Intelligent reconstruction and prediction method and system for ocean three-dimensional flow field

Abstract

The invention discloses an intelligent reconstruction and prediction method and system for an ocean three-dimensional flow field, and relates to the technical field of data processing, and the method comprises the steps: obtaining the multi-source observation data of an ocean through a large language model and an edge monitoring device; preprocessing and fusing the multi-source observation data to obtain fused observation data; taking the target observation parameters in the fused observation data as nodes, and taking the space-time correlation and physical quantity coupling relationship among the target observation parameters as edges to construct a space-time correlation map; reconstructing and determining reconstruction data corresponding to the multi-source observation data according to the space-time correlation atlas; the three-dimensional flow field of the ocean is obtained through prediction according to the reconstruction data, and a visual three-dimensional flow field is output. According to the method, the multi-source observation data are fused and reconstructed, and the accuracy of predicting the ocean three-dimensional flow field can be improved.

Description

Technical Field

[0001] This application relates to the field of data processing technologies, and in particular, to a method and system for intelligent reconstruction and prediction of ocean three-dimensional flow fields. Background Art

[0002] Ocean three-dimensional flow field prediction is a core area of ocean dynamics research and is of great value for climate change simulation, pollution diffusion prediction, resource development, and military strategies. However, traditional prediction methods face multiple technical bottlenecks: ① Scarcity of observational data: Although satellite remote sensing technology can cover the globe, it can only obtain surface information. The reconstruction of the underwater three-dimensional flow field relies on high-cost and high-platform-stability profile measurements (such as ADCP and current meters), and data resources are scarce. ② High computational complexity of the model: Traditional numerical models (such as POM and HYCOM) need to solve based on physical equations, with extremely high computational costs and limited accuracy, making it difficult to meet real-time requirements. ③ Difficulty in multi-variable coupling: The ocean flow field involves the spatio-temporal coupling of multiple variables such as temperature, salinity, and wind stress. Traditional statistical methods (such as ARIMA and SVR) are difficult to capture non-linear relationships and have short prediction time horizons. In recent years, artificial intelligence technology has provided new ideas for ocean flow field prediction, but existing solutions still have limitations: ① Limitation of single modality: Some methods only focus on the reconstruction of the surface flow field or vertical flow velocity, lacking an overall representation of the three-dimensional full flow field. ② Strong data dependence: Deep learning models require a large amount of historical data support, are prone to overfitting, and have poor interpretability. Summary of the Invention

[0003] In view of this, embodiments of this application provide a method and system for intelligent reconstruction and prediction of ocean three-dimensional flow fields to improve the accuracy of predicting ocean three-dimensional flow fields.

[0004] One aspect of the embodiments of this application provides a method for intelligent reconstruction and prediction of ocean three-dimensional flow fields, and the method includes the following steps:

[0005] Obtain multi-source observational data of the ocean through a large language model and edge monitoring devices;

[0006] Preprocess and fuse the multi-source observational data to obtain fused observational data;

[0007] Construct a spatio-temporal correlation map with the target observational parameters in the fused observational data as nodes and the spatio-temporal correlation and physical quantity coupling relationships between the target observational parameters as edges;

[0008] Reconstruct and determine the reconstructed data corresponding to the multi-source observational data according to the spatio-temporal correlation map;

[0009] Predict the three-dimensional flow field of the ocean based on the reconstructed data and output the visualized three-dimensional flow field.

[0010] In some embodiments, obtaining multi-source observation data of the ocean through a large language model and edge monitoring devices includes the following steps:

[0011] Use the large language model to parse the user instructions, and then obtain the profile observation data corresponding to the user instructions; project the profile observation data into the spatio-temporal grid of the satellite altimeter, and then calculate the spatio-temporal matching degree of the profile observation data; delete the profile observation data with the spatio-temporal matching degree lower than the set threshold to obtain the first observation data;

[0012] Use the ADCP device in the edge monitoring device to obtain underway ADCP data; use Kalman filtering to remove the noise of the underway ADCP data, and then perform flow velocity profile reconstruction on the underway ADCP data through the edge GPU to obtain the second observation data;

[0013] Wherein, the first observation data and the second observation data are used as the multi-source observation data.

[0014] In some embodiments, preprocessing and fusing the multi-source observation data to obtain fused observation data includes the following steps:

[0015] Extract barotropic mode features and first baroclinic mode features from the multi-source observation data;

[0016] Set the three-dimensional grid and boundary conditions of the FVCOM model;

[0017] Project the barotropic mode features and the first baroclinic mode features into the three-dimensional grid of the FVCOM numerical model for spatio-temporal alignment;

[0018] Adjust the quality threshold according to the task type, and perform data cleaning on the barotropic mode features and the first baroclinic mode features after spatio-temporal alignment according to the quality threshold to obtain quality assessment features;

[0019] Use the edge monitoring device to extract vertical mode features from the obtained multi-source observation data;

[0020] Extract Ekman flow features coupled with the barotropic mode features from the multi-source observation data obtained by the satellite in the edge monitoring device;

[0021] Fuse the quality assessment features, the vertical mode features and the Ekman flow features to obtain the fused observation data.

[0022] In some embodiments, the method further includes the following steps:

[0023] Adjust the feature weights of the target observation parameters according to the task requirements to adjust the corresponding edges in the spatio-temporal association graph.

[0024] In some embodiments, reconstructing the reconstruction data corresponding to the multi-source observation data according to the spatio-temporal correlation map includes the following steps:

[0025] Output a gridded hydrodynamic field through the FVCOM model based on the spatio-temporal correlation map;

[0026] Input the hydrodynamic field in the form of time series data into the LSTM network, embed physical constraint terms into the LSTM network, and dynamically adjust the sampling frequency through an event trigger mechanism;

[0027] Calculate the runoff contribution based on the SCS-CN model, simulate the vertical velocity distribution through the generalized unit hydrograph, introduce a multi-scale attention mechanism, and dynamically allocate computing resources to the mesoscale eddy region;

[0028] Optimize the parameters of the FVCOM model.

[0029] In some embodiments, optimizing the parameters of the FVCOM model includes the following steps:

[0030] Extract the features of the three-dimensional flow field output by the FVCOM model and the fusion observation data respectively and calculate the residuals;

[0031] Construct a state space according to the residuals and enable the FVCOM model to perform reinforcement learning;

[0032] Dynamically adjust the weights of the FVCOM model through a policy network and a value network; wherein, the policy network is used to input the residual feature map and output the action probability distribution; the value network is used to input the state vector and output the expected reward;

[0033] Coordinate and optimize the FVCOM model based on the residual feature map transmitted by the edge monitoring device.

[0034] In some embodiments, outputting the visualized three-dimensional flow field includes the following steps:

[0035] Output the contour line or vector map of the three-dimensional flow field through a digital twin engine, an AR / VR visualization terminal, a BIM-GIS collaboration platform or the edge monitoring device.

