Near-shore sea wave forecasting method and system based on deep learning

Abstract

The invention belongs to the technical field of marine environment prediction, and discloses a near-shore sea wave forecasting method and system based on deep learning. The method comprises the following steps: constructing a near-shore adjacent wave forecasting model NWNN training data set based on deep learning based on open source environment data; a near-shore adjacent wave forecasting model NWNN is constructed; using PyTorch to train a near-shore adjacent wave forecasting model NWNN (Neighbor Wave Nearest Neighbor); and performing accuracy verification on a forecast result by using a buoy measured value which is not trained by the near-shore adjacent wave forecast model NWNN. According to the method, more accurate and reliable data support can be provided for prediction of marine weather events, so that marine economy and life and property safety of the masses can be guaranteed, and the trend and influence of climate change can be known.

Description

Technical Field

[0001] The present invention belongs to the technical field of marine environment prediction, and in particular relates to a method and system for forecasting nearshore waves based on deep learning. Background Art

[0002] Coastal areas are increasingly threatened by extreme wave events due to climate change, sea level rise and intensified storm systems. Accurate forecasting of nearshore wave conditions is critical to ensuring maritime safety and effective disaster response. The current wave forecasting framework mainly relies on wind fields derived from atmospheric models to drive wave predictions. However, due to the inherent assumptions in atmospheric and wave models, errors will continue to accumulate, which often leads to differences between predicted and observed conditions. In addition, the complex coupling relationship between atmospheric driving and ocean processes is often simplified (especially in coastal areas), which limits the accuracy of the model.

[0003] Many current studies have explored deep learning architectures that use time series of buoy data to predict wave conditions. However, these models often suffer from phase deviations between predictions and observations, especially during sudden extreme wave events. This limitation stems from their reliance on historical buoy observations, which do not fully account for the effects of real-time wind fields and long-distance swells, causing predictions to lag behind actual conditions. This time lag is particularly severe during rapidly changing events, such as storm-induced wave surges.

[0004] Invention patent: A method for forecasting nearshore single-point wave height based on deep learning (Announcement No. CN113283588B), disclosed: Step 1: Specify regional marine hydrological and meteorological data set; Step 2: Data cleaning and processing to obtain valid data and divide into training set and test set; Step 3: Train the model to obtain the wave height prediction model after training; Step 4: Verify the accuracy of the wave height prediction model and predict the wave height result. This method is based on the LSTM model to learn the changing characteristics of the measured data of the buoy and forecast the waves at a single point.

[0005] Through the above analysis, the problems and defects of the existing technology are as follows: the existing technology only predicts the wave change trend of the buoy in the future based on the buoy time series data. However, since the waves at a given location are significantly affected by the propagation of swells in the adjacent sea area and the local wind forcing, when wave prediction only relies on single-point time series data, no matter how complex the model is, the prediction accuracy is still low, and it is impossible to predict sudden waves. Summary of the invention

[0006] In order to overcome the problems existing in the related art, the disclosed embodiments of the present invention provide a nearshore wave forecasting method and system based on deep learning.

[0007] The technical solution is as follows: A method for forecasting nearshore waves based on deep learning, comprising the steps of: S1, based on open source environmental data, build a NWNN training dataset for the nearshore nearwave forecasting model based on deep learning; S2, based on the constructed training data set, construct the nearshore nearwave forecast model NWNN; S3, using PyTorch to train the nearshore nearwave forecast model NWNN; S4, using the measured values ​​of the buoys that have not been trained by the nearshore nowwave forecasting model NWNN, the accuracy of the forecast results is verified.

[0008] In step S1, a training data set of a nearshore nearwave forecast model NWNN based on deep learning is constructed, including: obtaining single-point measured data through buoys, and obtaining regional wind and wave data through numerical models; the regional wind and wave data include single-point and regional significant wave heights, and wind speeds in the UV direction; for missing and outliers in the measured data, the outliers are removed and the missing values ​​are filled.

