A high performance computing based marine environment forecasting method and system

Abstract

The application relates to the field of environment prediction, and discloses a marine environment prediction method and system based on high-performance computing, which are used for forming a marine environment prediction technology system. The method comprises the following steps: converting physical marine observation data into initial field data, and initializing a particle distribution function set; performing lattice Boltzmann evolution calculation in a high-performance computing cluster parallel environment, coupling a collision operator and a continuous transmission process, and dynamically generating an adaptive grid; performing global synchronization on updated marine state data; and converting the marine state data of the whole sea area into a prediction field with physical dimensions. The application compares and analyzes the prediction field and the measured data, generates a quantitative prediction product with credibility information, provides auxiliary decision-making information for maritime activities, and effectively improves the precision, efficiency and practicability of marine environment prediction.

Description

Technical Field

[0001] This invention relates to the field of environmental forecasting, and more particularly to a marine environmental forecasting method and system based on high-performance computing. Background Technology

[0002] With the intensification of global climate change and the increasing frequency of activities such as marine resource development, maritime transport, and marine engineering, the need for accurate forecasting of the marine environment is becoming increasingly urgent. The marine environment is a complex dynamic system influenced by the interaction of multiple physical processes, including ocean dynamics, thermohaline processes, and the Coriolis effect caused by the Earth's rotation. These processes are coupled with each other, making the marine environment highly nonlinear and complex, posing a significant challenge to accurately simulating and forecasting its state.

[0003] Traditional marine environmental forecasting methods are mainly based on numerical models of continuum mechanics, such as the finite difference method, the finite element method, and the finite volume method. These methods simulate marine physical processes by discretizing the oceanic governing equations and solving them numerically on a fixed grid. However, with the continuous deepening of marine scientific research and the increasing demands for forecast accuracy and timeliness, traditional methods have gradually revealed some limitations.

[0004] Traditional methods have difficulty handling complex ocean boundaries and irregular geometries. Fixed meshes are difficult to accurately fit complex terrains such as coastlines and islands, resulting in large errors in simulation results near the boundaries.

[0005] For some physical phenomena in the ocean with local high gradient characteristics, such as fronts and vortices, traditional fixed-grid methods often require extremely high grid resolution to accurately capture them. This undoubtedly increases the amount of computation and the computational cost significantly, limiting its application in large-scale ocean simulations.

[0006] Traditional methods also face challenges in parallel computing. While partitioned parallel computing can improve computational efficiency, data synchronization and communication between different computing nodes can easily lead to decreased computational efficiency due to the large amount of data exchange and complex communication modes, making it difficult to fully leverage the advantages of high-performance computing clusters.

[0007] Therefore, we propose a high-performance computing-based method and system for marine environmental forecasting to address the aforementioned problems. Summary of the Invention

[0008] This invention provides a marine environment forecasting method and system based on high-performance computing, which can be used to form a marine environment forecasting technology system.

[0009] The first aspect of this invention provides a marine environment forecasting method based on high-performance computing. The method includes: converting physical oceanographic observation data of a target sea area into initial field data; initializing a particle distribution function set based on the initial field data; wherein the particle distribution function set includes an initial equilibrium set of the main distribution function, an initial set of scalar distribution functions, and a set of Coriolis force source terms; and, in a parallel environment of a high-performance computing cluster, performing evolutionary calculations based on the particle distribution function set to update the ocean state. The evolutionary calculations couple collision operators and continuous transport processes for the marine environment and dynamically generate adaptive meshes based on flow field characteristics, including: performing local collision calculations in parallel within each sub-computational domain based on the particle distribution function set, wherein the collision calculations couple the rotational effect provided by the set of Coriolis force source terms. The buoyancy effect induced by the initial set of scalar distribution functions generates an updated set of distribution functions; the updated set of distribution functions is continuously transmitted along a preset discrete velocity direction based on floating-point coordinates to generate a transmitted set of distribution functions; based on the transmitted set of distribution functions, intermediate ocean state data including flow velocity, temperature, and salinity information in each sub-computation domain is calculated; adaptive grid control parameters are generated based on the gradient characteristics of the intermediate ocean state data; based on the adaptive grid control parameters, the computational grid of each sub-computation domain is dynamically refined or coarsened to generate an updated adaptive grid structure and a corresponding set of particle distribution functions and ocean state data; the updated ocean state data distributed across each computation node is globally synchronized to obtain full-ocean-area state data; and a marine environmental forecast field is obtained based on the full-ocean-area state data.

[0010] Optionally, in a first implementation of the first aspect of the present invention, the method includes: acquiring historical observation data or reanalysis data of the ocean dynamic field and ocean thermohaline field of the target sea area; determining characteristic length, characteristic time, and characteristic velocity based on the physical characteristics of the target sea area; processing the historical observation data or reanalysis data of the ocean dynamic field and ocean thermohaline field using the characteristic length, characteristic time, and characteristic velocity as characteristic scale quantities to generate a dimensionless ocean parameter field; and interpolating and mapping the dimensionless ocean parameter field onto a preset continuous computation grid to form initial field data.

[0011] Optionally, in a second implementation of the first aspect of the present invention, based on the intermediate ocean state data, the vorticity value of the velocity field, the gradient value of the temperature field, and the gradient value of the salinity field are calculated respectively to generate multiphysics gradient feature data; according to a preset grid adjustment physical criterion, the multiphysics gradient feature data is compared with the corresponding dynamic threshold to generate a preliminary grid adjustment region identification field; based on the grid adjustment region identification field and combined with a preset grid resolution level rule, a target resolution level is determined for each identified region to generate a set of grid refinement or coarsening instructions; the set of grid refinement or coarsening instructions is processed to generate adaptive grid control parameters.

