and the forecast data corresponding to the fourth forecast sub-region is determined as the positive sample data in the positive sample set, wherein the other forecast data includes: the forecast data corresponding to the fourth forecast sub-region; the positive sample set is constructed based on the positive sample data;

[0043] Understandably, the steps to construct a positive sample set include:

[0044] Determine the second forecast sub-region corresponding to the center sample data: Assume that a large forecast region is being processed, which is divided into multiple k×k sub-regions. Select a specific sub-region (the second forecast sub-region) from these sub-regions as the center sample data;

[0045] Identify the third forecast subregion corresponding to other forecast data: Observe other forecast data in the dataset and determine which third forecast subregion each of them corresponds to.

[0046] Identifying overlapping regions: In other third forecast sub-regions, find fourth forecast sub-regions that overlap with the second forecast sub-region corresponding to the central sample data. The area of ​​the overlapping region needs to be greater than or equal to a preset area. The forecast data of the fourth forecast sub-regions whose overlapping area with the second forecast sub-region meets the condition are identified as positive sample data, and these positive sample data are collected to construct a positive sample set.

[0047] By using the above technical solution, by selecting forecast sub-regions that have sufficient overlap with the central sample data as positive samples, the model can capture the wind field patterns shared between different small regions, thereby enhancing the forecasting ability for unseen similar regions. Since the construction of the positive sample set emphasizes regional correlation, the first model can learn to more accurately identify wind field characteristics related to specific geographical locations.

[0048] 2) Construct a negative sample set:

[0049] A second forecast sub-region corresponding to the central sample data is determined, and a third forecast sub-region corresponding to each of the other forecast data (excluding the central sample data) among the plurality of first forecast data is determined, wherein the plurality of first forecast sub-regions include: the second forecast sub-region and the plurality of third forecast sub-regions; a fifth forecast sub-region is determined among the plurality of third forecast sub-regions whose area of ​​a first overlapping region with the second forecast sub-region is smaller than a preset area, and the forecast data corresponding to the fifth forecast sub-region is determined as the negative sample data in the negative sample set, wherein the other forecast data includes: the forecast data corresponding to the fifth forecast sub-region; the negative sample set is constructed based on the negative sample data;

[0050] Understandably, the technical solutions for constructing negative sample sets include:

[0051] The second forecast sub-region corresponding to the central sample data is determined, and each of the multiple first forecast data (excluding the central sample data) corresponds to a different third forecast sub-region.

[0052] Fifth forecast sub-region and negative sample data: Identify the fifth forecast sub-region that has geographical overlap with the second forecast sub-region (the sub-region where the central sample is located), but the overlap area is smaller than the preset area.

[0053] Constructing a negative sample set: Forecast data corresponding to all fifth forecast sub-regions are classified as negative sample data. These data constitute part of the negative sample set in the contrastive learning process.

[0054] Understandably, in wind field forecasting, small geographical differences can lead to significant changes in wind speed and direction. By including data in the negative sample set that have slight overlap with the central sample but differ in meteorological conditions, the first model can more accurately capture and reflect the regional characteristics of wind fields, thereby improving the overall accuracy and reliability of wind field forecasts.

[0055] (2) Calculate the first similarity between the central sample data and each positive sample in the positive sample set, and calculate the second similarity between the central sample data and each negative sample in the negative sample set;

[0056] Understandably, it is necessary to calculate the first and second similarities. Specifically:

[0057] First similarity (positive sample similarity): Calculates the similarity between the central sample data and each sample in the positive sample set. This can be achieved using some distance metric (e.g., cosine similarity, Euclidean distance, etc.), and the result reflects the degree of similarity between the central sample and its geographically neighboring regions (positive samples).

[0058] The second similarity (negative sample similarity): calculates the similarity between the center sample data and each sample in the negative sample set. This helps the first model understand the differences between the center sample and other geographical locations (negative samples).

[0059] (3) Input the first similarity and the second similarity into the target loss function to obtain the loss value corresponding to the first model;

[0060] The loss value of the first model also needs to be determined: the first and second similarities are taken as inputs and calculated into the target loss function. The loss function is used to maximize the similarity of positive samples (first similarity) while minimizing the similarity of negative samples (second similarity). Based on these similarity values, the loss function will produce a loss value that reflects the validity of the first model's learning of the regional feature representation of the center samples under the current parameter settings.

