Method, device and medium for reconstructing flow velocity field of isolated wave in ocean

Abstract

The present application relates to a method, device, equipment and medium for reconstructing internal solitary wave flow velocity field in the ocean, and relates to the technical field of internal solitary wave reconstruction. The method comprises: constructing a data set according to internal solitary waves generated by the gravitational collapse method; constructing a flow velocity field reconstruction model, including an encoder and a decoder; wherein the encoder comprises a hierarchically connected DoubleConv module and a pooling layer, the decoder comprises a hierarchically connected transposed convolution layer and a DoubleConv module, and each DoubleConv module is integrated into a channel attention module and a spatial attention module; based on a preset loss function, the flow velocity field reconstruction model is trained through the data set until the model converges, obtaining a trained flow velocity field reconstruction model; inputting the waveform profile line of the internal solitary wave to be reconstructed into the trained flow velocity field reconstruction model for processing, and outputting the internal solitary wave flow velocity field reconstruction result in the ocean. The present application can improve the reconstruction accuracy and efficiency of the flow velocity field.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of internal solitary wave reconstruction, in particular to a method, device, equipment and medium for reconstructing an ocean internal solitary wave flow field. BACKGROUND

[0002] As a typical internal wave propagating along the density jump of the marginal sea, the internal solitary wave is generated by the nonlinear steepening of the internal tide encountering the seamount or continental shelf slope, and its amplitude can reach hundreds of meters. The strong current induced in the propagation process not only deeply affects the internal environmental changes of the ocean, but also continuously threatens the safe operation of marine engineering equipment and underwater vehicles. Therefore, accurately obtaining the internal solitary wave flow field information is of great significance to marine scientific research and engineering safety protection.

[0003] Traditionally, the internal solitary wave flow field is mainly obtained by two types of technical paths: one is the direct measurement based on the underwater submersible carrying ADCP instruments, which collects flow velocity data at different depths and filters the background flow interference to achieve observation. Although this method can provide reliable data support for related research, it is difficult to achieve large-scale and efficient flow field measurement due to the complex marine observation environment, inevitable ocean current interference, uncertainty of internal solitary wave excitation, and high deployment and maintenance cost of observation instruments. The other is the method relying on visible light or SAR satellite remote sensing observation combined with dynamics theory inversion. The internal solitary wave in the satellite image presents alternating light and dark stripes due to the modulation of sea surface roughness by the induced ocean current. Based on the distribution of these stripes, the wave parameters can be inverted and the flow velocity field can be calculated through the nonlinear internal solitary wave theory model. However, this path has a strong dependence on real-time environmental field information, and requires temperature, salinity, density and other element data at full water depth during the propagation of internal solitary waves. It is difficult to achieve efficient and stable real-time data collection in full water depth range by marine and underwater observation platforms. In practical applications, it relies on ocean models or reanalysis data, but is limited by the spatial and temporal resolution of the data. At the same time, the weak nonlinear theory has inherent limitations in describing the flow field accuracy as it does not consider the flow line displacement induced by wave motion.

[0004] In recent years, the flow velocity field reconstruction model technology has been tried to be applied to the identification and parameter estimation of internal solitary waves in satellite images due to its learning ability for complex features and patterns, such as support vector machine, YOLO model, etc. However, existing researches mainly focus on the existence detection and waveform parameter extraction of internal solitary waves, and the core flow velocity field reconstruction capability is still lacking in-depth exploration, and there is no flow field reconstruction scheme that can be independent of environmental field information and dynamics theory support. SUMMARY

[0005] Therefore, it is necessary to provide a method, device, equipment and medium for reconstructing a flow velocity field of an internal solitary wave in the ocean, which can be independent of real-time environmental field information and dynamic theory support, and can balance the reconstruction accuracy and efficiency.

[0006] A method for reconstructing a flow velocity field of an internal solitary wave in the ocean, the method comprising:

[0007] An internal solitary wave is generated according to a gravity collapse method, and a waveform profile line and a corresponding wave-induced flow velocity field of the internal solitary wave are synchronously measured to construct a data set;

[0008] A flow velocity field reconstruction model is constructed, the flow velocity field reconstruction model comprising an encoder and a decoder; wherein the encoder comprises hierarchically connected DoubleConv modules and a pooling layer, the decoder comprises hierarchically connected transposed convolution layers and DoubleConv modules, and each DoubleConv module incorporates a channel attention module and a spatial attention module;

[0009] The flow velocity field reconstruction model is trained based on a preset loss function through the data set until the model converges, and a trained flow velocity field reconstruction model is obtained;

[0010] The waveform profile line of an internal solitary wave to be reconstructed is input into the trained flow velocity field reconstruction model for processing, and a flow velocity field reconstruction result of the internal solitary wave in the ocean is output.