[0036] Another aspect of the embodiments of the present application further provides a system for intelligent reconstruction and prediction of an ocean three-dimensional flow field, the system includes:

[0037] A data acquisition unit, configured to obtain multi-source observation data of the ocean through a large language model and an edge monitoring device;

[0038] A data fusion unit for preprocessing and fusing the multi-source observation data to obtain fused observation data;

[0039] A map construction unit for constructing a spatio-temporal correlation map with the target observation parameters in the fused observation data as nodes and the spatio-temporal correlation and physical quantity coupling relationships between the target observation parameters as edges;

[0040] A data reconstruction unit for reconstructing and determining the reconstructed data corresponding to the multi-source observation data according to the spatio-temporal correlation map;

[0041] A prediction output unit for predicting the three-dimensional flow field of the ocean based on the reconstructed data and outputting the visualized three-dimensional flow field.

[0042] Another aspect of the embodiments of the present application further provides an electronic device, including a processor and a memory;

[0043] The memory is used to store programs;

[0044] The processor executes the program to implement the method described in any one of the above.

[0045] Another aspect of the embodiments of the present application further provides a computer-readable storage medium, where the storage medium stores a program, and the program is executed by a processor to implement the method described in any one of the above.

[0046] The present application at least includes the following beneficial effects:

[0047] The present application can obtain multi-source observation data of the ocean through a large language model and edge monitoring devices; preprocess and fuse the multi-source observation data to obtain fused observation data; construct a spatio-temporal correlation map with the target observation parameters in the fused observation data as nodes and the spatio-temporal correlation and physical quantity coupling relationships between the target observation parameters as edges; reconstruct and determine the reconstructed data corresponding to the multi-source observation data according to the spatio-temporal correlation map; predict the three-dimensional flow field of the ocean based on the reconstructed data and output the visualized three-dimensional flow field. The present application performs fusion reconstruction on multi-source observation data, which can improve the accuracy of predicting the three-dimensional flow field of the ocean. Description of the Drawings

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0049] Figure 1 It is a schematic flowchart of a method for intelligent reconstruction and prediction of a three-dimensional ocean flow field provided by an embodiment of the present application;

[0050] Figure 2 This is a schematic structural diagram of a system for intelligent reconstruction and prediction of a three-dimensional ocean current field provided by an embodiment of the present application. Detailed implementation manners

[0051] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, but not to limit the present application.

[0052] Before elaborating on the embodiments of the present application, some related technologies involved in the embodiments of the present application are described as follows:

[0053] Related technologies:

[0054] (1) Satellite remote sensing and Argo buoy combined reconstruction method:

[0055] Based on the geostrophic current and Ekman current inverted by satellite altimeter and scatterometer, combined with the 1000-meter depth data of Argo buoys, a three-dimensional current field is reconstructed by extracting the barotropic mode and the first baroclinic mode. This method verifies the mapping relationship between satellite data and Argo data, but relies on multi-source data fusion and does not solve the problem of direct remote sensing detection of sea surface vector currents.

[0056] (2) Traditional three-dimensional current field numerical simulation:

[0057] Numerical models such as the Princeton Ocean Model (POM) are used to simulate the current field by discretely solving the motion equations. Its advantage lies in the clear physical mechanism, but it consumes a large amount of computing resources and it is difficult to achieve high-resolution real-time prediction.

[0058] Some other related technologies:

[0059] (1) EOF-LSTM coupled model for surface current field prediction:

[0060] Existing methods extract spatial modes through empirical orthogonal functions (EOF) and combine long short-term memory networks (LSTM) to process time series data for ocean surface current field prediction. However, this method does not consider multi-variable coupling and is only applicable to two-dimensional space.

[0061] (2) Deep learning model for vertical velocity reconstruction:

[0062] A deep convolutional neural network (CNN) based on sea surface height field is proposed to directly predict the vertical velocity field. Compared with traditional dynamic diagnostic methods (such as eSQG), it has higher accuracy and smaller data requirements, but only focuses on vertical velocity reconstruction.

[0063] (3)Transformer Model for Multivariable Three-Dimensional Field Prediction:

[0064] Construct the 3D-Geoformer model, which uses spatio-temporal attention mechanism to simulate the multi-variable three-dimensional fields of the ocean and atmosphere (such as ENSO-related sea surface temperature field and wind stress field), and the prediction time limit exceeds 18 months. This method breaks through the limitations of traditional RNNs, but it relies on multi-variable datasets in specific regions.

[0065] The existing technologies have made progress in satellite data fusion, numerical simulation optimization, and deep learning applications, but there are still problems such as strong data dependence, insufficient multi-variable coupling, and separation of vertical flow fields and full three-dimensional flow fields. This application aims to combine multi-source data and intelligent algorithms to solve the above technical bottlenecks and achieve intelligent prediction of ocean three-dimensional flow fields with high precision and high timeliness.

[0066] Disadvantages of the existing technologies:

[0067] (1)Strong data dependence and high acquisition cost: Traditional three-dimensional flow field reconstruction relies on high-cost profile measurements (such as ADCP, current meters) with high platform stability, and also requires the fusion of multi-source data (satellite + Argo), and data resources are scarce.

[0068] (2)High computational complexity and difficult to predict in real time: Numerical models (such as POM) need to solve based on physical equations, consuming a large amount of computing resources and making it difficult to achieve high-resolution real-time prediction.

[0069] (3)Insufficient multi-variable coupling and three-dimensional full flow field characterization: Existing methods (such as EOF-LSTM, vertical velocity reconstruction models) only focus on a single variable (such as surface flow field or vertical velocity), and do not achieve the overall characterization of the three-dimensional full flow field and the spatio-temporal coupling of multi-variables.

[0070] (4)Poor model interpretability and easy to overfit: Although deep learning models can handle complex relationships, they rely on large-scale historical data, are easy to overfit, and lack physical mechanism explanations.

[0071] (5)Lack of direct remote sensing detection technology for sea surface vector flow: Traditional methods rely on indirect inversion (such as geostrophic flow + Ekman flow) and do not solve the problem of high-precision direct observation of sea surface vector flow.

[0072] The invention objective of this application is, aiming at the above disadvantages, this application aims to:

[0073] (1)Reduce data dependence and acquisition cost: By optimizing the mapping relationship between satellite remote sensing data and Argo buoy data, reduce the dependence on high-cost profile measurement data and achieve efficient three-dimensional flow field reconstruction.

[0074] (2) Improve computational efficiency and real-time performance: Combine intelligent algorithms (such as deep learning) to replace the solution of some physical equations, reduce computational complexity, and meet the requirements of high-resolution real-time prediction.

[0075] (3) Achieve three-dimensional full flow field and multi-variable coupling characterization: Through multi-modal data fusion and intelligent algorithm design, break through the limitations of traditional models on the coupling of vertical flow field, horizontal flow field and multi-variables (temperature, salinity, wind stress), and construct a full three-dimensional flow field prediction system.

[0076] (4) Enhance model interpretability and stability: Introduce physical prior knowledge into the deep learning framework, balance data-driven and physical constraints, improve the generalization ability of the model and reduce the risk of overfitting.

[0077] (5) Directly detect sea surface vector flow: Develop a vector flow inversion technology based on satellite altimeters and scatterometers to solve the problem of direct remote sensing observation of sea surface flow fields and improve the quality of data sources.