[0061] (4) Perform gradient calculation on the model parameters in the first model based on the loss value to train the first model.

[0062] Furthermore, the first model can be trained based on the first loss value: the backpropagation algorithm calculates the gradient based on the calculated loss value. An optimization algorithm is then used to update the model parameters of the first model based on the gradient, with the aim of reducing the loss value, i.e., improving the model's ability to distinguish and represent regional features.

[0063] Through the above technical solution, the first model will learn to focus on the subtle differences between different regions. Even if there is no clear topographic information in other parts of the dataset, it can still distinguish them through the learned representations, thereby improving the generalization ability and computational efficiency of the first model.

[0064] Optionally, after training the target prediction model in step S206 above, the method further includes: acquiring the third prediction data of the plurality of first prediction sub-regions at the current time, and inputting the plurality of third prediction data into the trained target prediction model so that the target prediction model predicts the fourth prediction data of each first prediction sub-region at the third time, wherein the third time is the time after the current time; and concatenating the plurality of fourth prediction data to obtain the predicted prediction data of the target prediction region at the third time.

[0065] Understandably, trained models can be used for wind field forecasting. By processing the current data of each segmented sub-region, the wind field conditions of each sub-region at the next moment can be predicted, and these predictions can be integrated to provide forecast data for the entire target forecast area. Specifically:

[0066] Acquire current data for multiple first forecast sub-regions: Assume the target forecast area is divided into 100 first forecast sub-regions. Collect third forecast data for each first forecast sub-region at the current time, i.e., data points containing wind speed, wind direction, and possible other meteorological parameters at that time.

[0067] Input to the target forecast model: The collected third forecast data is used as input to the target forecast model, which has been optimized through a contrastive learning strategy and a regional segmentation training method. The target forecast model uses its learned regional characteristics and general meteorological field representation to predict the fourth forecast data for each first forecast sub-region at the next time step, i.e., the wind field prediction value for each sub-region at the next time step.

[0068] Data splicing: After completing the prediction for each sub-region, the fourth prediction data of all the first prediction sub-regions are spliced ​​together according to the original segmentation order and position to form a comprehensive prediction data for the third time of the entire target prediction region.

[0069] By using the above technical solution, a large area can be divided into multiple smaller sub-regions for independent forecasting, and then the results can be stitched together. This can handle the complexity and diversity of wind fields over a large area, while reducing the amount of computation and improving forecasting efficiency.

[0070] Optionally, after training the target prediction model in step S206 above, the method further includes: if it is determined that the fifth prediction data of the region to be predicted in the target prediction region needs to be predicted at a third time, determining whether the size of the second region of the region to be predicted is greater than the preset threshold, wherein the third time is the time after the current time; if it is determined that the size of the second region is less than or equal to the preset threshold, inputting the sixth prediction data of the region to be predicted at the current time into the trained target prediction model to obtain the fifth prediction data.

[0071] It is understandable that the fifth forecast data for the area to be forecasted can be predicted at the next time step from the current time. Specifically:

[0072] In the forecasting process, it is assumed that the meteorological conditions of the area to be forecasted in the target forecast area at the next time point in the current time range have been determined (i.e., the fifth forecast data).

[0073] Assess the size of the region to be predicted: Check if the size of the second region of the region to be predicted is greater than the preset threshold. The preset threshold is used to determine whether further segmentation is needed or whether to directly use the model for prediction.

[0074] Direct forecasting of small areas: If the size of the second region to be forecasted is less than or equal to a preset threshold, this means that the region to be forecasted is small enough that further segmentation is not necessary. In this case, the sixth forecast data of the region at the current time (i.e., the meteorological data of the region at the current time, such as wind speed and wind direction) can be directly input into the trained target forecast model to obtain the fifth forecast data of the region at the next time.

[0075] The above technical solution allows for the direct use of the trained target prediction model for small-area forecasts without additional preprocessing or segmentation, thus improving the efficiency of the forecasting process.

