Modeling method and device of marine vector environmental field, storage medium and electronic equipment

Abstract

A modeling method and device of a marine vector environmental field, a storage medium and an electronic device, relate to the technical field of marine environment modeling. The modeling method comprises: preprocessing topographic data, forcing field, initial field, boundary field and multi-source observation data of a target sea area collected to obtain preprocessed standard data; determining a basic marine dynamic environmental field according to a pre-set regional marine model of the target sea area and the preprocessed standard data; determining a future time period environmental field prediction result according to a continuous environmental field sample generation rule of a pre-constructed time series prediction network and the basic marine dynamic environmental field; and performing multi-scale data assimilation on the multi-source observation data, the basic marine dynamic environmental field and the future time period environmental field prediction result to obtain a target marine vector environmental field. The modeling method can still construct a continuous and updateable three-dimensional target marine vector environmental field under sparse observation conditions, thereby improving the availability and integrity of complex sea area environmental information.

Description

Technical Field

[0001] This application relates to the field of marine environment modeling technology, and more specifically, to a method, apparatus, storage medium, and electronic device for modeling marine vector environmental fields. Background Technology

[0002] While existing ocean observation data sources are gradually becoming more abundant, including various types of data such as ocean currents, temperature, salinity, meteorology, and topography, problems such as uneven spatial distribution, insufficient temporal continuity, inconsistent observation scales, and heterogeneous data formats are still prevalent in engineering applications. For specific mission areas, local observations often cannot independently form a complete three-dimensional continuous environmental model, and are particularly difficult to reflect local flows, thermocline changes, and multi-depth layer coupling characteristics near complex seabed topography.

[0003] Regional ocean numerical models can continuously simulate the current field, temperature-salinity structure, and their temporal evolution in target sea areas based on dynamic equations, thus serving as an important means of constructing fundamental marine environmental fields. However, relying solely on numerical models still has significant limitations. For example, the governing equations typically require approximation and parameterization in engineering implementation, making it difficult to fully analyze small-scale processes. Furthermore, errors in boundary conditions, initial field biases, and discrete solution errors can lead to deviations between simulation results and the real marine environment, especially in complex coastal areas, island and reef regions, and areas with strong topographic relief.

[0004] On the other hand, predicting the marine environment using only traditional two-dimensional convolutional networks or simple temporal extrapolation methods usually only emphasizes the spatial texture at a certain moment, making it difficult to simultaneously preserve the spatial structure information and temporal evolution patterns of the continuous environmental field. When the target sea area exhibits seasonal, interannual, or even decadal scale variations, the prediction accuracy and stability of the prediction network for future flow fields and temperature-salinity structures will be limited if the network lacks a targeted spatiotemporal joint modeling mechanism.

[0005] While existing technologies have explored marine environment modeling from the perspectives of numerical simulation, deep learning prediction, and data assimilation, they generally suffer from a fragmented approach: some schemes can only provide basic simulation results, making it difficult to improve the predictive ability for future periods; some schemes emphasize prediction accuracy but lack constraints consistent with the physical model; and some schemes, while introducing observational corrections, fail to further organize the topographic coupling results into a 3D environmental product that is easy for AUVs to directly use. Therefore, existing technologies struggle to simultaneously achieve physical interpretability, continuous prediction capability, state correction capability, and engineering usability.

[0006] The content in the background section is merely technology known to the public and does not necessarily represent existing technology in this field. Summary of the Invention

[0007] This application aims to provide a method, apparatus, storage medium, and electronic device for modeling marine vector environmental fields to solve at least one of the aforementioned technical problems.

[0008] According to one aspect of this application, this application provides a modeling method for a marine vector environmental field. The modeling method includes: preprocessing the collected topographic data, forcing field, initial field, boundary field, and multi-source observation data of the target sea area to obtain preprocessed standard data; determining the basic marine dynamic environmental field based on a preset regional marine model of the target sea area and the preprocessed standard data; determining the environmental field prediction results for future periods based on the continuous environmental field sample generation rules of a pre-constructed time-series prediction network and the basic marine dynamic environmental field; and performing multi-scale data assimilation on the multi-source observation data, the basic marine dynamic environmental field, and the environmental field prediction results for future periods to obtain the target marine vector environmental field.

[0009] According to some embodiments of this application, the collected topographic data, forcing field, initial field, boundary field, and multi-source observation data of the target sea area are preprocessed to obtain preprocessed standard data. This includes: projecting the topographic data, forcing field, initial field, boundary field, and multi-source observation data onto a unified coordinate system to obtain first standard data; performing unified time step mapping on the first standard data to obtain second standard data; converting the format of the second standard data to obtain third standard data that meets the input requirements; and determining the preprocessed standard data based on the third standard data and preset quality rules.

[0010] According to some embodiments of this application, a basic marine dynamic environment field is determined based on a preset regional marine model and preprocessed standard data of the target sea area, including: determining a preset regional marine model based on the target sea area and computing resources; and establishing a three-dimensional numerical grid based on the determined preset regional marine model and preprocessed standard data to obtain the basic marine dynamic environment field.

