Ocean current velocity global spatio-temporal field reconstruction method and device based on underway ADCP

Abstract

The application discloses a kind of ocean current velocity global space-time field reconstruction method and device based on walk navigation ADCP, method includes: determining target sea area, constructs global space-time grid, obtains each grid node;According to walk navigation ADCP, the observation data of target sea area is preprocessed, the multidimensional feature set of the observation point corresponding to each observation data after preprocessing is set, and each observation point is mapped to global space-time grid, based on the multidimensional feature set of each observation point, grid node feature is set, and grid node feature matrix is obtained;Based on the time dimension of grid node feature matrix, dynamic adjacency matrix is constructed;Grid node feature matrix and dynamic adjacency matrix are input into preset space-time feature extraction network, and space-time representation feature matrix is obtained;Space-time representation feature matrix, auxiliary environmental data are input into preset regionalization physical constraint model, based on nearshore area physical constraint and ocean area physical constraint, flow velocity field reconstruction is carried out, and global reconstruction flow velocity field is obtained.

Description

Technical Field

[0001] This invention relates to the field of marine dynamic environment observation and intelligent data reconstruction technology, specifically to a method and device for reconstructing the global spatiotemporal field of ocean current velocity based on a mobile ADCP. Background Technology

[0002] Global Ocean Surface Currents (GOSC) are a core element of the marine dynamic environment, playing a crucial role in areas such as heat transfer, material transport, ecosystem evolution, and marine pollution tracking. Accurately acquiring continuous, spatiotemporal ocean current fields across the entire ocean is of great significance for understanding ocean dynamic mechanisms, ensuring maritime safety, and carrying out marine environmental protection.

[0003] Currently, ocean current velocity observation methods mainly include satellite remote sensing inversion, buoy observation, and mobile ADCP (Acoustic Doppler Current Profiler) observation. Among these, mobile ADCP has become an indispensable core device in marine surveys due to its ability to acquire vertical velocity profile data and its high observation efficiency. However, due to the limitations of a single vessel's track, mobile ADCP observation data has inherent defects such as "spatiotemporal sparsity, asynchronous sampling, and discrete profiles," and can only obtain local velocity profiles along the track, unable to directly provide a continuous spatiotemporal field of ocean current velocity across the entire region.

[0004] In existing technologies, methods for reconstructing the global spatiotemporal field of ocean current velocity can be mainly divided into three categories: Traditional numerical modeling methods construct global or regional ocean models based on hydrodynamic equations and output velocity fields by assimilating limited observation data. However, these methods rely on high-precision boundary conditions and initial fields, have high computational costs, poor real-time performance, and limited simulation accuracy in nearshore complex terrain regions. Purely data-driven deep learning methods utilize neural networks to fit the spatiotemporal correlations of observed data to reconstruct the velocity field. For example, the existing PhyGOSCR-Net model uses geostrophic currents, Ekman currents, and Stokes drift as inputs, combined with Argo buoy and drifting buoy data as labels, to reconstruct the surface velocity field. However, this type of method has the following shortcomings: 1. It does not address the vertical characteristics of ADCP profile data, only utilizing surface velocity, thus wasting the vertical dynamic information contained in the profile; 2. It lacks targeted processing for sparse asynchronous ADCP track data, and directly using general grid interpolation easily introduces spurious errors; 3. Physical constraints only introduce general fluid equations, without considering local dynamic characteristics such as topography and tides in the ADCP observation area, resulting in low physical consistency and low matching degree with actual sea conditions. Simple interpolation completion method: This method completes the sparse observation points of ADCP using traditional spatial interpolation methods such as Kriging interpolation and inverse distance weighting. However, this type of method ignores the spatiotemporal continuity and physical correlation of ocean current velocities, and the reconstruction results are prone to local anomalies, failing to reflect the true ocean dynamic processes. Summary of the Invention

[0005] In view of the above problems, embodiments of the present invention are proposed to provide a method and apparatus for global spatiotemporal field reconstruction of ocean current velocity based on a mobile ADCP to overcome or at least partially solve the above problems.

[0006] According to one aspect of the present invention, a method for reconstructing the global spatiotemporal field of ocean current velocity based on a mobile ADCP is provided, the method comprising: The target sea area is determined, a global spatiotemporal grid is constructed, and each grid node is obtained; the global spatiotemporal grid is a four-dimensional domain including longitude, latitude, depth, and time series. Based on the observation data of the target sea area obtained by the mobile ADCP, the observation data is preprocessed, a multi-dimensional feature set of the observation points corresponding to each preprocessed observation data is set, and each observation point is mapped to each grid node in the global spatiotemporal grid based on coordinates. Based on the multi-dimensional feature set of each observation point, the grid node features are set to obtain the grid node feature matrix. Based on the time dimension of the mapped grid node feature matrix, a dynamic adjacency matrix is ​​constructed, where each matrix element represents the association weight between grid nodes. The grid node feature matrix and dynamic adjacency matrix are input into a pre-defined spatiotemporal feature extraction network to obtain a spatiotemporal representation feature matrix. The spatiotemporal representation feature matrix and auxiliary environmental data are then input into a pre-defined regionalized physical constraint model. Based on nearshore and oceanic physical constraints, the current velocity field is reconstructed to obtain a globally reconstructed current velocity field. The nearshore and oceanic regions are determined based on topographic depth. The pre-defined spatiotemporal feature extraction network and the pre-defined regionalized physical constraint model are trained and optimized based on a constructed multi-loss function.

[0007] According to another aspect of the present invention, a global spatiotemporal field reconstruction device for ocean current velocity based on a mobile ADCP is provided, comprising: The initial grid module is suitable for determining the target sea area, constructing a global spatiotemporal grid, and obtaining each grid node; the global spatiotemporal grid is a four-dimensional domain including longitude, latitude, depth, and time series; The grid matrix module is suitable for acquiring observation data of the target sea area based on the mobile ADCP, preprocessing the observation data, setting the multi-dimensional feature set of the observation points corresponding to each preprocessed observation data, mapping each observation point to each grid node in the global spatiotemporal grid based on coordinates, setting grid node features based on the multi-dimensional feature set of each observation point, and obtaining the grid node feature matrix. The dynamic adjacency matrix module is suitable for constructing dynamic adjacency matrices based on the time dimension of the mapped grid node feature matrix. Each matrix element in the dynamic adjacency matrix represents the association weight between grid nodes. The reconstruction module is suitable for inputting the grid node feature matrix and dynamic adjacency matrix into a preset spatiotemporal feature extraction network to obtain a spatiotemporal representation feature matrix; and for reconstructing the current velocity field based on the spatiotemporal representation feature matrix and auxiliary environmental data into a preset regionalized physical constraint model, based on the physical constraints of the nearshore area and the ocean area, to obtain a global reconstructed current velocity field; wherein, the nearshore area and the ocean area are determined according to the terrain depth; the preset spatiotemporal feature extraction network and the preset regionalized physical constraint model are trained and optimized based on the constructed multiple loss functions.

[0008] According to another aspect of the present invention, a computing device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus; The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the above-described method for reconstructing the global spatiotemporal field of ocean current velocity based on a mobile ADCP.

[0009] According to another aspect of the present invention, a computer storage medium is provided, wherein at least one executable instruction is stored therein, the executable instruction causing a processor to perform operations corresponding to the above-described method for reconstructing the global spatiotemporal field of ocean current velocity based on a mobile ADCP.

