Intelligent prediction method, device and equipment for two-dimensional effective wave height of typhoon wave

Abstract

This invention provides an intelligent prediction method, apparatus, and device for two-dimensional typhoon wave significant height. The method includes: acquiring forecast wind field data and historical wave field data for a target sea area; preprocessing the forecast wind field data and historical wave field data to obtain normalized wind field data and normalized wave field data; inputting the normalized wind field data and normalized wave field data into a trained prediction model to obtain the predicted two-dimensional typhoon wave significant height; wherein the prediction model is trained using a training dataset and validated using a validation dataset, and the training dataset and the validation dataset include independent grade-level forecast wind field data and historical wave field data. This solution, by optimizing the training dataset and providing a prediction model that balances speed and effectiveness, can reduce computation time, meet the real-time forecasting requirements of operations, and achieve higher accuracy in two-dimensional significant wave height forecasts under extreme weather conditions such as typhoons.

Description

Technical Field

[0001] This invention relates to the field of wave prediction technology, and in particular to an intelligent prediction method, device and equipment for two-dimensional typhoon wave significant height. Background Technology

[0002] Predicting the significant wave height of ocean waves is of great importance in fields such as marine engineering and meteorological disaster early warning. Traditional wave prediction methods mainly rely on numerical models, such as the third-generation wave model (WAVEWATCH III). While these models can provide relatively comprehensive wave information, they are computationally complex and time-consuming. In recent years, machine learning methods have been introduced into the field of wave prediction. However, due to the scarcity of training samples under extreme weather conditions, conventional machine learning models struggle to accurately learn the complex nonlinear relationship between wind and wave fields under severe sea states such as typhoons. This results in generally low predictions of the extreme values ​​of significant wave height, which is precisely a key requirement for disaster prevention and mitigation.

[0003] In existing technologies, those based on Convolutional Neural Networks (CNNs) or Long Short-Term Memory (LSTM) networks... While LSTM (Laser-Sensitive Memory) prediction models have improved prediction efficiency to some extent, they still suffer from limitations in the receptive field or insufficient capture of long-range dependencies when processing two-dimensional spatial data. Furthermore, these models are poorly adapted to extreme weather events such as typhoons, and their predictions often deviate significantly from actual observations. Therefore, there is an urgent need for an intelligent wave prediction method that can balance computational efficiency and prediction accuracy. Summary of the Invention

[0004] This invention provides an intelligent prediction method, device, and equipment for two-dimensional typhoon wave significant wave height, which solves the problem of insufficient prediction accuracy of sea wave significant wave height under extreme weather conditions.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: This invention provides an intelligent prediction method for two-dimensional typhoon wave significant height, comprising: Obtain forecast wind field data and historical wave field data for the target sea area; The predicted wind field data and the historical wave field data are preprocessed to obtain normalized wind field data and normalized wave field data, respectively. The normalized wind field data and normalized wave field data are input into the trained prediction model to obtain the predicted two-dimensional typhoon wave significant height. The prediction model is trained using a training dataset and then validated using a validation dataset. The training dataset and the validation dataset include independent grade-level forecast wind field data and historical wave field data.

[0006] Optionally, obtain forecast wind field data and historical wave field data for the target sea area, including: Based on the forecast wind field of the global forecast system, the forecast wind field data for the target sea area is obtained; Historical wave field data for the target sea area were obtained based on the reanalysis of the wave field from the National Marine Environmental Forecasting Center.

[0007] Optionally, the predicted wind field data and the historical wave field data are preprocessed to obtain normalized wind field data and normalized wave field data, including: The predicted wind field data and the historical wave field data are resampled onto a target regular grid; the predicted wind field data and the historical wave field data exist at spatial intervals of [missing information]. The original mesh; The time resolution of the predicted wind field data and the historical wave field data on the target rule grid is unified to hourly to obtain normalized wind field data and normalized wave field data.

[0008] Optionally, resampling the forecast wind field data and the historical wave field data onto the target regular grid includes: Create a spatial interval within the target sea area. A uniform, regular grid is used as the target grid; The predicted wind field data and historical wave field data on the original grid are spatially mapped to the grid points to be determined on the target grid to obtain the target regular grid; the spatial interval of the target regular grid is... .

[0009] Optionally, the predicted wind field data and the historical wave field data on the original grid are spatially mapped to the grid points to be determined on the target grid to obtain a target regular grid, including: Locate the position of the grid point to be determined on the original grid, and obtain the original grid points in the four directions closest to the grid point to be determined on the original grid; Based on the original grid points, in the longitude direction... and Thus, the first intermediate point and the second intermediate point are obtained; Based on the first midpoint and the second midpoint, in the latitudinal direction... This yields the desired grid points on the target regular grid. The target rule grid is obtained based on the grid points to be determined on each target rule grid. Among them, the This indicates that the grid point to be determined is located at the lower left corner of the original grid. , ); the This indicates that the grid point to be determined is located at the lower right corner of the original grid. , ); the This indicates that the grid point to be determined is located at the top left corner of the original grid. , ); the This indicates that the grid point to be determined is located at the upper right corner of the original grid. , ); and These represent the first intermediate point and the second intermediate point, respectively. Represents the lattice point to be determined ( , ); t represents the normalized distance from the left side line in the longitude direction, i.e. ; s represents the normalized distance from the base line in the latitudinal direction, i.e. .

