A high spatial resolution wave element prediction method

Abstract

The present invention discloses a high spatial resolution wave element prediction method, which relates to the technical field of nearshore wave model element prediction, and includes the following steps: S1: determine the target sea area, obtain meteorological and wave reanalysis data, and prepare initial field files and boundary files; S2: construct a computing grid of the target sea area, and use the SWAN model to simulate and calculate high spatial resolution wave elements according to meteorological and wave reanalysis data, as well as the initial field files and boundary files; S3: construct a database; S4: construct a wave element prediction model, and iteratively train the wave element prediction model until convergence; S5: use the iteratively trained wave element prediction model to predict high spatial resolution wave elements. The method of the present invention can realize the calculation of wave elements from low spatial resolution reanalysis data to high spatial resolution, and reduces the demand for computing resources, improves computing efficiency, and accelerates response speed.

Description

Technical Field

[0001] The invention relates to the field of nearshore wave mode element prediction technology, and in particular to a high spatial resolution wave element prediction method. Background Art

[0002] Nearshore wave elements are of great significance to coastal infrastructure, ship mooring stability, and marine engineering disaster prevention and control. In practical engineering applications, wave parameters with high temporal and spatial resolution are needed to achieve accurate wave prediction, thereby providing a reliable reference for engineering design, offshore operations, etc.

[0003] At present, the acquisition of high spatial resolution wave elements at complex nearshore locations usually adopts the traditional numerical simulation method, which provides boundaries and initial files through low spatial resolution reanalysis data, and establishes nested grids in complex areas for encrypted calculations. This method has high demand for computing resources, low computing efficiency and slow response speed. Especially when real-time response and rapid prediction and forecast of sea condition warnings are required, this type of numerical simulation method is often difficult to respond quickly and difficult to apply in practice.

[0004] In view of this, this invention is proposed. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a high spatial resolution wave element prediction method, which uses publicly available low spatial resolution reanalysis data to quickly calculate high spatial resolution wave elements in complex offshore environments. Compared with traditional numerical simulation methods, it reduces the demand for computing resources, improves computing efficiency, and speeds up response speed.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] A high spatial resolution wave element prediction method comprises the following steps:

[0008] S1: Determine the target sea area, obtain the public meteorological reanalysis data and wave reanalysis data according to the scope of the target sea area, and prepare the initial field file and boundary file for the simulation calculation of the shallow water wave numerical model in the target sea area;

[0009] S2: Construct the computational grid of the target sea area, and use the shallow water wave numerical model to simulate and calculate the wave elements with high spatial resolution based on the meteorological reanalysis data and wave reanalysis data obtained by S1, as well as the initial field file and boundary file constructed by S1;

[0010] S3: Build a database based on S1 meteorological reanalysis data and wave reanalysis data, as well as high spatial resolution wave elements simulated and calculated by S2;

[0011] S4: A wave element prediction model is constructed based on the Conv-LSTM model and the AutoEncoder model. The wave element prediction model includes a sequentially connected input layer, a compression structure, and a reconstruction structure. The wave element prediction model is iteratively trained using the database constructed by S3, so that the wave element prediction model learns the nonlinear mapping relationship between meteorological reanalysis data and wave reanalysis data and high spatial resolution wave elements until convergence;

[0012] S5: Use the wave element prediction model iteratively trained in S4 to predict wave elements with high spatial resolution.

[0013] Further, the S1 comprises the following steps:

[0014] S11: Determine the target sea area, obtain the land boundary of the target sea area, and the water depth and altitude corresponding to each longitude and latitude point in the ocean area of ​​the target sea area;

[0015] S12: According to the scope of the target sea area, obtain low spatial resolution meteorological reanalysis data and wave reanalysis data of the preset year from the public ERA5;

[0016] S13: Based on the land boundary of the target sea area obtained in S11, the water depth and altitude corresponding to each longitude and latitude point in the ocean area of ​​the target sea area, and the low spatial resolution meteorological reanalysis data and wave reanalysis data of the preset year obtained in S12, an initial field file and a boundary file for simulating the wave elements of the target sea area data are prepared.

