A method and device for analyzing spatiotemporal characteristics of mesoscale eddies based on neural networks

Abstract

The present application relates to the field of marine information processing technology, and specifically to a method and device for analyzing the spatiotemporal characteristics of mesoscale eddies based on a neural network. The method comprises the following steps: obtaining the type, eddy center position, eddy intensity, and temperature and sound velocity anomaly characteristics of each layer of mesoscale eddies based on three-dimensional ocean current velocity data in a target area; and predicting the eddy center position, eddy intensity, and temperature and sound velocity anomaly characteristics of mesoscale eddies at future times based on the type, eddy center position, eddy intensity, and temperature and sound velocity anomaly characteristics of each layer of mesoscale eddies at historical times. This method solves the problems in the prior art of difficulty in capturing small-scale eddies or rapidly changing eddy characteristics, difficulty in coping with large-scale data processing requirements, and poor robustness.

Description

Technical Field

[0001] The present application relates to the field of marine information processing technology, and in particular to a method and device for analyzing the spatiotemporal characteristics of mesoscale eddies based on a neural network. Background Art

[0002] Mesoscale eddies are a typical manifestation of mesoscale motion in the ocean. They typically range in size from tens to hundreds of kilometers, with lifecycles ranging from days to months. They possess strong material transport and energy conversion capabilities, and have significant implications for ocean energy cycles, ecosystem maintenance, air-sea interactions, and underwater communications and navigation. Therefore, the identification, structural analysis, and spatiotemporal evolution of mesoscale eddies have long been key topics in physical oceanography and underwater information processing.

[0003] Currently, methods for analyzing mesoscale eddies primarily include image recognition techniques based on remote sensing data and traditional statistical analysis methods based on physical fields. The former relies on remote sensing observations such as satellite altimeter data and sea surface temperature (SST), and uses image processing techniques such as edge detection, Hough transform, and waveform fitting to identify eddy boundaries. The latter, based on the velocity field or vorticity field, employs techniques such as empirical orthogonal function decomposition (EOF), spectral analysis, and rotational velocity distribution analysis to construct an eddy identification index system. While these methods have achieved some progress in practical applications, they still face the following technical bottlenecks:

[0004] Traditional remote sensing data has limited spatial and temporal resolution, making it difficult to accurately capture and track rapidly changing small-scale or multi-layered mesoscale eddies. It also struggles to capture the characteristics of small-scale eddies or rapidly changing eddies, and struggles to cope with large-scale data processing requirements. Furthermore, it lacks adaptability to complex ocean environments, and suffers from poor applicability and robustness in complex conditions such as multi-eddy interference and strong background currents. Summary of the Invention

[0005] The present application provides a method, device / system, equipment and computer-readable storage medium for analyzing the spatiotemporal characteristics of mesoscale vortices based on neural networks, which can solve the problems in the existing technology of difficulty in capturing small-scale vortices or rapidly changing vortex characteristics, difficulty in coping with large-scale data processing needs, and poor robustness.

[0006] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0007] In one aspect, the present invention provides a method for analyzing the spatiotemporal characteristics of mesoscale eddies based on a neural network, comprising the following steps:

[0008] Based on the three-dimensional ocean current velocity data in the target area, the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale eddies are obtained;

[0009] The trained neural network predicts the vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of mesoscale vortices at future moments based on the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale vortices at historical moments.

[0010] In some alternative embodiments, training the neural network includes the following steps:

[0011] Based on the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale vortex at different times, a graph structure is constructed. The graph nodes of the graph structure include the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each mesoscale vortex, and the edges of the graph structure are the adjacency relationships between the graph nodes of each mesoscale vortex.

[0012] When the parameters of the neural network are updated iteratively at each step, the graph nodes and edges corresponding to the graph structure at the previous moment are input into the graph neural network, and the predicted type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of the mesoscale vortex at the next moment are output. The loss value of the iterative step is obtained in combination with the label data, and the parameters of the graph neural network are updated based on the loss value.

[0013] In some optional solutions, the loss value is calculated according to the loss function Determine, among which, 、 、 、 , are the hyperparameter weights of strength regression loss, structural continuity constraint loss, temperature regression and sound speed regression loss, respectively. is the total loss value, is the intensity regression loss with respect to eddy strength, is the structural continuity constraint loss about the vortex center position, is the temperature regression loss for temperature anomalies, About the sound velocity regression loss of sound velocity anomaly.