[0009] Furthermore, the outlier is a value greater than 99. After the removal, the vacant positions and missing values ​​are filled by bilinear interpolation. The source of the filling data is numerical simulation or open source reanalysis data. The specific implementation method of bilinear interpolation is: Set the matrix midpoint and the four points around it , , , , and the corresponding values ​​are , , , , ,but .

[0010] In step S2, a nearshore nearwave forecast model NWNN is constructed, including: Construct the buoy feature extraction module and the regional wind and wave feature extraction module. The buoy feature extraction module consists of four fully connected layers, and each layer is activated using the ReLU function; The regional wind and wave feature extraction module consists of two convolutional pooling layers. After each convolution, the ReLU function is used for activation. The resulting matrix vector is flattened and input into a fully connected layer. The results of the two modules are concatenated and passed through a fully connected layer to generate predicted waves, completing the construction of the nearshore wave forecast model NWNN. The input of the nearshore wave forecast model is the measured data of the buoy in the past 24 hours and the wind field and wave grid data of the area in the past 2 hours. The input dimensions are two-dimensional tensor B×T and four-dimensional tensor B×(T×V)×H×W respectively; the output is the wave data of the next 24 hours, and the output dimension is two-dimensional tensor B×T; among them, B is the batch size, T is the time length, V is the number of variables, D is the radial length, H is the longitude length, and W is the latitudinal length.

[0011] Furthermore, the first layer of the buoy feature extraction module maps the input of length 24 to a feature vector of length 64, and performs ReLU activation; the subsequent layers expand to a feature vector of length 128, which is reduced to a feature vector of length 64 to capture temporal features; and the feature dimension is reduced to 32; The specific expression of full connection is: ; In the formula, is the input vector, is the weight matrix, is the bias vector, is the output vector; The specific expression of the ReLU activation function is: ; In the formula, is the input vector.

[0012] Furthermore, the first layer of the regional wind and wave feature extraction module applies 64 convolution kernels of size 3×3, which are activated by ReLU and 2×2 maximum pooling layers to reduce the spatial dimension; the second convolution layer applies 128 convolution kernels with the same kernel size and padding strategy, which are activated by ReLU and 2×2 maximum pooling again; after the last pooling operation, the feature map is flattened into a 1-dimensional vector and then input into the fully connected layer to reduce the feature dimension to 32; the specific expression of convolution is: ; In the formula, For the The sample in The convolution result under the convolution kernel is For the The bias term of the convolution kernel, is the warp length, is the latitudinal length, for The index of For the The convolution kernel is The weight value when For the Samples in Get data value; ; In the formula, is the element in the output feature map after pooling, represents the position of the pooling window in the input feature map, represents the range of the pooling window, Indicates that the maximum value in the window is selected as the output value. and is the index range of the pooling window, is the position in the output feature map after pooling The value of is the position in the input feature map after pooling The value of .

[0013] Furthermore, the results of the buoy feature extraction module and the regional wind and wave feature extraction module are concatenated, including: concatenating the two columns of feature vectors in the second dimension, and the expression is: ; In the formula, is the output vector of the buoy feature extraction module, is the output vector of the regional wind and wave feature extraction module, is the concatenated vector, To concatenate a and b, The dimension of the stitching.

[0014] In step S3, the nearshore nearwave forecast model NWNN is trained using PyTorch, including the following steps: S301, the nearshore nearwave forecast model NWNN takes the mean absolute error MAE as the optimization target; ; In the formula, and are the true value and the predicted value, is the total amount of data; S302, the learning rate follows the cosine annealing algorithm during the entire training process; the specific expression of cosine annealing is: ; In the formula, For the The learning rate for the number of training rounds, is the total number of training rounds, is the learning rate decay factor, is the initial learning rate.

[0015] In step S4, the accuracy of the forecast results is verified by using the measured values ​​of the buoys that have not been trained by the nearshore nearwave forecast model NWNN, including: Using correlation coefficient , Root Mean Square Error , mean absolute percentage error and skill scores To evaluate the prediction performance of the nearshore nearwave forecast model NWNN and the existing untrained models; the expression is: ; ; ; ; In the formula, is the true value, is the predicted value, is the average of the predicted values, is the average of the true values, is the true value of the previous time step.