[0010] Optionally, the normalized wind field data and normalized wave field data are input into the trained prediction model to obtain the predicted two-dimensional typhoon wave significant height, including: The normalized wind field data and normalized wave field data are segmented to obtain spatially complete and independent image blocks; Features are extracted from the image patch to obtain a multi-scale spatial feature map; The multi-scale spatial feature map is reconstructed onto the target regular grid to obtain the predicted two-dimensional typhoon wave height.

[0011] Optional, the training process for the predictive model includes: Obtain the actual wave height data in the target sea area corresponding to the training dataset and the validation dataset; The predicted wind field data and historical wave field data in the training dataset are input into the prediction model in batches to obtain the training predicted wave height; The prediction model error is obtained by comparing the predicted wave height from the training with the actual wave height data. The parameters of the prediction model are updated based on the prediction model error to obtain the trained prediction model. The predicted wind field data and historical wave field data in the validation dataset are input into the trained prediction model in batches to obtain the validation predicted wave height. When the similarity between the verified predicted wave height and the actual wave height data is greater than a preset threshold, the trained prediction model is output.

[0012] This invention also provides an intelligent prediction device for two-dimensional typhoon wave significant height, comprising: The acquisition module is used to acquire forecast wind field data and historical wave field data for the target sea area; The processing module is used to preprocess the forecast wind field data and the historical wave field data respectively to obtain normalized wind field data and normalized wave field data; input the normalized wind field data and normalized wave field data into the trained prediction model to obtain the predicted two-dimensional typhoon wave significant height; wherein, the prediction model is trained on the training dataset and the prediction model is verified on the validation dataset after training, and the training dataset and the validation dataset include independent grade-level forecast wind field data and historical wave field data.

[0013] This invention also provides a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when run by the processor, executes the above-described method.

[0014] This invention also provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described method.

[0015] The technical solution of the present invention has at least the following effects: The above-mentioned solution of the present invention acquires forecast wind field data and historical wave field data of the target sea area; preprocesses the forecast wind field data and the historical wave field data respectively to obtain normalized wind field data and normalized wave field data; inputs the normalized wind field data and normalized wave field data into a trained prediction model to obtain the predicted two-dimensional typhoon wave significant height; wherein, the prediction model is trained using a training dataset, and the prediction model is validated using a validation dataset after training, and the training dataset and the validation dataset include independent grade-level forecast wind field data and historical wave field data. This can reduce computation time, meet the real-time forecasting requirements of operations, and achieve higher accuracy in two-dimensional significant wave height forecasting under extreme weather conditions such as typhoons, realizing full-process intelligent prediction of sea wave height. Attached Figure Description

[0016] Figure 1 This is a flowchart of the intelligent prediction method for two-dimensional typhoon wave significant height provided in an embodiment of the present invention; Figure 2 This is a structural diagram of the intelligent prediction device for two-dimensional typhoon wave effective wave height provided in an embodiment of the present invention; Figure 3This is a schematic diagram of the structure of the computing device provided in an embodiment of the present invention; Figure 4 This is a display of real ocean wave data and wave field provided by the embodiments of the present invention; Figure 5 The wave field data is predicted using the conventional model provided in this embodiment of the invention. Figure 6 The wave field data is predicted by the prediction model provided in this embodiment of the invention. Detailed Implementation

[0017] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0018] like Figure 1 As shown, embodiments of the present invention propose an intelligent prediction method for two-dimensional typhoon wave significant height, comprising: Step 11: Obtain forecast wind field data and historical wave field data for the target sea area; Step 12: Preprocess the predicted wind field data and the historical wave field data respectively to obtain normalized wind field data and normalized wave field data; Step 13: Input the normalized wind field data and normalized wave field data into the trained prediction model to obtain the predicted two-dimensional typhoon wave significant height; The prediction model is trained using a training dataset and then validated using a validation dataset. The training dataset and the validation dataset include independent grade-level forecast wind field data and historical wave field data.

[0019] In this embodiment, in step 11, the forecast wind field data and historical wave field data of the target sea area are obtained respectively: the forecast wind field data is the wind field prediction data of the target sea area during the forecast period, including zonal wind and meridional wind; the historical wave field data is the wave data of the target sea area for a period of time before the forecast period, including significant wave height, average wave direction and average period, etc.; the historical wave field data provides the wave basis for the forecast period, and the wave field of the target sea area during the forecast period can be obtained by superimposing the wind force changes of the forecast wind field data on the wave basis.