[0078] Refer to Figure 1 , the embodiments of the present application provide a method for intelligent reconstruction and prediction of ocean three-dimensional flow fields, specifically including the following steps S100~S140:

[0079] S100: Obtain multi-source observation data of the ocean through a large language model and edge monitoring devices;

[0080] S110: Preprocess and fuse the multi-source observation data to obtain fused observation data;

[0081] S120: Construct a spatio-temporal correlation map with the target observation parameters in the fused observation data as nodes and the spatio-temporal correlation and physical quantity coupling relationships between the target observation parameters as edges;

[0082] S130: Reconstruct and determine the reconstruction data corresponding to the multi-source observation data according to the spatio-temporal correlation map;

[0083] S140: Predict the three-dimensional flow field of the ocean based on the reconstruction data and output the visualized three-dimensional flow field.

[0084] Optionally, the obtaining of multi-source observation data of the ocean through a large language model and edge monitoring devices includes the following steps:

[0085] Use the large language model to parse the user instructions, and then obtain the profile observation data corresponding to the user instructions; project the profile observation data into the spatio-temporal grid of the satellite altimeter, and then calculate the spatio-temporal matching degree of the profile observation data; delete the profile observation data with the spatio-temporal matching degree lower than the set threshold to obtain the first observation data;

[0086] Obtain the moving ADCP data using the ADCP device in the edge monitoring device; use Kalman filtering to remove the noise of the moving ADCP data, and then perform flow velocity profile reconstruction on the moving ADCP data through the edge GPU to obtain the second observation data;

[0087] Among them, the first observation data and the second observation data are used as the multi-source observation data.

[0088] Optionally, the preprocessing and fusion of the multi-source observation data to obtain the fused observation data includes the following steps:

[0089] Extract the barotropic mode features and the first baroclinic mode features from the multi-source observation data;

[0090] Set the three-dimensional grid and boundary conditions of the FVCOM model;

[0091] Project the barotropic mode features and the first baroclinic mode features onto the three-dimensional grid of the FVCOM numerical model for spatio-temporal alignment;

[0092] Adjust the quality threshold according to the task type, and perform data cleaning on the barotropic mode features and the first baroclinic mode features after spatio-temporal alignment according to the quality threshold to obtain the quality assessment features;

[0093] Use the edge monitoring device to extract the vertical mode features from the obtained multi-source observation data;

[0094] Extract the Ekman flow features coupled with the barotropic mode features from the multi-source observation data obtained by the satellite in the edge monitoring device;

[0095] Fuse the quality assessment features, the vertical mode features and the Ekman flow features to obtain the fused observation data.

[0096] Optionally, the method further includes the following steps:

[0097] Adjust the feature weights of the target observation parameters according to the task requirements to adjust the corresponding edges in the spatio-temporal association graph.

[0098] Optionally, the reconstruction of the multi-source observation data corresponding to the reconstructed data according to the spatio-temporal association graph reconstruction includes the following steps:

[0099] Output the gridded hydrodynamic field through the FVCOM model based on the spatio-temporal association graph;

[0100] Input the hydrodynamic field in the form of time series data into the LSTM network, embed physical constraint terms into the LSTM network, and dynamically adjust the sampling frequency through an event trigger mechanism;

[0101] Calculate the runoff contribution based on the SCS-CN model, simulate the vertical velocity distribution through the generalized unit hydrograph, introduce a multi-scale attention mechanism, and dynamically allocate computing resources to the mesoscale eddy region;

[0102] Optimize the parameters of the FVCOM model.

[0103] Optionally, the optimizing the parameters of the FVCOM model includes the following steps:

[0104] Extract the features of the three-dimensional flow field output by the FVCOM model and the fusion observation data respectively and calculate the residuals;

[0105] Construct a state space based on the residuals and enable the FVCOM model to perform reinforcement learning;

[0106] Dynamically adjust the weights of the FVCOM model through a policy network and a value network; wherein, the policy network is used to input the residual feature map and output the action probability distribution; the value network is used to input the state vector and output the expected reward;

[0107] Perform coordinated optimization on the FVCOM model based on the residual feature map transmitted by the edge monitoring device.

[0108] Optionally, the outputting the visualized three-dimensional flow field includes the following steps:

[0109] Output the contour line or vector map of the three-dimensional flow field through a digital twin engine, an AR / VR visualization terminal, a BIM-GIS collaborative platform or the edge monitoring device.

[0110] Next, specific application examples will be combined to introduce and explain the solution of the embodiment of the present application in detail.

[0111] The solution of this embodiment includes 5 core modules, and each module is introduced in detail as follows:

[0112] Core Module 1: Multi-source Observation Data Acquisition Layer.

[0113] Integrate data sources such as satellite remote sensing (altimeter / scatterometer), shore-based high-frequency ground wave radar, moving ship ADCP, meteorological station, Argo buoy, etc. Propose a "command-driven-dynamic evaluation-edge collaboration" trinity architecture, and for the first time realize the intelligent, adaptive and low-latency characteristics of ocean multi-source data acquisition. The constructed module components are:

[0114] ① Instruction parsing engine: A special model for the marine domain fine-tuned based on GPT-4, which parses user instructions to generate a data requirement JSON (such as "Obtain ADCP flow velocity data in the East China Sea region at 14:00 on March 11, 2025, with an accuracy of ±0.1 m / s").

[0115] ② Dynamic data quality assessment module:

[0116] Spatiotemporal consistency assessment: Calculate the spatiotemporal matching degree of data sources through the DUViT spatiotemporal attention mechanism (improvement of image registration technology).

[0117] Multi-dimensional quality indicators: Include multiple indicators such as satellite data inversion error (radiometric correction method), ADCP profile integrity (data integrity detection algorithm), etc.

[0118] ③ Edge collaborative network: Lightweight processing nodes deployed on shore-based radars and moving devices, which perform preprocessing tasks such as data compression and outlier filtering.

[0119] ④ Data fusion center: Construct a semantic association graph of multi-source data based on graph neural networks. The specific implementation steps for the multi-source observation data acquisition layer are as follows:

[0120] Step 1: Instruction parsing and task decomposition, input natural language instruction → Enhance semantic understanding through a domain dictionary (such as mapping "ADCP" to "Acoustic Doppler Current Profiler"), and generate a structured task.

[0121] Step 2: Dynamic data quality assessment, spatiotemporal consistency calculation: Project the ADCP profile data into the spatiotemporal grid of the satellite altimeter through the DUViT model, and calculate the spatiotemporal matching degree:

[0122] 。

[0123] Step 3: Edge collaborative data acquisition.

[0124] Intelligent task distribution: Give high-priority tasks to ADCP devices first (matching degree > 0.85 and accuracy meets the standard), and adopt an incremental update mode for meteorological stations (only transmit parameters that change by more than ±5%).

[0125] Edge preprocessing pipeline: Moving ADCP data: Use Kalman filtering to remove noise (data cleaning technology), and perform flow velocity profile reconstruction through an edge GPU (optimization of the spatial analysis engine).

[0126] Shore-based radar data: Automatically complete missing sectors based on the spatiotemporal association graph (improvement of image fusion method).

[0127] Step 4: Data fusion and quality verification.

