Sea wave spectrum intelligent correction method and system based on neural network

Abstract

The invention belongs to the technical field of marine environment prediction, and discloses a neural network-based sea wave spectrum intelligent correction method and system. The method comprises the following steps: setting a target sea area and acquiring environmental data; performing numerical calculation on the sea wave spectrum in the target sea area through a sea wave numerical mode; a sea wave spectrum correction model comprising a numerical calculation module, an actual measurement processing module and an intelligent correction module is constructed through a deep learning method; and carrying out precision verification and adaptability evaluation on the constructed sea wave spectrum correction model. According to the model, the problem of errors caused by physical process simplification in the simulation process of a traditional method is solved. Meanwhile, the problem that an existing data assimilation method is limited by observation data is solved. According to the method, the simulation precision of the sea wave mode can be effectively improved, and the method has good sea area adaptability.

Description

Technical Field

[0001] This invention belongs to the field of marine environment prediction technology, and in particular relates to a method and system for intelligent correction of ocean wave spectrum based on neural networks. Background Technology

[0002] Wave spectra are crucial tools for describing ocean wave characteristics, reflecting the energy distribution of waves across different frequencies and wave numbers. They also reveal wave components, including the generation and development of local wind waves, the propagation of open-ocean swells, and the interactions between multiple wave systems. Furthermore, information such as the location of energy peaks and spectral width within the wave spectrum reveals typical wave characteristics, providing a foundation for predicting wave height and period. Therefore, accurate descriptions of wave spectra offer more precise results for characterizing ocean waves, further supporting research on marine climate change and maritime navigation safety.

[0003] Current methods for correcting ocean wave data include traditional data assimilation and optimization using machine learning. Traditional data assimilation methods correct numerical simulation results by incorporating measured ocean wave data, thereby reducing simulation bias and improving forecast accuracy to some extent. Representative methods include variational assimilation, Kalman filtering, and optimal interpolation. Among them, variational assimilation constructs a cost function and minimizes the difference between observations and the model to achieve optimal estimation of the model state; Kalman filtering, based on a state-space model and dynamic updates of error covariance, can gradually improve the reliability of simulation results through iterative fusion of forecasts and observations; optimal interpolation, based on statistical principles, linearly combines observation and model results through weight allocation to achieve spatial and temporal consistency correction. These methods have been widely used in large-scale ocean models and operational ocean wave forecasting, but they have high computational complexity and are somewhat dependent on the spatiotemporal coverage of observation data.

[0004] In recent years, with the development of artificial intelligence, machine learning methods have been gradually introduced into the optimization and correction of marine environmental information to compensate for the shortcomings of traditional assimilation methods in terms of computational efficiency and handling of complex nonlinear relationships. In wave simulation and forecasting, machine learning methods typically construct neural networks or other nonlinear regression models to fit and learn the deviations between numerical simulation results and buoy observation data. This process can effectively capture the systematic errors of numerical models in specific regions or scenarios and correct the results in real time in new forecasts. Compared with traditional methods, machine learning methods have stronger generalization ability and higher computational efficiency, and can improve the accuracy of model simulation and forecasting of wave elements within the sea area under limited observation conditions. Therefore, machine learning-based correction is gradually becoming an important supplementary approach to improving the accuracy of wave models.

[0005] Based on the above analysis, the problems and shortcomings of the existing technology are as follows: Traditional data assimilation methods are computationally complex and rely on ocean observation information. When unavailable observational data for assimilation, especially in remote sea areas or under extreme weather conditions, the applicability of traditional assimilation methods is challenged. Therefore, there is a problem of difficulty in correcting ocean environmental data when observational data is scarce.

[0006] Machine learning-based wave data correction methods currently focus primarily on global wave parameters, such as significant wave height and wave period. These parameters can reflect the overall characteristics of waves to some extent, but they cannot characterize the spatial distribution of wave energy in terms of frequency and direction. Currently, no systematic correction method for wave spectra has been explicitly proposed. Summary of the Invention

[0007] To overcome the problems existing in related technologies, the present invention discloses an intelligent wave spectrum correction method and system based on neural networks.

[0008] The technical solution is as follows: A wave spectrum intelligent correction method based on neural networks, comprising the following steps: S1, Target sea area setting and environmental data acquisition; S2, through the numerical wave model, performs numerical calculations on the wave spectrum within the target sea area; S3 constructs a wave spectrum correction model using deep learning methods, including a numerical calculation module, a field measurement processing module, and an intelligent correction module. For the wave spectrum calculated by the wave numerical model, the numerical calculation module constructs numerical simulation wave spectrum data, and the field measurement processing module constructs buoy observation wave spectrum data. Using the buoy observation wave spectrum data as a benchmark, the intelligent correction module constructs the deviation between the numerical simulation wave spectrum data and the buoy observation wave spectrum data, and outputs the corrected wave spectrum. S4 is used to verify the accuracy and evaluate the adaptability of the constructed wave spectrum correction model.