[0017] Further, the S2 comprises the following steps:

[0018] S21: construct a computational grid for the target sea area, where the grid is refined near the coast;

[0019] S22: Based on the meteorological reanalysis data and wave reanalysis data obtained in S12, and the initial field file and boundary file constructed in S13, the SWAN model is used to simulate and calculate the high spatial resolution wave elements of the preset year;

[0020] S23: Obtain high spatial resolution wave elements of a preset year simulated and calculated by the SWAN model in S22, and remove invalid values.

[0021] Further, the S3 comprises the following steps:

[0022] S31: the low spatial resolution meteorological reanalysis data and wave reanalysis data of the preset year obtained in S12 and the high spatial resolution wave elements of the preset year obtained in S23 are matched in time with each other;

[0023] S32: The low spatial resolution meteorological reanalysis data and wave reanalysis data after time correspondence, as well as the high spatial resolution wave elements, are divided into windows and stacked, and combined with the longitude and latitude points of the target sea area, a database with 12 hours of data as a set of samples is constructed.

[0024] Furthermore, the input layer of the wave element prediction model is a Conv-LSTM model, the compression structure includes 7 layers, which are alternating three-dimensional convolution layers and pooling layers, and the reconstruction structure includes 12 layers, which are alternating upsampling layers and three-dimensional convolution layers.

[0025] Furthermore, the data in the database constructed in S3 is divided by year. In S4, the wave element prediction model is trained year by year using transfer learning. After each training, the training parameters are adjusted according to the test results.

[0026] Furthermore, in transfer learning, the mean square error (MSE) is calculated by the following steps:

[0027] S41: Constructing a three-dimensional mean square error matrix , where m represents the number of moments in each set of data in the database, and n represents the spatial resolution of the high spatial resolution wave elements at a single moment output by the wave element prediction model. Medium Element The calculation formula is as follows:

[0028] ,

[0029] In the formula, The high spatial resolution wave elements output by the wave element prediction model. is the wave element with high spatial resolution in the database. Point time on land is 0, when Point time on the ocean is 1;

[0030] S42: Based on the constructed three-dimensional mean square error matrix Calculate the mean squared error (MSE).

[0031] Furthermore, in transfer learning, the mean square error (MSE) is calculated by the following steps:

[0032] S41: Constructing the initial three-dimensional mean square error matrix , where m represents the number of moments in each set of data in the database, and n represents the spatial resolution of the high spatial resolution wave elements at a single moment output by the wave element prediction model. Medium Element The calculation formula is as follows:

[0033] ,

[0034] In the formula, The high spatial resolution wave elements output by the wave element prediction model. is the wave element with high spatial resolution in the database. Point time on land is 0, when Point time on the ocean is 1;

[0035] S42: Calculate the mean and variance of the high spatial resolution wave elements at each moment of the first point in the current group of data;

[0036] S43: Determine in turn whether the wave element with high spatial resolution at each moment of the first point in the current group of data is greater than the sum of the mean and variance calculated in S42. If so, set , if otherwise ;

[0037] S44: Execute S42 and S43 cyclically, according to the calculated Constructing a three-dimensional mean square error matrix ;

[0038] S45: Based on the constructed three-dimensional mean square error matrix Calculate the mean squared error (MSE).

[0039] Further, the S5 comprises the following steps:

[0040] S51: Obtaining the public meteorological reanalysis data and wave reanalysis data of the target sea area within the time range to be predicted;

[0041] S52: Based on the meteorological reanalysis data and wave reanalysis data obtained in S51, the wave element prediction model iteratively trained in S4 is used to predict the wave elements with high spatial resolution within the time range to be predicted.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] 1. Through the public low-spatial-resolution reanalysis data, the high-spatial-resolution wave elements (such as significant wave height, average wave direction, and average wave period) in complex offshore environments can be quickly calculated and predicted in both hindcasts and forecasts.

