County-level sea area sea wave correction forecasting method, equipment, medium and product

Abstract

The invention discloses a county-level sea area sea wave correction forecasting method and device, a medium and a product, and relates to the field of sea forecasting, and the method comprises the steps: obtaining n feature vectors, and transmitting the n feature vectors to a pre-trained TCN model, enabling the pre-trained TCN model to output n county-level sea area sea wave forecasting error prediction values after the report starting time a, the n eigenvectors are eigenvectors corresponding to N forecast time periods of the target meteorological elements within a predetermined range with the target county-level sea area as the center after the report starting time a; for each forecast moment after the report starting time a, the difference between the forecast value of the sea waves in the county-level sea area and the forecast error of the sea waves in the county-level sea area is determined, and a preliminary correction forecast value of the sea waves in the county-level sea area is obtained; and sending the preliminary correction forecast value to a pre-trained KAN model, so that the KAN model outputs final correction forecast values of the sea waves in the n county-level sea areas after the report starting time a. The accuracy of sea wave forecasting in county-level sea areas is improved.

Description

Technical Field

[0001] The present application relates to the field of ocean forecasting technology, and in particular to a method, equipment, medium and product for correcting and forecasting ocean waves in county-level sea areas. Background Art

[0002] Currently, county-level ocean wave forecasts are based on large-scale numerical wave forecast products, combined with regional statistical methods for comprehensive forecasting. However, due to uncertainties in physical parameters, errors in initial and boundary conditions, numerical wave forecasts often suffer from multiple errors, which in turn affect the accuracy of county-level ocean wave forecasts.

[0003] To improve forecast accuracy, forecast error correction is a widely used and effective method. Traditional error correction methods include model output statistics (MOS), multi-model combination, and empirical formula correction. However, with the development of artificial intelligence (AI), intelligent correction methods based on multi-source data fusion are becoming an important means to improve the accuracy of wave forecasts. AI-based correction technologies offer significant advantages, including the ability to exploit nonlinear correlations in wave data, breaking through linear limitations; and the ability to process gridded data, comprehensively analyze spatial distributions, and capture regional synergistic effects. However, wave evolution is influenced by nonlinear dynamic mechanisms and multiscale ocean-atmosphere interactions, resulting in significant spatiotemporal complexity. Existing intelligent wave forecast correction methods based on neural networks such as BPNN (Backpropagation Neural Network), CNN (Convolutional Neural Network), and LSTM (Long Short-Term Memory), as well as their combination models, face the following technical bottlenecks in implementation: Regarding feature input, single time series data cannot effectively represent the long-term characteristics of wave changes, leading to inaccurate forecasts. Summary of the Invention

[0004] The purpose of this application is to provide a county-level sea area wave correction forecast method, equipment, medium and product to solve the problem of inaccurate wave forecast described in the background technology.

[0005] To achieve the above objectives, this application provides the following solutions: In the first aspect, this application provides a county-level sea wave correction forecast method, including: Obtain n eigenvectors and send them to a pre-trained model based on a Temporal Convolutional Network (TCN) so that the pre-trained TCN model outputs predicted values of the n county-level sea area wave forecasting errors after the starting forecast time a. The n eigenvectors are eigenvectors corresponding to N forecast periods of target meteorological elements within a predetermined range centered on the target county-level sea area after the starting forecast time a; For each forecast moment after the starting forecast time a, determine the difference between the forecast value of the county-level sea area wave and the forecast error of the county-level sea area wave to obtain a preliminary revised forecast value of the county-level sea area wave; Send the preliminary revised forecast value to a pre-trained Kolmogorov-Arnold Network (KAN) model so that the KAN model outputs the final revised forecast values of the n county-level sea area waves after the starting forecast time a.

[0006] Optionally, the target meteorological elements include any one of sea level pressure, zonal wind speed at a first predetermined altitude, and meridional wind speed at a second predetermined altitude.

[0007] Optionally, the predetermined range includes a predetermined number of longitude and latitude grid regions centered on the target county-level sea area.

[0008] Optionally, the n eigenvectors are generated by the following method: For the target meteorological elements, obtain meteorological forecast data for N forecast periods after the starting forecast time a; For the meteorological forecast data for the N forecast periods, slide according to a window size W and a sliding step length L. Each time of sliding generates an eigenvector, and the n eigenvectors are obtained, where n = .

[0009] Optionally, before obtaining the meteorological forecast data for the N forecast periods of the target meteorological elements, the method further includes the following method: Send the meteorological forecast data for the N forecast periods corresponding to a first predetermined number of longitude and latitude grid regions to a Principal Component Analysis (PCA) model so that the PCA model performs dimensionality reduction processing on the meteorological forecast data for the N forecast periods to obtain the meteorological forecast data for the N forecast periods of a second predetermined number of longitude and latitude grid regions after dimensionality reduction, where the second predetermined number is less than the first predetermined number.

