Marine hydrometeorological data fusion processing method based on deep learning fusion

Abstract

The invention relates to the technical field of data processing, in particular to a deep learning fusion-based marine hydro-meteorological data fusion processing method, which comprises the following steps: S1, data acquisition: monitoring marine hydro-meteorological data through a distributed marine monitoring network; s2, constructing a regional collaborative fusion network; s3, space-time consistency optimization: introducing a multi-scale space-time attention mechanism, capturing a local short-term change and a global long-term trend which exist in the data at the same time, and dynamically adjusting weights of different time scales; and optimizing a data fusion result by taking marine physical characteristics as constraint conditions, wherein the marine physical characteristics comprise tide and ocean current periodicity laws. According to the method, through introduction of a multi-scale space-time attention mechanism, accurate capture of short-term dynamic change and long-term trend is realized on the basis of global fusion data, and through a dynamic time window adjustment mechanism, the method can adapt to change characteristics of different time scales.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to a fusion processing method for ocean hydro-meteorological data based on deep learning fusion. Background Art

[0002] With the increasing global climate change and the growing demand for ocean environmental monitoring, the collection and processing of hydro-meteorological data play a crucial role in ocean scientific research. The accurate fusion processing of ocean hydro-meteorological data is of great significance for realizing extreme weather warning, climate trend prediction, and ocean resource development. However, due to the complexity, variability, and extremity of the ocean environment, the existing data processing methods face the following challenges:

[0003] The spatio-temporal changes of ocean data are complex and unstable. Extreme events (such as tropical cyclones, tsunamis) may occur in the short term, and periodic patterns (such as tidal changes, ocean current paths) are presented in the long term. There are obvious deficiencies in the existing methods in terms of the balance between dealing with short-term sudden dynamics and long-term trend changes.

[0004] Ocean hydro-meteorological data is strongly restricted by natural physical laws (such as tidal cycles, ocean current diffusion, etc.). However, most of the existing deep learning models rely on data-driven and fail to fully combine domain knowledge, resulting in poor physical consistency of data results. Summary of the Invention

[0005] The present invention provides a fusion processing method for ocean hydro-meteorological data based on deep learning fusion.

[0006] The fusion processing method for ocean hydro-meteorological data based on deep learning fusion includes the following steps:

[0007] S1, data collection: Monitor ocean hydro-meteorological data through a distributed ocean monitoring network. The distributed ocean monitoring network includes multiple monitoring nodes, and each monitoring node is equipped with observation equipment to collect hydro-meteorological data, including temperature, salinity, ocean current, wind speed, air pressure, and precipitation;

[0008] S2, construction of regional collaborative fusion network:

[0009] S21, node feature embedding: Generate node features from the hydro-meteorological data collected within the coverage of each monitoring node, including the time feature and space feature of the node, and represent the time and space correlation of the data through the edge connection between nodes;

[0010] S22, edge weight definition: Define the weight of the edge according to the spatial distance component and data correlation component between monitoring nodes;

[0011] S23, Graph Convolutional Network Modeling: Based on the global topological structure of the monitoring network, use the graph convolutional network to jointly model all nodes, and comprehensively learn the node features and edge connection relationships through convolutional operations to generate a globally fused representation of ocean hydrometeorological data;

[0012] S3, Spatiotemporal Consistency Optimization: Introduce a multi-scale spatiotemporal attention mechanism to capture the local short-term changes and global long-term trends that coexist in the data, dynamically adjust the weights of different time scales, and classify and weight-adjust the fused data through the spatiotemporal attention mechanism to ensure the coordination of spatiotemporal consistency between sudden change data and regular change data; Combine ocean physical characteristics as constraint conditions to optimize the data fusion result, and the ocean physical characteristics include tidal and ocean current periodic laws to ensure that the consistency of the data conforms to the actual physical meaning.

[0013] Optionally, the monitoring nodes include deep-sea monitoring nodes, shallow-sea monitoring nodes, and near-shore monitoring nodes to ensure that the monitoring network covers the entire target sea area;

[0014] Each monitoring node is equipped with observation equipment, including:

[0015] A temperature sensor for monitoring the change of water temperature in the sea area;

[0016] A salinity sensor for real-time measurement of the salt concentration in the water body;

[0017] An anemometer for monitoring the speed and direction of ocean currents;

[0018] An anemometer for collecting the change of sea surface wind speed and direction;

[0019] A barometric sensor for recording the change of atmospheric pressure;

[0020] A rain gauge for measuring the precipitation in the sea area.

