Ocean digital twin engine system

Abstract

The invention relates to the technical field of ocean digital twin engine systems, and provides an ocean digital twin engine system to solve key bottlenecks in ocean data visualization, dynamic simulation and engineering application in the prior art. Comprising a data layer, a model layer, a scene layer, a service layer and an application layer. The data layer is used for uniformly accessing multi-source ocean data and carrying out space-time alignment and quality control; the model layer is used for scheduling an ocean numerical mode and supporting multi-physics field joint simulation; the scene layer is used for realizing true three-dimensional light tracking rendering of the ocean phenomenon; the service layer is used for issuing lightweight scene service and supporting 100,000-level concurrent access; and the application layer is used for enabling application development.

Description

Technical Field

[0001] This invention relates to the field of marine digital twin engine system technology, specifically to a marine digital twin engine system. Background Technology

[0002] Marine data visualization aims to transform complex, multidimensional, and dynamic marine environmental data into intuitive and interactive graphical representations. This technology integrates scientific computing, geographic information systems, and computer graphics methods, relying on specialized algorithms and high-performance rendering to achieve multi-scale visual analysis of marine phenomena, from macroscopic trends to microscopic mechanisms. It is a key technology supporting marine scientific research, environmental monitoring and forecasting, and intelligent marine decision-making. Currently, existing technologies mainly include the following categories:

[0003] 1. Techniques primarily focused on scientific data visualization are typically implemented using programming languages ​​such as MATLAB, Python, or Fortran. They rely on numerical computation tools (such as Simulink for MATLAB, Matplotlib / Plotly for Python, and PGPLOT for Fortran) to visualize multidimensional data (such as ocean temperature fields and current fields).

[0004] However, scientific computing visualization technologies (such as CFD simulation and remote sensing data rendering) are mainly geared towards offline batch processing scenarios, and have the following three core drawbacks:

[0005] 1) Insufficient real-time capability: Lacks online interactive parsing capability for large-scale scientific data (such as ROMS / FVCOM model output), resulting in significant spatiotemporal delay;

[0006] 2) Dimensional fusion defects: Insufficient ability to couple and express multi-dimensional (physical, chemical, biological parameters) and multi-scale (watershed, grid, particle) dynamic data;

[0007] 3) High professional barriers: The visualization results rely on manual interpretation by domain experts, which has a significant gap with the engineering decision-making system and makes it difficult to support rapid response business applications.

[0008] 2. Geographic Information System (GIS) technologies, such as ArcGIS, SuperMap, and QGIS (open-source desktop GIS software), are suitable for visualizing geospatial data (such as typhoon paths and ship trajectories), presenting data visualization effects on maps, and primarily serving engineering applications.

[0009] The application of geographic information systems in marine environments has also revealed the following key issues:

[0010] 1) Inadequate 3D representation: Lack of native support for scenarios such as vertical profile temperature and salinity fields, underwater topography, and integrated water body modeling;

[0011] 2) Distortion of dynamic processes: The phenomenon representation algorithm does not embed ocean dynamic constraints (such as using still water approximation to handle storm surge flooding and ignoring the Coriolis force and wave climb effect);

[0012] 3) Low model integration: Insufficient standardization of data interfaces with professional ocean numerical models leads to a break in the simulation and visualization links, making it difficult to unify spatiotemporal benchmarks.

[0013] At the same time, existing visualization systems also have the following common technical shortcomings:

[0014] 1) Lack of data-driven approach: Water dynamics (such as eddy migration trajectories and pollutant diffusion fronts) lack physical modeling based on first principles;

[0015] 2) Spatial-temporal dimensional fragmentation: The rendering efficiency of four-dimensional data fields (longitude, latitude, depth, and time) is low, making it difficult to achieve dynamic simulation with second-level updates;

[0016] 3) Insufficient optical realism: The optical characteristics such as the two-way reflection distribution function (BRDF) of the sea surface and water volume scattering are not considered, resulting in poor comparability between virtual ocean scenes and remote sensing observation data. Summary of the Invention

[0017] One of the objectives of this invention is to propose a marine digital twin engine system to address the key bottlenecks in the existing technologies for marine data visualization, dynamic simulation, and engineering applications.

[0018] The technical solution of the present invention is as follows:

[0019] A marine digital twin engine system includes: a data layer, a model layer, a scene layer, a service layer, and an application layer;

[0020] The data layer is used to uniformly access multi-source ocean data and perform spatiotemporal alignment and quality control.

[0021] The model layer is used to schedule ocean numerical models and supports multiphysics joint simulation.

[0022] The scene layer is used to achieve true 3D ray tracing rendering of marine phenomena;

[0023] The service layer is used to publish lightweight scenario services, supporting 100,000 concurrent accesses.

[0024] The application layer is used to enable application development.

[0025] Furthermore, the data layer operates as follows:

[0026] S110: Input raw data;

[0027] S120: Classify data according to its source and information;

[0028] S130: Preprocessing based on data type:

[0029] Satellite remote sensing data is converted into spectral reflectance or spectral radiance using calibration coefficients.

[0030] For numerical pattern data, read the NetCDF file structure, parse the information, and extract the multidimensional array;

[0031] Perform CRC check on the buoy data: calculate the cyclic redundancy check code and compare it with the check code attached to the data transmission packet;

[0032] S140: Unify all the above data onto the same spatiotemporal reference.

[0033] S150: Perform anomaly detection on the data;

[0034] S160: For normal data that passes the detection, perform feature extraction and calculate key feature indicators;

[0035] For abnormal data, GAN repair is performed first, and then feature extraction is performed on the repaired data;

[0036] S170: Perform EOF compression;

[0037] S180: Organize multidimensional data into data cubes;

[0038] S190: Output the data cube in different formats according to the needs of downstream applications.

