Ocean forecasting method and device and computing equipment cluster

Abstract

The ocean forecasting method comprises the steps that first ocean environment data and a first ocean boundary in a first sea area are obtained, and the first ocean environment data and the first ocean boundary are both historical data; iteratively processing the fragmented data through a neural network to predict future marine environment data and future marine boundaries; wherein in different rounds of iteration processes, the fragmented data is obtained by segmenting the first marine environment data and the first marine boundary based on different input block sizes, in the first round of iteration process, the first input block size is the initial size, and in any round of iteration process except the first round of iteration, the first input block size is the initial size; the size of the first input block is obtained by adjusting the initial size based on the future ocean boundary obtained by the previous round of iteration; and outputting future marine environment data obtained in the current round of iteration under the condition that the continuity of the future marine boundary obtained in the current round of iteration meets a requirement or the number of iterations reaches a threshold value. The method can improve the ocean forecasting precision.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence (AI) technology, and in particular to a marine forecasting method, apparatus and computing equipment cluster. Background Technology

[0002] With the increasing depletion of land resources, countries are gradually turning their attention to the ocean in search of new resources. This has not only accelerated the progress of marine research but also made ocean forecasting increasingly important. Ocean forecasting is based on the past and present state and evolution patterns of the ocean, combined with numerical models, observational data, and data assimilation methods, to predict ocean phenomena and conditions (seawater temperature and salinity, ocean currents, tides, waves, and sea ice, etc.) at different spatiotemporal scales. Although numerical simulation methods have made ocean forecasting possible and brought commercial value to the marine field, they require a large amount of computing resources for high-performance computing. As the data resolution increases linearly, the required computational cost increases exponentially. Conversely, using low-resolution data makes it difficult to learn the characteristics of mesoscale eddies in the ocean, resulting in poor accuracy of the prediction results. Summary of the Invention

[0003] This application provides a marine forecasting method, apparatus, computing device cluster, computer storage medium, and computer product that can improve the accuracy of marine forecasting results.

[0004] In a first aspect, this application provides a marine forecasting method, comprising: acquiring first marine environmental data and a first marine boundary in a first sea area, wherein both the first marine environmental data and the first marine boundary are historical data; processing piecewise data iteratively through a neural network to predict future marine environmental data and future marine boundaries; wherein, in different iterations, the piecewise data is obtained by dividing the first marine environmental data and the first marine boundary based on different input block sizes, and in the first iteration, the first input block size is the initial size, and in any iteration other than the first iteration, the first input block size is adjusted based on the future marine boundary obtained in the previous iteration; and outputting the future marine environmental data obtained in the current iteration when the continuity of the future marine boundary obtained in the current iteration meets the requirements or the number of iterations reaches a threshold.

[0005] In this way, by introducing information about ocean boundaries as a basis for adaptive patch size adjustment during ocean forecasting, the neural network can learn features of different granularities, thereby ensuring that the neural network can learn the mesoscale eddy features in the ocean and thus improve the accuracy of ocean forecasting.

[0006] In one possible implementation, before iteratively processing the fragmented data through a neural network, the method further includes: mapping a first variable from the first marine environmental data to a spatiotemporal coordinate system, wherein the spatial coordinates of the spatiotemporal coordinate system are divided into grids, and at least one value of the first variable is filled in a grid, wherein the first variable is any variable in the first marine environmental data; constructing a spatiotemporal neighborhood centered on the first grid in the spatiotemporal coordinate system, wherein the first grid is a grid in the spatiotemporal coordinate system that is not filled with data; calculating the value to be filled in the first grid based on the values ​​filled in the grids contained in the spatiotemporal neighborhood, and filling the first grid. In this way, the spatiotemporally sparse marine environmental data can be transformed into dense data, thereby improving the quality of inference data and thus improving the accuracy of marine forecasts.

[0007] In one possible implementation, the temporal neighborhood radius and spatial neighborhood radius of the spatiotemporal neighborhood are determined based on the density of effective grid points in the spatiotemporal coordinate system, where effective grid points are grids filled with the values ​​of the first variable.

[0008] In one possible implementation, the fragmented data is processed iteratively using a neural network. This includes: transforming the fragmented data from the time domain to the frequency domain in any iteration, extracting features to obtain frequency domain features, and then using an attention mechanism to process the frequency domain features to predict future marine environmental data and future ocean boundaries. In this way, when processing fragmented data through a neural network, the temporal characteristics can be better learned, thereby improving the accuracy of ocean forecasting.

[0009] In one possible implementation, the neural network is built based on a transformer architecture. During the attention calculation process in the neural network, the query vector matrix is ​​obtained from slices of data related to the first marine environmental data, while the key vector matrix and value vector matrix are obtained from slices of data related to the first marine boundary. This allows the marine boundary to be integrated into the marine environmental data, improving the accuracy of marine forecasting.

[0010] In one possible implementation, the first marine environmental data includes: sea level, and one or more of temperature, salinity, zonal current velocity, and meridional current velocity at different sea level heights.

[0011] Secondly, this application provides a marine forecasting device, comprising: an acquisition module and a processing module. The acquisition module is used to acquire first marine environmental data and a first marine boundary in a first sea area, wherein both the first marine environmental data and the first marine boundary are historical data. The processing module is used to iteratively process the fragmented data through a neural network to predict future marine environmental data and future marine boundaries; wherein, in different iterations, the fragmented data is obtained by dividing the first marine environmental data and the first marine boundary based on different input block sizes, and in the first iteration, the first input block size is the initial size; in any iteration other than the first iteration, the first input block size is adjusted based on the future marine boundary obtained in the previous iteration. The processing module is also used to output the future marine environmental data obtained in the current iteration when the continuity of the future marine boundary obtained in the current iteration meets the requirements or the number of iterations reaches a threshold.

[0012] In one possible implementation, before iteratively processing the fragmented data through a neural network, the processing module is further configured to: map a first variable in the first marine environmental data to a spatiotemporal coordinate system, wherein the spatial coordinates of the spatiotemporal coordinate system are divided into grids, at least one value of the first variable is filled in a grid, and the first variable is any variable in the first marine environmental data; construct a spatiotemporal neighborhood centered on the first grid in the spatiotemporal coordinate system, wherein the first grid is a grid in the spatiotemporal coordinate system that is not filled with data; calculate the value to be filled in the first grid based on the value filled in the grid contained in the spatiotemporal neighborhood, and fill the first grid.