[0011] According to some embodiments of this application, the regional ocean model is configured to support curved orthogonal horizontal grids and S-coordinate vertical stratification.

[0012] According to some embodiments of this application, the prediction results of the environmental field for future periods are determined based on the continuous environmental field sample generation rules of the pre-constructed temporal prediction network and the basic marine dynamic environmental field. This includes: constructing the basic marine dynamic environmental field into a continuous sample sequence in chronological order; performing three-dimensional convolutional feature extraction on the continuous sample sequence to obtain a spatiotemporal correlation feature tensor; and inputting the spatiotemporal correlation feature tensor into the temporal prediction network to obtain the prediction results of the environmental field for future periods according to the continuous environmental field sample generation rules.

[0013] According to some embodiments of this application, multi-scale data assimilation is performed on multi-source observation data, basic marine dynamic environment field, and future time period environment field prediction results to obtain a target marine vector environment field. This includes: spatiotemporally aligning the multi-source observation data to obtain aligned multi-source observation data; and performing parameter estimation and recursive correction on the aligned multi-source observation data, basic marine dynamic environment field, and future time period environment field prediction results to obtain the target marine vector environment field.

[0014] According to some embodiments of this application, after multi-scale data assimilation of multi-source observation data, basic marine dynamic environment field and future time period environment field prediction results to obtain the target marine vector environment field, the modeling method further includes: coupling the target marine vector environment field and the seabed topography model of the target sea area to obtain the three-dimensional environment model of the target sea area and the target output results; wherein, the target output results include the multi-depth layer current velocity vector, temperature distribution and salinity distribution of the target sea area.

[0015] According to one aspect of this application, a modeling apparatus for a marine vector environmental field is provided. The modeling apparatus includes: a preprocessing module, a regional marine model construction module, a time-series prediction module, and a data assimilation and correction module. The preprocessing module preprocesses the collected topographic data, forcing field, initial field, boundary field, and multi-source observation data of the target sea area to obtain preprocessed standard data. The regional marine model construction module determines the basic marine dynamic environmental field based on a pre-defined regional marine model of the target sea area and the preprocessed standard data. The time-series prediction module determines the environmental field prediction results for future periods based on the continuous environmental field sample generation rules of a pre-constructed time-series prediction network and the basic marine dynamic environmental field. The data assimilation and correction module performs multi-scale data assimilation on the multi-source observation data, the basic marine dynamic environmental field, and the environmental field prediction results for future periods to obtain the target marine vector environmental field.

[0016] According to another aspect of this application, this application also provides a non-volatile computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is capable of implementing the modeling method for the marine vector environmental field as described above.

[0017] According to another aspect of this application, this application also provides an electronic device, including: one or more processors; and a storage device for storing one or more programs, which, when executed by the one or more processors, enable the one or more processors to implement the modeling method for the marine vector environmental field as described above.

[0018] The technical solution of this application can preprocess the topography, forcing field, initial field, boundary field, and multi-source observation data of the target sea area to obtain preprocessed standard data. The technical solution of this application determines the basic marine dynamic environment field through a pre-defined regional ocean model of the target sea area and the preprocessed standard data.

[0019] The technical solution of this application determines the environmental field prediction results for future periods by using the continuous environmental field sample generation rules of a pre-constructed time-series prediction network and the basic marine dynamic environmental field. This technical solution can perform multi-scale data assimilation on multi-source observation data, the basic marine dynamic environmental field, and the environmental field prediction results for future periods to obtain the target marine vector environmental field.

[0020] The technical solution of this application can integrate multi-source observation data, regional ocean models, time-series prediction networks, and data assimilation into a unified link. Even under conditions of sparse observation, it can still construct a continuous and updatable three-dimensional target ocean vector environmental field, thereby improving the accessibility and completeness of environmental information in complex marine areas.

[0021] The technical solution of this application utilizes the physical interpretability of regional ocean models and the temporal prediction capability of temporal prediction networks, which can better balance the continuous evolution of the environmental field with the prediction needs of future periods, thereby improving the accuracy and foresight of the refined modeling results. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart illustrating a modeling method 1000 according to an embodiment of this application is shown; Figure 2 A flowchart illustrating step S100 according to an embodiment of this application is shown; Figure 3 A flowchart illustrating step S200 according to an embodiment of this application is shown; Figure 4 A flowchart illustrating step S300 according to an embodiment of this application is shown; Figure 5 A flowchart illustrating step S400 according to an embodiment of this application is shown; Figure 6 This illustrates another flowchart of a modeling method 1000 according to an embodiment of this application; Figure 7A schematic diagram of the modeling apparatus according to an embodiment of this application is shown; Figure 8 Another structural schematic diagram of the modeling apparatus according to an embodiment of this application is shown.

[0024] Explanation of reference numerals in the attached figures: Modeling device 10; preprocessing module 11; regional ocean model construction module 12; time series prediction module 13; data assimilation and correction module 14; terrain coupling and output module 15. Detailed Implementation

[0025] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, they are provided so that this application will be thorough and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted.

