A sea wave prediction method and device based on deep learning

Abstract

The application provides a sea wave prediction method and device based on deep learning, and relates to the technical field of sea wave prediction. The method determines physical derived quantities related to the generation, breaking and dissipation processes of sea waves according to wind field data and water depth data, uses the physical derived quantities, the wind field data, the water depth data and the sea wave state at the previous moment to jointly construct multi-channel input data, and further constructs a physical reference sea wave state at the target moment according to the physical derived quantities and the sea wave state at the previous moment. Thus, the same physical information can simultaneously act on the input process and the training constraint process of the sea wave prediction network, so that the sea wave prediction network is constrained by the sea wave evolution law while learning the sea wave change relationship in the sample data, thereby reducing the prediction results that do not conform to the generation, breaking and dissipation rules of sea waves, and improving the stability and physical rationality of the sea wave prediction results.

Description

Technical Field

[0001] This application relates to the field of wave prediction technology, and more specifically, to a wave prediction method and apparatus based on deep learning. Background Technology

[0002] Ocean wave conditions are a crucial environmental factor affecting ship navigation, marine engineering construction, offshore operations, and coastal disaster prevention and mitigation. Accurate prediction of wave parameters such as significant wave height, peak wave period, and peak wave direction is of great significance for marine environmental forecasting and maritime safety. Current wave prediction methods typically employ numerical wave models, obtaining wave parameters at different times and spatial locations by numerically solving the wave evolution process. However, these methods are computationally complex and computationally intensive, making it difficult to balance prediction efficiency and spatial resolution.

[0003] With the development of artificial intelligence technology, deep learning models are increasingly being used for wave prediction. Deep learning models can learn the nonlinear relationship between environmental factors and wave states based on historical data, exhibiting high inference efficiency. However, existing deep learning models primarily rely on sample data for training, making their prediction results susceptible to the distribution and quality of the training data. Their ability to represent the evolutionary patterns of waves is limited, potentially leading to predictions that do not conform to actual wave variations. Therefore, improving the stability and physical plausibility of deep learning wave prediction results is a problem that needs to be addressed. Summary of the Invention

[0004] In view of this, this application provides a method and apparatus for predicting ocean waves based on deep learning.

[0005] Specifically, this application is implemented through the following technical solution: In a first aspect, this application provides a deep learning-based wave prediction method, comprising: Acquire wind field data, water depth data, wave state at the previous moment, and sample wave state at the target moment for the target sea area. The wave state includes significant wave height, peak wave period, and peak wave direction. The wind energy input potential is determined based on the wind field data, the water depth limit breakage threshold and bottom friction indication are determined based on the water depth data, and multi-channel input data is constructed based on the wind field data, the water depth data, the previous wave state, the wind energy input potential, the water depth limit breakage threshold and the bottom friction indication. The multi-channel input data is input into the wave prediction network to obtain the predicted wave state at the target time; The physical reference wave state at the target time is constructed based on the wave state at the previous moment, the wind energy input potential, the water depth limit breakage threshold, and the bottom friction indicator. A joint loss is determined based on the predicted wave state, the sample wave state at the target time, and the physical reference wave state. The wave prediction network is then trained based on the joint loss. Using the trained wave prediction network, the predicted wave state for the time to be predicted is output based on the wind field data, water depth data, and wave state at the time to be predicted and the previous time.

[0006] In one possible implementation, constructing the multi-channel input data includes: Obtain the geographical location and time information of the target sea area; The wind field data, water depth data, and wave state are correlated temporally and spatially, missing and invalid values ​​are processed, and land grids are identified based on the water depth data to establish a land-sea mask. The wind field data, water depth data, and wave state are normalized. Based on the wind field data, the Pearson-Moscowitz reference wave parameters are determined, and the geographic location code and time code are determined based on the geographic location and time information, respectively. The multi-channel input data is constructed based on the processed data, the Pearson-Moscowitz reference wave parameters, the geographic location encoding, and the time encoding.

[0007] In one possible implementation, determining the wind energy input potential, the water depth-limited breakage threshold, and the bottom friction indicator includes: The wind friction speed is determined based on the wind speed and drag coefficient at 10m above the sea surface, and the wind energy input potential is determined based on the wind friction speed, air density, and seawater density. The water depth data is subjected to amplitude limiting processing to obtain the effective water depth; The water depth limit breakage threshold is determined based on the effective water depth and the water depth limit breakage coefficient, and the bottom friction indication is determined based on the effective water depth and the bottom friction coefficient.

[0008] In one possible implementation, constructing the physical reference wave state includes: The wave energy at the previous moment is determined based on the effective wave height in the wave state at the previous moment; The wave energy growth term is determined based on the wind energy input potential, the bottom friction dissipation term is determined based on the bottom friction indicator and the wave energy at the previous moment, and the breakage dissipation term is determined based on the water depth limit breakage threshold and the wave energy at the previous moment. The physical reference wave energy is determined based on the wave energy at the previous moment, the wave energy growth term, the bottom friction dissipation term, and the fragmentation dissipation term. The effective wave height of the physical reference can be determined based on the physical reference wave energy.

[0009] In one possible implementation, constructing the physical reference wave state further includes: The physical reference peak wave period is determined based on the peak wave period of the previous moment and the reference peak wave period corresponding to the wind field data. The physical reference peak wave direction is determined based on the peak wave direction of the previous moment and the wind direction corresponding to the wind field data. The physical reference wave state is obtained by coupling constraints on the physical reference effective wave height and the physical reference peak wave period according to the wave steepness relation, coupling constraints on the physical reference peak wave period and the water depth data according to the dispersion relation, and coupling constraints on the physical reference peak wave direction, the wave propagation direction along the group velocity direction and the water depth gradient of the water depth data according to the refraction relation.

[0010] In one possible implementation, the multi-channel input data is input into a wave prediction network to obtain the predicted wave state, including: The residual U-Net network is used to perform multi-scale feature extraction and feature fusion on the multi-channel input data to obtain the sine and cosine components of the predicted significant wave height, predicted peak wave period, and predicted peak wave direction. The predicted peak wave direction is determined based on the sine and cosine components.

[0011] In one possible implementation, determining the joint loss includes: The data fitting loss is determined based on the difference between the predicted wave state and the wave state of the sample at the target time. The wave state evolution loss is determined based on the difference between the predicted wave state and the physical reference wave state. Based on the predicted wave state, determine at least one of the following: depth-limited breakage loss, wave steepness loss, wave energy propagation loss, and spatial smoothing loss; Based on the land-sea mask, the data fitting loss, the wave state evolution loss, and at least one determined loss are weighted within the ocean grid to obtain the joint loss.