[0013] In one possible implementation, the temporal neighborhood radius and spatial neighborhood radius of the spatiotemporal neighborhood are determined based on the density of effective grid points in the spatiotemporal coordinate system, where effective grid points are grids filled with the values ​​of the first variable.

[0014] In one possible implementation, when the processing module iteratively processes the fragmented data through a neural network, it specifically performs the following: in any iteration, it transforms the fragmented data from the time domain to the frequency domain and extracts features to obtain frequency domain features; and it uses an attention mechanism to process the frequency domain features to predict future marine environmental data and future marine boundaries.

[0015] In one possible implementation, the neural network is built based on a transformer architecture. During the attention calculation process in the neural network, the query vector matrix is ​​obtained from slices of data related to the first ocean environment data, while the key vector matrix and value vector matrix are obtained from slices of data related to the first ocean boundary.

[0016] In one possible implementation, the first marine environmental data includes: sea level, and one or more of temperature, salinity, zonal current velocity, and meridional current velocity at different sea level heights.

[0017] Thirdly, this application provides a computing device cluster, including at least one computing device, each computing device including a processor and a memory; the processor of the at least one computing device is used to execute instructions stored in the memory of the at least one computing device, so that the computing device cluster performs the method described in the first aspect or any possible implementation of the first aspect.

[0018] Fourthly, this application provides a computer-readable storage medium including computer program instructions, which, when executed by a computing device, perform the method described in the first aspect or any possible implementation thereof; or, when executed by a cluster of computing devices, the cluster of computing devices performs the method described in the first aspect or any possible implementation thereof. Exemplarily, the cluster of computing devices may include one or more computing devices.

[0019] Fifthly, this application provides a computer program product containing instructions that, when executed by a computing device, cause the computing device to perform the method described in the first aspect or any possible implementation thereof; or, when executed by a cluster of computing devices, cause the cluster of computing devices to perform the method described in the first aspect or any possible implementation thereof. Exemplarily, a cluster of computing devices may include one or more computing devices.

[0020] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the architecture of a marine forecasting system provided in an embodiment of this application;

[0022] Figure 2 This is a schematic diagram of a grid divided in a three-dimensional spatiotemporal coordinate system, and a spatiotemporal neighborhood determined by the grid to be filled, provided in an embodiment of this application.

[0023] Figure 3 Yes Figure 1 The diagram shown illustrates the training process of the neural operator prediction module.

[0024] Figure 4 This is a schematic diagram illustrating the interaction between a tenant and a cloud computing platform, provided in an embodiment of this application.

[0025] Figure 5This is a schematic flowchart of a marine forecasting method provided in an embodiment of this application;

[0026] Figure 6 This is a schematic diagram illustrating a step of populating ocean data according to an embodiment of this application;

[0027] Figure 7 This is a schematic diagram of the structure of a marine forecasting device provided in an embodiment of this application;

[0028] Figure 8 This is a schematic diagram of the structure of a computing device provided in an embodiment of this application;

[0029] Figure 9 This is a schematic diagram of the structure of a computing device cluster provided in an embodiment of this application;

[0030] Figure 10 This is a schematic diagram of another computing device cluster structure provided in an embodiment of this application. Detailed Implementation

[0031] In this article, the term "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. The symbol " / " in this article indicates that the related objects are in an "or" relationship; for example, A / B means A or B.

[0032] The terms "first" and "second," etc., used in the specification and claims herein are used to distinguish different objects, not to describe a specific order of objects. For example, "first response message" and "second response message," etc., are used to distinguish different response messages, not to describe a specific order of response messages.

[0033] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0034] In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more, for example, multiple processing units means two or more processing units, multiple elements means two or more elements, etc.

[0035] First, the relevant technical terms involved in the technical solution provided in this application will be introduced.

[0036] (1) Input block size

[0037] Patch size is an important hyperparameter in computer vision used to train neural networks. It represents the size of a unit image in the input model, that is, how many small image patches a complete image is divided into for training. It determines the level of detail in which the model learns image features: the finer the patch size, the more fine-grained the model can learn fine-grained features, but this may result in weak global feature learning ability; the larger the patch size, the better the model can learn macroscopic image features, but the ability to learn fine-grained features is reduced.

[0038] (2) Mesoscale eddies

[0039] Mesoscale eddies are oceanic currents present in all oceans; it's rare to find areas entirely devoid of them. They significantly influence the physical properties of the ocean. They can cause changes in sea surface temperature, affecting the temperature distribution of seawater through the upwelling and downwelling of cyclonic and anticyclonic eddies, thus forming cold and warm eddies. Furthermore, their intense stirring action can create "cold filaments" and "warm filaments" of sea surface temperature, affecting ocean heat flux. In terms of the chemical and biological environment, they carry nutrients to the euphotic layer through upwelling, promoting increased primary productivity in the ocean. Simultaneously, their stirring action and particle transport effects also influence the marine chemical and biological environment. The formation mechanism of mesoscale eddies is complex, possibly related to factors such as monsoons, topographic interactions, and wind stress curl. Their role in the ocean extends beyond local areas, potentially adjusting overall ocean circulation and serving as a crucial link in the transport and mixing of ocean matter and energy. Therefore, mesoscale eddies are an important characteristic for describing ocean behavior. Because mesoscale eddies vary in thickness and size, rotate in different directions (left or right), and have varying seawater temperatures at their centers, their study is quite challenging.