[0044] 2. It can serve the operation and maintenance and disaster warning of large offshore ports and docks, help optimize the berthing and dispatching arrangements of ships, improve the operation and maintenance efficiency and safety of ports and docks, enhance the ability to warn and respond to marine disasters, and ensure the safety of coastal infrastructure and personnel; in addition, it can also serve the intelligent wave management system in complex geographical environments such as energy bases.

[0045] 3. A wave element prediction model was established based on the convolutional long short-term memory (Conv-LSTM) and autoencoder (AutoEncoder) neural network; the Conv-LSTM model was used to improve the processing capabilities of spatial relationship data and temporal relationship data, and the AutoEncoder model was used to further improve the processing capabilities of spatial relationship data, and the spatial resolution of wave elements was improved; the wave element prediction model constructed a nonlinear mapping relationship between low spatial resolution meteorological reanalysis data and wave reanalysis data and high spatial resolution wave elements, effectively replacing the complex grid nested calculation process in the traditional numerical model, and realizing the prediction from low spatial resolution wave parameters to high spatial resolution wave elements, that is, the rapid calculation of fine wave characteristics near the shore; and compared with the traditional numerical simulation method, it reduces the demand for computing resources, improves computing efficiency, and speeds up response speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flow chart of the high spatial resolution wave element prediction method of embodiment 1;

[0047] Figure 2 It is a structural diagram of the wave element prediction model of the first embodiment;

[0048] Figure 3 A computational grid diagram of the target sea area for simulation in Example 1;

[0049] Figure 4 Comparison of the field diagrams of the simulation calculation of Example 1 Figure 1 , wherein A is a wave field element field map composed of high spatial resolution significant wave height at a certain moment in 2023 calculated by the wave element prediction model of the embodiment, and B is a wave field element field map composed of high spatial resolution significant wave height at the corresponding moment calculated by the SWAN model;

[0050] Figure 5 Comparison of the field diagrams of the simulation calculation of Example 1 Figure 2 , wherein a is a wave field element map composed of low spatial resolution significant wave height reanalysis data at a certain moment in 2023 input into the wave element prediction model, b is a wave field element map composed of high spatial resolution significant wave height calculated by the wave element prediction model of the embodiment, c is an enlarged view of point A in figure a, and d is an enlarged view of point B in figure b;;

[0051] Figure 6 The effective wave height prediction time series of the simulation calculation of Example 1 Figure 1 ;

[0052] Figure 7 The effective wave height prediction time series of the simulation calculation of Example 1 Figure 2 . DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0054] Embodiment 1:

[0055] A high spatial resolution wave element prediction method comprises the following steps:

[0056] S1: Determine the target sea area, obtain the public meteorological reanalysis data and wave reanalysis data according to the scope of the target sea area, and prepare the initial field file and boundary file for the simulation calculation of the shallow water wave numerical model in the target sea area;

[0057] S2: Construct the computational grid of the target sea area, and use the shallow water wave numerical model to simulate and calculate the wave elements with high spatial resolution based on the meteorological reanalysis data and wave reanalysis data obtained by S1, as well as the initial field file and boundary file constructed by S1;

[0058] S3: Build a database based on S1 meteorological reanalysis data and wave reanalysis data, as well as high spatial resolution wave elements simulated and calculated by S2;

[0059] S4: A wave element prediction model is constructed based on the Conv-LSTM model and the AutoEncoder model. The wave element prediction model includes a sequentially connected input layer, a compression structure, and a reconstruction structure. The wave element prediction model is iteratively trained using the database constructed by S3, so that the wave element prediction model learns the nonlinear mapping relationship between meteorological reanalysis data and wave reanalysis data and high spatial resolution wave elements until convergence;

[0060] S5: Use the wave element prediction model iteratively trained in S4 to predict wave elements with high spatial resolution.

[0061] The high spatial resolution wave element prediction method of this embodiment uses publicly available low spatial resolution reanalysis data to quickly calculate high spatial resolution wave elements (such as significant wave height, average wave direction, and average wave period) in complex offshore environments, which can be used for prediction in both hindcasts and forecasts.