[0009] In step S1, the target sea area setting and environmental data acquisition include: S101, set the target sea area according to requirements; S102, acquire environmental data within the sea area, including two parts: buoy data and open-source environmental field data within the target sea area; buoy data includes buoy location, observation time range, wave spectrum data, and wave height; open-source environmental field data includes wind speed field and water depth information within the target sea area.

[0010] In step S2, the wave spectrum of the target sea area is numerically calculated using a wave numerical model, including: S201. Based on the research sea area, the model computation domain is selected, and the model computation boundary field is set with full consideration of the wave transmission characteristics to achieve high-precision calculation of the waves. During the calculation process, a large computation domain is set based on the target sea area, and nested calculations are performed on the target sea area. The large computation domain is set to be a computation domain that expands by 20° in the four directions of east, west, south, and north of the target sea area, thereby providing boundary field information for the calculation of the target sea area by the model. S202: Using the acquired environmental data such as wind speed, a model is generated to calculate the forced field; based on the wind speed information in the east-west and north-south directions 10 meters from the sea surface in the sea area, the forced field in the sea area is calculated, and the wind field data is interpolated according to the model calculation grid to obtain forced field data that is consistent with the time and spatial dimension step size set in the calculation process. S203 sets the numerical discretization method in the calculation process of the wave numerical model, including spatial discretization of frequency and wave direction and long-term discretization of global time step; it completes the numerical solution of the wave spectrum in the sea area and outputs the wave height at the target location.

[0011] In step S3, the numerical calculation module processes the spectral frequency of the wave spectrum simulated by the wave numerical model using linear interpolation to construct numerically simulated wave spectrum data that is consistent with the spectral frequency of the wave spectrum observed by the buoy. The linear interpolation calculation method is as follows: ; In the formula, For the soundtrack of the ocean waves, For buoy observation frequency, For the simulated spectral frequencies of the wave mode, The wave spectrum at the frequency observed by the buoy. The simulated spectrum of ocean wave patterns at a certain frequency. The simulated spectrum of the wave mode at adjacent spectral frequencies; After frequency unification, the numerical simulation wave spectrum data were obtained. .

[0012] Furthermore, to address the issues of missing measurements and data loss in buoy data, the time series of wave spectra was processed using measured data as a reference.

[0013] First, a numerical simulation timeline was created based on the time steps of the wave model numerical calculations. Second, a buoy observation timeline was created in the same format. Due to data gaps caused by weather and equipment limitations, the numerical simulation data was deleted according to the missing time series and the timeline.

[0014] In step S3, the measurement processing module smooths the abnormal fluctuation positions in the original observation data of the buoy-observed wave spectrum using a Gaussian filtering method; and retains the dominant frequency and energy distribution characteristics of the spectrum using a peak recovery method. The specific calculation method is shown below: ; ; ; ; In the formula, For Gaussian kernel, This is the result after Gaussian filtering. This is the result after peak recovery. For the calculation process of the numerical function, For At the index position, with The original observation results for the sliding window offset; for The Gaussian kernel calculation results at the location, The peak value of the function. The peak value of the function after Gaussian filtering. For peak index, The peak value in the original signal. The standard deviation of the Gaussian kernel. As the independent variable, Let the radius be the Gaussian kernel. This is the offset of the sliding window. The weighting coefficients for peak recovery. The output signal after Gaussian filtering. For index position, The original signal, This is the final output result after peak recovery; Filtered and processed wave spectrum data from actual measurements, resulting in smooth and high-fidelity data. .

[0015] Furthermore, data gaps due to extreme weather or equipment malfunctions are removed. This removal is done by directly deleting missing data points along the observation timeline.

[0016] In step S3, the intelligent correction module uses the numerical simulation wave spectrum data constructed by the numerical calculation module. As input, buoy-observed wave spectrum data were constructed using the measured processing module. For output; and The data dimensions are all 44×6324; 44 represents the number of frequencies and 6324 represents the number of time points; the input and output layers each contain 44 neurons, representing the wave spectrum results corresponding to 44 frequencies at a single time point; A three-layer fully connected hidden layer is used, with 200, 200, and 100 neurons respectively. All hidden layers use the ReLU activation function to enhance non-linear expression. The output layer uses linear activation. During training, the input data is first batch-processed, then read and fed into the network time-by-time for forward propagation, outputting the corrected wave spectrum. The mean squared error (MSE) is used as the loss function to measure the difference between the output and the observed spectrum. The Adam optimizer is used to adaptively adjust the learning rate during the optimization process, ultimately achieving the correction of the wave spectrum.