[0026] The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a full understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced without one or more of these specific details, or other methods, components, materials, devices, etc. In these cases, well-known structures, methods, devices, implementations, materials, or operations will not be shown or described in detail.

[0027] Furthermore, the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus.

[0028] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order.

[0029] The technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0030] The English terms used in this application, their full English names, and their corresponding Chinese definitions are as follows: AUV, Autonomous Underwater Vehicle; ETOPO, EarthTOPOgraphy, is a global seabed topography dataset. WGS-84, World Geodetic System 1984; NetCDF, NetworkCommonDataForm, is a common data format for networks. HDF5, Hierarchical Data Format, version 5; Hierarchical Data Format (Fifth Edition); NumPy, NumericPython, is a Python numerical computation extension library. ROMS, Regional Ocean Modeling System; FVCOM, FiniteVolumeCommunityOceanModel; MITgcm, MIT General Circulation Model; HYCOM, HybridCoordinateModel, is a hybrid coordinate ocean model. Conv-LSTM, Convolutional Long Short-Term Memory network; ConvGRU, ConvolutionalGatedRecurrentUnit.

[0031] See Figure 7 The marine vector environmental field modeling device 10 provided in this application includes a preprocessing module 11, a regional marine model construction module 12, a time-series prediction module 13, and a data assimilation and correction module 14. See also... Figure 8 The modeling device 10 also includes a terrain coupling and output module 15.

[0032] The following is combined Figure 7 and Figure 8 This application describes a modeling method 1000 for marine vector environmental fields. See also... Figure 1 The modeling method 1000 includes steps S100-S400.

[0033] In step S100, the collected topographic data, forcing field, initial field, boundary field and multi-source observation data of the target sea area are preprocessed to obtain preprocessed standard data.

[0034] According to the example embodiment, the target sea area can be a specific marine spatial region pre-delineated based on actual application needs (such as AUV operation area, environmental monitoring range, etc.). Topographic data can be digitized elevation information characterizing the spatial distribution and undulations of the seabed depth in the target sea area, including longitude, latitude, and water depth values. For example, the topographic data can utilize the ETOPO series datasets to ensure accurate representation of the coastline and seabed undulations.

[0035] The forcing field can characterize the external driving conditions of the sea surface in the regional ocean model corresponding to the target sea area, including sea surface wind stress, heat flux, and freshwater flux. The initial field can characterize the initial state field of the regional ocean model corresponding to the target sea area, including environmental state data such as temperature and salinity at the time of model calculation. The boundary field can characterize the boundary condition field of the regional ocean model corresponding to the target sea area, including boundary constraint data such as sea surface height, three-dimensional velocity, temperature, and salinity.

[0036] Multi-source observation data can be multi-parameter discrete sampling records from different ocean observation platforms used to characterize the true physical state of the target sea area. Multi-source observation data can include flow field, temperature, salinity and meteorological observations.

[0037] Preprocessing can be a process of standardizing and reconstructing acquired multi-source data through format conversion and normalization. For example, preprocessing may include operations such as coordinate and time unification, format conversion, and quality checks.

[0038] Preprocessing standard data can be obtained by preprocessing topographic data, forcing fields, initial fields, boundary fields, and multi-source observation data.

[0039] For example, in step S100, the preprocessing module 11 preprocesses the collected topographic data, forcing field, initial field, boundary field and multi-source observation data of the target sea area to obtain preprocessed standard data.

[0040] The preprocessing module 11 can unify the coordinates and time of topographic data, forcing field, initial field, boundary field and multi-source observation data, convert the format to the format that the regional ocean model can read, and remove, mark or interpolate abnormal data such as missing, out-of-boundary, abrupt or duplicate records, so as to obtain preprocessed standard data.

[0041] In step S200, the basic marine dynamic environment field is determined based on the preset regional marine model and preprocessed standard data of the target sea area.

[0042] According to the example embodiment, the regional ocean model can be a marine dynamic model for conducting three-dimensional numerical simulations of a target sea area. The basic marine dynamic environment field can be a three-dimensional continuous gridded benchmark dataset representing the large-scale background state of the target sea area, output by the regional ocean model after numerical solution under controlled boundary conditions and initial driving forces.

[0043] For example, in step S200, the regional ocean model construction module 12 determines the basic marine dynamic environment field based on the preset regional ocean model and preprocessed standard data of the target sea area.

[0044] The regional ocean model construction module 12 establishes a three-dimensional numerical grid for the regional ocean model, inputs preprocessed standard data into the regional ocean model for simulation, and outputs the basic marine dynamic environment field.

[0045] In step S300, the environmental field prediction results for future periods are determined based on the continuous environmental field sample generation rules of the pre-constructed time-series prediction network and the basic marine dynamic environmental field.