[0010] According to another aspect of the present invention, a computer program product is provided, including at least one executable instruction that causes a processor to perform operations corresponding to the above-described method for reconstructing the global spatiotemporal field of ocean current velocity based on a mobile ADCP.

[0011] The ocean current velocity global spatiotemporal field reconstruction method and apparatus based on mobile ADCP provided in this invention fully exploits the vertical physical correlation of current velocity profiles in mobile ADCP, integrates profile layering features into global reconstruction, improves reconstruction accuracy, adapts to the spatiotemporal characteristics of sparse asynchronous ADCP track data, solves the reconstruction discontinuity problem caused by spatiotemporal sparsity, constructs regional adaptive physical constraints, and combines auxiliary environmental data of the target sea area to ensure the physical consistency of reconstruction results and adaptability to actual sea conditions, thereby achieving high-precision reconstruction of ocean current velocity global spatiotemporal field from mobile ADCP sparse asynchronous observation data to high resolution, physical consistency, and spatiotemporal continuity.

[0012] The above description is merely an overview of the technical solutions of the embodiments of the present invention. In order to better understand the technical means of the embodiments of the present invention and to implement them in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the embodiments of the present invention more obvious and understandable, specific implementation methods of the embodiments of the present invention are described below. Attached Figure Description

[0013] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the embodiments of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart of a method for reconstructing the global spatiotemporal field of ocean current velocity based on a mobile ADCP according to an embodiment of the present invention is shown. Figure 2 A schematic diagram of the pre-defined spatiotemporal feature extraction network structure is shown; Figure 3 The flowcharts illustrating the training and reconstruction of the global spatiotemporal field for each model are shown. Figure 4 A schematic diagram of the model update mechanism is shown; Figure 5 A schematic diagram of a global spatiotemporal field reconstruction device for ocean current velocity based on a mobile ADCP according to an embodiment of the present invention is shown. Figure 6 A schematic diagram of the structure of a computing device according to an embodiment of the present invention is shown. Detailed Implementation

[0014] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0015] Figure 1 A flowchart of a method for reconstructing the global spatiotemporal field of ocean current velocity based on a mobile ADCP according to an embodiment of the present invention is shown, as follows: Figure 1 As shown, the method includes the following steps: Step S101: Determine the target sea area, construct a global spatiotemporal grid, and obtain each grid node.

[0016] Existing technologies do not consider vertical features during processing. This embodiment uses a global spatiotemporal grid, a four-dimensional domain including longitude, latitude, depth, and time series, to physically correlate the vertical current velocity of the underway ADCP profile and integrate the profile's layered features into the global reconstruction. Specifically, the boundary of the target sea area is determined, resulting in the longitude, latitude, and vertical depth intervals, and the time series interval is determined, thus obtaining a four-dimensional global spatiotemporal grid. The two-dimensional planar spatial domain of the target sea area is... Defined by latitude and longitude boundaries Where x is longitude and y is latitude, x min and x max Determine the longitude interval based on the minimum and maximum longitude values ​​corresponding to the target sea area, y min and y max The minimum and maximum latitude values ​​corresponding to the target sea area are used to determine the latitude range. The vertical depth domain is... z is the depth, z max This represents the maximum value in the vertical depth range, which is the maximum vertical observation depth of the underway ADCP. The time series domain is... t is time, t start and t end A continuous time series interval is constructed, with the starting and ending points as the time references. The global spatiotemporal grid is... This represents the entire range of the reconstructed ocean current field. Here, the topographic depth matrix h(x, y) of the target sea area can also be obtained based on ocean topographic data such as ETOPO1.

[0017] The global spatiotemporal grid D is discretized into a regular spatiotemporal depth grid. This discretization is based on a preset longitude and latitude resolution, the depth cells of the mobile ADCP, and a preset time resolution, resulting in each grid node. The preset longitude and latitude resolution includes... △x is the preset longitude resolution, which can be as low as 0.05; △y is the preset latitude resolution, which can also be as low as 0.05. Spatial grid node set. Where P is the number of grid nodes after discretization along the longitude direction, and Q is the number of grid nodes after discretization along the latitude direction. For longitude direction index, For latitude direction index, The vertical resolution is based on the depth cells of the mobile ADCP. Determine the set of vertical grid nodes. (Same as the depth cell bin in ADCP), K is the number of depth layers discretized in the vertical depth direction. Indexed in the depth direction. Preset time resolution. The value is, for example, 1 hour / step, which represents the set of time grid nodes. N is the number of time points. Indexed by time nodes. The set of global grid nodes after discretization is: , which is the target mesh for reconstruction.

[0018] Step S102: Obtain observation data of the target sea area based on the mobile ADCP, preprocess the observation data, set the multi-dimensional feature set of the observation points corresponding to each preprocessed observation data, map each observation point to each grid node in the global spatiotemporal grid based on coordinates, set the grid node features based on the multi-dimensional feature set of each observation point, and obtain the grid node feature matrix.

[0019] The observation data from the underway ADCP is a sparse, asynchronous, discrete point set. By acquiring observation data from the target sea area, the original observation point set of the ADCP is obtained. ,in This represents the effective number of observation points, and due to sparsity, the number of observation points is far less than the set of nodes in the discretized global grid. ,Right now Each observation point Corresponding to a unique spatiotemporal depth coordinate , It is a global spatiotemporal grid, and Due to its asynchronous nature, the spatiotemporal sampling interval is irregular, and there is no fixed sampling pattern. Each observation point... Corresponding observed velocity vector ,in These represent the latitudinal, meridional, and vertical velocity components, respectively. Each observation point also has a data quality weight, ranging from [0,1], calculated based on the ADCP echo intensity and correlation, characterizing the reliability of the observation data; 0 represents invalid data, and 1 represents high-quality data. Due to the surface density of the observation points in the spatial domain... , For a two-dimensional planar spatial domain, The area of ​​the spatial domain is less than the minimum observation density threshold required for continuous reconstruction of the entire domain, which makes it impossible for traditional interpolation and fitting methods to achieve effective reconstruction at this density.

[0020] Based on the above description, this embodiment acquires observational data and auxiliary environmental data of the target sea area using a mobile ADCP based on preset mobile ADCP parameters. The preset mobile ADCP parameters include parameters such as sampling frequency, vertical depth cell bin size, and transducer draft. The acquired observational data includes spatiotemporal data, current velocity data, mass characteristic data, and auxiliary observational data; the spatiotemporal data includes observation timestamps. Longitude and latitude and transducer depth The velocity data includes the zonal velocity components of each depth bin. Meridional velocity component and vertical velocity component Three-dimensional flow velocity Let i and k represent the data corresponding to the i-th observation point at the k-th depth layer. The quality characteristic data includes the echo intensity of each depth cell. Correlation data and quality marking Auxiliary observation data includes ship speed. and heading data Auxiliary environmental data includes tidal data and wind field data (such as obtaining the zonal component of the wind field at a height of 10m). Meridional component (These parameters can be obtained based on satellite wind fields, etc.) and seawater physical parameters (such as seawater density). Gravitational acceleration g, Coriolis parameter Coriolis parameter meridional gradient etc., Coriolis parameters Coriolis parameter meridional gradient (It can be obtained based on existing geophysical fluid dynamics and physical oceanography formulas); tidal data includes tidal phases. and high tide For example, data can be obtained from tidal models such as TPXO9.

