Method and system for correcting wave height data of a sea wave model using satellite data

Abstract

The application discloses a method for correcting wave height data of a sea wave model by using satellite data, and comprises the following steps: traversing the satellite data, and matching the satellite data with corresponding sea wave model data; establishing a preset full connection neural network with the sea wave integral parameter of the wave height of the sea wave component as input and the corrected component wave height as output; establishing a preset full connection neural network with the component residual wave height as input and the corrected component residual wave height as output; connecting the plurality of preset component wave height correction networks and the preset residual wave height correction network in parallel, and taking the square root of the sum of the output items of each network as the total output, namely the corrected overall significant wave height; training the preset parallel network by minimizing the mean square error of the overall significant wave height output by the parallel network and the matched satellite altimeter significant wave height, so as to obtain a neural network model for correcting the significant wave height data of the sea wave model; and applying the model to the output result of the sea wave model which needs to be corrected, so as to improve the accuracy of the sea wave model data.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of oceanography, and in particular to a method and system for correcting wave height data of a sea wave model using satellite data. BACKGROUND

[0002] Sea wave data is widely used in the field of oceanography and plays an important role in various marine research and applications such as marine engineering technology, marine physics, marine remote sensing, marine disaster assessment and prediction. Currently, there are two main methods for obtaining sea wave data: field observation and numerical simulation. Field observation includes buoy observation and satellite remote sensing measurement. The sea wave data obtained by field observation has certain spatial and temporal limitations. For example, field buoys are expensive and are usually concentrated in near-shore areas. They often have data gaps due to malfunctions or maintenance, resulting in limited temporal and spatial coverage of the observed data. Satellite remote sensing data can cover almost the entire globe, but the spatial coverage and temporal resolution are low. For a specific location, data can only be obtained for a short period of time when each satellite passes. Numerical simulation can effectively reduce the spatial and temporal limitations of field observation, but the accuracy of the sea wave model data obtained by numerical simulation needs to be improved due to errors in the calculation process.

[0003] Using satellite data for field observation to improve the accuracy of sea wave model wave height data can improve the accuracy of sea wave model wave height data while ensuring spatial and temporal coverage of sea wave wave height data, which has great application prospects. Currently, there is a lack of research on methods for correcting sea wave model wave height data in China. SUMMARY

[0004] The main purpose of the present application is to improve the accuracy of sea wave model wave height data, so that the sea wave model wave height data with high spatial and temporal coverage has high accuracy.

[0005] The technical solution adopted by the present application is:

[0006] A method for correcting sea wave model wave height data using satellite data is provided, comprising the following steps:

[0007] S1, obtaining satellite data and sea wave model data in the same time period, traversing the satellite data, extracting sea wave model data consistent with the spatial and temporal positions, and forming a data pair as a training sample set for a pre-set neural network model;

[0008] S2, establishing a plurality of first pre-set fully connected neural networks for correcting sea wave model component wave height data, the input of each sea wave component in the sea wave model being a sea wave integral parameter, and the output result of each component wave height being a corrected sea wave model component;

[0009] S3, a second preset fully connected neural network for correcting residual wave height data of a component in the sea wave mode is established, the input is the residual wave height of the component in the sea wave mode, and the output result is the corrected residual wave height of the component in the sea wave mode;

[0010] S4, a third preset fully connected neural network for correcting significant wave height data of the sea wave mode is established, specifically, a plurality of first preset fully connected neural networks and the second preset fully connected neural network are connected in parallel, and the square sum of the output items of the first and second preset fully connected neural networks is taken as the total output of the parallel neural network after being taken as the square root, as the corrected total significant wave height;

[0011] S5, the third preset fully connected neural network is trained, and the minimization of the mean square error of the total significant wave height output by the preset parallel neural network and the significant wave height of the matched satellite data in step S1 is taken as the training target, to obtain a final neural network model for correcting the significant wave height data of the sea wave mode;

[0012] S6, the final neural network model is applied to the output result of the sea wave mode that needs to be corrected, and each component wave height and the total significant wave height in the sea wave mode are corrected.

[0013] According to the above technical scheme, the satellite data is significant wave height data observed by a satellite, and the type of the on-board sensor used is a microwave altimeter or a sea wave spectrometer.

[0014] According to the above technical scheme, the sea wave mode includes WAVEWATCH III, SWAN, and WAM, and the output parameters of the sea wave mode include the sea wave integral parameters of each sea wave component after two-dimensional spectrum segmentation and the total significant wave height.