[0076] The method further includes, after determining whether the size of the second region of the region to be predicted is greater than the preset threshold, the method further includes: if the size of the second region is greater than the preset threshold, traversing the region to be predicted using a sliding window with a preset step size to divide the region to be predicted into multiple second prediction sub-regions; inputting the seventh prediction data of the multiple second prediction sub-regions at the current time into the trained target prediction model to obtain the eighth prediction data of each second prediction sub-region at the third time, and determining whether there is a second overlapping region among the multiple second prediction sub-regions; if the second overlapping region is determined to exist among the multiple second prediction sub-regions, determining the average value of the eighth prediction data of the multiple third prediction sub-regions corresponding to the second overlapping region at the third time as the prediction data of the second overlapping region at the third time, wherein the multiple second prediction sub-regions include the multiple third prediction sub-regions; and concatenating the multiple eighth prediction data and the prediction data of the second overlapping region at the third time to obtain the fifth prediction data.

[0077] Understandably, when the area to be predicted exceeds a preset threshold size, forecast data can be generated through more detailed region segmentation and a trained target prediction model. Specifically:

[0078] Assess the size of the region to be predicted: Determine whether the size of the region to be predicted exceeds the preset threshold. If it does, more detailed segmentation is required.

[0079] Segmentation using a sliding window: The region to be predicted is traversed using a sliding window strategy with a preset step size, and it is further subdivided into multiple smaller second prediction sub-regions.

[0080] Input data for prediction: Input the seventh forecast data (i.e., the current wind speed, wind direction and other meteorological data of each sub-region) of each second forecast sub-region at the current time into the already trained target forecast model to predict the eighth forecast data of each sub-region at the next time.

[0081] Handling overlapping regions: During the segmentation process, some sub-regions may overlap, i.e., there are second overlapping regions. Special processing is required for these regions: the average of the eighth forecast data of all segmented sub-regions in the overlapping region is used as the forecast data for the second overlapping region at the third time step.

[0082] The forecast data is generated by splicing together the eighth forecast data of all non-overlapping second forecast sub-regions and the forecast data of the second overlapping region to generate the fifth forecast data of the entire forecast area at the next time step.

[0083] Based on the above technical solution, a sliding window strategy is used to traverse and segment the area to be predicted, avoiding the high computational costs that may result from directly predicting large areas and improving forecast efficiency. Furthermore, by calculating the average value of forecast data in overlapping areas, the geographical consistency and accuracy of the forecast results are ensured, reducing prediction bias.

[0084] To better understand the training process of the above-mentioned forecast model, the implementation flow of the training method of the above-mentioned forecast model will be described below in conjunction with optional embodiments, but this is not intended to limit the technical solution of the embodiments of this application.

[0085] To address the shortcomings of related technologies that treat the forecast area as a whole (high computational cost and poor generalization), this application proposes an optional embodiment of a wind field forecasting method based on contrastive learning to mine the regional characteristics of wind fields in different regions. By employing a region segmentation training strategy, the computational cost of a single forecast is reduced, while the model can complete forecasts for different regions, enhancing generalization. Secondly, this optional embodiment does not rely on regional historical wind resource assessments or topographical inputs; it is purely data-driven, mining the regional characteristics of wind fields in different areas and pre-training a wind field feature representation model that reflects these regional characteristics. Finally, based on the pre-trained representation model, a wind field forecasting model is constructed to achieve future wind field forecasts.

[0086] Figure 3 This is a flowchart of a wind field forecasting method based on contrastive learning to mine regional wind field characteristics in different areas, according to an optional embodiment of this application, as shown below. Figure 3 As shown:

[0087] 1. Data preprocessing strategy and task definition for region segmentation:

[0088] Step S301: Determine the size of the given region E;

[0089] Suppose that the range of the given region E (i.e. the target prediction region) is (X1, Y1)-(X2, Y2).

[0090] Step S302: Determine whether |X2-X1| > K, and determine whether |Y2-Y1| > K;

[0091] If it is determined that |X2-X1|>K and |Y2-Y1|>K, then proceed to step S303;

[0092] If it is determined that |X2-X1|≤K, and / or |Y2-Y1|≤K, proceed to step S306.