[0062] Based on this, the high spatial resolution wave element prediction method of this embodiment can serve the operation and maintenance and disaster warning of large offshore ports and docks, help optimize the berthing and scheduling arrangements of ships, improve the operation and maintenance efficiency and safety of ports and docks, enhance the ability to warn and respond to marine disasters, and ensure the safety of coastal infrastructure and personnel; in addition, it can also serve the intelligent wave management system in complex geographical environments such as energy bases.

[0063] Specifically, the high spatial resolution wave element prediction method of this embodiment establishes a wave element prediction model based on the convolutional long short-term memory (Conv-LSTM) and autoencoder (AutoEncoder) neural network; the Conv-LSTM model is used to improve the processing capabilities of spatial relationship data and temporal relationship data, and the AutoEncoder model is used to further improve the processing capabilities of spatial relationship data, and the spatial resolution of wave elements is improved; the wave element prediction model constructs a nonlinear mapping relationship between low spatial resolution meteorological reanalysis data and wave reanalysis data and high spatial resolution wave elements, effectively replacing the complex grid nested calculation process in the traditional numerical model, and realizing the prediction from low spatial resolution wave parameters to high spatial resolution wave elements, that is, the rapid calculation of fine wave characteristics near the shore; and compared with the traditional numerical simulation method, it reduces the demand for computing resources, improves computing efficiency, and accelerates response speed.

[0064] In an optional embodiment, the S1 comprises the following steps:

[0065] S11: Determine the target sea area, obtain the land boundary of the target sea area, and the water depth and altitude corresponding to each longitude and latitude point in the ocean area of ​​the target sea area;

[0066] S12: According to the scope of the target sea area, obtain low spatial resolution meteorological reanalysis data and wave reanalysis data of the preset year from the public ERA5;

[0067] S13: Based on the land boundary of the target sea area obtained in S11, the water depth and altitude corresponding to each longitude and latitude point in the ocean area of ​​the target sea area, and the low spatial resolution meteorological reanalysis data and wave reanalysis data of the preset year obtained in S12, an initial field file and a boundary file for simulating the wave elements of the target sea area data are prepared.

[0068] In an optional embodiment, the S2 comprises the following steps:

[0069] S21: construct a computational grid for the target sea area, where the grid is refined near the coast;

[0070] S22: Based on the meteorological reanalysis data and wave reanalysis data obtained in S12, and the initial field file and boundary file constructed in S13, the SWAN model is used to simulate and calculate the high spatial resolution wave elements of the preset year;

[0071] S23: Obtain high spatial resolution wave elements of a preset year simulated and calculated by the SWAN model in S22, and remove invalid values.

[0072] In an optional embodiment, S3 includes the following steps:

[0073] S31: the low spatial resolution meteorological reanalysis data and wave reanalysis data of the preset year obtained in S12 and the high spatial resolution wave elements of the preset year obtained in S23 are matched in time with each other;

[0074] S32: The low spatial resolution meteorological reanalysis data and wave reanalysis data after time correspondence, as well as the high spatial resolution wave elements, are divided into windows and stacked, and combined with the longitude and latitude points of the target sea area, a database with 12 hours of data as a set of samples is constructed.

[0075] In an optional embodiment, the input layer of the wave element prediction model is a Conv-LSTM model, the compression structure includes 7 layers, which are alternating three-dimensional convolution layers and pooling layers, and the reconstruction structure includes 12 layers, which are alternating upsampling layers and three-dimensional convolution layers.

[0076] In an optional embodiment, the data in the database constructed in S3 is divided by year, and in S4, the wave element prediction model is trained year by year using transfer learning, and the training parameters are adjusted according to the test results after each training is completed.

[0077] In an optional embodiment, the mean square error (MSE) is calculated in transfer learning by the following steps:

[0078] S41: Constructing a three-dimensional mean square error matrix , where m represents the number of moments in each set of data in the database, and n represents the spatial resolution of the high spatial resolution wave elements output by the wave element prediction model. Medium Element The calculation formula is as follows:

[0079] ,

[0080] In the formula, The high spatial resolution wave elements output by the wave element prediction model. is the wave element with high spatial resolution in the database. Point time on land is 0, when Point time on the ocean is 1;

[0081] S42: Based on the constructed three-dimensional mean square error matrix Calculate the mean squared error (MSE).