[0012] In one possible implementation, training the wave prediction network includes: The multi-channel input data is constructed using the sample wave state of the previous moment or the predicted wave state of the previous moment output by the wave prediction network. As the number of training rounds increases, the proportion of multi-channel input data constructed using the predicted wave state from the previous moment is increased; According to the stated ratio, the weight of the wave state evolution loss and at least one of the depth-limited breaking loss, wave steepness loss, wave energy propagation loss and spatial smoothing loss is increased in the joint loss.

[0013] Secondly, this application provides a deep learning-based wave prediction device, comprising: The acquisition module is used to acquire wind field data, water depth data, wave state at the previous moment and sample wave state at the target moment for the target sea area. The wave state includes significant wave height, peak wave period and peak wave direction. The determination module is used to determine the wind energy input potential based on the wind field data, determine the water depth limit breakage threshold and bottom friction indication based on the water depth data, and construct multi-channel input data based on the wind field data, the water depth data, the previous wave state, the wind energy input potential, the water depth limit breakage threshold and the bottom friction indication. The first prediction module is used to input the multi-channel input data into the wave prediction network to obtain the predicted wave state at the target time. The training module is used to construct the physical reference wave state at the target time based on the wave state at the previous moment, the wind energy input potential, the water depth limit breakage threshold, and the bottom friction indicator, and to determine the joint loss based on the predicted wave state, the sample wave state at the target time, and the physical reference wave state, and to train the wave prediction network based on the joint loss. The second prediction module is used to utilize the trained wave prediction network to output the predicted wave state at the time to be predicted, based on the wind field data, water depth data, and wave state at the previous time.

[0014] In one possible implementation, the determining module is used to: Obtain the geographical location and time information of the target sea area; The wind field data, water depth data, and wave state are correlated temporally and spatially, missing and invalid values ​​are processed, and land grids are identified based on the water depth data to establish a land-sea mask. The wind field data, water depth data, and wave state are normalized. Based on the wind field data, the Pearson-Moscowitz reference wave parameters are determined, and the geographic location code and time code are determined based on the geographic location and time information, respectively. The multi-channel input data is constructed based on the processed data, the Pearson-Moscowitz reference wave parameters, the geographic location encoding, and the time encoding.

[0015] In one possible implementation, the determining module is used to: The wind friction speed is determined based on the wind speed and drag coefficient at 10m above the sea surface, and the wind energy input potential is determined based on the wind friction speed, air density, and seawater density. The water depth data is subjected to amplitude limiting processing to obtain the effective water depth; The water depth limit breakage threshold is determined based on the effective water depth and the water depth limit breakage coefficient, and the bottom friction indication is determined based on the effective water depth and the bottom friction coefficient.

[0016] In one possible implementation, the training module is used to: The wave energy at the previous moment is determined based on the effective wave height in the wave state at the previous moment; The wave energy growth term is determined based on the wind energy input potential, the bottom friction dissipation term is determined based on the bottom friction indicator and the wave energy at the previous moment, and the breakage dissipation term is determined based on the water depth limit breakage threshold and the wave energy at the previous moment. The physical reference wave energy is determined based on the wave energy at the previous moment, the wave energy growth term, the bottom friction dissipation term, and the fragmentation dissipation term. The effective wave height of the physical reference can be determined based on the physical reference wave energy.

[0017] In one possible implementation, the training module is also used for: The physical reference peak wave period is determined based on the peak wave period of the previous moment and the reference peak wave period corresponding to the wind field data. The physical reference peak wave direction is determined based on the peak wave direction of the previous moment and the wind direction corresponding to the wind field data. The physical reference wave state is obtained by coupling constraints on the physical reference effective wave height and the physical reference peak wave period according to the wave steepness relation, coupling constraints on the physical reference peak wave period and the water depth data according to the dispersion relation, and coupling constraints on the physical reference peak wave direction, the wave propagation direction along the group velocity direction and the water depth gradient of the water depth data according to the refraction relation.

[0018] In one possible implementation, the first prediction module is used for: The residual U-Net network is used to perform multi-scale feature extraction and feature fusion on the multi-channel input data to obtain the sine and cosine components of the predicted significant wave height, predicted peak wave period, and predicted peak wave direction. The predicted peak wave direction is determined based on the sine and cosine components.

[0019] In one possible implementation, the training module is used to: The data fitting loss is determined based on the difference between the predicted wave state and the wave state of the sample at the target time. The wave state evolution loss is determined based on the difference between the predicted wave state and the physical reference wave state. Based on the predicted wave state, determine at least one of the following: depth-limited breakage loss, wave steepness loss, wave energy propagation loss, and spatial smoothing loss; Based on the land-sea mask, the data fitting loss, the wave state evolution loss, and at least one determined loss are weighted within the ocean grid to obtain the joint loss.

[0020] In one possible implementation, the training module is used to: The multi-channel input data is constructed using the sample wave state of the previous moment or the predicted wave state of the previous moment output by the wave prediction network. As the number of training rounds increases, the proportion of multi-channel input data constructed using the predicted wave state from the previous moment is increased; According to the stated ratio, the weight of the wave state evolution loss and at least one of the depth-limited breaking loss, wave steepness loss, wave energy propagation loss and spatial smoothing loss is increased in the joint loss.

[0021] Thirdly, embodiments of this application also provide a computer device, including: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor communicates with the memory via the bus, and when the machine-readable instructions are executed by the processor, the steps of the first aspect above, or any possible implementation of the first aspect, are performed.

[0022] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps described in the first aspect or any possible implementation of the first aspect.

[0023] The deep learning-based wave prediction method and apparatus provided in this application can determine physical derivatives related to wave generation, breaking, and dissipation processes based on wind field data and water depth data. These physical derivatives, along with wind field data, water depth data, and the wave state at the previous moment, are used to construct multi-channel input data. Furthermore, a physical reference wave state for the target moment is constructed based on the physical derivatives and the wave state at the previous moment. Thus, the same physical information can simultaneously influence the input process and training constraint process of the wave prediction network. This allows the wave prediction network to learn the wave variation relationships in the sample data while being constrained by the laws of wave evolution, thereby reducing prediction results that do not conform to the laws of wave generation, breaking, and dissipation, and improving the stability and physical rationality of the wave prediction results.

[0024] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating an exemplary embodiment of a deep learning-based wave prediction method according to this application; Figure 2 This is a schematic diagram of a deep learning-based wave prediction device shown in an exemplary embodiment of this application; Figure 3 This is a schematic diagram of a computer device illustrated in an exemplary embodiment of this application. Detailed Implementation

[0026] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0027] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0028] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0029] Research has revealed that existing deep learning models primarily rely on sample data for training, making their predictions susceptible to the distribution and quality of the training data. This results in limited ability to represent the evolution of ocean waves, potentially leading to predictions that do not accurately reflect actual wave patterns. Therefore, improving the stability and physical plausibility of deep learning wave predictions is a crucial issue that needs to be addressed.