[0093] Step S301 mainly targets scenarios with a large forecast area. For example, it forecasts for the whole province of a certain province or the entire territory of a certain country. When the forecast area is small, data preprocessing for regional segmentation can be skipped, and forecasting can be directly performed.

[0094] Step S303, execute the regional segmentation strategy;

[0095] Assume that the size of each region after segmentation is K×K. When |X2 - X1| > K and |Y2 - Y1| > K, execute the regional segmentation strategy. The regional segmentation strategy is different in the training stage and the inference forecasting stage.

[0096] In the training stage, in order to expand the data volume as much as possible, the data preprocessing strategy for constructing the training dataset is as follows: Figure 4 For the schematic diagram of segmenting region E according to an optional embodiment of the present application ( Figure 4 Region E is segmented into eight small regions 1 - 8), as Figure 4 shown, move a window of size K×K across the entire region Ε with a step size of s (i.e., the preset step size) each time, obtaining regions (i.e., multiple first forecast sub-regions). The data of each small region at each moment t is used as a sample. When the data sample at moment t is used as the input x of the forecast model (i.e., the target forecast model), then the data sample at moment t + 1 corresponds to the label of the output y of the forecast model for model supervision. When the time span of the complete data is [T1, T2], through this strategy, samples can be constructed as the training dataset.

[0097] In the inference forecasting stage, it is necessary to minimize the computational amount and speed up the forecasting speed. Figure 5 is the schematic diagram of segmentation in the inference forecasting stage according to an optional embodiment of the present application ( Figure 5 Region E is segmented into six small regions), as Figure 5 shown: Move a window of size K×K across the entire region Ε with a step size of K each time, obtaining small regions (the schematic diagram is shown as the gray region in [[ID=3l]] Figure 3 ). Specifically, when or , take (X2 - K, Y1)-(X2, Y1 + K), (X2 - K, Y1 + K)-(X2, Y1 + 2×K), ……, (X2 - K, Y1 + (n - 1)×K)-(X2, Y1 + n×K), (X2 - K, Y2 - K)-(X2, Y2) as the segmentation regions of the parts that cannot be completely segmented (where, Y1 + n×K < Y2). Obtain the forecast results of each small region after segmentation through the forecast model, and splice the forecast results in the segmentation order to obtain the forecast result of the entire region Ε. Specifically, when or When there is overlap in the edge region, the average of the forecast results for the overlapping part is used as the output forecast result.

[0098] Step S304: Obtain the training dataset.

[0099] The training dataset is obtained by segmenting region E using the method described above during the training phase.

[0100] Next, using the feature representation method and prediction model constructed below, we can achieve prediction for each small region. Therefore, the corresponding task can be defined as: given a sample x of size K×K in any region within the dataset, x∈R layers×K×K Where layers represent the height layer, assuming the sample corresponds to time t, then by constructing model f, we obtain f(x) = y, where y represents the predicted value within a region of size K×K at the same location as x at time t+1.

[0101] 2. A method for mining regional characteristics of wind fields in different areas based on comparative learning.

[0102] Step S305: Design the agent task, construct the positive sample set and the negative sample set, and train the base model F (i.e. the first model) based on contrastive learning.

[0103] Contrastive learning is essentially a deep learning sample construction and training method. Therefore, the first step of contrastive learning is to construct positive and negative samples for each sample by constructing surrogate tasks. By narrowing the gap between the sample and the positive sample and widening the gap between the sample and the negative sample during training, the model learns the distribution of the samples in the representation space.

[0104] Taking any sample α (i.e., the central sample data) in the dataset constructed using the above technical solution as an example, assuming that the time corresponding to α is t, and the geographical area is... The area is K×K. A positive sample is defined as follows: for any sample β in the dataset, if the geographical area corresponding to β is exactly [missing value], then [missing value]. Then β is a positive sample of α. All positive samples of α obtained from the dataset are called the positive sample set of α. A negative sample corresponding to sample α is defined as: for any sample θ in the dataset, if the geographical region corresponding to θ is... The overlapping area is smaller than μ can be set to 4, or other positive real numbers, then θ is the negative sample of α. All negative samples of α obtained from the dataset are called the negative sample set of α.