[0011] In another aspect, a device for reconstructing a flow velocity field of an internal solitary wave in the ocean is also provided, comprising:

[0012] A data set construction module is configured to generate an internal solitary wave according to a gravity collapse method, synchronously measure a waveform profile line and a corresponding wave-induced flow velocity field of the internal solitary wave, and construct a data set;

[0013] A flow velocity field reconstruction model construction module is configured to construct a flow velocity field reconstruction model, the flow velocity field reconstruction model comprising an encoder and a decoder; wherein the encoder comprises hierarchically connected DoubleConv modules and a pooling layer, the decoder comprises hierarchically connected transposed convolution layers and DoubleConv modules, and each DoubleConv module incorporates a channel attention module and a spatial attention module;

[0014] A model training module is configured to train the flow velocity field reconstruction model based on a preset loss function through the data set until the model converges, and obtain a trained flow velocity field reconstruction model;

[0015] A flow velocity field reconstruction module is configured to input the waveform profile line of an internal solitary wave to be reconstructed into the trained flow velocity field reconstruction model for processing, and output a flow velocity field reconstruction result of the internal solitary wave in the ocean.

[0016] On another front, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-mentioned method for reconstructing the velocity field of isolated waves in the ocean.

[0017] Furthermore, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the above-described method for reconstructing the velocity field of isolated waves in the ocean.

[0018] Compared with existing technologies, the ocean solitary wave velocity field reconstruction method, apparatus, equipment, and medium provided by this invention have the following beneficial effects:

[0019] 1. Internal isolated waves are generated using the gravity collapse method to construct a dataset containing waveform contour lines and corresponding wave-induced velocity fields. The stratified fluid parameters can be precisely controlled through the experimental environment, avoiding uncontrollable interference from the actual marine environment. Synchronous measurements directly establish the correspondence between waveform contour lines and wave-induced velocity fields, with high data matching degree and low noise, providing high-quality and highly reliable basic data for model training.

[0020] 2. The constructed velocity field reconstruction model adopts an encoder and decoder structure to achieve layer-by-layer extraction from shallow waveform details to deep abstract features. Furthermore, the DoubleConv module incorporates a channel attention layer and a spatial attention layer, which can specifically enhance channel features and spatial locations that are strongly correlated with the flow field distribution. This can strengthen feature capture and nonlinear mapping capabilities, thereby improving the accuracy and efficiency of velocity field reconstruction.

[0021] 3. The method proposed in this invention only requires the waveform contour of the isolated wave to be reconstructed, and the velocity field reconstruction result can be directly output through the model. It does not require real-time environmental field parameter support, nor does it require calling complex dynamic theory models for inversion calculation. It gets rid of the dependence on environmental field and dynamic theory, simplifies the velocity field reconstruction process, and improves practicality and scenario adaptability.

[0022] 4. By generating datasets in a controllable environment and training the model with a preset loss function, the model can learn the data patterns without deviating from the physical essence of the internal solitary wave flow field. The model has strong stability and generalization ability after training convergence. It can not only stably reconstruct the flow field of laboratory scenarios, but also transfer it to real sea scenarios. It can output stable and accurate flow velocity field results for internal solitary waves of different nonlinear intensities. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention, and those skilled in the art can obtain other related drawings based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating the ocean solitary wave velocity field reconstruction method in Example 1.

[0025] Figure 2 This is a schematic diagram of the stratified fluid density profile of the stratified water flow channel in Example 1;

[0026] Figure 3 This is a schematic diagram of the dimensionless waveform contour in Example 1;

[0027] Figure 4 This is a schematic diagram of the dimensionless horizontal velocity output from the velocity field reconstruction model in Example 1;

[0028] Figure 5 This is a schematic diagram of the dimensionless vertical velocity output from the velocity field reconstruction model in Example 1.