[0021] After obtaining the above data, the speed of the boat can be used as a basis. and heading data The flow velocity data is corrected for hull motion to obtain the corrected flow velocity data:

[0022] in, To obtain the raw flow rate data of ADCP, This is the corrected flow velocity data.

[0023] Furthermore, for each acquired observation data point, those meeting preset rejection criteria are removed to obtain preprocessed observation data. Preset rejection criteria include, for example, marking a data point as invalid (…). (Observational data), echo intensity is below the echo intensity threshold ,like =50, or the correlation data is below the correlation data threshold. ,like =0.7; Outliers in the velocity data at each depth layer are determined and removed based on the 3σ criterion (e.g., for the velocity data at each depth layer, determine the mean and standard deviation σ, and take three times the standard deviation σ above and below the mean as the normal velocity range, such as [mean value - 3σ, mean value + 3σ]. Values ​​outside the normal velocity range are considered outliers in the velocity data, etc.) Through the above preprocessing steps such as correction and elimination, a multi-dimensional feature set is constructed for each observation point corresponding to the preprocessed observation data. The multi-dimensional feature set includes vertical profile features. Planar spatiotemporal characteristics and quality perception features .

[0024] Vertical profile features This includes corrected velocity data, vertical gradient, and vertical statistical characteristics; the vertical gradient is determined by the ratio of the difference between the corrected zonal velocity components of adjacent depth cells at the observation point to the depth cell value, such as the vertical gradient. ; The data represents the zonal current velocity at the k-th depth layer after correction for the i-th observation point. This represents the zonal velocity data corrected at the (k-1)th depth layer for the i-th observation point. Vertical statistical characteristics include the mean and variance of the corrected velocity data for each depth cell at the observation point.

[0025] Planar spatiotemporal characteristics Including the coordinate normalization values ​​of the spatiotemporal data of the observation point, tidal phase Wind field data ( ).

[0026] Quality perception characteristics Including quality weight The mass weight is based on the echo intensity at the observation point. Maximum echo intensity of the depth cell where the observation point is located The ratio and correlation data of observation points Confirmed, such as ; This represents the maximum echo intensity at each valid observation point within that depth cell. 0.6 and 0.4 are illustrative examples; adjustments can be made based on implementation conditions, and are not limited here.

[0027] After the above preprocessing steps, the observation points corresponding to each preprocessed observation data are mapped to grid nodes in the global spatiotemporal grid according to the coordinates of the spatiotemporal data. Grid node features are set based on the multi-dimensional feature set of each observation point, resulting in a grid node feature matrix. During the mapping according to the coordinates of the spatiotemporal data, it is necessary to map each valid ADCP observation point... Mapped to the corresponding grid nodes in the global spatiotemporal grid Specifically, the coordinates of the grid nodes in the global spatiotemporal grid are compared with the spatiotemporal data of the corresponding observation points in the preprocessed observation data. If they do not match, meaning the grid node has no corresponding observation point, the grid node feature is set to 0. If they match, it is necessary to further determine whether there are multiple matching grid nodes. If so, it means that the grid node corresponds to multiple observation points, and the quality weights of the multi-dimensional feature sets of the multiple matching grid nodes need to be adjusted. The weighted average is then set as the grid node feature, such as the final grid node feature. ,in, To map to this grid node A set of multiple observation points. If a grid node corresponds to an observation point, the multi-dimensional feature set of the matched grid node is set as the grid node feature.

[0028] Step S103: Construct a dynamic adjacency matrix based on the time dimension of the mapped grid node feature matrix.

[0029] Based on the time dimension of the grid node feature matrix, determine each time step Spatial grid Determine the relationship between any two grid nodes and Spatial weights and temporal weights.

[0030] Spatial and temporal weights are fused based on a preset weight ratio to obtain matrix elements representing the association weights between grid nodes. A dynamic adjacency matrix corresponding to the nth time step is then constructed based on these matrix elements. Where n represents the time step index, P represents the number of grid nodes discretized along the longitude direction, Q represents the number of grid nodes discretized along the latitude direction, and P×Q represents the total number of grid nodes in the horizontal spatial domain of the target sea area. Dynamic Adjacency Matrix Each row and each column corresponds to a grid node in the horizontal spatial domain, and the matrix elements... Represents the grid node at the nth time step. and grid nodes The correlation weights between them are given, where p1 and p2 both range from 1 to P, and q1 and q2 both range from 1 to Q.

[0031] The spatial weights are shown below:

[0032] Where d is the Euclidean distance function, which calculates the distance between any two grid nodes. and The distance between, This is the spatial bandwidth parameter, with a value such as 0.25.

[0033] The time weights are as follows:

[0034] in, This is the time bandwidth parameter, with a value such as 5Δt, which is 5 times the preset time resolution. The t function is used to obtain the timestamp of the nearest observation point for the corresponding grid node. For grid nodes The timestamp of the most recent observation point For grid nodes The timestamp of the most recent observation point.

[0035] Spatial and temporal weights are fused based on a preset weight ratio to obtain matrix elements representing the association weights between grid nodes. A dynamic adjacency matrix corresponding to the nth time step is then constructed based on these matrix elements. As shown below:

[0036] The above 0.7 and 0.3 are preset weight ratios, which are for illustrative purposes only and can be adjusted according to the implementation situation. No limit is set here.

[0037] Step S104: Input the grid node feature matrix and dynamic adjacency matrix into the preset spatiotemporal feature extraction network to obtain the spatiotemporal representation feature matrix; and input the spatiotemporal representation feature matrix and auxiliary environmental data into the preset regionalized physical constraint model to reconstruct the velocity field based on the physical constraints of the nearshore area and the physical constraints of the ocean area to obtain the global reconstructed velocity field.

[0038] The obtained dynamic adjacency matrix and at the nth time step and the kth depth layer The input matrix determined by the feature matrix of the horizontal grid nodes. The inputs are fed into a pre-defined spatiotemporal feature extraction network. The depth coordinates corresponding to the k-th vertical depth layer are represented by k, where k represents the vertical depth layer index and the value of k is 1-K; K represents the number of depth layers after discretization in the vertical depth direction; C represents the input feature dimension of each grid node at the corresponding time step and the corresponding depth layer. The input features include at least one of the corrected flow velocity data, vertical gradient, vertical statistical features, planar spatiotemporal features, and quality-aware features.

[0039] The pre-defined spatiotemporal feature extraction network can be a network such as Graph WaveNet++, specifically as follows: Figure 2 As shown, the input matrix ( Figure 2 The input features (feature tensor) and dynamic adjacency matrix are processed by graph convolutional layers for spatial neighborhood association, followed by dilated causal convolutional layers for multi-scale temporal dependency extraction, vertical feature fusion layers for feature fusion at different depths, and finally a quality attention layer for suppressing low-quality observation features, resulting in a spatiotemporal representation feature matrix. Where P×Q represents the total number of grid nodes in the horizontal spatial domain of the target sea area, and K represents the number of depth layers after discretization in the vertical depth direction. To hide the feature dimension, the spatiotemporal representation feature matrix can be used for subsequent velocity field prediction and physical residual calculation in a predefined regionalized physical constraint model.