[0010] Optionally, the wave forecast error in the county-level sea area is the wave height forecast error of the waves in the county-level sea area.

[0011] Optionally, the first predetermined height is 10 meters, and the second predetermined height is 10 meters.

[0012] In a second aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the steps of the method according to any one of the above first aspects.

[0013] In a third aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method according to any one of the above first aspects are implemented.

[0014] In a fourth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of the method according to any one of the above first aspects are implemented.

[0015] According to the specific embodiments provided by the present application, the following technical effects are disclosed in the present application: The county-level sea area wave correction forecast method provided by the embodiments of the present application predicts the wave forecast error in the county-level sea area through a TCN model; then preliminarily corrects the forecast based on the forecast error to obtain a preliminary corrected forecast value; and then sends the preliminary corrected forecast value to a pre-trained KAN model, so that the KAN model outputs the final corrected forecast values of the waves in n county-level sea areas after the starting time a. That is, the present application constructs a TCN-KAN hybrid model to improve the accuracy of the wave forecast in the county-level sea area. Description of the Drawings

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0017] Figure 1 It is an application environment diagram of a county-level sea area wave correction forecast method in an embodiment of the present application; Figure 2 It is a flowchart of a county-level sea area wave correction forecast method provided by an embodiment of the present application; Figure 3 It is a flowchart of a method for generating n feature vectors provided by an embodiment of the present application; Figure 4A raster schematic diagram of reanalysis data and meteorological numerical prediction products provided by another embodiment of the present application; Figure 5 A schematic diagram of a method for generating n eigenvectors provided by another embodiment of the present application; Figure 6 A schematic diagram of the construction and training of a TCN model provided by an embodiment of the present application; Figure 7 A schematic diagram of hybrid feature generation provided by another embodiment of the present application; Figure 8 A schematic diagram of the construction and training of a KAN model provided by an embodiment of the present application; Figure 9 A comparison schematic diagram between the sea wave prediction result of the present application and the prediction results of other prediction methods provided by an embodiment of the present application; Figure 10 Another comparison schematic diagram between the sea wave prediction result of the present application and the prediction results of other prediction methods provided by an embodiment of the present application; Figure 11 A schematic diagram of the processing flow of an intelligent correction method for sea wave forecasting in county-level sea areas provided by an exemplary embodiment of the present application; Figure 12 A schematic diagram of the structure of a computer device provided by an embodiment of the present application. Detailed implementation manners

[0018] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0019] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below with reference to the accompanying drawings and specific implementation manners.

[0020] The sea wave correction forecasting method provided by the embodiments of the present application can be applied to, for example Figure 1In the application environment shown, the application environment includes a terminal and a server. Among them, the terminal communicates with the server through a network. The data storage system can store the data that the server needs to process. The data storage system can be set up separately, integrated on the server, placed on the cloud or other servers. The terminal can send the data to be processed to the server. After receiving the data to be processed, the server can first store the data to be processed, and when it is necessary to process it, retrieve it from the storage location, or, processing tasks can be executed while storing. In addition, in some embodiments, the method for correcting and forecasting sea waves in county-level sea areas can also be implemented separately by the server or the terminal. For example, the terminal can directly perform the correction and forecasting of sea waves in county-level sea areas for the data to be processed, or the server can obtain the data to be processed from the data storage system and perform the correction and forecasting of sea waves in county-level sea areas for the data to be processed.

[0021] Among them, the terminal can be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, Internet of Things devices, and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. The portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server can be implemented by an independent server or a server cluster composed of multiple servers, and can also be a cloud server.

[0022] In an exemplary embodiment, refer to Figure 2 as shown, a method for correcting and forecasting sea waves in county-level sea areas is provided. This method is executed by a computer device, and specifically can be executed separately by a computer device such as a terminal or a server, or jointly executed by the terminal and the server. In the embodiments of the present application, taking this method applied to Figure 1 the server in as an example for illustration, it includes the following steps 201 to step 203: Before describing the specific steps, it should also be noted in advance that this method uses the numerical forecast data of a certain target meteorological element on a certain day to calculate the final corrected forecast value of the sea waves in the county-level sea area. For other target meteorological elements, the method for correcting and forecasting sea waves in the county-level sea area provided in the present application can also be used. The method is similar to that of the present application, and the present application will not elaborate one by one.