[0016] Another object of the present invention is to provide a nearshore wave forecasting system based on deep learning, which implements the nearshore wave forecasting method based on deep learning, and the system comprises: The training dataset construction module is used to construct the NWNN training dataset for the nearshore nearwave forecasting model based on deep learning based on open source environmental data; A nearshore nearwave forecast model construction module is used to construct a nearshore nearwave forecast model NWNN based on the constructed training data set; Training module, used to train the nearshore nearwave forecasting model NWNN using PyTorch; The verification module is used to verify the accuracy of the forecast results using the actual measured values ​​of the buoy that have not been trained by the nearshore nearwave forecast model NWNN.

[0017] In combination with all the above technical solutions, the beneficial effects of the present invention are as follows: First, based on the problems existing in current wave prediction, the present invention proposes a nearshore wave forecasting method that integrates buoy observations and regional wind and wave environment. Since the waves at a given location are significantly affected by the propagation of swell waves in the adjacent sea area and local wind forcing, accurate prediction must not only consider the temporal characteristics of single-point waves, but also the wind and wave conditions in the surrounding areas. The present invention constructs a nearshore wave nowcasting model (Nearshore Wave Nowcasting Network, NWNN), and extracts the buoy characteristics of a single point and the wind and wave characteristics of the region at the same time, making up for the shortcomings that the current single-point wave forecast cannot take into account the wave propagation characteristics, and the regional wave forecast cannot be combined with measured data, providing key support for nearshore disaster prevention, maritime safety, etc.

[0018] Second, the present invention constructs a dynamically controlled nearshore wave nowcasting model (NWNN), innovatively integrates the buoy time series characteristics and regional wind and wave information, and achieves accurate simulation of wave generation and propagation and accurate forecast for the next 24 hours. Its technical application will significantly enhance the early warning capability in extreme weather events and provide strong data support for marine disaster preparedness, shipping safety, nearshore engineering and marine resource development.

[0019] Third, the technical solution of the present invention successfully solves the technical problems that have long been difficult to overcome in the field of ocean wave forecasting. Due to the limitations of single-point buoy data or regional models, traditional methods cannot accurately capture sudden extreme ocean wave events, especially under rapidly changing storm surges, and the prediction results often deviate significantly from the actual situation. The present invention innovatively combines the timing characteristics of buoys and regional wind and wave characteristics to construct a dynamic deep learning model (NWNN), which comprehensively considers the complex physical characteristics of wave generation and propagation, not only significantly improving the forecast accuracy, but also breaking through the phase deviation problem between the predicted value and the measured value. This technological breakthrough fills the gap between single-point and regional models, provides a new solution for accurate ocean wave forecasting in extreme weather, and meets people's long-standing urgent needs for high-precision marine disaster warnings and safety assurance. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings herein are incorporated in and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description, serve to explain the principles of the present disclosure; Figure 1 is a schematic diagram of a nearshore wave forecasting method based on deep learning provided by an embodiment of the present invention; Figure 2 is a flow chart of a method for forecasting nearshore waves based on deep learning provided by an embodiment of the present invention; Figure 3 It is a structural diagram of a nearshore nearwave forecast model NWNN provided by an embodiment of the present invention; Figure 4 This is a comparison chart of the predicted values ​​and measured values ​​of the most commonly used models; Figure 5 This is the error diagram between the predicted values ​​and the measured values ​​of the most commonly used models; Figure 6 Skill score diagram of existing methods at different forecast durations; Figure 7 A comparison chart of the forecast results of the NWNN of the present invention and the ANN of the prior art 24 hours in advance; Figure 8 A comparison chart of the prediction effects of the NWNN of the present invention and the ANN of the prior art on extreme waves; Fig. 9 is a schematic diagram of a nearshore wave forecasting system based on deep learning provided by an embodiment of the present invention; In the figure: 1. Training data set construction module; 2. Nearshore nearwave forecast model construction module; 3. Training module; 4. Verification module. DETAILED DESCRIPTION

[0021] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific implementation disclosed below.