[0029] Figure 6 This is a schematic diagram illustrating the performance of the trained velocity field reconstruction model in Example 1 on the test set, where, Figure 6 (a) is a schematic diagram of the experimental measurement results of the velocity field vector and magnitude distribution of the internal solitary wave. Figure 6 (b) is a schematic diagram of the velocity field reconstruction results. Figure 6 (c) Schematic diagram of the difference distribution between the reconstructed wave-induced velocity field and the actual wave-induced velocity field. Figure 6 (d) is a graph showing the correlation coefficient, MSE, and MAE between the reconstructed velocity field and the experimental flow field in the test set;

[0030] Figure 7 This is a schematic diagram illustrating the performance of the velocity field reconstruction model trained in Example 1 applied to the reconstruction of the isolated wave flow field within the observation. Figure 7 (a) is a schematic diagram of the observation results of the horizontal velocity field of the internal solitary wave from the buoy. Figure 7 (b) is a schematic diagram of the reconstruction results of the velocity field reconstruction model of the horizontal velocity field of the internal solitary wave. Figure 7 (c) is a schematic diagram of the observation results of the vertical velocity field of the internal solitary wave from the buoy. Figure 7 (d) is a schematic diagram of the reconstruction result of the velocity field reconstruction model of the vertical velocity field of the internal solitary wave. Figure 7 (e) is a schematic diagram showing the vertical density distribution results, velocity field reconstruction model, KdV theoretical model, eKdV theoretical model, and the vertical distribution results of horizontal velocity at the valley of the inner isolated wave obtained from the buoy system. Figure 7 (f) is a graph showing the MSE and MAE statistical results between the vertical distribution of horizontal velocity at the valley of the inner isolated wave obtained by the velocity field reconstruction model, the KdV theoretical model and the eKdV theoretical model and the field observation.

[0031] Figure 8This is a structural block diagram of the ocean solitary wave velocity field reconstruction device in Example 2;

[0032] Figure 9 This is a diagram of the internal structure of the computer device in Example 3.

[0033] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0035] It should be noted that in this invention, the use of terms such as "first," "second," etc., is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0036] It is understood that the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0037] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0038] Example 1

[0039] like Figure 1 As shown, this embodiment provides a method for reconstructing the velocity field of isolated waves in the ocean, including the following steps:

[0040] Step 201: Generate an internal isolated wave using the gravity collapse method, and simultaneously measure the waveform profile of the internal isolated wave and the corresponding wave-induced velocity field to construct a dataset.

[0041] Step 202: Construct a flow field reconstruction model, which includes an encoder and a decoder. The encoder includes a hierarchically connected DoubleConv module and a pooling layer, and the decoder includes a hierarchically connected transposed convolutional layer and a DoubleConv module. Each DoubleConv module incorporates a channel attention module and a spatial attention module.

[0042] Step 203: Based on the preset loss function, train the velocity field reconstruction model using the dataset until the model converges, and obtain the trained velocity field reconstruction model.

[0043] Step 204: Input the waveform contour of the isolated wave to be reconstructed into the trained velocity field reconstruction model for processing, and output the velocity field reconstruction result of the isolated wave in the ocean.

[0044] In the specific implementation of step 201, the experimental setup is first constructed. Internal solitary wave gravity collapse wave generators are installed on both sides of a stratified flow channel with geometric dimensions of 1200cm × 40cm × 60cm (length × width × height). The stratified flow channel is also equipped with a stratified water preparation device and a triangular wedge-shaped wave-damping device. The stratified water preparation device includes a water tank with a stirrer and a mushroom-shaped structure located below the channel, used to prepare a fluid environment simulating ocean stratification. The gravity collapse wave generator uses a movable gate with the same width as the channel and a "lock-unlock" method to generate internal solitary waves. The triangular wedge-shaped wave-damping device is installed at the end of the channel to dissipate the propagation energy of the internal solitary waves and prevent reflected waves from interfering with the measurements.

[0045] To facilitate the description of internal solitary waves and the processing of experimental data, a two-dimensional fixed coordinate system for the stratified flow tank is defined as follows: The origin is set at the free water surface at the left end of the tank, defining a two-dimensional motion coordinate system for the internal solitary wave. The origin is set at the center of the waveform trough.

[0046] Subsequently, parameter measurements and data acquisition were performed: a conductivity probe vertically driven by a stepper motor was used to measure the density profile of the stratified fluid to determine the stratification environment parameters, such as... Figure 2 As shown, the vertical axis represents the thickness. The horizontal axis represents density. It can be seen that the fluid thickness ratio between the upper and lower layers varies between 0.2 and 0.4, and the density ratio varies between 0.981 and 0.983 in the stratified environmental parameters; based on the horizontal velocity shear of the flow particles near the internal isolated wave waveform... and the direction angle of flow velocity The feature with the greatest change is obtained by inversion as follows: Figure 3 The diagram shows the contour of an internal isolated wave, where the horizontal axis is... Represents the normalized propagation time history curve; the vertical axis is... Indicates normalized vertical water depth; Indicates vertical depth.