[0040] When reconstructing the velocity field, the nearshore and ocean regions are first divided based on topographic depth. For example, based on the topographic depth matrix h(x, y), and using 50m as the dividing condition, the spatial domain Ω is divided into the nearshore and ocean regions. The nearshore region... Oceania Generate a region mask M(x,y), where M=1 for the nearshore region and M=0 for the ocean region, corresponding to the spatial grid node set. Divided into discrete grid nearshore areas Discrete grid ocean region .

[0041] The spatiotemporal representation feature matrix and auxiliary environmental data are input into the preset regional physical constraint model, and physical constraints in the nearshore area and the ocean area are constructed to reconstruct the horizontal and vertical current velocities in the nearshore and ocean areas, and then verify and adjust them to obtain the global reconstructed velocity field.

[0042] Specifically, the spatiotemporal representation feature matrix, regional mask, and auxiliary environmental data such as topographic water depth information, tidal data, and wind field data are input into the preset regionalized physical constraint model. The preset regionalized physical constraint model can adopt a network such as Regional-PINN to predict the horizontal and vertical velocity components on the global grid, thereby constructing the regionalized physical residuals of the nearshore area and the ocean area to obtain the final global reconstructed velocity field.

[0043] The nearshore region introduces physical equations that include shallow water dynamic constraints and bottom friction constraints. The physical constraints in the nearshore region are constructed based on the shallow water equations and bottom friction, and for any effective spatiotemporal depth grid node... To reconstruct the horizontal component of the flow velocity , and The reconstructed velocity components for longitude and latitude in the two horizontal directions are shown below:

[0044] in, The spatial domain is defined as the nearshore area, Z represents the vertical depth domain, and T represents the time domain. The term representing the local time variation of horizontal flow velocity, The nonlinear advection term representing the horizontal velocity. This represents the plane gradient operator, where g is the acceleration due to gravity. This indicates the sea level anomaly obtained from the reconstruction. This indicates the external water level correction term, which is based on the tide height. Determined by water level observation data or topographic boundary conditions; These represent viscosity-related parameters, used to characterize velocity diffusion or viscous dissipation in nearshore areas. denoted as the bottom friction coefficient, which is a regional physical parameter learned by the Regional-PINN network; This represents the magnitude of the horizontal velocity vector. Indicates the bottom friction term; Indicates wind stress, , Indicates the wind stress drag coefficient. Indicates air density; Represents the near-sea surface wind speed vector; Indicates wind speed modulus; Indicates the density of reference seawater; This represents the physical residual vector in the nearshore region, used to measure the degree to which the reconstructed velocity field satisfies the nearshore shallow water dynamic constraints and bottom friction constraints.

[0045] When satisfied At that time, it can be determined that the nearshore area reconstruction results meet the given physical consistency requirements. This indicates the preset physical residual threshold for the nearshore area.

[0046] A depth-averaged momentum constraint equation is introduced for the ocean region, and the physical constraints for the ocean region are constructed based on the two-dimensional depth-averaged momentum equation. For any valid spatiotemporal depth grid node... To reconstruct the horizontal component of the flow velocity , and The reconstructed velocity components for longitude and latitude in the two horizontal directions are shown below:

[0047] in, For the ocean region; Coriolis parameters, Represents a vertical unit vector. represents the wind stress coefficient, which is a parameter learned by the Regional-PINN network; s is an unmodeled term, which is a parameter learned by the Regional-PINN network. This represents the physical residual vector for the ocean region. For other parameters, please refer to the corresponding parameter descriptions in the physical constraint formulas for the nearshore region.

[0048] When satisfied At that time, it can be determined that the reconstruction results of the ocean region meet the given physical consistency requirements. This indicates the preset physical residual threshold for the ocean region.

[0049] In addition to the above reconstruction of the horizontal velocity components in nearshore and ocean regions, the regionalized physical constraint model also includes incompressible continuity constraints applicable to both nearshore and ocean regions. For any valid spatiotemporal depth grid node, the reconstructed three-dimensional velocity components... To satisfy:

[0050] in, and These represent the reconstructed meridional, zonal, and vertical velocity components, respectively. Continuity constraints are used to ensure the flux continuity of the reconstructed velocity field in the horizontal and vertical directions, and to guarantee that the reconstructed velocity field in the nearshore and open ocean regions satisfies the incompressibility of the water body. For the nearshore region, the reconstruction results can be further modified by incorporating topographic depth boundaries and bottom friction conditions.

[0051] The above-mentioned pre-set spatiotemporal feature extraction network and pre-set regionalized physical constraint model are trained and optimized based on the constructed multiple loss functions.

[0052] Multiple loss functions include data loss term, physical loss term, and regularization loss term.

[0053] The data loss term represents the observation fitting error, as shown below:

[0054] in, This indicates that the reconstructed velocity field is achieved through interpolation operators. At the observation point The obtained three-dimensional predicted velocity components correspond to longitude, latitude, and vertical velocity components, respectively. Interpolation operator. First, distance weights are constructed based on the spatial, depth, and temporal distances between the observation point and adjacent grid nodes. Then, the reconstructed velocity components of adjacent grid nodes are weighted and summed based on these distance weights. The distance weights are determined according to a preset distance attenuation function, the parameters of which are determined based on the preset latitude and longitude resolution of the global spatiotemporal grid, the depth cells of the mobile ADCP, and the preset temporal resolution. When the observation point coincides with a grid node, the interpolation operator takes the value of the reconstructed velocity component of the corresponding grid node. The vertical velocity loss weight is set to 0.1. For quality weights.

[0055] The physical loss term represents the deviation from physical laws, as shown below:

[0056] in, It is the weight of regional physical loss, and the specific value can be adjusted according to the implementation situation; It represents the physical residual vector on the discrete grid nodes.

[0057] The regularization loss term represents the smoothness of the model and its parameters, as shown below:

[0058] in, This represents a regionalized physical parameter vector, used to characterize the local dynamic features corresponding to its location and time. The subscript... Indicates the corresponding position and time in the longitude and latitude directions, respectively corresponding to the position in the _______th ... The grid node and the first At each time step. The other subscripts correspond to the position and time along the longitude and latitude directions indicated by the subscripts. For the nearshore area, This can include viscosity-related parameters, etc. Bottom friction coefficient Parameters related to wind stress, etc.; for the ocean region, This can include viscosity-related parameters, etc. Wind stress coefficient Unmodeled items, etc. This represents the learnable network weight parameters in the model. Here, the model includes a pre-defined spatiotemporal feature extraction network and a pre-defined regionalized physical constraint model. The learnable network weight parameters in the model include the weight parameters in the pre-defined spatiotemporal feature extraction network. and the weight parameters in the preset regionalized physical constraint model . Used to extract spatiotemporal representation feature matrices from dynamic spatiotemporal graphs and multidimensional feature data. It is used to generate a reconstructed velocity field that satisfies regional physical constraints based on the spatiotemporal representation feature matrix, region mask, and auxiliary environmental data. The regularization term is used to regularize the learnable network weights to suppress model overfitting. The regularization term for each regional physical parameter vector is used to constrain the continuous changes of physical parameters in space and time, avoiding unreasonable abrupt changes in physical parameters between adjacent regions or adjacent time steps. and These represent the physical parameter smoothing regularization coefficient and the model weight regularization coefficient, respectively.

[0059] Based on the above multiple loss functions, the joint loss function is obtained as follows:

[0060] in, and This is the weighting coefficient for the loss item, and its value can be, for example, 0.3 or 0.1. It can also be adjusted according to the implementation situation, and there is no limitation here.

