Sea surface temperature prediction method and system based on distributed cross-scale joint learning

Abstract

The invention relates to the technical field of ocean temperature prediction, and discloses a sea surface temperature prediction method and system based on distributed cross-scale joint learning, a sea surface temperature prediction model based on distributed cross-scale joint learning is constructed, and the prediction model comprises a local feature learning module deployed at each local node in a local system, the local node is used for carrying out local feature learning on the sea surface temperature image sequence stored on the local node and outputting a sea surface temperature prediction result of the local node; and the cross-scale joint learning module is deployed in a distributed system, learns sea surface temperature image features of different scales on different local nodes through the graph learning module, dynamically optimizes the communication weight between the local nodes, realizes gradient aggregation of different local nodes, and guides cross-scale joint learning. According to the method, image data is not directly transmitted in the training process, data leakage is avoided, the communication weight matrix is dynamically optimized, and the cross-scale feature of the sea temperature data is fully considered.

Description

Technical Field

[0001] The present application relates to the technical field of ocean temperature prediction, and specifically to a method and system for predicting sea surface temperature based on distributed cross-scale joint learning. Background Art

[0002] Sea surface temperature prediction is an important research topic in the field of marine science. Most sea surface temperature data comes from satellite remote sensing, buoys and other sensors. The data volume is large and diverse, and involves sensitive information. The prediction of sea surface temperature spatiotemporal data field based on deep neural networks is an important attempt to apply artificial intelligence to the field of marine science. Common methods include physical model-based and data-driven machine learning methods. Physical model-based methods are usually unable to accurately model certain complex nonlinear relationships, while data-driven machine learning predicts sea surface temperature more accurately by learning patterns and features from a large amount of historical ocean data, thereby making more accurate forecasts.

[0003] At present, deep learning-based sea surface temperature prediction uses a centralized learning training method, and mostly adopts recurrent neural networks (RNNs) to process time-series sea surface temperature data, including long short-term memory (LSTM) and gated recurrent unit (GRU) variants, to model the temporal relationship between SST time series data, and apply a fully connected layer to map to the final prediction result, achieving good prediction results.

[0004] However, there are several problems with the current method: First, the cross-scale characteristics of sea temperature data are not considered. For example, the current method only trains single-scale images and does not consider letting the model learn the cross-scale characteristics of sea temperature. Second, the deep learning model uses centralized training, that is, all data is first transferred to the processor and then processed centrally, which occupies network transmission resources and leads to information leakage risks. For example, the sea temperature image data is transmitted to the central server and the model is trained at the central node. During the data upload process, the experimental data is neither encrypted nor added to ensure the security of the transmission process. During the data transmission process, the network bandwidth resources are occupied and the real data may be leaked. Third, the existing federated learning architecture cannot be directly applied to sea temperature image prediction. The federated learning method allows data to be trained in parallel on multiple nodes, significantly reducing the training time, making more effective use of existing computing resources, and ensuring data privacy and security. However, marine science data has a complex structure and has requirements in processing and application. There is no distributed learning architecture that can be directly applied to sea temperature prediction. Summary of the invention

[0005] The technical problem to be solved by this application is to overcome the shortcomings of the existing technology and provide a sea surface temperature prediction method and system based on distributed cross-scale joint learning. Under the premise of ensuring the local storage of original data, cross-node joint feature learning is realized through distributed model parameter interaction, thus avoiding security issues such as data leakage and expanding the applicable scenarios of the prediction model.

[0006] To achieve the above-mentioned purpose, the first aspect of the present application provides a sea surface temperature prediction method based on distributed cross-scale joint learning, and constructs a sea surface temperature prediction model based on distributed cross-scale joint learning. The prediction model adopts a local system and a distributed system architecture. The local system includes a plurality of local nodes. The communication relationship between each local node is defined by a connection matrix. The matrix elements contain 0 or 1 to indicate whether the local nodes can communicate. Each local node is distributed with a sea surface temperature image sequence from different sources but with the same data structure as the input of the prediction model. The prediction model includes: A local feature learning module, deployed on each local node in the local system, for performing local feature learning on a sea surface temperature image sequence stored on the local node and outputting a sea surface temperature prediction result of the local node; The cross-scale joint learning module is deployed in the distributed system, and learns the sea surface temperature image features of different scales on different local nodes through the graph learning module, dynamically optimizes the communication weight between each local node, realizes the gradient aggregation of different local nodes, obtains the global gradient update, and distributes the updated gradient value to each local node to guide the cross-scale joint learning.

[0007] Furthermore, the local feature learning module performs local feature learning on the sea surface temperature image sequence stored on the local node, including: constructing the sampling points of the sea surface temperature image as graph nodes, each sampling point includes latitude and longitude coordinates and temperature values, and generating an adjacency matrix based on the Euclidean distance between the sampling points. The adjacency matrix is ​​expressed as: ; in, represents the Euclidean distance between sampling points i and j, It is expressed as a threshold distance. When the distance between two sampling points is less than or equal to the threshold distance, the edge is connected. When the distance between two sampling points is greater than the threshold distance, the edge is not connected.