[0047] Simultaneously, a PIV measurement system was used to acquire the wave-induced velocity field corresponding to the waveform profile. This PIV system included 40 μm diameter polystyrene tracer particles, a laser with an output wavelength of 532 nm, and a CCD camera with a frame rate of 25 fps. The CCD camera recorded raw flow field images in 2×60×501 dimensions (horizontal×vertical×spatial sampling points). The internal isolated wave-induced velocity field data were obtained after processing the raw flow field images. In this process, polystyrene tracer particles are fed into the water tank along with stratified water, a laser is suspended above the center of the tank to provide a light source, and a CCD camera captures the raw images of the horizontal and vertical flow fields.

[0048] Next, data preprocessing is performed: the acquired waveform contour lines and wave-induced velocity fields are dimensionless, specifically including the horizontal coordinates of the wave-induced velocity field are converted to linear coordinates. Vertical coordinates According to the preset length benchmark , Dimensionless transformation, the result of dimensionless transformation is as follows Figure 4 and Figure 5 As shown; the waveform outline According to the preset amplitude benchmark Dimensionless; the horizontal velocity of the wave-induced flow field Vertical velocity Maximum gradient value according to waveform horizontal variation Dimensionless transformation ultimately yields a dimensionless waveform profile and a dimensionless wave-induced velocity field.

[0049] Finally, a dataset was constructed: 600 sets of dimensionless waveform contour lines and dimensionless wave-induced velocity field data were collected and divided into training and testing sets in an 8:2 ratio to form an internal isolated wave velocity field reconstruction model dataset for model training.

[0050] This step generates internal solitary waves in a controlled laboratory environment, enabling precise control of stratified fluid parameters and avoiding problems such as uncontrollable ocean current interference and uncertainties in internal solitary wave excitation in real marine environments. Simultaneous measurements directly establish a one-to-one correspondence between waveform contour lines and wave-induced velocity fields, resulting in high data matching and low noise. This provides high-quality and highly reliable basic data for subsequent model training, solving problems such as the difficulty in acquiring traditional field measurement data and the reliance of remote sensing inversion on environmental field information.

[0051] In the specific implementation of step 202, the encoder structure is first constructed: the encoder adopts a hierarchical connection structure, the number of layers is determined according to the requirements, and each layer includes a DoubleConv module and a pooling layer connected in sequence. The DoubleConv module processes the dimensionless waveform contour of the input and outputs the first feature map; the pooling layer processes the first feature map and outputs the second feature map; the processing flow of the DoubleConv module and the pooling layer is repeated, downsampling layer by layer and increasing the number of feature channels to form the hierarchical connection structure of the encoder, and finally outputs the deep abstract feature map.

[0052] The decoder also adopts a hierarchical connection structure. Each layer includes a transposed convolutional layer and a DoubleConv module connected in sequence. The transposed convolutional layer processes the input deep abstract feature map and outputs the third feature map. The DoubleConv module processes the third feature map and outputs the fourth feature map. The processing flow of the transposed convolutional layer and the DoubleConv module is repeated to improve the spatial resolution of the feature map layer by layer, forming the hierarchical connection structure of the decoder, and finally outputting the velocity field reconstruction result.

[0053] The DoubleConv module in both the encoder and decoder has the same structure. The DoubleConv module includes a first convolutional layer, a second convolutional layer, a batch normalization layer, an activation layer, a channel attention layer, and a spatial attention layer, connected sequentially. The first and second convolutional layers perform convolution processing on the input feature map, mapping the number of channels in the input feature map to a preset number of channels. The batch normalization layer performs batch normalization processing on the input feature map, eliminating the gradient vanishing problem caused by differences in feature value distribution. The activation layer performs non-linear activation on the input feature map, enhancing the model's feature representation ability. The channel attention layer performs channel attention weighting processing on the input feature map. The spatial attention layer performs spatial attention weighting processing on the input feature map. Through the channel attention layer and the spatial attention layer, the flow field reconstruction model's ability to focus on key features is improved.

[0054] The velocity field reconstruction model constructed in this step uses an encoder and decoder structure to achieve layer-by-layer extraction and recovery from shallow waveform details to deep abstract features. The channel attention and spatial attention modules integrated in the DoubleConv module can specifically enhance channel features and spatial positions that are strongly correlated with the flow field distribution, effectively improving the model's learning accuracy of the nonlinear mapping relationship between the waveform and the flow field, and solving the problems of inaccurate feature capture and large reconstruction deviation in key areas in traditional models.

[0055] In the specific implementation of step 203, a loss function is first preset. This loss function is a composite loss function, and its expression is:

[0056] ;

[0057] In the formula, , , Indicates the weighting coefficient; This represents the mean squared error loss function; This represents the mean absolute error loss function; This represents the physical information loss function.