[0023] Step , obtain n feature vectors and send them to a pre-trained TCN model, so that the pre-trained TCN model outputs the predicted values of the n sea wave forecast errors in the county-level sea area after the starting time a. The n feature vectors are the feature vectors corresponding to the N forecast periods of the target meteorological element within a predetermined range centered on the target county-level sea area after the starting time a.

[0024] Step 202: For each forecast time after the starting time a, determine the difference between the forecast value of the sea wave in the county-level sea area and the forecast error of the sea wave in the county-level sea area, so as to obtain the preliminary corrected forecast value of the sea wave in the county-level sea area.

[0025] Step 203: Send the preliminary corrected forecast value to the pre-trained KAN model, so that the KAN model outputs the final corrected forecast values of the sea waves in n county-level sea areas after the starting time a.

[0026] It should be noted here that in this application, the forecast data of the target meteorological elements in the same county-level sea area and the same time period are used to predict the forecast error of the sea waves in the same county-level sea area and the same time period.

[0027] The method for correcting the forecast of sea waves in the county-level sea area provided by the embodiment of this application calculates the forecast error of the sea waves in the county-level sea area through the TCN model; then preliminarily corrects the forecast based on the forecast error to obtain the preliminary corrected forecast value; and then sends the preliminary corrected forecast value to the pre-trained KAN model, so that the KAN model outputs the final corrected forecast values of the sea waves in n county-level sea areas after the starting time a. That is, this application constructs a TCN-KAN hybrid model to improve the accuracy of the forecast of sea waves in the county-level sea area.

[0028] Optionally, in another exemplary embodiment of this application, the target meteorological elements include any one of sea level air pressure, zonal wind speed at the first predetermined height, and meridional wind speed at the second predetermined height.

[0029] Wherein, the first predetermined height and the second predetermined height are the predetermined heights from the sea level, and this height can be an empirical value. For example, both are the heights of 10 meters.

[0030] Of course, in actual use, other target meteorological elements may also be included, and this application does not limit this.

[0031] Optionally, in another exemplary embodiment of this application, the predetermined range includes a predetermined number of longitude and latitude grid regions centered on the target county-level sea area.

[0032] This application uses this grid region as the meteorological characteristic background field information.

[0033] Optionally, referring to Figure 3 , in another exemplary embodiment of this application, the n feature vectors are obtained through the following steps 401 and 402: Step 401: For the target meteorological element, obtain the meteorological forecast data for N forecast periods after the starting time a.

[0034] Step 402 : Slide the N weather forecast data of the forecast time according to the window size W and the sliding step size L, and generate a feature vector each time the sliding occurs, thereby obtaining the n feature vectors.

[0035] Wherein, n= .

[0036] Each time the window slides, the weather forecast data corresponding to the window becomes a feature vector, that is, the weather forecast data included in the feature vector is the same as the size of the window.

[0037] Optionally, in another exemplary embodiment of the present application, before step 401, the method further includes the following step 501: Step 501: Send the weather forecast data for the N forecast periods corresponding to a first predetermined number of latitude and longitude grid areas to a PCA model, so that the PCA model performs dimensionality reduction processing on the weather forecast data for the N forecast periods to obtain the weather forecast data for the N forecast periods for a second predetermined number of latitude and longitude grid areas after dimensionality reduction, wherein the second predetermined number is less than the first predetermined number.

[0038] Among them, dimensionality reduction is to reduce the number of regional grids.

[0039] Furthermore, the process of dimensionality reduction using PCA is as follows: Step 1: Flatten the two-dimensional grid data into a one-dimensional array. Q , flatten all the numerical forecast two-dimensional grid data of the starting time and forecast time into a one-dimensional array one by one, and stack them into the original matrix .

[0040] Step 2: The matrix is diagonalized. The covariance matrix of ,and is a matrix composed of rows, let ,but for right The data after basis transformation. The covariance matrix of ,but and The relationship is expressed as the following formula (1): (1) in, is the number of samples. Therefore, the optimization goal is converted to finding a matrix ,satisfy It is a diagonal matrix, and the diagonal elements are arranged in descending order.

[0041] Step 3: Determine the feature vector after dimensionality reduction. Calculate the contribution ratio of the principal component variance of each meteorological element . When the sum of the contribution ratios of the principal component variances of the first principal components of all meteorological elements is greater than 85%, then determine the first diagonal elements as the feature vector of the meteorological element after dimensionality reduction of the grid data.

[0042] Exemplarily, for each element of mean sea level pressure (MSLP), 10-meter zonal wind speed (U10), and 10-meter meridional wind speed (V10), the regional background data extracted for each forecast time in each forecast file is flattened into a one-dimensional matrix with a length of 225, and then stacked by forecast into a two-dimensional matrix of (m * 119, 225) (where m is the total number of numerical forecast files collected). Through the PCA method, first extract 4 eigenvalues from 225 dimensions, and calculate the contribution ratio of the principal component variance of each eigenvalue, as shown in Table 1 below.