[0061] Based on the joint loss function above, the data can be divided into training and validation sets in an 8:2 ratio. The model is implemented using a framework such as PyTorch, with the AdamW optimizer (initial learning rate 0.001), batch size of 8, training epochs of 100, and the CosineAnnealingLR learning rate scheduler. An early stopping mechanism is triggered to save the optimal model. The optimization objective is:

[0062] in, For a multi-source input set, This indicates observational data acquired based on ADCP. This represents auxiliary environment data. It represents marine topographic data, etc. This represents the raw flow rate data. Represents a regionalized physical parameter vector; This represents the learnable network weight parameters in the model. This is the optimal global reconstructed velocity field obtained after training based on historical observation data.

[0063] After training the pre-set spatiotemporal feature extraction network and the pre-set regionalized physical constraint model, the global spatiotemporal grid of the target sea area and auxiliary environmental information are used as input information. Through dynamic adjacency matrix construction, Graph WaveNet++ feature encoding, and Regional-PINN physical constraint optimization, the predicted current velocity is obtained. After obtaining the predicted current velocity, further spatiotemporal smoothing is performed using a 3×3×3 Gaussian filter kernel, and the sea area boundary nodes are corrected using a mirror symmetry method, followed by physical consistency verification. If the nearshore physical residual vector... satisfy Or the physical residual vector of the ocean region satisfy Then, the physical parameters of the corresponding region are locally fine-tuned and recalculated to finally obtain the global reconstructed velocity field.

[0064] Furthermore, existing technologies generally lack online adaptive update capabilities, making it impossible to interface with real-time observation data from mobile ADCP systems and thus difficult to meet the needs of real-time maritime operations (such as emergency pollution tracking and dynamic navigation planning). This embodiment acquires newly added observation data from the mobile ADCP system, obtains a new grid node feature matrix based on the new observation data, and updates and optimizes the weight parameters of the preset regionalized physical constraint model based on the new grid node feature matrix. Specifically, for the newly added real-time ADCP observation point set... , Let j be the index of the newly added observation point. This represents the number of newly added valid observation points. The newly added observation data is... . Let J be the three-dimensional measured velocity vector of the j-th newly added observation point. Assign a data quality weight to the j-th newly added observation point. Following steps S101-S102, generate a real-time multi-dimensional feature set to obtain the feature matrix of the newly added grid nodes. Perform a lightweight online update on the preset spatiotemporal feature extraction network and the preset regionalized physical constraint model. For example, define the updated input set as... , For the newly added observation dataset, the updated time window is: , This defines the time window corresponding to the newly added observation data. A base layer fixation and adaptation layer update strategy is adopted, fixing the weight parameters of the core feature extraction layer of the preset spatiotemporal feature extraction network Graph WaveNet++. Only update the weight parameters of the regional adaptation layer of the preset regionalized physical constraint model Regional-PINN. and regionalized physical parameter vector The learning rate can be set to, for example, 0.0001, the number of training epochs to, for example, 10 epochs, and the optimization objective is:

[0065] in, This represents the reconstructed velocity field obtained based on newly added observation data. This represents the weight parameters of the updated preset regionalized physical constraint model. This represents the updated regionalized physical parameter vector. Each updated... , , The value of the joint loss function that satisfies the minimum requirement, and For a fixed constant The value of const is set according to the implementation situation and is not limited here.

[0066] After updating the weight parameters, the time window containing newly added observation data can be applied based on the updated preset regionalized physical constraint model. Corresponding grid nodes The flow velocity field is reconstructed, in which... To determine the time window based on newly added observation data The newly obtained time grid node set of observation data is used to reconstruct the velocity field and replace the original reconstruction results, ensuring the accuracy of the velocity field reconstruction across the entire domain.

[0067] In an optional embodiment, for the nearshore area, a certain latitude and longitude boundary is as follows: For the target sea area, vertical depth domain Time series domain The grid nodes of the global spatiotemporal grid are determined, with a preset longitude and latitude resolution of 0.05 degrees, a preset time resolution of 1 hour / step, and a vertical resolution of 2 meters. Auxiliary environmental data can be obtained from sources such as TPXO9 tidal model data and CCMP 6-hour wind field data. Observational data from a mobile ADCP system are collected to obtain a sparse set of observation points. With a ship speed range of 5-8 knots and a heading of 30-150 degrees, after processing including ship motion correction and quality control elimination, 3200 valid observation points were selected, and a multi-dimensional feature set was constructed. Vertical gradient calculation used a 2-meter depth difference. A dynamic adjacency matrix was constructed, with a spatial bandwidth parameter of 0.25 and a temporal bandwidth parameter of 5 hours, resulting in spatial weights greater than or equal to 0.8 and temporal weights greater than or equal to 0.7. The training and validation sets were divided in an 8:2 ratio. The AdamW optimizer was used (initial learning rate 0.001), with a batch size of 8 and 80 training epochs (triggered by an early stop mechanism). The joint loss function used loss term weighting coefficients of 0.3 and 0.1, respectively. A comparison was made between the globally reconstructed velocity field and the anchorage ADCP observation data obtained from the trained model: the mean absolute error (MAE) of the zonal velocity was 3.2 cm / s, the root mean square error (RMSE) was 4.8 cm / s, and the correlation coefficient was... =0.89; Mean absolute error (MAE) of meridional velocity = 2.8 cm / s, root mean square error (RMSE) = 4.2 cm / s, correlation coefficient =0.91; Physical residual =0.08m / s 2 When 2 hours of ADCP real-time observation data are added. With 120 valid points, the model update took 3.5 minutes, and the root mean square error (RMSE) of the reconstructed result was further reduced by 8% after the update.

[0068] For the ocean region, a certain latitude and longitude boundary is as follows: For the target sea area, vertical depth domain Time series domain The grid nodes of the global spatiotemporal grid are determined, with a preset longitude and latitude resolution of 0.05 degrees, a preset time resolution of 1 hour / step, and a vertical resolution of 4 meters. Auxiliary environmental data can be obtained from sources such as TPXO9 tidal model data and CCMP 6-hour wind field data. Observational data from a mobile ADCP system are collected to obtain a sparse set of observation points. With a ship speed range of 8-12 knots and a heading of 60-120 degrees, after processing including ship motion correction and quality control elimination, 2800 valid observation points were obtained. A multi-dimensional feature set and a dynamic adjacency matrix were constructed. During model training, 90 training rounds were conducted. The weighting coefficients for the loss term in the joint loss function were 0.25 and 0.1, respectively. Comparison of the reconstruction results with Argo buoy observation data showed: mean absolute error (MAE) of zonal current velocity = 4.1 cm / s, root mean square error (RMSE) = 5.9 cm / s, and correlation coefficient... =0.87; Mean absolute error (MAE) of meridional velocity = 3.6 cm / s, Root mean square error (RMSE) = 5.1 cm / s, Correlation coefficient =0.89. The above embodiments demonstrate that this embodiment can achieve high-precision, physically consistent full-domain spatiotemporal field reconstruction of current velocity in nearshore and ocean areas based on mobile ADCP sparse observation data, and has real-time update capability, which is significantly better than the existing technology.