[0017] Furthermore, the specific calculation method for the loss function is as follows: ; In the formula, For loss function, For the sample size, These are the actual value and the model prediction, respectively.

[0018] Another objective of this invention is to provide a neural network-based intelligent wave spectrum correction system, which implements the neural network-based intelligent wave spectrum correction method. The system includes: The data acquisition module is used for setting the target sea area and acquiring environmental data; The wave spectrum numerical calculation module is used to perform numerical calculations on the wave spectrum within the target sea area using a wave numerical model. The wave spectrum correction model construction module is used to construct a wave spectrum correction model, including a numerical calculation module, a field measurement processing module, and an intelligent correction module, using deep learning methods. For the wave spectrum calculated by the wave numerical model, the numerical calculation module is used to construct numerically simulated wave spectrum data, and the field measurement processing module is used to construct buoy-observed wave spectrum data. Using the buoy-observed wave spectrum data as a benchmark, the intelligent correction module is used to construct the deviation between the numerically simulated wave spectrum data and the buoy-observed wave spectrum data, and outputs the corrected wave spectrum. The model validation and evaluation module is used to validate the accuracy and evaluate the adaptability of the constructed wave spectrum correction model.

[0019] Combining all the above technical solutions, the beneficial effects of this invention are as follows: First, this invention proposes a neural network-based intelligent wave spectrum correction method and system. The model uses buoy observation data as a benchmark to construct a model that corrects the deviation between the simulated wave spectrum and the actual wave spectrum, thus achieving effective correction of the wave spectrum. This model solves the error problem caused by the simplification of physical processes in traditional methods during simulation. It also overcomes the limitation of existing data assimilation methods by the availability of observational data. This invention can effectively improve the simulation accuracy of wave models and has good adaptability to different sea areas. This invention provides a novel and effective method and strategy for the numerical calculation of wave spectra in real sea areas.

[0020] Secondly, this invention can optimize the frequency-space results of ocean waves, more accurately describe the distribution of ocean wave energy, and provide more accurate data input for research on ocean climate change and maritime navigation safety. This invention can effectively correct the ocean wave spectrum in numerical simulations using only a small amount of measured data. It solves the error problem caused by the simplification of physical processes in traditional methods. Furthermore, once the model is built, subsequent correction of the ocean wave spectrum no longer needs to rely on a large amount of real-world ocean wave spectrum data. Compared to existing data assimilation methods, it overcomes the limitation of observational data. Moreover, compared to existing technologies that correct single ocean wave parameters such as wave height and wave period, this method can optimize the distribution characteristics of ocean waves in the frequency space, more comprehensively reflecting ocean wave characteristics.

[0021] Third, traditional wave models describe wave generation and dissipation processes using physical expressions, which simplifies the process compared to reality. This invention employs a neural network and utilizes limited real-world sea observation data to correct numerical simulation results, addressing the discrepancy between simulated and real wave spectra. Furthermore, existing wave data correction methods only address global parameters such as wave height and period, lacking comprehensive correction of the wave spectrum. This invention aims to optimize the wave spectrum and comprehensively evaluates the model correction effect using two dimensions: spectral shape and spectral parameters (spectral distance and peak). It employs four indicators—wave spectrum, wave height, peak frequency, and peak value—to optimize the frequency space results of the wave spectrum and more accurately describe the distribution of wave energy. This invention solves the problem of simplification in the calculation of wave generation, dissipation, and energy transfer processes in traditional wave numerical simulation methods, effectively correcting numerical simulation results using limited real-world sea observation data. Attached Figure Description