[0030] In view of this, this application provides a deep learning-based wave prediction method and apparatus. Based on wind field data and water depth data, physical derivatives related to wave generation, breaking, and dissipation processes are determined. These physical derivatives, along with wind field data, water depth data, and the wave state at the previous moment, are used to construct multi-channel input data. Furthermore, a physical reference wave state for the target moment is constructed based on the physical derivatives and the wave state at the previous moment. Thus, the same physical information can simultaneously influence the input process and training constraint process of the wave prediction network. This allows the wave prediction network to learn the wave variation relationships in the sample data while being constrained by the laws of wave evolution, thereby reducing prediction results that do not conform to the laws of wave generation, breaking, and dissipation, and improving the stability and physical rationality of the wave prediction results.

[0031] To facilitate understanding of this embodiment, the deep learning-based wave prediction method disclosed in this application will first be described in detail. The execution entity of the deep learning-based wave prediction method provided in this application is generally a computer device with a certain computing power. In some possible implementations, this deep learning-based wave prediction method can be implemented by a processor calling computer-readable instructions stored in memory.

[0032] See Figure 1 , Figure 1 The flowchart illustrates a deep learning-based wave prediction method as an exemplary embodiment of this application. The method includes steps S101 to S105. Wherein: S101. Obtain wind field data, water depth data, wave state at the previous moment, and sample wave state at the target moment for the target sea area. The wave state includes significant wave height, peak wave period, and peak wave direction.

[0033] The wave state includes significant wave height, peak wave period, and peak wave direction. The wave state at the previous moment is used to characterize the wave state before the target moment, and the sample wave state at the target moment is used as the training target for the wave prediction network.

[0034] Specifically, wind field data and wave state data of the target sea area at multiple consecutive moments can be obtained, and the data from two adjacent moments can be combined into training samples according to time order. Among them, the wave state at the earlier moment is used as the wave state at the previous moment, and the wave state at the later moment is used as the wave state of the target moment sample.

[0035] In the specific implementation process, wind field data, water depth data, geographical location information, time information and wave state data of a certain time period and a certain area of ​​nearshore waters can be obtained.

[0036] Wind field data can be obtained using the fifth-generation European Centre for Medium-Range Weather Forecasts Atmospheric Reanalysis dataset (ECMWFReanalysis v5, ERA5), which includes zonal and meridional wind speed components at 10 m above sea level, with a temporal resolution of 1 h and a spatial resolution of 0.25°×0.25°.

[0037] Ocean wave state data can be obtained from the output data of the ROMS-SWAN coupled model, which consists of the Regional Ocean Modeling System (ROMS) and the Simulating WAves Nearshore (SWAN) model. The data includes significant wave height, peak wave period, and peak wave direction, with a time resolution of 1 h and a spatial resolution of 5 km.

[0038] After acquiring the above data, time and spatial mappings were performed on the wind field data, water depth data, and wave state data, transforming data from different sources into corresponding time series and spatial grids. The water depth data was positiveized, and land grids were identified and a land-sea mask was established based on the processed water depth data. Missing and invalid values ​​were handled, and the wind field data, water depth data, and wave state data were normalized.

[0039] Geographic location information can include the longitude and latitude corresponding to each spatial grid. Temporal information can be processed using periodic encoding. For example, sine and cosine functions can be used to encode intra-year and intra-day times respectively to characterize the periodic changes in ocean wave states with seasons and time.

[0040] S102. Determine the wind energy input potential based on the wind field data, determine the water depth limit breakage threshold and bottom friction indication based on the water depth data, and construct multi-channel input data based on the wind field data, the water depth data, the previous wave state, the wind energy input potential, the water depth limit breakage threshold and the bottom friction indication.

[0041] The zonal and meridional wind speed components at a point 10 m above sea level are denoted as follows: and The wind speed at 10 m above sea level is determined using the following formula:

[0042] in, This represents the zonal wind speed component at a location 10 meters above sea level. This represents the meridional wind speed component at a distance of 10 m above sea level. This indicates the wind speed at a height of 10 meters above sea level; the subscript "10" indicates that the reference height for the corresponding wind speed or wind speed component is 10 meters above sea level. When using the International System of Units (SI), , and The units are all m / s.

[0043] Determine the Pierson-Moskowitz (PM) reference significant wave height and reference peak wave period based on wind speed:

[0044]

[0045] in, Indicates the PM reference significant wave height. Indicates the period of the PM reference peak wave. Indicates gravitational acceleration; subscript Indicates significant wave height, subscript Indicates the peak value in the peak wave period, subscript This indicates that the corresponding parameter is a reference value determined based on the PM relationship.

[0046] The wind friction velocity is determined based on wind speed and drag coefficient, and the wind energy input potential is determined based on wind friction velocity, air density, and seawater density.

[0047]

[0048]

[0049] in, Indicates the wind energy input potential. and These represent the air density and seawater density, respectively. Indicates the speed of wind friction. Indicates the drag coefficient; subscript Indicates wind energy input, subscript Indicates air, subscript Indicates seawater, subscript It indicates dragging.

[0050] The water depth data is subjected to amplitude limiting to obtain the effective water depth:

[0051] in, Indicates water depth. Indicates effective water depth, subscript This represents the effective value after amplitude limiting. When the water depth is expressed in meters (m), the above formula limits the effective water depth to the range of 0.5 m to 50 m.

[0052] Determine the water depth limit breakage threshold based on the effective water depth and the water depth limit breakage coefficient:

[0053] in, Indicates the water depth-limited breakage threshold, subscript Indicates the upper limit of wave height, This represents the water depth-limited breakage coefficient. In this embodiment, Take 0.73.

[0054] Determine the bottom friction indication based on the effective water depth and bottom friction coefficient:

[0055] in, Indicates the bottom friction level. Indicates the coefficient of friction at the bottom, subscript Indicates seabed friction, This indicates taking the absolute value.

[0056] Subsequently, the wind field data, water depth data, previous wave state, wind energy input potential, water depth limit breakage threshold, and bottom friction indication are combined to obtain multi-channel input data.

[0057] In one implementation, the multi-channel input data includes 18 input channels, namely 2 wind field channels, 2 PM reference wave parameter channels, 4 previous wave state channels, 3 physical derivative channels, 1 water depth channel, 2 geographic location channels, and 4 time-coded channels.