[0128] Constructing a semantic association graph for multi-source data (extension of semantic alignment technology): Nodes: ADCP flow velocity, satellite altimeter geostrophic flow, meteorological station wind stress; Edges: spatio-temporal correlation (Pearson correlation coefficient), physical quantity coupling relationship.

[0129] Quality verification: If the quality score of a data source is < 0.6, the backup data source (such as Argo buoy) is automatically triggered for supplementation, and the data source weights are dynamically adjusted through reinforcement learning (real-time data warehouse technology).

[0130] Core module 2: Data preprocessing and fusion layer.

[0131] Satellite data: Inverting the surface current field through geostrophic flow + Ekman flow;

[0132] Profile data: ADCP and Argo buoy data are used for vertical mode extraction;

[0133] Meteorological data: Input of driving factors such as wind stress and air pressure.

[0134] Propose a "trinity" architecture of "dynamic mode selection - spatio-temporal alignment enhancement - edge collaborative enhancement" to achieve the intelligent, adaptive and high-precision characteristics of ocean multi-source data preprocessing.

[0135] Main modules for constructing the data preprocessing and fusion layer system:

[0136] ① Dynamic mode selection engine: An improved algorithm based on CEOF analysis to automatically identify the barotropic mode (90% kinetic energy) and the first baroclinic mode.

[0137] ② Spatio-temporal alignment enhancement module: Satellite data: Projected onto the FVCOM grid through the DUViT model; Profile data: Spatio-temporal interpolation is performed using Kalman filter (error < 0.3 m / s).

[0138] ③ Edge collaborative enhancement network: Lightweight processing nodes deployed on ADCP and shore-based radars to perform flow velocity profile reconstruction and feature extraction.

[0139] ④ Multi-modal fusion center: Construct a data semantic association graph based on graph neural network (nodes include multi-dimensional features such as flow velocity, temperature, salinity, etc.).

[0140] The specific implementation steps are as follows:

[0141] Step 1: Dynamic mode selection and initialization. Input the original data set (satellite altimeter, ADCP, meteorological station, etc.), extract the barotropic mode (eigenvalue λ1 > λ2) and the first baroclinic mode (eigenvalue λ2 > λ3) through CEOF analysis, initialize the FVCOM three-dimensional grid, and set boundary conditions (such as tides, wind stress).

[0142] Step 2: Spatiotemporal alignment enhancement and satellite data alignment: Geostrophic currents inverted from altimeter data are projected onto the FVCOM grid, with the spatial resolution increased to 2 km × 2 km. The Ekman currents inverted from scatterometer data are spatiotemporally matched with Argo buoy data through the DUViT model (matching degree > 0.85); Profile data enhancement: ADCP data: Noise is removed using Kalman filtering at the edge nodes (residual < 0.1 m / s), and for underway ADCP, the flow velocity profile reconstruction is performed through edge GPUs.

[0143] Step 3: Dynamic quality assessment of multimodal data, construction of a multi-dimensional quality scoring system, and dynamic threshold setting: The quality threshold is automatically adjusted according to the task type (such as oil spill diffusion prediction).

[0144] Step 4: Edge collaborative modal decomposition, ADCP data: EOF decomposition is performed at the edge nodes to extract vertical modes (the first 3 main modes explain 85% of the energy), and the modal weights are optimized through reinforcement learning (the reward function is based on the FVCOM simulation error).

[0145] Satellite data: After the scatterometer data undergoes complex conversion (us,t = us,t + ivs,t), CEOD decomposition is performed to extract the Ekman flow characteristics coupled with the barotropic mode.

[0146] Step 5: Semantic fusion of multimodal data, construction of a semantic association graph with 18-dimensional features: Nodes: Flow velocity, temperature, salinity, wind stress, air pressure, etc.; Edges: Coupling relationships of physical quantities, spatiotemporal correlations (Pearson coefficient calculation > 0.7).

[0147] Dynamic graph attention mechanism: Automatically adjusts the feature weights according to the task requirements (such as pollution diffusion).

[0148] Step 6: Data assimilation and quality verification, residual analysis: Calculate the residuals between the observed data and the model predictions (threshold ±0.2 m / s). If the residual of a certain data source > the threshold, the backup data source (such as Argo buoy) is automatically triggered for supplementation; Reinforcement learning optimization: Dynamically adjusts the coupling weights between FVCOM and the intelligent model through the PPO algorithm (convergence speed is improved).

[0149] Core module 3: Intelligent reconstruction engine.

[0150] PI-LSTM time series model: Processes multivariate time series data (such as temperature, salinity, flow velocity), and captures non-linear dynamic characteristics;

[0151] Generalized unit hydrograph module: Simulates the coupling of hydrological processes (such as runoff, evaporation) and three-dimensional flow fields;

[0152] FVCOM physical model: Serves as the basic dynamic framework and provides a gridded hydrodynamic field.

[0153] Propose a "command-driven - dynamic evaluation - edge collaboration" trinity architecture to achieve the intelligent, adaptive, and low-latency characteristics of ocean multi-source data collection.

[0154] Intelligent reconstruction engine system architecture:

[0155] (1) Instruction parsing engine: A domain-specific model for the ocean field fine-tuned based on GPT-4, which parses user instructions to generate a data requirement JSON (such as "Obtain ADCP flow velocity data in the East China Sea region at 14:00 on March 11, 2025, with an accuracy of ±0.1 m / s").

[0156] (2) Dynamic data quality assessment module: Spatiotemporal consistency assessment: Calculate the spatiotemporal matching degree of data sources through the DUViT spatiotemporal attention mechanism (improvement of image registration technology); Multi-dimensional quality indicators: Include multiple indicators such as satellite data inversion error (radiometric correction method), ADCP profile integrity (data integrity detection algorithm), etc.

[0157] (3) Edge collaboration network: Lightweight processing nodes deployed on shore-based radars and moving devices, which perform preprocessing tasks such as data compression (adaptive compression algorithm) and outlier filtering.

[0158] (4) Data fusion center: Construct a semantic association graph of multi-source data based on graph neural networks (expansion of semantic alignment technology).

[0159] The specific implementation steps are as follows:

[0160] Step 1: Instruction parsing and task decomposition: Input natural language instructions → Enhance semantic understanding through a domain dictionary (such as mapping "ADCP" to "Acoustic Doppler Current Profiler").

[0161] Step 2: Dynamic data quality assessment, spatiotemporal consistency calculation: Project ADCP profile data into the spatiotemporal grid of satellite altimeters through the DUViT model and calculate the spatiotemporal matching degree.

[0162] Step 3: Edge collaboration data collection, intelligent task distribution: Send high-priority tasks to ADCP devices first (matching degree > 0.85 and accuracy meets the standard), and adopt an incremental update mode for weather stations (only transmit parameters with a change exceeding ±5%); Edge preprocessing pipeline: Moving ADCP data: Use Kalman filtering to remove noise (data cleaning technology), and perform flow velocity profile reconstruction through an edge GPU (optimization of the spatial analysis engine); Shore-based radar data: Automatically complete missing sectors based on the spatiotemporal association graph (improvement of image fusion method).