[0015] According to the above technical scheme, in step S1, the spatio-temporal grid in which the satellite data is located in the sea wave mode is specifically found, and the data of the spatio-temporal grid is matched with the satellite observed data.

[0016] According to the above technical scheme, the correction refers to improving the accuracy of the data, so that the data output by the sea wave mode is closer to the actual true (observed) value; the component wave height of each component of the corrected sea wave mode, the corrected residual wave height of the component of the sea wave mode, and the corrected total significant wave height are the component wave height, the component residual wave height, and the total significant wave height that are closer to the satellite data observed on site (i.e. the accuracy is improved) after applying the method.

[0017] According to the above technical scheme, the sea wave integral parameters of each sea wave component after two-dimensional spectrum segmentation include the component wave height of each sea wave component, and the component period, component wavelength, component wave direction, and component wave spread of each sea wave component.

[0018] According to the technical scheme, the full connection neural network updates network parameters itself for realizing a training target according to input features; the first and second preset full connection neural networks have different parameters due to different inputs, but the first and second preset full connection neural networks all include an input layer, at least one hidden layer and an output layer, and have the same structure (number of layers and number of neurons in each layer).

[0019] According to the technical scheme, the third preset full connection neural network and the final neural network model are both composed of multiple full connection neural networks in parallel, wherein the number of full connection neural networks used for parallel connection is N+1, which is the number of partitions of the sea wave mode output data:

[0020] Wherein the component wave height, component wave direction, component period and component wave spread of each sea wave component are respectively input of one of the first preset full connection neural networks in the first N preset full connection neural networks, and the last second preset full connection neural network takes the residual wave height ε as input, and the residual wave height ε is defined as follows:

[0021]

[0022] Wherein Hs is the overall significant wave height of a position of the mode output, pHs i is the component wave height of the i-th sea wave component at the position, and N is the number of sea wave components of the mode output;

[0023] The (N+1) neural networks are connected in parallel, and the square root of the sum of the (N+1) output results is minimized as a training target with the mean square error of the satellite data significant wave height, and the final neural network model of the corrected sea wave mode wave height data is obtained by training the parallel neural networks.

[0024] According to the technical scheme, the parallel connection refers to a connection mode between the (N+1) neural networks, and there is no connection between the neurons between the (N+1) neural networks, wherein the change of the input of one network will not affect the output result of another network, but will affect the output result of itself and the sum of squares of all network output results; the reason for using this connection mode is that each sea wave component in the ocean satisfies the energy superposition principle and the independent propagation principle, and each sea wave component has no influence on each other; therefore, in the process of correcting by using the neural network, the information of each sea wave component is taken as an independent branch of the parallel neural network, and the component wave height of the sea wave component is corrected.

[0025] According to the technical scheme, the training of the third preset full connection neural network is that the multiple first preset full connection neural networks and the second preset full connection neural network connected in parallel adjust the weight values of their respective networks to obtain their respective output results, and minimize the square root of the sum of the output results with the mean square error of the satellite data significant wave height.

[0026] According to the technical solution, the third preset full connection neural network training target is minimization of mean square error of the corrected total significant wave height and the matched satellite data significant wave height, and the smaller the mean square error is, the closer the values of the total significant wave height and the matched satellite data significant wave height are, and the higher the accuracy of the corrected sea wave mode data is, that is, the better the correction effect is.

[0027] According to the technical solution, the results of the first N neural networks are component wave heights of the corrected components, the result of the (N+1)th neural network is a corrected residual wave height, and the square root of the sum of the results of the (N+1) parallel networks is a corrected total significant wave height; because the output result of the third preset full connection neural network is the square root of the sum of the output results of the multiple first preset full connection neural networks and the second preset full connection neural network, when the third preset full connection neural network can obtain a more accurate total significant wave height result, the corresponding multiple first preset full connection neural networks will also output more accurate component wave height results.

[0028] The application also provides a system for correcting sea wave mode wave height data by using satellite data, which comprises:

[0029] A training sample set collection module is configured to obtain satellite data and sea wave mode data in the same time period, traverse the satellite data, extract sea wave mode data consistent with the satellite data in terms of spatial and temporal positions, and form data pairs as a training sample set of a preset neural network model.

[0030] A first neural network module is configured to establish multiple first preset full connection neural networks for correcting sea wave mode component wave height data, and the inputs are sea wave integral parameters of each sea wave component in the sea wave mode, and the outputs are component wave heights of the corrected components in the sea wave mode.