[0058] Among them, the two wind field channels correspond to the zonal wind speed component and the meridional wind speed component, respectively; the two PM reference wave parameter channels correspond to the PM reference significant wave height and the PM reference peak wave period, respectively; the four previous wave state channels correspond to the significant wave height, peak wave period, peak wave direction cosine component and peak wave direction sine component, respectively; the three physical derivative channels correspond to the wind energy input potential, water depth limit breakage threshold and bottom friction indicator, respectively; the two geographic location channels correspond to the normalized longitude and latitude, respectively; and the four time coding channels correspond to the sine coding and cosine coding of the time within the year and the time within the day, respectively.

[0059] It should be noted that the aforementioned 18 channels represent one data organization method. In other implementations, the encoding methods for time information, geographic location information, and peak wave direction can be adjusted according to the input format of the wave prediction network.

[0060] Wind energy input potential, water depth-limited fracturing threshold, and bottom friction indicator have a dual function: firstly, they serve as input channels in the multi-channel input data, providing the wave prediction network with information related to wind energy input, shallow water fracturing, and bottom friction dissipation; secondly, they are used to construct the physical reference wave state at the target time by combining the wave state at the previous moment. Thus, the same physical derivatives can simultaneously influence both the input process and the training constraint process of the wave prediction network.

[0061] In cases where the wave state of the previous moment is missing in the first moment, the PM reference wave state can be used as the initial wave state.

[0062] S103. Input the multi-channel input data into the wave prediction network to obtain the predicted wave state at the target time.

[0063] In this embodiment, the wave prediction network adopts a residual U-Net. The residual U-Net includes an encoder, a bottleneck layer, and a decoder.

[0064] Multi-channel input data can be divided into spatial data blocks according to the spatial grid of the target sea area. The encoding end performs downsampling of the spatial data blocks step by step to extract spatial correlation features between wind field, water depth, previous wave state, and physical derivatives at different spatial scales. The bottleneck layer is used to fuse the high-level spatial features extracted by the encoding end. The decoding end recovers the spatial scale of the feature map step by step, and the features of the corresponding spatial scale of the encoding end are fused by connecting them at the same scale to obtain the predicted wave state corresponding to the spatial grid of the target sea area.

[0065] The output of the wave prediction network can be represented as:

[0066] in, Indicates the target time The network output, Indicates the target time Predicted effective wave height, Indicates the target time The predicted peak wave period, Indicates the target time The predicted peak direction; the pointed cap symbol above the variable indicates that the corresponding parameter is a network prediction value, and the subscript... Indicates the target time.

[0067] Determining the predicted peak wave direction by using the cosine and sine components of the predicted peak wave direction can avoid numerical discontinuities in the peak wave direction near 0° and 360°. Therefore, the predicted significant wave height, predicted peak wave period, and predicted peak wave direction for each ocean grid in the target sea area can be obtained.

[0068] S104. Construct a physical reference wave state for the target time based on the wave state at the previous moment, the wind energy input potential, the water depth limit breakage threshold, and the bottom friction indicator. Determine a joint loss based on the predicted wave state, the sample wave state at the target time, and the physical reference wave state. Train the wave prediction network based on the joint loss.

[0069] The physical reference wave state includes the physical reference significant wave height, the physical reference peak wave period, and the physical reference peak wave direction. The following uses... Indicate the target time, with Indicates the time preceding the target time, indicated by a superscript. This indicates that the corresponding parameters are reference values ​​constructed based on physical relationships.

[0070] First, determine the wave energy at the previous moment based on the significant wave height in the wave state at the previous moment:

[0071] in, Indicates the significant wave height at the previous moment. It represents the wave energy characteristic quantity of the previous moment.

[0072] Determine the wave energy growth term per unit time based on the wind energy input potential:

[0073] in, Indicates the target time The wind energy input potential.

[0074] Determine the bottom friction dissipation term per unit time based on the bottom friction indicator and the wave energy at the previous moment:

[0075] Determine the breakage dissipation term per unit time based on the water depth-limited breakage threshold and the wave energy at the previous moment:

[0076] in, , and These represent the wave energy growth term, bottom friction dissipation term, and fragmentation dissipation term per unit time corresponding to the target time, respectively. , and These represent the preset wave energy growth coefficient, bottom friction dissipation coefficient, and fragmentation dissipation coefficient, respectively. This represents the corrected linear unit (ReLU) function. (Subscript) , and These represent wind energy input, bottom friction dissipation, and fragmentation dissipation, respectively.

[0077] The physical reference wave energy at the target time is determined based on the wave energy at the previous moment, the wave energy growth term per unit time, the bottom friction dissipation term per unit time, and the fragmentation dissipation term per unit time.

[0078] in, The physical reference wave energy at the target time. Indicates the time interval between adjacent moments. This represents a lower limit constant that is greater than zero.

[0079] Determining the physical reference effective wave height based on the physical reference wave energy:

[0080] in, This represents the physical reference effective wave height at the target time. Therefore, the wind energy input potential is used to determine the degree of wave energy growth, the bottom friction indicator is used to determine the degree of wave energy dissipation, and the depth-limited breakage threshold is used to determine the wave height range at which depth-limited breakage occurs.

[0081] Furthermore, the physical reference peak wave period is determined based on the peak wave period of the previous moment and the PM reference peak wave period:

[0082] in, This indicates the peak wave period at the previous moment. This represents the period of the PM reference peak wave, determined based on the wind speed at the target time. The period of the physical reference peak wave at the target time. This represents the preset period adjustment coefficient. This calculation causes the peak wave period to change under the influence of the wind field towards the PM reference peak wave period corresponding to the current wind speed.

[0083] Based on the wave steepness relationship, the physical reference significant wave height and the physical reference peak wave period are coupled and constrained:

[0084] in, Indicates the physical reference effective wave height. Indicates the period of the physical reference peak wave. Indicates the preset upper limit of wave steepness, subscript This indicates the maximum allowed value. In this embodiment, Take 0.142.

[0085] The variation of peak wave period with water depth is constrained based on the linear dispersion relation:

[0086] in, , This represents the angular frequency corresponding to the period of the physical reference peak wave. Indicates wave number, Indicates water depth. This represents the hyperbolic tangent function. Using the above dispersion relation, the influence of water depth on wave propagation can be considered when determining the period of the physical reference peak wave.

[0087] Determine the wind direction at the target time based on the zonal and meridional wind speed components:

[0088] in, and These represent the zonal and meridional wind speed components at a location 10 m above the sea surface at the target time, respectively. Indicates the wind direction at the target time, subscript Indicates wind direction. This represents the bivariate arctangent function that determines the azimuth angle based on two wind speed components.

[0089] Determine the initial physical reference peak wave direction based on the peak wave direction at the previous moment and the wind direction at the target moment:

[0090] in, Indicates the peak wave direction at the previous moment. Indicates the physical reference peak wave direction at the target time. This indicates the preset wave direction adjustment coefficient.