[0046] According to the example implementation, the continuous environmental field sample generation rule can be a rule that organizes the environmental field under a unified grid and a unified time step into network input and output labels using a fixed window. For example, the continuous environmental field sample generation rule can include: using the flow velocity, temperature, and salinity fields of six consecutive time steps as input to predict the environmental field of the next time step or several future time steps; using variables such as east-west flow velocity, north-south flow velocity, vertical flow velocity, temperature, and salinity at each depth layer and each grid point as tensor channels; and performing normalization, missing measurement mask labeling, and training / validation set partitioning during sample generation to ensure consistency between the network input dimension and physical meaning.

[0047] The environmental field prediction results for future periods can be the ocean vector environmental field prediction results for one or more future time steps output by the time series prediction network.

[0048] For example, in step S300, the time series prediction module 13 determines the environmental field prediction results for future periods based on the continuous environmental field sample generation rules of the pre-constructed time series prediction network and the basic marine dynamic environmental field.

[0049] The temporal prediction module 13 constructs a continuous sample sequence of the basic marine dynamic environment field in chronological order. The temporal prediction module 13 performs three-dimensional convolutional feature extraction on the continuous sample sequence to obtain a spatiotemporal correlation feature tensor. The temporal prediction module 13 inputs the spatiotemporal correlation feature tensor into the temporal prediction network to obtain the environmental field prediction results for future periods based on the continuous environmental field sample generation rules.

[0050] In step S400, multi-scale data assimilation is performed on multi-source observation data, basic marine dynamic environment field, and future time period environment field prediction results to obtain the target marine vector environment field.

[0051] According to the example embodiment, the target ocean vector environment field can be the magnitude and direction of the current velocity at different times and depths in the target sea area, and can also include a comprehensive expression of environmental factors such as temperature and salinity.

[0052] For example, in step S400, the data assimilation and correction module 14 performs multi-scale data assimilation on multi-source observation data, basic marine dynamic environment field and future time period environment field prediction results to obtain the target marine vector environment field.

[0053] The data assimilation and correction module 14 performs spatiotemporal alignment on the multi-source observation data to obtain aligned multi-source observation data. The data assimilation and correction module 14 then performs parameter estimation and recursive correction on the aligned multi-source observation data, the basic marine dynamic environment field, and the predicted environment field for future periods to obtain the target marine vector environment field.

[0054] Through the above embodiments, the technical solution of this application can preprocess the topography, forcing field, initial field, boundary field, and multi-source observation data of the target sea area to obtain preprocessed standard data. The technical solution of this application determines the basic marine dynamic environment field using a pre-defined regional ocean model of the target sea area and the preprocessed standard data.

[0055] The technical solution of this application determines the environmental field prediction results for future periods by using the continuous environmental field sample generation rules of a pre-constructed time-series prediction network and the basic marine dynamic environmental field. This technical solution can perform multi-scale data assimilation on multi-source observation data, the basic marine dynamic environmental field, and the environmental field prediction results for future periods to obtain the target marine vector environmental field.

[0056] The technical solution of this application can integrate multi-source observation data, regional ocean models, time-series prediction networks, and data assimilation into a unified link. Even under conditions of sparse observation, it can still construct a continuous and updatable three-dimensional target ocean vector environmental field, thereby improving the accessibility and completeness of environmental information in complex marine areas.

[0057] The technical solution of this application utilizes the physical interpretability of regional ocean models and the temporal prediction capability of temporal prediction networks, which can better balance the continuous evolution of the environmental field with the prediction needs of future periods, thereby improving the accuracy and foresight of the refined modeling results.

[0058] Optionally, see Figure 2 Step S100 includes steps S110-S140.

[0059] In step S110, the terrain data, forcing field, initial field, boundary field and multi-source observation data are projected onto a unified coordinate system to obtain the first standard data.

[0060] According to the example embodiment, the first standard data can be topographic data, forcing field, initial field, boundary field, and multi-source observation data projected onto a unified coordinate system. The unified coordinate system can be the WGS-84 latitude-longitude-depth reference coordinate system, local northeast-sky coordinates, or local northeast-ground coordinates, etc.

[0061] For example, in step S110, the preprocessing module 11 projects the terrain data, forcing field, initial field, boundary field and multi-source observation data onto a unified coordinate system to obtain the first standard data.

[0062] The preprocessing module 11 projects the terrain data, forcing field, initial field, boundary field and multi-source observation data onto the WGS-84 latitude-longitude-depth reference coordinate system, and further maps them onto the curve orthogonal horizontal grid and S-coordinate vertical grid to obtain the first standard data.

[0063] In step S120, the first standard data is mapped to a uniform time step to obtain the second standard data.

[0064] According to the example embodiment, the second standard data can be data mapped to the first standard data using a unified time step. The unified time step can be a preset hourly, daily, or other task-level time step.

[0065] For example, in step S120, the preprocessing module 11 performs a unified time step mapping on the first standard data to obtain the second standard data. The preprocessing module 11 performs a unified mapping on the first standard data to an hourly time step to obtain the second standard data.

[0066] In step S130, the second standard data is format-converted to obtain the third standard data that meets the input requirements.

[0067] According to the example embodiment, the third standard data can be data converted from the second standard data format. The input requirements can be regional ocean model reading requirements or time series prediction network reading requirements.