[0031] A second neural network module is configured to establish a second preset full connection neural network for correcting sea wave mode component residual wave height data, and the input is a component residual wave height in the sea wave mode, and the output is a corrected sea wave mode component residual wave height.

[0032] A third neural network module is configured to establish a third preset full connection neural network for correcting sea wave mode significant wave height data, and the multiple first preset full connection neural networks and the second preset full connection neural network are connected in parallel, the square root of the sum of the outputs of the first and second preset full connection neural networks is taken as the total output of the parallel neural network, and the total output is taken as a corrected total significant wave height.

[0033] a final neural network model module, configured to train the third preset full connection neural network, take minimization of mean square error between the overall significant wave height output by the parallel neural network and the matched satellite data significant wave height in step S1 as a training target, and obtain a final neural network model for correcting the significant wave height data of the sea wave model;

[0034] a final correction module, configured to apply the final neural network model to the output result of the sea wave model in need of correction, and correct the component wave height and the overall significant wave height in the sea wave model.

[0035] The application further provides a computer storage medium, which stores a computer program executable by a processor, and the computer program executes the method for correcting the wave height data of the sea wave model by using satellite data in the technical solution.

[0036] The application has the following beneficial effects: the application takes the satellite data and the sea wave model data matched with the satellite data in space and time as a training sample set, establishes N preset full connection neural networks for correcting the component wave height data of the sea wave model, establishes a preset full connection neural network for correcting the component residual wave height data of the sea wave model, trains the (N+1) full connection neural networks in parallel, and obtains a neural network model for correcting the wave height data of the sea wave model. The application considers the feature differences of the component data of the sea wave model, establishes full connection neural networks for different sea wave components, trains the full connection neural networks in parallel, obtains the component wave height of each component after correction, the residual wave height after correction, and the overall significant wave height data after correction, improves the accuracy of the sea wave model data, and makes the corrected sea wave model data have high space-time resolution and high accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0037] The application will be further described below with reference to the drawings and embodiments. In the drawings:

[0038] Figure 1 The application is an embodiment of a method for correcting sea wave model data by using satellite data, and a flowchart of the method is shown in the figure.

[0039] Figure 2 The application is an embodiment of a method for correcting sea wave model data by using satellite data, and a flowchart of the method is shown in the figure.

[0040] Figure 3 The application is an embodiment of a method for correcting sea wave model data by using satellite data, and a flowchart of the method is shown in the figure. DETAILED DESCRIPTION

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0042] like Figure 1 As shown, in one embodiment of the present invention, the method for correcting ocean wave pattern data using satellite data includes the following steps:

[0043] S1. Obtain satellite data and wave pattern data for the same year, traverse altimeter satellite data, read the corresponding wave pattern data, form data pairs, and use them as training sample sets for the pre-set neural network model.

[0044] Specifically, in one embodiment of the present invention, the satellite data is the effective wave height data observed by the satellite, and the type of spaceborne sensor used can be a microwave altimeter or a wave spectrometer; the wave model data includes WAVEWATCH III, SWAN, and WAM, and the output parameters of the wave model include the wave integral parameters of each wave component after two-dimensional spectral segmentation and the overall effective wave height. The wave integral parameters of each wave component after two-dimensional spectral segmentation need to include the component wave height of each wave component, and optionally include the component period, component wavelength, component wave direction, component wave span, and other integral parameters of each wave component.

[0045] In one embodiment of the present invention, the effective wave height data of the altimeter satellite is read, and the data is traversed according to the time dimension to find the spatiotemporal grid in the wave pattern. The wave pattern data of the spatiotemporal grid is matched with the data observed by the satellite to form a data pair, which is stored in a two-dimensional matrix as a training sample set.

[0046] S2. Establish a first pre-set fully connected neural network with N corrected wave heights of wave model components: take the wave integral parameter of a certain wave component of the wave model as input and the corrected wave height of the wave component as output.

[0047] S3. Establish a second pre-set fully connected neural network for correcting the residual wave height of the wave pattern: take the residual wave height of the wave pattern component as input and the corrected residual wave height of the component as output.

[0048] Correction refers to improving the accuracy of data, making the data output by the wave model closer to the actual (observed) values. The component wave heights, component residual wave heights, and overall significant wave heights of each component of the corrected wave model are component wave heights, component residual wave heights, and overall significant wave heights that are closer to the on-site observation satellite data (i.e., the accuracy is improved) after applying the method of this patent.