[0012] Optionally, in a third implementation of the first aspect of the present invention, the method includes: determining overlapping boundary regions that need to be exchanged between sub-computation domains based on the updated adaptive grid structure; extracting data corresponding to the overlapping boundary regions from the ocean state data stored in each computing node to form boundary data blocks to be exchanged; exchanging the boundary data blocks to be exchanged between computing nodes through a parallel communication interface; processing the received boundary data blocks from adjacent sub-computation domains with the ocean state data of the current sub-computation domain to generate full sub-computation domain state data; and collecting the full sub-computation domain state data from all computing nodes to form full ocean state data.

[0013] Optionally, in the fourth implementation of the first aspect of the present invention, based on the updated adaptive mesh structure, the geometric features and data scale of the overlapping boundary regions of each sub-computation domain are analyzed to generate an optimized list of parallel communication tasks; according to the list of parallel communication tasks, each computing node is scheduled to send and receive the boundary data blocks to be exchanged through non-blocking communication mode to generate a set of received data blocks; the external boundary data in the set of received data blocks and the local boundary data are processed to generate a boundary consistency processing result; the boundary consistency processing result is verified and checked to generate a synchronization completion flag.

[0014] Optionally, in the fifth implementation of the first aspect of the present invention, the verification check of the boundary consistency processing result includes: performing spatial momentum conservation interpolation to verify the continuity of momentum flux between adjacent overlapping grid layers; calculating the absolute difference between the local momentum flux and the remote interpolated momentum flux at the boundary of adjacent sub-computation domains; if the absolute difference is lower than a preset tolerance threshold, determining that data synchronization and exchange across resolution boundaries are continuous; if the absolute difference is not lower than the preset tolerance threshold, triggering boundary data retransmission or adjusting the interpolation weights.

[0015] Optionally, in the sixth implementation of the first aspect of the present invention, the method includes: using a predetermined characteristic scale quantity to perform dimensional reduction calculation on the full-ocean state data to generate ocean flow velocity field, temperature field and salinity field data; organizing the flow velocity field, temperature field and salinity field data according to a preset spatial grid structure and forecast time series to generate marine environmental element forecast field data; and encapsulating the marine environmental element forecast field data and writing it into a forecast data file to form a marine environmental forecast field.

[0016] Optionally, in the seventh implementation of the first aspect of the present invention, the method further includes: acquiring measured observation data of the target sea area matching the forecast period; performing a matching and comparative analysis between the marine environmental forecast field and the measured observation data to generate forecast error assessment data; calculating the deterministic index and uncertainty range of the forecast result based on the forecast error assessment data to generate a quantitative forecast product with credibility information; and generating auxiliary decision-making information based on the quantitative forecast product with credibility information and a preset marine application scenario decision-making rule.

[0017] A second aspect of the present invention provides a marine environment forecasting system based on high-performance computing. The system includes: a conversion module for converting physical oceanographic observation data of a target sea area into initial field data; an initialization module for initializing a set of particle distribution functions based on the initial field data; wherein the set of particle distribution functions includes an initial equilibrium set of the main distribution function, an initial set of scalar distribution functions, and a set of Coriolis force source terms; and a parallel module for performing evolutionary calculations based on the set of particle distribution functions in a parallel environment of a high-performance computing cluster to update the ocean state. The evolutionary calculations couple collision operators and continuous transport processes for the ocean environment and dynamically generate adaptive meshes based on flow field characteristics, including: performing local collision calculations in parallel within each sub-computational domain based on the set of particle distribution functions, wherein the collision calculations are coupled with rotational forces provided by the set of Coriolis force source terms. The system generates an updated distribution function set by considering the transfer effect and the buoyancy effect induced by the initial set of scalar distribution functions. This updated distribution function set is then continuously transmitted along a preset discrete velocity direction using floating-point coordinates to generate a transmitted distribution function set. Based on the transmitted distribution function set, intermediate ocean state data, including flow velocity, temperature, and salinity information, is calculated for each sub-computation domain. Adaptive grid control parameters are generated based on the gradient characteristics of the intermediate ocean state data. According to the adaptive grid control parameters, the computational grid of each sub-computation domain is dynamically refined or coarsened to generate an updated adaptive grid structure and its corresponding particle distribution function set and ocean state data. A synchronization module is used to globally synchronize the updated ocean state data distributed across each computation node to obtain full-ocean-area state data. A forecasting module is used to obtain a marine environmental forecast field based on the full-ocean-area state data.

[0018] The mechanism of this invention is as follows: It solves the fundamental contradiction between computational accuracy and resource consumption through dynamic grid technology, and describes the interaction process of multiple ocean spheres in a unified manner with mesoscopic particle evolution, thereby significantly improving the accuracy of forecasts and computational efficiency. Beneficial effects: It can capture the interaction between multiple physical fields in the ocean more accurately, break through the limitations of traditional single-component simulation, reduce the error caused by model simplification, make the forecast results closer to the actual ocean state, and provide a more reliable basis for ocean activities; It dynamically generates adaptive meshes based on flow field characteristics. It can automatically adjust the mesh resolution according to the physical complexity of different ocean regions, concentrate computing resources in high-gradient regions, avoid wasting resources in calm regions, improve the utilization rate of computing resources, accurately simulate complex flow fields, further improve forecast accuracy, and meet the needs of real-time forecasting; In a high-performance computing cluster parallel environment, a complete and efficient parallel computing process is formed, from local collision calculation to boundary data exchange and fusion. This strategy reduces communication bottlenecks, fully leverages the powerful computing capabilities of the cluster, significantly shortens forecast calculation time, greatly enhances computational efficiency, and ensures the timeliness of forecasts. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of one embodiment of the marine environment forecasting method based on high-performance computing in this invention.

[0020] Figure 2 This diagram illustrates the process of converting physical oceanographic observation data (ocean dynamic field and thermohaline field) into initial field data.

[0021] Figure 3 This is a schematic diagram of another embodiment of the marine environment forecasting method based on high-performance computing in this invention.