[0039] Furthermore, the operating principle of the model layer is as follows:

[0040] S210: Input the data required for the model to run;

[0041] S220: Automatically generate or start standardized container instances based on model configuration and requirements;

[0042] S230: Based on the characteristics of the model algorithm, automatically determine and select the optimal computing resource deployment strategy;

[0043] For CPU-intensive tasks, choose MPI cluster deployment;

[0044] For GPU-accelerated tasks, choose CUDA container deployment;

[0045] S240: Perform the core numerical computations of the model on the allocated computing resources;

[0046] S250: Determine whether bidirectional coupling with other modes is required based on the simulation task configuration;

[0047] If necessary, proceed to the next step and perform the coupling loop process;

[0048] If not, proceed to the final result output process;

[0049] S260: Coordinated via WSMF coupler;

[0050] S270: Perform data mapping;

[0051] S280: Perform time synchronization;

[0052] S290: After performing coupled data exchange, return to S240 until the simulation is complete.

[0053] Furthermore, the operating principle of the scene layer is as follows:

[0054] S310: Obtain the raw data to be rendered;

[0055] S320: Analyze the raw data, determine the data type, and classify the data type into hydrological element field data and ocean dynamic phenomenon data;

[0056] S330: Performs rendering processing based on different data;

[0057] S340: Combines the parallel rendering results into a scene image;

[0058] S350: Based on the distance of objects in the scene, it performs a level of detail assessment and applies corresponding rendering optimization strategies;

[0059] S360: Outputs the rendered image data in the frame buffer according to the settings.

[0060] Furthermore, the rendering process in step S330 includes:

[0061] If the data is from a hydrological element field, then proceed as follows:

[0062] S331: Reconstruct discrete scalar field data into continuous three-dimensional volume data;

[0063] S332: Based on a physical model, calculate the optical properties of light as it propagates within volume data;

[0064] S333: Perform light step rendering;

[0065] If the data pertains to ocean dynamic phenomena, then proceed as follows:

[0066] S334: Generate a set of particles representing individuals based on the original data;

[0067] S335: Updates the state of all particles in each frame according to the laws of physics;

[0068] S336: Perform particle resampling.

[0069] Furthermore, step S350 includes:

[0070] S351: Divide the scene into foreground and background;

[0071] S352: For close-up shots, first perform high-precision rendering: use more detailed models, higher texture resolution, and smaller ray step length for calculation, and then perform subsurface scattering: simulate the physical effect of light scattering inside a translucent material;

[0072] For distant views, first perform simplified rendering, then replace the Impostor.

[0073] Furthermore, the service layer operates as follows:

[0074] S410: Receives client requests;

[0075] S420: Parse the request content and determine whether the request type is a real-time interactive request or a batch calculation request;

[0076] For real-time interactive requests, differential compression is used for transmission, and the request is scheduled to the edge computing node closest to the user's geographical location.

[0077] If it is a batch computing request, the request will be scheduled to a central cloud with powerful computing and storage capabilities, mobilizing a large amount of resources to process the full amount of data;

[0078] S430: Request to access the internal microservice network through the service network;

[0079] S440: Conduct a safety inspection. If the inspection is passed, proceed to the next step.

[0080] If the request fails the inspection, the request processing flow is terminated and recorded in the audit log;

[0081] S450: Dynamically assemble (or orchestrate) the required microservice chain based on the request type;

[0082] S460: Allocate appropriate computing resources in the cluster based on the type and size of the request;

[0083] If it is a GPU-first real-time rendering service, allocate a GPU instance to the request, start the real-time rendering service, and enable dynamic LOD control;

[0084] If it is a CPU-based elastic analytics service, then allocate a CPU cluster instance, start the analytics service, and cache the calculation results;

[0085] S470: Returns the processing results to the client via streaming;

[0086] S480: The client receives the data stream and performs the final decoding and presentation.

[0087] Furthermore, the operating principle of the application layer is as follows:

[0088] S510: Developer inputs requirements;

[0089] S520: Determine the development mode:

[0090] If it is determined to be a web application, load the corresponding plugin and initialize the web rendering environment;

[0091] If it is determined to be a native application, then activate the bridging plugin with Unity or Unreal Engine;

[0092] S530: Transforms the developer's requirements into specific data service call parameters and selects the required data source for configuration;

[0093] S540: Based on the developer's input intent, match the application template from the template library, inject the configured data source and user parameters, and generate the basic scene;

[0094] S550: Determines whether the application requires dynamic, energy-intensive calculations or simulations in the backend;

[0095] If needed, for lightweight computing, the WebAssembly module is scheduled to perform calculations directly on the browser side; for heavy simulations, tasks are submitted to the service layer via API, and the service layer schedules high-performance computing clusters.

[0096] If not, then enter pure rendering mode;

[0097] S560: Associate the dynamic result data generated by S550 with the basic scene generated in S540;

[0098] S570: Detects the hardware capabilities of the running terminal;

[0099] S580: Based on device capability adaptation, adjust the on / off state and level of visual effects according to user preferences or scenario requirements;

[0100] S590: Encapsulates the configuration, data, logic, and rendering results generated in the above steps and outputs a visualization component.

[0101] Furthermore, the output format in step S190 includes:

[0102] PARQUET Data Lake: Outputs data in columnar storage format, suitable for efficient and low-cost batch processing and ad-hoc queries in big data analytics frameworks (such as Spark and Hive);

[0103] HDF5 numerical model: Outputs in a hierarchical format that is standard in the field of scientific computing, retaining complete metadata and dimensional information for subsequent numerical simulation or in-depth analysis;

[0104] GEOJSON application platform: Outputs point data or aggregated polygon data in geospatial vector formats such as GEOJSON for visualization and interaction in WebGIS (Geographic Information System based on Internet technology) application platforms.

[0105] Furthermore, the final result output process in step S250 includes:

[0106] S251: Output the final result;

[0107] Output formats include VTK or NetCDF.

[0108] S252: Generate visual analysis results.

[0109] The beneficial effects of this invention are as follows:

[0110] 1. Construct a real-time interactive marine visualization system to support online dynamic analysis and interactive analysis of scientific data, achieving a leap from "offline post-processing" to "online twin".

[0111] 2. Achieve high-fidelity dynamic marine scene modeling, integrate marine dynamic constraints, and enhance the physical realism of phenomena such as storm surges, ocean currents, and temperature and salinity fields, avoiding the static distortion problem of traditional GIS.

[0112] 3. Establish a complete chain of "numerical model, visualization, and decision support" and achieve deep integration with mainstream ocean models (such as FVCOM and ADCIRC) through standardized interfaces to form a closed loop of "simulation, rendering, and application".