[0163] Step 4: Data Fusion and Quality Verification, Constructing a Spatiotemporal Association Map of Multi-source Data (Extension of Semantic Alignment Technology): Nodes: ADCP Flow Velocity, Satellite Altimeter Geostrophic Flow, Meteorological Station Wind Stress; Edges: Spatiotemporal Correlation (Pearson Correlation Coefficient), Physical Quantity Coupling Relationship. Quality Verification: If the quality score of a data source is <0.6, the backup data source (such as Argo buoy) is automatically triggered for supplementation, and the data source weights are dynamically adjusted through reinforcement learning (real-time data warehouse technology).

[0164] Step 5: Core Operations of the Intelligent Reconstruction Engine:

[0165] ① Initialization of the FVCOM Physical Model: Generate an unstructured triangular grid based on SMS, set boundary conditions such as tides and wind stress, and extract the barotropic mode (90% kinetic energy) and the first baroclinic mode through CEOF analysis.

[0166] ② PI-LSTM Time Series Modeling: Input time series data such as temperature, salinity, and wind stress into the LSTM network, embed physical constraint terms (such as the momentum equation), and dynamically adjust the sampling frequency through an event trigger mechanism (such as increasing to 1Hz when the flow velocity changes abruptly).

[0167] ③ Generalized Unit Hydrograph Coupling: Calculate the runoff contribution based on the SCS-CN model, simulate the vertical flow velocity distribution through the generalized unit hydrograph (Guh distribution), and introduce a multi-scale attention mechanism to dynamically allocate computing resources to the mesoscale eddy region.

[0168] ④ Data Assimilation and Optimization: Use the PPO algorithm to optimize the coupling weights between FVCOM and the intelligent model, set the residual threshold to ±0.2m / s, and adopt the adaptive grid refinement technology: reduce the grid size from 2km to 500m in the area where the flow velocity gradient >0.5m / s.

[0169] Step 6: Prediction and Visualization Output, Outputting a 72-hour three-dimensional flow field prediction (contour / vector diagram), integrating the pollution diffusion module, and simulating the schematic diagram of the oil spill transport path.

[0170] Core Module 4: Data Assimilation and Optimization Layer.

[0171] The reinforcement learning algorithm (such as PPO) optimizes the model parameters and inverses based on the residuals of the observed data;

[0172] Multi-scale Attention Mechanism: Balancing the global mode (such as the barotropic mode) and local details (such as mesoscale eddies).

[0173] Composition of the Module System Architecture of the Data Assimilation and Optimization Layer:

[0174] (1) Residual Feature Extractor: A multi-scale residual analysis module based on a spatiotemporal attention mechanism.

[0175] (2) Reinforcement learning optimizer: A dynamic weight adjustment engine using the PPO algorithm.

[0176] (3) Multi-scale fusion center: A fusion interface that integrates FVCOM grid data and the prediction results of the intelligent model.

[0177] (4) Edge-center communication layer: Supports lightweight feature transmission.

[0178] Specific implementation process:

[0179] Step 1: Residual calculation and feature extraction. Input the three-dimensional flow field Qpred(x, y, z, t) predicted by the FVCOM model and multi-source observation data O(x, y, z, t) (including satellites, ADCP, Argo, etc.), and output the residual field E(x, y, z, t) = O(x, y, z, t) - Qpred(x, y, z, t).

[0180] Step 2: Construction of the reinforcement learning environment.

[0181] State space:

[0182] The global residual variance is as follows:

[0183] ;

[0184] The local vortex intensity is as follows:

[0185] ;

[0186] Action space:

[0187] The weight adjustment coefficient wd of the data source ∈ [0.1, 0.9];

[0188] The perturbation amount of the model parameters is as follows:

[0189] .

[0190] Step 3: Dynamic weight adjustment, policy network (Actor): Input the residual feature map and output the action probability distribution.

[0191] The policy network (Actor) is the core component in reinforcement learning, and its function is to generate the probability distribution of actions according to the current environmental state. In this embodiment, the policy network adopts a multi-layer perceptron (MLP) architecture enhanced by spatio-temporal attention, and the specific structure is as follows:

[0192] ① Input layer: Receive the multi-scale residual feature map (such as the low-frequency, mid-frequency, and high-frequency features after spatial pyramid decomposition);

[0193] ② Spatiotemporal Attention Layer: Perform multi-scale downsampling on the residual features using three-dimensional spatial pyramid decomposition (scale factors are 1×1×1, 2×2×2, 4×4×4); introduce a spatiotemporal attention module (STAttention), and focus on the key flow field error regions (such as mesoscale vortex regions) through dynamic weight allocation;

[0194] ③ Policy Network Layer: Consists of 3 fully connected layers (FC1: 256→128, FC2: 128→64, FC3: 64→32), with the activation function being Swish; the output layer uses the Softmax function to generate the action probability distribution, and the action space includes: the data source weight adjustment coefficient wd∈[0.1,0.9]; the perturbation amount of model parameters (continuous action).

[0195] The dynamic weight adjustment of the policy network in the dynamic weight adjustment mechanism is the core innovation point of this embodiment, and the specific implementation is as follows:

[0196] ① State Space Modeling:

[0197] Global State: The global variance of the flow field residual is:

[0198] .

[0199] Local State: The vortex intensity is:

[0200] ;

[0201] Extracted through the eigenvalue decomposition of the Hessian matrix.

[0202] ② Action Space Design:

[0203] Data Source Weight Adjustment: Prevent weight degradation through the KL divergence constraint (λ = 0.01) of the PPO algorithm; Dynamically allocate weights: Increase the weight of satellite data in the frontal region and increase the weight of ADCP data in the vortex region; The perturbation range of model parameters is adaptively adjusted according to the residual intensity:

[0204] ;

[0205] (α is 0.1 in the calm region and can reach 0.5 in the strong vortex region).

[0206] ③ Reinforcement Learning Training: The optimization algorithm uses the truncated policy gradient of PPO (clip ratio = 0.2), and updates 500 times per round of iteration.

[0207] Value Network (Critic): Input state vector, output expected reward. Composition of the value network system architecture: ① Spatiotemporal attention layer: Adopt three-dimensional spatial pyramid decomposition (scale factors 1×1×1, 2×2×2, 4×4×4) and STAttention module; ② Multimodal fusion layer: Integrate the semantic association map of 18-dimensional features such as flow velocity, temperature, salinity, and wind stress; ③ Dynamic value evaluation layer: Include physical constraint terms (such as momentum equations) and adaptive weight allocation module; ④ Edge collaboration interface: Support feature quantization compression (SNORM encoding, code rate ≤ 0.1bpp) and secure transmission protocol.

[0208] Specific implementation process of the value network

[0209] Step 1: Input of residual features, input multi-scale residual field E(x, y, z, t) (including three resolutions of 1km, 500m, and 200m); Preprocessing: Extract local spatial features through a three-dimensional convolutional layer, and normalize the residual field to [-1, 1].

[0210] Step 2: Enhancement of spatiotemporal attention, spatial pyramid decomposition and spatiotemporal attention calculation.