[0022] Figure 4 This is a schematic diagram of one embodiment of the marine environment forecasting system based on high-performance computing in this invention.

[0023] Figure 5 This is a schematic diagram of one embodiment of a marine environmental forecasting device based on high-performance computing according to the present invention. Detailed Implementation

[0024] This invention provides a marine environmental forecasting method and system based on high-performance computing, used to form a marine environmental forecasting technology system. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the marine environment forecasting method based on high-performance computing in this invention includes: 101. Convert the physical oceanographic observation data of the target sea area into initial field data.

[0026] It is understood that the executing entity of this invention can be a marine environmental forecasting system based on high-performance computing, or it can be a terminal or a server; no specific limitation is made here. This embodiment of the invention will be described using a server as an example.

[0027] It should be noted that real-time observation data of the target bay is retrieved from the maritime monitoring database. Assuming a buoy station in the center of the bay, the raw physical data measured by the sensor is: eastward surface current velocity u. x =0.5 meters per second, northward flow velocity u y =0.2 meters per second, measured seawater density kilograms per cubic meter.

[0028] To perform high-performance computing, a dimensionless reference system needs to be established. The characteristic length (grid step size) of the basic level (coarse grid) is set to 100 meters, and the characteristic time (time step size) is set to 1 second. Thus, the grid reference velocity is 100 meters per second, and the standard seawater density of 1000 kilograms per cubic meter is set as the characteristic reference density.

[0029] A dimensionless transformation is performed: the actual flow velocity is divided by the reference velocity to obtain a dimensionless eastward flow velocity of 0.005 (i.e., 0.5 / 100) and a dimensionless northward flow velocity of 0.002 (i.e., 0.2 / 100); the actual density is divided by the reference density to obtain a dimensionless density value of 1.026. In the LBM method, the physical speed of sound in real seawater is extremely high (approximately 1500 m / s). To accelerate numerical simulation and maintain computational stability, an amplified reference velocity is typically used to introduce a moderate degree of artificial compressibility. The dimensionless transformation in this scheme ensures that the computational Mach number (the ratio of flow velocity to lattice sound velocity) during numerical solution is much less than 1.

[0030] These dimensionless values ​​are precisely mapped to the corresponding nodes of the computational grid. For blank areas without monitoring equipment, the system uses a spatial kriging interpolation algorithm, combined with the shoreline boundary friction coefficient, to generate a dimensionless initial field matrix covering the entire bay area. This matrix removes physical dimensions, transforming the complex ocean conditions into a purely numerical sequence.

[0031] 102. Initialize the particle distribution function set using the initial field data.

[0032] It should be noted that the macroscopic dimensionless physical field is decomposed into the distribution state of the underlying microscopic particles. The system adopts a two-dimensional nine-velocity (D2Q9) model based on a single relaxation time (BGK) collision operator, that is, the fluid state at each grid node is represented by particle distribution functions in nine discrete directions.

[0033] Theoretical weights are assigned to the nine directions: the stationary direction has a weight of 4 / 9, the orthogonal directions (east, west, south, and north) each have a weight of 1 / 9, and the diagonal directions each have a weight of 1 / 36.

[0034] Taking the central grid point of the bay as an example, its input data includes a dimensionless density of 1.026, an eastward flow velocity of 0.005, and a northward flow velocity of 0.002. The server performs numerical calculations using the equilibrium distribution equation: ; in For direction weights, For discrete velocity directions, Let be the lattice sound velocity. The system substitutes the dimensionless density and velocity vector into the formula to initialize the distribution function in each direction. The calculated function value for the stationary direction at this point is approximately 0.4560. Furthermore, for island and coastline areas, the system employs a half-step bounce scheme to handle solid boundary conditions. For the eastward direction, due to the strengthening effect of the positive velocity flow, the calculated value is 0.1157; while the corresponding westward direction, due to the weakening effect of the countercurrent, has a value reduced to 0.1123. Similarly, the calculated value for the northward direction is 0.1147.

[0035] Such computations are executed in parallel within the distributed memory of a high-performance computing cluster, ensuring that hundreds of millions of grid points across the entire ocean area receive the corresponding nine-axis distribution function values. The sum of these function values ​​at each node is exactly equal to the dimensionless density at that point. ).

[0036] The physical state of the sea area has been completely transformed into a set of microscopic particles that can be identified by high-performance computing clusters, thus providing memory pre-storage for parallel evolutionary computing.

[0037] 103. In the parallel environment of a high-performance computing cluster, lattice Boltzmann evolution calculations are performed based on a set of particle distribution functions to update the ocean state. The evolution calculations couple collision operators with continuous transport processes for the ocean environment and dynamically generate adaptive grids based on flow field characteristics.

[0038] It should be noted that this step is the stage for deriving the core dynamics of sea state, which involves the dynamic adjustment of the grid and the coupling of the physical environment.

[0039] The cluster detected a sharp flow field gradient in the bay mouth region, triggering adaptive refinement of the local grid, reducing the grid step size from 100 meters to 25 meters. To maintain physical consistency across different levels of evolution, the system simultaneously shortened the time step in this region to 0.25 seconds, ensuring a consistent grid velocity between the finer grid and the basic coarser grid. Strictly maintain constancy and consistency.

[0040] Real-time meteorological data is introduced, assuming the sea surface is currently affected by a southwest wind (blowing northeast) with a speed of 8 m / s. The server converts the corresponding wind stress into microscopic discrete force terms. During the collision process, particles originally moving northeast are propelled by the downwind, and their distribution function value is enhanced; the eastward distribution function value of a certain node is updated from 0.1157 to 0.1168 through collision evolution. Simultaneously, the system automatically incorporates the Coriolis effect at 30 degrees North latitude, performing macroscopic trajectory deflection corrections for long-term forecasts and accurately calculating the rotational deflection effect in short-term forecasts.