[0068] File format conversion typically includes converting terrain data, forcing fields, initial fields, and boundary fields into gridded model input files such as NetCDF; organizing multi-source observation data into tables or NetCDF files with time, latitude, longitude, depth, and observation variable fields; and converting continuous environmental field samples for time series prediction networks into HDF5, NumPy arrays, or equivalent tensor formats.

[0069] For example, in step S130, the preprocessing module 11 performs format conversion on the second standard data to obtain third standard data that meets the input requirements. The preprocessing module 11 converts the second standard data into NetCDF file format to obtain the third standard data.

[0070] In step S140, preprocessing standard data is determined based on the third standard data and preset quality rules.

[0071] According to the example embodiment, the preset quality rules can be data control rules that satisfy the realism of marine physics and spatiotemporal logic. For example, the preset quality rules can comprehensively evaluate the validity of each data point based on the inherent variation range of marine physical parameters, the gradient continuity characteristics of adjacent spatiotemporal domains, and seabed topographic constraints, and automatically trigger different subsequent operations such as data retention, anomaly marking, removal, or interpolation compensation based on the evaluation results.

[0072] Preset quality rules include, but are not limited to: setting reasonable threshold ranges for physical quantities such as temperature, salinity, and current velocity, and removing records that are significantly outside the range of ocean physics; checking the abrupt change amplitude of adjacent time points or adjacent grid points, and marking isolated spikes that do not conform to the continuous change pattern; and removing invalid observations that fall on land areas or exceed the depth of the seabed based on land-sea mask and water depth constraints.

[0073] For example, in step S140, the preprocessing module 11 determines the preprocessing standard data based on the third standard data and preset quality rules. For abnormal data such as missing data, out-of-bounds data, abrupt changes, or duplicate records, the preprocessing module 11 can use quality control rules such as threshold range checks, adjacent spatiotemporal consistency checks, land-sea mask checks, and physical quantity continuity checks to remove, mark, or interpolate and compensate for the abnormal data, ensuring that subsequent numerical simulations and assimilation chains can directly call upon the data.

[0074] Through the above embodiments, the technical solution of this application can form a unified input dataset suitable for regional ocean models and data assimilation by preprocessing different data, so that the same set of input data can support three types of processing links: physical simulation, data-driven prediction and observation correction.

[0075] Optionally, see Figure 3 Step S200 may include steps S210-S220.

[0076] In step S210, a preset regional ocean model is determined based on the target sea area and computing resources.

[0077] According to the example embodiment, computing resources can be the hardware and software computing resources of the modeling device. The regional ocean model can be predetermined based on the target sea area and computing resources.

[0078] Optionally, the regional ocean model is configured to support curved orthogonal horizontal grids and S-coordinate vertical stratification. The regional ocean model can employ a curved orthogonal coordinate system in the horizontal direction to enhance the representation of complex coastlines and local sea boundaries. In the vertical direction, the regional ocean model can employ S-coordinate vertical stratification to locally refine the surface mixing layer, thermocline, or bottom boundary current regions according to the research focus. This balances overall modeling efficiency with improved resolution of key vertical processes. For example, the regional ocean model can be a model such as ROMS.

[0079] For example, regional ocean models can also be configured regionally for FVCOM, MITgcm, HYCOM, or other equivalent models capable of simulating the three-dimensional dynamic environment field of the target sea area.

[0080] For example, in step S210, the regional ocean model construction module 12 determines a preset regional ocean model based on the target sea area and computing resources. The regional ocean model construction module 12 determines the regional ocean model to be a ROMS model based on the target sea area and computing resources.

[0081] In step S220, a three-dimensional numerical grid is established based on the predetermined regional ocean model and preprocessed standard data to obtain the basic marine dynamic environment field.

[0082] For example, in step S220, the regional ocean model construction module 12 establishes a three-dimensional numerical grid based on the predetermined regional ocean model and preprocessed standard data to obtain the basic ocean dynamic environment field.

[0083] The regional ocean model construction module 12 establishes a curved orthogonal horizontal grid and S-coordinate vertical layers within the regional ocean model. To balance the ability to represent complex seafloor topography with the resolution of key areas, local layers can be densified according to modeling requirements such as the surface mixing layer, bottom boundary current, and thermocline. The regional ocean model construction module 12 takes the corresponding topographic data, forcing field, initial field, and boundary field standard data from the preprocessed standard data as input to the regional ocean model. The regional ocean model performs three-dimensional numerical simulations of the current field, temperature and salinity structure, and their temporal evolution in the target sea area. The simulation output includes velocity vectors, temperature fields, and salinity fields at least at multiple depth layers, forming a basic marine dynamic environment field covering the entire mission sea area and simulation time window.

[0084] Through the above embodiments, the technical solution of this application can establish a three-dimensional numerical grid using a regional ocean model, thereby obtaining a basic ocean dynamic environment field. Compared with schemes that rely solely on discrete observations, the basic ocean dynamic environment field of this application has the advantages of good continuity, wide coverage, and ease of rolling updates. Simultaneously, the basic ocean dynamic environment field provides a clear and physically meaningful training and inference foundation for subsequent time-series prediction networks.