[0020] In step 12, the original resolution of the forecast wind field data and the historical wave field data is 0.25°×0.25°. However, the original resolution cannot accurately capture the small-scale and tightly structured weather system of typhoons. It is easy for key structures such as the typhoon eye and the dense spiral rainbands to be smoothed out in the low-resolution data, resulting in bias in the information input to the model. Therefore, it is necessary to increase the resolution from 0.25°×0.25° to 0.1°×0.1° so that the model can receive richer and more detailed information on the typhoon wind field structure and ocean wave gradient. In addition, it is necessary to preprocess the zonal wind, meridional wind and historical wave field data into a unified multi-channel tensor in the channel dimension, and then scale the data to make it fall into a unified numerical range. Time alignment is used to ensure that the input "historical wave field" time series corresponds precisely to the start time of the "forecast wind field".

[0021] In step 13, the preprocessed standard data can be processed by a pre-trained prediction model to predict ocean wave information within the target time period. The prediction model is trained using a training dataset and then validated using a validation dataset. Once validation is successful, the trained prediction model is output for use in the actual prediction process. Simultaneously, the parameters of the prediction model can be continuously corrected by comparing each actual prediction with subsequent actual ocean wave field data. The training and validation datasets include non-repeating forecast wind field data and historical wave field data for different years. For example, the training dataset might use data from 2023, the validation dataset from 2024, and the actual prediction process could predict data from 2025. Figures 4 to 6 As shown, Figure 4 For actual ocean wave data at a specific moment, Figure 5 This refers to wave field data predicted using a traditional model. Figure 6 The attached figure shows the wave field data predicted by this prediction model. As can be clearly seen from the figure, compared with the traditional model, the wave field prediction data of this prediction model is closer to the actual wave data, which can solve the problem that the traditional model predicts the extreme value of significant wave height too low.

[0022] This technical solution improves the prediction accuracy and computational efficiency of typhoon waves, effectively overcomes the inherent defect of traditional machine learning models that predict the extreme value of significant wave height too low, and can efficiently extract the long-range spatial dependence between wind field and wave field, thereby achieving accurate prediction of the spatial distribution pattern of sea wave field in a large area of ​​sea.

[0023] In an optional embodiment of the present invention, step 11, obtaining forecast wind field data and historical wave field data for the target sea area, may include: Step 111: Obtain the forecast wind field data for the target sea area based on the forecast wind field of the global forecast system; Step 112: Obtain historical wave field data for the target sea area based on the reanalysis wave field from the National Marine Environmental Forecasting Center.

[0024] In this embodiment, in step 111, forecast wind field data from the Global Forecast System (GFS) can be obtained through the online data service platform provided by the official website. Through the graphical interface and simple query tools provided by the system, data retrieval can be performed by selecting parameters such as the geographical location and time range of the target sea area as needed.

[0025] In step 112, the National Marine Environmental Forecasting Center will regularly release reanalysis wave field data. The reanalysis wave field refers to a set of historical wave field datasets with high accuracy and consistency generated by reprocessing and analyzing historical observation data and combining them with advanced numerical models. These datasets integrate observation data from multiple sources, including buoy observations and satellite remote sensing data. After strict quality control and data assimilation processing, they can more accurately reflect the characteristics of ocean surface waves over a period of time and can be used as a benchmark for predicting ocean wave fields.

[0026] In an optional embodiment of the present invention, step 12, which involves preprocessing the predicted wind field data and the historical wave field data to obtain normalized wind field data and normalized wave field data, may include: Step 121: Resample the predicted wind field data and the historical wave field data onto the target regular grid; the predicted wind field data and the historical wave field data exist at spatial intervals of... The original mesh; Step 122: Unify the time resolution of the predicted wind field data and the historical wave field data on the target rule grid to hourly to obtain normalized wind field data and normalized wave field data.

[0027] In this embodiment, in step 121, the original resolution of both the forecast wind field data and the historical wave field data is [missing information]. That is, the predicted wind field data and the historical wave field data exist at a spatial interval of [missing information]. The original mesh has a low resolution, meaning it lacks detail. When a prediction model is input with low-resolution original data, significant detail loss and smoothing of small-scale features result in a prediction that deviates further from the true value after subsequent processing. Therefore, it is necessary to reduce the original resolution of the data. Upgraded to This improves the accuracy of the data input into the prediction model.

[0028] In step 122, by unifying the time resolution to hourly, the wind field data and wave field data can be correlated, so that the model can know the evolution of the ocean waves in the future time series based on the continuous historical wave field data and the forecast wind field data.

[0029] In an optional embodiment of the present invention, step 121, resampling the forecast wind field data and the historical wave field data onto the target regular grid, may include: Step 1211, create a spatial interval of within the target sea area. A uniform, regular grid is used as the target grid; Step 1212: Map the predicted wind field data and historical wave field data on the original grid to the desired grid points on the target grid to obtain a target regular grid; the spatial interval of the target regular grid is... .