[0049] The full connection neural network of the embodiment of the application continuously updates its network parameters according to input features to achieve a training target; the first and second preset full connection neural networks have different parameters due to different inputs, but both of them include an input layer, at least one hidden layer and an output layer, and have the same structure (number of layers and number of neurons in each layer).

[0050] Specifically, in an embodiment of the application, as shown in Figure 3 There are three sea wave mode sea wave components, and the sea wave integral parameters of each sea wave component include four, pHs0, pHs1, pHs2, pDir0, pDir1, pDir2, pSpr0, pSpr1, pSpr2, pTp0, pTp1, pTp2. The integral parameters are divided into three matrices with a dimension of (4, 1) according to the sea wave components (pHs0, pDir0, pSpr0, pTp0), (pHs1, pDir1, pSpr1, pTp1), (pHs2, pDir2, pSpr2, pTp2), which are used as inputs of the preset full connection neural network for correcting the wave height of the first three sea wave mode components. Remain_Hs(ε) is a component residual wave height, which is a matrix with a dimension of (1, 1) and is used as an input of the preset full connection neural network for correcting the residual wave height of the last sea wave mode. The definition of Remain_Hs(ε) is as follows.

[0051]

[0052] In the formula, Hs is the overall significant wave height of a certain position of the mode output, pHs i is the component wave height of the i-th sea wave component at the position, and N is the number of sea wave components of the mode output.

[0053] S4, the N preset component wave height correction networks (i.e., the first preset full connection neural network) and the preset residual wave height correction network (i.e., the second preset full connection neural network) are connected in parallel to obtain a third preset full connection neural network for correcting the sea wave mode significant wave height data. The square root of the sum of the outputs of each network is taken as the total output of the parallel network, i.e., the corrected overall significant wave height.

[0054] Specifically, in an embodiment of the application, as shown in Figure 2 the structural diagram of the neural network, the four full connection neural networks are connected in parallel, and each full connection neural network includes an input layer, at least one hidden layer and an output layer.

[0055] The connection mode between the (N+1) parallel neural networks is that there is no connection between neurons between the (N+1) neural networks, wherein the change of the input of one network does not affect the output result of another network, but affects the output result of itself and the square sum of all network output results; the connection mode is used because the energy superposition principle and the independent propagation principle are met between each sea wave component in the ocean, and each sea wave component has no influence on each other; therefore, in the process of correction by using the neural network, the information of each sea wave component is taken as an independent branch of the parallel neural network, and the component wave height of the sea wave component is corrected.

[0056] The training of the third preset full connection neural network is that the multiple first preset full connection neural networks and the second preset full connection neural network are adjusted respectively to obtain respective output results, and the square root of the square sum of the output results is minimized to the mean square error of the satellite data effective wave height. The training target of the third preset full connection neural network is to minimize the mean square error of the corrected total effective wave height and the matched satellite data effective wave height, and the smaller the mean square error is, the closer the values of the total effective wave height and the matched satellite data effective wave height are, and the higher the accuracy of the corrected sea wave mode data is, that is, the better the correction effect is.

[0057] The results of the first N neural networks are respectively the component wave heights of the corrected components, the result of the (N+1)th neural network is the corrected residual wave height, and the square root of the square sum of the output results of the (N+1) parallel networks is the total effective wave height after correction; because the output result of the third preset full connection neural network is the square root of the square sum of the output results of the multiple first preset full connection neural networks and the second preset full connection neural network, when the third preset full connection neural network can obtain a more accurate total effective wave height result, the multiple first preset full connection neural networks will also output more accurate component wave height results.

[0058] S5, minimizing the mean square error of the total effective wave height output by the parallel neural network and the matched satellite effective wave height data, training the preset parallel neural network to obtain a neural network model for correcting the sea wave mode effective wave height data.

[0059] Specifically, in an embodiment of the present application, as shown in Figure 3 the sea wave integral parameter matrix of the four sea wave components in step S3 is taken as the input of the four full connection neural networks, and the square root of the square sum of the four dimension (1, 1) matrices output by the four full connection neural networks: pHs0_cal, pHs1_cal, pHs2_cal, pHsR_cal is taken as the total output Alt_Hs of the parallel network:

[0060]

[0061] wherein pHsi_cal is the corrected component wave height output by the i-th sea wave component as input, and pHsR_cal is the corrected component residual wave height.