[0041] After a collision, the particles move to adjacent grids according to their discrete directions. Computational nodes (CPU cores) responsible for different regions perform high-speed boundary exchanges through the InfiniBand network. When a particle representing water flowing eastward crosses from the computational block responsible for core A to the computational block responsible for core B, the system completes data packaging and synchronization with extremely low network latency, ensuring the continuity of physical flow.

[0042] Through this high-frequency iteration, the system simulates the complex interaction of microscopic particles under the combined action of wind field, rotation force and pressure gradient, thereby deducing the dynamic migration process of macroscopic flow field.

[0043] 104. Perform global synchronization of the updated ocean state data distributed across all computing nodes to obtain ocean state data for the entire sea area.

[0044] It should be noted that the cluster has completed the predetermined number of evolutionary calculations, and the latest sea state data for the entire bay area is distributed and stored in the independent memory of 10,000 computing nodes. Step 104 aims to restore these fragmented local data into a consistent global state.

[0045] A global synchronization command is issued, and each node begins to execute data reduction operations. Taking node 512, which is responsible for refining the grid at the bay mouth, as an example, it packages the latest distribution function of 40,000 fine grid points in memory and attaches precise geospatial labels.

[0046] Through the cluster's internal 10 Gigabit switching network, 10,000 nodes simultaneously send data packets to the master storage node. The master system constructs a macroscopic topology matrix covering the entire bay area in memory, piecing together data packets based on their tags, much like stitching together a puzzle. Figure 1 This allows local data to be filled into the global coordinate system.

[0047] The key lies in boundary consistency handling: due to the existence of the adaptive grid, at the boundary between the 100-meter step size region and the 25-meter step size region, the system performs spatial interpolation smoothing to eliminate numerical jumps caused by resolution differences. Simultaneously, the system removes redundant shadow grid data retained between nodes for communication purposes. The server obtains a dimensionless distribution function full matrix covering the entire sea area, including hundreds of millions of grid points, and is logically continuous. This matrix represents the unified forecast status of the entire sea area at the current moment.

[0048] 105. Convert the state data of the entire sea area into a marine environmental forecast field with physical dimensions.

[0049] It should be noted that this involves remapping abstract dimensionless numerical values ​​back to the real physical world.

[0050] Moment operations are performed on the globally synchronized distribution function set. For each grid point, the numerical values ​​in nine directions are summed to restore the macroscopic dimensionless density, and the momentum in each direction is weighted and summed to restore the dimensionless velocity component.

[0051] Taking the center of the bay as an example, after one hour of southwest wind blowing, the system calculated that the dimensionless density at this point remains conserved at 1.0260, the dimensionless eastward flow velocity increases to 0.006 due to wind stress, and the northward flow velocity increases to 0.003.

[0052] Perform a reverse conversion using the reference standard from step 101: 1. Density reduction: (Density remains stable under the assumption of incompressible fluid).

[0053] 2. Velocity Reduction: Dimensionless Eastward meters per second; dimensionless northward direction meters per second.

[0054] Table 1: Evolution of Forecast Results Table 1 Initial time (T=0) 1.0260 0.005 / 0.002 1026.0 0.5 / 0.2 Static observation initial field Forecast time (T+1h) 1.0260 0.006 / 0.003 1026.0 0.6 / 0.3 Accelerating with tailwind (southwest wind) These restored data are used to generate standard gridded forecast products (ocean current distribution map, density and salinity gradient map), providing reliable quantitative decision-making basis for shipping wave avoidance and oil spill diffusion prediction.

[0055] Please see Figure 2 and Figure 3 Another embodiment of the marine environment forecasting method based on high-performance computing in this invention includes: 201. Convert the physical oceanographic observation data of the target sea area into initial field data.

[0056] Specifically, historical observation data or reanalysis data of the ocean dynamic field and ocean thermohaline field of the target sea area are acquired; based on the physical characteristics of the target sea area, the characteristic length, characteristic time, and characteristic velocity are determined; the ocean dynamic field and ocean thermohaline field data are dimensionless processed using characteristic scale quantities to generate a dimensionless ocean parameter field; the dimensionless ocean parameter field is interpolated and mapped onto a preset continuous computation grid to form initial field data.

[0057] It should be noted that this involves mapping a specific sea area in the western Pacific Ocean (20°N, 130°E) from physical observations to a numerical matrix. The server retrieved data from the Global Ocean Reanalysis System to identify the background field before the cyclone's influence: surface seawater temperature was 28.5℃, salinity was 34.80 PSU, and measured seawater density was 1022.5 kg / m³. 3 Influenced by the background circulation, the ocean current flows slowly to the northwest at a speed of 0.5 m / s.

[0058] To adapt to the Lattice Boltzmann (LBM) multi-component model, a characteristic reference system was established: the grid step size (characteristic length) was set to 500 meters, and the time step size (characteristic time) to 5 seconds, thus deriving a lattice reference velocity of 100 m / s. Simultaneously, a reference density of 1000 kg / m³ was set. 3 The reference temperature difference is 30℃, and the reference salinity is 35PSU.

[0059] Perform dimensionless processing: 1. Dynamic field: The actual flow velocity 0.5 divided by the reference velocity 100 yields a dimensionless flow velocity component of 0.005.

[0060] 2. Hot salt field: Temperature 28.5 divided by 30 gives a dimensionless temperature of 0.95; Salinity 34.80 divided by 35 gives a dimensionless salinity of 0.9943.

[0061] 3. Density field: The density 1022.5 divided by 1000 gives the dimensionless density 1.0225.

[0062] A Cartesian grid with a step size of 500 meters is constructed, and the aforementioned dimensionless parameters are mapped to global nodes using trilinear interpolation. For island and reef regions, the system uses masking techniques to handle solid boundaries. The generated four-dimensional dimensionless initial field matrix provides standardized, order-aligned underlying data for subsequent initialization.

[0063] 202. Initialize the particle distribution function set using the initial field data.