[0021] Optionally, the generation of the time features includes: extracting periodic, trend, and sudden change features based on the time series hydrometeorological data collected within the coverage of the monitoring nodes; modeling the time series data through a long short-term memory network to generate the time feature vector of the nodes, describing the dynamic change law of the data;

[0022] The generation of the spatial features includes: according to the spatial distribution within the coverage of the monitoring nodes, using a convolutional neural network to extract the spatial change patterns of the data, and converting the spatial features including local temperature gradients and ocean current diffusion paths into spatial feature vectors to describe the local characteristics of the area where the nodes are located;

[0023] Concatenate the time features and the spatial features, and generate a unified high-dimensional node feature representation through dimensionality reduction embedding to construct the node feature vector of each monitoring node.

[0024] Optionally, the spatial distance component is calculated by the spherical distance formula to calculate the spatial distance between node v i and node v j .

[0025] The data correlation component is calculated using the correlation coefficient to calculate the time series correlation between node v i and v j .

[0026] The edge weight combines the spatial distance and data correlation, and the edge weight is defined as:

[0027] where d ij represents the spatial distance between node v i and v j , r ij is the correlation coefficient of the time series of node v i and v j , w ij is the edge weight between node v i and v j , and α, β are adjustment parameters used to balance the importance of distance and correlation.

[0028] Optionally, the graph convolutional network learns the high-order representation of the node feature vector:

[0029] where H (l) is the node feature matrix of the l-th layer, initialized as [F 1 , F 2 ,..., F n , is the weighted adjacency matrix, constructed from the edge weight w ij , is the degree matrix of the weighted adjacency matrix, W (l) is the learnable weight matrix of the l-th layer, and σ is the activation function.

[0030] Optionally, the globally fused ocean hydrometeorological data is represented as:

[0031] F global = Aggregate(H (L) ), where F global is the representation of the globally fused ocean hydrometeorological data, H (L) is the node feature matrix of the last layer of the graph convolutional network, and Aggregate(·) is an aggregation operation, such as global average pooling or max pooling.

[0032] Optionally, the S3 includes multi-scale time window partitioning:

[0033] According to the time characteristics of the monitoring data, the time series is divided into multiple time windows (short-term window, long-term window) for capturing local short-term changes and global long-term trends respectively.

[0034] Optionally, the multi-scale spatio-temporal attention mechanism includes the calculation of temporal attention weights and spatial attention weights;

[0035] The temporal attention weight α t is calculated based on multi-scale time window partitioning for the data segment x in each time window t ;

[0036] The spatial attention weight β ij is calculated by computing the data correlation between each monitoring node v i and its neighboring nodes;

[0037] The classification and weighted adjustment of the fused data through the spatio-temporal attention mechanism specifically include:

[0038] Spatio-temporal feature fusion: Fusing the temporal attention weight α t and the spatial attention weight β ij to generate the spatio-temporal joint weight γ ij,t ;

[0039] Based on the dynamic fluctuation of the joint weight γ ij,t , setting the classification threshold θ:

[0040] If γ ij,t > θ, it is determined as sudden change data;

[0041] If γ ij,t ≤ θ, it is determined as regular change data.

[0042] Assign different weights to the classification results, with higher weights for sudden change data to ensure capturing its dynamic change characteristics.

[0043] Optionally, the optimization of the data fusion result by combining ocean physical characteristics as a constraint condition specifically includes:

[0044] Embedding periodic laws: Extracting the tidal period and ocean current period laws of the monitoring area and constructing a constraint function: L physics = ∑ t ‖F t - f physics (t)‖ 2 , where F t is the representation of the globally fused ocean hydro-meteorological data at time t, and f physics (t) is the physical law function describing the periodic behavior of tides or ocean currents.

[0045] Optionally, it further includes an overall optimization objective: after the joint weight adjustment, the marine physical property constraint is added as a loss function to the overall optimization objective: L total = L attention + λ·L physics , where L attention is the optimization loss of the attention mechanism, and λ is a regulation parameter that controls the weight of the marine physical property constraint.