[0014] In some alternative schemes, the intensity regression loss is calculated with respect to the eddy strength ,

[0015] in, is the true value of eddy strength, is the predicted value of eddy intensity, N is the number of mesoscale eddies;

[0016] Structural continuity constraint loss at the vortex center ,

[0017] in, is the predicted position of the vortex center at time step t, is the predicted position of the vortex center at time step t+1, and T is the total number of time steps in the tracking sequence;

[0018] Temperature regression loss on temperature anomalies ,

[0019] in, is the actual observed temperature anomaly value; is the predicted temperature anomaly value;

[0020] About the sound velocity regression loss of sound velocity anomaly

[0021] in, For actual observation of sound velocity anomaly; To predict the sound speed anomaly.

[0022] In some optional solutions, the method of obtaining the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale eddies based on the three-dimensional ocean current velocity data in the target area includes:

[0023] Based on the three-dimensional velocity data of the ocean current in the target area, the types of ocean currents and the eddy strength of each layer are determined, and the eddy core of the mesoscale eddy is screened based on the velocity field and constraint conditions in the eddy strength.

[0024] The mesoscale vortices in each layer are equivalent to irregular disks, and the three-dimensional vortex structure is constructed based on the depth.

[0025] Based on the relative displacement, maximum tangential velocity and circulation area of ​​the three-dimensional vortex structure at consecutive moments, the evolution trend of the mesoscale vortex is identified;

[0026] Based on the velocity and temperature data of a certain day at a certain moment, the velocity and temperature data of the same day in the adjacent set years are subtracted to extract the temperature and sound speed anomaly characteristics of the mesoscale eddy.

[0027] In some optional solutions, determining the type of ocean currents and eddy strengths of each layer based on the three-dimensional ocean current velocity data of the target area includes:

[0028] The positions corresponding to when the east-west velocity component v and the north-south velocity component u in each layer of ocean current velocity data change from positive to negative or from negative to positive are marked as mesoscale eddy boundaries;

[0029] The type of mesoscale vortex is determined based on whether the velocity vector within the mesoscale vortex boundary is clockwise or counterclockwise;

[0030] The mesoscale eddy velocity field is determined based on the east-west velocity component v and the north-south velocity component u of the data points within the mesoscale eddy boundary.

[0031] In some optional schemes, the point that meets the conditions that the directions of the v component in the east-west direction are opposite in sign and the amplitude increases linearly with distance, the directions of the u component in the north-south direction are opposite in sign and the amplitude increases linearly with distance, the velocity value of the mesoscale vortex boundary is minimum, and the directions of the surrounding adjacent velocity vectors are consistent, is determined as the vortex center of the mesoscale vortex.

[0032] In some optional solutions, the method of equating each layer of mesoscale vortexes to an irregular disk and constructing a three-dimensional vortex structure based on depth includes:

[0033] Adjacent disks with the same vortex type and a vertical distance between their centers less than one-fourth of the radius of the previous vortex are considered to belong to the same mesoscale vortex.

[0034] The disks belonging to the same mesoscale vortex are constructed into a three-dimensional vortex structure.

[0035] In some optional schemes, before obtaining the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale eddies based on the three-dimensional ocean current velocity data in the target area, the three-dimensional ocean current data is traversed point by point to eliminate points with incomplete or abnormal data.

[0036] In a second aspect, the present invention provides a device for analyzing the spatiotemporal characteristics of mesoscale eddies based on a neural network, comprising:

[0037] A historical feature acquisition module is used to obtain the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale eddies based on the three-dimensional ocean current velocity data in the target area;

[0038] The neural network prediction module is used to predict the vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of mesoscale vortices at future moments based on the types, vortex center positions, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale vortices at historical moments.