[0211] Spatial pyramid decomposition is the core module for realizing multi-scale feature extraction. Its design goal is to capture the global and local features of the flow field error through spatial partitioning at different scales. The specific implementation is as follows: ① Multi-scale spatial partitioning, adopt the three-dimensional spatial pyramid decomposition strategy, and divide the input residual field into three scales: Fine scale (1×1×1): Retain high-resolution detail features (such as the flow velocity gradient in the frontal region); Medium scale (2×2×2): Balance the calculation efficiency and feature expression ability (suitable for the mesoscale vortex region); Coarse scale (4×4×4): Capture large-scale flow field structures (such as the barotropic mode driven by tides).

[0212] The features of each scale are generated through adaptive average pooling, avoiding the problem of losing edge information in traditional max pooling. ② Feature fusion mechanism, adopt the weighted summation strategy to fuse multi-scale features:

[0213] ;

[0214] Among them, the weight coefficients are dynamically adjusted through reinforcement learning, and residual connections are introduced to alleviate the problem of gradient disappearance.

[0215] The spatiotemporal attention mechanism focuses on key error regions through dynamic weight allocation. The specific implementation is as follows:

[0216] ①Three-dimensional spatio-temporal attention module (STAttention). Spatial attention: Three-dimensional convolution is used to extract local spatial features (convolution kernel size 3×3×3), and the correlation within the feature map is calculated through self-attention:

[0217] ;

[0218] Among them, Q, K, and V are generated from the feature maps of flow velocity, temperature-salinity, and wind stress respectively. Temporal attention: The time series is divided into blocks of size b, and long-range dependencies are captured through attention calculation across time steps (such as the influence of tidal cycles).

[0219] ②Dynamic weight allocation strategy. Physical quantity coupling evaluation: Calculate the cross-correlation coefficient of flow velocity-temperature salinity, and dynamically adjust the weight of the vortex intensity index; Reinforcement learning optimization: Use the PPO algorithm to optimize the attention weight, set the residual threshold to ±0.1 m / s, and introduce the KL divergence constraint (λ = 0.02) to prevent uneven weight distribution.

[0220] The innovative technological breakthroughs are as follows:

[0221] (1) Three-dimensional spatio-temporal pyramid attention mechanism, which combines three-dimensional space decomposition with self-attention mechanism to achieve global-local feature adaptive focusing of flow field errors.

[0222] (2) Attention enhancement with multi-physical field coupling, introducing dynamic indicators such as vortex intensity and frontal position to dynamically adjust the attention weight.

[0223] (3) Edge-center collaborative computing architecture, where lightweight attention feature extraction is performed at edge nodes (FLOPs < 1G), and the central server completes parameter optimization.

[0224] Step 3: Multi-modal value fusion. Physical quantity coupling evaluation: Calculate the cross-correlation coefficient of flow velocity-temperature salinity (when the threshold > 0.6, increase the weight of temperature-salinity features by 10%); Dynamically adjust the weight of the vortex intensity index; The value function is the core index in reinforcement learning used to evaluate the expected cumulative reward of a state or state-action pair. Its essence is to quantify the achievement degree of the system goal through mathematical modeling. In ocean flow field reconstruction, the value function is used to dynamically evaluate the rationality of data assimilation results, and guide model parameter optimization by minimizing the deviation of the prediction error from physical constraints.

[0225] Step 4: Edge-center collaborative optimization. Edge node tasks: Perform lightweight attention feature extraction (FLOPs < 1G) and transmit the feature map through the QUIC protocol (end-to-end latency < 5ms); Central server tasks: Update the value network parameters after receiving the features (using the PPO algorithm, clip ratio = 0.3), and send the optimized weights to the edge nodes (quantization precision 8-bit fixed-point number).

[0226] Step 5: Multi-scale fusion and parameter update:

[0227] Parameter perturbation injection: Apply a perturbation δp to the FVCOM model parameters (such as the turbulence coefficient), and the perturbation range is controlled by an adaptive threshold: δp = σ p ⋅ tanh(α ⋅ ∣E∣).

[0228] Step 6: Edge-center collaborative optimization. Edge node tasks: Perform lightweight residual feature extraction and transmit the feature map through quantization compression (SNORM); Central server tasks: Complete the reinforcement learning parameter update after receiving the feature map and send the optimized weights to the edge nodes.

[0229] Core module 5: Prediction and visualization output layer.

[0230] Real-time three-dimensional flow field visualization (contour / vector map); Simulation of scenarios such as oil spill diffusion and pollution migration. Composition of the prediction and visualization output layer system: ① Digital twin engine: Integrate the FVCOM model and the deep learning prediction results to support dynamic parameter deduction; ② AR / VR visualization terminal: Equipped with a three-dimensional flow field rendering engine (OSGEarth) and an interactive SDK; ③ BIM-GIS collaborative platform: Support real-time spatial analysis of IFC format building models and ocean flow field data; ④ Edge computing node: Deploy a lightweight model (FLOPs < 5G) to achieve real-time update of flow field data.

[0231] The specific implementation process is as follows:

[0232] Step 1: Dynamic spatio-temporal alignment of multi-source data. Input data: Satellite altimeter data (spatial resolution 25km × 25km, temporal resolution 1 day), Argo float data (spatial resolution 1000m, temporal resolution 10 days), ADCP profile data (spatial resolution 500m, temporal resolution 1 hour), etc.

[0233] Dynamic weighted fusion algorithm:

[0234] Step 2: Deployment of Multimodal Prediction Model, Model Architecture: ① Physical Model Layer: FVCOM three-dimensional flow field model, coupled with dynamic parameters such as tides and wind stress; ② AI Enhancement Layer: Spatiotemporal sequence prediction network based on Transformer, with the spatiotemporal sequence of satellite altimeter and Argo data as input; ③ The Hybrid Input Layer is the core module for realizing the dynamic fusion of multi-source ocean flow field data. Its design goal is to break through the limitations of traditional single data sources through multimodal data alignment and dynamic weight allocation. The specific components include a multi-source data interface module, a dynamic routing network, and a feature fusion module. The multi-source data interface module supports the standardized access of multiple types of ocean observation data such as satellite altimeters (25km resolution), Argo floats (1km resolution), and ADCP profile data (500m resolution); the spatiotemporal alignment algorithm (adaptive weighting based on residual field intensity) is used to achieve the dynamic spatiotemporal matching of multi-source data. The dynamic routing network (Router) adopts a Sparse Mixture of Experts (Sparse MoE) architecture, including 4 expert modules (each expert is an independent feedforward network) and a Top-K routing strategy; the output probability distribution of the router determines the data flow direction. The feature fusion module adopts a three-dimensional spatiotemporal pyramid fusion strategy, with spatial dimensions: 1×1×1 (detail features), 2×2×2 (mesoscale eddies), 4×4×4 (large-scale flow field); in the time dimension, long-range dependencies such as tidal cycles are captured through cross-time-step attention.

[0235] Step 3: AR / VR Interactive Visualization, AR Scene Construction: Scan the coastline through the mobile phone camera to automatically load the BIM model of the corresponding area; overlay the three-dimensional flow field vectors in the AR interface (color represents flow velocity, and arrow density represents eddy intensity);

[0236] Users can view the flow velocity profile at a specific location through gesture operations (delay < 200ms); VR Scene Design: Wear a head-mounted display device to enter the virtual ocean laboratory and real-time control the virtual ship to observe the flow field structure; collect flow field node data through a data glove to automatically generate a correlation heat map of flow velocity-temperature / salinity.