[0105] The training process of contrastive learning requires reducing the gap between a sample and its corresponding positive sample and increasing the gap between a sample and its corresponding negative sample. Therefore, the Information Noise-Contrastive Estimation (infoNCE) loss function is used, with the following formula:

[0106]

[0107] Where α represents any sample, This indicates the base model used for contrastive learning. Examples include the Shifted Window Transformer (Swin-Transformer) and Residual Networks (ResNet). Represents the characterization of α, α + Let S(α) denote any positive sample of α, and let S(α) denote the set of negative samples of α. - Let S(α) denote any negative sample of α, and τ be a temperature parameter used to adjust the degree of attention given to difficult samples. That is, the smaller the temperature parameter, the more attention is paid to separating the corresponding original samples, positive samples, and negative samples from the most similar difficult samples, thereby obtaining a more uniform representation.

[0108] The training is primarily for training the base model used in contrastive learning. In each training iteration, a sample α is randomly selected, and a positive sample α is formed from sample α. + And N negative samples S(α) of sample α, to obtain the loss. Then gradient calculation is performed. After training is complete, a trained base model will be obtained. go through The resulting representation can include the regional characteristics of the wind field.

[0109] Step S306: Construct the wind field forecast model (i.e., the target forecast model);

[0110] A well-trained base model F has the ability to represent regional characteristics; however, it will correspondingly weaken the ability to represent general meteorological fields. Figure 6 This is a schematic diagram of the wind field forecasting model architecture according to an optional embodiment of this application, such as... Figure 6 As shown, a single Swin-Transformer layer (i.e., the second model) is used to learn the general meteorological field representation M_G. Then, the general representation is concatenated with the regional representation M_R obtained through F, and then connected to a convolution-based UNet network (U-shaped network). Finally, the forecast result is output (when the forecast area is small, M_R can be represented by a zero vector, and the overall model architecture remains unchanged).

[0111] Step S307: Complete wind field forecasting.

[0112] In summary, the optional embodiments of this application propose a region segmentation training wind field forecasting strategy for scenarios with large forecast areas, including a region segmentation and sample expansion strategy in the training phase and a region segmentation forecasting strategy in the inference phase; and propose a comparative learning-based method for mining regional features and wind field representations of different regions, which can mine regional features of wind field in different regions, with a key feature being the proposal of a new surrogate task to construct positive and negative samples; and propose a wind field forecasting model that combines comparative learning representation features containing regional features with general features.

[0113] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0114] This embodiment also provides a training device for a prediction model, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0115] Figure 7 This is a structural block diagram of a training device for a prediction model according to an embodiment of this application, such as... Figure 7 As shown, the device includes:

[0116] The segmentation module 72 is used to segment the target forecast region into multiple first forecast sub-regions by traversing the target forecast region through a sliding window with a preset step size when the size of the first region of the target forecast region is determined to be greater than a preset threshold.

[0117] Construction module 74 is used to construct a target forecast model based on a first model and a second model, wherein the first model is used to characterize the regional characteristics of each first forecast sub-region, and the second model is used to characterize the general characteristics of the target forecast region;

[0118] The input module 76 is used to train the target forecast model by taking the first forecast data of the plurality of first forecast sub-regions at a first time as input data and the second forecast data of the plurality of first forecast sub-regions at a second time as output data, wherein the second time is the time after the first time.

[0119] Using the aforementioned apparatus, when the size of the first region of the target forecast area is determined to be greater than a preset threshold, a sliding window with a preset step size traverses the target forecast area to divide it into multiple first forecast sub-regions. A target forecast model is constructed based on a first model representing the regional characteristics of each first forecast sub-region and a second model representing the general characteristics of the target forecast area. First forecast data of the multiple first forecast sub-regions at a first time step are used as input data, and second forecast data of the multiple first forecast sub-regions at the next time step are used as output data to train the target forecast model. In other words, this application divides the target forecast area into multiple first forecast sub-regions and constructs a target forecast model based on a first model representing the regional characteristics of each first forecast sub-region and a second model representing the general characteristics of the target forecast area, thereby training the target forecast model. According to the embodiments of this application, the problem of poor generalization and insufficient forecast accuracy caused by forecast models in related technologies predicting the forecast area as a whole can be solved. This improves the generalization of the target forecast model and enhances forecast accuracy.