[0064] Specifically, a multi-component structure for the particle distribution function is defined, including a principal distribution function describing the flow field, a scalar distribution function describing the temperature field, and a scalar distribution function describing the salinity field. Based on the macroscopic flow parameters in the initial field data, the initial equilibrium set of the principal distribution function is calculated according to the equilibrium function relationship. Based on the temperature and salinity parameters in the initial field data, the scalar distribution functions for the temperature field and the salinity field are initialized respectively to form an initial set of scalar distribution functions. Combining the Earth's rotation parameters, the set of Coriolis force source terms acting on the principal distribution function is calculated. The initial equilibrium set of the principal distribution function, the initial set of scalar distribution functions, and the set of Coriolis force source terms are combined into a set of particle distribution functions.

[0065] It should be noted that this step aims to particleize the macroscopic parameters in 201. In the multi-component LBM model, the system needs to initialize three independent distribution function layers for the main flow field, temperature field, and salinity field at each grid point, and inject the Earth's rotation effect.

[0066] A D2Q9 structure is allocated in memory for each node. For the main flow field, the dimensionless density of 1.0225 and the flow velocity of 0.005 are read. Based on the equilibrium distribution function, the system calculates the particle proportions in nine discrete directions. Since the water flow is northwest, the calculated function value for the stationary direction (weight 4 / 9) is approximately 0.4544, while the particle distribution values ​​for the west and north directions are enhanced to 0.1141 and 0.1139 respectively due to momentum contributions.

[0067] For the thermosalinous scalar field, the system employs isotropic initialization. Dimensionless temperature (0.95) and dimensionless salinity (0.9943) are weighted and distributed to the discrete directions of their respective levels. At this point, the scalar sum of the distribution function at each level is strictly equal to the macroscopic initial value.

[0068] Considering the geographical location at 20°N latitude, the server incorporates Coriolis parameters. The system calculates acceleration compensation affected by rotational inertia based on the initial flow velocity vector. Due to the rightward deflection of forces in the Northern Hemisphere, the system generates a correction value for a microscopic force source term deflected northeastward and superimposes it onto the main distribution function layer. The server encapsulates these three layers of distribution functions and the Coriolis force source term into a composite particle set. The static ocean observation slice has been transformed into a dynamic microstate with momentum, energy, and inertial properties, ready to enter the parallel evolution stage.

[0069] 203. In the parallel environment of a high-performance computing cluster, lattice Boltzmann evolution calculations are performed based on a set of particle distribution functions to update the ocean state. The evolution calculations couple collision operators with continuous transport processes for the ocean environment and dynamically generate adaptive meshes based on flow field characteristics.

[0070] Specifically, based on the particle distribution function set, local collision calculations are performed in parallel within each sub-computational domain. These collision calculations couple the rotational effect provided by the Coriolis force source term set and the buoyancy effect induced by the scalar distribution function, generating an updated distribution function set. This updated distribution function set is then continuously transmitted along a preset discrete velocity direction using floating-point coordinates, generating a transmitted distribution function set. Based on the transmitted distribution function set, intermediate ocean state data, including flow velocity, temperature, and salinity information, is calculated for each sub-computational domain. Based on the gradient characteristics of the intermediate ocean state data, the physical complexity of the flow field is evaluated, generating adaptive grid control parameters. According to the adaptive grid control parameters, the computational grid of each sub-computational domain is dynamically refined or coarsened, and interpolation methods are used to migrate data to maintain continuity, generating an updated adaptive grid structure and the corresponding particle distribution function set and ocean state data.

[0071] Furthermore, based on intermediate ocean state data, the vorticity values ​​of the velocity field, the gradient values ​​of the temperature field, and the gradient values ​​of the salinity field are calculated to generate multiphysics gradient feature data. According to the preset grid adjustment physical criteria, the multiphysics gradient feature data is compared with the corresponding dynamic thresholds to identify high-intensity change regions that require grid refinement and low-intensity change regions that can be coarsened, generating a preliminary grid adjustment region identification field. Based on the grid adjustment region identification field and combined with the preset grid resolution level rules, a target resolution level is determined for each identified region, and the geometric transformation parameters required from the current grid to the target grid are calculated to generate a set of grid refinement or coarsening instructions. The set of grid refinement or coarsening instructions is then subjected to topological consistency verification and smooth transition processing of adjacent regions to generate adaptive grid control parameters.

[0072] It should be noted that high-performance computing power is used to extrapolate sea state dynamics. In cyclonic environments, evolution calculations require a high degree of coupling between wind field dynamics and adaptive grid adjustment.

[0073] Each computation node performs collision calculations on the particle distribution function. The system introduces the Ekman suction force generated by a strong cyclone wind field to simulate the vertical motion of surface seawater caused by wind stress divergence. The collision operator not only handles momentum exchange but also calculates buoyancy corrections based on the density differences induced by temperature and salinity distributions. Assuming a node near the eyewall, the system calculates that the suction effect causes an instantaneous increase of the upward particle distribution probability by 0.0003, simulating the upward trend of deep cold water.

[0074] After the collision, the particles move towards adjacent grid points along discrete directions. A particle carrying northwest momentum (value 0.1168) completes its cross-grid migration within one time step. Subsequently, the system summarizes and extracts intermediate states.

[0075] The velocity gradient in the cyclone's central region was detected to increase sharply from 0.005 to 0.012, along with a dramatic increase in the temperature gradient. Based on these gradient characteristics, the system issued a mesh refinement command, reducing the mesh size in the cyclone's eyewall region from 500 meters to 125 meters.

[0076] The particle distribution function is transferred to the new mesh using Lagrange interpolation. At the boundary of the fine-grid region, the system performs momentum conservation smoothing to ensure that the vortex structures captured by the fine mesh are accurately fed back to the global flow field.

[0077] 204. Perform global synchronization of the updated ocean state data distributed across all computing nodes to obtain ocean state data for the entire sea area.