[0022] The innovation of the present invention is that it organically combines the time series characteristics of buoys with the characteristics of regional wind and waves for the first time, comprehensively captures the generation and propagation characteristics of waves, and realizes high-precision nowcasting of nearshore waves. Specifically, the present invention processes single-point measured data and regional environmental information respectively through buoy feature extraction modules and regional wind and wave feature extraction modules, and uses a multi-layer network to fuse these two types of features, thereby significantly improving the accuracy of predictions, especially in extreme wave events, far exceeding traditional models, solving the technical difficulties of traditional methods in phase deviation and prediction delay, and providing important support for nearshore disaster prevention and control and marine resource development.

[0023] Example 1: The nearshore wave forecasting method based on deep learning provided by the embodiment of the present invention proposes a nearshore wave nowcasting model, which fully considers the propagation and generation characteristics of waves, combines the timing characteristics of buoys with the wind and wave characteristics of the region, and accurately forecasts nearshore waves. Its forecasting accuracy far exceeds that of existing models, and it also has excellent forecasting effects for large waves. The present invention can provide reliable safety guarantees for marine activities, help prevent and reduce disasters, improve the development and utilization efficiency of marine resources, and provide scientific support for coastal management and ecological protection.

[0024] like Figure 1 As shown, the nearshore wave forecasting method based on deep learning provided by the embodiment of the present invention first preprocesses the buoy data, including but not limited to interpolating missing buoy data, removing abnormal buoy data, transforming the spatial and temporal resolution of regional wind and wave data, and other operations to unify the data format. After preprocessing the data source, a data set for deep learning model training is obtained.

[0025] Secondly, a nearshore wave nowcasting model (NWNN) is constructed based on the constructed buoy and regional wind and wave datasets. NWNN consists of two modules: the buoy feature extraction module and the regional wind and wave feature extraction module. The buoy feature extraction module consists of four fully connected layers, each layer is activated by the ReLU function, and finally a column of feature vectors is formed. The regional wind and wave feature extraction module consists of two convolutional pooling layers, each layer is activated by the ReLU function, and finally the formed matrix vector is flattened, input into a fully connected layer and a feature vector with the same output length as the buoy feature extraction module is generated. Finally, the feature vectors generated by the two modules are spliced ​​and passed through a fully connected layer to generate the predicted waves. The input of the model is the measured data of the buoy in the past 24 hours and the wind field and wave grid data of the region in the past 2 hours, and the output is the wave data for the next 24 hours.

[0026] Finally, the model output results are compared with the measured data to verify the forecast accuracy. The evaluation indicators include: root mean square error, mean relative error, mean absolute error, correlation coefficient, and skill score.

[0027] Example 2, as another embodiment of the present invention, Figure 2 As shown, the nearshore wave forecasting method based on deep learning provided by the embodiment of the present invention includes the following steps: S1, based on open source environmental data, build a NWNN training dataset for the nearshore nearwave forecasting model based on deep learning; Single-point measured data is obtained through buoys, and regional wind and wave data is obtained through numerical models, including single-point and regional significant wave heights and wind speed in the UV direction. Since the measured data may be missing or abnormal, it is necessary to remove abnormal values ​​and fill in missing values. The numerical model includes operations such as temporal and spatial resolution transformation to unify the data format. Among the data sets, 70% is used as a training set and 30% is used as a test set.

[0028] S2, based on the constructed training data set, construct the nearshore nearwave forecast model NWNN; Construct the buoy feature extraction module and the regional wind and wave feature extraction module. The buoy feature extraction module consists of four fully connected layers, each of which is activated by the ReLU function. The regional wind and wave feature extraction module consists of two convolutional pooling layers, which are activated by the ReLU function after each convolution. Finally, the matrix vector formed is flattened and input into a fully connected layer. Finally, the results of the two modules are concatenated and a fully connected layer is used to generate the predicted waves; the construction of the nearshore nearshore wave forecast model NWNN is completed.

