Search

CN-122023701-A - Multi-core parallel particle tracking acceleration method for large-range ocean drift target

CN122023701ACN 122023701 ACN122023701 ACN 122023701ACN-122023701-A

Abstract

The invention belongs to the technical field of offshore target drift tracking, and particularly discloses a multi-core parallel particle tracking acceleration method for a large-range ocean drift target. The method comprises the steps of firstly constructing a marine target drift prediction model, obtaining satellite remote sensing images of a prediction area to obtain initial distribution positions of the marine target, setting model calculation areas, particle numbers and prediction duration model parameters, obtaining model driving fields including wind fields, flow fields and the like, performing parallel setting on a multi-core processor, including calculating the number of CPUs and the physical core numbers thereof, setting the running line numbers, dividing tasks, creating a thread pool once, setting a scheduling strategy, initializing a particle state, executing time propulsion and drift numerical simulation processes, performing parallel tracking calculation on large-scale drift particles, and finally predicting to obtain a particle track set. The method of the invention realizes high-fidelity and fine drift prediction within minute-level time limit.

Inventors

  • XU JIANGLING
  • WU LINGJUAN
  • WU CHANGMAO
  • CAO RUICHEN
  • HUANG JUAN
  • SONG YAN

Assignees

  • 自然资源部北海预报减灾中心(自然资源部青岛海洋中心)

Dates

Publication Date
20260512
Application Date
20260410

Claims (10)