[0091] The initial physical reference peak wave direction is further corrected based on the propagation direction of the waves along the group velocity direction, and the water depth gradient is also considered. Refraction correction is applied to the propagation-corrected physical reference peak wave direction. Among other things, Represents the spatial gradient operator, The spatial gradient represents the water depth. The above refraction correction causes the physical reference peak wave direction in the nearshore area to deflect towards the isobath normal as the water depth changes, resulting in the final physical reference peak wave direction.

[0092] Thus, the significant wave height and peak wave period are coupled through the wave steepness relationship, the peak wave period and water depth are coupled through the dispersion relationship, and the peak wave direction is coupled with the wind direction, the wave propagation direction and the water depth gradient through the propagation and refraction relationship.

[0093] The joint loss includes data fitting loss and physical constraint loss:

[0094] in, Indicates joint loss, This represents the data fitting loss. Represents physical constraint loss. and These represent the weights of the data fitting loss and the physical constraint loss, respectively; subscripts. , and These represent joint, data fitting, and physical constraints, respectively.

[0095] The data fitting loss, used to characterize the difference between the predicted wave state and the sample wave state at the target time, can be expressed as:

[0096] in, This represents the fitting loss for the effective wave height. This represents the loss in peak-wave period fitting. This represents the peak-to-wavelength fitting loss. This indicates the peak wave loss to the unit circle; , , and These represent the weights of the four types of losses mentioned above.

[0097] The fitting loss for effective wave height and peak wave period can be calculated using smoothed L1 loss. The fitting loss for peak wave direction can be determined based on the cosine and sine components of the predicted peak wave direction and the sample peak wave direction, respectively.

[0098] The peak wave loss to the unit circle can be expressed as:

[0099] in, and These represent the predicted peak wave direction cosine and sine components, respectively, of the network output. Indicates the ocean grid area, This indicates that the values ​​corresponding to each grid within the ocean grid area are averaged.

[0100] The effective wave height evolution loss is determined based on the difference between the predicted effective wave height and the physical reference effective wave height:

[0101] in, This represents the significant wave height evolution loss. Similarly, the peak wave period evolution loss can be determined based on the difference between the predicted peak wave period and the physical reference peak wave period. The peak wave direction evolution loss is determined based on the cosine and sine components of the predicted peak wave direction and the physical reference peak wave direction, respectively. .

[0102] Physical constraint losses may also include at least one of the following: non-negative effective wave height loss, minimum peak wave period loss, water depth constraint breakage loss, wave steepness loss, wind speed saturation loss, low wind speed constraint loss, wave energy propagation loss, and spatial smoothing loss.

[0103] Among them, the non-negative significant wave height loss is used to penalize predicted significant wave heights that are less than zero; the minimum peak wave period loss is used to penalize predicted peak wave periods that are less than the preset lower limit of the period; the water depth limit breakage loss is used to penalize predicted significant wave heights that exceed the water depth limit breakage threshold; and the wave steepness loss is used to penalize predicted significant wave heights and predicted peak wave periods that do not meet the preset wave steepness relationship.

[0104] The depth-limited breakage loss can be expressed as:

[0105] in, This indicates the breakage loss due to water depth limitations.

[0106] Wind speed saturation loss is used to constrain the relationship between the predicted wave state and the fully developed wave state corresponding to the current wind field; low wind speed constraint loss is used to limit the occurrence of large predicted effective wave heights that do not match the wind field under low wind speed conditions; wave energy propagation loss is used to constrain the wave energy changes at adjacent time points to match the wave propagation process along the group velocity direction; spatial smoothing loss is used to constrain the variation amplitude of the predicted wave state between adjacent ocean grids to reduce local anomalous abrupt changes in spatial distribution.

[0107] The physical constraint loss can be expressed as:

[0108]

[0109]

[0110]

[0111] in, This indicates the loss in the evolution of wave state. This indicates the loss related to propagation and smoothing. Represents the loss due to physical boundaries and state constraints; Indicates wave energy propagation loss, Indicates spatial smoothing loss, This indicates a non-negative loss due to significant wave height. This represents the minimum loss of the peak wave period. This indicates that water depth limits breakage losses. Indicates wave steepness loss. Indicates wind speed saturation loss. This indicates the loss due to low wind speed constraint.

[0112] The various losses in the joint loss are calculated in conjunction with the land-sea mask, and only the loss corresponding to the ocean grid is used to update the parameters of the wave prediction network.

[0113] For example, wave state data can be divided into training, validation, and test sets at ratios of 70%, 15%, and 15%, respectively. The parameters of the wave prediction network are then updated using an Adaptive Moment Estimation (Adam) optimizer, with the initial learning rate set to... The learning rate is reduced using cosine annealing. Training is stopped and the network parameters are saved when the joint loss on the validation set no longer decreases within a preset number of training epochs.

[0114] The training process includes a teacher-mandated training phase and a recursive training phase. In the teacher-mandated training phase, multi-channel input data is constructed using the wave state samples from the previous moment; in the recursive training phase, multi-channel input data is constructed using the predicted wave state from the previous moment output by the wave prediction network.

[0115] Specifically, a planned sampling method can be used to set the predicted state usage ratio. Predicted state usage ratio Indicates the first The proportion of multi-channel input data constructed using the predicted wave state from the previous moment in each training round; subscript This indicates the number of training rounds. The proportion of predicted states used is gradually increased as the number of training rounds increases. This allows the training process to gradually transition from teacher-mandated training to recursive training.

[0116] The weight of the physical constraint loss increases as the proportion of predicted states used increases:

[0117] in, Indicates the first The physical constraint loss weights used in each training round The initial weights represent the physical constraint loss. This represents the final weight of the physical constraint loss. As the proportion of the model's own predicted state in the multi-channel input data increases, the weight of the physical constraint loss is increased accordingly to strengthen the constraint on the recursive predicted state.

[0118] In some embodiments, the degree of violation of each physical constraint can also be statistically analyzed during continuous recursive training. The degree of violation may include the extent to which the predicted effective wave height exceeds the water depth limit breaking threshold, the extent to which the wave energy change between adjacent time moments does not conform to a preset wave energy relationship, and the extent to which the predicted wave state exceeds a preset wave steepness upper limit. The weight of the corresponding physical loss term is increased according to each degree of violation to adjust the impact of different physical constraints on network parameter updates.

[0119] S105. Using the trained wave prediction network, output the predicted wave state at the time to be predicted based on the wind field data, water depth data, and wave state at the previous time.