[0029] The input of the nearshore near-shore wave forecast model is the measured data of the buoy in the past 24 hours and the wind field and wave grid data of the area in the past 2 hours. The input dimensions are two-dimensional tensor B×T and four-dimensional tensor B×(T×V)×H×W respectively; the output is the wave data of the next 24 hours, and the output dimension is two-dimensional tensor B×T; among them, B is the batch size, T is the time length, V is the number of variables, D is the radial length, H is the longitude length, and W is the latitudinal length; S3, using PyTorch to train the nearshore nearwave forecast model NWNN; The nearshore nearwave forecast model NWNN is trained for a total of 500 rounds with a batch size of 32. In each round of training, the model parameters are updated according to the gradient of the mean absolute error loss function, which measures the difference between the predicted value and the true value. The learning rate scheduler dynamically adjusts the learning rate in each round to ensure efficient learning throughout the training process. In order to monitor the progress of training, the nearshore nearwave forecast model NWNN is evaluated on the test set after each round. If the test loss improves, the current nearshore nearwave forecast model NWNN model weights are saved as the best performing nearshore nearwave forecast model NWNN to prevent overfitting and ensure optimal performance.

[0030] S4, using the measured values ​​of the buoys that have not been trained by the nearshore nowwave forecasting model NWNN, the accuracy of the forecast results is verified.

[0031] Exemplarily, the outliers in the measured values ​​in step S1 are defined as values ​​greater than 99. The vacant positions and other missing values ​​after elimination are filled by bilinear interpolation. The source of the filling data is numerical simulation or open source reanalysis data. The invention innovatively proposes that the specific implementation method of bilinear interpolation is: set the midpoint of the matrix and the four points around it , , , , and the corresponding values ​​are , , , , ,but .

[0032] For example, Figure 3 As shown, in step S2, a nearshore nearwave forecast model NWNN is constructed, including: Construct the buoy feature extraction module and the regional wind and wave feature extraction module. The buoy feature extraction module consists of four fully connected layers, and each layer is activated using the ReLU function; S201, the float feature extraction module consists of four fully connected layers. The first layer maps the input of length 24 to a feature vector of length 64, followed by ReLU activation. Subsequent layers expand to a feature vector of length 128, and then reduce to a feature vector of length 64, ensuring that the temporal features are fully captured. Finally, the feature dimension is reduced to 32.

[0033] The innovative proposal of the present invention is that the specific expression of full connection is: ; Where: is the input vector, is the weight matrix, is the bias vector, is the output vector.

[0034] The specific expression of the ReLU activation function is: ; In the formula, is the input vector.

[0035] S202, the regional wind and wave feature extraction module uses two convolution layers. The first layer applies 64 convolution kernels of size 3×3, followed by ReLU activation and 2×2 maximum pooling layers to reduce the spatial dimension. The second convolution layer applies 128 convolution kernels with the same kernel size and padding strategy, followed by ReLU activation and 2×2 maximum pooling again. After the last pooling operation, the feature map is flattened into a 1-dimensional vector and then input into the fully connected layer to reduce the feature dimension to 32. The present invention innovatively proposes that the specific expression of convolution is: ; Where: is the warp length, is the latitudinal length, for The index of For the The convolution kernel is The weight value when For the The samples are in Get the data value, For the The bias term of the convolution kernel, For the The sample in Convolution result under convolution kernel.

[0036] The specific expression of pooling is: ; In the formula, is the element in the output feature map after pooling, represents the position of the pooling window in the input feature map, represents the range of the pooling window, Indicates that the maximum value in the selection window is used as the output value.

[0037] S203, concatenating the results of the two modules means concatenating the two columns of feature vectors in the second dimension. Suppose the output vector of the buoy feature extraction module is , the output vector of the regional wind and wave feature extraction module is , the innovative proposal of the present invention, then the concatenated vector for: ; Exemplarily, the step S3 uses PyTorch to train the nearshore nearwave forecast model NWNN, specifically comprising: S301, the nearshore nearwave forecast model NWNN takes the mean absolute error (MAE) as the optimization target: ; in, and are the true value and the predicted value, The total amount of data.