[0082] In an optional embodiment, the mean square error (MSE) is calculated in transfer learning by the following steps:

[0083] S41: Constructing the initial three-dimensional mean square error matrix , where m represents the number of moments in each set of data in the database, and n represents the spatial resolution of the high spatial resolution wave elements output by the wave element prediction model. Medium Element The calculation formula is as follows:

[0084] ,

[0085] In the formula, The high spatial resolution wave elements output by the wave element prediction model. is the wave element with high spatial resolution in the database. Point time on land is 0, when Point time on the ocean is 1;

[0086] S42: Calculate the mean and variance of the high spatial resolution wave elements at each moment of the first point in the current group of data;

[0087] S43: Determine in turn whether the wave element with high spatial resolution at each moment of the first point in the current group of data is greater than the sum of the mean and variance calculated in S42. If so, set , if otherwise ;

[0088] S44: Execute S42 and S43 cyclically, according to the calculated Constructing a three-dimensional mean square error matrix ;

[0089] S45: Based on the constructed three-dimensional mean square error matrix Calculate the mean squared error (MSE).

[0090] In an optional embodiment, the S5 comprises the following steps:

[0091] S51: Obtaining the public meteorological reanalysis data and wave reanalysis data of the target sea area within the time range to be predicted;

[0092] S52: Based on the meteorological reanalysis data and wave reanalysis data obtained in S51, the wave element prediction model iteratively trained in S4 is used to predict the wave elements with high spatial resolution within the time range to be predicted.

[0093] In order to verify the effectiveness of the method of this embodiment, the coastal waters of the southern Shandong Peninsula were selected as an example as the target sea area to verify the method of this embodiment. The core areas of the coastal waters of the southern Shandong Peninsula were Dongjiakou Port and Qianwan, which are key areas for practical application of coastal engineering.

[0094] In S1, the land boundary of the target sea area, as well as the water depth and altitude corresponding to each latitude and longitude point in the ocean area of ​​the target sea area are obtained; thereafter, low spatial resolution meteorological reanalysis data and wave reanalysis data for a total of 30 years from 1994 to 2023 are obtained from the public ERA5 library. The meteorological reanalysis data include: sea surface wind speed and sea surface wind direction, and the wave reanalysis data include: effective wave height, average wave direction, and wave average period; in addition, it should be noted that the spatial range for obtaining meteorological reanalysis data and wave reanalysis data must be larger than the range of the target sea area; thereafter, the initial field file and boundary file for the simulation calculation of the numerical model of shallow water waves in the target sea area are prepared. It should be understood that in other specific implementation processes, other public meteorological reanalysis data and wave reanalysis data can be selected as needed.

[0095] In S2, the computational grid of the target sea area is constructed, such as Figure 3 As shown; thereafter, the SWAN model is selected, and the wave elements with high spatial resolution of the target sea area from 1994 to 2023 are calculated according to the initial field file and boundary file for the simulation calculation of the target sea area wave numerical model produced by S1, including: effective wave height, average wave direction, and average wave period; then invalid values ​​are eliminated; it should be understood that the size of the grid can be set as needed, and the shallow water wave numerical model can be selected.

[0096] In S3, low spatial resolution meteorological reanalysis data and wave reanalysis data at the same time between 1994 and 2023, as well as high spatial resolution wave elements, are collected in units of hours (one moment corresponds to one hour); subsequently, window division and stacking are performed, and a database with 12 hours of data as a group of samples is constructed in combination with the longitude and latitude points of the target sea area. Specifically, the database includes data from 1994 to 2023, each group of data includes 12 hours of data, and each group of data is divided into 12 groups, each group includes meteorological reanalysis data and wave reanalysis data corresponding to each longitude and latitude point at low spatial resolution at the corresponding hour, and wave elements corresponding to each longitude and latitude point at high spatial resolution; that is, each group of data includes meteorological reanalysis data and wave reanalysis data corresponding to each longitude and latitude point at low spatial resolution at 12 moments, and wave elements corresponding to each longitude and latitude point at high spatial resolution. In this embodiment, a database with 12 hours of data as a group of samples is constructed, but the wave element prediction model constructed by S4 has continuous time characteristics in the wave elements calculated and output after learning.