[0085] Optionally, see Figure 4 Step S300 may include steps S310-S330.

[0086] In step S310, the basic marine dynamic environment field is constructed into a continuous sample sequence in chronological order.

[0087] According to the example embodiment, a continuous sample sequence can be the smallest data unit of a time-series prediction network.

[0088] For example, in step S310, the time-series prediction module 13 constructs the basic marine dynamic environment field into a continuous sample sequence according to time order. Each sample in the continuous sample sequence can be composed of environmental field tensors at multiple consecutive time points. The environmental field tensor can be a multidimensional array formed by arranging marine environmental variables within a certain time window according to time, depth, horizontal grid, and variable channels. The environmental field tensor can be represented as: [time step, depth layer, latitudinal grid, radial grid, variable channel] or an equivalent format.

[0089] The environmental field tensor can be obtained by interpolating the basic marine dynamic environmental field to a unified grid. The channels of the environmental field tensor can include major environmental elements such as east-west current velocity, north-south current velocity, vertical current velocity, temperature, and salinity, and other environmental elements such as sea surface height, density, turbulence mixing parameters, wind stress, heat flux, freshwater flux, seabed depth, or land-sea mask can be added as needed for the mission.

[0090] In step S320, three-dimensional convolutional feature extraction is performed on the continuous sample sequence to obtain the spatiotemporal correlation feature tensor.

[0091] According to the example embodiment, the spatiotemporal correlation feature tensor can be a local spatiotemporal feature tensor after three-dimensional convolution feature extraction of a continuous sample sequence.

[0092] For example, in step S320, the temporal prediction module 13 performs three-dimensional convolutional feature extraction on the continuous sample sequence to obtain a spatiotemporal correlation feature tensor. The temporal prediction module 13 can extract local spatiotemporal feature tensors by performing joint convolution processing on the continuous sample sequence through the three-dimensional convolutional feature extraction layer.

[0093] In step S330, the spatiotemporal correlation feature tensor is input into the time series prediction network to obtain the environmental field prediction results for future time periods based on the continuous environmental field sample generation rules.

[0094] According to the example implementation, the temporal prediction network can be a Conv-LSTM network (Convolutional Long Short-Term Memory network). The Conv-LSTM network possesses both convolutional feature extraction and temporal memory capabilities. The temporal prediction network can also be a ConvGRU, a combination of 3D convolutional and recurrent networks, or other equivalent networks.

[0095] In step S330, the time series prediction module 13 inputs the spatiotemporal correlation feature tensor into the time series prediction network to obtain the environmental field prediction results for future time periods according to the continuous environmental field sample generation rules.

[0096] The temporal prediction module 13 inputs the spatiotemporal correlation feature tensor into the Conv-LSTM network. While maintaining the spatial convolutional structure, the Conv-LSTM learns the temporal evolution law (i.e., the continuous environmental field sample generation rule) and outputs the marine vector environmental field prediction results for one or more future time steps (i.e., the environmental field prediction results for future periods).

[0097] When the target sea area has obvious seasonal, interannual or decadal scale variation characteristics, the Conv-LSTM network can introduce an attention mechanism to assign higher weights to features that better match the actual evolution of the current target sea area, thereby suppressing irrelevant or mismatched features.

[0098] For example, in nearshore waters significantly influenced by monsoons, the weight of features similar to historical samples of the same season can be increased, while the weight of features associated with off-season flow patterns can be decreased, based on the current month or season label. Similarly, in waters exhibiting multi-year periodic temperature and salinity variations, an attention coefficient can be calculated based on the similarity between the current observation profile and historical interannual samples, allowing features that more closely resemble current measured trends in thermocline location, flow direction changes, or salinity gradients to occupy a larger proportion of the predictions.

[0099] Through the above embodiments, the technical solution of this application can construct a continuous sample sequence of the basic marine dynamic environment field in chronological order, and then perform three-dimensional convolutional feature extraction to obtain a spatiotemporal correlation feature tensor. The technical solution of this application can input the spatiotemporal correlation feature tensor into a temporal prediction network to obtain the environmental field prediction results for future periods according to the continuous environmental field sample generation rules.

[0100] The technical solution of this application can take into account both spatial local structure and temporal neighborhood correlation through the spatiotemporal correlation feature tensor obtained by 3D convolution feature extraction. The technical solution of this application can utilize a temporal prediction network, which can better balance the continuous evolution of the environmental field and the prediction needs of future periods, thereby improving the accuracy and foresight of subsequent modeling results.

[0101] Optionally, see Figure 5 Step S400 includes steps S410-S420.

[0102] In step S410, the multi-source observation data is spatiotemporally aligned to obtain aligned multi-source observation data.

[0103] According to the example embodiment, the aligned multi-source observation data can be the data after spatiotemporal alignment of the multi-source observation data.

[0104] Multi-source observation data can include flow field, temperature, salinity and meteorological observation data, as well as relevant observations obtained from ADCP, DVL, profile buoys, CTD, satellite remote sensing, shore-based radar, ocean stations or shipborne sensors.