[0040] (3) Marine environmental data

[0041] Marine environmental data refers to datasets describing the physical, chemical, and biological characteristics of the ocean, obtained through satellite observation, site observation, simulation, model inference, or analysis of publicly available datasets. These data reflect the ocean's historical state, current state (i.e., present condition), future state, historical changes, or future changes. Marine environmental data can be global, covering the entire Earth's ocean regions; regional, focusing on specific sea areas or regions, such as a particular ocean or basin, or specific ecosystems like coastal areas, estuaries, or coral reefs; or local, such as data from ports, nearshore areas, or specific marine protected areas. For example, variables in marine environmental data may include sea level, and one or more of the following at different sea level heights: temperature, salinity, zonal current velocity, and meridional current velocity. Furthermore, the scale of the acquired marine environmental data may vary. For example, there can be one-dimensional ocean environmental data (such as real-time wave height data measured from a buoy), two-dimensional ocean environmental data (such as a surface salinity distribution map of a sea area), and three-dimensional ocean environmental data (such as a vertical temperature profile of a sea area). For instance, to accurately predict the ocean's thermocline and mixing layer, temperature, salinity, zonal current velocity, and meridional current velocity at 15 depths (0m, 6m, 10m, 20m, 30m, 50m, 70m, 100m, 125m, 150m, 200m, 250m, 300m, 400m, 500m) can be selected.

[0042] (3) Historical marine environmental data

[0043] Historical marine environmental data refers to marine environmental data that has already been acquired, and it can be included in historical data.

[0044] Next, the technical solution provided in this application will be introduced.

[0045] For example, Figure 1 A schematic diagram of the architecture of a marine forecasting system provided in an embodiment of this application is shown. Figure 1 As shown, the ocean forecasting system 100 includes: a spatiotemporal coordinate system construction module 110, a multi-scale spatiotemporal weighted fusion module 120, a patch segmentation module 130, a neural operator prediction module 140, and a patch size adjustment module 150.

[0046] The spatiotemporal coordinate system construction module 110 is mainly used to map data of different scales in historical marine environmental data to the same spatiotemporal coordinate system for subsequent processing. The spatiotemporal coordinate system can include multi-dimensional (two-dimensional or higher) spatial coordinates and one-dimensional time coordinates. In this embodiment, a point in the spatiotemporal coordinate system can be determined by at least three coordinate values. For example, when a point in the spatiotemporal coordinate system is determined by three coordinate values, it can be denoted as (x, y, w). x and y represent spatial location, which can be represented by longitude and latitude; w represents a time point. When a point in the spatiotemporal coordinate system is determined by four coordinate values, it can be denoted as (x, y, z, w). x, y, and z represent spatial location, which can be represented by longitude, latitude, and sea surface height; w represents a time point. For example, the spatial coordinates of the spatiotemporal coordinate system can be divided into grids, so that after mapping historical marine environmental data to the spatiotemporal coordinate system, the historical marine environmental data can be converted into grid-like data. The grids in the spatiotemporal coordinate system can be, but are not limited to, based on the global ocean domain dimension, and each grid can represent a location in the ocean. After mapping historical ocean environmental data to the spatiotemporal coordinate system, the value of any variable in the historical ocean environmental data can be located within a grid, i.e., filled within a grid. In some embodiments, any variable in the historical ocean environmental data can be represented as a three-dimensional shape of C×720×1400 at each time point. Here, 720×1400 is the predicted dimension of the global ocean domain, which can guarantee a spatial resolution of 1 / 4×1 / 4°. The spatial resolution of 1 / 4×1 / 4° refers to dividing the Earth's surface into a grid, with each grid approximately 0.25 degrees longitude and 0.25 degrees latitude. Furthermore, different variables at the same time point can be stacked and stitched together in the C dimension.

[0047] The multi-scale spatiotemporal weighted fusion module 120 is mainly used to numerically fill historical marine environmental data based on a sparse spatiotemporal coordinate system to obtain historical marine environmental data in a dense spatiotemporal coordinate system. This allows for the acquisition of high-quality training or inference data. When the training data is high-quality, a high-precision driving model can be trained, which can then be used for high-quality inference. For example, for any variable in the historical marine environmental data, after projecting it onto a spatiotemporal coordinate system, gridding can be performed in dimensions other than the time dimension to convert the historical marine environmental data into grid-like data. The size of each grid can be determined according to the actual situation and is not limited here. For example, when the spatiotemporal coordinate system is a three-dimensional coordinate system, gridding can be performed along the latitude and longitude, with each grid approximately 0.25 degrees in the longitude direction and 0.25 degrees in the latitude direction. Furthermore, for any grid to be filled (i.e., a grid lacking data), a spatiotemporal neighborhood (i.e., a time domain and a spatial neighborhood) can be constructed centered on that grid. The spatiotemporal neighborhood is primarily determined by the density of effective grid points (i.e., grids with sufficient data). In other words, the temporal and spatial neighborhood radii can be determined by the density of effective grid points. Finally, a spatiotemporal differential weighted fusion calculation is performed on the data within this spatiotemporal neighborhood to obtain the required filling value for the grid to be filled. That is, when filling the grid, the temporal and spatial neighborhood radii can first be determined based on the density of effective grid points to construct a spatiotemporal neighborhood centered on the grid to be filled. Then, a spatiotemporal differential fusion method is used to fill the grid with a comprehensive value, satisfying the differentiability of the time and spatial dimensions, to obtain the updated target grid value (i.e., the value of the grid to be filled). For example, taking a three-dimensional (3D) spatiotemporal coordinate system as an example, such as... Figure 2 As shown, for a certain type of observation data, grids can be divided in the longitudinal and latitudinal directions, and a spatiotemporal neighborhood can be constructed centered on a grid to be filled at time t. The required filling value in that grid is then calculated. Figure 2 R in t Represents the radius of the time neighborhood.

[0048] For example, when the spatiotemporal coordinate system is a four-dimensional (4D) spatiotemporal coordinate system, a spatiotemporal neighborhood centered on the target grid is selected for each target grid (i.e., the grid to be filled). Its spatial neighborhood radius is R s The time neighborhood radius is R tTherefore, in the spatiotemporal coordinate system of a certain type of observation data (such as sea surface temperature or salinity), the data required to fill the i-th layer of the target grid is calculated as follows:

[0049]

[0050] N represents the spatiotemporal neighborhood of the target grid. The total number of observational data (such as sea surface temperature or salinity); Describe a spatiotemporally differentiable convex function that satisfies the following conditions in its spatiotemporal neighborhood: Minimize the weighted average of valid grid points within the inner area; (x k ,y k ,z k ,t k () represents the coordinates of the k-th observation data; Representing the spatiotemporal neighborhood The value of the k-th observation; w k express The spatiotemporal weighted interpolation. Where w k The calculation is as follows:

[0051]

[0052] It should be understood that when the spacetime coordinate system is a coordinate system with other dimensions, the coordinates (x, y, y) of the k-th observation data in "Formula 1" and "Formula 2" above will be... k ,y k ,z k ,t k Replace ) with the coordinates of the kth observation data in the new coordinate system.