[0078] Specifically, based on the updated adaptive grid structure, the overlapping boundary regions that need to be exchanged between each sub-computation domain are determined; data corresponding to the overlapping boundary regions are extracted from the ocean state data stored in each computing node to form boundary data blocks to be exchanged; the boundary data blocks to be exchanged are exchanged between each computing node through a parallel communication interface; the boundary data blocks received from adjacent sub-computation domains are fused and consistency-processed with the ocean state data of the current sub-computation domain to generate locally consistent full sub-computation domain state data; the locally consistent full sub-computation domain state data on all computing nodes are collected to form full ocean state data.

[0079] Furthermore, based on the updated adaptive grid structure, the geometric characteristics and data scale of the overlapping boundary regions of each sub-computation domain are analyzed to generate an optimized list of parallel communication tasks. According to the list of parallel communication tasks, each computing node is scheduled to send and receive boundary data blocks to be exchanged through non-blocking communication mode, generating a set of received data blocks including data from adjacent sub-computation domains. The external boundary data in the received data block set is fused and weighted averaged with the local boundary data to generate a boundary consistency processing result. The boundary consistency processing result is then subjected to integrity verification and numerical range rationality checks to generate a synchronization completion flag.

[0080] It should be noted that in an adaptive mesh (AMR) environment, the data exchanged between nodes varies in scale, and physical consistency synchronization needs to be achieved through non-blocking communication.

[0081] Identify the boundaries between each sub-computational domain. Node A is responsible for a 125-meter refined grid, and node B is responsible for a 250-meter transition grid. The system extracts boundary data blocks with a width of 3 grid layers from the edge of node A, covering the main body, temperature, and salinity distribution functions.

[0082] The cluster utilizes MPI asynchronous communication. Node A sends approximately 15.6MB of data packets to neighboring nodes while simultaneously receiving feedback from the external domain. This mode allows computation and communication to overlap, significantly improving forecast timeliness.

[0083] The verification and check of the boundary consistency processing results includes: performing spatial momentum conservation interpolation to verify the continuity of momentum flux between adjacent overlapping grid layers; calculating the absolute difference between the local momentum flux and the remote interpolated momentum flux at the boundary of adjacent sub-computation domains; if the absolute difference is lower than a preset tolerance threshold, determining that the data synchronization and exchange across resolution boundaries are continuous; if the absolute difference is not lower than the preset tolerance threshold, triggering boundary data retransmission or adjusting the interpolation weights. The system performs a weighted average of the overlapping points of A and B. The dimensionless temperature at a certain point at the boundary is 0.932 in A, 0.928 in B, and 0.930 in the system fusion.

[0084] The entire sea area data range is checked to ensure that the dimensionless salinity is within a reasonable range (0.9~1.1). After verification, the master control node generates a synchronization completion flag and integrates the fragmented data in the 10,000-core memory into a logically unified dimensionless state matrix for the entire sea area.

[0085] 205. Convert the state data of the entire sea area into a marine environmental forecast field with physical dimensions.

[0086] Specifically, the system receives state data for the entire sea area; uses pre-determined characteristic scales to perform dimensional reduction calculations on the state data for the entire sea area, generating sea flow velocity field, temperature field, and salinity field data with physical units; organizes the flow velocity field, temperature field, and salinity field data with physical units according to a preset spatial grid structure and forecast time series to generate structured marine environmental element forecast field data; and encapsulates the structured marine environmental element forecast field data into a forecast data file conforming to a standard format to form a marine environmental forecast field.

[0087] It should be noted that the dimensionless numerical matrix is ​​mapped back to the physical world to generate a deliverable forecast field.

[0088] Perform moment calculations across the entire grid. Taking the center point of the cyclone eyewall as an example, the system accumulates the particle distribution function to obtain the latest dimensionless state: density 1.0232, northwest velocity component 0.012, temperature 0.93, salinity 0.9949.

[0089] Restore the feature reference value from step 201: 1. Flow rate restoration: m / s (northwest direction).

[0090] 2. Temperature reduction: ℃ (consistent with the cooling caused by cold suction).

[0091] 3. Salinity reduction: PSU.

[0092] 4. Density reduction: kg / m 3 .

[0093] The structured forecast elements are shown in Table 2 (T+24h) below: Table 2 Surface flow velocity 0.0120 100 1.2m / s Maximum velocity zone on the periphery of the cyclone Seawater temperature 0.9300 30 27.9℃ Cyclone passage causes temperature drop due to updraft seawater salinity 0.9949 35 34.82 PSU Upwelling carries deep, high-salinity water. Seawater density 1.0232 1000 <![CDATA[1023.2kg / m 3 ]]> Temperature drop and salt rise together lead to increased density Encapsulating the fully restored grid data into a standard NetCDF file, including the three-dimensional spatiotemporal distribution of the flow field and temperature-salinity field, marks the completion of the forecast calculation task.

[0094] 206. Obtain measured observation data of the target sea area that matches the forecast period; This process involves spatiotemporal matching and comparative analysis of marine environmental forecast fields and measured observation data to generate forecast error assessment data. Based on this data, the certainty index and uncertainty range of the forecast results are calculated, generating quantitative forecast products with credibility information. Then, based on these products and pre-defined marine application scenario decision-making rules, auxiliary decision-making information for specific maritime activities is generated. Further, the measured observation data undergoes quality control and format standardization to generate a standardized validation dataset. The marine environmental forecast field is aligned and interpolated with the standardized validation dataset in both temporal and spatial dimensions to generate a set of forecast-observation data pairs with a common spatiotemporal reference. Based on this set, the absolute and relative deviations between forecast and observed values ​​are calculated for each grid point, generating a spatial distribution field of forecast errors. Regional statistical analysis of this spatial distribution field is performed to calculate the average level and dispersion of forecast errors in different sea areas, generating a regional error statistics table. Finally, the spatial distribution field and the regional error statistics table are integrated to form the forecast error assessment data.