[0039] Compared with the existing technology, the advantages of the present invention are: in this scheme, the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale eddy are obtained based on the three-dimensional ocean current velocity data in the target area, which is equivalent to a round of preprocessing of the three-dimensional ocean current velocity data, which can eliminate the influence of invalid or abnormal data, and the amount of data will be greatly reduced, which can improve the prediction efficiency. By integrating the mesoscale eddy characteristics driven by ocean current data with the learning and reasoning capabilities of neural networks, the bottleneck problems of traditional mesoscale eddy identification and analysis methods in terms of accuracy, stability and dynamic tracking are comprehensively solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0041] Figure 1 Flowchart of a method for analyzing spatiotemporal characteristics of mesoscale eddies based on a neural network in an embodiment of the present invention;

[0042] Figure 2 is a diagram of the ocean current velocity field in an embodiment of the present invention;

[0043] Figure 3 A three-dimensional structure diagram of a mesoscale vortex in an embodiment of the present invention;

[0044] Figure 4 A comparison diagram of the sound velocity profiles in an embodiment of the present invention;

[0045] Figure 5 A comparison diagram of propagation loss in an embodiment of the present invention;

[0046] Figure 6 This is a schematic diagram of the hardware structure of the mesoscale vortex spatiotemporal characteristics analysis device involved in the embodiment of the present application. DETAILED DESCRIPTION

[0047] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0048] like Figure 1 As shown, on the one hand, the present invention provides a method for analyzing the spatiotemporal characteristics of mesoscale vortices based on a neural network, comprising the following steps:

[0049] S1: Based on the three-dimensional ocean current velocity data in the target area, the type, eddy center position, eddy intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale eddies are obtained.

[0050] In this scheme, in order to use neural networks to predict the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of future mesoscale vortices, and eliminate the influence of abnormal ocean current data, and improve the prediction efficiency of the neural network, the three-dimensional ocean current velocity data of the target area at the historical moment is obtained, and the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale vortex at the historical moment are obtained. This is used as the input of the neural network to predict the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of the mesoscale vortex at the future moment. In this way, it is equivalent to having performed a round of preprocessing on the three-dimensional ocean current velocity data, which can eliminate the influence of invalid or abnormal data, and the amount of data will be greatly reduced, which can improve the prediction efficiency.

[0051] Specifically, step S1 includes the following steps:

[0052] S11: Determine the current type and eddy strength of each layer based on the three-dimensional velocity data of the ocean current in the target area, and screen out the eddy core of the mesoscale eddy based on the velocity field and constraint conditions in the eddy strength.

[0053] In this example, the eddy strength includes the circulation area and the maximum tangential velocity. The circulation area is determined based on the mesoscale eddy boundary; the maximum tangential velocity is determined based on the mesoscale eddy velocity field. Determining the current type and eddy strength of each layer includes:

[0054] A: The positions corresponding to when the east-west velocity component v and the north-south velocity component u in each layer of ocean current velocity data change from positive to negative or from negative to positive are marked as mesoscale eddy boundaries.

[0055] In this example, before marking the mesoscale eddy boundary, the three-dimensional ocean current data are traversed point by point, and points with incomplete or abnormal data are removed to ensure the accuracy and robustness of the analysis.

[0056] B: Determine the type of mesoscale vortex based on whether the velocity vector within the mesoscale vortex boundary is clockwise or counterclockwise.

[0057] Specifically, if the local velocity vector within the boundary of the mesoscale vortex rotates counterclockwise, it is a cold vortex; if it rotates clockwise, it is a warm vortex.

[0058] C: Determine the mesoscale eddy velocity field based on the east-west velocity component v and the north-south velocity component u of the data points within the mesoscale eddy boundary.

[0059] like Figure 2 As shown, within the identified boundary points, the velocity field is calculated:

[0060]

[0061] Find the point with the minimum velocity as the candidate vortex center point, and perform the following four constraint checks to confirm the vortex center location:

[0062] Along the east-west direction, the v component has opposite signs on both sides and its amplitude increases linearly with distance;

[0063] Along the north-south direction, the u component has opposite signs on both sides and its amplitude increases linearly with distance;

[0064] There is a global minimum speed value in the candidate area;

[0065] The directions of adjacent velocity vectors around the mesoscale vortex center are consistent and fall in the same quadrant or adjacent quadrants.

[0066] In this example, the vortex center of the mesoscale vortex is screened out based on the velocity field and constraint conditions in the vortex intensity, including: the points that meet the conditions that the directions of the v component in the east-west direction are opposite in sign and the amplitude increases linearly with distance, the directions of the u component in the north-south direction are opposite in sign and the amplitude increases linearly with distance, the velocity value of the mesoscale vortex boundary is the smallest, and the directions of the surrounding adjacent velocity vectors are consistent are determined as the vortex center of the mesoscale vortex.