[0043] Table 1 Contribution ratio of principal component variance of each feature after PCA dimensionality reduction

[0044] According to the requirement that the total contribution ratio of the principal components of the features is above 85%, select the first 4 features to form the feature vector after dimensionality reduction. Thus, the original sample of (m * 119, 225) dimensions is reduced to (m * 119, 4), greatly reducing the amount of data while retaining most of the features.

[0045] This application collects county-level sea wave forecast data, county-level sea area neighborhood reanalysis data, and meteorological numerical forecast products. Specifically, the county-level sea wave forecast data can be the wave height forecast data of the county-level sea area, the county-level sea area neighborhood reanalysis data is the reanalysis data of the wave height in the county-level sea area neighborhood, and the meteorological numerical forecast products include the numerical forecast data of 3 elements: mean sea level pressure, 10-meter zonal wind speed, and 10-meter meridional wind speed.

[0046] Exemplarily, collect the county-level wave forecast data in the sea area of N65 (near the east coast of Rongcheng) (latitude range 36.5° - 37.5°N, longitude range 122.5° - 123°E) from January 1, 2020 to July 31, 2024. The forecast validity period is 119 hours, and the forecast time interval is 1 hour. Download the reanalysis data in the same time period from the European Centre for Medium-Range Weather Forecasts (ECMWF). The spatial resolution of the reanalysis data is 0.5°, and the time resolution is 1 hour. In addition, download the meteorological numerical forecast products of the Copernicus Atmosphere Monitoring Service (CAMS) at 00:00 (Universal Time) in the same time period, including three elements: sea-level pressure, 10-meter zonal wind speed, and 10-meter meridional wind speed. The spatial resolution of the meteorological numerical forecast products is 0.4°, the forecast validity period is 119 hours, and the forecast time interval is 1 hour.

[0047] After collecting the data, it is necessary to preprocess the data to facilitate the subsequent processing process. The preprocessing process is as follows: For the reanalysis data, use the zonal statistical analysis method to extract the reanalysis grid data with pixel centers located inside the county vector surface, and establish the spatial mapping relationship between the county-level region and the reanalysis data. See Figure 4 As shown, it is a schematic diagram of a county-level data (vector surface), from Figure 4 it can be seen that the arithmetic mean method can be used, that is, the formula S→E(e) = (e1 + e2 + e3) / 3 is used to calculate and obtain the reanalysis time series data of the sea waves in the county-level sea area. Among them, S is the reanalysis time series data, which can also be denoted as E(e); e1~e9 are the reanalysis grid data.

[0048] For the meteorological numerical forecast products, delimit a certain size of neighborhood range, with the geometric center of the county-level sea area polygon as the center point, and extract the grid data of each element and each forecast validity period in the specified neighborhood range in the meteorological numerical forecast products as the meteorological characteristic background field information.

[0049] Exemplarily, use the zonal statistical analysis method to establish the spatial mapping relationship between the county-level sea area and the reanalysis data, and extract the grids of the reanalysis data corresponding to each county-level sea area. For the meteorological numerical forecast products, with the grid where the N65 sea area is located as the center, take 7 grids in each of the four directions, and delimit it as the regional range of the meteorological characteristic background field of the county-level sea area, that is, a 15-row and 15-column square area centered on the N65 sea area.

[0050] Optionally, see Figure 11 In the left half of, in another exemplary embodiment of the present application, the process of constructing, training, and correcting the TCN model is as follows: Step 1: Organize the meteorological numerical forecast feature dataset. For the meteorological element , the starting time at the The eigenvector after dimensionality reduction of the numerical prediction data at each moment is denoted as (As can be seen from the process of "dimensionality reduction by PCA", is a one-dimensional array with a length of ). Let the sliding time window length be W, the sliding step be L, the number of meteorological elements be C, and the maximum prediction time limit of the numerical prediction product be N. The schematic diagram of sample organization can be seen in Figure 5 as shown. When the time step taken at the end of the sliding time window is reached, the eigenvectors of the previous day's numerical prediction product advanced by the corresponding time steps are used for supplementation. Therefore, for each meteorological numerical prediction product, eigenvectors with a dimension of ( ) are organized and generated.

[0051] After the characteristic data is organized, is the nth characteristic record organized and generated in the prediction product file of meteorological element at the starting time a. The wave height prediction error (the predicted value of the county-level sea area wave minus the reanalysis data value) at the end prediction time of each characteristic record is extracted as the label data, denoted as , as in Figure 5 the corresponding characteristic record time in is . The label matrix extracted from each prediction product file is denoted as , and its dimension is ( 1).