[0022] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure; Figure 1 This is a flowchart of the intelligent wave spectrum correction method based on neural networks provided in this embodiment of the invention; Figure 2 This is a schematic diagram of the intelligent wave spectrum correction method based on neural networks provided in this embodiment of the invention; Figure 3 This is a comparison chart of the interpolation results of the WW3 wave spectrum before and after time 1; Figure 4 This is a comparison chart of the interpolation results of the WW3 wave spectrum at time 2 before and after the interpolation. Figure 5 Comparison of WW3 wave spectrum filtering results before and after time 1; Figure 6 Comparison of wave spectrum filtering results before and after WW3 at time 2; Figure 7 Comparison of wave spectrum before and after correction of the wave spectrum correction model at time 1; Figure 8 Comparison of wave spectrum before and after correction of the wave spectrum correction model at time 2; Figure 9 Comparison of wave spectrum before and after correction of the wave spectrum correction model at time 3; Figure 10 Comparison of wave spectrum before and after correction of the wave spectrum correction model at time 4; Figure 11 A comparison of wave spectrum accuracy before and after wave spectrum correction model correction results; Figure 12 A graph showing the correlation results of wave spectrum before and after correction in numerical simulation of wave model; Figure 13 This is a graph showing the correlation results of wave spectrum before and after correction of the wave spectrum correction model of this invention; Figure 14 A comparison of wave height parameters and observed values ​​before and after model correction; Figure 15 A graph showing the comparison between the spectral peak frequency parameters and observed values ​​before and after model correction; Figure 16 A graph showing the comparison between the peak spectral parameters and observed values ​​before and after model correction; Figure 17 Comparison of wave spectra before and after model correction at two locations on the buoy; Figure 18 Comparison of wave spectra before and after model correction at three locations for the buoy; Figure 19 Comparison of wave spectra before and after model correction at four locations for the buoy; Figure 20 Comparison of wave spectra before and after model correction at 5 locations for the buoy; Figure 21 A comparison of wave height parameters and observed values ​​before and after model correction at different buoy locations; Figure 22 A graph showing the comparison between the spectral peak frequency parameters and observed values ​​before and after model correction at different buoy locations; Figure 23 A comparison of the peak spectral parameters and observed values ​​before and after model correction at different buoy locations. Detailed Implementation

[0023] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0024] The innovation of this invention lies in its proposal of a neural network-based spectral bias correction model. This model uses a neural network to correct the accuracy of wave spectra in numerical simulations, enabling a more precise description of wave energy distribution. The model comprehensively verifies its correction performance through two dimensions of spectral parameters (spectral shape and distance, spectral peaks), and four quantitative indicators (wave spectrum, wave height, peak frequency, and peak value). This model solves the error problem caused by the simplification of physical processes in traditional methods during simulation. It also overcomes the limitation of existing data assimilation methods by observational data. This invention can effectively improve the simulation accuracy of wave models and has good adaptability to different sea areas. This invention provides a novel and effective method and strategy for the numerical calculation of wave spectra in real sea areas.

[0025] Example 1, as Figure 1 As shown, the intelligent wave spectrum correction method based on neural networks provided in this embodiment of the invention includes: S1, Target sea area setting and environmental data acquisition; S2, through the numerical wave model, performs numerical calculations on the wave spectrum within the target sea area; S3 constructs a wave spectrum correction model using deep learning methods, including a numerical calculation module, a field measurement processing module, and an intelligent correction module. For the wave spectrum calculated by the wave numerical model, the numerical calculation module constructs numerical simulation wave spectrum data, and the field measurement processing module constructs buoy observation wave spectrum data. Using the buoy observation wave spectrum data as a benchmark, the intelligent correction module constructs the deviation between the numerical simulation wave spectrum data and the buoy observation wave spectrum data, and outputs the corrected wave spectrum. S4 is used to verify the accuracy and evaluate the adaptability of the constructed wave spectrum correction model.

[0026] As demonstrated by the above examples, traditional data assimilation methods involve complex calculations and rely heavily on ocean observation information. Current machine learning-based wave data correction methods primarily focus on global wave parameters and cannot characterize the spatial distribution of wave energy in terms of frequency and direction.

[0027] This invention proposes a deep learning-based spectral bias correction model. By correcting the wave spectrum, this model can more accurately describe the distribution of wave energy. The model's correction performance is comprehensively verified using four quantitative indicators, focusing on two dimensions: spectral shape and distance, and spectral peaks. This model significantly improves the simulation accuracy of numerical wave models and exhibits good adaptability to various sea areas.

[0028] Furthermore, this invention proposes a neural network-based intelligent wave spectrum correction method and system. This model uses buoy observation data as a benchmark to construct a model that corrects the deviation between the simulated wave spectrum and the actual spectrum, thus achieving effective correction of the wave spectrum. This method can more accurately describe the distribution of wave energy in the frequency space. Simultaneously, it can perform high-precision calculations of spectral parameters such as wave height, peak frequency, and peak value. This model can effectively improve the simulation accuracy of wave models and has good adaptability to different sea areas. This invention provides a novel and effective method and strategy for the rapid and accurate calculation of wave spectra.

[0029] Example 2: This invention provides a neural network-based intelligent wave spectrum correction system, which includes: The data acquisition module is used for setting the target sea area and acquiring environmental data; The wave spectrum numerical calculation module is used to perform numerical calculations on the wave spectrum within the target sea area using a wave numerical model. The wave spectrum correction model construction module is used to construct a wave spectrum correction model, including a numerical calculation module, a field measurement processing module, and an intelligent correction module, using deep learning methods. For the wave spectrum calculated by the wave numerical model, the numerical calculation module is used to construct numerically simulated wave spectrum data, and the field measurement processing module is used to construct buoy-observed wave spectrum data. Using the buoy-observed wave spectrum data as a benchmark, the intelligent correction module is used to construct the deviation between the numerically simulated wave spectrum data and the buoy-observed wave spectrum data, and outputs the corrected wave spectrum. The model validation and evaluation module is used to validate the accuracy and evaluate the adaptability of the constructed wave spectrum correction model.