[0046] Advantages of the present invention:

[0047] In the present invention, by introducing a multi-scale spatio-temporal attention mechanism, accurate capture of short-term dynamic changes and long-term trends is achieved on the basis of global fusion of data. Through the dynamic time window adjustment mechanism, it can adapt to the change characteristics of different time scales; combined with the calculation of spatial correlation between nodes, it ensures the consistency optimization of the spatio-temporal characteristics of monitoring nodes in the global network. At the same time, the step of spatio-temporal feature weighted adjustment significantly improves the coordination between data sudden changes and regular changes, and the generated global fusion representation has higher spatio-temporal consistency and dynamic adaptability.

[0048] In the present invention, by embedding constraint conditions of marine physical properties (such as tidal cycle, ocean current path, etc.), the global consistency of data is further optimized. The physical law constraint not only effectively suppresses the influence of abnormal data, but also ensures that the optimized data conforms to the actual physical meaning of the marine hydro-meteorological system. This method combining deep learning with domain knowledge is difficult to achieve in traditional deep learning models, and greatly improves the scientificity and credibility of data results.

[0049] In the present invention, by constructing a regional collaborative fusion network based on a graph convolutional network, global correlation modeling of data between distributed monitoring nodes is realized. The combination of node feature embedding and dynamic definition of edge weights ensures the integration ability of data in multi-source heterogeneous and complex environments. The finally generated global fusion data representation, after spatio-temporal consistency optimization, can be directly used in various scenarios such as real-time monitoring, disaster warning, and climate trend analysis, significantly improving the usability and diversity of data, and providing strong technical support for accurate decision-making in the field of marine hydro-meteorology. Description of the Drawings

[0050] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only those of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts.

[0051] Figure 1Schematic diagram of the method steps of the embodiments of the present invention. Detailed implementation manners

[0052] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments. For some well-known technologies, those skilled in the art can also adopt other alternative ways for implementation; moreover, the accompanying drawings are only for more specifically describing the embodiments and are not intended to specifically limit the present invention.

[0053] It should be noted that in the specification, references to "an embodiment", "embodiments", "exemplary embodiments", "some embodiments", etc. indicate that the described embodiments may include specific features, structures, or characteristics, but not every embodiment necessarily includes the specific feature, structure, or characteristic. Additionally, when combining embodiments to describe a specific feature, structure, or characteristic, implementing such a feature, structure, or characteristic in combination with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the relevant art.

[0054] Generally, terms can be understood at least in part from their use in context. For example, at least in part depending on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in a singular sense, or can be used to describe a combination of features, structures, or characteristics in a plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey a set of exclusive factors, but rather, at least in part depending on the context, can allow for the existence of other factors that may not be explicitly described.

[0055] As Figure 1 shown, the fusion processing method for ocean hydrometeorological data based on deep learning fusion includes the following steps:

[0056] S1, Data acquisition: Monitor ocean hydrometeorological data through a distributed ocean monitoring network. The distributed ocean monitoring network includes multiple monitoring nodes, and each monitoring node is equipped with observation equipment to collect hydrometeorological data, including temperature, salinity, ocean current, wind speed, air pressure, and precipitation;

[0057] S2, Construction of regional collaborative fusion network:

[0058] S21, Node feature embedding: Generate node features from the hydrometeorological data collected within the coverage range of each monitoring node, including the time feature and space feature of the node, and represent the time and space correlation of the data through the edges connecting between the nodes;

[0059] S22, Edge weight definition: Define the weight of the edge according to the spatial distance component and data correlation component between the monitoring nodes;

[0060] S23, Graph Convolutional Network Modeling: Based on the global topological structure of the monitoring network, use the graph convolutional network to jointly model all nodes, and comprehensively learn the node features and edge connection relationships through convolutional operations to generate a globally fused representation of ocean hydrometeorological data;

[0061] S3, Spatiotemporal Consistency Optimization: Introduce a multi-scale spatiotemporal attention mechanism to capture the local short-term changes and global long-term trends that coexist in the data, dynamically adjust the weights of different time scales, classify and weight-adjust the fused data through the spatiotemporal attention mechanism to ensure the coordination of spatiotemporal consistency between sudden change data and regular change data; Combine ocean physical characteristics as constraint conditions to optimize the data fusion results. Ocean physical characteristics include tides and the periodic laws of ocean currents to ensure that the consistency of the data conforms to the actual physical meaning.