[0058] The mean squared error loss function is expressed as follows:

[0059] ;

[0060] The expression for the mean absolute error loss function is:

[0061] ;

[0062] The physical information loss function is expressed as follows:

[0063] ;

[0064] In the formula, This indicates the number of sets of waveform contour lines and corresponding wave-induced velocity fields in the dataset; Represents the normalized i-th A set of real wave-induced velocity fields; Represents the normalized i-th Reconstruct the wave-induced velocity field; Indicates the first Reconstruct the wave-induced velocity field; Indicates the horizontal reconstructed flow rate; Indicates the vertical reconstructed flow velocity; Indicates the first Group theory wave-induced velocity field; Indicates the theoretical horizontal flow velocity; This represents the theoretical vertical velocity. The actual wave-induced velocity field is also represented. To reconstruct the dimensionless wave-induced velocity field obtained through the experimental setup. The output of the velocity field reconstruction model is the theoretical wave-induced velocity field obtained using the internal solitary wave DJL theory.

[0065] In this composite loss function, the mean squared error loss function With mean absolute error loss function Used to measure network performance; physical information loss function This allows for the measurement of the difference between predicted and theoretically calculated data under the same operating conditions from a physical information perspective, thereby guiding the model to focus on key features of the flow field data. This is achieved through weighting coefficients. , , Weighted summation is performed to synthesize the total loss function, thereby balancing the contributions of each loss function in the model optimization process and improving the performance and accuracy of the model in specific application areas.

[0066] The model was then trained: the Adam optimizer was used to adjust the parameters of the velocity field reconstruction model, and a learning rate scheduler was introduced during training to accelerate model convergence; the training set was input into the model in batches for iterative training; during training, the output was normalized, scaling its value range to [-1, 1] to facilitate loss calculation and model optimization; during prediction, the output was denormalized to restore its original physical dimensions. Early stopping was used during training to prevent overfitting; training was stopped when the convergence condition was met, resulting in a converged velocity field reconstruction model. The performance of the converged velocity field reconstruction model was then validated using a test set; successful validation yielded the trained velocity field reconstruction model.

[0067] This step combines data fitting loss with physical theory loss through a composite loss function, ensuring that the model does not deviate from the physical essence of the internal solitary wave flow field while learning the data patterns. By combining the Adam optimizer, learning rate scheduler, and early stopping method, the training efficiency and stability of the model are effectively improved, overfitting is avoided, and the trained model has good generalization ability, laying the foundation for subsequent applications in real-world marine scenarios.

[0068] In the specific implementation of step 204, the waveform contour of the inner solitary wave to be reconstructed is first obtained: the observation data of the inner solitary wave to be reconstructed is obtained, which can come from the temperature observation data of the buoy or the inversion data of the satellite remote sensing image, and the waveform contour of the inner solitary wave to be reconstructed is extracted based on the observation data.

[0069] Then, preprocessing is performed. The dimensionless method used in step 201 when constructing the dataset is used to perform dimensionless preprocessing on the waveform contour lines of the extracted internal solitary waves to be reconstructed, so as to ensure that the input data format is consistent with the input format during model training.

[0070] Next, velocity field prediction is performed. The preprocessed dimensionless waveform contour is input into the trained velocity field reconstruction model. The encoder extracts features and the decoder recovers features, outputting dimensionless horizontal and vertical velocity field prediction values. At the same time, the model can repair the missing velocity field information in the near-surface layer of the observed flow field, improving the completeness of the prediction results.

[0071] Finally, the results are transformed: the dimensionless horizontal and vertical velocity field predictions output by the model are denormalized, and based on the reference in step 201 when the dimensions were made dimensionless, they are restored to physical dimensions with actual physical meaning, so as to obtain the final prediction results of the ocean solitary wave velocity field.

[0072] This step only requires inputting the waveform contour of the internal solitary wave to be reconstructed. The trained model can directly output the velocity field prediction results without the need for real-time environmental field parameters or complex dynamic theoretical models for inversion calculations. This greatly simplifies the velocity field reconstruction process and reduces the dependence on external data and theoretical tools. At the same time, the trained model can be directly transferred to real-world marine scenarios and can output stable and accurate velocity field results for internal solitary waves of different nonlinear intensities.

[0073] In one embodiment, the model's performance is validated. For example, using a set of internal solitary wave velocity field reconstruction results from the test set, such as... Figure 6 (a) to Figure 6 As shown in (c), the direction and magnitude distribution of the velocity vector in the reconstructed flow field of the internal solitary wave are consistent with the actual flow field.