[0052] Step 2: Data normalization and dataset division. The extreme value normalization method is used to perform per-characteristic normalization processing on the organized and generated characteristic data and label data, expressed as the following formulas (2) and (3): (2) (3) In the formula, is the i th characteristic in the above characteristic data, ; is the maximum value in the sample space, is the minimum value in the sample space; is the normalized characteristic data; is the above label data, is the maximum value in the label space, is the minimum value in the label space; It is the standardized label data. Then, a data subset of the last consecutive period of time is extracted from the normalized dataset as the test set, and the previous data subset is randomly divided into a training set and a validation set according to a certain ratio (such as 8:2) to form the final sample dataset.

[0053] Step 3: Construct the TCN model. The network structure and parameters of the TCN model are initially determined, and the above training set data is input into the TCN model for iterative training, and the training results are verified using the validation set data. The network structure of the TCN model is set as Figure 6 shown as follows: Figure 6 Among them, the input data includes feature data and label data, and the dimensions are ( None, W, C×K ) and ( None, 1 ); the TCN receives the input data, designs 3 hidden layers, and the number of channels is H1, H2, and H3 respectively. Residual connections are used between the hidden layers; the convolution method of the hidden layer uses dilated convolution, that is, the more layers towards the upper layer, the more "holes" in the convolution window, and the larger the dilation convolution coefficient d. The output data dimension of the hidden layer H3 is ( None, W, H3 ), which is flattened into ( None, W×H3 ) and input into the fully connected (FCN) layer; after the FCN layer operation, the model result is output through the activation function, that is, the corrected wave height prediction error .

[0054] Step 4: Train the TCN model. The model parameters (such as the number of nodes in hidden layers H1 to H3, the dilation convolution coefficient, the convolution kernel size, etc.) are tuned using the Grid Search method. By enumerating multiple parameter combinations, the combination with the smallest corrected prediction error on the validation set is selected as the final parameter. The activation function of the model is selected as ReLU, and the loss function is selected as the mean square error MSE. Determine the number of model training epochs. After the model training is completed, save the model structure, optimal parameters, and weight data to a file.

[0055] Step 5: Initial correction of sea waves in county-level sea areas. Run the optimal model obtained from the above training, input the training set, validation set, and test set, calculate the prediction error value for each feature record in the sample dataset, and subtract the prediction error value from the corresponding prediction value to obtain the preliminary correction value, denoted as . Among them, a is the starting time of the forecast product, . For the forecast products with the same starting time, a total of correction values are calculated, and the time interval between adjacent correction values is .

[0056] Exemplarily: For each meteorological numerical weather prediction file, by means of a sliding window method, set the sliding step size to 1 and the sliding window length to 24 (for times less than 24 hours, supplement with the prediction data at the corresponding time in the previous day's numerical weather prediction file), and slide to extract the feature vectors after dimensionality reduction of each element every 24 hours. In this way, a matrix with dimensions (119, 24, 12) is obtained for each numerical weather prediction file. After extracting from each numerical weather prediction file one by one, stack them according to the first dimension to generate feature data with dimensions (119*m, 24, 12). At the same time, extract the significant wave height at the end time of each sliding window and organize it as label data. Then, perform independent extreme value normalization on the third dimension of the feature data and the label data in order to generate a normalized dataset. Use the data from October 1, 2023 to July 31, 2024 as the test set, and randomly divide the data between January 1, 2020 and September 30, 2023 into a training set and a validation set according to a ratio of 8:2. Construct a TCN model with an input dimension of (None, 24, 12), and determine that the output channel numbers H1, H2, and H3 of the three hidden layers of the TCN are 128, 32, and 8 respectively through the grid search algorithm. Then, through a fully connected layer, output preliminary error prediction data with dimensions (None, 1). During the model training process, the maximum number of iterations is 1000, the minimum batch size is 400, the learning rate is 0.0001, the optimizer selects the Adam algorithm, the activation function uses ReLU, the loss function uses MSE, and no early stopping mechanism is set. Input the processed training set into the model for training, and use the validation set to verify the training results until the training ends, and save the optimal model structure, parameters, and weights. After training is completed, read and run the model file, perform error prediction on all feature records in the training set, validation set, and test set, and save the preliminary error prediction results in the order of the numerical weather prediction starting time and prediction time limit, with dimensions (m, 119).