[0062] The monitoring nodes include deep-sea monitoring nodes, shallow-sea monitoring nodes, and inshore monitoring nodes to ensure that the monitoring network covers the entire target sea area;

[0063] Each monitoring node is equipped with observation equipment, including:

[0064] A temperature sensor for monitoring the change in water temperature in the sea area;

[0065] A salinity sensor for real-time measurement of the salt concentration in the water body;

[0066] An anemometer for monitoring the speed and direction of ocean currents;

[0067] An anemometer and wind vane for collecting the change in wind speed and direction on the sea surface;

[0068] A barometric sensor for recording the change in atmospheric pressure;

[0069] A rain gauge for measuring the precipitation in the sea area.

[0070] The generation of time features includes: Based on the time series hydrometeorological data collected within the coverage of the monitoring nodes, extract periodic, trend, and sudden change features; Model the time series data through a long short-term memory network to generate the time feature vector of the node, describing the dynamic change law of the data;

[0071] The generation of spatial features includes: According to the spatial distribution within the coverage of the monitoring nodes, use a convolutional neural network to extract the spatial change patterns of the data, and convert the spatial features including local temperature gradients and ocean current diffusion paths into spatial feature vectors to describe the local characteristics of the area where the node is located;

[0072] Concatenate the time features and spatial features, and generate a unified high-dimensional node feature representation through dimensionality reduction embedding to construct the node feature vector of each monitoring node.

[0073] Temporal feature generation: Modeling node v using a Long Short-Term Memory network (LSTM) i for its dynamic characteristics in the time series, with the formula as follows: where is the input of hydrometeorological data of node v i collected at time t, is the hidden state at time t, representing the temporal feature, represents the hidden state and memory state at the previous time, and LSTM(·) represents the activation function of the Long Short-Term Memory network;

[0074] Spatial feature generation: Extracting the spatial characteristics of node v using a Convolutional Neural Network (CNN) i as follows:

[0075] where represents the spatial data matrix within the coverage of node v, including the local temperature gradient and ocean current diffusion path, S i represents the spatial feature representation of node v i and CNN(·) represents the Convolutional Neural Network operation for extracting spatial features; i

[0076] Node feature vector: Concatenating the temporal feature and the spatial feature S i and generating a unified node feature vector through dimensionality reduction: where F i is the final feature vector of node v i and W is the feature dimensionality reduction weight matrix, i represents the feature concatenation operation.

[0077] Calculating the spatial distance component by calculating the spatial distance between node v i and node v j using the spherical distance formula:

[0078] d ij = R·arccos(sinφ i ·sinφ j + cosφ i ·cosφ j ·cos(λ i - λ j ))), where

[0079] d ij represents the spatial distance between node v i and v j , R is the radius of the Earth, φ i , φ j are for node vi and v j latitude, λ i , λ j is the longitude of node v i and v j ;

[0080] The calculation of the data correlation component uses the correlation coefficient to calculate the time series correlation of node v i and v j :

[0081] where r ij is the correlation coefficient of the time series of node v i and v j , is the observed value of node v i and v j at time t, is the average value of the time series of node v i and v j ;

[0082] The edge weight combines the spatial distance and data correlation, and the edge weight is defined as:

[0083] where d ij represents the spatial distance between node v i and v j , r ij is the correlation coefficient of the time series of node v i and v j , w ij is the edge weight between node v i and v j , and α, β are adjustment parameters used to balance the importance of distance and correlation.

[0084] The graph convolutional network learns the high-order representation of the node feature vector:

[0085] where H (l) is the node feature matrix of the l-th layer, initialized as [F 1 , F 2 ,..., F n , is the weighted adjacency matrix, constructed by the edge weight w ij , is the degree matrix of the weighted adjacency matrix, W (l) is the learnable weight matrix of the l-th layer, and σ is the activation function.