[0113] 4. Enhance the immersive simulation capabilities of marine scenes. Based on realistic optical models and efficient GPU rendering, achieve high-resolution, high-frame-rate four-dimensional marine evolution visualization to support operational applications such as marine environmental monitoring and disaster prevention and mitigation.

[0114] Compared with existing GIS technologies, this invention also has the following advantages:

[0115] 1. Deep specialization of marine data processing: In terms of data compatibility, it supports access to marine data in multiple formats; in terms of spatiotemporal data processing, it has established a vertically layered + time-series four-dimensional data cube, while GIS technology only supports two-dimensional / 2.5D spatial data.

[0116] 2. Physical Realism of Ocean Phenomenon Modeling: The use of a hydrodynamic model driven by nonlinear shallow water equations to replace the static flooding analysis method of "flooding from flat ground" greatly improves the accuracy of flooding analysis.

[0117] 3. Dynamic interactive capabilities of visualization: GPU-accelerated volume rendering + real-time ray tracing replaces CPU symbolic rendering, resulting in a significant improvement in frame rate; supports four-dimensional data field dynamic inference (GIS) technology (static snapshots or multi-temporal overlay); optical effects use sea surface BRDF + water body subsurface scattering method instead of simple color filling.

[0118] 4. Deep integration of ocean numerical models: Containerized integration of various ocean numerical models, supporting real-time linkage of parameter adjustment, simulation, and rendering.

[0119] Therefore, the core value of this invention lies in the deep integration of scientific computing visualization, GIS spatial analysis, ocean dynamics simulation and real-time rendering technologies to build a full-stack visualization engine for ocean digital twins, and promote the evolution of ocean scientific research and engineering applications from "static analysis" to "dynamic twins". Attached Figure Description

[0120] Figure 1 This is a diagram showing the relative positions, connections, functions, and logical interactions between the modules in Example 1.

[0121] Figure 2 This is a diagram illustrating the operational principle of the data layer.

[0122] Figure 3 This is a diagram illustrating the operational principle of the model layer.

[0123] Figure 4 This is a diagram illustrating the operational principle of the scene layer.

[0124] Figure 5 This is a diagram illustrating the operational principles of the service layer.

[0125] Figure 6 This is a diagram illustrating the operational principles of the application layer.

[0126] Figure 7 This is the interface for ocean data processing tools in the data layer;

[0127] Figure 8 The interface for ocean numerical model and algorithm management tools at the model layer;

[0128] Figure 9 This is the interface for a twin scene editor for the scene layer;

[0129] Figure 10 This is the interface of the cloud service publishing platform for the service layer;

[0130] Figure 11 A toolkit interface to support full-stack development at the application layer.

[0131] In the picture:

[0132] Apache Flink is a distributed stream processing framework;

[0133] QoS assurance is a technical mechanism used to guarantee the quality, stability, and priority of data transmission.

[0134] Ray Marching is a ray-stepping rendering technique.

[0135] The RTX 4090 is a graphics card from NVIDIA based on the Ada Lovelace architecture;

[0136] React / Vue is a front-end JavaScript framework;

[0137] ARCore is an augmented reality development platform launched by Google.

[0138] OpenXR is a cross-platform standard;

[0139] Dierctx12 is a graphical application programming interface (API) developed by Microsoft.

[0140] Three.js is a lightweight, cross-browser JavaScript 3D library;

[0141] HPC stands for High Performance Computing;

[0142] WebGL2 is the standard for web graphics. Detailed Implementation

[0143] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0144] Example 1

[0145] like Figure 1 , 7As shown in Figure 11, a marine digital twin engine system combines marine observation, artificial intelligence, advanced simulation, and supercomputing to achieve a realistic, high-resolution, multi-dimensional, and interactive real-time virtual representation of the ocean. By constructing a virtual ocean system, historical and real-time observation data, along with human understanding, are fully integrated into this system, forming a capability for real-time simulation, analysis, and extrapolation.

[0146] The main differences between marine digital twins and existing "traditional simulations" and "digital oceans" are as follows:

[0147] 1. Simulation technology is typically an abstract imitation of real-world systems, used for scientific experiments and decision support, and cannot be correlated with actual physical objects in real time. In the marine digital twin, simulation technology is merely a creation and operation technique.

[0148] 2. Unlike "Digital Ocean," which focuses on data integration and static analysis services, marine digital twins utilize advanced sensing technologies, artificial intelligence, and big data analytics to achieve real-time dynamic updates and interaction. They focus on refined management and real-time decision support, exhibiting high interactivity. Through methods such as field measurements, simulations, and data analysis, marine digital twins can perceive, diagnose, and predict the state of physical entities in real time, and adjust the behavior of these entities by optimizing instructions.

[0149] Specifically, this embodiment adopts a five-layer collaborative architecture consisting of "Data Layer (DTO Data), Model Layer (DTO Model), Scenario Layer (DTO Explorer), Service Layer (DTO Cloud), and Application Layer (DTO SDK)," which comprises five core components covering data governance, model management, scenario construction, cloud services, and development support. This forms a complete digital twin technology stack, namely the data layer, model layer, scenario layer, service layer, and application layer, enabling closed-loop processing of the entire marine digital twin process.

[0150] The data layer serves as a standardized marine data governance hub, enabling unified access to multi-source marine data, including satellite remote sensing data, marine buoy networks (via real-time TCP transmission control protocol streams), marine survey data, basic geographic data, digital model products, and unmanned equipment. It performs spatiotemporal alignment and quality control, and standardizes access.

[0151] Core capabilities: Supports automated cleaning, fusion, and standardized output of multimodal marine elements (wind field, current field, tide level, temperature and salinity, etc.); compatible with heterogeneous geographic data (vector / raster, BIM (Building Information Modeling), laser point cloud, oblique photography, etc.) to achieve integrated land and sea data infrastructure construction; provides spatiotemporal index optimization and data lake docking interface to support efficient data retrieval in high-concurrency scenarios.

[0152] Model layer: Positions the numerical simulation and visualization algorithm integration engine, used to schedule ocean numerical models, support multiphysics joint simulation, and perform coupled calculations.