[0105] For example, in step S410, the data assimilation and correction module 14 performs spatiotemporal alignment on the multi-source observation data to obtain aligned multi-source observation data.

[0106] The data assimilation and correction module 14 maps multi-source observation data onto a model grid and a unified time axis. For example, multi-source observation data is first mapped to a model grid and a unified time axis consistent with the basic ocean dynamic environment field and the predicted environment field for future periods. For spatially discrete observation data, the data assimilation and correction module 14 can perform two-dimensional interpolation sequentially by depth layer, and then perform one-dimensional interpolation along the vertical direction. For temporally discrete observation data, the data assimilation and correction module 14 can perform interpolation processing on a time series basis at each grid point or sampling trajectory. For example, the data assimilation and correction module 14 can use cubic spline interpolation to preserve the trend of environmental field changes while ensuring smoothness.

[0107] In step S420, parameter estimation and recursive correction are performed on the aligned multi-source observation data, the basic marine dynamic environment field, and the predicted environment field for future periods to obtain the target marine vector environment field.

[0108] For example, in step S420, the data assimilation and correction module 14 performs parameter estimation and recursive correction on the aligned multi-source observation data, the basic marine dynamic environment field, and the environmental field prediction results for future periods to obtain the target marine vector environment field.

[0109] The data assimilation and correction module 14 estimates key model parameters using gradient descent. These key model parameters can be adjustable parameters that affect the environmental field output in the regional ocean model and multi-scale data assimilation process. They can also include weighting parameters when network prediction results are fused. For example, key model parameters may include the vertical mixing coefficient, horizontal diffusion coefficient, bottom friction coefficient, boundary relaxation coefficient, forcing field correction coefficient, observation error covariance, model error covariance, and the fusion weights of numerical simulation results and network prediction results. The data assimilation and correction module 14 recursively corrects state variables such as the flow field, temperature field, and salinity field using Kalman filtering, ultimately obtaining the target ocean vector environmental field. The target ocean vector environmental field is more consistent with the observation data and more coordinated with the physical model.

[0110] For existing multi-source observation data corresponding to specific time points, the data assimilation and correction module 14 can use the multi-source observation data to perform parameter estimation and recursive correction to obtain the target ocean vector environment field. Future time points are then recursively updated based on the corrected target ocean vector environment field.

[0111] Through the above embodiments, the technical solution of this application can perform spatiotemporal alignment of multi-source observation data to obtain aligned multi-source observation data. The technical solution of this application can perform parameter estimation and recursive correction on the aligned multi-source observation data, the basic marine dynamic environment field, and the predicted environmental field for future periods to obtain the target marine vector environment field.

[0112] The technical solution of this application can reduce the systematic bias caused by simple numerical simulation and avoid the problem of lack of physical constraints in simple network prediction.

[0113] Optionally, see Figure 6 After step S400, the modeling method also includes step S500.

[0114] In step S500, the target ocean vector environment field and the seabed topography model of the target sea area are coupled to obtain the three-dimensional environment model of the target sea area and the target output results.

[0115] According to the example embodiment, the seabed topography model can be a three-dimensional bottom boundary numerical model generated from the original topographic and water depth data of the target sea area, which corresponds one-to-one with the computational grid within the regional ocean model on the horizontal node index. For example, the seabed topography model can be generated from ETOPO series topographic data or other water depth measurement data. Through steps such as cropping the target sea area, coordinate unification, grid interpolation, land-sea mask generation, and water depth anomaly correction, a seabed elevation or water depth raster consistent with the regional ocean model grid is obtained.

[0116] The 3D environment model can be a coupled environmental field model of the target ocean area, derived from the vector environmental field of the target ocean and the seabed topography model of the target sea area. The target output results include multi-depth layer current velocity vectors, temperature distribution, and salinity distribution of the target sea area.

[0117] For example, in step S500, the terrain coupling and output module 15 couples the target ocean vector environment field and the seabed topography model of the target sea area to obtain the three-dimensional environment model of the target sea area and the target output result.

[0118] The terrain coupling and output module 15 couples the target ocean vector environment field and the seabed topography model of the target sea area to obtain the three-dimensional environment model and target output results.

[0119] The terrain coupling and output module 15 can continuously update the 3D environment model according to a preset time step, and also supports offline batch generation for specific task sea areas.

[0120] The 3D environment model and target output results can be sent to the AUV navigation compensation module, path planning module, mission simulation module, or environment visualization module via standard interfaces. For example, in the navigation compensation scenario, the flow velocity vectors at different depth layers can be directly called to correct the AUV trajectory; in the path planning scenario, the 3D environment model can be used to evaluate the environmental costs of different routes; in the mission simulation scenario, the 3D environment model can be used as an external condition for game or decision simulation.

[0121] Through the above embodiments, the technical solution of this application can couple the target ocean vector environment field and the seabed topography model of the target sea area to obtain the three-dimensional environment model of the target sea area and the target output result.