[0062] S6, apply the neural network model for correcting the significant wave height data of the sea wave mode to the sea wave mode data in need of correction, correct the component wave heights and the overall significant wave height in the sea wave mode, and improve the data accuracy.

[0063] The system for correcting the wave height data of the sea wave mode according to the embodiments of the present application is mainly used for implementing the above method embodiments, and the system comprises:

[0064] The training sample set collection module is used for acquiring satellite data and sea wave mode data in the same time period, traversing the satellite data, extracting the sea wave mode data consistent with the satellite data in the spatial and temporal positions, and composing data pairs as the training sample set of the preset neural network model.

[0065] The first neural network module is used for establishing a plurality of first preset fully connected neural networks for correcting the component wave height data of the sea wave mode, the inputs of which are the sea wave integral parameters of each sea wave component in the sea wave mode, and the output results of which are the component wave heights of the corrected sea wave mode components.

[0066] The second neural network module is used for establishing a second preset fully connected neural network for correcting the component residual wave height data of the sea wave mode, the input of which is the component residual wave height of the sea wave mode, and the output result of which is the corrected component residual wave height of the sea wave mode.

[0067] The third neural network module is used for establishing a third preset fully connected neural network for correcting the significant wave height data of the sea wave mode, specifically, the plurality of first preset fully connected neural networks and the second preset fully connected neural network are connected in parallel, and the square sum of the output items of the first and second preset fully connected neural networks is taken as the total output of the parallel neural network after being taken the square root, as the corrected overall significant wave height.

[0068] The final neural network model module is used for training the third preset fully connected neural network, taking the minimization of the mean square error between the overall significant wave height output by the preset parallel neural network and the significant wave height of the matched satellite data in step S1 as the training target, and obtaining the final neural network model for correcting the significant wave height data of the sea wave mode.

[0069] The final correction module is used for applying the final neural network model to the output result of the sea wave mode in need of correction, and correcting the component wave heights and the overall significant wave height in the sea wave mode.

[0070] The functions of the various modules and each step of the method embodiments are relative, and will not be repeated here.

[0071] The application further provides a computer readable storage medium, such as a flash memory, a hard disk, a multimedia card, a card memory (for example, an SD or DX memory, etc.), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, an optical disk, a server, an App application store, and the like, which stores a computer program, and the program is executed by a processor to realize corresponding functions. The computer readable storage medium of the embodiment is executed by the processor to realize the method for correcting wave height data of a sea wave model by using satellite data.

[0072] To sum up, the application considers the feature differences of each sea wave component data in the sea wave model, establishes a full connection neural network for training for different sea wave components, obtains the component wave height of each component after correction, the residual wave height after correction, and the overall effective wave height data after correction by connecting the multiple full connection neural networks in parallel, improves the accuracy of the sea wave model data, and makes the corrected sea wave model data have high spatial and temporal resolution and high accuracy.

[0073] It should be understood that, for those skilled in the art, improvements or changes can be made according to the above description, and all these improvements and changes shall belong to the protection scope of the appended claims of the application.

Claims

1. A method for correcting wave height data of ocean wave patterns using satellite data, characterized in that, Includes the following steps: S1. Acquire satellite data and wave pattern data within the same time period, traverse the satellite data, extract wave pattern data that is consistent with it in time and space, and form data pairs as training sample sets for the pre-set neural network model. S2. Establish multiple first pre-set fully connected neural networks for correcting wave height data of wave model components. The inputs are the wave integral parameters of each wave component in the wave model, and the outputs are the component wave heights of each component of the corrected wave model. S3. Establish a second pre-set fully connected neural network for correcting the residual wave height data of wave model components. The input is the residual wave height of the wave model components, and the output is the corrected residual wave height of the wave model components. S4. Establish a third pre-set fully connected neural network for correcting the effective wave height data of the wave pattern. Specifically, multiple first pre-set fully connected neural networks and second pre-set fully connected neural networks are connected in parallel. The sum of the squares of the output terms of the first and second pre-set fully connected neural networks, after taking the square root, is used as the total output of the parallel neural network, which is used as the corrected overall effective wave height. S5. Train the third pre-set fully connected neural network, taking the minimization of the mean square error between the total effective wave height output by the pre-set parallel neural network and the effective wave height of the matched satellite data in step S1 as the training objective, to obtain the final neural network model for correcting the effective wave height data of the wave model. S6. Apply the final neural network model to the output of the wave pattern that needs correction, and correct the wave height of each component and the overall significant wave height in the wave pattern.