[0069] An implementation of the overall process of this embodiment is as follows: Figure 3As shown, based on the raw ADCP observation data, data preprocessing includes hull correction, quality control (i.e., removing data that meets preset removal conditions), and feature construction (i.e., constructing a multi-dimensional feature set), mapping to obtain gridded initial features (i.e., the grid node feature matrix). A temporal spatiotemporal graph (i.e., a dynamic adjacency matrix) is constructed based on the gridded initial features. Based on the gridded initial features, the temporal spatiotemporal graph, and auxiliary data such as topography, tides, and wind fields, the data is input into the Graph WaveNet++ feature encoding network. Through graph convolutional layers, dilated convolutional layers, vertical feature fusion, and quality attention layers, spatiotemporal encoded features (i.e., the spatiotemporal representation feature matrix) are obtained. Furthermore, based on Regional-PINN physical constraints and joint optimization, physical equations for the nearshore area and the ocean area are constructed for constraint to achieve current field reconstruction. During reconstruction, a joint loss function is used for optimization (the joint loss function also includes optimization of the weight parameters of GraphWaveNet++, not shown in the figure). The joint loss function consists of three parts: data loss, physical loss, and regularization loss. When the joint loss function is optimized based on the reconstructed global spatiotemporal field of ocean current velocity, model parameters are updated through backpropagation, and optimized weight parameters are obtained through training iterations. After reconstructing the global spatiotemporal field of ocean current velocity, physical verification and smoothing can also be performed based on spatiotemporal coding features. The above is an example; specific adjustments can be made according to the implementation situation, and no limitations are imposed here.

[0070] For online updates of the model, it can be done as follows: Figure 4 Based on existing historical observations and auxiliary data, an adaptive dynamic spatiotemporal map of the flight path and initial features of the target sea area grid (i.e., dynamic adjacency matrix and grid node feature matrix) are obtained. Through lightweight online updates, the weight parameters of the Graph WaveNet++ spatiotemporal feature extraction network are fixed and not updated; only the weight parameters of the Regional-PINN regionalized physical constraint model are updated. Figure 4 As shown, only the parameters of the regional adaptation layer are updated. Backpropagation is performed online using a joint loss function based on real-time observation data from the mobile ADCP, and the model parameters are updated for iterative training. This allows for the reception of real-time observation data and, by updating only the adaptation layer parameters, rapid iterative reconstruction of the current field is achieved, resulting in the global spatiotemporal field of ocean current velocity. The above is an example; specific settings should be implemented according to the actual situation, and no limitations are specified here.

[0071] The ocean current velocity global spatiotemporal field reconstruction method based on mobile ADCP provided by the present invention fully explores the vertical physical correlation of current velocity in mobile ADCP profiles, integrates profile layering features into global reconstruction, improves reconstruction accuracy, adapts to the spatiotemporal characteristics of ADCP sparse asynchronous track data, solves the reconstruction discontinuity problem caused by spatiotemporal sparsity, constructs regional adaptive physical constraints, and combines auxiliary environmental data of the target sea area to ensure the physical consistency of reconstruction results and adaptability to actual sea conditions, thereby achieving high-precision reconstruction of ocean current velocity global spatiotemporal field from mobile ADCP sparse asynchronous observation data to high resolution, physical consistency, and spatiotemporal continuity.

[0072] Figure 5 A schematic diagram of the structure of the ocean current velocity global spatiotemporal field reconstruction device based on a mobile ADCP provided in an embodiment of the present invention is shown. Figure 5 As shown, the device includes: The initial grid module 510 is suitable for determining the target sea area, constructing a global spatiotemporal grid, and obtaining each grid node; the global spatiotemporal grid is a four-dimensional domain including longitude, latitude, depth, and time series; The grid matrix module 520 is suitable for acquiring observation data of the target sea area based on the underway ADCP, preprocessing the observation data, setting the multi-dimensional feature set of the observation points corresponding to each preprocessed observation data, mapping each observation point to each grid node in the global spatiotemporal grid based on coordinates, setting grid node features based on the multi-dimensional feature set of each observation point, and obtaining the grid node feature matrix. The dynamic adjacency matrix module 530 is suitable for constructing a dynamic adjacency matrix based on the time dimension of the mapped grid node feature matrix. Each matrix element in the dynamic adjacency matrix represents the association weight between grid nodes. The reconstruction module 540 is adapted to input the grid node feature matrix and dynamic adjacency matrix into a preset spatiotemporal feature extraction network to obtain a spatiotemporal representation feature matrix; and, based on the spatiotemporal representation feature matrix and auxiliary environmental data into a preset regionalized physical constraint model, to reconstruct the current velocity field based on the physical constraints of the nearshore area and the ocean area, to obtain a global reconstructed current velocity field; wherein, the nearshore area and the ocean area are determined according to the terrain depth; the preset spatiotemporal feature extraction network and the preset regionalized physical constraint model are trained and optimized based on the constructed multiple loss functions.

[0073] Optionally, the initial mesh module 510 is further adapted to: The boundary of the target sea area is determined, and the longitude, latitude, and vertical depth intervals of the target sea area are obtained. The time series interval is also determined to obtain a four-dimensional global spatiotemporal grid. Among them, the maximum value of the vertical depth interval is the maximum vertical observation depth of the underway ADCP. Based on the preset longitude and latitude resolution, the depth unit of the mobile ADCP, and the preset time resolution, the global spatiotemporal grid is discretized to obtain each grid node.

[0074] Optionally, the grid matrix module 520 is further adapted to: Based on preset mobile ADCP parameters, observational data and auxiliary environmental data of the target sea area are acquired using mobile ADCP. The observational data includes spatiotemporal data, current velocity data, mass characteristic data, and auxiliary observation data. The auxiliary observation data includes ship speed and heading data. The mass characteristic data includes echo intensity, correlation data, and mass markers for each depth unit. The current velocity data includes three-dimensional current velocities of zonal, meridional, and vertical current components. The spatiotemporal data includes observation timestamps, longitude, latitude, and depth. The auxiliary environmental data includes tidal data, wind field data, and seawater physical parameters. The tidal data includes tidal phase and tidal height. The current velocity data is corrected for ship motion based on ship speed and heading data to obtain the corrected current velocity data. Observational data that meet the preset rejection criteria are removed to obtain preprocessed observational data. The preset rejection criteria include invalid quality marking, echo intensity below the echo intensity threshold or correlation data below the correlation data threshold, and outliers in the velocity data of each depth layer determined based on the 3σ criterion. A multi-dimensional feature set is constructed for each observation point corresponding to the preprocessed observation data. This multi-dimensional feature set includes vertical profile features, planar spatiotemporal features, and quality-perceived features. Vertical profile features include corrected velocity data, vertical gradient, and vertical statistical features. The vertical gradient is determined based on the ratio of the difference between the corrected zonal velocity components of adjacent depth units of the observation point to the depth unit. Vertical statistical features include the mean and variance of the corrected velocity data for each depth unit of the observation point. Planar spatiotemporal features include the coordinate normalized values, tidal phase, and wind field data of the spatiotemporal data of the observation point. Quality-perceived features include quality weights, which are determined based on the ratio of the echo intensity of the observation point to the maximum echo intensity of the depth unit where the observation point is located, and the correlation data of the observation point. The observation points corresponding to each preprocessed observation data are mapped to each grid node in the global spatiotemporal grid according to the coordinates of the spatiotemporal data. The grid node features are set based on the multi-dimensional feature set of each observation point to obtain the grid node feature matrix.