[0086] The globally fused ocean hydrometeorological data is represented as:

[0087] F global = Aggregate(H (L) ), where F global is the globally fused representation of ocean hydro-meteorological data, H (L) is the node feature matrix of the last layer of the graph convolutional network, and Aggregate(·) is an aggregation operation such as global average pooling or max pooling.

[0088] The global representation generated by the graph convolutional network synthesizes the data features and mutual relationships of all monitoring nodes into a unified high-dimensional vector, reflecting the spatio-temporal dynamic changes of the entire monitoring network. The globally fused representation retains the spatio-temporal correlation information between monitoring nodes, ensuring that subsequent steps can process sudden changes and long-term trends based on the context of the overall network.

[0089] In spatio-temporal consistency optimization, the globally fused representation provides a complete spatio-temporal feature vector. Spatio-temporal consistency optimization can perform further classification and weight adjustment based on these features to ensure the coordination of sudden changes and regular changes. The high-dimensional feature vector in the globally fused representation can be combined with ocean physical laws (such as tidal and ocean current periods) to provide specific data support for the constraint conditions in spatio-temporal consistency optimization.

[0090] During the spatio-temporal consistency optimization process, the input data of the multi-scale spatio-temporal attention mechanism is the globally fused representation generated by the graph convolutional network. These global features represent the spatial relationships and temporal dynamics of the entire monitoring network, ensuring that the optimization steps operate directly based on the global context information.

[0091] The globally fused features already contain the time characteristics (short-term and long-term trends) of each node. The multi-scale spatio-temporal attention mechanism further refines these time characteristics, dynamically adjusting the weights at different time scales to make the optimization results more in line with the global dynamic changes. In the globally fused representation, the spatial relationships between nodes (correlations defined by edge weights) are passed to the spatio-temporal attention mechanism as the basis for calculating the spatial attention weights. These global spatial relationships ensure that during the optimization process, the influence distribution between different nodes reflects the real ocean hydro-meteorological distribution pattern. S3 includes multi-scale time window partitioning:

[0092] According to the time characteristics of the monitoring data, the time series is divided into multiple time windows (short-term window, long-term window) for capturing local short-term changes and global long-term trends respectively. The length of the time window is dynamically adjusted, determined by the data volatility within the window. The higher the volatility, the shorter the window; the lower the volatility, the longer the window.

[0093] The multi-scale spatio-temporal attention mechanism includes the calculation of time attention weights and spatial attention weights;

[0094] Temporal attention weight α t is calculated based on multi-scale temporal window partitioning for each data segment x in a temporal window t , and the temporal attention weight is calculated as follows: where α t represents the attention weight at time t, and W q , W k are the temporal attention query and key mapping matrices, and d k is the dimension of the key vector;

[0095] Spatial attention weight β ij is calculated by computing the data correlation between each monitoring node v i and its neighboring nodes: where β ij represents the spatial attention weight between nodes v i and v j , F i , F j are the node feature vectors, and W q , W k are the spatial attention query and key mapping matrices;

[0096] Classifying and weighted adjusting the fused data through the spatio-temporal attention mechanism specifically includes:

[0097] Spatio-temporal feature fusion: Fusing the temporal attention weight α t and the spatial attention weight β ij to generate the spatio-temporal joint weight γ ij,t : γ ij,t =α t ·β ij , where γ ij,t represents the joint weight between nodes v i and v j within the time t and spatial range, and classifying sudden changes and regular changes:

[0098] Based on the dynamic fluctuation of the joint weight γ ij,t , set the classification threshold θ:

[0099] If γ ij,t >θ, it is determined as sudden change data;

[0100] If γ ij,t ≤θ, it is determined as regular change data.

[0101] Assign different weights to the classification results, with higher weights for sudden change data to ensure capturing its dynamic change characteristics.

[0102] The classification threshold θ should be dynamically adjusted according to the statistical analysis of historical data. Based on the standard deviation threshold, the formula is as follows: θ = μ + k·σ, where μ is the mean of the joint weight γ ij,t and σ is the standard deviation of the joint weight γ ij,t . k is an adjustment factor (take 1.52, or it can be adjusted according to the actual data sensitivity).