[0057] Optionally, referring to Figure 11 the right half of, in another exemplary embodiment of the present application, the process of constructing, training, and correcting the KAN model is as follows: Step 1: Organize the preliminary corrected county-level sea area wave time series dataset. For each meteorological numerical weather prediction file, organize the correction values calculated in the "process of constructing, training, and correcting the TCN model", p where Figure 7 is the total number of meteorological numerical weather prediction products, and the organization method is as shown: Thus, the dimension of the feature data is (None, 1, XL ), denoted as . The label data is the final multi-step correction value, denoted as For moments between 1 and N ( is the time interval of the revised forecast), such as in the 3-hourly correction . Thus, the dimension of the labeled data is (None, ).

[0058] Step 2: Data normalization and dataset partitioning, using the same method as in the "Process of TCN Model Construction, Training and Revision" to generate the training set, validation set and test set.

[0059] Step 3: KAN model construction. Preliminary determine the network structure and parameters of the KAN model, input the above training set data into the KAN model for iterative training, and use the validation set data to verify the training results. The network structure settings of the KAN model are as Figure 8 shown: Figure 8 In the model, the input layer is the ( )-dimensional feature data XL, the output layer is the ( )-dimensional labeled data YL, KAN adopts a 3-layer network structure design, and the dimensions of each layer are H1, H2, and H3 respectively. The KAN model directly outputs the revised data, and the dimension is ( ).

[0060] Step 4: Model training, parameter tuning and training method are the same as in the "Process of TCN Model Construction, Training and Revision" to generate the model structure, parameters and weight files.

[0061] Step 5: Output of the revised result. Run the optimal model obtained from the above training, input the test set, calculate the revised value for each feature record in the test set, and denote it as . In, a is the starting time of the forecast product, n is the forecast time limit of the model of the present application, .

[0062] Exemplarily, taking the starting time of each numerical weather prediction file as the starting time of the sample data of this model, for the 119 error prediction values within the corresponding time range in the above "preliminary error prediction results", subtract the error prediction value from the county-level sea area wave prediction value to obtain the preliminary correction value, and organize and generate feature data with a dimension of (None, 119). Select the 3-hourly significant wave height reanalysis data from the 3rd to 117th hour after the starting time of this model as the label data. Perform normalization processing on the generated feature data and label data using the extreme value normalization method, and then divide the sample dataset into a training set, a validation set, and a test set in the same manner as the above "preliminary error prediction results". Construct a KAN model with an input dimension of (None, 119). Determine that the dimensions of the three hidden layers H1, H2, and H3 are 128, 32, and 8 respectively through the grid search algorithm. Connect different layers using the third-order B-spline activation function, define the grid size of the B-spline interval as 5, and the output dimension is (None, 39). During the model training process, the maximum number of iterations is 1000, the minimum batch size is 100, the learning rate is 0.0001, the optimizer selects the Adam algorithm, the loss function uses MSE, and no early stopping mechanism is set. Input the training set data into the model for training, and use the validation set to verify the training results until the training is completed, and save the optimal model structure, parameters, and weights. After training is completed, read and run the model file, input the feature data in the test set into the model, output the results and perform denormalization to obtain the final correction value.

[0063] Optionally, for the correction results of the above two models, the following method can be used for evaluation: Use the Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) indicators to evaluate the results of the model of this application. The evaluation distinguishes different prediction time ranges, and the calculation formulas are as follows in formulas (4) and (5): (4) (5) In the formula, is the total number of samples in the model test set, is the sample i 's significant wave height reanalysis data value, is the sample i 's corresponding significant wave height correction value.

[0064] The RMSE and MAE metrics are used to test and evaluate the model of this application. Within the 119-hour forecast lead time, the 6-hourly RMSE and MAE metric values between the corrected and uncorrected model of this application and the reanalysis data are calculated, and the statistical results are shown in the following table. As can be seen from the table, compared with the prediction results before correction, the prediction results of the model of this application after correction have higher accuracy at each lead time. The average RMSE error after correction decreased by 29.61% compared with that before correction. ； The average MAE error after correction decreased by 34.66% compared with that before correction. 。 The results show that the accuracy of the county-level sea area wave forecast results has been greatly improved after being corrected by the intelligent correction method for county-level sea area waves of this application.