[0053] The patch segmentation module 130 is mainly used to segment historical marine environmental data after data filling. In addition, the patch segmentation module 130 can also be used to segment historical marine boundaries related to historical marine environmental data.

[0054] The neural operator prediction module 140 primarily processes the segmented historical ocean environment data and historical ocean boundaries to predict future ocean environment data and future ocean boundaries. For example, the neural operator prediction module 140 can be, but is not limited to, a neural network built based on a transformer architecture. In some embodiments, to enable the neural operator prediction module 140 to better learn temporal features, it can employ a frequency domain transformer to process the data. Specifically, the neural operator prediction module 140 can first use a Fourier transform to convert the acquired input signal from the time domain to the frequency domain, that is, project the segmentation result output by the patch segmentation module 130 onto the frequency domain space; then, feature extraction is performed in the frequency domain, and attention mechanisms are used to process frequency domain features, capturing patterns and dependencies in the sequence data; finally, an inverse Fourier transform is used to convert the processed signal back from the frequency domain to the time domain to obtain the final prediction result. In some embodiments, when the neural operator prediction module 140 performs attention calculation, the query vector (Q) matrix is ​​obtained based on the segmented historical ocean environment data, and the key vector (K) matrix and value vector (V) matrix are both obtained based on the segmented historical ocean boundaries.

[0055] The patch size adjustment module 150 primarily determines the continuity of future ocean boundaries. If the continuity requirement is not met, the patch size adjustment module 150 can adjust the patch size used by the patch segmentation module 130 when segmenting the data. For example, it can increase or decrease the patch size. Then, based on the adjusted patch size, the patch segmentation module 130 can segment the historical ocean environment data and historical ocean boundaries, and input the segmented data into the neural operator prediction module 140 for processing. After processing the re-segmented data, the neural operator prediction module 140 can output its latest predicted future ocean environment data and future ocean boundaries. Next, the patch size adjustment module 150 can determine the continuity of the latest obtained future ocean boundaries. This iterative process continues until the continuity of the latest obtained future ocean boundaries meets the requirement. If the continuity requirement is met, the patch size adjustment module 150 can output the latest obtained future ocean environment data and / or future ocean boundaries. For example, future ocean boundary continuity means that the future ocean boundary is a continuous boundary (without breakpoints). The continuity of the future ocean boundary is calculated as the break length / (break length + boundary length). In some embodiments, the patch size adjustment module 150 can calculate the total distance of the break portions in the future ocean boundary. When the total break distance is larger, it indicates that the prediction quality is poor and the patch size needs to be adjusted.

[0056] It should be noted that the patch size adjustment module 150 can be added or not depending on the actual situation, and there is no limitation here. However, considering the highly nonlinear nature and complex boundaries of ocean models, the model needs sufficient generalization ability; and the ocean, as a complex fluid, has the characteristic of mesoscale eddies, which means the model needs to be able to resolve mesoscale eddies to ensure the accuracy of global-scale simulations. Therefore, it is best to add the patch size adjustment module 150 so that the neural operator prediction module 140 can learn the mesoscale eddy feature by continuously adjusting the patch size, thereby improving the accuracy of predictions. When the patch size adjustment module 150 is not added to the ocean forecasting system 100, the data input to the neural operator prediction module 140 may not include historical ocean boundaries related to historical ocean environmental data.

[0057] As can be seen from the above description of the marine forecasting system 100 provided in the embodiments of this application, after acquiring multi-scale marine environmental data, this system can project these data onto the same spatiotemporal coordinate system and fill in sparse marine environmental data to obtain dense marine environmental data. This solves the problem of scarce marine data and the lack of high-quality datasets due to multi-scale data. Furthermore, by introducing marine boundaries related to the marine environmental data as a basis for adaptive patch size adjustment, the model can learn mesoscale eddy characteristics, improving the model's prediction accuracy.

[0058] The following is about Figure 1 The training process of the neural operator prediction module 140 in the shown ocean forecasting system 100 is described. For example... Figure 3 As shown, the historical ocean environment data in the sample data is first padded to make it denser. Then, the historical ocean boundary and dense historical ocean environment data are used as input, and this data is segmented based on the initial patch size. Next, the segmented historical ocean boundary patches are randomly masked, and all the segmented data are projected into the frequency domain space, and the prediction results (i.e., future ocean environment data and future ocean boundary) are output through the Fourier autoencoder module. Next, the continuity of the future ocean boundary is determined, and the patch size is adaptively adjusted according to the determined continuity. Specifically, when the continuity requirement is not met, the patch size is adjusted, and the input data is re-segmented with the adjusted patch size; when the continuity is met, the prediction results are output. Finally, the loss function is used to calculate the sample labels and the output prediction results, and the parameters in the neural operator prediction module 140 are adjusted with the goal of minimizing the calculated result. It should be understood that the random mask used in the training process can be selected according to the actual situation and is not limited here.