[0153] It features built-in containerized encapsulation and parameterized configuration management of mainstream ocean models such as ROMS, ADCIRC, and SWAN, as well as management of large AI models such as Pangu and Xihe; it integrates professional algorithms such as particle systems and volume rendering to support high-fidelity simulation of dynamic processes such as storm surge flooding, pollutant diffusion, and vortex evolution; and it supports model coupling and embedding of third-party algorithms.

[0154] Scene layer: A location data-driven 3D dynamic scene building tool used to achieve true 3D ray tracing rendering of marine phenomena and perform dynamic rendering.

[0155] Visual representation driven by physical mechanisms such as typhoon path prediction and storm surge flooding dynamics; water environment simulation: dynamic rendering of three-dimensional ocean water based on coupled temperature, salinity, and flow field data; supports land-sea three-dimensional fusion, and seamless integration of oblique photogrammetry real-scene modeling and elevation + image digital elevation model (DEM) to achieve high-precision construction of coastal zone scenes.

[0156] Service Layer: Positioned as a cloud-based service tool for twin scenarios, used to publish lightweight scenario services, supporting 100,000 concurrent accesses, and providing cloud-based services.

[0157] It offers one-click cloud publishing capabilities, converting 3D scenes into lightweight formats such as WebGL / 3D Tiles; it supports distributed rendering and multi-level LOD (Level of Detail) optimization to ensure low-latency access to large-scale scenes; and it integrates OGC standard services (map services, feature services) to achieve interconnection with GIS platforms.

[0158] Application Layer: Positioning a rapid development framework for twin applications, used to enable rapid development of multi-platform applications such as Web, mobile, AR, or VR.

[0159] It provides dual-mode development interfaces for C / S (server-client) and B / S (browser / server) modes, and supports quick invocation for secondary development.

[0160] The data flow relationships of the above architectures are as follows:

[0161] 1. Vertical data flow: External data source → Data layer → Model layer → Scene layer → Service layer → Application layer → Terminal application.

[0162] Horizontal interaction: The data layer interacts bidirectionally with the data lake to realize the storage of raw data and the output of standardized products; the model layer interacts with high-performance computing clusters and GPU computing clusters to complete distributed numerical computation; the scene layer calls rendering cluster resources to achieve real-time visualization; and the service layer distributes scene services through the CDN network (Content Delivery Network).

[0163] 2. Functional coupling relationship

[0164] The data layer provides underlying support: it provides standardized input data to the model layer and background geographic data to the scene layer.

[0165] The model layer serves as the core computing engine, outputting dynamic simulation results to the scene layer.

[0166] The scenario layer acts as a central converter: it integrates the basic scenario data from the data layer and the dynamic data from the model layer; and it provides publishable scenario packages for the service layer.

[0167] The service layer acts as a service hub: it provides API (Application Programming Interface) service endpoints to the application layer and receives scene outputs from the scene layer.

[0168] The application layer acts as a bridge for applications: it encapsulates the service interfaces of the service layer and provides development components to the final application.

[0169] 3. Logical Interaction Relationships

[0170] Users initiate requests through the application layer → the service layer schedules resources → the scenario layer generates a scenario → the model layer performs calculations → the data layer prepares data.

[0171] This embodiment innovates in multiple aspects through system architecture design:

[0172] Intelligent fusion and dynamic visualization technology for multi-source marine data: It is the first to create a multimodal data spatiotemporal alignment engine for marine digital twins, which supports automatic coordinate transformation and time synchronization of heterogeneous data such as buoy observation, satellite remote sensing, and numerical model output; it breaks through the limitations of traditional static symbolization in GIS and proposes physical modeling methods for marine phenomena, including particle systems based on the Navier-Stokes equations (SPH method) and nonlinear hydrodynamic collision models, which improve the simulation accuracy of typical marine phenomena.

[0173] A containerized coupled computing engine for ocean numerical models. An ocean model scheduling platform based on Docker (an open-source platform) and Kubernetes (used to manage containerized applications across multiple hosts in a cloud platform) is built to enable plug-and-play deployment of ocean numerical models, supporting parameterized configuration templates and elastic resource allocation. A hybrid programming model of MPI (a cross-language communication protocol) + CUDA (a parallel computing platform and programming model) is adopted to achieve parallel computing and GPU acceleration for tens of millions of grid cells.

[0174] A lightweight service architecture for cloud-based collaboration. It adopts a marine data differential compression transmission protocol to achieve lightweight data processing and fast loading, and supports dual-channel access for OGC standard services (Geographic Information Interoperability Technical Specification) (WMS map service / WFS feature service) and custom APIs (REST expressive state transfer / gRPC open-source high-performance remote procedure call framework).

[0175] Example 2

[0176] like Figure 2 , 6 As shown in Example 1, this example demonstrates the operating principles of each architecture in the marine digital twin engine system.

[0177] The data layer operates as follows:

[0178] S110: Input raw data;

[0179] For example, satellite remote sensing data includes: L0 level data (raw data), L1 level data (radiometrically corrected data), and L2 level data (geometrically corrected data).

[0180] Numerical model data: NetCDF (Network Common Data Format) and other format files;

[0181] Field observation data: raw data streams or files transmitted from buoys, stations, etc.;

[0182] S120: Classify data according to its source and information;

[0183] S130: Preprocessing based on data type:

[0184] Satellite remote sensing data is converted into spectral reflectance or spectral radiance using calibration coefficients.

[0185] For numerical model data, the NetCDF file structure is read, and information such as dimensions, variables, and attributes is parsed to extract scientific data values ​​into multidimensional arrays (such as temperature fields and pressure fields) that can be used by the program.

[0186] Perform CRC (Cyclic Redundancy Check) check on the buoy data: Calculate the cyclic redundancy check code and compare it with the check code attached to the data transmission packet;

[0187] S140: Unifying all the above data onto the same spatiotemporal benchmark, so that multi-source data are comparable and fusionable in spatiotemporal terms, is a prerequisite for joint analysis.

[0188] S150: Perform anomaly detection on the data;

[0189] S160: For normal data that passes the detection, perform feature extraction and calculate key feature indicators;

[0190] For abnormal data, first perform GAN repair (Generative Adversarial Network repair): use a generative adversarial network to generate reasonable content based on the image context to fill in the missing or damaged areas, and then extract features from the repaired data.