[0030] Example 3, as another embodiment of the present invention, such as Figure 2 As shown, the intelligent wave spectrum correction method based on neural networks provided in this embodiment of the invention includes: S1, Target sea area setting and environmental data acquisition; specifically including: S101, set the target sea area according to requirements; S102, Acquiring marine environmental data, specifically divided into two parts: buoy data within the target sea area and open-source environmental field data. The buoy data includes buoy location, observation time range, wave spectrum data, and wave height. The open-source environmental field data includes wind speed field and water depth information within the target sea area.

[0031] S2, Wave data calculation based on wave models: Numerical calculation of the wave spectrum within the target sea area is performed using wave numerical models; specifically including: S201, Select the model computation domain based on the research sea area range, and set the model computation boundary field with full consideration of the wave transmission characteristics; During model calculations, the external sea area influences the waves within the target sea area. Therefore, a boundary field needs to be set during the calculation process to achieve high-precision wave calculations. Based on the target sea area set in S1, a larger computational domain is further established, and nested calculations are performed on the target sea area. The larger computational domain is set to expand by 20° in each of the four directions (east, west, south, and north) of the target sea area. This serves as the boundary field information for the model calculations of the target sea area.

[0032] S202 uses acquired environmental data such as wind speed to generate a model to calculate the forced field; Based on the wind speed information in the east-west and north-south directions at 10 meters above the sea surface, the forced field in the sea area is calculated. The wind field data is interpolated according to the model calculation grid to obtain forced field data that is consistent with the time and spatial dimension step size set in the calculation process.

[0033] S203 specifies the numerical discretization method used in the calculation process of a numerical wave model (including forced field, boundary field, and numerical discretization, which are all steps in the calculation process of a numerical wave model). This includes spatial discretization of frequency and wave direction, as well as temporal discretization such as global time step. This allows for the calculation of the wave spectrum within the sea area, and simultaneously outputs the wave height at the target location.

[0034] S3, Wave Spectrum Correction Model Construction: This invention constructs a wave spectrum correction model using deep learning methods. For the wave spectrum calculated by numerical wave models, buoy observation data is used as a benchmark. By constructing the deviation between the two, effective correction of the simulated wave spectrum is achieved. The wave spectrum correction model consists of three parts: a numerical calculation module, a field measurement processing module, and an intelligent correction module. Specifically, it includes: S301, Numerical Calculation Module: For the wave spectrum simulated by the wave numerical model in step S2, the spectral frequency is processed by linear interpolation to construct numerically simulated wave spectrum data that is consistent with the spectral frequency of the wave spectrum observed by the buoy.

[0035] This invention innovatively proposes the following linear interpolation calculation method: ; In the formula, For the soundtrack of the ocean waves, For buoy observation frequency, and For the simulated spectral frequencies of the wave mode, The wave spectrum at the frequency observed by the buoy. The simulated spectrum of ocean wave patterns at a certain frequency. The simulated spectrum of the wave mode at adjacent spectral frequencies; After frequency unification, the numerical simulation wave spectrum data were obtained. .

[0036] The expression for calculating the wave spectrum using a numerical model is: ; In the formula, For wave action, For wave energy, Relative frequency; ; In the formula, For wave number, As direction, Wavenumber directional spectrum, For wave action density spectrum, This represents the wave action density spectrum as a function of wave number and direction. The wave number direction spectrum represents the wave number and direction-related information; the wave spectrum is the result of integrating the direction component of the wave direction spectrum. express; After frequency unification, the numerical simulation wave spectrum data were obtained. .

[0037] like Figure 3 As shown, the comparison results of the WW3 wave spectrum interpolation before and after at time 1, and as shown in the figure... Figure 4 The results of the wave spectrum interpolation before and after WW3 at time 2 are shown.

[0038] Furthermore, to address the issues of missing measurements and data loss in buoy data, the time series of wave spectra was processed using measured data as a reference, corresponding to the buoy location or any self-selected location. The specific process was as follows: First, a numerical simulation timeline was created based on the wave model's numerical calculation time step. Second, the buoy observation timeline was created in the same format. Due to data gaps caused by weather and equipment limitations, the numerical simulation data was deleted according to the missing time series and the timeline.