[0038] S302, in order to improve convergence, the learning rate follows the cosine annealing algorithm during the entire training process. The invention innovatively proposes that the specific expression of cosine annealing is: ; Where: For the The learning rate for the number of training rounds, is the total number of training rounds, is the learning rate decay factor, is the initial learning rate.

[0039] Exemplarily, the step S4 of verifying the sea wave forecast result specifically includes: S401, using the correlation coefficient (r), root mean square error (RMSE), mean absolute percentage error (MAPE) proposed by the present invention and the skill score (Skill) innovatively proposed by the present invention to evaluate the prediction performance of different models.

[0040] ; ; ; ; In the formula, is the true value, is the predicted value, is the average of the predicted values, is the average of the true values, is the true value of the previous time step.

[0041] S402, Figure 4 This is a comparison chart of the predicted values ​​and measured values ​​of several most commonly used models. Figure 5 This is the error diagram between the predicted values ​​and the measured values ​​of several most commonly used models.

[0042] The limitations of these models are particularly evident during extreme weather events. For example, all models failed to capture the sudden rise in SWH around March 13 to March 16, 2023, showing a delayed response. Even advanced models such as Transformers and Long Short-Term Memory Networks (LSTM) did not show a clear advantage over simple models such as autoregressive models (AR) and artificial neural networks (ANN) in predicting these rapid changes.

[0043] S403, comparing the prediction results of several currently most commonly used models with the persistence model. Figure 6 Skill scores of existing methods at different forecast horizons. The five-pointed star represents the highest-scoring persistence model for that forecast horizon; the baseline for evaluating forecast models is the assumption that future wave conditions will be the same as current observations. Existing methods do not significantly improve on the persistence model, and in some cases even perform worse. For example, the Transformer model has a skill score of -0.15 for a 2-hour forecast.

[0044] S404, Figure 7 The NWNN of the present invention is compared with the prior art ANN 24h ahead forecast results. The performance gap between NWNN and existing methods is very obvious. Many predictions of ANN deviate significantly from the "perfect prediction" diagonal (red dashed line), especially for extreme waves above 3 meters, but NWNN provides more consistent predictions, and the skill score significantly exceeds ANN and other existing methods. And NWNN shows stronger performance in extreme wave conditions, compared with the purple marks (representing the large errors of ANN), the orange marks (representing the large errors of NWNN) are significantly less.

[0045] in, Figure 7 The middle dashed line is the best fit line, the blue solid line is the NWNN fit line, and the green solid line is the ANN fit line. The orange points are points where the NWNN error is large (error is greater than 1m). The purple points are points where the ANN error is large. The lower right corner is the skill score of the NWNN.

[0046] S405, Figure 8 The prediction effects of the NWNN of the present invention and the ANN of the prior art for extreme waves (wave height greater than 3m) are compared. The error of the NWNN is more biased towards the 0 axis than that of the ANN, and the ANN will significantly underestimate the wave intensity as the forecast time increases. Figure 8 In the figure, the solid line is the NWNN prediction result, and the dotted line is the ANN prediction result.

[0047] S406, the present invention compares the forecast results of NWNN with the most advanced European Meteorological Center. The forecast results of NWNN are comparable to those of the European Meteorological Center, and the forecast accuracy of some locations even exceeds that of the European Meteorological Center. At different locations, such as: Based on the measured value of 3.04m, the predicted value of the present invention can reach 2.31m, while the predicted value of the European Meteorological Center is only 1.95m; Based on the measured value of 2.76m, the predicted value of the present invention can reach 2.33m, while the predicted value of the European Meteorological Center is only 1.98m; Based on the measured value of 1.49m, the predicted value of the present invention can reach 1.41m, while the predicted value of the European Meteorological Center is only 1.39m; Based on the measured value of 1.89m, the predicted value of the present invention can reach 1.62m, while the predicted value of the European Meteorological Center is only 1.59m; Based on the measured value of 1.31m, the predicted value of the present invention can reach 1.28m, while the predicted value of the European Meteorological Center is only 1.90m; Based on the measured value of 1.63m, the predicted value of the present invention can reach 1.46m, while the predicted value of the European Meteorological Center is only 1.76m; Based on the measured value of 1.45m, the predicted value of the present invention can reach 1.39m, while the predicted value of the European Meteorological Center is only 1.58m; Based on the measured value of 0.86m, the predicted value of the present invention can reach 0.91m, while the predicted value of the European Meteorological Center is only 1.08m; Based on the measured value of 0.52m, the predicted value of the present invention can reach 0.60m, while the predicted value of the European Meteorological Center is only 0.53m.