[0097] In S4, the Keras tool library of the TensorFlow toolkit disclosed in the Python language is used to build a wave element prediction model, whose input layer is a Conv-LSTM model for receiving input information; layers 2-8 are alternating three-dimensional convolutional layers and pooling layers for decomposing and compressing the input information layer by layer; layers 9-20 are alternating upsampling layers and three-dimensional convolutional layers for downscaling and reconstructing the input information. The pooling layer and upsampling layer in the wave element prediction model of this embodiment are regularized reduction and amplification of the matrix data in the target sea area. For example, the grid wave data of 24*24 size is 12*12 layers after passing through a 2*2 pooling layer, and similarly, it is restored to 24*24 layers after passing through a 2*2 upsampling layer; it should be understood that in other specific implementation processes, the number of layers of the wave element prediction model can be set according to the actual size of the numerical grid in the target sea area.

[0098] In addition, in S4, transfer learning is used to train the wave element prediction model year by year, and the training parameters are adjusted according to the test results after each training. The transfer learning method is a prior art and will not be elaborated here; specifically, the wave element prediction model is first iteratively trained using the 1994 data in the database, and the training parameters are adjusted according to the test results after training, such as the error magnification factor, and then the wave element prediction model is iteratively trained using the 1995 data in the database, and so on.

[0099] In addition, in S4, the mean square error MSE is calculated in transfer learning by the following steps:

[0100] S41: Constructing the initial three-dimensional mean square error matrix , where 12 means that each set of data described above includes 12 hours of data, and 192*192 means the spatial resolution of the wave elements with high spatial resolution at a single moment output by the wave element prediction model. Specifically, when a set of low spatial resolution reanalysis data is input to the wave element prediction model, the data output by the wave element prediction model includes 12 192*192 two-dimensional matrices, each of which corresponds to a moment, and each element in the two-dimensional matrix corresponds to a wave element at a latitude and longitude point, such as significant wave height; Medium Element The calculation formula is as follows:

[0101] ,

[0102] In the formula, k=1...12, i=1...192, j=1...192, The high spatial resolution wave elements output by the wave element prediction model. is the wave element with high spatial resolution in the database. Point time on land is 0, when Point time on the ocean is 1; through this step, the points corresponding to the land are set to zero, and then the prediction center of the wave element prediction model is shifted to the ocean area.

[0103] S42: Calculate the mean and variance of the high spatial resolution wave elements at each moment of the first point in the current group of data. S43: Determine in turn whether the high spatial resolution wave elements at each moment of the first point in the current group of data are greater than the sum of the mean and variance calculated in S42. If so, set , if otherwise Each set of data described above includes 12 hours of data, that is, the high spatial resolution wave elements in each set of data are a three-dimensional matrix composed of a 192*192 two-dimensional matrix at 12 moments. In S42 and S43, the first longitude and latitude point of 192*192 is selected according to the preset order. First, the mean deviation and variance of the values ​​of the wave elements at the point at 12 moments are calculated. Then, it is calculated whether the wave elements at the point at 12 moments are greater than the sum of the calculated mean and variance. If so, let , if otherwise .

[0104] S44: Execute S42 and S43 cyclically, according to the calculated Constructing a three-dimensional mean square error matrix Through this step, the attention of the wave element prediction model can be adjusted to the larger value of effective waves in the target sea area.

[0105] S45: Based on the constructed three-dimensional mean square error matrix The mean square error MSE is calculated and learning is performed in S4 using the mean square error MSE.

[0106] In this embodiment, the wave element prediction model is trained using data from 1994 to 2022, and then the data from 2023 is used to verify the effectiveness of the trained wave element prediction model.