[0067] The above-mentioned type, eddy intensity and eddy center are calculated for each layer of three-dimensional ocean current data to obtain the type, eddy intensity and eddy center of each layer of ocean current.

[0068] like Figure 3 As shown, S12: each layer of mesoscale vortex is equivalent to an irregular disk, and a three-dimensional vortex structure is constructed based on the depth.

[0069] In the above steps, the boundaries of each layer of mesoscale vortices have been obtained. According to the boundaries of each layer of mesoscale vortices, each layer of mesoscale vortices can be equivalent to an irregular disk to obtain the two-dimensional structure of each layer of mesoscale vortices. By matching the two-dimensional structures of each layer of mesoscale vortices, a three-dimensional vortex structure can be constructed based on depth.

[0070] Specifically, step S12 includes:

[0071] S121: Adjacent disks with the same vortex type and a vertical distance between their centers less than one-fourth of the radius of the previous vortex are determined to belong to the same mesoscale vortex.

[0072] Specifically, to determine whether adjacent layer vortex structures belong to the same three-dimensional mesoscale vortex, the following conditions must be met:

[0073] Consistent type: adjacent layers are cold vortices or warm vortices.

[0074] Spatial continuity: The vertical distance between the centers of adjacent disks is less than one-fourth of the radius of the previous vortex.

[0075] If the above conditions are met, the disks in each layer are classified as the same three-dimensional mesoscale vortex structure.

[0076] The three-dimensional structure assembly is completed based on the consistency of vortex types and spatial distance constraints, realizing the construction of vortex bodies from two-dimensional cross-sections to three-dimensional structures, breaking through the limitation of traditional layered methods that are difficult to characterize the overall structure.

[0077] A tracking strategy based on maximum propagation distance and polarity consistency was designed to effectively ensure the continuous expression of vortices in time series, which is suitable for multi-vortex interference or complex background flow field environments.

[0078] S122: Construct the disks belonging to the same mesoscale vortex into a three-dimensional vortex structure.

[0079] S13: Based on the relative displacement, maximum tangential velocity, and circulation area of ​​the three-dimensional vortex structure at consecutive moments, the evolution trend of the mesoscale vortex is identified.

[0080] At subsequent moments, the algorithm first determines whether the three-dimensional vortex structures at consecutive moments are the same. Specifically, it sets the maximum possible range of movement of the current vortex center and searches for mesoscale vortex centers of the same type (polarity) within this range. The mesoscale vortex closest to the mesoscale vortex center at the previous moment is found and used as a matching target, enabling continuous vortex tracking.

[0081] When the three-dimensional vortex structure at consecutive moments is judged to be the same three-dimensional vortex structure, the evolution trend of the mesoscale vortex is identified based on the relative displacement, maximum tangential velocity and circulation area of ​​the three-dimensional vortex structure at consecutive moments.

[0082] S14: Based on the velocity and temperature data of a certain day, subtract the velocity and temperature data of the same day in the adjacent set years to extract the temperature and sound speed anomaly characteristics of the mesoscale eddy.

[0083] In this example, the method of subtracting the multi-year average (e.g., five-year) data of the same day from the daily data is used to eliminate the influence of seasonal cycles and extract the abnormal velocity field and temperature field of the mesoscale eddy.

[0084] like Figure 4 and Figure 5 As shown, the temperature field profile and sound velocity field profile corresponding to the mesoscale vortex are constructed to analyze the impact of the vortex on the sound propagation environment.

[0085] S2: The trained neural network predicts the vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of mesoscale vortices at future moments based on the types, vortex center positions, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale vortices at historical moments.

[0086] In this example, the neural network uses a graph neural network. Specifically, a stacked graph attention network (GAT) is used as the basic inference model. The model structure is as follows:

[0087] Input layer: receiving node feature matrix , where N is the number of vortices, which is the number of mesoscale vortices identified in step S1, and F is the node feature dimension, which includes node features such as vortex type, vortex center geographical location (longitude, latitude, depth), vortex intensity, temperature anomaly characteristics, and sound speed anomaly characteristics. That is, the N×F matrix is ​​the input node feature, which is obtained based on the vortex feature extraction in the data preprocessing stage in step S1.