[0237] Step 4: BIM-GIS Collaborative Application, Data Interface: Spatially match the pier pile foundation coordinates in the BIM model with the flow field prediction results; calculate the hydrodynamic parameters when the ship berths (such as impact force, mooring cable tension); Decision Support: Generate a conflict heat map of typhoon paths and offshore wind farm layouts; provide dynamic early warnings of the scour risk of cross-sea bridge piers (based on CFD-DEM coupled simulation).

[0238] Step 5: Edge Computing and Real-time Update, Lightweight Deployment: Deploy edge servers (NVIDIA Jetson AGX Orin) at coastal observation stations to perform preprocessing of flow field data; transmit the compressed flow field data (bit rate ≤ 0.3bpp) to the central server via the QUIC protocol; Dynamic update mechanism: When satellite or buoy data is updated, trigger local model retraining (using the federated learning framework); Synchronize global digital twin parameters every 10 minutes.

[0239] In summary, this embodiment includes the following technical solutions:

[0240] (1) Multi-source data dynamic spatio-temporal alignment weighted algorithm, proposing an adaptive weight allocation mechanism based on residual field intensity to break through the static spatio-temporal matching limitations of traditional KNN registration methods. Dynamically adjust the weights of satellite, Argo, and ADCP data through a reinforcement learning policy network, improving accuracy.

[0241] (2) 3D spatio-temporal pyramid attention enhancement mechanism, proposing a spatio-temporal feature focusing technology that combines 3D space decomposition (1×1×1, 2×2×2, 4×4×4) with self-attention to break through the capture limitations of traditional 2D attention for the global-local feature coupling of flow field errors, improving the error matching degree.

[0242] (3) Value evaluation model for multi-physical field coupling, introducing dynamic indicators such as vortex intensity and frontal position to dynamically adjust the weights of value functions. Achieve an increase in the weight of temperature-salinity characteristics when the cross-correlation coefficient of flow velocity-temperature-salinity > 0.6.

[0243] (4) Edge-center collaborative training architecture, designing lightweight edge nodes (FLOPs < 1G) to perform feature extraction, and the central server optimizes global parameters through the federated learning framework. Compared with traditional centralized computing, the communication bandwidth utilization rate is improved, and the video memory occupancy is reduced.

[0244] (5) Adaptive weight distribution penalty term, introducing KL divergence constraint (λ = 0.02) in the value network to prevent the degradation of data source weights and improve policy stability.

[0245] (6) AR / VR virtual-real collaborative analysis system, dynamically linking and displaying the flow field characteristics with engineering facilities (such as simulating the structural stress of a wharf driven by the typhoon path), breaking through the static visualization limitations of the VolumeLIC algorithm. The user operation response speed is improved, and the frame rate of flow field rendering in the VR scene is stable.

[0246] (7) BIM-GIS deep integration platform, constructing a BIM model containing 18 types of marine environmental parameters to achieve joint simulation of the siting of offshore wind farms and flow field optimization. Break through the limitations of traditional GIS that only supports spatial analysis and support real-time calculation of hydrodynamic parameters for ship berthing (such as impact force and mooring cable tension).

[0247] (8) Dynamic noise injection routing strategy, adding Gaussian noise (σ = 0.1) in the sparse Mixture of Experts (MoE) routing calculation to avoid over-concentration of the strategy. Compared with the noise-free method, the convergence speed is improved and the routing diversity is increased.

[0248] (9) Multimodal feature semantic alignment graph. Construct a semantic association graph of 18-dimensional features such as flow velocity, temperature, and salinity through a graph neural network to improve the accuracy of node similarity calculation.

[0249] (10) Edge inference under the federated learning framework, executing a lightweight model at the edge node, protecting data privacy through differential privacy, and simultaneously optimizing the parameters of the global digital twin.

[0250] (11) Multiphysical field coupling evaluation index, introducing a hydrodynamics-thermodynamics coupling index (such as the correlation between the depth of seawater stratification and the flow velocity gradient) to optimize the design of the heat exchange system of an offshore platform. Breaking through the traditional evaluation method that only focuses on a single physical quantity.

[0251] (12) Adaptive perturbation range dynamic adjustment, proposing a method to dynamically adjust the perturbation range of model parameters based on the residual strength.

[0252] (13) Multiple method fusion intelligent prediction, proposing an intelligent prediction method for ocean current fields that fuses the PI-LSTM machine learning model, the FVCOM three-dimensional hydrodynamic model, and the generalized unit hydrograph method, integrating the advantages of different methods to achieve complementary advantages.

[0253] (14) Sparse Mixture of Experts (MoE) routing with adaptive noise injection, based on the Transformer architecture, proposing to add Gaussian noise in the routing calculation to avoid over-concentration of the strategy. Compared with the noise-free method, the convergence speed is improved and the routing diversity is increased.

[0254] (15) Multimodal feature semantic alignment graph, constructing a semantic association graph of 18-dimensional features such as flow velocity, temperature, and salinity through a graph neural network, and the accuracy of node similarity calculation. Breaking through the feature fusion method that only relies on modal decomposition.

[0255] Refer to Figure 2 , an embodiment of the present application provides a system for intelligent reconstruction and prediction of ocean three-dimensional current fields, the system includes:

[0256] A data acquisition unit, configured to obtain multi-source observation data of the ocean through a large language model and edge monitoring devices;

[0257] A data fusion unit, configured to preprocess and fuse the multi-source observation data to obtain fused observation data;

[0258] A spectrum construction unit for constructing a spatio-temporal correlation spectrum with the target observation parameters in the fused observation data as nodes and the spatio-temporal correlation and physical quantity coupling relationships between the target observation parameters as edges;

[0259] A data reconstruction unit for reconstructing and determining the reconstructed data corresponding to the multi-source observation data according to the spatio-temporal correlation spectrum;

[0260] A prediction output unit for predicting the three-dimensional flow field of the ocean based on the reconstructed data and outputting the visualized three-dimensional flow field.

[0261] It can be understood that the content in the above method embodiments is applicable to the system embodiments of the present application. The functions specifically implemented by the system embodiments of the present application are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those in the above method embodiments.

[0262] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order mentioned in the operation diagrams. For example, depending on the functions / operations involved, two consecutive blocks shown may actually be executed substantially simultaneously, or the blocks can sometimes be executed in the reverse order. In addition, the embodiments presented and described in the flowcharts of the present application are provided by way of example for the purpose of providing a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logical flows presented herein. Alternative embodiments are contemplated, in which the order of various operations is changed and the sub-operations described as part of a larger operation are executed independently.

[0263] In addition, although the present application is described in the context of functional modules, it should be understood that, unless otherwise stated to the contrary, one or more of the functions and / or features may be integrated in a single physical system and / or software module, or one or more functions and / or features may be implemented in separate physical systems or software modules. It can also be understood that a detailed discussion of the actual implementation of each module is not necessary for understanding the present application. More precisely, considering the attributes, functions, and internal relationships of the various functional modules in the system disclosed herein, the actual implementation of the module will be understood within the ordinary skills of an engineer. Therefore, those skilled in the art can implement the present application as set forth in the claims without undue experimentation. It can also be understood that the specific concepts disclosed are illustrative only and are not intended to limit the scope of the present application, which is determined by the full scope of the appended claims and their equivalents.