[0107] like Figure 4 As shown, Figure 4 A is a wave field element map composed of high spatial resolution significant wave height at a certain time in 2023 calculated by the wave element prediction model of this embodiment, Figure 4 B is a wave field element map composed of high spatial resolution significant wave heights at the corresponding moment calculated by the SWAN model; it can be seen that the wave element prediction model of this embodiment can accurately predict the distribution of wave field elements.

[0108] like Figure 5 As shown, Figure 5 a is the wave field element map composed of low spatial resolution significant wave height reanalysis data at a certain time in 2023 input into the wave element prediction model. Figure 5 b A wave field element diagram composed of high spatial resolution significant wave heights calculated by the wave element prediction model of this embodiment, Figure 5 c is Figure 5 The enlarged view of point A in a. Figure 5 d is Figure 5 b is an enlarged view of point B; it can be seen that in the wave field element field map composed of the low spatial resolution effective wave height reanalysis data, the details of the nearshore area are obscured, the area where Qianwan and Dongjiakou Port are located is marked as land, and there is no wave element data, while the wave element prediction model of this embodiment can effectively predict the distribution of wave field elements.

[0109] like Figure 6 As shown, the hourly time series diagram of the effective wave height at Dongjiakou location in 2023, where the blue solid line is the calculation result of the wave element prediction model of this embodiment, and the red solid line is the calculation result of the SWAN model. Figure 6 The highest RMSE value is 0.095(m), and the highest RMSE value for significant wave heights above 0.75m is 0.1221(m). Figure 7 As shown, the hourly time series diagram of the effective wave height at Qianwan location throughout 2023, where the blue solid line is the calculation result of the wave element prediction model of this embodiment, and the red solid line is the calculation result of the SWAN model. Figure 7The highest RMSE value is 0.091 (m), and the highest RMSE value for significant wave heights above 0.75 m is 0.2289 (m). It can be seen that the wave element prediction model of this embodiment can accurately predict the distribution of wave field elements.

[0110] In addition, for the calculation of one year's data, the operation time of the traditional SWAN model is 54 hours, and the required hardware resources are CPU (58 cores). The operation time of the wave element prediction model of this embodiment is 19.7 seconds, and the required hardware resources are GPU (NVIDIA-RTX4070TI SUPER). Compared with the SWAN model, the wave element prediction model of this embodiment reduces the demand for computing resources, improves computing efficiency, and speeds up response speed.

[0111] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A high spatial resolution wave element prediction method, characterized in that: The steps include: S1: Determine the target sea area, obtain the public meteorological reanalysis data and wave reanalysis data according to the scope of the target sea area, and prepare the initial field file and boundary file for the simulation calculation of the shallow water wave numerical model in the target sea area; S2: Construct the computational grid of the target sea area, and use the shallow water wave numerical model to simulate and calculate the wave elements with high spatial resolution based on the meteorological reanalysis data and wave reanalysis data obtained by S1, as well as the initial field file and boundary file constructed by S1; S3: Build a database based on S1 meteorological reanalysis data and wave reanalysis data, as well as high spatial resolution wave elements simulated and calculated by S2; S4: A wave element prediction model is constructed based on the Conv-LSTM model and the AutoEncoder model. The wave element prediction model includes a sequentially connected input layer, a compression structure, and a reconstruction structure. The wave element prediction model is iteratively trained using the database constructed by S3, so that the wave element prediction model learns the nonlinear mapping relationship between meteorological reanalysis data and wave reanalysis data and high spatial resolution wave elements until convergence; S5: Use the wave element prediction model iteratively trained in S4 to predict wave elements with high spatial resolution.

2. A high spatial resolution wave element prediction method according to claim 1, characterized in that: The S1 comprises the following steps: S11: Determine the target sea area, obtain the land boundary of the target sea area, and the water depth and altitude corresponding to each longitude and latitude point in the ocean area of ​​the target sea area; S12: According to the scope of the target sea area, obtain low spatial resolution meteorological reanalysis data and wave reanalysis data of the preset year from the public ERA5; S13: Based on the land boundary of the target sea area obtained in S11, the water depth and altitude corresponding to each longitude and latitude point in the ocean area of ​​the target sea area, and the low spatial resolution meteorological reanalysis data and wave reanalysis data of the preset year obtained in S12, an initial field file and a boundary file for simulating the wave elements of the target sea area data are prepared.