[0103] Assignment of classification weights:

[0104] For sudden change data: A higher weight is assigned, and w 突发 is set to 0.7 to 0.9 to ensure that the impact of sudden changes in optimization is more significant;

[0105] For regular change data: A lower weight is assigned, and w 常规 is set to 0.3 to 0.5 to maintain the background characteristics of regular data.

[0106] Optimizing the data fusion result by combining ocean physical characteristics as a constraint condition specifically includes:

[0107] Embedding periodic rules: Extract the tidal period and ocean current period rules of the monitoring area, and construct a constraint function: L physics = ∑ t ‖F t - f physics (t)‖ 2 , where F t is the representation of ocean hydrometeorological data for global fusion at time t, belonging to the time slice of F global . In spatio-temporal consistency optimization, in order to combine time dynamic characteristics and physical law constraints, F global needs to be disassembled according to the time dimension, and the global data slice corresponding to time t is extracted, denoted as F t , and f physics (t) is a physical law function that describes the periodic behavior of tides or ocean currents.

[0108] It also includes the overall optimization objective: After adjusting the joint weight, the ocean physical characteristic constraint is added to the overall optimization objective as a loss function: L total = L attention + λ·L physics , where L attention is the optimization loss of the attention mechanism, and λ is an adjustment parameter that controls the weight of the ocean physical characteristic constraint.

[0109] Generate a spatio-temporally consistent representation of ocean hydrometeorological data according to the optimized spatio-temporal feature weights and classification adjustment results for subsequent scenario applications.

[0110] The present invention covers any alternatives, modifications, equivalent methods, and solutions made to the essence and scope of the present invention. To enable the public to have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, and those skilled in the art can fully understand the present invention without the description of these details. In addition, well-known methods, processes, procedures, components, and circuits are not described in detail to avoid unnecessary confusion to the essence of the present invention.

[0111] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A fusion processing method for ocean hydrological and meteorological data based on deep learning fusion, characterized in that: The following steps are involved: S1, data collection: monitoring the ocean hydrological and meteorological data through a distributed ocean monitoring network, wherein the distributed ocean monitoring network includes multiple monitoring nodes, each of which is equipped with observation equipment to collect hydrological and meteorological data, including temperature, salinity, ocean currents, wind speed, air pressure and precipitation; S2, regional collaborative integration network construction: S21, node feature embedding: Generate node features from the hydrological and meteorological data collected within the coverage area of ​​each monitoring node, including the time and space features of the node, and characterize the time and space correlation of the data through the edge connection between nodes; S22, edge weight definition: define the edge weight according to the spatial distance component and data correlation component between monitoring nodes; S23, graph convolutional network modeling: Based on the global topological structure of the monitoring network, all nodes are jointly modeled using graph convolutional networks. The node features and edge connection relationships are comprehensively learned through convolution operations to generate a globally integrated representation of ocean hydrological and meteorological data. S3, spatiotemporal consistency optimization: Introduce a multi-scale spatiotemporal attention mechanism to capture the local short-term changes and global long-term trends that exist simultaneously in the data, dynamically adjust the weights of different time scales, classify and weight the fused data through the spatiotemporal attention mechanism, and ensure the coordination of sudden change data and regular change data in spatiotemporal consistency; combine the physical characteristics of the ocean as constraints to optimize the data fusion results, and the physical characteristics of the ocean include tides and the periodic laws of ocean currents.

2. The method for fusion processing of ocean hydrological and meteorological data based on deep learning fusion according to claim 1 is characterized in that: The monitoring nodes include deep-sea monitoring nodes, shallow-sea monitoring nodes and near-shore monitoring nodes; Each monitoring node is equipped with observation equipment, including: Temperature sensor, used to monitor changes in sea water temperature; Salinity sensor, used to measure the salt concentration in water in real time; current meters, which monitor the speed and direction of ocean currents; Anemometer, used to collect sea surface wind speed and direction changes; Barometric pressure sensor, used to record changes in atmospheric pressure; Precipitation meter, used to measure precipitation in the sea area.