[0095] It should be noted that this step is the quality control and application stage of the forecast results. The system acquires the measured data from Argo buoys in the western Pacific corresponding to the forecast period. For the aforementioned cyclone eyewall point, the measured values ​​are: flow velocity 1.15 m / s, temperature 28.1℃, and salinity 34.78 PSU.

[0096] Three-dimensional linear interpolation was performed to align the predicted grid points with the observed coordinates. The calculations yielded the following results: absolute deviation of flow velocity 0.05 m / s, temperature deviation 0.2℃, and salinity deviation 0.04 PSU.

[0097] The multi-dimensional error assessment table 3 is as follows: Table 3: Surface flow velocity 1.20m / s 1.15m / s 0.05m / s 4.3% High (meets business requirements) Surface temperature 27.9℃ 28.1℃ 0.2℃ 0.7% Extremely high (consistent with the cooling trend) Surface salinity 34.82 PSU 34.78 PSU 0.04 PSU 0.1% Extremely high Based on the assessment results showing deviations all within 5%, the forecast product was determined to have high reliability. Combined with the safety manual for oil and gas platforms in the target sea area, the system triggered decision recommendations: Warning: The ocean currents caused by the cyclone have accelerated to 1.2 m / s.

[0098] Decision action: It is recommended that the platform mooring system undergo a secondary hardening measure, and all diving and underwater robot (ROV) operations be prohibited for the next 24 hours.

[0099] The entire process is now closed, achieving an effective transformation from microscopic particle simulation to macroscopic maritime decision-making.

[0100] The above describes the marine environment forecasting method based on high-performance computing in the embodiments of the present invention. The following describes the marine environment forecasting system based on high-performance computing in the embodiments of the present invention. Please refer to [link / reference]. Figure 4 One embodiment of the marine environment forecasting system based on high-performance computing in this invention includes: a conversion module 301, used to convert physical oceanographic observation data of a target sea area into initial field data; an initialization module 302, used to initialize a set of particle distribution functions based on the initial field data; a parallel module 303, used to perform evolutionary calculations based on the set of particle distribution functions in a parallel environment of a high-performance computing cluster to update the ocean state, wherein the evolutionary calculations couple collision operators and continuous transmission processes for the ocean environment and dynamically generate adaptive grids based on flow field characteristics; a synchronization module 304, used to globally synchronize the updated ocean state data distributed across each computing node to obtain full-ocean-area state data; and a forecasting module 305, used to obtain a marine environment forecasting field based on the full-ocean-area state data.

[0101] above Figure 4 The marine environment forecasting system based on high-performance computing in this embodiment of the invention will be described in detail from the perspective of modular functional entities. The marine environment forecasting equipment based on high-performance computing in this embodiment of the invention will be described in detail from the perspective of hardware processing.

[0102] Figure 5This is a schematic diagram of a high-performance computing-based marine environment forecasting device 400 provided in an embodiment of the present invention. The high-performance computing-based marine environment forecasting device 400 can vary significantly due to different configurations or performance characteristics. It may include one or more central processing units (CPUs) 410 (e.g., one or more processors) and a memory 420, and one or more storage media 430 (e.g., one or more mass storage devices) for storing application programs 433 or data 432. The memory 420 and storage media 430 can be temporary or persistent storage. The program stored in the storage media 430 may include one or more modules (not shown in the diagram), each module may include a series of instruction operations on the high-performance computing-based marine environment forecasting device 400. Furthermore, the processor 410 may be configured to communicate with the storage media 430 and execute the series of instruction operations in the storage media 430 on the high-performance computing-based marine environment forecasting device 400.

[0103] The high-performance computing-based marine environmental forecasting device 400 may also include one or more power supplies 440, one or more wired or wireless network interfaces 450, one or more input / output interfaces 460, and / or one or more operating systems 431, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 5 The illustrated structure of a high-performance computing-based marine environment forecasting device does not constitute a limitation on high-performance computing-based marine environment forecasting devices. It may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0104] The present invention also provides a marine environment forecasting device based on high-performance computing. The marine environment forecasting device based on high-performance computing includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor performs the steps of the marine environment forecasting method based on high-performance computing in the above embodiments.

[0105] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the marine environment forecasting method based on high-performance computing.

[0106] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0107] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0108] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A marine environment forecasting method based on high-performance computing, characterized in that, include: Convert physical oceanographic observation data of the target sea area into initial field data; The particle distribution function set is initialized based on the initial field data; wherein, the particle distribution function set includes the initial equilibrium state set of the main distribution function, the initial set of the scalar distribution function, and the Coriolis force source term set; In a parallel environment of a high-performance computing cluster, the ocean state is updated based on the evolutionary computation of the particle distribution function set. This evolutionary computation couples a collision operator with a continuous transport process for the ocean environment and dynamically generates an adaptive mesh based on flow field characteristics, including: Based on the set of particle distribution functions, local collision calculations are performed in parallel within each sub-computation domain. The collision calculations are coupled with the rotational effect provided by the set of Coriolis force source terms and the buoyancy effect induced by the initial set of scalar distribution functions, resulting in an updated set of distribution functions. The updated distribution function set is continuously transmitted along a preset discrete velocity direction based on floating-point coordinates to generate the transmitted distribution function set. Based on the transmitted distribution function set, intermediate ocean state data including flow velocity, temperature and salinity information in each sub-computation domain are calculated. Adaptive grid control parameters are generated based on the gradient features of the intermediate ocean state data. Based on the adaptive grid control parameters, the computational grid of each sub-computation domain is dynamically refined or coarsened to generate an updated adaptive grid structure and the corresponding set of particle distribution functions and ocean state data. The updated ocean state data distributed across all computing nodes is globally synchronized to obtain ocean state data for the entire sea area. A marine environmental forecast field is obtained based on the aforementioned full-area state data.