[0191] Maximize the value of data and ensure the quality and continuity of data used in downstream analysis.

[0192] S170: Perform EOF compression, that is, perform empirical orthogonal function analysis. Through powerful data compression and dimensionality reduction methods, use a few main modes and their time variations to characterize the main information of the original complex data field, greatly reducing the amount of data.

[0193] S180: Organize processed multidimensional data with the same dimensions into data cubes to facilitate slicing, dicing, and aggregation analysis of any dimension;

[0194] S190: A standardized data product that outputs data cubes in different formats to meet the needs of different application scenarios, based on the requirements of downstream applications.

[0195] Output formats include:

[0196] PARQUET Data Lake: A data type that outputs data in a columnar storage format, suitable for efficient and low-cost batch processing and ad-hoc queries in big data analytics frameworks;

[0197] HDF5 numerical model: A file format for storing and organizing large-scale data, outputting in a hierarchical format that is standard in the field of scientific computing, preserving complete metadata and dimensional information for subsequent numerical simulation or in-depth analysis.

[0198] GEOJSON Application Platform: A format for encoding various geographic data structures, outputting point data or aggregated polygon data in geospatial vector formats such as GEOJSON for visualization and interaction in WebGIS application platforms.

[0199] In summary, the data layer uses an ETL (Extract-Transform-Load, the core data processing flow in a data warehouse) engine to automatically clean, unify coordinates, and align geospatial data (DEM, BIM, etc.) including marine observation data (buoys, satellite remote sensing, etc.), numerical model outputs (NetCDF, HDF5 format, etc.), and geospatial data (DEM, BIM, etc.). It employs GeoHash (a geocoding system that converts latitude and longitude coordinates into sortable strings) and timestamps to construct a multidimensional data cube.

[0200] In terms of spatial matching, an improved GeoHash-T algorithm (an algorithm for converting latitude and longitude coordinates on the Earth's surface into one-dimensional strings) is used to unify ocean data from different coordinate systems (such as the geocentric coordinate system used by the WGS84 global satellite positioning system and the UTM plane rectangular coordinate system) into a standard grid (resolution adjustable). This supports seamless fusion of vertically layered data (such as temperature and salinity profiles) under curved Earth coordinate systems with planar remote sensing data. Regarding spatiotemporal downscaling, EOF analysis (empirical orthogonal function) is used to downscale low-resolution model data (such as 1°×1°) to high-resolution data (such as 100m×100m) while preserving key ocean features.

[0201] Structured data (such as temperature and salinity profiles) is stored using standardized JSON / Parquet; unstructured data is optimized through block compression and pyramid layering, improving data governance efficiency. Standardized processing speed for multi-source data is increased by 5 times, with actual preprocessing time for 1TB of data less than 2 hours; spatiotemporal query latency is less than 50ms, a 90% reduction compared to traditional GIS.

[0202] The operating principle of the model layer in this embodiment is as follows:

[0203] S210: Input the data required for the model to run, such as the data cube output by the data layer, NetCDF files, etc.

[0204] S220: Automatically generate or start standardized container instances based on model configuration and requirements to achieve environment isolation, dependency management, version control, and deployment reproducibility;

[0205] S230: Based on the characteristics of the model algorithm, automatically determine and select the optimal computing resource deployment strategy;

[0206] For CPU-intensive tasks, choose MPI cluster deployment: perform distributed parallel computing on multiple CPU nodes through message passing interface;

[0207] For GPU-accelerated tasks, choose CUDA container deployment: Deploy the computing task to a container configured with a CUDA environment to perform large-scale parallel computing using thousands of GPU cores.

[0208] S240: Perform the core numerical computations of the model on the allocated computing resources;

[0209] S250: Based on the simulation task configuration, determine whether bidirectional coupling with other modes is required, such as ADCIRC+SWAN or other modes;

[0210] If necessary, proceed to the next step and perform the coupling loop process;

[0211] If not, the process will proceed to the final result output, which includes:

[0212] S251: Output the final result;

[0213] Output formats include VTK format (General Visualization Data Format) or NetCDF format;

[0214] S252: Generate visual analysis results for evaluation and analysis;

[0215] S260: Coordinates communication between different models by using a WSMF coupler (wavelength division multiplexing single-mode fiber coupler), that is, by using a Web service modeling framework as middleware.

[0216] S270: Perform data mapping, which means mapping the output data of different models (which may have different grid resolutions, grid types and data structures) to the input format and grid acceptable to the other model through operations such as spatial interpolation and regridding.

[0217] S280: Performs time synchronization, manages the time steps between coupled models, and ensures that they exchange data at the correct points in time;

[0218] S290: After performing coupled data exchange, return to S240 until the simulation is complete.

[0219] In summary, the model layer encapsulates traditional ocean numerical models such as ROMS (Regional Ocean Numerical Model), SWAN (Wave Model), and ADCIRC (Flooding Model) as well as AI numerical models using Docker containers, and leverages Kubernetes cluster scheduling for parallel computing. The particle tracking algorithm employs the SPH (Smoothed Particle Hydrodynamics) method, and the volume rendering algorithm is based on GPU-accelerated Ray Casting technology. For templated parameter generation, YAML (a format used to express data serialization) is used to define pattern configurations (such as grid resolution, time step, and physical parameterization scheme), supporting preset configurations for common ocean models.

[0220] The model outputs gridded data products by driving the allocation and use of computing resources through operator pipelines. These data products can be connected to a data lake or stored independently.

[0221] It achieves breakthroughs in simulation and rendering performance. It supports real-time execution of high-resolution mesh ocean numerical models; GPU-accelerated volume rendering enables smooth 60FPS rendering of hundreds of millions of particles.

[0222] The scene layer operates as follows in this embodiment:

[0223] S310: Obtain the raw data to be rendered;

[0224] S320: Analyze the raw data, determine the data type, and divide the data type into hydrological element field data (temperature field or salinity field, etc.) and ocean dynamic phenomenon data (typhoon or ocean current, etc.) to ensure that the metadata is rich enough and standardized.