[0065] Table 2 Comparison of prediction errors between the model of this application and existing inventions

[0066] Select two cold air processes on November 4, 2023 and December 13, 2023 as sample periods. The difference results in the accuracy of the county-level sea area wave forecasts before and after correction by the model of this application are as Figure 9 、 Figure 10 shown in the prediction curve graph. For these two cold air processes, large wave processes (significant wave height > 3m) occurred in the N65 sea area. As can be seen from the graph, after being corrected by the model of this application, the prediction results of the two cold air processes are closer to the true values, and are more accurate in extreme value prediction. After statistics, during the cold air process on November 4, 2023, the RMSE of the prediction results before and after correction by the model of this application were 0.28m and 0.37m respectively; during the cold air process on December 13, 2023, the RMSE of the prediction results before and after correction by the model of this application were 0.19m and 0.40m respectively. Before correction, the predicted values of the significant wave height in the county-level sea area wave numerical forecast had systematic overestimation within the 119-hour forecast lead time, especially the maximum significant wave height in the wave height extreme value area was significantly higher than the true value; after being corrected by the method of this application, the degree of coincidence between the predicted significant wave height sequence and the process curve of the true value was significantly improved. After being corrected by the model of this application, the overall prediction error (RMSE) of the prediction results in the two cold air processes decreased by 24.3% and 52.5% respectively compared with that before correction.

[0067] In summary, this application proposes an intelligent correction and forecasting method for sea waves in county-level sea areas. After being corrected by the model of this application, the accuracy of the sea wave forecasting results in county-level sea areas has been effectively improved. This application adopts a "two-step" modeling idea. It uses the PCA method to reduce the dimensionality of the meteorological background field characteristic data, constructs a sample data set using the same data organization method, uses the TCN model for preliminary correction, and uses the KAN model for final correction, thus constructing a PCA-TCN-KAN hybrid model. Examples prove that the model of this application can effectively correct the sea wave forecasting for a long time period, improve the forecasting accuracy, and the RMSE can be reduced by more than 20% compared with the original sea wave forecasting in county-level sea areas within 119 hours of the time period. Moreover, it can show better fitting ability during the disaster process.

[0068] In addition, the programming language used in this application is Python 3.9.13, the machine learning framework is Pytorch1.12.1, and the operating system is Windows Server 2019 Standard. The server hardware conditions used in this application are as follows: Intel(R) Xeon(R) Gold 6226 CPU @ 2.70GHz, 256GB of memory.

[0069] As mentioned in the background technology of this application, there are varying degrees of data redundancy in multi-source spatio-temporal numerical forecasting data, and there is a problem of temporal sequence mismatch in data organization; second, the gradient stability is insufficient, making it difficult to effectively capture long-distance temporal sequence dependencies, which in turn affects the convergence of model training, and it lacks a powerful function approximation ability, which may cause model underfitting or similar mean regression phenomena. The above problems jointly restrict the further improvement of the accuracy of the artificial intelligence correction results of sea waves in county-level sea areas, making its development face a bottleneck.

[0070] From the content of the specific embodiments of this application, it can be seen that this application proposes an intelligent correction and forecasting model for sea waves in county-level sea areas using TCN and KAN in response to the problems mentioned in the background technology. This model takes multi-source meteorological and hydrological spatio-temporal sequences as inputs. First, it uses PCA for data dimensionality reduction to eliminate redundant information and organizes spatio-temporal sequence feature samples based on meteorological numerical forecasting products; then it uses the dilated convolutional structure of TCN to capture the long-distance temporal sequence dependence characteristics of sea wave evolution; finally, it uses the general function approximation characteristics of the KAN model to achieve high-precision mapping of the non-linear dynamic process, thus constructing an intelligent correction and forecasting system for sea waves in county-level sea areas that integrates multi-source data characteristics. While maintaining the computational efficiency, this method successfully breaks through the bottleneck of the existing intelligent correction methods for sea wave forecasting in county-level sea areas with a more superior model architecture and greatly improves its accuracy.

[0071] The present application provides an intelligent correction and forecasting method for sea waves in county-level sea areas, aiming to solve the following two problems and improve the accuracy of sea wave forecasting in county-level sea areas. First, the characteristics of the meteorological background field are introduced and dimensionality reduction is performed to solve the problems of time-consuming model training and difficult convergence. Meteorological numerical forecast products are regional grid data, which contain spatio-temporal background information as well as a large amount of redundant information. Directly inputting them into the intelligent model will lead to problems such as excessive data volume, time-consuming training, and difficult convergence. Therefore, the present application performs dimensionality reduction on the original meteorological numerical forecast data through PCA. While retaining the main characteristics of the regional background field, it reduces the amount of model input data, lowers the requirements for computer hardware conditions for model training, and improves the model training speed and convergence. Second, a two-step correction method combining TCN and KAN is adopted to solve problems such as the weak long-sequence processing ability and function representation ability of existing models. TCN flexibly controls the receptive field of the model by stacking convolutional layers, adjusting the convolutional kernel size, and using dilated convolutions. Through an exponentially increasing dilation factor, TCN obtains a large receptive field with fewer layers, thus efficiently capturing long-term dependencies. At the same time, KAN has a learnable activation function (such as a spline function) at the edge of the neural network and has a stronger function representation ability than a network with a fixed activation function. Therefore, in modeling tasks related to physical processes, KAN can achieve higher accuracy with fewer parameters. By combining TCN and KAN for modeling, using TCN as a powerful spatio-temporal sequence feature extractor, and then inputting the features extracted by TCN and the time series observation data into KAN for final mapping, a more accurate function fitting is obtained, and further, the problem of gradient disappearance / explosion during the training process caused by the long-time dependence relationship and multi-scale features of the data itself is better solved, and the long-distance temporal sequence dependence features of sea wave evolution can be better captured, etc.