[0120] In an exemplary embodiment, the construction module 74 is further configured to: construct a positive sample set and a negative sample set corresponding to each first forecast data as the center sample data; calculate a first similarity between the center sample data and each positive sample in the positive sample set, and calculate a second similarity between the center sample data and each negative sample in the negative sample set; input the first similarity and the second similarity into a target loss function to obtain a loss value corresponding to the first model; and perform gradient calculation on the model parameters in the first model according to the loss value to train the first model.

[0121] In an exemplary embodiment, the construction module 74 is further configured to determine a second forecast sub-region corresponding to the central sample data, and to determine each third forecast sub-region corresponding to the other forecast data besides the central sample data among the plurality of first forecast data, wherein the plurality of first forecast sub-regions include: the second forecast sub-region and the plurality of third forecast sub-regions; determine a fourth forecast sub-region among the plurality of third forecast sub-regions whose area of ​​a first overlapping region with the second forecast sub-region is greater than or equal to a preset area, and determine the forecast data corresponding to the fourth forecast sub-region as the positive sample data in the positive sample set, wherein the other forecast data includes: the forecast data corresponding to the fourth forecast sub-region; and construct the positive sample set based on the positive sample data.

[0122] In an exemplary embodiment, the construction module 74 is further configured to determine a second forecast sub-region corresponding to the central sample data, and to determine each third forecast sub-region corresponding to the other forecast data (excluding the central sample data) among the plurality of first forecast data, wherein the plurality of first forecast sub-regions include: the second forecast sub-region and the plurality of third forecast sub-regions; to determine a fifth forecast sub-region among the plurality of third forecast sub-regions whose area of ​​a first overlapping region with the second forecast sub-region is less than a preset area, and to determine the forecast data corresponding to the fifth forecast sub-region as the negative sample data in the negative sample set, wherein the other forecast data includes: the forecast data corresponding to the fifth forecast sub-region; and to construct the negative sample set based on the negative sample data.

[0123] In an exemplary embodiment, the input module 76 is further configured to acquire the third forecast data of the plurality of first forecast sub-regions at the current time, and input the plurality of third forecast data into the trained target forecast model, so that the target forecast model predicts the fourth forecast data of each first forecast sub-region at the third time, wherein the third time is the time after the current time; and concatenate the plurality of fourth forecast data to obtain the predicted forecast data of the target forecast region at the third time.

[0124] In an exemplary embodiment, the input module 76 is further configured to, when determining that it is necessary to predict the fifth forecast data of the region to be predicted in the target forecast region at a third time, determine whether the size of the second region to be predicted is greater than the preset threshold, wherein the third time is the time after the current time; and when determining that the size of the second region is less than or equal to the preset threshold, input the sixth forecast data of the region to be predicted at the current time into the trained target forecast model to obtain the fifth forecast data.

[0125] In an exemplary embodiment, the input module 76 is further configured to, when determining that the size of the second region is greater than the preset threshold, traverse the region to be predicted using a sliding window with a preset step size to divide the region to be predicted into multiple second prediction sub-regions; input the seventh prediction data of the multiple second prediction sub-regions at the current time into the trained target prediction model to obtain the eighth prediction data of each second prediction sub-region at the third time, and determine whether there is a second overlapping region among the multiple second prediction sub-regions; when determining that there is a second overlapping region among the multiple second prediction sub-regions, determine the average value of the eighth prediction data of the multiple third prediction sub-regions corresponding to the second overlapping region at the third time as the prediction data of the second overlapping region at the third time, wherein the multiple second prediction sub-regions include: the multiple third prediction sub-regions; and concatenate the multiple eighth prediction data and the prediction data of the second overlapping region at the third time to obtain the fifth prediction data.

[0126] It should be noted that the above modules can be implemented by software or hardware. For the latter, they can be implemented in the following ways, but are not limited to: all the above modules are located in the same processor; or, the above modules are located in different processors in any combination.

[0127] Embodiments of this application also provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above method embodiments when run.