[0088] And the receiving edge feature matrix , represents the connection relationship between the node features used to describe each vortex. The elements in the adjacency matrix are It is used to characterize whether there is an adjacency relationship between node feature i and node feature j, where It indicates that there is an edge connection between the two node features. For example, when the vortex intensity is large (>1.2) and the temperature anomaly is high (>1°C), the sound speed anomaly will also increase at the next moment; when the vortex center position shifts northward, it is usually accompanied by a weakening of the intensity. Indicates that there is no connection between the two nodes.

[0089] If there is a structural connection between node feature i and feature node j (such as spatial adjacency or temporal continuity), then Aij=1, otherwise Aij=0. l In the layer, the neighbor set determined by the adjacency matrix A , perform weighted fusion on neighbor node features to complete the feature update operation of target node feature i, specifically using the multi-head attention mechanism to calculate the relationship between adjacent nodes The weighted expression is:

[0090]

[0091] in, is the attention weight, which is the node feature i and the neighbor node feature j in the l The attention weight of the layer indicates the contribution of the adjacent node features to the update of node feature i. is a linear transformation matrix used to perform feature mapping on input node features. Activation functions (such as ELU) are used to enhance the expressive power of the model. is the node feature i in the l+1 The node feature representation of the layer, Indicates the adjacent node feature j in the l The feature representation of the layer, is the set of adjacent node features of node feature i, which is all node features directly connected to node feature i in the graph structure.

[0092] The lth layer refers to the lth neural network computing unit in the graph neural network, that is, the lth propagation layer in the multi-layer graph neural network (GAT).

[0093] The adjacency matrix and the node feature matrix are jointly input into the graph neural network as the basis for the model to learn node feature propagation and feature update. The intermediate layer can stack 2-4 layers of GAT to capture vortex correlations across time and space.

[0094] Output layer: Output according to the task type: the type of mesoscale vortex in each layer, the position of the vortex center, the vortex intensity, and the temperature and sound speed anomaly characteristics.

[0095] Before using the graph neural network, the neural network needs to be trained. Training the neural network includes the following steps:

[0096] Based on the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale vortex at different times, a graph structure is constructed, wherein the graph nodes of the graph structure include the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each mesoscale vortex, and the edges of the graph structure are the adjacency relationships between the graph nodes of each mesoscale vortex.

[0097] In the graph neural network processing process, the input graph nodes are the node feature matrix, and the edges of the graph structure are the edge feature matrix.

[0098] In this example, the training samples are obtained by obtaining three-dimensional ocean current data of a set time period in a set sea area, and the method of obtaining the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale vortex at different times is consistent with the steps in step S1.

[0099] The vortex centers that meet the spatial distance constraint at adjacent moments are connected, referring to the current vortex center and the vortex center at the previous moment. The distance constraint sets the maximum possible range of movement of the current vortex center, within which the search is conducted for mesoscale vortex centers of the same type (polarity). Vortices that meet the type consistency and vertical continuity constraints at adjacent depth layers at the same moment are defined as two two-dimensional vortex structures in adjacent depth layers that are determined to be the same mesoscale vortex.

[0100] When the parameters of the neural network are updated iteratively at each step, the graph nodes and edges corresponding to the graph structure at the previous moment are input into the graph neural network, and the predicted vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of the mesoscale vortex at the next moment are output. The loss value of the iterative step is obtained in combination with the label data, and the parameters of the graph neural network are updated according to the loss value.

[0101] During training, the graph nodes and edges of the graph structure corresponding to the previous moment are input into the graph neural network, and the predicted type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of the mesoscale vortex at the next moment are output. Since the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of the mesoscale vortex corresponding to all acquisition moments have been obtained when obtaining the training samples, the vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of the actual mesoscale vortex corresponding to the vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of the mesoscale vortex at the next moment predicted by the neural network can be used as label data to obtain the loss value of the iterative step, and the parameters of the graph neural network are updated according to the loss value.

[0102] In this embodiment, the loss value is calculated based on the loss function Determine, among which, 、 、 and are the hyperparameter weights of strength regression loss, structural continuity constraint loss, temperature and sound speed loss, respectively. is the total loss value, is the intensity regression loss with respect to eddy strength, is the structural continuity constraint loss about the vortex center position, is the temperature regression loss for temperature anomalies, About the sound velocity regression loss of sound velocity anomaly.