[0072] The present application adopts a "two-step" modeling strategy: First, the original meteorological numerical forecast data is processed by the PCA model for dimensionality reduction, and the dimensionally reduced data is organized into meteorological spatio-temporal feature samples, and the TCN model is used for preliminary correction; Second, the preliminary correction result of the TCN model is used as the input data, and the KAN model is used for the final correction of the multi-step sea wave forecasting in the county-level sea area.

[0073] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be seen Figure 12As shown in the figure. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data related to the correction and prediction of sea waves in county-level sea areas. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, a method for correcting and predicting sea waves in county-level sea areas can be implemented.

[0074] Those skilled in the art can understand that Figure 12 the structure shown in the figure is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0075] In an exemplary embodiment, a computer device is further provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0076] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0077] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0078] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of the relevant data need to comply with the relevant regulations.

[0079] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, in the various embodiments provided in the present application, any reference to a memory, a database, or other media can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memories (ReRAM), magnetoresistive random access memories (MRAM), ferroelectric random access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0080] In the various embodiments provided in the present application, the databases involved can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. In the various embodiments provided in the present application, the processors involved can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logics, data processing logics based on quantum computing, etc., without limitation.

[0081] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the various technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0082] Specific examples are used in this article to elaborate on the principles and implementation manners of the present application. The descriptions of the above embodiments are only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A method for correcting and forecasting sea waves in a county-level sea area, characterized in that, Including: Obtain n feature vectors and send them to a pre-trained TCN model, so that the pre-trained TCN model outputs predicted values of the wave forecasting errors of n county-level sea areas after the starting time a. The n feature vectors are feature vectors corresponding to N forecasting time periods of target meteorological elements within a predetermined range centered on the target county-level sea area after the starting time a; For each forecasting moment after the starting time a, determine the difference between the forecast value of the sea waves in the county-level sea area and the forecasting error of the sea waves in the county-level sea area, and obtain a preliminary corrected forecast value of the sea waves in the county-level sea area; Send the preliminary corrected forecast value to a pre-trained KAN model, so that the KAN model outputs the final corrected forecast values of the sea waves in n county-level sea areas after the starting time a.

2. The method for correcting and predicting sea waves in county-level sea areas according to claim 1, wherein The target meteorological elements include any one of sea level pressure, zonal wind speed at a first predetermined height, and meridional wind speed at a second predetermined height.

3. The method for correcting and forecasting sea waves in a county-level sea area according to claim 1, wherein The predetermined range includes a predetermined number of longitude and latitude grid areas centered on the target county-level sea area.

4. The method for correcting and forecasting sea waves in county-level sea areas according to claim 1, wherein The n feature vectors are generated by the following method: For the target meteorological elements, obtain meteorological forecast data for N forecasting time periods after the starting time a; For the meteorological forecast data of the N forecast lead times, slide according to the window size W and the sliding step length L, generate a feature vector each time of sliding, and obtain the n feature vectors, where n = .

5. The county-level sea area wave correction forecasting method according to claim 4, wherein Before obtaining the meteorological forecast data for N forecasting time periods after the starting time a for the target meteorological elements, the method further includes the following method: Send the meteorological forecast data for N forecasting time periods corresponding to a first predetermined number of longitude and latitude grid areas to a PCA model, so that the PCA model performs dimensionality reduction processing on the meteorological forecast data for N forecasting time periods to obtain the meteorological forecast data for N forecasting time periods of a second predetermined number of longitude and latitude grid areas after dimensionality reduction, where the second predetermined number is less than the first predetermined number.

6. The method for correcting and forecasting sea waves in a county-level sea area according to claim 5, wherein, The wave forecasting error of the county-level sea area is the wave height forecasting error of the sea waves in the county-level sea area.

7. The method for correcting and forecasting sea waves in county-level sea areas according to claim 2, wherein The first predetermined height is 10 meters, and the second predetermined height is 10 meters.

8. A computer device, comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the county-level sea area wave correction forecasting method according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the county-level sea area wave correction forecasting method according to any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the county-level sea area wave correction forecasting method according to any one of claims 1-7.

Images