[0059] The above is a description of the marine forecasting system 100 provided in the embodiments of this application. Exemplarily, the marine forecasting system 100 described above can be configured on a cloud computing platform, for example, deployed on at least one instance such as a virtual machine or container, so that the cloud computing platform can provide marine forecasting services. Of course, the marine forecasting system 100 can also be configured on nodes other than the cloud computing platform, for example, deployed in at least one data center or on at least one server, depending on the actual situation, and is not limited here. The cloud computing platform can provide pages related to public cloud services for tenants to remotely access public cloud services. In this embodiment, tenants (also referred to as "users") can pre-purchase the marine forecasting services provided by the marine forecasting system 100 on the cloud computing platform. For ease of understanding, the interaction between the tenant and the cloud computing platform is described below. Figure 4 As shown, the interaction between the tenant and the cloud computing platform mainly includes: the tenant logs into the cloud computing platform 400 through a client webpage, selects and purchases cloud services (i.e., marine forecasting services) related to the marine forecasting system 100 within the cloud computing platform 400, and after purchase, the tenant can perform marine forecasts on the cloud computing platform 400 based on the functions provided by the marine forecasting service. The cloud computing platform 400 is primarily used to manage the infrastructure for running the marine forecasting service. For example, the infrastructure for running the marine forecasting service may include multiple data centers located in different regions, each data center including multiple servers. Data centers can provide basic resources for the marine forecasting service, such as computing resources and storage resources. Therefore, when purchasing and using the marine forecasting service, the tenant mainly pays for the resources used. When using the marine forecasting service, the tenant can input historical marine environmental data and / or historical marine boundaries through the configuration interface, application program interface (API), or other interfaces provided by the cloud computing platform 400. The cloud computing platform 200 can then perform marine forecasts according to the tenant's input.

[0060] Next, based on the above, a marine forecasting method provided by an embodiment of this application will be introduced.

[0061] For example, Figure 5 This diagram illustrates a flow chart of a marine forecasting method provided in an embodiment of this application. It is understood that this method can be executed by any device, equipment, platform, or cluster of devices with computing and processing capabilities. For example, the method can be executed by a marine forecasting device, which can be implemented by software and / or hardware, and can be configured in, but is not limited to, electronic devices or servers; typically, it can be configured on a cloud computing platform. For ease of description, a cloud computing platform will be used as the execution entity in the following description. Figure 5As shown, this ocean forecasting method may include the following steps:

[0062] S501, the cloud computing platform acquires the first marine environmental data and the first marine boundary in the first sea area, wherein the first marine environmental data and the first marine boundary are historical data.

[0063] In this embodiment, the user can upload first marine environmental data and a first marine boundary through a client associated with the cloud computing platform. The client can then transmit the first marine environmental data and the first marine boundary to the cloud computing platform. Both the first marine environmental data and the first marine boundary are historical data. Furthermore, there is a one-to-one correspondence between marine environmental data at a given point in time and marine boundary at that point in time; that is, obtaining marine environmental data at a given point in time also requires obtaining the marine boundary at that point in time. For example, the first marine environmental data includes: sea surface height, and one or more of the following at different sea surface heights: temperature, salinity, zonal current velocity, and meridional current velocity. For example, the client associated with the cloud computing platform can be a desktop application, mobile application, web application, or web-based application, etc.

[0064] S502, the cloud computing platform uses neural networks to iteratively process fragmented data to predict future marine environmental data and future marine boundaries; in different iterations, the fragmented data is obtained by dividing the first marine environmental data and the first marine boundary based on different input block sizes, and in the first iteration, the first input block size is the initial size, and in any iteration other than the first iteration, the first input block size is adjusted based on the future marine boundary obtained in the previous iteration.

[0065] In this embodiment, after acquiring the first marine environment data and the first marine boundary, the cloud computing platform can iteratively process these two types of data through a neural network to predict future marine environment data and future marine boundaries. In any iteration, the cloud computing platform segments the first marine environment data and the first marine boundary based on the patch size used in that iteration to obtain the data chunks that the neural network needs to process. The patch size differs in different iterations. Furthermore, the patch size in the first iteration can be an initial size, which can be user-configured or a system default; this is not limited here. In any iteration other than the first iteration, the first input block size is obtained by adjusting the initial size based on the future marine boundary obtained in the previous iteration. It should be understood that adjusting the patch size used in the current iteration using the predicted future marine boundary from the previous iteration allows the neural network to learn features of different granularities, thereby ensuring that the neural network can learn mesoscale eddy features in the ocean and improving the accuracy of ocean forecasting.

[0066] As one possible implementation, during any iteration, the cloud computing platform can transform the sharded data used in that iteration from the time domain to the frequency domain and extract features to obtain frequency domain features. Then, the cloud computing platform can use an attention mechanism to process these frequency domain features to predict future marine environmental data and future ocean boundaries. In this way, when processing sharded data through a neural network, the cloud computing platform can better learn temporal characteristics, thereby improving the accuracy of ocean forecasting. For example, this neural network can be built based on a transformer architecture. Specifically, during the attention calculation process in the neural network, the query vector matrix is ​​obtained based on sharded data related to the first marine environmental data, while the key vector matrix and value vector matrix are obtained based on sharded data related to the first ocean boundary, in order to integrate the ocean boundary into the marine environmental data and improve the accuracy of ocean forecasting.

[0067] S503. The cloud computing platform determines whether the continuity of the future ocean boundary obtained in the current iteration meets the requirements.

[0068] In this embodiment, during the current iteration, the cloud computing platform can determine the continuity of the future ocean boundary obtained from the current iteration. For example, the cloud computing platform can calculate the total distance of the broken parts in the future ocean boundary. The larger the total distance of the broken parts, the worse the prediction quality, meaning that the continuity of the future ocean boundary obtained from the current iteration does not meet the requirements, and in this case, S504 will be executed. If the continuity of the future ocean boundary obtained from the current iteration meets the requirements, then S505 will be executed.

[0069] S504, the cloud computing platform determines whether the number of iterations has reached the preset number.

[0070] In this embodiment, if the continuity of the future ocean boundary obtained in the current iteration does not meet the requirements, the cloud computing platform can determine whether the current iteration count has reached a preset number. If the preset number has not been reached, it returns to S502 for the next iteration. If the preset number has been reached, it executes S505.

[0071] S505, the cloud computing platform outputs future marine environmental data obtained in the current iteration.

[0072] In this embodiment, if the continuity of the future ocean boundary obtained in the current iteration meets the requirements or the number of iterations reaches a threshold, the cloud computing platform outputs the future ocean environment data obtained in the current iteration. For example, the cloud computing platform can transmit the future ocean environment data obtained in the current iteration to its associated client to display the prediction results to the user through the client. Of course, the cloud computing platform can also transmit the future ocean boundary obtained in the current iteration to the client to display it to the user through the client.