  1. 1. The multi-core parallel particle tracking acceleration method for the large-range ocean drift target is characterized by comprising the following steps of: Step 1, constructing an offshore target drift prediction model by considering the action of wind, waves and current on an offshore target based on a Lagrangian particle tracking method; step 2, acquiring a satellite remote sensing image of a forecast area, and acquiring an initial distribution position of an offshore target object based on the satellite remote sensing image; step 3, setting model parameters including a model calculation area, the number of particles, a prediction duration and a time step; step 4, obtaining a model driving field including a stokes drift caused by interaction of a wind field, a flow field and a wave flow; Step 5, performing parallel setting on the multi-core processor, wherein the parallel setting comprises the steps of calculating the number of the CPUs and the physical core number thereof, setting the running thread number, dividing tasks, creating a thread pool at one time and setting a scheduling strategy; and step 6, initializing the particle state, then executing a time propulsion and drift numerical simulation process, carrying out parallel tracking calculation on large-scale drift particles, and finally predicting to obtain all particle track sets.
  2. 2. The multi-core parallel particle tracking acceleration method for large-range ocean drift targets according to claim 1, wherein in the step 1, the particle tracking method adopts a random walk mode of particles to simulate the movement of the particles, the displacement variable of each particle is determined by a lagrangian equation, and an offshore target drift prediction model is constructed, as shown in formula (1): (1) Wherein the method comprises the steps of After the displacement of the single particle in unit time is determined for the displacement of the single particle, the unit displacement is subjected to time integration to realize dynamic particle integration; Is the wind action coefficient; is surface ocean current; is wind at the height of the sea surface 10m, A stokes drift caused by wave-current interactions; A diffusion coefficient for an individual particle; is an independent random number.
  3. 3. The multi-core parallel particle tracking acceleration method for large-range ocean drift targets according to claim 1, wherein in the step 2, the process of obtaining the initial distribution position of the ocean targets is as follows: Firstly, after a satellite remote sensing image is obtained, carrying out radiation calibration and atmospheric correction pretreatment on the satellite remote sensing image to obtain an atmospheric bottom layer emissivity product; And then, performing visual interpretation on the calculated image, selecting a boundary value of a pixel which can be judged as an offshore target object, and performing offshore target object related extraction by taking a detection index of the boundary value as a threshold value to obtain an offshore target object initial distribution position.
  4. 4. The multi-core parallel particle tracking acceleration method for large-range ocean drift targets according to claim 1, wherein in the step 3, specific setting of model parameters is as follows: I. The space range of the model calculation area is larger than the range where the marine target object is monitored, and the sea area where all the marine target objects potentially drift during the drift numerical simulation; II, predicting time length, namely setting the predicted time length to be more than 72 hours; III, counting the number of particle points according to the initial distribution position of the offshore target object to obtain the total particle number ; IV. time step size setting The value range is set as 。
  5. 5. The multi-core parallel particle tracking acceleration method for large-range ocean drift targets according to claim 1, wherein in the step 4, the process of obtaining the model driving field is as follows: acquiring hydrological meteorological data of surface sea current, sea surface wind and stokes drift; Wherein the surface layer flows And the wind at the height of the sea surface 10 m From measured data of sea area where marine target is located or from numerical simulation results based on ocean and meteorological models, stokes drift The method comprises the steps of calculating through a marine numerical model; The data duration of all the model driving fields is not shorter than the drift numerical simulation duration, i.e. the predicted duration.
  6. 6. The multi-core parallel particle tracking acceleration method for large-range ocean drift targets according to claim 1, wherein in the step 5, the parallel setting steps are as follows: Step 5.1, firstly detecting the number of CPU of an operation node, and recording the physical core number of the CPU; Step 5.2, if the user sets the number of threads participating in calculation, the running thread number is set to be the user-specified number, otherwise, the running thread number is set to be the physical core number of the running node; step 5.3, if the particle block size is specified by the user, setting according to the specification of the user, otherwise adopting default equipartition setting; Step 5.4, creating a thread pool at one time for the calculation task call during the whole drift numerical simulation period; and 5.5, if the scheduling strategy is specified by the user, setting the scheduling strategy according to the user, otherwise defaulting to the static scheduling strategy.
  7. 7. The method for multi-core parallel particle tracking acceleration for large-scale ocean drift targets according to claim 6, wherein in step 5.5, the user scheduling policy includes a static scheduling policy and a dynamic scheduling policy.
  8. 8. The method for multi-core parallel particle tracking acceleration for large-scale ocean drift targets according to claim 1, wherein in step 6, the particle state is initialized, i.e. given particle At the position of The longitude and latitude information of moment, and a model driving field for drift numerical simulation initial moment is provided at the same time, wherein the model driving field comprises a stokes drift caused by interaction of a wind field, a flow field and a wave flow; Wherein, the , As the total number of particles, The start time of the drift numerical simulation time period is indicated.
  9. 9. The multi-core parallel particle tracking acceleration method for large-scale ocean drift targets of claim 8, wherein in the step 6, the time advance and drift numerical simulation process is as follows: initializing particle states Acquiring a model driving field and setting a time step ; Step 6.1. Main circulation; For a pair of Step by step Advancing, wherein The drift numerical simulation time length is represented, namely, the starting time of the prediction time length; the end time of the drift numerical simulation duration; step 6.2, field interpolation is carried out, and a physical driving field is calculated; for each particle position Obtaining flow velocity, wind vector and wave flow interaction item by bilinear/trilinear interpolation, wherein the flow velocity is The wind vector is Wave-stream interaction is ; Step 6.3, executing multi-thread parallel particle updating; I. advection: ; II, wind and wave drift additional terms: ; Wherein, the Is the wind action coefficient; and III, random diffusion: From a normal distribution array based on isotropic or anisotropic diffusion tensor Mid-sampling diffuse noise ; ; Shoreline/boundary: If particles Hit shoreline/boundary location, then execute stay, adsorb or absorb strategy; step 6.4. Migration across subfields: adding particles into a migration buffer area of a corresponding tile according to the positions of the particles, and moving the particles to a task queue of a target tile once after local aggregation in a thread; Step 6.5, field missing treatment; If the model driving field is in a lack of detection, performing time extrapolation or neighborhood spatial interpolation on related variables, wherein the related variables refer to stokes drifting caused by interaction of wind fields, flow fields or wave flows; step 6.6, if the writing time point is reached, writing the track data of the particles predicted by each thread in the current time step into an output file in batches, and compressing the track data according to the particle blocks or the tile batches so as to reduce the writing jitter; Step 6.7, if the check point interval is reached, periodically storing the particle state; step 6.8. When the drift numerical simulation is complete, i.e. when When the particle trajectory set and the checkpoint data are output.
  10. 10. The multi-core parallel particle tracking acceleration method for a large-scale ocean drift target according to claim 1, wherein the offshore target comprises spilled oil, floating dangerous goods, or enteromorpha green tide.