[0075] Optionally, the grid matrix module 520 is further adapted to: The system is based on whether the coordinates of the grid nodes in the global spatiotemporal grid match the spatiotemporal data of the observation points corresponding to the preprocessed observation data. If there is no match, set the grid node feature to 0; If a match is found, determine whether there are multiple matching grid nodes; If so, the quality weights of the multi-dimensional feature sets of the matched grid nodes are weighted and averaged to set the grid node features. If not, set the multi-dimensional feature set of the matched grid nodes as the grid node features.

[0076] Optionally, the dynamic adjacency matrix module 530 is further adapted to: Based on the time dimension of the grid node feature matrix, the spatial weight and time weight between any two grid nodes are determined. The spatial weight and time weight are then fused based on a preset weight ratio to obtain matrix elements that characterize the correlation weight between grid nodes. A dynamic adjacency matrix is ​​constructed based on each matrix element.

[0077] Optionally, the refactoring module 540 is further adapted to: The input matrix and the dynamic adjacency matrix are determined by the depth dimension of the grid node feature matrix and fed into the preset spatiotemporal feature extraction network. The spatiotemporal representation feature matrix is ​​obtained through graph convolutional layer, dilated causal convolutional layer, vertical feature fusion layer and quality attention layer. Based on topographic depth, the nearshore zone and the ocean zone are divided; The spatiotemporal representation feature matrix and auxiliary environmental data are input into a preset regionalized physical constraint model, and physical constraints for the nearshore area and the ocean area are constructed to reconstruct the horizontal and vertical current velocities in the nearshore and ocean areas, and then verify and adjust them to obtain a global reconstructed velocity field. Among them, physical equations including shallow water dynamic constraints and bottom friction constraints are introduced for the nearshore area, and depth-average momentum constraint equations are introduced for the ocean area.

[0078] Optionally, the device further includes: an updated model 550, adapted to acquire new observation data of the mobile ADCP, obtain a new grid node feature matrix based on the new observation data, and update and optimize the weight parameters of the preset regionalized physical constraint model based on the new grid node feature matrix.

[0079] The descriptions of the above modules refer to the corresponding descriptions in the method embodiments, and will not be repeated here.

[0080] This invention also provides a non-volatile computer storage medium storing at least one executable instruction that can perform the operation corresponding to the ocean current velocity global spatiotemporal field reconstruction method based on mobile ADCP in any of the above method embodiments.

[0081] This application provides a computer program product, which includes at least one executable instruction or computer program that enables a processor to perform the operation corresponding to the ocean current velocity global spatiotemporal field reconstruction method based on mobile ADCP in any of the above method embodiments.

[0082] Figure 6 The diagram illustrates the structure of a computing device according to an embodiment of the present invention. The specific embodiments of the present invention do not limit the specific implementation of the computing device.

[0083] like Figure 6 As shown, the computing device may include: a processor 602, a communication interface 604, a memory 606, and a communication bus 608.

[0084] in: The processor 602, communication interface 604, and memory 606 communicate with each other via communication bus 608.

[0085] Communication interface 604 is used to communicate with other network elements such as clients or other servers.

[0086] The processor 602 is used to execute program 610, which can specifically execute the relevant steps in the above embodiment of the ocean current velocity global spatiotemporal field reconstruction method based on the mobile ADCP.

[0087] Specifically, program 610 may include program code that includes computer operation instructions.

[0088] Processor 602 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The computing device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0089] Memory 606 is used to store program 610. Memory 606 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0090] Specifically, program 610 can be used to cause processor 602 to execute the ocean current velocity global spatiotemporal field reconstruction method based on mobile ADCP in any of the above method embodiments. The specific implementation of each step in program 610 can be found in the corresponding steps and units described in the above embodiments of ocean current velocity global spatiotemporal field reconstruction based on mobile ADCP, and will not be repeated here. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the devices and modules described above can be referred to the corresponding process descriptions in the foregoing method embodiments, and will not be repeated here.

[0091] The algorithms or displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, the embodiments of the present invention are not directed to any particular programming language. It should be understood that the embodiments of the present invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing preferred embodiments of the present invention.

[0092] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0093] Similarly, it should be understood that, in order to streamline the embodiments of the invention and aid in understanding one or more of the various inventive aspects, features of the embodiments of the invention are sometimes grouped together in a single embodiment, figure, or description thereof in the above description of exemplary embodiments of the invention. However, this disclosure should not be construed as reflecting an intention that the claimed embodiments of the invention require more features than are expressly recited in each claim. Rather, as reflected in the following claims, inventive aspects lie in fewer than all features of a single foregoing disclosed embodiment. Therefore, the claims following the detailed description are hereby expressly incorporated into that detailed description, wherein each claim itself is a separate embodiment of the invention.

[0094] Those skilled in the art will understand that modules in the device of the embodiments can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiments can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components. Except where at least some of such features and / or processes or units are mutually exclusive, any combination can be used to combine all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or device so disclosed. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0095] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of the invention and form different embodiments. For example, in the following claims, any of the claimed embodiments can be used in any combination.

[0096] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components according to the embodiments of the present invention. The embodiments of the present invention can also be implemented as device or apparatus programs (e.g., computer programs and computer program products) for performing part or all of the methods described herein. Such programs implementing the embodiments of the present invention can be stored on a computer-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0097] It should be noted that the above embodiments are illustrative of the present invention and not restrictive of the invention, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. Embodiments of the present invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names. The steps in the above embodiments, unless otherwise specified, should not be construed as limiting the order of execution.

Claims

1. A method for global spatiotemporal reconstruction of ocean current velocity based on underway ADCP, characterized in that the method... include: The target sea area is determined, a global spatiotemporal grid is constructed, and each grid node is obtained; the global spatiotemporal grid is a four-dimensional domain including longitude, latitude, depth, and time series. The observation data of the target sea area is obtained by the mobile ADCP. The observation data is preprocessed, a multi-dimensional feature set of the observation points corresponding to each preprocessed observation data is set, and each observation point is mapped to each grid node in the global spatiotemporal grid based on coordinates. The grid node features are set based on the multi-dimensional feature set of each observation point to obtain the grid node feature matrix. Based on the time dimension of the mapped grid node feature matrix, a dynamic adjacency matrix is ​​constructed, where each matrix element represents the association weight between grid nodes. The grid node feature matrix and the dynamic adjacency matrix are input into a preset spatiotemporal feature extraction network to obtain a spatiotemporal representation feature matrix; Furthermore, the spatiotemporal representation feature matrix and auxiliary environmental data are input into a preset regionalized physical constraint model, and the velocity field is reconstructed based on the physical constraints of the nearshore area and the ocean area to obtain a global reconstructed velocity field; wherein, the nearshore area and the ocean area are determined according to the terrain depth; the preset spatiotemporal feature extraction network and the preset regionalized physical constraint model are trained and optimized based on the constructed multiple loss functions.

2. The method according to claim 1, characterized in that, The process of determining the target sea area, constructing a global spatiotemporal grid, and obtaining each grid node further includes: The boundary of the target sea area is determined, and the longitude interval, latitude interval and vertical depth interval of the target sea area are obtained. The time series interval is also determined to obtain a four-dimensional global spatiotemporal grid. Among them, the maximum value of the vertical depth interval is the maximum vertical observation depth of the underway ADCP. The global spatiotemporal grid is discretized based on the preset longitude and latitude resolution, the depth unit of the mobile ADCP, and the preset time resolution to obtain each grid node.