2. The method for correcting ocean wave pattern wave height data using satellite data according to claim 1, characterized in that, The satellite data is effective wave height data observed by the satellite, and the types of onboard sensors used are microwave altimeters or wave spectrometers.

3. The method for correcting ocean wave pattern wave height data using satellite data according to claim 1, characterized in that, The wave models include WAVEWATCH III, SWAN, and WAM. The output parameters of the wave models include the wave integral parameters of each wave component after two-dimensional spectral segmentation and the total significant wave height.

4. The method for correcting ocean wave pattern wave height data using satellite data according to claim 1, characterized in that, In step S1, the spatiotemporal grid in which the satellite data is located in the wave pattern is specifically located, and the data in that spatiotemporal grid is matched with the data observed by the satellite.

5. The method for correcting ocean wave pattern wave height data using satellite data according to claim 3, characterized in that, The wave integral parameters of each wave component after the two-dimensional spectrum segmentation include the component wave height, component period, component wavelength, component wave direction, and component wave span of each wave component.

6. The method for correcting ocean wave pattern wave height data using satellite data according to claim 1, characterized in that, The first and second pre-built fully connected neural networks each include an input layer, at least one hidden layer, and an output layer.

7. The method for correcting wave height data of ocean wave patterns using satellite data according to claim 1, characterized in that, Both the third pre-built fully connected neural network and the final neural network model are composed of multiple fully connected neural networks connected in parallel. The number of fully connected neural networks used for parallel connection is N plus one, which is the number of partitions in the wave pattern output data. The component wave height, component wave direction, component period, and component wave span of each wave component are respectively the inputs of one of the first N pre-set fully connected neural networks. The last second pre-set fully connected neural network takes the residual wave height ε as input, and the residual wave height ε is defined as follows; In the formula, Hs is the overall effective wavelength at a certain position in the mode output, and pHs is... i It is the wave height of the i-th wave component at that position, and N is the number of wave components output by the mode; (N+1) neural networks are connected in parallel, and the training objective is to minimize the mean square error of the sum of the squares of their (N+1) outputs and the effective wave height of the satellite data. The parallel neural networks are then trained to obtain the final neural network model for correcting the wave height data of the wave pattern.

8. The method for correcting ocean wave pattern wave height data using satellite data according to claim 7, characterized in that, The outputs of the first N neural networks are the component wave heights of each component after correction, the output of the (N+1)th neural network is the residual wave height after correction, and the square root of the sum of squares of the outputs of the (N+1)th parallel network is the overall effective wave height after correction.

9. A system for correcting wave height data of ocean wave patterns using satellite data, characterized in that, include: The training sample set collection module is used to acquire satellite data and wave pattern data within the same time period, traverse the satellite data, extract wave pattern data that is consistent with it in time and space, and form data pairs as the training sample set for the pre-set neural network model. The first neural network module is used to establish multiple first pre-set fully connected neural networks for correcting wave height data of wave model components. The inputs are the wave integral parameters of each wave component in the wave model, and the outputs are the component wave heights of each component of the corrected wave model. The second neural network module is used to establish a second pre-set fully connected neural network for correcting the residual wave height data of wave model components. The input is the residual wave height of the components in the wave model, and the output is the corrected residual wave height of the wave model components. The third neural network module is used to establish a third preset fully connected neural network for correcting the effective wave height data of the wave pattern. Specifically, multiple first preset fully connected neural networks and second preset fully connected neural networks are connected in parallel. The sum of the squares of the output terms of the first and second preset fully connected neural networks, after taking the square root, is used as the total output of the parallel neural network, which is used as the corrected overall effective wave height. The final neural network model module is used to train the third pre-set fully connected neural network. The training objective is to minimize the mean square error between the total significant wave height output by the pre-set parallel neural network and the significant wave height of the matched satellite data in step S1, so as to obtain the final neural network model for correcting the significant wave height data of the wave model. The final correction module is used to apply the final neural network model to the output of the wave pattern that needs correction, and to correct the wave height of each component and the overall significant wave height in the wave pattern.

10. A computer storage medium, characterized in that, It contains a computer program that can be executed by a processor, which performs the method of correcting wave height data of ocean wave patterns using satellite data as described in any one of claims 1-8.

Images