[0122] The technical solution of this application incorporates data assimilation and terrain coupling into the overall technical approach, making the final output results closer to the real environment, and can directly serve AUV navigation compensation, path planning and mission simulation, and has strong engineering application value.

[0123] According to another aspect of this application, this application also provides a non-volatile computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is capable of implementing the modeling method for the marine vector environmental field as described above.

[0124] According to another aspect of this application, this application also provides an electronic device, including: one or more processors; and a storage device for storing one or more programs, which, when executed by the one or more processors, enable the one or more processors to implement the modeling method for the marine vector environmental field as described above.

[0125] According to another aspect of this application, this application also provides a computer program product, comprising: a computer program stored on a computer-readable storage medium; the computer program includes program instructions that, when executed by a computer, cause the computer to perform the modeling method for the marine vector environmental field as described above.

[0126] Finally, it should be noted that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions of the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for modeling a marine vector environmental field, characterized in that, The modeling method includes: The collected topographic data, forcing field, initial field, boundary field, and multi-source observation data of the target sea area are preprocessed to obtain preprocessed standard data; Based on the preset regional ocean model of the target sea area and the preprocessed standard data, the basic marine dynamic environment field is determined; Based on the continuous environmental field sample generation rules of the pre-constructed time-series prediction network and the basic marine dynamic environmental field, the environmental field prediction results for future periods are determined. The multi-source observation data, the basic marine dynamic environment field, and the predicted environmental field for the future time period are assimilated at multiple scales to obtain the target marine vector environment field.

2. The modeling method according to claim 1, characterized in that, The collected topographic data, forcing field, initial field, boundary field, and multi-source observation data of the target sea area are preprocessed to obtain preprocessed standard data, including: The terrain data, the forcing field, the initial field, the boundary field, and the multi-source observation data are projected onto a unified coordinate system to obtain the first standard data; The first standard data is mapped to a unified time step to obtain the second standard data; The second standard data is format-converted to obtain the third standard data that meets the input requirements; The preprocessing standard data is determined based on the third standard data and the preset quality rules.

3. The modeling method according to claim 1, characterized in that, The step of determining the basic marine dynamic environment field based on the preset regional ocean model of the target sea area and the preprocessed standard data includes: The preset regional ocean model is determined based on the target sea area and computing resources; Based on the predetermined regional ocean model and the preprocessed standard data, a three-dimensional numerical grid is established to obtain the basic marine dynamic environment field.

4. The modeling method according to claim 3, characterized in that, The regional ocean model is configured to support curved orthogonal horizontal grids and S-coordinate vertical stratification.

5. The modeling method according to claim 1, characterized in that, The step of determining the environmental field prediction results for future periods based on the continuous environmental field sample generation rules of the pre-constructed time-series prediction network and the basic marine dynamic environmental field includes: The basic marine dynamic environment field is constructed as a continuous sample sequence in chronological order; Three-dimensional convolutional feature extraction is performed on the continuous sample sequence to obtain a spatiotemporal correlation feature tensor; The spatiotemporal correlation feature tensor is input into the time series prediction network to obtain the environmental field prediction result for the future time period according to the continuous environmental field sample generation rule.

6. The modeling method according to claim 1, characterized in that, The process of assimilating the multi-source observation data, the basic marine dynamic environment field, and the predicted environmental field for future time periods into a multi-scale data assimilation method to obtain the target marine vector environment field includes: The multi-source observation data is spatiotemporally aligned to obtain aligned multi-source observation data; The aligned multi-source observation data, the basic marine dynamic environment field, and the predicted environmental field for the future time period are subjected to parameter estimation and recursive correction to obtain the target marine vector environment field.

7. The modeling method according to claim 1, characterized in that, After performing multi-scale data assimilation on the multi-source observation data, the basic ocean dynamic environment field, and the predicted environmental field for the future time period to obtain the target ocean vector environment field, the modeling method further includes: The target ocean vector environment field and the seabed topography model of the target sea area are coupled to obtain the three-dimensional environment model of the target sea area and the target output result; The target output results include multi-depth layer current velocity vectors, temperature distribution, and salinity distribution of the target sea area.

8. A modeling device for marine vector environmental fields, characterized in that, The modeling device includes: The preprocessing module preprocesses the collected topographic data, forcing field, initial field, boundary field, and multi-source observation data of the target sea area to obtain preprocessed standard data. The regional ocean model construction module determines the basic marine dynamic environment field based on the preset regional ocean model of the target sea area and the preprocessed standard data. The time-series prediction module determines the environmental field prediction results for future periods based on the continuous environmental field sample generation rules of the pre-constructed time-series prediction network and the basic marine dynamic environmental field. The data assimilation and correction module performs multi-scale data assimilation on the multi-source observation data, the basic marine dynamic environment field, and the predicted environmental field for the future period to obtain the target marine vector environment field.

9. A non-volatile computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the modeling method for the marine vector environmental field as described in any one of claims 1-7.

10. An electronic device, characterized in that, include: One or more processors; Storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method for modeling the marine vector environmental field as described in any one of claims 1-7.

Images