[0030] In this embodiment, in step 1211, a brand-new virtual grid system is created as the target grid within the geographical boundary of the target sea area. The grid consists of grid points that are evenly distributed in the longitude and latitude directions, with each grid point having a fixed interval of 0.1 degrees in the longitude and latitude directions.

[0031] In step 1212, the book data on the original grid is mapped onto the target grid to obtain a target regular grid. This results in normalized wind field data and normalized wave field data with a unified spatial resolution of 0.1°×0.1° and perfectly aligned grid points. This effectively eliminates grid differences from the original data source and provides the model with more refined spatial input features. The target regular grid is the grid containing the predicted wind field data and the historical wave field data.

[0032] In an optional embodiment of the present invention, step 1212, mapping the predicted wind field data and the historical wave field data on the original grid to the desired grid points of the target grid to obtain a target regular grid, may include: Step 12121: Locate the position of the grid point to be determined on the original grid, and obtain the original grid points in the four directions closest to the grid point to be determined on the original grid; Step 12122: Based on the original grid points, pass through the longitude direction... and Thus, the first intermediate point and the second intermediate point are obtained; Step 12123: Based on the first intermediate point and the second intermediate point, traverse the latitude direction... This yields the desired grid points on the target regular grid. Step 12124: Obtain the target regular grid based on the grid points to be determined on each target regular grid. Among them, the This indicates that the grid point to be determined is located at the lower left corner of the original grid. , ); the This indicates that the grid point to be determined is located at the lower right corner of the original grid. , ); the This indicates that the grid point to be determined is located at the top left corner of the original grid. , ); the This indicates that the grid point to be determined is located at the upper right corner of the original grid. , ); and These represent the first intermediate point and the second intermediate point, respectively. Represents the lattice point to be determined ( , ); t represents the normalized distance from the left side line in the longitude direction, i.e. ; s represents the normalized distance from the base line in the latitudinal direction, i.e. .

[0033] In this embodiment, in step 12121, each data point on the original grid is a known data point. In order to find the grid point to be found on the target grid corresponding to the data point on the original grid, it is necessary to first obtain the nearest surrounding grid point corresponding to the data point on the original grid, namely the upper left corner point, the lower left corner point, the upper right corner point, and the lower right corner point. The data point on the original grid is located in the closed space enclosed by the upper left corner point, the lower left corner point, the upper right corner point, and the lower right corner point.

[0034] In step 12122, two intermediate points are obtained by calculating the intermediate point along the longitude direction, which are used for subsequent calculations. The calculation formula is as follows: and The With the That is, the first midpoint and the second midpoint.

[0035] In step 12123, passing through in the latitudinal direction This process yields the desired grid point on the target regular grid, i.e., the corresponding mapped coordinates of a point on the original grid onto the target regular grid. It should be noted that the order of obtaining the longitude and latitude directions in steps 12122 and 12123 is not required. Alternatively, two midpoints in the latitude direction can be obtained in step 12122, and the desired grid point in the longitude direction can be obtained in step 12123. The final coordinates of the desired grid point obtained in both orders should be identical.

[0036] In step 12124, for each data point on the original grid, the corresponding grid point to be determined on the target regular grid is obtained, that is, the target regular grid for each data point is clearly defined.

[0037] In an optional embodiment of the present invention, step 13, inputting the normalized wind field data and normalized wave field data into the trained prediction model to obtain the predicted two-dimensional typhoon wave significant height, may include: Step 131: The normalized wind field data and the normalized wave field data are segmented to obtain spatially complete and independent image blocks; Step 132: Extract features from the image patch to obtain a multi-scale spatial feature map; Step 133: Reconstruct the multi-scale spatial feature map onto the target regular grid to obtain the predicted two-dimensional typhoon wave height.

[0038] In this embodiment, in step 131, the multi-channel input data [batch size, time step, height, width, channel] after spatial and temporal normalization is divided into multiple non-overlapping image blocks in the spatial dimension (height and width). A fixed 5×5 pixel size is used as the size of a single image block. That is, the image segmentation layer in the model slides in the spatial dimension with a step size of 5, dividing the entire image composed of all data into (H / 5)×(W / 5) image blocks. Each 5×5 image block contains information of all channels in the local area (such as future zonal wind, meridional wind, and historical significant wave height sequence). After these image blocks are flattened, they form a sequence with dimensions of [(H / 5)×(W / 5), 5×5×C], converting the two-dimensional spatial data into a visual sequence.

[0039] In step 132, the image patch sequence is input to the feature extraction layer. Through hierarchical processing of multiple stages, spatial features from local to global are gradually extracted and fused. Specifically, this includes: Stage 1 (local feature extraction): The prediction model uses a window self-attention mechanism to calculate the relationship between image patches within each independent local window, enabling the model to efficiently learn the direct, local nonlinear mapping between wind and wave fields within a small sea area; and through a shift window self-attention mechanism, information can be exchanged between adjacent windows, initially expanding the receptive field of the prediction model. At the end of this stage, the block merging layer merges the features of adjacent 2x2 image patches into one, achieving a 4x downsampling, halving the spatial size (H,W), but increasing the number of feature channels, generating the first set of multi-scale feature maps.