[0225] S330: Performs rendering processing based on different data;

[0226] Specifically, if the data is from a hydrological element field, then the following steps are taken:

[0227] S331: Reconstruct discrete scalar field data into continuous three-dimensional volume data;

[0228] S332: Based on a physical model, calculate the optical properties of light, such as absorption, scattering, and emission, as light propagates within volume data.

[0229] S333: Performs ray step rendering, for example: emitting a ray from each pixel on the screen, passing through volume data, accumulating color and transparency at step sampling points, and finally synthesizing the color value of the pixel, achieving a balance between rendering quality and performance;

[0230] If the data pertains to ocean dynamic phenomena, then proceed as follows:

[0231] S334: Generate a set of particles representing individuals (such as water droplets and airflow particles) based on the original data (such as wind vector points and typhoon path points);

[0232] S335: Updates the state of all particles in each frame according to physical laws (such as Newtonian mechanics and fluid dynamics);

[0233] S336: Perform particle resampling.

[0234] S340: Combines parallel rendering results (volume rendering image and particle system image) into a scene image;

[0235] S350: Based on the distance of objects in the scene, determine the level of detail (LOD) and apply corresponding rendering optimization strategies;

[0236] Specifically, step S351: divide the scene into foreground and background;

[0237] S352: For close-up scenes, high-precision rendering is performed first: calculations are performed using more detailed models, higher texture resolution, and smaller ray step lengths, and then subsurface scattering is performed: simulating the physical effect of light scattering inside translucent materials, greatly enhancing the realism of the rendering;

[0238] For distant scenes, a simplified rendering is first performed: simplified models, low-resolution textures, larger light step lengths, and even a reduction in the number of particles are used. Then, an Impostor replacement (or pseudo-particle replacement) is performed: complex 3D objects are replaced with pre-rendered 2D sprite images, thereby significantly reducing the overhead of drawing calls and computation.

[0239] S360: Outputs the rendered image data in the frame buffer according to the settings.

[0240] In summary, the scene layer maps the flow field data output by the data layer into dynamic textures and implements optical effects through shader programming. In the case of land-sea integrated scenes, the oblique photography and ocean water data are unified into a spatial reference based on the 3D model standard, and the rendering load is optimized by using LOD (Level of Detail) technology.

[0241] The specific data processing method involves streaming a four-dimensional data field (longitude, latitude, depth / height, and time) via WebGL (a 3D graphics protocol). This enables physically realistic dynamic rendering and integrated land-sea scene fusion. Based on data-driven dynamic simulation of ocean water bodies, it achieves millimeter-level precision visualization of complex ocean phenomena, improving dynamic process fidelity by over 90% compared to the static representation of traditional GIS. It employs GPU-accelerated volume rendering and a particle system (SPH) to support real-time rendering of hundreds of millions of grid data points. Through 3D Tiles+ dynamic texture mapping technology, it achieves seamless fusion of oblique photogrammetry models and ocean water body data, improving the efficiency of coastal zone scene construction and supporting centimeter-level terrain matching. Built-in LOD optimization reduces the loading time of large-scale scenes to within 3 seconds while maintaining optical realism.

[0242] The service layer operates as follows in this embodiment:

[0243] S410: Receives client requests;

[0244] S420: Parse the request content and determine whether the request type is a real-time interactive request or a batch calculation request;

[0245] For real-time interactive requests, differential compression transmission is used to schedule the request to the edge computing node closest to the user's geographical location, which greatly reduces latency and bandwidth consumption.

[0246] If it is a batch computing request, the request will be scheduled to a central cloud with powerful computing and storage capabilities, mobilizing a large amount of resources to process the full amount of data;

[0247] S430: Requests enter the internal microservice network through the service network, providing basic capabilities such as service discovery, load balancing, circuit breaking and degradation, ensuring that requests can be reliably distributed to the correct backend service instances;

[0248] S440: Perform security checks (such as requesting identity authentication, permission verification, and security scanning). If the checks pass, proceed to the next step.

[0249] If the request fails the verification, the request processing flow will be terminated, and the illegal or invalid access attempt will be logged in the audit log for security analysis and tracing.

[0250] S450: Dynamically assemble (or orchestrate) the required microservice chain based on the request type;

[0251] S460: Allocate appropriate computing resources in the cluster based on the type and size of the request;

[0252] For GPU-first real-time rendering services, allocate a GPU instance to the request, start the real-time rendering service, and enable dynamic LOD control to adjust rendering details in real time based on user interaction to ensure smooth frame rates.

[0253] If it is a CPU-based elastic analysis and computing service, then allocate a CPU cluster instance, start the analysis and computing service, and cache the calculation results to avoid repeated calculations for requests with the same parameters and improve response speed.

[0254] S470: The processing results are returned to the client via streaming, avoiding waiting for all processing to be completed before returning, realizing processing and transmission at the same time, and improving the user experience;

[0255] S480: The client receives the data stream and performs the final decoding and presentation for visualization analysis and evaluation.

[0256] In summary, the service layer uses CDN (Content Delivery Network) edge computing to distribute 3D scene data, and enables multi-user collaborative editing through the WebSocket protocol (communication protocol); it adopts the lightweight format of 3D Tiles (an open specification for streaming and rendering large-scale, heterogeneous 3D spatial data) + glTF (3D model format), and supports GPU instantiation rendering on the browser side.

[0257] Scene deployment can automatically generate service metadata (compatible with OGC standard services); dynamic updates use differential data push. It achieves lightweight and high-concurrency support; data volume is reduced by 80%, supporting high-concurrency online access; it integrates OGC standard services (map services / feature services) and is 100% compatible with third-party GIS platforms, such as ArcGIS (the platform developed by Esri) and SuperMap software, reducing the cost of business system integration.