[0128] Optionally, in this embodiment, the storage medium may be configured to store program code for performing the following steps:

[0129] S1, when the size of the first region of the target forecast area is determined to be greater than a preset threshold, the target forecast area is traversed by a sliding window based on a preset step size to divide the target forecast area into multiple first forecast sub-regions;

[0130] S2, construct a target forecast model based on the first model and the second model, wherein the first model is used to characterize the regional characteristics of each first forecast sub-region, and the second model is used to characterize the general characteristics of the target forecast region;

[0131] S3, the target forecast model is trained by taking the first forecast data of the multiple first forecast sub-regions at the first time as input data and the second forecast data of the multiple first forecast sub-regions at the second time as output data, wherein the second time is the time after the first time.

[0132] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0133] Embodiments of this application also provide an electronic device, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0134] In one exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor and the input / output device is connected to the processor.

[0135] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program:

[0136] S1, when the size of the first region of the target forecast area is determined to be greater than a preset threshold, the target forecast area is traversed by a sliding window based on a preset step size to divide the target forecast area into multiple first forecast sub-regions;

[0137] S2, construct a target forecast model based on the first model and the second model, wherein the first model is used to characterize the regional characteristics of each first forecast sub-region, and the second model is used to characterize the general characteristics of the target forecast region;

[0138] S3, the target forecast model is trained by taking the first forecast data of the multiple first forecast sub-regions at the first time as input data and the second forecast data of the multiple first forecast sub-regions at the second time as output data, wherein the second time is the time after the first time.

[0139] Embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above method embodiments.

[0140] Embodiments of this application also provide another computer program product, including a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in any of the above method embodiments.

[0141] Embodiments of this application also provide a computer program that includes computer instructions stored in a computer-readable storage medium; a processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps in any of the above method embodiments.

[0142] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program:

[0143] S1, when the size of the first region of the target forecast area is determined to be greater than a preset threshold, the target forecast area is traversed by a sliding window based on a preset step size to divide the target forecast area into multiple first forecast sub-regions;

[0144] S2, construct a target forecast model based on the first model and the second model, wherein the first model is used to characterize the regional characteristics of each first forecast sub-region, and the second model is used to characterize the general characteristics of the target forecast region;

[0145] S3, the target forecast model is trained by taking the first forecast data of the multiple first forecast sub-regions at the first time as input data and the second forecast data of the multiple first forecast sub-regions at the second time as output data, wherein the second time is the time after the first time.

[0146] Specific examples in this embodiment can be found in the examples described in the above embodiments and exemplary implementations, and will not be repeated here.

[0147] Obviously, those skilled in the art should understand that the modules or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. They can be implemented using computer-executable program code, and thus can be stored in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0148] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A method for training a forecasting model, characterized in that, include: If the size of the first region of the target forecast area is determined to be greater than a preset threshold, the target forecast area is traversed by a sliding window with a preset step size to divide the target forecast area into multiple first forecast sub-regions. A target forecasting model is constructed based on the first model and the second model, wherein the first model is used to characterize the regional characteristics of each first forecasting sub-region, and the second model is used to characterize the general characteristics of the target forecasting region; The target forecast model is trained by using the first forecast data of the plurality of first forecast sub-regions at a first time as input data and the second forecast data of the plurality of first forecast sub-regions at a second time as output data, wherein the second time is the time after the first time.

2. The training method for the prediction model according to claim 1, characterized in that, Before constructing the target prediction model based on the first model and the second model, the method further includes: Each first forecast data point is used as the central sample data point, and a positive sample set and a negative sample set corresponding to the central sample data point are constructed based on the multiple first forecast data points. Calculate the first similarity between the center sample data and each positive sample in the positive sample set, and calculate the second similarity between the center sample data and each negative sample in the negative sample set; The first similarity and the second similarity are input into the target loss function to obtain the loss value corresponding to the first model; The gradient of the model parameters in the first model is calculated based on the loss value in order to train the first model.