[0074] Figure 6 (d) shows the maximum absolute value of horizontal velocity in the reconstructed flow field of 120 groups in the test set. Maximum absolute value of vertical velocity And the corresponding correlation coefficient (R), MSE and MAE statistical results, as shown in the figure. This represents the dimensionless horizontal velocity magnitude of the internal solitary wave. This represents the dimensionless vertical velocity magnitude of the internal isolated wave. Statistical analysis shows that the correlation coefficients for the horizontal and vertical velocities in the test set are 0.97 and 0.81, respectively, with relatively small errors compared to the true values. The MSEs are 0.014 and 0.020, and the MAEs are 0.083 and 0.099, respectively. These results clearly demonstrate that the velocity field reconstruction model successfully grasps the nonlinear correspondence between the internal isolated wave waveform profile and the wave-induced velocity field, possessing the ability to receive any internal isolated wave waveform profile and output reconstructed horizontal and vertical velocity fields matching that waveform.

[0075] In one embodiment, the trained velocity field reconstruction model is extended to the reconstruction of internal solitary wave velocity fields in the actual ocean to examine the performance of the velocity field reconstruction model trained on laboratory data in real-world ocean applications.

[0076] Internal solitary wave data were obtained from a mooring system located southwest of the Dongsha Islands (117°44.7'E, 20°44.2'N). The potential density profile at the observation point was calculated from the temperature and salinity data of WOA18. Figure 7(e) shows the gray curve. The 16 temperature sensors and 3 ADCPs arranged along the water depth on the mooring system can capture the temperature and velocity information when the internal solitary wave passes by. After interpolating the temperature and velocity information along the water depth, the waveform contour of the internal solitary wave is identified from the observed temperature data, while the wave-induced velocity field is obtained by removing the average velocity of the background current in the 30 minutes before the arrival of the internal solitary wave.

[0077] Dimensionless waveform contours from a set of typical internal solitary wave observation data are taken and input into the velocity field reconstruction model to obtain dimensionless prediction results of horizontal and vertical velocity fields, which are then compared with the actual observation results. Figure 7 (c) The dimensionless reconstructed horizontal flow field can describe Figure 7 (a) Characteristics of the shear distribution of the true horizontal velocity along the waveform profile. Figure 7 (d) The dimensionless prediction of the vertical flow field can describe Figure 7 (b) shows the characteristics of the true vertical velocity distributed in opposite directions along the trough, while the velocity field reconstruction model repairs the missing velocity field information in the near-surface layer of the observed flow field.

[0078] Furthermore, to compare the reconstruction performance of the velocity field reconstruction model and the classical theoretical model on the observed flow field of the internal solitary wave, the observed flow field is taken near the center of the wave trough ( Figure 7 (a) The vertical profile of the horizontal velocity at the location indicated by the red dashed line was calculated using KdV and eKdV theoretical models combined with environmental field information, respectively, to obtain the vertical profile of the horizontal velocity at the same location under the same working conditions. Figure 7 As shown in (e), the trend of horizontal velocity increasing with water depth in the reconstruction results of the velocity field reconstruction model is better than that of the theoretical model reconstruction results. Its error is mainly caused by the influence of complex boundary conditions, complex background shear flow and instrument stability under complex sea conditions in the actual ocean. Data with these interference characteristics are not available in the experimental dataset used by the velocity field reconstruction model.

[0079] Wave amplitude is an important wave parameter characterizing the strength of nonlinearity in internal solitary waves. To verify the performance of the velocity field reconstruction model for reconstructing internal solitary wave current fields with varying nonlinearities in the actual ocean, five sets of observed internal solitary wave waveforms with different amplitudes were taken. The corresponding reconstructed flow fields were obtained using the velocity field reconstruction model (Pre), the KdV model, and the eKdV model, respectively. The MSE and MAE of the vertical profile of the horizontal velocity at the wave trough in the reconstructed flow field were calculated compared with the observed results, as shown below. Figure 7As shown in (f), the velocity field reconstruction model, without the support of environmental field information, significantly outperforms the traditional theoretical model that incorporates environmental field information in its prediction results. For internal solitary waves with different amplitudes, the MSE and MAE of both theoretical models increase with increasing amplitude, indicating that the applicability of the weak nonlinear theory for reconstructing internal solitary wave flow fields weakens as the nonlinearity of the internal solitary wave increases. However, the MSE and MAE of the velocity field reconstruction model change very little with amplitude, indicating that the velocity field reconstruction model has good generalization ability and is suitable for reconstructing internal solitary wave flow fields with different nonlinear strengths.