[0264] When the above - mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer - readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read - only memories (ROMs), random - access memories (RAMs), magnetic disks, or optical discs, etc., which can store program codes of various kinds.

[0265] The logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer - readable medium for use by an instruction - execution system, system, or device (such as a computer - based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction - execution system, system, or device), or in combination with these instruction - execution systems, systems, or devices. For the purposes of this specification, a "computer - readable medium" can be any system that can contain, store, communicate, propagate, or transport a program for use by or in combination with an instruction - execution system, system, or device.

[0266] More specific examples (nonexhaustive list) of computer - readable media include the following: electrical connection parts with one or more wirings (electronic systems), portable computer disk cartridges (magnetic systems), random - access memories (RAMs), read - only memories (ROMs), erasable programmable read - only memories (EPROMs or flash memories), fiber - optic systems, and portable compact disc read - only memories (CDROMs). Additionally, a computer - readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, then editing, interpreting, or otherwise processing it as appropriate, and then storing it in a computer memory.

[0267] It should be understood that each part of the present application can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0268] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc. means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0269] Although the embodiments of the present application have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and purposes of the present application, and the scope of the present application is defined by the claims and their equivalents.

[0270] The above is a specific description of the preferred embodiments of the present application, but the present application is not limited to the described embodiments. Those skilled in the art can also make various equivalent deformations or substitutions without departing from the spirit of the present application, and these equivalent deformations or substitutions are all included within the scope defined by the claims of the present application.

Claims

1. A method for intelligent reconstruction and prediction of three-dimensional ocean current fields, characterized in that, The method includes the following steps: Obtain multi-source observation data of the ocean through a large language model and edge monitoring devices; Preprocess and fuse the multi-source observation data to obtain fused observation data; Construct a spatio-temporal correlation map with the target observation parameters in the fused observation data as nodes and the spatio-temporal correlation and physical quantity coupling relationships between the target observation parameters as edges; Reconstruct and determine the reconstructed data corresponding to the multi-source observation data according to the spatio-temporal correlation map; Predict the three-dimensional flow field of the ocean based on the reconstructed data and output the visualized three-dimensional flow field.

2. The method for intelligent reconstruction and prediction of a three-dimensional ocean flow field according to claim 1, wherein, The step of obtaining multi-source observation data of the ocean through a large language model and edge monitoring devices includes the following steps: Use the large language model to parse user instructions, and then obtain the profile observation data corresponding to the user instructions; project the profile observation data into the spatio-temporal grid of the satellite altimeter, and then calculate the spatio-temporal matching degree of the profile observation data; delete the profile observation data with the spatio-temporal matching degree lower than the set threshold to obtain the first observation data; Use the ADCP device in the edge monitoring device to obtain shipborne ADCP data; use Kalman filtering to remove the noise of the shipborne ADCP data, and then perform flow velocity profile reconstruction on the shipborne ADCP data through the edge GPU to obtain the second observation data; Wherein, the first observation data and the second observation data are used as the multi-source observation data.

3. The method for intelligent reconstruction and prediction of an ocean three-dimensional flow field according to claim 1, characterized in that, The step of preprocessing and fusing the multi-source observation data to obtain fused observation data includes the following steps: Extract barotropic mode features and first baroclinic mode features from the multi-source observation data; Set the three-dimensional grid and boundary conditions of the FVCOM model; Project the barotropic mode features and the first baroclinic mode features into the three-dimensional grid of the FVCOM numerical model for spatio-temporal alignment; Adjust the quality threshold according to the task type, and perform data cleaning on the barotropic mode features and the first baroclinic mode features after spatio-temporal alignment according to the quality threshold to obtain quality evaluation features; Use the edge monitoring device to extract vertical mode features from the obtained multi-source observation data; Extract Ekman flow features coupled with the barotropic mode features from the multi-source observation data obtained by the satellite in the edge monitoring device; Fuse the quality evaluation features, the vertical mode features and the Ekman flow features to obtain the fused observation data.

4. The method for intelligent reconstruction and prediction of a three-dimensional ocean flow field according to claim 1, characterized in that The method further includes the following steps: Adjust the feature weights of the target observation parameters according to the task requirements to adjust the corresponding edges in the spatio-temporal correlation map.

5. A method for intelligent reconstruction and prediction of a three-dimensional ocean flow field according to claim 1, characterized in that, The step of reconstructing and determining the reconstructed data corresponding to the multi-source observation data according to the spatio-temporal correlation map includes the following steps: Output a gridded hydrodynamic field through the FVCOM model based on the spatio-temporal correlation map; Input the hydrodynamic field in the form of time series data into the LSTM network, embed physical constraint terms into the LSTM network, and dynamically adjust the sampling frequency through an event trigger mechanism; Calculate the runoff contribution based on the SCS-CN model, simulate the vertical velocity distribution through the generalized unit hydrograph, introduce a multi-scale attention mechanism, and dynamically allocate computing resources to the mesoscale eddy region; Optimize the parameters of the FVCOM model.

6. The method for intelligent reconstruction and prediction of a three-dimensional ocean flow field according to claim 5, wherein The step of optimizing the parameters of the FVCOM model includes the following steps: Extract the features of the three-dimensional flow field output by the FVCOM model and the fusion observation data respectively and calculate the residuals; Construct a state space based on the residuals and enable the FVCOM model to perform reinforcement learning; Dynamically adjust the weights of the FVCOM model through a policy network and a value network; wherein, the policy network is used to input the residual feature map and output the action probability distribution; the value network is used to input the state vector and output the expected reward; Perform coordinated optimization on the FVCOM model based on the residual feature map transmitted by the edge monitoring device.

7. A method for intelligent reconstruction and prediction of a three-dimensional ocean flow field according to any one of claims 1 to 6, characterized in that, The step of outputting the visualized three-dimensional flow field includes the following steps: Output the contour line or vector map of the three-dimensional flow field through a digital twin engine, an AR / VR visualization terminal, a BIM-GIS collaborative platform or the edge monitoring device.

8. An intelligent reconstruction and prediction system for a three-dimensional ocean current field, characterized in that, The system includes: A data acquisition unit for acquiring multi-source observation data of the ocean through a large language model and an edge monitoring device; A data fusion unit for preprocessing and fusing the multi-source observation data to obtain fusion observation data; A spectrum graph construction unit for constructing a spatio-temporal correlation graph with the target observation parameters in the fusion observation data as nodes and the spatio-temporal correlation and physical quantity coupling relationship between the target observation parameters as edges; A data reconstruction unit for reconstructing and determining the reconstructed data corresponding to the multi-source observation data according to the spatio-temporal correlation graph; A prediction output unit for predicting the three-dimensional flow field of the ocean based on the reconstructed data and outputting the visualized three-dimensional flow field.

9. An electronic device, characterized in that, The electronic device includes a processor and a memory; The memory is used to store programs; The processor executes the program to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a program, and the program is executed by the processor to implement the method according to any one of claims 1 to 7.

Images