2. The marine environment forecasting method based on high-performance computing according to claim 1, characterized in that, include: To acquire historical observational or reanalysis data of the ocean dynamic field and ocean thermohaline field in the target sea area; Based on the physical characteristics of the target sea area, determine the feature length, feature time, and feature velocity; The historical observation data or reanalysis data of the ocean dynamic field and ocean thermohaline field are processed using the characteristic length, characteristic time and characteristic velocity as characteristic scale quantities to generate a dimensionless ocean parameter field; The dimensionless ocean parameter field is interpolated and mapped onto a preset continuous computational grid to form initial field data.

3. The marine environment forecasting method based on high-performance computing according to claim 1, characterized in that, Based on the intermediate ocean state data, the vorticity value of the velocity field, the gradient value of the temperature field, and the gradient value of the salinity field are calculated respectively to generate multiphysics gradient feature data. Based on the preset grid adjustment physical criteria, the multiphysics gradient feature data is compared with the corresponding dynamic threshold to generate a preliminary grid adjustment region identification field. Based on the grid adjustment area identification field and combined with the preset grid resolution level rules, a target resolution level is determined for each identified area, and a set of grid refinement or coarsening instructions is generated. The set of mesh refinement or coarsening instructions is processed to generate adaptive mesh control parameters.

4. The marine environment forecasting method based on high-performance computing according to claim 2, characterized in that, include: Based on the updated adaptive mesh structure, the overlapping boundary regions where each sub-computation domain needs to be exchanged are determined. Data corresponding to the overlapping boundary region is extracted from the ocean state data stored in each computing node to form boundary data blocks to be exchanged; The boundary data blocks to be exchanged are exchanged between the computing nodes through a parallel communication interface; The received boundary data blocks from adjacent sub-computation domains are processed together with the ocean state data of this sub-computation domain to generate state data for the entire sub-computation domain. The state data of all sub-computation domains on all computing nodes are collected to form the state data of the entire sea area.

5. The marine environment forecasting method based on high-performance computing according to claim 4, characterized in that, Based on the updated adaptive mesh structure, the geometric characteristics and data scale of the overlapping boundary regions of each sub-computation domain are analyzed to generate an optimized list of parallel communication tasks. According to the parallel communication task list, each computing node is scheduled to send and receive the boundary data blocks to be exchanged through non-blocking communication mode, thereby generating a set of received data blocks; The external boundary data in the received data block set is processed together with the local boundary data to generate a boundary consistency processing result; The boundary consistency processing result is verified and checked to generate a synchronization completion flag.

6. The marine environment forecasting method based on high-performance computing according to claim 5, characterized in that, The verification and check of the boundary consistency processing result includes: Perform spatial momentum conservation interpolation to verify the continuity of momentum flux between adjacent overlapping grid layers; Calculate the absolute difference between the local momentum flux and the long-range interpolated momentum flux at the boundary of adjacent sub-computation domains; If the absolute difference is lower than a preset tolerance threshold, it is determined that data synchronization and exchange across resolution boundaries are continuous. If the absolute difference is not lower than the preset tolerance threshold, then boundary data retransmission or interpolation weight adjustment is triggered.

7. The marine environment forecasting method based on high-performance computing according to claim 4, characterized in that, include: Using predetermined characteristic scales, dimensional reduction calculations are performed on the state data of the entire sea area to generate sea flow velocity field, temperature field and salinity field data; The flow velocity field, temperature field, and salinity field data are organized according to a preset spatial grid structure and forecast time series to generate marine environmental element forecast field data. The marine environmental element forecast field data is encapsulated and written into a forecast data file to form a marine environmental forecast field.

8. The marine environment forecasting method based on high-performance computing according to any one of claims 1-7, characterized in that, Also includes: Acquire measured observation data of the target sea area that matches the forecast period; The marine environment forecast field is matched and compared with the measured observation data to generate forecast error assessment data; Based on the forecast error assessment data, the certainty index and uncertainty range of the forecast results are calculated, and a quantitative forecast product with credibility information is generated. Based on quantitative forecast products with credibility information, and combined with pre-set decision-making rules for marine application scenarios, auxiliary decision-making information is generated.

9. A marine environmental forecasting system based on high-performance computing, characterized in that, The marine environment forecasting system based on high-performance computing includes: The conversion module is used to convert physical oceanographic observation data of the target sea area into initial field data; An initialization module is used to initialize a set of particle distribution functions based on the initial field data; wherein, the set of particle distribution functions includes an initial equilibrium state set of the main distribution function, an initial set of scalar distribution functions, and a set of Coriolis force source terms; A parallel module is used to update the ocean state based on the particle distribution function set evolution calculation in a parallel environment of a high-performance computing cluster. This evolution calculation couples a collision operator with a continuous transport process for the ocean environment and dynamically generates an adaptive mesh based on flow field characteristics, including: Based on the set of particle distribution functions, local collision calculations are performed in parallel within each sub-computation domain. The collision calculations are coupled with the rotational effect provided by the set of Coriolis force source terms and the buoyancy effect induced by the initial set of scalar distribution functions, resulting in an updated set of distribution functions. The updated distribution function set is continuously transmitted along a preset discrete velocity direction based on floating-point coordinates to generate the transmitted distribution function set. Based on the transmitted distribution function set, intermediate ocean state data including flow velocity, temperature and salinity information in each sub-computation domain are calculated. Adaptive grid control parameters are generated based on the gradient features of the intermediate ocean state data. Based on the adaptive grid control parameters, the computational grid of each sub-computation domain is dynamically refined or coarsened to generate an updated adaptive grid structure and the corresponding set of particle distribution functions and ocean state data. The synchronization module is used to globally synchronize the updated ocean state data distributed across various computing nodes to obtain ocean state data for the entire sea area. The forecast module is used to obtain a marine environmental forecast field based on the full-area state data.

Images