[0048] It can be seen from the above embodiments that the present invention can accurately predict the waves in the nearshore area in the next 24 hours, especially the sudden large waves. Compared with the existing methods, the present invention can well simulate the generation and propagation characteristics of the waves by extracting the dual features of the measured data and the regional wind and wave information, solve the phase deviation problem between the predicted value and the measured value of the existing method, and effectively improve the prediction accuracy. The present invention can provide more accurate and reliable data support for the prediction of marine weather events, thereby helping to protect the marine economy and the safety of life and property of the people, and understand the trend and impact of climate change.

[0049] Embodiment 3, as Fig. 9 As shown, the nearshore wave forecasting system based on deep learning provided by the embodiment of the present invention includes: Training data set construction module 1 is used to construct the NWNN training data set of the nearshore nearwave forecasting model based on deep learning based on open source environmental data; A nearshore nearwave forecasting model building module 2 is used to build a nearshore nearwave forecasting model NWNN based on the constructed training data set; Training module 3, used to train the nearshore nearwave forecasting model NWNN using PyTorch; Verification module 4 is used to verify the accuracy of the forecast results using the measured values ​​of the buoys that have not been trained by the nearshore nearwave forecast model NWNN.

[0050] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for forecasting nearshore waves based on deep learning, characterized in that: The method comprises the steps of: S1, based on open source environmental data, build a training dataset for the nearshore nearwave forecasting model NWNN based on deep learning; S2, based on the constructed training data set, construct the nearshore nearwave forecast model NWNN; S3, using PyTorch to train the nearshore nearwave forecast model NWNN; S4, the accuracy of the forecast results is verified using the measured values ​​of the buoys that have not been trained by the nearshore nowwave forecast model NWNN.

2. The method for forecasting nearshore waves based on deep learning according to claim 1, characterized in that: In step S1, a training data set of a nearshore nearwave forecast model NWNN based on deep learning is constructed, including: obtaining single-point measured data through buoys, and obtaining regional wind and wave data through numerical models; the regional wind and wave data include single-point and regional significant wave heights, and wind speeds in the UV direction; for missing and outliers in the measured data, the outliers are removed and the missing values ​​are filled.

3. The method for forecasting nearshore waves based on deep learning according to claim 2, characterized in that: The outliers are values ​​greater than 99. After being removed, the vacant positions and missing values ​​are filled by bilinear interpolation. The source of the filling data is numerical simulation or open source reanalysis data. The specific implementation method of bilinear interpolation is: set the midpoint of the matrix and the four points around it , , , , and the corresponding values ​​are , , , , ,but .

4. The method for forecasting nearshore waves based on deep learning according to claim 1, characterized in that: In step S2, a nearshore nearwave forecast model NWNN is constructed, including: Construct the buoy feature extraction module and the regional wind and wave feature extraction module. The buoy feature extraction module consists of four fully connected layers, and each layer is activated using the ReLU function; The regional wind and wave feature extraction module consists of two convolutional pooling layers. After each convolution, the ReLU function is used for activation. The resulting matrix vector is flattened and input into a fully connected layer. The results of the two modules are concatenated and passed through a fully connected layer to generate predicted waves, completing the construction of the nearshore wave forecast model NWNN. The input of the nearshore wave forecast model is the measured data of the buoy in the past 24 hours and the wind field and wave grid data of the area in the past 2 hours. The input dimensions are two-dimensional tensor B×T and four-dimensional tensor B×(T×V)×H×W respectively; the output is the wave data of the next 24 hours, and the output dimension is two-dimensional tensor B×T; among them, B is the batch size, T is the time length, V is the number of variables, D is the radial length, H is the longitude length, and W is the latitudinal length.