[0040] Stage 2 (Global Feature Extraction): The input is the feature map downsampled from Stage 1. The model repeats windowing and shifting window attention calculations at a lower spatial resolution. Since each feature point now represents a larger area of ​​ocean information from the original input, this stage effectively captures long-range spatial dependencies, such as the energy transfer effect of the strong wind zone at the typhoon center on distant swells. After further block merging and downsampling, a higher-level and more abstract feature map is output. Through these two stages, the feature extraction layer outputs a set of multi-scale spatial feature maps, encoding the complex interactions between wind and wave fields at different scales, enabling the model to make accurate predictions.

[0041] In step 133, the multi-scale spatial feature map obtained in step 132 (whose spatial resolution is now much lower than the original input) is processed by the image reconstruction layer to restore its spatial resolution to the same H×W size as the input data, and the final predicted value is output. The specific implementation includes: upsampling, using transposed convolution or pixel shuffling to gradually increase the spatial size of the feature map; convolutional thinning, inserting standard convolutional layers between the upsampling layers to refine the features, eliminate the "checkerboard" artifacts that may be caused by upsampling, and further fuse feature information to make the prediction result spatially smoother and physically more reasonable; and an output layer, finally using a 1×1 convolutional layer to map the number of channels to 1, i.e., predicting a significant wave height value for each grid point. After the above processing, a two-dimensional typhoon wave significant wave height prediction field with a spatial resolution of 0.1°×0.1° is generated, perfectly aligned with the input target regular grid, visually displaying the spatial distribution of wave height in the entire target sea area at a specific future time.

[0042] In an optional embodiment of the present invention, the training process of the prediction model may include: (1) Obtain the actual wave height data in the target sea area corresponding to the training dataset and the validation dataset; (2) Input the forecast wind field data and historical wave field data in the training dataset into the prediction model in batches to obtain the training prediction wave height; (3) Based on the training predicted wave height and the actual wave height data, the prediction model error is obtained; (4) Update the parameters of the prediction model according to the prediction model error to obtain the trained prediction model; (5) Input the forecast wind field data and historical wave field data in the verification dataset into the trained prediction model in batches to obtain the verification predicted wave height; (6) When the similarity between the verified predicted wave height and the actual wave height data is greater than a preset threshold, the trained prediction model is output.

[0043] In this embodiment, the prepared training and validation datasets are used to train and optimize the prediction model, ensuring that it ultimately has reliable and accurate prediction capabilities.

[0044] Obtain the real wave height data in the target sea area corresponding to the training dataset and the validation dataset: For each sample in the training dataset and the validation dataset (i.e., each set of input wind field and past wave field data), there must be corresponding valid wave height data that is observed in the real world or obtained through reanalysis. This is the standard answer for model learning. This data serves as the label in supervised learning, is the target of model learning, and is also the only benchmark for measuring the quality of model prediction results.

[0045] The forecast wind field data and historical wave field data in the training dataset are input into the prediction model in batches to obtain the training predicted wave height. This process is the model learning process. In order to update the model parameters more stably, the training data is divided into multiple small batches. Each time, a small batch of data is taken and input into the model. The data flows from the input layer to the output layer in the model. Finally, the model calculates the training predicted wave height of this batch of data based on its current parameters.

[0046] The prediction model error is obtained by comparing the predicted wave heights from the training data with the actual wave height data. This step evaluates the model's current performance by comparing the prediction results with the corresponding "standard answer." The root mean square error (RMSE) is used to quantify the difference between the predicted and actual values, thus amplifying the impact of larger errors and making the model more suitable for predicting extreme values ​​(such as typhoon wave extremes). This yields a numerical value specifically representing the inaccuracy of the current model's predictions, i.e., the prediction model error. The prediction model error is obtained through... The calculation is performed, where L is the prediction model error, reflecting the overall inaccuracy of the model's predictions under the current parameters; MSE() is the error function; P is the trained predicted wave height; T is the actual wave height data; and N is the total number of valid wave heights at all spatial grid points in a batch. It is the model's prediction of the first Effective wave height of each grid point; It is the true wave height corresponding to the i-th grid point; The prediction model's parameters are updated based on its error to obtain the trained prediction model. This step allows the model to continuously learn and improve itself. Optimization is performed based on the prediction model's error, calculating how to fine-tune millions or even tens of millions of parameters (such as weights and biases) to reduce the error. Following the calculated adjustment direction and magnitude, all model parameters are updated, and this process is repeated cyclically. The calculated error guides the direction and magnitude of model parameter adjustments to reduce future prediction errors, gradually improving the model's predictive ability. Specifically, this includes: calculating the prediction model error. Relative to each trainable parameter of the model Gradients of (e.g., convolutional kernel weights, fully connected layer weights, etc.) ,Right now The gradient This indicates the steepest rising direction of the prediction model error in the parameter space; therefore, its opposite direction... This refers to the direction in which the loss decreases the fastest; updating the model parameters along the opposite direction of the gradient, using the following update formula (taking the most basic stochastic gradient descent as an example): ,in: and These are the values ​​before and after the parameter update, respectively. The learning rate determines the step size for parameter updates; This is the gradient of the parameter. By repeating steps (3) and (4) thousands of times, the model parameters are continuously fine-tuned, reducing the prediction model error. As the value of gradually decreases, the predictive ability of the model gradually improves, eventually resulting in a high-precision prediction model after training.