[0073] In this way, by introducing information about ocean boundaries as a basis for adaptive patch size adjustment during ocean forecasting, the neural network can learn features of different granularities, thereby ensuring that the neural network can learn the mesoscale eddy features in the ocean and thus improve the accuracy of ocean forecasting.

[0074] In some embodiments, due to the high cost of marine environmental data observation, the wide range of data sources, and the spatiotemporal sparsity (gap in both the temporal and spatial dimensions), deep-sea datasets often have significant gaps. This results in poor data quality for inference, limiting the accuracy of marine forecasting. Therefore, in this embodiment, before iteratively processing the fragmented data through a neural network, the first marine environmental data can be padded to obtain denser marine environmental data. For example, the padding step is as follows... Figure 6As shown, in S601, the first variable in the first marine environmental data is mapped to a spatiotemporal coordinate system. The spatiotemporal coordinate system is divided into grids, and at least one value of the first variable is filled in one grid. This first variable can be any variable in the first marine environmental data. In S602, a spatiotemporal neighborhood is constructed centered on the first grid in the spatiotemporal coordinate system. The first grid is a grid in the spatiotemporal coordinate system that is not filled with data. The temporal neighborhood radius and spatial neighborhood radius of the spatiotemporal neighborhood are determined based on the density of effective grid points in the spatiotemporal coordinate system. Effective grid points are grids filled with values ​​of the first variable. In S603, based on the values ​​filled in the grids included in the spatiotemporal neighborhood, the required filling value in the first grid is calculated, and the first grid is filled. For the specific processes of S601 to S603, please refer to the preceding text. Figure 1 The relevant descriptions of the spatiotemporal coordinate system construction module 110 and the multi-scale spatiotemporal weighted fusion module 120 will not be repeated here.

[0075] It is understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. In addition, the various embodiments or technical features involved in the embodiments described above can be combined according to the actual situation, and the combined solution is still within the protection scope of this application.

[0076] Based on the methods in the above embodiments, this application also provides a marine forecasting device.

[0077] For example, Figure 7 A schematic diagram of the structure of a marine forecasting device provided in an embodiment of this application is shown. Figure 7 As shown, the marine forecasting device 700 includes an acquisition module 701 and a processing module 702. The acquisition module 701 acquires first marine environmental data and a first marine boundary in a first sea area, wherein both the first marine environmental data and the first marine boundary are historical data. The processing module 702 iteratively processes the fragmented data using a neural network to predict future marine environmental data and future marine boundaries. In different iterations, the fragmented data is obtained by dividing the first marine environmental data and the first marine boundary based on different input block sizes. In the first iteration, the first input block size is the initial size; in any iteration other than the first iteration, the first input block size is adjusted based on the future marine boundary obtained in the previous iteration. The processing module 702 also outputs the future marine environmental data obtained in the current iteration when the continuity of the future marine boundary obtained in the current iteration meets the requirements or the number of iterations reaches a threshold.

[0078] In one possible implementation, before processing the fragmented data iteratively through a neural network, the processing module 702 is further configured to: map a first variable in the first marine environmental data to a spatiotemporal coordinate system, wherein the spatial coordinates of the spatiotemporal coordinate system are divided into grids, at least one value of the first variable is filled in a grid, and the first variable is any variable in the first marine environmental data; construct a spatiotemporal neighborhood centered on the first grid in the spatiotemporal coordinate system, wherein the first grid is a grid in the spatiotemporal coordinate system that is not filled with data; calculate the value to be filled in the first grid based on the value filled in the grid contained in the spatiotemporal neighborhood, and fill the first grid.

[0079] In one possible implementation, the temporal neighborhood radius and spatial neighborhood radius of the spatiotemporal neighborhood are determined based on the density of effective grid points in the spatiotemporal coordinate system, where effective grid points are grids filled with the values ​​of the first variable.

[0080] In one possible implementation, when processing the sliced ​​data iteratively through the neural network, the processing module 702 is specifically used to: transform the sliced ​​data from the time domain space to the frequency domain space in any iteration process, and perform feature extraction to obtain frequency domain features; and use an attention mechanism to process the frequency domain features to predict future marine environmental data and future marine boundaries.

[0081] In one possible implementation, the neural network is built based on a transformer architecture. During the attention calculation process in the neural network, the query vector matrix is ​​obtained from slices of data related to the first ocean environment data, while the key vector matrix and value vector matrix are obtained from slices of data related to the first ocean boundary.

[0082] In one possible implementation, the first marine environmental data includes: sea level, and one or more of temperature, salinity, zonal current velocity, and meridional current velocity at different sea level heights.

[0083] In some embodiments, Figure 7 Both the acquisition module 701 and the processing module 702 shown can be implemented in software or in hardware. For example, the implementation of the acquisition module 701 will be described below. Similarly, the implementation of the processing module 702 can refer to the implementation of the acquisition module 701.

[0084] As an example of a software functional unit, module 701 may include code running on a computing instance. The computing instance may include at least one of a physical host (computing device), a virtual machine, or a container. Further, the aforementioned computing instance may be one or more. For example, module 701 may include code running on multiple hosts / virtual machines / containers. It should be noted that the multiple hosts / virtual machines / containers used to run the code may be distributed within the same region or in different regions. Further, the multiple hosts / virtual machines / containers used to run the code may be distributed within the same availability zone (AZ) or in different AZs, each AZ including one or more geographically proximate data centers. Typically, a region may include multiple AZs.

[0085] Similarly, multiple hosts / virtual machines / containers used to run this code can be distributed within the same Virtual Private Cloud (VPC) or across multiple VPCs. Typically, a VPC is set up within a region. Communication between two VPCs within the same region, as well as between VPCs in different regions, requires a communication gateway to be set up within each VPC to enable interconnection between VPCs.

[0086] As an example of a hardware functional unit, the acquisition module 701 may include at least one computing device, such as a server. Alternatively, the acquisition module 701 may also be a device implemented using an application-specific integrated circuit (ASIC) or a programmable logic device (PLD). The PLD may be implemented using a complex programmable logical device (CPLD), a field-programmable gate array (FPGA), generic array logic (GAL), or any combination thereof.