3. The method according to claim 1, characterized in that, The process of acquiring observation data of the target sea area based on the underway ADCP, preprocessing the observation data, setting a multi-dimensional feature set for each observation point corresponding to each preprocessed observation data, mapping each observation point to each grid node in the global spatiotemporal grid based on coordinates, and setting grid node features based on the multi-dimensional feature set of each observation point to obtain a grid node feature matrix further includes: Based on preset underway ADCP parameters, observational data and auxiliary environmental data for the target sea area are acquired using underway ADCP. The observational data includes spatiotemporal data, current velocity data, mass characteristic data, and auxiliary observation data. The auxiliary observation data includes ship speed and heading data. The mass characteristic data includes echo intensity, correlation data, and mass markers for each depth unit. The current velocity data includes three-dimensional current velocities of zonal, meridional, and vertical current components. The spatiotemporal data includes observation timestamps, longitude, latitude, and depth. The auxiliary environmental data includes tidal data, wind field data, and seawater physical parameters. The tidal data includes tidal phase and tidal height. Based on the ship speed and heading data, the flow velocity data is corrected for ship motion to obtain the corrected flow velocity data. Observational data that meet the preset rejection criteria are removed to obtain preprocessed observational data. The preset rejection criteria include quality marking as invalid, echo intensity being lower than the echo intensity threshold or correlation data being lower than the correlation data threshold, and outliers in the velocity data of each depth layer being determined based on the 3σ criterion. A multi-dimensional feature set is constructed for each observation point corresponding to the preprocessed observation data. This multi-dimensional feature set includes vertical profile features, planar spatiotemporal features, and quality-aware features. The vertical profile features include corrected velocity data, vertical gradient, and vertical statistical features. The vertical gradient is determined based on the ratio of the difference between the corrected zonal velocity components of adjacent depth units of the observation point to the depth unit. The vertical statistical features include the mean and variance of the corrected velocity data for each depth unit of the observation point. The planar spatiotemporal features include the coordinate normalized values, tidal phase, and wind field data of the spatiotemporal data of the observation point. The quality-aware features include quality weights, which are determined based on the ratio of the echo intensity of the observation point to the maximum echo intensity of the depth unit where the observation point is located, and the correlation data of the observation point. The observation points corresponding to each preprocessed observation data are mapped to each grid node in the global spatiotemporal grid according to the coordinates of the spatiotemporal data. The grid node features are set based on the multi-dimensional feature set of each observation point to obtain the grid node feature matrix.

4. The method according to claim 3, characterized in that, The step of mapping the observation points corresponding to each preprocessed observation data to each grid node in the global spatiotemporal grid according to the coordinates of the spatiotemporal data, and setting grid node features based on the multi-dimensional feature set of each observation point to obtain the grid node feature matrix further includes: The system is based on whether the coordinates of the grid nodes in the global spatiotemporal grid match the spatiotemporal data of the observation points corresponding to the preprocessed observation data. If there is no match, set the grid node feature to 0; If a match is found, determine whether there are multiple matching grid nodes; If so, the quality weights of the multi-dimensional feature sets of the matched grid nodes are weighted and averaged to set the grid node features. If not, set the multi-dimensional feature set of the matched grid nodes as the grid node features.

5. The method according to claim 1, characterized in that, The construction of the dynamic adjacency matrix based on the time dimension of the mapped grid node feature matrix further includes: Based on the time dimension of the grid node feature matrix, the spatial weight and time weight between any two grid nodes are determined, and the spatial weight and time weight are fused based on a preset weight ratio to obtain matrix elements that characterize the correlation weight between grid nodes. A dynamic adjacency matrix is ​​constructed based on each matrix element.

6. The method according to claim 1, characterized in that, The step of inputting the grid node feature matrix and the dynamic adjacency matrix into a preset spatiotemporal feature extraction network to obtain a spatiotemporal representation feature matrix; and inputting the spatiotemporal representation feature matrix and auxiliary environmental data into a preset regionalized physical constraint model, and reconstructing the velocity field based on nearshore physical constraints and oceanic physical constraints to obtain a globally reconstructed velocity field, further includes: The input matrix, determined by the depth dimension of the grid node feature matrix, and the dynamic adjacency matrix are jointly input into a preset spatiotemporal feature extraction network. The spatiotemporal representation feature matrix is ​​obtained through graph convolutional layers, dilated causal convolutional layers, vertical feature fusion layers, and quality attention layers. Based on topographic depth, the nearshore zone and the ocean zone are divided; The spatiotemporal representation feature matrix and auxiliary environmental data are input into a preset regionalized physical constraint model, and physical constraints for the nearshore area and the ocean area are constructed to reconstruct the horizontal and vertical current velocities in the nearshore and ocean areas, and then verify and adjust them to obtain a global reconstructed velocity field. Among them, the physical equations including shallow water dynamic constraints and bottom friction constraints are introduced for the nearshore area, and the depth-average momentum constraint equations are introduced for the ocean area.

7. The method according to claim 1, characterized in that, The method further includes: New observation data of the mobile ADCP is acquired, and a new grid node feature matrix is ​​obtained based on the new observation data. The weight parameters of the preset regionalized physical constraint model are then updated and optimized based on the new grid node feature matrix.

8. A device for reconstructing the global spatiotemporal field of ocean current velocity based on a mobile ADCP, characterized in that the device... include: An initial grid module is suitable for determining the target sea area, constructing a global spatiotemporal grid, and obtaining each grid node; the global spatiotemporal grid is a four-dimensional domain including longitude, latitude, depth, and time series; The grid matrix module is suitable for acquiring observation data of the target sea area based on the mobile ADCP, preprocessing the observation data, setting a multi-dimensional feature set of the observation points corresponding to each preprocessed observation data, mapping each observation point to each grid node in the global spatiotemporal grid based on coordinates, setting grid node features based on the multi-dimensional feature set of each observation point, and obtaining a grid node feature matrix. The dynamic adjacency matrix module is suitable for constructing a dynamic adjacency matrix based on the time dimension of the mapped grid node feature matrix, wherein each matrix element in the dynamic adjacency matrix represents the association weight between grid nodes; The reconstruction module is adapted to input the grid node feature matrix and the dynamic adjacency matrix into a preset spatiotemporal feature extraction network to obtain a spatiotemporal representation feature matrix; Furthermore, based on the spatiotemporal representation feature matrix and auxiliary environmental data to the preset regionalized physical constraint model, the velocity field is reconstructed based on the physical constraints of the nearshore area and the ocean area to obtain the global reconstructed velocity field; wherein, the nearshore area and the ocean area are determined according to the terrain depth; the preset spatiotemporal feature extraction network and the preset regionalized physical constraint model are trained and optimized based on the constructed multiple loss functions.

9. A computing device, characterized in that, include: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the ocean current velocity global spatiotemporal field reconstruction method based on any one of claims 1-7.

10. A computer storage medium, characterized in that, The storage medium stores at least one executable instruction that causes the processor to perform the operation corresponding to the ocean current velocity global spatiotemporal field reconstruction method based on the underway ADCP as described in any one of claims 1-7.

11. A computer program product, characterized in that, It includes at least one executable instruction that causes the processor to perform the operation corresponding to the ocean current velocity global spatiotemporal field reconstruction method based on any one of claims 1-7.

Images