[0080] The results clearly demonstrate that the velocity field reconstruction model successfully predicted horizontal and vertical flow fields based on waveform contour features from the experimental dataset. It can also be applied to flow field reconstruction in real-world internal solitary wave observations. Its simple architecture and implementation mean that the network can be applied to predicting or reconstructing internal solitary wave flow fields in other future observation scenarios. However, it is worth noting that this data-driven approach relies heavily on the quality of the training data. Poor-quality datasets may contain noise, outliers, or imbalanced distributions. Therefore, it is necessary to enrich the experimental dataset to encompass internal solitary waves and their flow field characteristics in real-world ocean environments to improve the generalization ability and accuracy of the velocity field reconstruction model.

[0081] It should be understood that, although this embodiment Figure 1 The steps are shown sequentially as indicated by the arrows, but they are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order in which these steps are performed; they can be executed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0082] Example 2

[0083] Based on the ocean solitary wave velocity field reconstruction method in Example 1, this example discloses an ocean solitary wave velocity field reconstruction device, such as... Figure 8 As shown, the ocean solitary wave current velocity field reconstruction device includes: a dataset construction module 401, a current velocity field reconstruction model construction module 402, a model training module 403, and a current velocity field reconstruction module 404, wherein:

[0084] The dataset construction module 401 is used to generate internal isolated waves according to the gravity collapse method, and simultaneously measure the waveform profile of the internal isolated waves and the corresponding wave-induced velocity field to construct the dataset.

[0085] The velocity field reconstruction model building module 402 is used to construct the velocity field reconstruction model, which includes an encoder and a decoder. The encoder includes a hierarchically connected DoubleConv module and a pooling layer, and the decoder includes a hierarchically connected transposed convolutional layer and a DoubleConv module. Each DoubleConv module incorporates a channel attention module and a spatial attention module.

[0086] The model training module 403 is used to train the flow field reconstruction model based on a preset loss function and a dataset until the model converges, thus obtaining the trained flow field reconstruction model.

[0087] The velocity field reconstruction module 404 is used to input the waveform contour of the isolated wave to be reconstructed into the trained velocity field reconstruction model for processing, and output the velocity field reconstruction result of the isolated wave in the ocean.

[0088] In this embodiment, the specific working process and working principle of the dataset construction module 401, the flow velocity field reconstruction model construction module 402, the model training module 403, and the flow velocity field reconstruction module 404 are the same as those in Embodiment 1, and therefore will not be described again in this embodiment. Each unit module can be implemented entirely or partially through software, hardware, or a combination thereof. Each unit module can be embedded in or independent of the processor in a computer device in hardware form, or it can be stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each of the above unit modules.

[0089] Example 3

[0090] like Figure 9 The diagram illustrates a terminal device disclosed in this embodiment, comprising a transmitter, a receiver, a memory, and a processor. The transmitter transmits instructions and data, the receiver receives instructions and data, the memory stores computer-executed instructions, and the processor executes the computer-executed instructions stored in the memory to implement the method described in Embodiment 1 above.

[0091] It is important to note that the aforementioned memory can be either standalone or integrated with the processor. When the memory is set up independently, the terminal device also includes a bus for connecting the memory and the processor.

[0092] Example 4

[0093] This embodiment discloses a computer-readable storage medium storing computer-executable instructions. When a processor executes the computer-executable instructions, it implements the method in Embodiment 1 above.

[0094] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0095] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0096] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A method of reconstructing the flow velocity field of a solitary wave in the ocean, characterized in that, The method comprises: According to the gravity collapse method, an internal solitary wave is generated, and the waveform profile line and the corresponding wave-induced flow velocity field of the internal solitary wave are synchronously measured to construct a data set; A flow velocity field reconstruction model is constructed, the flow velocity field reconstruction model comprising an encoder and a decoder; wherein the encoder comprises a DoubleConv module and a pooling layer connected in a hierarchical manner, the decoder comprises a transposed convolution layer and a DoubleConv module connected in a hierarchical manner, and each DoubleConv module incorporates a channel attention module and a spatial attention module; Based on a preset loss function, the flow velocity field reconstruction model is trained through the data set until the model converges, and a trained flow velocity field reconstruction model is obtained; The waveform profile line of an internal solitary wave to be reconstructed is input into the trained flow velocity field reconstruction model for processing, and an ocean internal solitary wave flow velocity field reconstruction result is output. The loss function is a composite loss function, and the expression is: ； In the formula, , , represents a weight coefficient; represents a mean square error loss function; represents a mean absolute error loss function; represents a physical information loss function; The physical information loss function expression is: ； wherein, represents the number of groups of wave profile lines and corresponding wave-induced current fields in the data set; represents the first group of reconstructed wave-induced current fields; represents the horizontal reconstructed current; represents the vertical reconstructed current; represents the first group of theoretical wave-induced current fields; represents the horizontal theoretical current; represents the vertical theoretical current.