[0039] S302, Actual Measurement Processing Module: For buoy-observed wave spectra, Gaussian filtering is used to smooth out anomalous wave locations in the raw observation data. Peak recovery is then employed to preserve the dominant frequency and energy distribution characteristics of the spectrum, avoiding the loss of crucial wave information. The specific calculation method is shown below: ; ; ; ; In the formula, For Gaussian kernel, This is the result after Gaussian filtering. This is the result after peak recovery. This refers to the calculation process of an exponential function. For At the index position, with The original observation results for the sliding window offset. for The Gaussian kernel calculation results at the location, The peak value of the function. The peak value of the function after Gaussian filtering. For peak index, The peak value in the original signal. The standard deviation of the Gaussian kernel. As the independent variable, Let the radius be the Gaussian kernel. This is the offset of the sliding window. The weighting coefficients for peak recovery. The output signal after Gaussian filtering. For index position, The original signal, This is the final output result after peak recovery; Filtered and processed wave spectrum data from actual measurements, resulting in smooth and high-fidelity data. .

[0040] like Figure 5 As shown, the comparison results of the WW3 wave spectrum before and after filtering at time 1 are as follows: Figure 6 Comparison of wave spectrum filtering results before and after WW3 at time 2. Furthermore, data missing periods due to extreme weather or equipment malfunctions were removed.

[0041] S303, Intelligent Correction Module: Numerical simulation wave spectrum data constructed using the numerical computation module. As input, buoy-observed wave spectrum data were constructed using the measured processing module. This is the output. In this model... and The data dimensions are all 44×6324. Specifically, 44 represents the number of frequencies, and 6324 represents the number of time points. Both the input and output layers contain 44 neurons, representing the wave spectrum results corresponding to the 44 frequencies at a single time point. The model employs three fully connected hidden layers with 200, 200, and 100 neurons respectively. All hidden layers use the ReLU activation function to enhance non-linear expressive power. The output layer uses linear activation. During training, the model first batch-processes the input data, reading it time-by-time and feeding it into the network for forward propagation, outputting the corrected wave spectrum. The model uses mean squared error (MSE) as the loss function to measure the difference between the model output and the observed spectrum. During optimization, the Adam optimizer is used to adaptively adjust the learning rate, improving training efficiency and convergence speed. Ultimately, the wave spectrum is corrected.

[0042] The specific calculation method for the loss function is as follows: ; In the formula, For loss function, For the sample size, These are the actual value and the model prediction, respectively.

[0043] S4, Model Accuracy Verification includes: evaluating the correction effect of the wave spectrum correction model using real wave spectrum data observed from buoys. This specifically includes two parts: wave spectrum accuracy and spectral parameter accuracy. First, the wave spectrum accuracy is verified using the mean absolute error (MAE), mean relative error (MAPE), root mean square error (RMSE), and coefficient of determination (r). Then, the wave heights of the model-corrected wave spectrum and the buoy-observed wave spectrum are calculated. Spectral peak frequency and spectral peak The error is calculated, and the model correction effect is jointly verified by multiple indicators. This enables a comprehensive evaluation of the model correction performance.

[0044] Accuracy verification calculation method: ; ; ; ; In the formula, Indicates the number of samples. and These represent the observed wave results and the calculated wave results, respectively.

[0045] Spectral parameters Spectral peak frequency , wave height The calculation method is as follows: ; ; ; In the formula, This represents the peak value of the wave spectrum. The frequency corresponding to the peak value. For the soundtrack of the ocean waves, For frequency.

[0046] For example, this invention trains and validates a model in a certain sea area in the middle of a certain ocean, and randomly selects four time points to display the wave spectrum results before and after the wave spectrum correction model. Figure 7 Comparison of wave spectra before and after correction of the wave spectrum correction model at time 1. Figure 8 Comparison of wave spectra before and after correction of the wave spectrum correction model at time 2. Figure 9 Comparison of wave spectra before and after correction of the wave spectrum correction model at time 3. Figure 10 Comparison of wave spectra before and after correction of the wave spectrum correction model at time 4; For example, the specific calculation accuracy of the method proposed in this invention is as follows: the MAE and RMSE of the simulated wave spectrum by the wave model are 0.62 and 1.13m, respectively. 2 Hz -1 The MAPE was 37.03%. After correction using the wave correction model, MAE, RMSE, and MAPE all decreased significantly compared to the wave model, with MAPE decreasing to 27.55%. Figure 11 The comparison of wave spectrum accuracy before and after wave spectrum correction model correction is shown. The spectral correlation increased from 0.83 to 0.94, as... Figure 12 Correlation results of wave spectrum before and after correction in numerical simulation of wave model Figure 13 The results show the correlation between wave spectrum before and after correction of the wave spectrum correction model of this invention; the results show that the current model effectively improves the simulation accuracy of wave spectrum.