3. The method for fusion processing of ocean hydrological and meteorological data based on deep learning fusion according to claim 1 is characterized in that: The generation of the time characteristics includes: extracting periodic, trend and sudden change characteristics based on the time series hydrological and meteorological data collected within the coverage area of ​​the monitoring node; modeling the time series data through a long short-term memory network to generate a time feature vector of the node; The generation of the spatial features includes: according to the spatial distribution within the coverage of the monitoring node, using a convolutional neural network to extract the spatial variation pattern of the data, converting the spatial features including the local temperature gradient and the ocean current diffusion path into a spatial feature vector to describe the local characteristics of the area where the node is located; The temporal features are concatenated with the spatial features, and a unified high-dimensional node feature representation is generated through dimensionality reduction embedding to construct the node feature vector of each monitoring node.

4. The method for fusion processing of ocean hydrological and meteorological data based on deep learning fusion according to claim 1 is characterized in that: The spatial distance component is calculated by calculating the spherical distance formula for node v i and node v j spatial distance; The data correlation component is calculated using the correlation coefficient calculation node v i and v j Time series correlation of The edge weight integrates spatial distance and data correlation, and defines the edge weight as: Among them, d ij Represents node v i and v j The spatial distance, r ij is node v i and v j The correlation coefficient of the time series, w ij is node v i and v j The edge weights between them, α, β are adjustment parameters used to balance the importance of distance and relevance.

5. The method for fusion processing of ocean hydrological and meteorological data based on deep learning fusion according to claim 4 is characterized in that: The graph convolutional network learns a high-order representation of the node feature vector: Among them, H (l) is the node feature matrix of the lth layer, initialized to [F1, F2, ..., F n ], is a weighted adjacency matrix consisting of edge weights w ij Build, is the degree matrix of the weighted adjacency matrix, W (l) is the learnable weight matrix of layer l, and σ is the activation function.

6. The method for fusion processing of ocean hydrological and meteorological data based on deep learning fusion according to claim 5 is characterized in that: The globally fused ocean hydrological and meteorological data is expressed as: F global =Aggregate(H (L) ), where F global is the global fusion of ocean hydrological and meteorological data, H (L) is the node feature matrix of the last layer of the graph convolutional network, and Aggregate(·) is the aggregation operation.

7. The method for fusion processing of ocean hydrological and meteorological data based on deep learning fusion according to claim 1 is characterized in that: The S3 includes multi-scale time window partitioning: According to the temporal characteristics of the monitoring data, the time series is divided into multiple time windows, which are used to capture local short-term changes and global long-term trends respectively.

8. The method for fusion processing of ocean hydrological and meteorological data based on deep learning fusion according to claim 7 is characterized in that: The multi-scale spatiotemporal attention mechanism includes the calculation of temporal attention weights and spatial attention weights; The temporal attention weight α t The calculation is based on multi-scale time window division, and the data segment x in each time window t ; The spatial attention weight β ij The calculation includes for each monitoring node v i , calculate its data correlation with surrounding nodes; The classification and weighted adjustment of the fused data through the spatiotemporal attention mechanism specifically includes: Spatiotemporal feature fusion: The temporal attention weight α t and the spatial attention weight β ij Fusion, generating spatiotemporal joint weights γ ij,t , based on the joint weight γ ij,t Dynamic fluctuations, set the classification threshold θ: If γ ij,t >θ, it is determined as sudden change data; If γ ij,t ≤θ, judged as normal variation data; Different weights are assigned to the classification results.

9. The method for fusion processing of ocean hydrological and meteorological data based on deep learning fusion according to claim 8, characterized in that: The optimization of the data fusion result by combining the ocean physical characteristics as constraint conditions specifically includes: Embedding periodic laws: Extract the tidal cycle and ocean current cycle laws in the monitoring area and construct the constraint function: L physics =Σ t ‖F t -f physics (t)‖ 2 , where F t is the global fusion of ocean hydrological and meteorological data at time t, f physics (t) is a physical law function that describes the periodic behavior of tides or ocean currents.

10. The method for fusion processing of ocean hydrological and meteorological data based on deep learning fusion according to claim 9, characterized in that: It also includes the overall optimization objective: After the joint weight adjustment, the ocean physical property constraints are added as the loss function to the overall optimization objective: L total =L attention +λ·L physics , where L attention is the attention mechanism optimization loss, and λ is the adjustment parameter that controls the weight of the ocean physical property constraints.

Images