[0103] Intensity regression loss on eddy strength ;in, is the true value of eddy strength, is the predicted value of eddy strength, ‖ ‖ represents the norm of the vector. In this example, the eddy strength is the maximum tangential velocity.

[0104] Structural continuity constraint loss at the vortex center ;in, is the predicted position of the vortex center at time step t, is the predicted position of the vortex center at time step t+1, and T is the total number of time steps in the complete tracking sequence. The goal is to constrain the smoothness of the trajectory, not to fit a specific annotation, but to teach the model to avoid sharp jumps in predictions, consistent with the physical continuity expected of ocean dynamics.

[0105] Temperature regression loss on temperature anomalies :

[0106]

[0107] is the actual observed temperature anomaly value; is the temperature anomaly value predicted by GNN; N is the number of mesoscale eddies.

[0108] About the sound velocity regression loss of sound velocity anomaly :

[0109]

[0110] in: For actual observation of sound velocity anomaly; is the predicted value.

[0111] In summary, this solution comprehensively solves the bottleneck problems of traditional mesoscale eddy identification and analysis methods in terms of accuracy, stability, three-dimensional modeling and dynamic tracking by integrating the physical mechanism modeling driven by ocean current data with the learning and reasoning capabilities of neural networks.

[0112] The mesoscale vortex identification results are modeled as a dynamic graph structure and combined with a multi-head attention mechanism to achieve intelligent prediction of vortex type, intensity and evolution trend. A composite loss function is designed to take into account classification, regression and path continuity goals, significantly improving the automation and adaptability of the system.

[0113] Characteristic anomaly extraction strategy to remove background field disturbances: Through the multi-year monthly mean field subtraction strategy, the seasonal cycle terms in the ocean field data are effectively eliminated, the mesoscale eddy anomaly characteristics are retained, and the physical interpretation ability of environmental factors such as temperature and sound speed is improved.

[0114] End-to-end analysis process integration design: A complete processing flow from ocean current data preprocessing, eddy identification, 3D modeling, tracking analysis, GNN reasoning to environmental response assessment has been built, with high modularity and task customization capabilities.

[0115] In the second aspect, a mesoscale vortex spatiotemporal characteristic analysis device based on a neural network includes: a historical feature acquisition module and a neural network prediction module.

[0116] Among them, the historical feature acquisition module is used to obtain the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale vortex based on the three-dimensional ocean current velocity data in the target area; the neural network prediction module is used to predict the vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of mesoscale vortexes at future moments based on the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale vortex at historical moments.

[0117] Among them, the functional implementation of each module in the above-mentioned neural network-based mesoscale eddy spatiotemporal characteristics analysis device corresponds to the various steps in the above-mentioned neural network-based mesoscale eddy spatiotemporal characteristics analysis method embodiment, and their functions and implementation processes will not be repeated here one by one.

[0118] In a third aspect, an embodiment of the present application provides a mesoscale vortex spatiotemporal characteristics analysis device based on a neural network. The mesoscale vortex spatiotemporal characteristics analysis device based on a neural network can be a personal computer (PC), a laptop computer, a server, or other device with data processing capabilities.

[0119] Reference Figure 6 , Figure 6 The hardware structure diagram of the neural network-based mesoscale eddy spatiotemporal characteristics analysis device involved in the embodiment of the present application is shown in FIG. In the embodiment of the present application, the neural network-based mesoscale eddy spatiotemporal characteristics analysis device may include a processor, a memory, a communication interface, and a communication bus.

[0120] The communication bus may be of any type and is used to interconnect the processor, memory, and communication interface.

[0121] Communication interfaces include input / output (I / O), physical, and logical interfaces, which are used to interconnect components within the neural network-based mesoscale eddy spatiotemporal characteristics analysis device, as well as to interconnect the device with other devices (such as other computing devices or user devices). Physical interfaces can include Ethernet, fiber optic, and ATM interfaces; user devices can include displays and keyboards.

[0122] The memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0123] The processor may be a general-purpose processor that can call a neural network-based mesoscale eddy spatiotemporal characteristics analysis program stored in a memory and execute the neural network-based mesoscale eddy spatiotemporal characteristics analysis method provided in the embodiments of the present application. For example, the general-purpose processor may be a central processing unit (CPU). The method executed when the neural network-based mesoscale eddy spatiotemporal characteristics analysis program is called can be referred to in the various embodiments of the neural network-based mesoscale eddy spatiotemporal characteristics analysis method of the present application, and will not be further described here.