[0258] The application layer operation principle in this embodiment is as follows:

[0259] S510: Developer inputs requirements;

[0260] S520: Determine the development mode, that is, determine the final form and runtime environment of the developer's target application:

[0261] If it is determined to be a web application (a computer program stored on a remote server and run by its users through a web browser), load the corresponding plugin (such as a visualization plugin) and initialize the web rendering environment;

[0262] If it is determined to be a native application, then the bridging plugin with Unity (a real-time 3D interactive content creation and operation platform) or Unreal Engine (a game engine) is activated to establish a communication link between the Web service layer and the game engine, and the data is handed over to the native engine for processing;

[0263] S530: Transforms the developer's requirements into specific data service call parameters and selects the required data source for configuration;

[0264] S540: Based on the developer's input intent, match application templates from the template library, inject the configured data source and user parameters, generate basic scenarios, avoid developing from scratch, and achieve rapid application construction and standardization;

[0265] S550: Determines whether the application requires dynamic, energy-intensive calculations or simulations in the backend;

[0266] If needed, for lightweight computing, the WebAssembly module (WASM for short, a virtual instruction set architecture designed for web pages and web platforms) is scheduled to perform calculations directly on the browser side; for heavy simulation, tasks are submitted to the service layer via API, and the service layer schedules high-performance computing clusters.

[0267] If not, it enters pure rendering mode, where the application only relies on pre-generated data (such as tiled maps and cached result files) for visualization rendering and does not trigger backend calculations;

[0268] S560: Associates the dynamic result data generated by S550 with the basic scene generated in S540, making the static scene "come alive", responding to data changes and updating the visualization effect in real time;

[0269] S570: Detects the hardware capabilities of the running terminal. For example, high-configuration devices can enable high-resolution textures and finer meshes, while ordinary devices can reduce rendering resolution and simplify model details.

[0270] S580: Based on device capability adaptation, adjust the on / off state and level of visual effects according to user preferences or scenario requirements to achieve the best balance between image quality and performance.

[0271] S590: Encapsulates the configuration, data, logic, and rendering results generated in the above steps and outputs a visualization component.

[0272] In summary, the application layer uses a B / S architecture to encapsulate components for typical marine phenomena such as typhoon visualization and flooding analysis, and provides declarative APIs; the C / S architecture provides secondary development through an entity component system framework.

[0273] The system aggregates multi-source data through service calls using the GraphQL interface (a query language and runtime for APIs that allows clients to precisely specify the data they need); local caching utilizes IndexedDB (a browser database) + Service Worker (a script that runs independently in the browser background) for offline support. This enables efficient cross-platform development. Developers can implement marine applications with minimal code, significantly improving development efficiency; combined with various marine numerical models and built-in algorithm modules, secondary development cycles are shortened, allowing for rapid response to disaster emergency decision-making needs.

[0274] Through the above implementation methods, this invention primarily addresses the issues of real-time interaction and multi-dimensional fusion in scientific data visualization. It breaks through the traditional offline batch processing model, solving the challenges of low-latency parsing and dynamic interaction in scientific data output, such as CFD, ROMS, and FVCOM models (three-dimensional primitive equation ocean numerical models). It achieves coupled visualization of multi-dimensional (physical, chemical, and biological parameters) and multi-scale (global, regional, and local) spatiotemporal data, enhancing the collaborative analysis capabilities of complex ocean phenomena. Furthermore, it lowers the professional threshold by using intelligent rendering and automatic feature extraction to reduce reliance on manual interpretation, thereby enhancing the direct application value of scientific data in engineering decision-making.

[0275] It also addresses the shortcomings of GIS technology in the dynamic three-dimensional representation of the ocean. It enhances the ability to create high-precision models integrating vertical profiles, underwater topography, and water bodies, compensating for the deficiencies of traditional GIS in representing the vertical dimension of the ocean; it improves the dynamic-driven visualization algorithms for ocean phenomena (such as storm surges, ocean currents, and thermohaline fields) to avoid distorted representations such as "water rising from flat ground," ensuring that the simulation conforms to the laws of fluid mechanics; and it constructs standardized interfaces to achieve seamless integration with professional ocean models such as ADCIRC and ROMS, ensuring the spatiotemporal consistency between numerical simulation and visualization.

[0276] This also addresses the challenges of dynamic rendering and optical realism in marine scenes. A data-driven dynamic model of the ocean water body is established to realistically depict the physical evolution of processes such as eddies and pollutant diffusion. Efficient rendering of four-dimensional (three-dimensional space + time) data fields is optimized to support dynamic simulations of the marine environment with second-level updates. Realistic marine optical models are introduced, such as the sea surface BRDF (biaxial reflectivity distribution function), water volume scattering, and refraction / reflection effects, to improve the comparability of remote sensing observations of virtual marine scenes and enhance simulation credibility.

[0277] Therefore, marine digital twins are, on the one hand, digital modeling and interaction with the physical world (real-time information and data exchange), and on the other hand, they also play the roles of "foresight" (computation based on clear mechanisms), "precognition" (inference based on unclear mechanisms), and achieving "shared intelligence" (multiple twins sharing wisdom). That is, through interactive feedback between virtual and physical entities, data fusion analysis, and iterative optimization of decision-making, digital twins can add new capabilities or expand the functions of physical entities, which is something that "traditional simulation" and "digital ocean" cannot achieve.

[0278] In addition, the above implementation methods fully consider the characteristics of various types of marine data in terms of dedicated digital twin engine architecture for marine applications, and have developed a dedicated access gateway for multi-source marine data; they have pioneered a containerization standard for marine models, and developed intelligent scheduling functions for model computing resources and unified model management functions.

[0279] A visualization methodology system for ocean observation parameters and typical ocean phenomena has been developed. An innovative dedicated renderer for ocean elements has been developed, employing volume rendering technology to visualize data such as temperature and salinity fields, and using particle tracking-based dynamic streamlines to represent flow fields and other information. A multi-dimensional representation method for the ocean surface environment, water body environment, and seabed topography environment has been established, supporting continuous rendering from the surface to the bottom layers.

[0280] The open marine twin scene construction platform provides a visual scene editor, a development scene template library, and multi-level development interfaces, enabling seamless integration of marine data, models, and applications.

[0281] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A marine digital twin engine system, characterized in that, include: Data layer, model layer, scenario layer, service layer, and application layer; The data layer is used to uniformly access multi-source ocean data and perform spatiotemporal alignment and quality control. The model layer is used to schedule ocean numerical models and supports multiphysics joint simulation. The scene layer is used to achieve true 3D ray tracing rendering of marine phenomena; The service layer is used to publish lightweight scenario services; The application layer is used to enable application development.