[0047] The predicted wind field data and historical wave field data from the validation dataset are input into the trained prediction model in batches to obtain the validation predicted wave height. During or after training, the model's true performance needs to be evaluated on a dataset that has never been used before (i.e., the validation dataset). Since the validation dataset does not participate in the model's parameter updates, it can be used to simulate the model's performance on future, unknown data. That is, the validation dataset data is input into the currently trained model to obtain the validation predicted wave height.

[0048] When the similarity between the verified predicted wave height and the actual wave height data is greater than a preset threshold, the trained prediction model is output. The preset threshold is used to measure the similarity between the verified predicted wave height and the actual wave height data. When the difference between the two is less than the preset threshold, it means that the similarity between the two has reached the expected effect. Only when the similarity reaches the expected effect is the model considered to have been successfully trained and can be used for actual business forecasting. If the threshold is not reached, the model structure, training data or hyperparameters need to be adjusted and retrained.

[0049] A specific embodiment of the intelligent prediction method for two-dimensional typhoon wave significant height provided by this invention is as follows: Step 1: Obtain the forecast wind field from GFS and the reanalysis wave field data from the National Marine Environmental Forecasting Center, and upgrade the original spatial resolution of 0.25°×0.25° to 0.1°×0.1°, with a temporal resolution of hourly.

[0050] Step 2: Use data from 2022 to 2023 to generate training samples, and select typhoon events during this period to add more than 100 additional typhoon weather training samples.

[0051] Step 3: Construct a two-dimensional wave prediction model. This model mainly consists of three parts: an image segmentation layer, a feature extraction layer, and a convolutional and upsampling layer. The image segmentation layer divides the input image into 5×5 local ocean area patches (each patch contains 25 pixels). This operation converts the image dimensions to [H / 5, W / 5, 25], providing a foundation for subsequent processing. The feature extraction layer extracts features in layers. This model uses two stages to progressively downsample (4 times) to generate multi-scale feature maps. The window attention mechanism reduces the computational complexity from quadratic to linear, while extracting spatially relevant features of the waves. The convolutional and upsampling layer further abstracts the image feature information and restores the image to its original size. This step also makes the image smoother and eliminates the checkerboard effect.

[0052] Step 4: Input the generated training data into the built model. Each sample is used 1000 times, and each training session uses 10 samples.

[0053] Step 5: Use the severe weather event in 2024 to test the effectiveness of this model, compare it with the prediction results of traditional models, evaluate the superiority of the prediction model, and confirm the credibility of the prediction model.

[0054] Step 6: Using this model, input the wind field for the next 24 hours and the historical wave field for the previous 3 moments, and output the hourly two-dimensional wave height field for the next 1 to 24 hours, thus obtaining the effective wave height of the two-dimensional typhoon waves.

[0055] The intelligent prediction method for two-dimensional typhoon wave significant wave height proposed in this invention improves the data preprocessing process and optimizes the composition of training samples, making the model structure more scientific and reasonable, thereby enhancing the global modeling capability of the model and significantly improving the prediction accuracy of sea waves under strong sea conditions such as typhoons.

[0056] like Figure 2 As shown, this embodiment of the invention also provides a two-dimensional intelligent prediction device 20 for the significant wave height of typhoon waves, comprising: The acquisition module 21 is used to acquire forecast wind field data and historical wave field data for the target sea area; The processing module 22 is used to preprocess the forecast wind field data and the historical wave field data respectively to obtain normalized wind field data and normalized wave field data; input the normalized wind field data and normalized wave field data into the trained prediction model to obtain the predicted two-dimensional typhoon wave significant height; wherein, the prediction model is trained through the training dataset, and the prediction model is verified through the verification dataset after training, and the training dataset and the verification dataset include independent grade-level forecast wind field data and historical wave field data.

[0057] Optionally, module 21 is specifically used for: Obtain the forecast wind field from the global forecasting system to get the forecast wind field data for the target sea area; The reanalysis wave field data from the National Marine Environmental Forecasting Center was obtained to acquire historical wave field data for the target sea area.