[0124] Those skilled in the art will understand that Figure 6 The hardware structure shown in the figure does not constitute a limitation to the present application and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0125] In a fourth aspect, an embodiment of the present application also provides a computer-readable storage medium.

[0126] The computer-readable storage medium of the present application stores a neural network-based mesoscale vortex spatiotemporal characteristics analysis program, wherein when the neural network-based mesoscale vortex spatiotemporal characteristics analysis program is executed by a processor, the steps of the neural network-based mesoscale vortex spatiotemporal characteristics analysis method as described above are implemented.

[0127] Among them, the method implemented when the neural network-based mesoscale eddy spatiotemporal characteristics analysis program is executed can refer to the various embodiments of the neural network-based mesoscale eddy spatiotemporal characteristics analysis method of this application, and will not be repeated here.

[0128] It should be noted that the serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0129] The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices. The terms "first", "second" and "third" are used to distinguish different objects, etc., and do not represent a sequence, nor do they limit the "first", "second" and "third" to different types.

[0130] In the description of the embodiments of this application, the words "exemplary," "for example," or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary," "for example," or "for example" in the embodiments of this application should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary," "for example," or "for example" is intended to present the relevant concepts in a concrete manner.

[0131] In the description of the embodiments of the present application, unless otherwise specified, “ / ” means or, for example, A / B can mean A or B; “and / or” in the text is merely a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, “multiple” refers to two or more than two.

[0132] In some processes described in the embodiments of the present application, multiple operations or steps are included that appear in a specific order. However, it should be understood that these operations or steps may not be performed in the order in which they appear in the embodiments of the present application or may be performed in parallel. The sequence numbers of the operations are only used to distinguish between different operations, and the sequence numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations or steps may be performed in sequence or in parallel, and these operations or steps may be combined.

[0133] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, or the part that contributes to the existing technology, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above and includes a number of instructions for enabling a terminal device to execute the methods described in each embodiment of this application.

[0134] The above are only preferred embodiments of the present application and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for analyzing the spatiotemporal characteristics of mesoscale eddies based on neural networks, characterized in that: The following steps are involved: Based on the three-dimensional ocean current velocity data in the target area, the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale eddies are obtained; The trained neural network predicts the type, center position, vortex intensity, and temperature and sound speed anomaly characteristics of mesoscale vortices in the future based on the type, center position, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale vortices at historical moments. Training a neural network involves the following steps: Based on the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale vortex at different times, a graph structure is constructed. The graph nodes of the graph structure include the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each mesoscale vortex, and the edges of the graph structure are the adjacency relationships between the graph nodes of each mesoscale vortex. When updating the parameters of the neural network at each iterative step, the graph nodes and edges corresponding to the graph structure at the previous moment are input into the graph neural network, and the predicted type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of the mesoscale vortex at the next moment are output. The loss value of the iterative step is obtained in combination with the label data, and the parameters of the graph neural network are updated according to the loss value; The loss value is based on the loss function Determine, among which, 、 、 and are the hyperparameter weights of strength regression loss, structural continuity constraint loss, temperature regression and sound speed regression loss, respectively. is the total loss value, is the intensity regression loss with respect to eddy strength, is the structural continuity constraint loss about the vortex center position, is the temperature regression loss for temperature anomalies, About the sound velocity regression loss of sound velocity anomaly.

2. The method for analyzing the spatiotemporal characteristics of mesoscale eddies based on a neural network according to claim 1, wherein: Intensity regression loss on eddy strength , in, is the true value of eddy strength, is the predicted value of eddy intensity, N is the number of mesoscale eddies; Structural continuity constraint loss at the vortex center , in, is the predicted position of the vortex center at time step t, is the predicted position of the vortex center at time step t+1, and T is the total number of time steps in the tracking sequence; Temperature regression loss on temperature anomalies , in, is the actual observed temperature anomaly value; is the predicted temperature anomaly value; About the sound velocity regression loss of sound velocity anomaly in, For actual observation of sound velocity anomaly; To predict the sound speed anomaly.