2. The marine digital twin engine system according to claim 1, characterized in that, The data layer operates as follows: S110: Input raw data; S120: Classify data according to its source and information; S130: Preprocessing based on data type: Convert satellite remote sensing data into spectral reflectance or spectral radiance using calibration coefficients. For numerical pattern data: Read the NetCDF file structure, parse the information, and extract the multidimensional array; Perform CRC check on the buoy data: calculate the cyclic redundancy check code and compare it with the check code attached to the data transmission packet; S140: Unify all the above data onto the same spatiotemporal reference; S150: Perform anomaly detection on the data; S160: For normal data that passes the detection, perform feature extraction and calculate key feature indicators; For abnormal data, GAN repair is performed first, and then feature extraction is performed on the repaired data; S170: Perform EOF compression; S180: Organize multidimensional data into data cubes; S190: Output the data cube in different formats according to the needs of downstream applications.

3. The marine digital twin engine system according to claim 2, characterized in that, The operating principle of the model layer is as follows: S210: Input the data required for the model to run; S220: Automatically generate or start standardized container instances based on model configuration and requirements; S230: Based on the characteristics of the model algorithm, automatically determine and select the optimal computing resource deployment strategy; For CPU-intensive tasks, choose MPI cluster deployment; For GPU-accelerated tasks, choose CUDA container deployment; S240: Perform the core numerical computations of the model on the allocated computing resources; S250: Determine whether bidirectional coupling with other modes is required based on the simulation task configuration; If necessary, proceed to the next step and perform the coupling loop process; If not, proceed to the final result output process; S260: Coordinated via WSMF coupler; S270: Perform data mapping; S280: Perform time synchronization; S290: Perform coupled data exchange, then return to S240 until the simulation is complete.

4. The marine digital twin engine system according to claim 3, characterized in that, The scene layer operates as follows: S310: Obtain the raw data to be rendered; S320: Analyze the raw data, determine the data type, and classify the data type into hydrological element field data and ocean dynamic phenomenon data; S330: Performs rendering processing based on different data; S340: Combines the parallel rendering results into a scene image; S350: Based on the distance of objects in the scene, it performs a level of detail assessment and applies corresponding rendering optimization strategies; S360: Outputs the rendered image data in the frame buffer according to the settings.

5. The marine digital twin engine system according to claim 4, characterized in that, The rendering process in step S330 includes: If the data is from a hydrological element field, then proceed as follows: S331: Reconstruct discrete scalar field data into continuous three-dimensional volume data; S332: Based on a physical model, calculate the optical properties of light as it propagates within volume data; S333: Perform light step rendering; If the data pertains to ocean dynamic phenomena, then proceed as follows: S334: Generate a set of particles representing individuals based on the original data; S335: Updates the state of all particles in each frame according to the laws of physics; S336: Perform particle resampling.

6. The marine digital twin engine system according to claim 5, characterized in that, Step S350 includes: S351: Divide the scene into foreground and background; S352: For close-up scenes, perform high-precision rendering first: use more detailed models, higher texture resolution, and smaller ray step length for calculation; Then, subsurface scattering is performed: simulating the physical effect of light scattering inside a translucent material; For distant views, first perform simplified rendering, then replace the Impostor.

7. A marine digital twin engine system according to claim 6, characterized in that, The service layer operates as follows: S410: Receives client requests; S420: Parse the request content and determine whether the request type is a real-time interactive request or a batch calculation request; For real-time interactive requests: differential compression transmission is used, and the request is scheduled to the edge computing node closest to the user's geographical location; If it is a batch computing request: the request will be scheduled to a central cloud with powerful computing and storage capabilities, and a large amount of resources will be mobilized to process the full amount of data; S430: Request to access the internal microservice network through the service network; S440: Conduct a safety inspection. If the inspection is passed, proceed to the next step. If the request fails the inspection, the request processing flow will be terminated and the information will be logged in the audit log for security analysis and tracing. S450: Dynamically assemble the required microservice chain based on the request type; S460: Allocate appropriate computing resources in the cluster based on the type and size of the request; If it is a GPU-first real-time rendering service, allocate a GPU instance to the request, start the real-time rendering service, and enable dynamic LOD control; If it is a CPU-based elastic analytics service, then allocate a CPU cluster instance, start the analytics service, and cache the calculation results; S470: Returns the processing results to the client via streaming; S480: The client receives the data stream and performs the final decoding and presentation.

8. A marine digital twin engine system according to claim 7, characterized in that, The application layer operates as follows: S510: Developer inputs requirements; S520: Determine the development mode: If it is determined to be a web application, load the corresponding plugin and initialize the web rendering environment; If it is determined to be a native application, then activate the bridging plugin with Unity or Unreal Engine; S530: Transforms the developer's requirements into specific data service call parameters and selects the required data source for configuration; S540: Based on the developer's input intent, match the application template from the template library, inject the configured data source and user parameters, and generate the basic scene; S550: Determines whether the application requires dynamic, energy-intensive calculations or simulations in the backend; If necessary, for lightweight computing, the WebAssembly module can be scheduled to perform the computation directly on the browser side; For heavy-duty simulations, tasks are submitted to the service layer via API, and the service layer schedules the high-performance computing cluster. If not, then enter pure rendering mode; S560: Associate the dynamic result data generated by S550 with the basic scene generated in S540; S570: Detects the hardware capabilities of the running terminal; S580: Based on device capability adaptation, adjust the on / off state and level of visual effects according to user preferences or scenario requirements; S590: Encapsulates the configuration, data, logic, and rendering results generated in the above steps and outputs a visualization component.

9. A marine digital twin engine system according to claim 2, characterized in that, The output format in step S190 includes: PARQUET Data Lake: Outputs data in a columnar storage format, suitable for efficient and low-cost batch processing and ad-hoc queries within big data analytics frameworks; HDF5 numerical model: Outputs in a hierarchical format that is standard in the field of scientific computing, retaining complete metadata and dimensional information for subsequent numerical simulation or in-depth analysis; GEOJSON application platform: Outputs point data or aggregated polygon data in geospatial vector format for visualization and interaction in WebGIS application platforms.

10. A marine digital twin engine system according to claim 3, characterized in that, The final result output process in step S250 includes: S251: Output the final result; Output formats include VTK or NetCDF. S252: Generate visual analysis results.

Images