[0047] Furthermore, this invention utilizes the spectral shape parameter wave height Spectral peak frequency and spectral peak The accuracy of both was evaluated. Table 1 shows that the model constructed in this invention accurately reflects wave height. Spectral peak frequency and spectral peak The calculated MAPE values ​​decreased from 16.42%, 16.18%, and 28.18% in WW3 to 11.11%, 9.82%, and 24.12%, respectively. The correlation coefficient r increased from 0.94, 0.645, and 0.88 to 0.97, 0.789, and 0.93, respectively. The correction effect on the wave parameters is shown in Figure 14-. Figure 16 As shown. Figure 14 The wave height parameters before and after model correction are compared with the observed values. Figure 15 The results show a comparison between the spectral peak frequency parameters before and after model correction and the observed values. Figure 16 The results show the comparison between the peak spectral parameters and the observed values ​​before and after model correction; The results show that the model effectively improves the calculation accuracy of spectral parameters and can achieve comprehensive correction of wave information.

[0048] Table 1 Comparison of spectral parameter accuracy before and after model correction

[0049] For example, step S4, model adaptability assessment, includes: based on the constructed model, further model adaptability analysis is conducted. Multiple buoys are selected within the target sea area, and the wave spectrum at different buoy locations is corrected. The model correction effect is then verified using measured data. This assesses the model's adaptability. Finally, a wave spectrum correction model with strong adaptability to the sea area is constructed.

[0050] This invention utilizes four other buoys within the sea area to evaluate the model's adaptability, with results shown in Figure 17- Figure 20 And as shown in Table 2. Among them Figure 17 Comparison of wave spectra before and after model correction at two locations on the buoy. Figure 18 Comparison of wave spectra before and after model correction at three locations on the buoy. Figure 19 Comparison of wave spectra before and after model correction at four locations on the buoy. Figure 20 Comparison of wave spectra before and after model correction at 5 locations for the buoy; Table 2 Wave spectra before and after model correction at different buoy positions ( Precision comparison

[0051] After model correction, The MAPE is no higher than 12.93%; The MAPE is not higher than 10.50%; The MAPE is no higher than 24.96%. Specific results are shown in Table 3. The changes in spectral parameters over time are detailed below. Figures 21-23 As shown. Figure 21 The wave height parameters before and after model correction are compared with the observed values ​​at different buoy locations. Figure 22The results show a comparison between the spectral peak frequency parameters and observed values ​​before and after model correction at different buoy locations. Figure 23 The results show the comparison between the peak spectral parameters and observed values ​​before and after model correction at different buoy locations.

[0052] Table 3 Comparison of wave spectrum parameters accuracy before and after model correction at different buoy locations.

[0053] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention and within the spirit and principles of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent correction of ocean wave spectra based on neural networks, characterized in that, The method includes the following steps: S1, Target sea area setting and environmental data acquisition; S2, through the numerical wave model, performs numerical calculations on the wave spectrum within the target sea area; S3 constructs a wave spectrum correction model using deep learning methods, including a numerical calculation module, a field measurement processing module, and an intelligent correction module. For the wave spectrum calculated by the wave numerical model, the numerical calculation module constructs numerical simulation wave spectrum data, and the field measurement processing module constructs buoy observation wave spectrum data. Using the buoy observation wave spectrum data as a benchmark, the intelligent correction module constructs the deviation between the numerical simulation wave spectrum data and the buoy observation wave spectrum data, and outputs the corrected wave spectrum. S4 is used to verify the accuracy and evaluate the adaptability of the constructed wave spectrum correction model.

2. The intelligent wave spectrum correction method based on neural networks according to claim 1, characterized in that, In step S1, the target sea area setting and environmental data acquisition include: S101, set the target sea area according to requirements; S102, acquire environmental data within the sea area, including two parts: buoy data and open-source environmental field data within the target sea area; buoy data includes buoy location, observation time range, wave spectrum data, and wave height; open-source environmental field data includes wind speed field and water depth information within the target sea area.

3. The intelligent wave spectrum correction method based on neural networks according to claim 1, characterized in that, In step S2, the wave spectrum of the target sea area is numerically calculated using a wave numerical model, including: S201. Based on the research sea area, the model computation domain is selected, and the model computation boundary field is set with full consideration of the wave transmission characteristics to achieve high-precision calculation of the waves. During the calculation process, a large computation domain is set based on the target sea area, and nested calculations are performed on the target sea area. The large computation domain is set to be a computation domain that expands by 20° in the four directions of east, west, south, and north of the target sea area, thereby providing boundary field information for the calculation of the target sea area by the model. S202: Using the acquired environmental data, a model is generated to calculate the forced field; based on the east-west and north-south wind speed information at 10 meters above the sea surface in the sea area, the forced field in the sea area is calculated, and the wind field data is interpolated according to the model calculation grid to obtain forced field data that is consistent with the time and spatial dimension step size set in the calculation process. S203 sets the numerical discretization method in the calculation process of the wave numerical model, including spatial discretization of frequency and wave direction and long-term discretization of global time step; it completes the numerical solution of the wave spectrum in the sea area and outputs the wave height at the target location.