[0058] Optionally, processing module 22 is specifically used for: The predicted wind field data and the historical wave field data are resampled onto a target regular grid; the predicted wind field data and the historical wave field data exist at spatial intervals of [missing information]. The original mesh; The time resolution of the predicted wind field data and the historical wave field data on the target rule grid is unified to hourly to obtain normalized wind field data and normalized wave field data.

[0059] Optionally, resampling the forecast wind field data and the historical wave field data onto the target regular grid includes: Create a spatial interval within the target sea area. A uniform, regular grid is used as the target grid; The predicted wind field data and historical wave field data on the original grid are spatially mapped to the grid points to be determined on the target grid to obtain the target regular grid; the spatial interval of the target regular grid is... .

[0060] Optionally, the predicted wind field data and the historical wave field data on the original grid are spatially mapped to the grid points to be determined on the target grid to obtain a target regular grid, including: Locate the position of the grid point to be determined in the original grid, and obtain the original grid points in the four directions closest to the grid point to be determined; Based on the original grid points, in the longitude or latitude direction... and Thus, the first intermediate point and the second intermediate point are obtained; Based on the first midpoint and the second midpoint, pass through in the latitude or longitude direction. This yields the desired grid points on the target regular grid. The target rule grid is obtained based on the grid points to be determined on each target rule grid. Among them, the This indicates that the grid point to be determined is located at the lower left corner of the original grid. , ); the This indicates that the grid point to be determined is located at the lower right corner of the original grid. , ); the This indicates that the grid point to be determined is located at the top left corner of the original grid. , ); the This indicates that the grid point to be determined is located at the upper right corner of the original grid. , ); and These represent the first intermediate point and the second intermediate point, respectively. Represents the lattice point to be determined ( , ); t represents the normalized distance in the longitude direction from the left side line or in the latitude direction from the bottom side line, i.e. or ;s represents the normalized distance in the latitudinal direction from the bottom edge or in the longitude direction from the left edge, i.e. or .

[0061] Optionally, the processing module 22 is also specifically used for: The normalized wind field data and normalized wave field data are segmented to obtain spatially complete and independent image blocks; Features are extracted from the image patch to obtain a multi-scale spatial feature map; The multi-scale spatial feature map is reconstructed onto the target regular grid to obtain the predicted two-dimensional typhoon wave height.

[0062] Optional, the training process for the predictive model includes: Obtain the actual wave height data in the target sea area corresponding to the training dataset and the validation dataset; The predicted wind field data and historical wave field data in the training dataset are input into the prediction model in batches to obtain the training predicted wave height; The prediction model error is obtained by comparing the predicted wave height from the training with the actual wave height data. The parameters of the prediction model are updated based on the prediction model error to obtain the trained prediction model. The predicted wind field data and historical wave field data in the validation dataset are input into the trained prediction model in batches to obtain the validation predicted wave height. When the similarity between the verified predicted wave height and the actual wave height data is greater than a preset threshold, the trained prediction model is output.

[0063] It should be noted that this device is a device corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0064] like Figure 3 As shown, this embodiment of the invention also provides a computing device 30, including a processor 31, a memory 32, and a program or instructions stored in the memory 32 and executable on the processor 31. When the program or instructions are executed by the processor 31, they implement the various processes of the above-described intelligent prediction method embodiment for two-dimensional typhoon wave effective height and achieve the same technical effect. To avoid repetition, they will not be described again here. It should be noted that the computing device in this embodiment of the invention includes the aforementioned mobile electronic devices and non-mobile electronic devices.

[0065] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0066] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0067] In the embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0068] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0069] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0070] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0071] Furthermore, it should be noted that in the apparatus and method of the present invention, it is obvious that the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent solutions of the present invention. Moreover, the steps performing the above series of processes can naturally be executed in the order described, but are not necessarily required to be executed in chronological order; some steps can be executed in parallel or independently of each other. Those skilled in the art will understand that all or any step or component of the method and apparatus of the present invention can be implemented in any computing device (including processors, storage media, etc.) or network of computing devices, in hardware, firmware, software, or a combination thereof. This is something that those skilled in the art can achieve by using their basic programming skills after reading the description of the present invention.

[0072] Therefore, the object of the present invention can also be achieved by running a program or a set of programs on any computing device. The computing device can be a known general-purpose device. Therefore, the object of the present invention can also be achieved simply by providing a program product containing program code for implementing the method or apparatus. That is, such a program product also constitutes the present invention, and the storage medium storing such a program product also constitutes the present invention. Obviously, the storage medium can be any known storage medium or any storage medium developed in the future. It should also be noted that in the apparatus and method of the present invention, it is obvious that the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent to the present invention. Furthermore, the steps for performing the above series of processes can naturally be performed in the order described, but are not necessarily required to be performed in chronological order. Some steps can be performed in parallel or independently of each other.