3. The method for analyzing the spatiotemporal characteristics of mesoscale eddies based on neural networks according to claim 1, wherein: The method of obtaining the type, vortex center position, vortex intensity, and temperature and sound velocity anomaly characteristics of each layer of mesoscale eddies based on the three-dimensional ocean current velocity data in the target area includes: Based on the three-dimensional velocity data of the ocean current in the target area, the types of ocean currents and the eddy strength of each layer are determined, and the eddy core of the mesoscale eddy is screened based on the velocity field and constraint conditions in the eddy strength. The mesoscale vortices in each layer are equivalent to irregular disks, and the three-dimensional vortex structure is constructed based on the depth. Based on the relative displacement, maximum tangential velocity and circulation area of ​​the three-dimensional vortex structure at consecutive moments, the evolution trend of the mesoscale vortex is identified; Based on the velocity and temperature data of a certain day at a certain moment, the velocity and temperature data of the same day in the adjacent set years are subtracted to extract the temperature and sound speed anomaly characteristics of the mesoscale eddy.

4. The method for analyzing the spatiotemporal characteristics of mesoscale eddies based on a neural network according to claim 3, wherein: The method of determining the types of ocean currents and eddy strengths of each layer based on the three-dimensional ocean current velocity data of the target area includes: The positions corresponding to when the east-west velocity component v and the north-south velocity component u in each layer of ocean current velocity data change from positive to negative or from negative to positive are marked as mesoscale eddy boundaries; The type of mesoscale vortex is determined based on whether the velocity vector within the mesoscale vortex boundary is clockwise or counterclockwise; The mesoscale eddy velocity field is determined based on the east-west velocity component v and the north-south velocity component u of the data points within the mesoscale eddy boundary.

5. The method for analyzing the spatiotemporal characteristics of mesoscale eddies based on neural networks according to claim 3, wherein: The point that meets the conditions that the directions of the v component on both sides in the east-west direction are opposite in sign and the amplitude increases linearly with distance, the directions of the u component on both sides in the north-south direction are opposite in sign and the amplitude increases linearly with distance, the velocity value of the mesoscale vortex boundary is the smallest, and the directions of the surrounding adjacent velocity vectors are consistent is determined as the vortex center of the mesoscale vortex.

6. The method for analyzing the spatiotemporal characteristics of mesoscale eddies based on neural networks according to claim 3, wherein: The method of converting each layer of mesoscale vortex into an irregular disk and constructing a three-dimensional vortex structure based on depth includes: Adjacent disks with the same vortex type and a vertical distance between their centers less than one-fourth of the radius of the previous vortex are considered to belong to the same mesoscale vortex. The disks belonging to the same mesoscale vortex are constructed into a three-dimensional vortex structure.

7. The method for analyzing the spatiotemporal characteristics of mesoscale eddies based on neural networks according to claim 1, wherein: Before obtaining the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale eddies based on the three-dimensional ocean current velocity data in the target area, the three-dimensional ocean current data is traversed point by point to eliminate points with incomplete or abnormal data.

8. A device for analyzing the spatiotemporal characteristics of mesoscale eddies based on neural networks, characterized in that: include: A historical feature acquisition module is used to obtain the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale eddies based on the three-dimensional ocean current velocity data in the target area; A neural network prediction module is used to predict the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of mesoscale vortices at future moments based on the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale vortices at historical moments; Training a neural network involves the following steps: Based on the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each layer of mesoscale vortex at different times, a graph structure is constructed. The graph nodes of the graph structure include the type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of each mesoscale vortex, and the edges of the graph structure are the adjacency relationships between the graph nodes of each mesoscale vortex. When updating the parameters of the neural network at each iterative step, the graph nodes and edges corresponding to the graph structure at the previous moment are input into the graph neural network, and the predicted type, vortex center position, vortex intensity, and temperature and sound speed anomaly characteristics of the mesoscale vortex at the next moment are output. The loss value of the iterative step is obtained in combination with the label data, and the parameters of the graph neural network are updated according to the loss value; The loss value is based on the loss function Determine, among which, 、 、 and are the hyperparameter weights of strength regression loss, structural continuity constraint loss, temperature regression and sound speed regression loss, respectively. is the total loss value, is the intensity regression loss with respect to eddy strength, is the structural continuity constraint loss about the vortex center position, is the temperature regression loss for temperature anomalies, About the sound velocity regression loss of sound velocity anomaly.

Images