4. The intelligent wave spectrum correction method based on neural networks according to claim 1, characterized in that, In step S3, the numerical calculation module processes the spectral frequency of the wave spectrum simulated by the wave numerical model using linear interpolation to construct numerically simulated wave spectrum data that is consistent with the spectral frequency of the wave spectrum observed by the buoy. The linear interpolation calculation method is as follows: ； In the formula, For the soundtrack of the ocean waves, For buoy observation frequency, and For the simulated spectral frequencies of the wave mode, The wave spectrum at the frequency observed by the buoy. The simulated spectrum of ocean wave patterns at a certain frequency. The simulated spectrum of the wave mode at adjacent spectral frequencies; After frequency unification, the numerical simulation wave spectrum data were obtained. .

5. The intelligent wave spectrum correction method based on neural networks according to claim 4, characterized in that, To address the issues of missing measurements and data loss in buoy data, the time series of wave spectra is processed using measured data as a reference.

6. The intelligent wave spectrum correction method based on neural networks according to claim 1, characterized in that, In step S3, the measurement processing module smooths the abnormal fluctuation positions in the original observation data of the buoy-observed wave spectrum using a Gaussian filtering method; and preserves the dominant frequency and energy distribution characteristics of the spectrum using a peak recovery method. The specific calculation method is shown below: ； ； ； ； In the formula, For Gaussian kernel, This is the result after Gaussian filtering. This is the result after peak recovery. For the calculation process of the numerical function, For At the index position, with The original observation results for the sliding window offset; for The Gaussian kernel calculation results at the location, The peak value of the function. The peak value of the function after Gaussian filtering. For peak index, The peak value in the original signal. The standard deviation of the Gaussian kernel. As the independent variable, Let the radius be the Gaussian kernel. This is the offset of the sliding window. The weighting coefficients for peak recovery. The output signal after Gaussian filtering. For index position, The original signal, This is the final output result after peak recovery; Filtered and processed wave spectrum data from actual measurements, resulting in smooth and high-fidelity data. .

7. The intelligent wave spectrum correction method based on neural networks according to claim 6, characterized in that, Data gaps caused by extreme weather or equipment malfunctions were removed.

8. The intelligent wave spectrum correction method based on neural networks according to claim 1, characterized in that, In step S3, the intelligent correction module uses the numerical simulation wave spectrum data constructed by the numerical calculation module. As input, buoy-observed wave spectrum data were constructed using the measured processing module. For output; and The data dimensions are all 44×6324; 44 represents the number of frequencies and 6324 represents the number of time points; the input and output layers each contain 44 neurons, representing the wave spectrum results corresponding to 44 frequencies at a single time point; A three-layer fully connected hidden layer is used, with 200, 200, and 100 neurons respectively. All hidden layers use the ReLU activation function to enhance non-linear expression. The output layer uses linear activation. During training, the input data is first batch-processed, then read and fed into the network time-by-time for forward propagation, outputting the corrected wave spectrum. The mean squared error (MSE) is used as the loss function to measure the difference between the output and the observed spectrum. The Adam optimizer is used to adaptively adjust the learning rate during the optimization process, ultimately achieving the correction of the wave spectrum.

9. The intelligent wave spectrum correction method based on neural networks according to claim 8, characterized in that, The specific calculation method for the loss function is as follows: ； In the formula, For loss function, For the sample size, These are the actual value and the model prediction, respectively.

10. A wave spectrum intelligent correction system based on neural networks, characterized in that, The system implements the intelligent wave spectrum correction method based on neural networks as described in any one of claims 1-9, and the system includes: The data acquisition module is used for setting the target sea area and acquiring environmental data; The wave spectrum numerical calculation module is used to perform numerical calculations on the wave spectrum within the target sea area using a wave numerical model. The wave spectrum correction model construction module is used to construct a wave spectrum correction model, including a numerical calculation module, a field measurement processing module, and an intelligent correction module, using deep learning methods. For the wave spectrum calculated by the wave numerical model, the numerical calculation module is used to construct numerically simulated wave spectrum data, and the field measurement processing module is used to construct buoy-observed wave spectrum data. Using the buoy-observed wave spectrum data as a benchmark, the intelligent correction module is used to construct the deviation between the numerically simulated wave spectrum data and the buoy-observed wave spectrum data, and outputs the corrected wave spectrum. The model validation and evaluation module is used to validate the accuracy and evaluate the adaptability of the constructed wave spectrum correction model.

Images