2. The method of oceanic internal solitary wave flow field reconstruction according to claim 1, wherein, According to the gravity collapse method, an internal solitary wave is generated, and the waveform profile line and the corresponding wave-induced flow velocity field of the internal solitary wave are synchronously measured to construct a data set, comprising: An internal solitary wave gravity collapse wave generating device is installed in a stratified water channel, and an internal solitary wave is generated by the gravity collapse wave generating device; The density profile of the stratified fluid is measured by using a conductivity probe to determine the stratified environmental parameters; and the internal solitary wave waveform profile line is obtained by inversion based on the stratified environmental parameters; The wave-induced flow velocity field corresponding to the waveform profile line is synchronously obtained by using a PIV measurement system; The waveform profile line and the wave-induced flow velocity field are subjected to dimensionless processing to obtain a dimensionless waveform profile line and a dimensionless wave-induced flow velocity field; A plurality of sets of dimensionless waveform profile lines and dimensionless wave-induced flow velocity fields are collected to construct a data set.

3. The method of claim 2, wherein, In the encoder, the input dimensionless waveform profile line is processed by the DoubleConv module to output a first feature map; and the first feature map is processed by the pooling layer to output a second feature map; The processing flow of the DoubleConv module and the pooling layer is repeated, the feature channel number is gradually down-sampled and increased, and a hierarchical connection structure of the encoder is formed, and finally a deep abstract feature map is output.

4. The method of claim 3, wherein, In the decoder, the input deep abstract feature map is processed by the transposed convolution layer to output a third feature map; and the third feature map is processed by the DoubleConv module to output a fourth feature map; The processing flow of the transposed convolution layer and the DoubleConv module is repeated, the spatial resolution of the feature map is gradually increased, a hierarchical connection structure of the decoder is formed, and finally a flow velocity field reconstruction result is output.

5. The method of claim 1 to 4, wherein, The DoubleConv module comprises a first convolution layer, a second convolution layer, a batch normalization layer, an activation layer, a channel attention layer and a spatial attention layer connected in sequence; wherein the first convolution layer and the second convolution layer perform convolution processing on the input feature map; the batch normalization layer performs batch normalization processing on the input feature map; the activation layer performs nonlinear activation on the input feature map; the channel attention layer performs channel attention weighting processing on the input feature map; and the spatial attention layer performs spatial attention weighting processing on the input feature map.

6. An apparatus for reconstructing the flow velocity field of a solitary wave in the ocean, characterized by The device comprises: a dataset construction module configured to generate an internal solitary wave according to a gravity collapse method, synchronously measure a waveform profile line of the internal solitary wave and a corresponding wave-induced flow velocity field, and construct a dataset; a flow velocity field reconstruction model construction module configured to construct a flow velocity field reconstruction model, wherein the flow velocity field reconstruction model comprises an encoder and a decoder; the encoder comprises DoubleConv modules and pooling layers connected in levels, and the decoder comprises transposed convolution layers and DoubleConv modules connected in levels, and each DoubleConv module is integrated with a channel attention module and a spatial attention module; a model training module configured to train the flow velocity field reconstruction model based on a preset loss function through the dataset until the model converges, so as to obtain a trained flow velocity field reconstruction model; a flow velocity field reconstruction module configured to input a waveform profile line of an internal solitary wave to be reconstructed into the trained flow velocity field reconstruction model for processing, and output a marine internal solitary wave flow velocity field reconstruction result; in the model training module, the loss function is a composite loss function, and the expression is: ； In the formula, , , represents a weight coefficient; represents a mean square error loss function; represents a mean absolute error loss function; represents a physical information loss function; the expression of the physical information loss function is: ； wherein, represents the number of groups of wave profile lines and corresponding wave-induced current fields in the data set; represents the first group of reconstructed wave-induced current fields; represents the horizontal reconstructed current; represents the vertical reconstructed current; represents the first group of theoretical wave-induced current fields; represents the horizontal theoretical current; represents the vertical theoretical current. 7.A computer device, comprising a memory and a processor, and characterized in that, The memory stores a computer program, and the processor implements the steps of the marine internal solitary wave flow velocity field reconstruction method in any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium, characterized in that, A computer program is stored thereon, and the computer program implements the steps of the marine internal solitary wave flow velocity field reconstruction method in any one of claims 1 to 5 when executed by a processor.

Images