3. The training method for the prediction model according to claim 2, characterized in that, Based on the multiple first forecast data, a positive sample set corresponding to the central sample data is constructed, including: A second forecast sub-region corresponding to the central sample data is determined, and a third forecast sub-region corresponding to each of the other forecast data (excluding the central sample data) among the plurality of first forecast data is determined, wherein the plurality of first forecast sub-regions include: the second forecast sub-region and the plurality of third forecast sub-regions; In the plurality of third forecast sub-regions, a fourth forecast sub-region is determined whose area of ​​the first overlapping region with the second forecast sub-region is greater than or equal to a preset area, and the forecast data corresponding to the fourth forecast sub-region is determined as the positive sample data in the positive sample set, wherein the other forecast data includes: the forecast data corresponding to the fourth forecast sub-region; The positive sample set is constructed based on the positive sample data.

4. The training method for the prediction model according to claim 2, characterized in that, Based on the multiple first forecast data, a negative sample set corresponding to the central sample data is constructed, including: A second forecast sub-region corresponding to the central sample data is determined, and a third forecast sub-region corresponding to each of the other forecast data (excluding the central sample data) among the plurality of first forecast data is determined, wherein the plurality of first forecast sub-regions include: the second forecast sub-region and the plurality of third forecast sub-regions; In the plurality of third forecast sub-regions, a fifth forecast sub-region is determined whose area of ​​the first overlapping region with the second forecast sub-region is smaller than a preset area, and the forecast data corresponding to the fifth forecast sub-region is determined as the negative sample data in the negative sample set, wherein the other forecast data includes: the forecast data corresponding to the fifth forecast sub-region; The negative sample set is constructed based on the negative sample data.

5. The training method for the prediction model according to claim 1, characterized in that, After training the target prediction model, the method further includes: The third forecast data of the plurality of first forecast sub-regions at the current time are obtained, and the plurality of third forecast data are input into the trained target forecast model so that the target forecast model predicts the fourth forecast data of each first forecast sub-region at the third time, wherein the third time is the next time after the current time; Multiple fourth forecast data are spliced ​​together to obtain the predicted forecast data of the target forecast area at the third time.

6. The training method for the prediction model according to claim 1, characterized in that, After training the target prediction model, the method further includes: If it is determined that the fifth forecast data of the region to be predicted in the target forecast region needs to be predicted at the third time, it is determined whether the size of the second region of the region to be predicted is greater than the preset threshold, wherein the third time is the time after the current time; If the size of the second region is determined to be less than or equal to the preset threshold, the sixth forecast data of the region to be predicted at the current time is input into the trained target forecast model to obtain the fifth forecast data.

7. The training method for the prediction model according to claim 6, characterized in that, After determining whether the size of the second region of the region to be predicted is greater than the preset threshold, the method further includes: If it is determined that the size of the second region is greater than the preset threshold, the region to be predicted is traversed by a sliding window based on the preset step size, so as to divide the region to be predicted into multiple second prediction sub-regions; The seventh forecast data of the plurality of second forecast sub-regions at the current time are input into the trained target forecast model to obtain the eighth forecast data of each second forecast sub-region at the third time, and to determine whether there is a second overlapping region among the plurality of second forecast sub-regions; If it is determined that there is a second overlapping region in the plurality of second forecast sub-regions, the average value of the eighth forecast data of the plurality of third forecast sub-regions corresponding to the second overlapping region at the third time is determined as the forecast data of the second overlapping region at the third time, wherein the plurality of second forecast sub-regions include the plurality of third forecast sub-regions; The forecast data of the fifth forecast is obtained by splicing together multiple eighth forecast data and the forecast data of the second overlapping region at the third time.

8. A training device for a forecasting model, characterized in that, include: The segmentation module is used to segment the target forecast region into multiple first forecast sub-regions by traversing the target forecast region through a sliding window with a preset step size when the size of the first region of the target forecast region is determined to be greater than a preset threshold. A construction module is used to construct a target forecast model based on a first model and a second model, wherein the first model is used to characterize the regional characteristics of each first forecast sub-region, and the second model is used to characterize the general characteristics of the target forecast region; The input module is used to train the target forecast model by taking the first forecast data of the plurality of first forecast sub-regions at a first time as input data and the second forecast data of the plurality of first forecast sub-regions at a second time as output data, wherein the second time is the time after the first time.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein the program, when executed, performs the method of any one of claims 1 to 7.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the method of any one of claims 1 to 7 through the computer program.

Images