Description

Multi-core parallel particle tracking acceleration method for large-range ocean drift target Technical Field The invention belongs to the technical field of offshore target drift tracking, and particularly relates to a multi-core parallel particle tracking acceleration method for a large-range ocean drift target, which is suitable for a scene of parallel tracking acceleration calculation of large-scale drift particles. Background The offshore oil spill event and the enteromorpha green tide disasters have the characteristics of wide source distribution, high sensitivity of a drift path to wind-wave-flow coupling, and difficult reversion of economic and ecological losses caused once logging in. In the actual emergency treatment process, the drift position, the influence area and the like of the target in the future of 3 to 7 days need to be mastered in time, and a treatment scheme is made in time. The existing target drift prediction technology generally adopts a single-line Cheng Lizi tracking model based on a Lagrangian frame to predict the track of an offshore drift target (such as an offshore target such as oil spill, floating dangerous goods, green tide of enteromorpha and the like). The model takes three-dimensional ocean currents, wind fields and wave fields as driving, and solves a particle motion equation through numerical integration. Because such a model needs to simulate the trajectories of thousands of particles at 168h (i.e. 7 days) or even longer, the traditional single-thread calculation is time-consuming and cannot meet the aging requirements of emergency treatment of offshore green tide disasters, pollution and the like. To output results in the "golden emergency window", existing target drift prediction systems typically sacrifice spatial resolution or particle size, resulting in larger drift position prediction errors for scattered particle points. To sum up, the current method for realizing the target drift prediction by using a single thread has the following defects: 1. The calculation efficiency is limited, the existing method generally adopts a single line Cheng Chuanhang to push the particle track, so that the utilization rate of the multi-core CPU resources is insufficient (typically lower than 10%), the large-scale long-time-effect simulation is difficult to support, and the expansibility is severely limited. 2. Parallel synchronization is inefficient, and in order to ensure data consistency, a single-threaded architecture often depends on a global synchronization mechanism, so that idle waiting occurs in the calculation process, and the inter-region migration delay of particles is obvious, so that the overall timeliness is affected. 3. And the memory access cost is high, and the access mode is discontinuous due to unreasonable layout of the particle data structure, so that the Cache miss rate exceeds 30%, and the memory bandwidth utilization efficiency and the CPU pipeline performance are obviously reduced. 4. The load balancing performance is insufficient, namely, under a large-range complex flow field, the single-thread mode cannot dynamically adapt to the characteristic of uneven particle distribution, and part of processing cores are in an idle state for a long time, so that the overall calculation load is seriously unbalanced. Disclosure of Invention The invention aims to provide a multi-core parallel particle tracking acceleration method for a large-range ocean drift target, which utilizes a multi-core computing platform to perform parallel tracking computation on the large-scale drift particles so as to complete drift prediction within a limited time limit. In order to achieve the above purpose, the invention adopts the following technical scheme: The multi-core parallel particle tracking acceleration method for the large-range ocean drift target comprises the following steps of: Step 1, constructing an offshore target drift prediction model by considering the action of wind, waves and current on an offshore target based on a Lagrangian particle tracking method; step 2, acquiring a satellite remote sensing image of a forecast area, and acquiring an initial distribution position of an offshore target object based on the satellite remote sensing image; step 3, setting model parameters including a model calculation area, the number of particles, a prediction duration and a time step; step 4, obtaining a model driving field including a stokes drift caused by interaction of a wind field, a flow field and a wave flow; Step 5, performing parallel setting on the multi-core processor, wherein the parallel setting comprises the steps of calculating the number of the CPUs and the physical core number thereof, setting the running thread number, dividing tasks, creating a thread pool at one time and setting a scheduling strategy; and step 6, initializing the particle state, then executing a time propulsion and drift numerical simulation process, carrying out parallel tracking calculatio