3. A high spatial resolution wave element prediction method according to claim 2, characterized in that: The S2 comprises the following steps: S21: construct a computational grid for the target sea area, where the grid is refined near the coast; S22: Based on the meteorological reanalysis data and wave reanalysis data obtained in S12, and the initial field file and boundary file constructed in S13, the SWAN model is used to simulate and calculate the high spatial resolution wave elements of the preset year; S23: Obtain high spatial resolution wave elements of a preset year simulated and calculated by the SWAN model in S22, and remove invalid values.

4. A high spatial resolution wave element prediction method according to claim 3, characterized in that: The S3 comprises the following steps: S31: the low spatial resolution meteorological reanalysis data and wave reanalysis data of the preset year obtained in S12 and the high spatial resolution wave elements of the preset year obtained in S23 are matched in time with each other; S32: The low spatial resolution meteorological reanalysis data and wave reanalysis data after time correspondence, as well as the high spatial resolution wave elements, are divided into windows and stacked, and combined with the longitude and latitude points of the target sea area, a database with 12 hours of data as a set of samples is constructed.

5. A high spatial resolution wave element prediction method according to claim 1, characterized in that: The input layer of the wave element prediction model is a Conv-LSTM model. The compression structure includes 7 layers, which are alternating three-dimensional convolutional layers and pooling layers. The reconstruction structure includes 12 layers, which are alternating upsampling layers and three-dimensional convolutional layers.

6. A high spatial resolution wave element prediction method according to claim 1, characterized in that: The data in the S3 constructed database is divided by year. In S4, transfer learning is used to train the wave element prediction model year by year. After each training, the training parameters are adjusted according to the test results.

7. A high spatial resolution wave element prediction method according to claim 6, characterized in that: In transfer learning, the mean square error MSE is calculated through the following steps: S41: Constructing a three-dimensional mean square error matrix , where m represents the number of moments in each set of data in the database, and n represents the spatial resolution of the high spatial resolution wave elements at a single moment output by the wave element prediction model. Medium Element The calculation formula is as follows: ， In the formula, The high spatial resolution wave elements output by the wave element prediction model. is the wave element with high spatial resolution in the database. Point time on land is 0, when Point time on the ocean is 1; S42: Based on the constructed three-dimensional mean square error matrix Calculate the mean squared error (MSE).

8. A high spatial resolution wave element prediction method according to claim 6, characterized in that: In transfer learning, the mean square error MSE is calculated through the following steps: S41: Constructing the initial three-dimensional mean square error matrix , where m represents the number of moments in each set of data in the database, and n represents the spatial resolution of the high spatial resolution wave elements at a single moment output by the wave element prediction model. Medium Element The calculation formula is as follows: ， In the formula, The high spatial resolution wave elements output by the wave element prediction model. is the wave element with high spatial resolution in the database. Point time on land is 0, when Point time on the ocean is 1; S42: Calculate the mean and variance of the high spatial resolution wave elements at each moment of the first point in the current group of data; S43: Determine in turn whether the wave element with high spatial resolution at each moment of the first point in the current group of data is greater than the sum of the mean and variance calculated in S42. If so, set , if otherwise ; S44: Execute S42 and S43 cyclically, according to the calculated Constructing a three-dimensional mean square error matrix ; S45: Based on the constructed three-dimensional mean square error matrix Calculate the mean squared error (MSE).

9. A high spatial resolution wave element prediction method according to claim 1, characterized in that: The S5 comprises the following steps: S51: Obtaining the public meteorological reanalysis data and wave reanalysis data of the target sea area within the time range to be predicted; S52: Based on the meteorological reanalysis data and wave reanalysis data obtained in S51, the wave element prediction model iteratively trained in S4 is used to predict the wave elements with high spatial resolution within the time range to be predicted.

Images