[0087] The multiple computing devices included in the acquisition module 701 can be distributed in the same region or in different regions. Similarly, the multiple computing devices included in the acquisition module 701 can be distributed in the same Availability Zone (AZ) or in different AZs. Likewise, the multiple computing devices included in the acquisition module 701 can be distributed in the same Virtual Private Cloud (VPC) or in multiple VPCs. These multiple computing devices can be any combination of computing devices such as servers, ASICs, PLDs, CPLDs, FPGAs, and GALs.

[0088] It should be noted that, in other embodiments, the acquisition module 701 can be used to execute any step in the marine forecasting method described in the above embodiments, and the processing module 702 can be used to execute any step in the marine forecasting method described in the above embodiments. Furthermore, the steps implemented by the acquisition module 701 and the processing module 702 can be specified as needed, and different steps in the marine forecasting method described in the above embodiments can be implemented by the acquisition module 701 and the processing module 702 respectively. Figure 7 The marine forecasting device 700 shown has all the functions of the device.

[0089] This application also provides a computing device 800. For example... Figure 8 As shown, the computing device 800 includes a bus 802, a processor 804, a memory 806, and a communication interface 808. The processor 804, the memory 806, and the communication interface 808 communicate with each other via the bus 802. The computing device 800 can be a server or an electronic device. It should be understood that this application does not limit the number of processors and memories in the computing device 800.

[0090] The 802 bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be divided into address buses, data buses, control buses, etc. For ease of representation, Figure 8 The bus 804 may be represented by a single line, but this does not mean that there is only one bus or one type of bus. The bus 804 may include a path for transmitting information between various components of the computing device 800 (e.g., memory 806, processor 804, communication interface 808).

[0091] Processor 804 may include any one or more processors such as a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0092] The memory 806 may include volatile memory, such as random access memory (RAM). The processor 804 may also include non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid state drive (SSD).

[0093] The memory 806 stores executable program code, and the processor 804 executes the executable program code to implement the aforementioned functions respectively. Figure 7 The functions of the acquisition module 701 and processing module 702 shown are used to implement the ocean forecasting method described in the above embodiments. That is, the memory 806 stores instructions for executing the ocean forecasting method described in the above embodiments.

[0094] Alternatively, the memory 806 stores executable code, and the processor 804 executes the executable code to implement the aforementioned functions respectively. Figure 7 The ocean forecasting device 700 shown in the diagram performs the functions of the ocean forecasting method described in the above embodiments. That is, the memory 806 stores instructions for executing the ocean forecasting method described in the above embodiments.

[0095] The communication interface 803 uses transceiver modules, such as, but not limited to, network interface cards and transceivers, to enable communication between the computing device 800 and other devices or communication networks.

[0096] This application also provides a computing device cluster. The computing device cluster includes at least one computing device. The computing device can be a server, such as a central server, an edge server, or a local server in a local data center. In some embodiments, the computing device can also be an electronic device such as a desktop computer, a laptop computer, or a smartphone.

[0097] like Figure 9As shown, the computing device cluster includes at least one computing device 800. The memory 806 of one or more computing devices 800 in the computing device cluster may store the same instructions for executing the ocean forecasting method described in the above embodiments.

[0098] In some possible implementations, the memory 806 of one or more computing devices 800 in the computing device cluster may also store partial instructions for executing the ocean forecasting method described in the above embodiments. In other words, a combination of one or more computing devices 800 can jointly execute instructions for executing the ocean forecasting method described in the above embodiments.

[0099] It should be noted that the memory 806 in different computing devices 800 within the computing device cluster can store different instructions, each used to execute the aforementioned instructions. Figure 7 The illustrated marine forecasting device 700 exhibits some of its functions. Specifically, the instructions stored in the memory 806 of the different computing devices 800 can implement the functions of one or more modules in the acquisition module 701 and processing module 702.

[0100] In some possible implementations, one or more computing devices in a computing device cluster can be connected via a network. This network can be a wide area network (WAN) or a local area network (LAN), etc. Figure 10 One possible implementation is shown. For example... Figure 10 As shown, two computing devices 800A and 800B are connected via a network. Specifically, they are connected to the network through communication interfaces in each computing device. In this possible implementation, the memory 806 in computing device 800A stores instructions for executing the functions of the acquisition module 701. Simultaneously, the memory 806 in computing device 800B stores instructions for executing the functions of the processing module 702.

[0101] It should be understood that Figure 10 The functions of the computing device 800A shown can also be performed by multiple computing devices 800. Similarly, the functions of the computing device 800B can also be performed by multiple computing devices 800.

[0102] This application also provides another computing device cluster. The connection relationships between the computing devices in this computing device cluster can be similarly referred to... Figure 9 and Figure 10 The connection method of the computing device cluster is different in that the memory 806 of one or more computing devices 800 in the computing device cluster can store the same instructions for executing the methods in the above embodiments.

[0103] In some possible implementations, the memory 806 of one or more computing devices 800 in the computing device cluster may also store partial instructions for executing the aforementioned data processing method. In other words, a combination of one or more computing devices 800 can jointly execute the instructions for executing the aforementioned data processing method.

[0104] Based on the methods in the above embodiments, this application provides a computer-readable storage medium including computer program instructions. When the computer program instructions are executed by a computing device, the computing device performs the methods in the above embodiments; or, when the computer program instructions are executed by a cluster of computing devices, the cluster of computing devices performs the methods in the above embodiments. Exemplarily, the computer-readable storage medium can be any available medium that the computing device can store, or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0105] Based on the methods in the above embodiments, this application provides a computer program product containing instructions that, when executed by a computing device, cause the computing device to perform the methods in the above embodiments, or, when executed by a cluster of computing devices, cause the cluster of computing devices to perform the methods in the above embodiments.

[0106] It is understood that the processor in the embodiments of this application can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.

[0107] The method steps in the embodiments of this application can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.