[0120] Obtain wind field data, water depth data, and wave state at the time to be predicted. Determine the wind energy input potential based on the wind field data at the time to be predicted. Determine the water depth limit breakage threshold and bottom friction indicator based on the water depth data. Construct multi-channel input data for prediction in the manner described in step S102.

[0121] The predicted wave height, predicted peak wave period, and predicted peak wave direction at the time to be predicted are obtained by inputting multi-channel input data into the trained wave prediction network.

[0122] When predicting ocean waves at multiple consecutive moments, the predicted ocean wave state at one moment is used as the ocean wave state at the previous moment of the next moment to be predicted. The wind field data and water depth data at the next moment to be predicted are combined to reconstruct the multi-channel input data for prediction, thereby continuously outputting the spatial distribution of ocean wave parameters in the target sea area at preset time intervals.

[0123] For example, hourly ERA5 wind field data, along with corresponding water depth data, geographic location information, and time information, can be input. For the first prediction time, the PM reference wave state is used as the initial wave state; for subsequent prediction times, the effective wave height, peak wave period, and peak wave direction predicted at the previous time are used as the input for the current time, and recursive predictions are performed at 1-hour time intervals.

[0124] At each prediction time, the spatial distribution of significant wave height, peak wave period, and peak wave direction for all ocean grids in the target sea area can be output through a forward inference of the trained wave prediction network. This prediction process obtains wave parameters through network computation after obtaining the trained network parameters, without the need for iterative operations in the frequency-direction spectrum space, thereby reducing the computational load required to output the spatial distribution of wave parameters in the target sea area hourly.

[0125] The deep learning-based wave prediction method provided in this application determines physical derivatives related to wave generation, breaking, and dissipation processes based on wind field and water depth data. These physical derivatives, along with wind field, water depth, and the previous wave state, are used to construct multi-channel input data. Furthermore, a physical reference wave state for the target time is constructed based on these physical derivatives and the previous wave state. Thus, the same physical information can simultaneously influence both the input and training constraints of the wave prediction network. This allows the network to learn wave variation relationships from sample data while being constrained by wave evolution laws, thereby reducing predictions that do not conform to wave generation, breaking, and dissipation patterns and improving the stability and physical plausibility of wave prediction results.

[0126] Those skilled in the art will understand that, in the above-described method of the specific implementation, the order in which each step is written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0127] Based on the same inventive concept, this application also provides a deep learning-based wave prediction device corresponding to the deep learning-based wave prediction method. Since the principle of the device in this application is similar to the deep learning-based wave prediction method described above in this application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0128] Reference Figure 2 As shown, Figure 2 A schematic diagram of a deep learning-based wave prediction device provided in this application embodiment, the device comprising: The acquisition module 210 is used to acquire wind field data, water depth data, wave state at the previous moment and sample wave state at the target moment in the target sea area. The wave state includes significant wave height, peak wave period and peak wave direction. The determination module 220 is used to determine the wind energy input potential based on the wind field data, determine the water depth limit breakage threshold and bottom friction indication based on the water depth data, and construct multi-channel input data based on the wind field data, the water depth data, the previous wave state, the wind energy input potential, the water depth limit breakage threshold and the bottom friction indication. The first prediction module 230 is used to input the multi-channel input data into the wave prediction network to obtain the predicted wave state at the target time. Training module 240 is used to construct a physical reference wave state at the target time based on the wave state at the previous moment, the wind energy input potential, the water depth limit breakage threshold and the bottom friction indicator, and to determine a joint loss based on the predicted wave state, the sample wave state at the target time and the physical reference wave state, and to train the wave prediction network based on the joint loss. The second prediction module 250 is used to utilize the trained wave prediction network to output the predicted wave state at the time to be predicted, based on the wind field data, water depth data, and wave state at the previous time.

[0129] In one possible implementation, the determining module 220 is used to: Obtain the geographical location and time information of the target sea area; The wind field data, water depth data, and wave state are correlated temporally and spatially, missing and invalid values ​​are processed, and land grids are identified based on the water depth data to establish a land-sea mask. The wind field data, water depth data, and wave state are normalized. Based on the wind field data, the Pearson-Moscowitz reference wave parameters are determined, and the geographic location code and time code are determined based on the geographic location and time information, respectively. The multi-channel input data is constructed based on the processed data, the Pearson-Moscowitz reference wave parameters, the geographic location encoding, and the time encoding.

[0130] In one possible implementation, the determining module 220 is used to: The wind friction speed is determined based on the wind speed and drag coefficient at 10m above the sea surface, and the wind energy input potential is determined based on the wind friction speed, air density, and seawater density. The water depth data is subjected to amplitude limiting processing to obtain the effective water depth; The water depth limit breakage threshold is determined based on the effective water depth and the water depth limit breakage coefficient, and the bottom friction indication is determined based on the effective water depth and the bottom friction coefficient.

[0131] In one possible implementation, the training module 240 is used for: The wave energy at the previous moment is determined based on the effective wave height in the wave state at the previous moment; The wave energy growth term is determined based on the wind energy input potential, the bottom friction dissipation term is determined based on the bottom friction indicator and the wave energy at the previous moment, and the breakage dissipation term is determined based on the water depth limit breakage threshold and the wave energy at the previous moment. The physical reference wave energy is determined based on the wave energy at the previous moment, the wave energy growth term, the bottom friction dissipation term, and the fragmentation dissipation term. The effective wave height of the physical reference can be determined based on the physical reference wave energy.

[0132] In one possible implementation, the training module 240 is further configured to: The physical reference peak wave period is determined based on the peak wave period of the previous moment and the reference peak wave period corresponding to the wind field data. The physical reference peak wave direction is determined based on the peak wave direction of the previous moment and the wind direction corresponding to the wind field data. The physical reference wave state is obtained by coupling constraints on the physical reference effective wave height and the physical reference peak wave period according to the wave steepness relation, coupling constraints on the physical reference peak wave period and the water depth data according to the dispersion relation, and coupling constraints on the physical reference peak wave direction, the wave propagation direction along the group velocity direction and the water depth gradient of the water depth data according to the refraction relation.

[0133] In one possible implementation, the first prediction module 230 is used to: The residual U-Net network is used to perform multi-scale feature extraction and feature fusion on the multi-channel input data to obtain the sine and cosine components of the predicted significant wave height, predicted peak wave period, and predicted peak wave direction. The predicted peak wave direction is determined based on the sine and cosine components.

[0134] In one possible implementation, the training module 240 is used for: The data fitting loss is determined based on the difference between the predicted wave state and the wave state of the sample at the target time. The wave state evolution loss is determined based on the difference between the predicted wave state and the physical reference wave state. Based on the predicted wave state, determine at least one of the following: depth-limited breakage loss, wave steepness loss, wave energy propagation loss, and spatial smoothing loss; Based on the land-sea mask, the data fitting loss, the wave state evolution loss, and at least one determined loss are weighted within the ocean grid to obtain the joint loss.