[0073] The above are preferred embodiments of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An intelligent prediction method of two-dimensional significant wave height of typhoon wave, characterized in that, include: Obtain forecast wind field data and historical wave field data for the target sea area; The predicted wind field data and the historical wave field data are preprocessed to obtain normalized wind field data and normalized wave field data, respectively. The normalized wind field data and normalized wave field data are input into the trained prediction model to obtain the predicted two-dimensional typhoon wave significant height. The prediction model is trained using a training dataset and then validated using a validation dataset. The training dataset and the validation dataset include independent grade-level forecast wind field data and historical wave field data. 2.The intelligent prediction method of two-dimensional significant wave height of typhoon wave according to claim 1, characterized in that, Obtain forecast wind field data and historical wave field data for the target sea area, including: Based on the forecast wind field of the global forecast system, the forecast wind field data for the target sea area is obtained; Historical wave field data for the target sea area were obtained based on the reanalysis of the wave field from the National Marine Environmental Forecasting Center. 3.The intelligent prediction method of two-dimensional typhoon wave significant wave height according to claim 1, characterized in that, The predicted wind field data and the historical wave field data are preprocessed to obtain normalized wind field data and normalized wave field data, including: resampling the forecast wind field data and the historical wave field data onto a target regular grid; the forecast wind field data and the historical wave field data exist in an original grid with a spatial interval of ; The time resolution of the predicted wind field data and the historical wave field data on the target rule grid is unified to hourly to obtain normalized wind field data and normalized wave field data.

4. The intelligent prediction method for two-dimensional typhoon wave significant height according to claim 3, characterized in that, Resampling the predicted wind field data and the historical wave field data onto the target regular grid includes: Create a spatial interval within the target sea area. A uniform, regular grid is used as the target grid; The predicted wind field data and historical wave field data on the original grid are spatially mapped to the grid points to be determined on the target grid to obtain the target regular grid; the spatial interval of the target regular grid is... .

5. The intelligent prediction method for two-dimensional typhoon wave significant height according to claim 4, characterized in that, The predicted wind field data and historical wave field data on the original grid are spatially mapped to the grid points to be determined on the target grid to obtain the target regular grid, including: Locate the position of the grid point to be determined on the original grid, and obtain the original grid points in the four directions closest to the grid point to be determined on the original grid; Based on the original grid points, in the longitude direction... and Thus, the first intermediate point and the second intermediate point are obtained; Based on the first midpoint and the second midpoint, in the latitudinal direction... This yields the desired grid points on the target regular grid. The target rule grid is obtained based on the grid points to be determined on each target rule grid. Among them, the This indicates that the grid point to be determined is located at the lower left corner of the original grid. , ); the This indicates that the grid point to be determined is located at the lower right corner of the original grid. , ); the This indicates that the grid point to be determined is located at the top left corner of the original grid. , ); the This indicates that the grid point to be determined is located at the upper right corner of the original grid. , ); and These represent the first intermediate point and the second intermediate point, respectively. Represents the lattice point to be determined ( , ); t represents the normalized distance from the left side line in the longitude direction, i.e. ; s represents the normalized distance from the base line in the latitudinal direction, i.e. .

6. The intelligent prediction method for two-dimensional typhoon wave significant height according to claim 3, characterized in that, The normalized wind field data and normalized wave field data are input into the trained prediction model to obtain the predicted two-dimensional typhoon wave significant height, including: The normalized wind field data and normalized wave field data are segmented to obtain spatially complete and independent image blocks; Features are extracted from the image patch to obtain a multi-scale spatial feature map; The multi-scale spatial feature map is reconstructed onto the target regular grid to obtain the predicted two-dimensional typhoon wave height.

7. The intelligent prediction method for two-dimensional typhoon wave significant height according to claim 1, characterized in that, The training process of the prediction model includes: Obtain the actual wave height data in the target sea area corresponding to the training dataset and the validation dataset; The predicted wind field data and historical wave field data in the training dataset are input into the prediction model in batches to obtain the training predicted wave height; The prediction model error is obtained by comparing the predicted wave height from the training with the actual wave height data. The parameters of the prediction model are updated based on the prediction model error to obtain the trained prediction model. The predicted wind field data and historical wave field data in the validation dataset are input into the trained prediction model in batches to obtain the validation predicted wave height. When the similarity between the verified predicted wave height and the actual wave height data is greater than a preset threshold, the trained prediction model is output.

8. A two-dimensional intelligent prediction device for the significant wave height of typhoon waves, characterized in that, include: The acquisition module is used to acquire forecast wind field data and historical wave field data for the target sea area; The processing module is used to preprocess the forecast wind field data and the historical wave field data respectively to obtain normalized wind field data and normalized wave field data. The normalized wind field data and normalized wave field data are input into the trained prediction model to obtain the predicted two-dimensional typhoon wave significant height; wherein, the prediction model is trained using a training dataset and then validated using a validation dataset, the training dataset and the validation dataset include independent grade-level forecast wind field data and historical wave field data.

9. A computing device, characterized in that, include: A processor, a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The system stores instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1 to 7.

Images