5. The method for forecasting nearshore waves based on deep learning according to claim 4, characterized in that: The first layer of the buoy feature extraction module maps the input of length 24 to a feature vector of length 64, and performs ReLU activation; the subsequent layers expand to a feature vector of length 128, which is reduced to a feature vector of length 64 to capture temporal features; reducing the feature dimension to 32; The specific expression of full connection is: ； In the formula, is the input vector, is the weight matrix, is the bias vector, is the output vector; The specific expression of the ReLU activation function is: 。 6. The method for forecasting nearshore waves based on deep learning according to claim 5, characterized in that: The first layer of the regional wind and wave feature extraction module applies 64 convolution kernels of size 3×3, which are activated by ReLU and 2×2 maximum pooling layers to reduce the spatial dimension; the second convolution layer applies 128 convolution kernels with the same kernel size and padding strategy, which are activated by ReLU and 2×2 maximum pooling again; after the last pooling operation, the feature map is flattened into a 1-dimensional vector and then input into the fully connected layer to reduce the feature dimension to 32; the convolution is specifically expressed as: ； In the formula, For the The sample in The convolution result under the convolution kernel is For the The bias term of the convolution kernel, is the warp length, is the latitudinal length, They are The index of For the The convolution kernel is The weight value when For the Samples in Get data value; ； In the formula, is the element in the output feature map after pooling, represents the position of the pooling window in the input feature map, represents the range of the pooling window, Indicates that the maximum value in the window is selected as the output value. and is the index range of the pooling window, is the position in the output feature map after pooling The value of is the position in the input feature map after pooling The value of .

7. The method for forecasting nearshore waves based on deep learning according to claim 4, characterized in that: The results of the buoy feature extraction module and the regional wind and wave feature extraction module are concatenated, including: concatenating the two columns of feature vectors in the second dimension, and the expression is: ； In the formula, is the output vector of the buoy feature extraction module, is the output vector of the regional wind and wave feature extraction module, is the concatenated vector, For the general and To splice, The dimension of the stitching.

8. The method for forecasting nearshore waves based on deep learning according to claim 1, characterized in that: In step S3, the nearshore nearwave forecast model NWNN is trained using PyTorch, including the following steps: S301, the nearshore nearwave forecast model NWNN takes the mean absolute error MAE as the optimization target; ； In the formula, and are the true value and the predicted value, is the total amount of data; S302, the learning rate follows the cosine annealing algorithm during the entire training process; the specific expression of cosine annealing is: ； In the formula, For the The learning rate for the number of training rounds, is the total number of training rounds, is the learning rate decay factor, is the initial learning rate.

9. The method for forecasting nearshore waves based on deep learning according to claim 1, characterized in that: In step S4, the accuracy of the forecast results is verified by using the measured values ​​of the buoys that have not been trained by the nearshore nearwave forecast model NWNN, including: Using correlation coefficient , Root Mean Square Error , mean absolute percentage error and skill scores To evaluate the prediction performance of the nearshore nearwave forecast model NWNN and the existing untrained models; the expression is: ； ； ； ； In the formula, is the true value, is the predicted value, is the average of the predicted values, is the average of the true values, is the true value of the previous time step.

10. A nearshore wave forecasting system based on deep learning, characterized in that: The system implements the nearshore wave forecasting method based on deep learning as claimed in any one of claims 1 to 9, and the system comprises: A training dataset construction module (1) is used to construct a deep learning-based nearshore nearwave forecasting model NWNN training dataset based on open source environmental data; A nearshore nearwave forecast model construction module (2) is used to construct a nearshore nearwave forecast model NWNN based on the constructed training data set; Training module (3), used to train the nearshore nearwave forecasting model NWNN using PyTorch; The verification module (4) is used to verify the accuracy of the forecast results using the measured values ​​of the buoys that have not been trained by the nearshore nearwave forecast model NWNN.

Images