[0135] In one possible implementation, the training module 240 is used for: The multi-channel input data is constructed using the sample wave state of the previous moment or the predicted wave state of the previous moment output by the wave prediction network. As the number of training rounds increases, the proportion of multi-channel input data constructed using the predicted wave state from the previous moment is increased; According to the stated ratio, the weight of the wave state evolution loss and at least one of the depth-limited breaking loss, wave steepness loss, wave energy propagation loss and spatial smoothing loss is increased in the joint loss.

[0136] The processing flow of each module in the device and the interaction flow between each module can be referred to the relevant descriptions in the above method embodiments, and will not be detailed here.

[0137] This application also provides a computer device, such as... Figure 3 As shown, Figure 3 This is a schematic diagram of a computer device structure provided in an embodiment of this application. The computer device includes: Processor 31 and memory 32; the memory 32 stores machine-readable instructions executable by the processor 31, and the processor 31 executes the machine-readable instructions stored in the memory 32. When the machine-readable instructions are executed by the processor 31, the processor 31 performs the following steps: Acquire wind field data, water depth data, wave state at the previous moment, and sample wave state at the target moment for the target sea area. The wave state includes significant wave height, peak wave period, and peak wave direction. The wind energy input potential is determined based on the wind field data, the water depth limit breakage threshold and bottom friction indication are determined based on the water depth data, and multi-channel input data is constructed based on the wind field data, the water depth data, the previous wave state, the wind energy input potential, the water depth limit breakage threshold and the bottom friction indication. The multi-channel input data is input into the wave prediction network to obtain the predicted wave state at the target time; The physical reference wave state at the target time is constructed based on the wave state at the previous moment, the wind energy input potential, the water depth limit breakage threshold, and the bottom friction indicator. A joint loss is determined based on the predicted wave state, the sample wave state at the target time, and the physical reference wave state. The wave prediction network is then trained based on the joint loss. Using the trained wave prediction network, the predicted wave state for the time to be predicted is output based on the wind field data, water depth data, and wave state at the time to be predicted and the previous time.

[0138] In one possible implementation, constructing the multi-channel input data includes: Obtain the geographical location and time information of the target sea area; The wind field data, water depth data, and wave state are correlated temporally and spatially, missing and invalid values ​​are processed, and land grids are identified based on the water depth data to establish a land-sea mask. The wind field data, water depth data, and wave state are normalized. Based on the wind field data, the Pearson-Moscowitz reference wave parameters are determined, and the geographic location code and time code are determined based on the geographic location and time information, respectively. The multi-channel input data is constructed based on the processed data, the Pearson-Moscowitz reference wave parameters, the geographic location encoding, and the time encoding.

[0139] In one possible implementation, determining the wind energy input potential, the water depth-limited breakage threshold, and the bottom friction indicator includes: The wind friction speed is determined based on the wind speed and drag coefficient at 10m above the sea surface, and the wind energy input potential is determined based on the wind friction speed, air density, and seawater density. The water depth data is subjected to amplitude limiting processing to obtain the effective water depth; The water depth limit breakage threshold is determined based on the effective water depth and the water depth limit breakage coefficient, and the bottom friction indication is determined based on the effective water depth and the bottom friction coefficient.

[0140] In one possible implementation, constructing the physical reference wave state includes: The wave energy at the previous moment is determined based on the effective wave height in the wave state at the previous moment; The wave energy growth term is determined based on the wind energy input potential, the bottom friction dissipation term is determined based on the bottom friction indicator and the wave energy at the previous moment, and the breakage dissipation term is determined based on the water depth limit breakage threshold and the wave energy at the previous moment. The physical reference wave energy is determined based on the wave energy at the previous moment, the wave energy growth term, the bottom friction dissipation term, and the fragmentation dissipation term. The effective wave height of the physical reference can be determined based on the physical reference wave energy.

[0141] In one possible implementation, constructing the physical reference wave state further includes: The physical reference peak wave period is determined based on the peak wave period of the previous moment and the reference peak wave period corresponding to the wind field data. The physical reference peak wave direction is determined based on the peak wave direction of the previous moment and the wind direction corresponding to the wind field data. The physical reference wave state is obtained by coupling constraints on the physical reference effective wave height and the physical reference peak wave period according to the wave steepness relation, coupling constraints on the physical reference peak wave period and the water depth data according to the dispersion relation, and coupling constraints on the physical reference peak wave direction, the wave propagation direction along the group velocity direction and the water depth gradient of the water depth data according to the refraction relation.

[0142] In one possible implementation, the multi-channel input data is input into a wave prediction network to obtain the predicted wave state, including: The residual U-Net network is used to perform multi-scale feature extraction and feature fusion on the multi-channel input data to obtain the sine and cosine components of the predicted significant wave height, predicted peak wave period, and predicted peak wave direction. The predicted peak wave direction is determined based on the sine and cosine components.

[0143] In one possible implementation, determining the joint loss includes: The data fitting loss is determined based on the difference between the predicted wave state and the wave state of the sample at the target time. The wave state evolution loss is determined based on the difference between the predicted wave state and the physical reference wave state. Based on the predicted wave state, determine at least one of the following: depth-limited breakage loss, wave steepness loss, wave energy propagation loss, and spatial smoothing loss; Based on the land-sea mask, the data fitting loss, the wave state evolution loss, and at least one determined loss are weighted within the ocean grid to obtain the joint loss.

[0144] In one possible implementation, training the wave prediction network includes: The multi-channel input data is constructed using the sample wave state of the previous moment or the predicted wave state of the previous moment output by the wave prediction network. As the number of training rounds increases, the proportion of multi-channel input data constructed using the predicted wave state from the previous moment is increased; According to the stated ratio, the weight of the wave state evolution loss and at least one of the depth-limited breaking loss, wave steepness loss, wave energy propagation loss and spatial smoothing loss is increased in the joint loss.

[0145] The aforementioned memory 32 includes a main memory 321 and an external memory 322; the main memory 321, also known as internal memory, is used to temporarily store the computational data in the processor 31, as well as the data exchanged with external memory 322 such as a hard disk. The processor 31 exchanges data with the external memory 322 through the main memory 321.

[0146] The specific execution process of the above instructions can be referred to the steps of the deep learning-based wave prediction method described in the embodiments of this application, and will not be repeated here.

[0147] This application also provides a computer-readable storage medium storing a computer program. When a processor runs the computer program, it executes the steps of the deep learning-based wave prediction method described in the above method embodiments. The storage medium can be either volatile or non-volatile.