[0108] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, cloud computing platform, or data center to another website, computer, cloud computing platform, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a cloud computing platform or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0109] It is understood that the various numerical designations used in the embodiments of this application are merely for descriptive convenience and are not intended to limit the scope of the embodiments of this application.

[0110] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application 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 protection scope of the technical solutions of the embodiments of this application.

Claims

1. A marine forecasting method, characterized in that, The method includes: Acquire first marine environmental data and first marine boundary in the first sea area, wherein the first marine environmental data and the first marine boundary are historical data; The fragmented data is processed iteratively by a neural network to predict future marine environmental data and future ocean boundaries. In different iterations, the fragmented data is obtained by dividing the first marine environmental data and the first ocean boundary based on different input block sizes. In the first iteration, the first input block size is the initial size. In any iteration other than the first iteration, the first input block size is adjusted based on the future ocean boundary obtained in the previous iteration. If the continuity of the future ocean boundary obtained in the current iteration meets the requirements or the number of iterations reaches the threshold, the future ocean environment data obtained in the current iteration is output.

2. The method according to claim 1, characterized in that, Before iteratively processing the fragmented data using a neural network, the following steps are also included: The first variable in the first marine environmental data is mapped to a spatiotemporal coordinate system, wherein the spatial coordinates of the spatiotemporal coordinate system are divided into grids, and at least one value of the first variable is filled in one of the grids, wherein the first variable is any variable in the first marine environmental data; A spatiotemporal neighborhood is constructed with the first grid in the spatiotemporal coordinate system as the center, wherein the first grid is a grid in the spatiotemporal coordinate system that is not filled with data; Based on the values ​​filled in the grid contained in the spatiotemporal neighborhood, the required values ​​to be filled in the first grid are calculated, and the first grid is filled.

3. The method according to claim 2, characterized in that, The temporal neighborhood radius and spatial neighborhood radius of the spatiotemporal neighborhood are determined based on the density of effective grid points in the spatiotemporal coordinate system, where effective grid points are grids filled with the values ​​of the first variable.

4. The method according to any one of claims 1-3, characterized in that, The iterative processing of fragmented data via neural networks includes: In any iteration, the fragmented data is transformed from the time domain to the frequency domain and features are extracted to obtain frequency domain features. Then, an attention mechanism is used to process the frequency domain features to predict the future marine environment data and the future marine boundary.

5. The method according to any one of claims 1-4, characterized in that, The neural network is constructed based on the transformer architecture. In the process of attention calculation in the neural network, the query vector matrix is ​​obtained based on the sliced ​​data related to the first marine environment data, and the key vector matrix and value vector matrix are obtained based on the sliced ​​data related to the first marine boundary.

6. The method according to any one of claims 1-5, characterized in that, The first marine environmental data includes: sea surface height, and one or more of the following at different sea surface heights: temperature, salinity, zonal current velocity, and meridional current velocity.

7. A marine forecasting device, characterized in that, include: The acquisition module is used to acquire first marine environmental data and first marine boundary in the first sea area, wherein the first marine environmental data and the first marine boundary are historical data; The processing module is used to iteratively process the fragmented data through a neural network to predict future marine environmental data and future ocean boundaries. In different iterations, the fragmented data is obtained by dividing the first marine environmental data and the first ocean boundary based on different input block sizes. Furthermore, in the first iteration, the first input block size is the initial size. In any iteration other than the first iteration, the first input block size is adjusted based on the future ocean boundary obtained in the previous iteration. The processing module is also used to output the future marine environment data obtained in the current iteration when the continuity of the future ocean boundary obtained in the current iteration meets the requirements or the number of iterations reaches a threshold.

8. The apparatus according to claim 7, characterized in that, Before iteratively processing the fragmented data through a neural network, the processing module is also used for: The first variable in the first marine environmental data is mapped to a spatiotemporal coordinate system, wherein the spatial coordinates of the spatiotemporal coordinate system are divided into grids, and at least one value of the first variable is filled in one of the grids, wherein the first variable is any variable in the first marine environmental data; A spatiotemporal neighborhood is constructed with the first grid in the spatiotemporal coordinate system as the center, wherein the first grid is a grid in the spatiotemporal coordinate system that is not filled with data; Based on the values ​​filled in the grid contained in the spatiotemporal neighborhood, the required values ​​to be filled in the first grid are calculated, and the first grid is filled.

9. The apparatus according to claim 8, characterized in that, The temporal neighborhood radius and spatial neighborhood radius of the spatiotemporal neighborhood are determined based on the density of effective grid points in the spatiotemporal coordinate system, where effective grid points are grids filled with the values ​​of the first variable.

10. The apparatus according to any one of claims 7-9, characterized in that, When the processing module iteratively processes the fragmented data through a neural network, it is specifically used for: In any iteration, the fragmented data is transformed from the time domain to the frequency domain and features are extracted to obtain frequency domain features. Then, an attention mechanism is used to process the frequency domain features to predict the future marine environment data and the future marine boundary.

11. The apparatus according to any one of claims 7-10, characterized in that, The neural network is constructed based on the transformer architecture. In the process of attention calculation in the neural network, the query vector matrix is ​​obtained based on the sliced ​​data related to the first marine environment data, and the key vector matrix and value vector matrix are obtained based on the sliced ​​data related to the first marine boundary.

12. The apparatus according to any one of claims 7-11, characterized in that, The first marine environmental data includes: sea surface height, and one or more of the following at different sea surface heights: temperature, salinity, zonal current velocity, and meridional current velocity.

13. A computing device cluster, characterized in that, It includes at least one computing device, each computing device including a processor and memory; The processor of the at least one computing device is configured to execute instructions stored in the memory of the at least one computing device to cause the cluster of computing devices to perform the method as described in any one of claims 1-6.

14. A computer-readable storage medium, characterized in that, The method includes computer program instructions that, when executed by a cluster of computing devices, cause the cluster of computing devices to perform the method as described in any one of claims 1-6, wherein the cluster of computing devices includes at least one computing device.

15. A computer program product containing instructions, characterized in that, When the instruction is executed by the computing device cluster, the computing device cluster causes the computing device cluster to perform the method as described in any one of claims 1-6, wherein the computing device cluster includes at least one computing device.

Images