[0148] This application also provides a computer program product carrying program code. The program code includes instructions that can be used to execute the steps of the deep learning-based wave prediction method described in the above method embodiments. For details, please refer to the above method embodiments, which will not be repeated here.

[0149] The aforementioned computer program product can be implemented through hardware, software, or a combination thereof. In one optional embodiment, the computer program product is specifically embodied in a computer storage medium; in another optional embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.

[0150] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division; in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection may be through some communication interfaces; the indirect coupling or communication connection of devices or units may be electrical, mechanical, or other forms.

[0151] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0152] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0153] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0154] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The scope of protection of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of protection of the claims.

Claims

1. A deep learning-based wave prediction method, characterized in that, include: Acquire wind field data, water depth data, wave state at the previous moment, and sample wave state at the target moment for the target sea area. The wave state includes significant wave height, peak wave period, and peak wave direction. The wind energy input potential is determined based on the wind field data, the water depth limit breakage threshold and bottom friction indication are determined based on the water depth data, and multi-channel input data is constructed based on the wind field data, the water depth data, the previous wave state, the wind energy input potential, the water depth limit breakage threshold and the bottom friction indication. The multi-channel input data is input into the wave prediction network to obtain the predicted wave state at the target time; The physical reference wave state at the target time is constructed based on the wave state at the previous moment, the wind energy input potential, the water depth limit breakage threshold, and the bottom friction indicator. A joint loss is determined based on the predicted wave state, the sample wave state at the target time, and the physical reference wave state. The wave prediction network is then trained based on the joint loss. Using the trained wave prediction network, the predicted wave state for the time to be predicted is output based on the wind field data, water depth data, and wave state at the time to be predicted and the previous time.

2. The method according to claim 1, characterized in that, Constructing the multi-channel input data includes: Obtain the geographical location and time information of the target sea area; The wind field data, water depth data, and wave state are correlated temporally and spatially, missing and invalid values ​​are processed, and land grids are identified based on the water depth data to establish a land-sea mask. The wind field data, water depth data, and wave state are normalized. Based on the wind field data, the Pearson-Moscowitz reference wave parameters are determined, and the geographic location code and time code are determined based on the geographic location and time information, respectively. The multi-channel input data is constructed based on the processed data, the Pearson-Moscowitz reference wave parameters, the geographic location encoding, and the time encoding.

3. The method according to claim 1, characterized in that, Determining the wind energy input potential, the water depth-limited breakage threshold, and the bottom friction indicator includes: The wind friction speed is determined based on the wind speed and drag coefficient at 10m above the sea surface, and the wind energy input potential is determined based on the wind friction speed, air density, and seawater density. The water depth data is subjected to amplitude limiting processing to obtain the effective water depth; The water depth limit breakage threshold is determined based on the effective water depth and the water depth limit breakage coefficient, and the bottom friction indication is determined based on the effective water depth and the bottom friction coefficient.

4. The method according to claim 3, characterized in that, Constructing the physical reference wave state includes: The wave energy at the previous moment is determined based on the effective wave height in the wave state at the previous moment; The wave energy growth term is determined based on the wind energy input potential, the bottom friction dissipation term is determined based on the bottom friction indicator and the wave energy at the previous moment, and the breakage dissipation term is determined based on the water depth limit breakage threshold and the wave energy at the previous moment. The physical reference wave energy is determined based on the wave energy at the previous moment, the wave energy growth term, the bottom friction dissipation term, and the fragmentation dissipation term. The effective wave height of the physical reference can be determined based on the physical reference wave energy.

5. The method according to claim 4, characterized in that, Constructing the physical reference wave state also includes: The physical reference peak wave period is determined based on the peak wave period of the previous moment and the reference peak wave period corresponding to the wind field data. The physical reference peak wave direction is determined based on the peak wave direction of the previous moment and the wind direction corresponding to the wind field data. The physical reference wave state is obtained by coupling constraints on the physical reference effective wave height and the physical reference peak wave period according to the wave steepness relation, coupling constraints on the physical reference peak wave period and the water depth data according to the dispersion relation, and coupling constraints on the physical reference peak wave direction, the wave propagation direction along the group velocity direction and the water depth gradient of the water depth data according to the refraction relation.

6. The method according to claim 1, characterized in that, The multi-channel input data is input into the wave prediction network to obtain the predicted wave state, including: The residual U-Net network is used to perform multi-scale feature extraction and feature fusion on the multi-channel input data to obtain the sine and cosine components of the predicted significant wave height, predicted peak wave period, and predicted peak wave direction. The predicted peak wave direction is determined based on the sine and cosine components.

7. The method according to claim 2, characterized in that, Determining the joint loss includes: The data fitting loss is determined based on the difference between the predicted wave state and the wave state of the sample at the target time. The wave state evolution loss is determined based on the difference between the predicted wave state and the physical reference wave state. Based on the predicted wave state, determine at least one of the following: depth-limited breakage loss, wave steepness loss, wave energy propagation loss, and spatial smoothing loss; Based on the land-sea mask, the data fitting loss, the wave state evolution loss, and at least one determined loss are weighted within the ocean grid to obtain the joint loss; Training the wave prediction network includes: The multi-channel input data is constructed using the sample wave state of the previous moment or the predicted wave state of the previous moment output by the wave prediction network. As the number of training rounds increases, the proportion of multi-channel input data constructed using the predicted wave state from the previous moment is increased; According to the stated ratio, the weight of the wave state evolution loss and at least one of the depth-limited breaking loss, wave steepness loss, wave energy propagation loss and spatial smoothing loss is increased in the joint loss.

8. A wave prediction device based on deep learning, characterized in that, The device includes: The acquisition module is used to acquire wind field data, water depth data, wave state at the previous moment and sample wave state at the target moment for the target sea area. The wave state includes significant wave height, peak wave period and peak wave direction. The determination module is used to determine the wind energy input potential based on the wind field data, determine the water depth limit breakage threshold and bottom friction indication based on the water depth data, and construct multi-channel input data based on the wind field data, the water depth data, the previous wave state, the wind energy input potential, the water depth limit breakage threshold and the bottom friction indication. The first prediction module is used to input the multi-channel input data into the wave prediction network to obtain the predicted wave state at the target time. The training module is used to construct the physical reference wave state at the target time based on the wave state at the previous moment, the wind energy input potential, the water depth limit breakage threshold, and the bottom friction indicator, and to determine the joint loss based on the predicted wave state, the sample wave state at the target time, and the physical reference wave state, and to train the wave prediction network based on the joint loss. The second prediction module is used to utilize the trained wave prediction network to output the predicted wave state at the time to be predicted, based on the wind field data, water depth data, and wave state at the previous time.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images