[0008] Furthermore, the local feature learning module performs local feature learning on the sea surface temperature image sequence stored on the local node, specifically including: Feature extraction is performed through a distributed graph convolution module, including the sea surface temperature image feature at time t on one of the local nodes and the adjacency matrix A as the input of the distributed graph convolution module, and the output of the local node i is obtained through GCN. , define the network weight as W, and the GCN is expressed as: ; Among them, the initial input , Represents the feature representation of the sea surface temperature image at time t on the local node i, and defines , is the identity matrix, D is the degree matrix, , is an activation function, where the model gradient of a local node i is defined as ; The operation is performed using a temporal feature learning module, and the output of the distributed graph convolution module is learned through an LSTM network; The fully connected layer is used to predict the output of the time series feature learning module and output the prediction result of the local node.

[0009] Furthermore, the cross-scale joint learning module learns the sea surface temperature image features of different scales on different local nodes through the graph learning module, including: the output of the distributed graph convolution module learns the weight of distributed gradient aggregation through the graph learning module, and cross-scale joint learning is performed according to the value returned by the graph learning module to optimize the network weight. The graph learning module is expressed as: ; in, represents the graph convolutional network output from local node i, Represents the F-norm of the matrix.

[0010] Furthermore, the cross-scale joint learning module dynamically optimizes the communication weight between each local node, including: optimizing the communication weight matrix through the graph learning module , the communication weight matrix Aggregation parameters used to guide each local node in gradient aggregation, the aggregation parameters include sea surface temperature image features at different scales, and distributed gradient aggregation is expressed as: ; in, represents the set of nodes that can communicate with the local node i,

[0011] Furthermore, the output features obtained by the distributed graph convolution module are It will be used as the input of the temporal feature learning module, and the temporal feature learning module adopts the LSTM network to learn the temporal information of the input features. The LSTM network performs operations including forget gate calculation, input gate calculation, candidate cell state generation, cell state update, output gate calculation and hidden state update.

[0012] Furthermore, the output of the time series feature learning module is predicted by a fully connected layer, which is composed of a multi-layer perceptron with two hidden layers, and the hidden state of the LSTM network output is mapped to the sea surface temperature prediction at the future moment. .

[0013] Furthermore, the loss function is expressed as: ; in, To predict the loss, is the loss of the cross-scale distributed joint learning module, and the parameters is a trade-off parameter; Predicting Losses It is expressed as: ; in, represents the prediction result on the local node i, represents the truth value; Loss of the Cross-Scale Distributed Joint Learning Module It is expressed as: ; in, represents the graph convolutional network output on the local node i, represents the results of the graph learning module, is a trade-off parameter.

[0014] To achieve the above-mentioned purpose, the second aspect of the present application provides a sea surface temperature prediction system based on distributed cross-scale joint learning, the prediction system comprising a sea surface temperature prediction model based on distributed cross-scale joint learning, the prediction model adopts a local system and a distributed system architecture, the local system comprises a plurality of local nodes, the communication relationship between each local node is defined by a connection matrix, the matrix elements contain 0 or 1 to indicate whether the local nodes can communicate, each local node is distributed with a sea surface temperature image sequence from different sources but with the same data structure as the input of the prediction model, the prediction model comprises: A local feature learning module, deployed on each local node in the local system, for performing local feature learning on a sea surface temperature image sequence stored on the local node and outputting a sea surface temperature prediction result of the local node; The cross-scale joint learning module is deployed in the distributed system, and learns the sea surface temperature image features of different scales on different local nodes through the graph learning module, dynamically optimizes the communication weight between each local node, realizes the gradient aggregation of different local nodes, obtains the global gradient update, and distributes the updated gradient value to each local node to guide the cross-scale joint learning.

[0015] After adopting the above technical solution, the present application has the following beneficial effects compared with the prior art: In this application, the prediction model uses a distributed federated learning framework to effectively ensure the security of image data. Under the premise of ensuring the local storage of original data, cross-node joint feature learning is achieved through distributed model parameter interaction. Compared with the traditional centralized training method, this framework uses an innovative privacy protection mechanism to not directly transmit image data during training, avoiding security issues such as data leakage. While avoiding the transmission of original data, it builds a globally optimized deep learning model, effectively solving the security bottleneck problem in ocean data sharing and expanding the applicable scenarios of the prediction model.

[0016] In this application, a feature fusion mechanism driven by a dynamic communication weight matrix is ​​constructed for the multi-scale spatiotemporal characteristics of the ocean temperature field, and the communication weight matrix is ​​used to guide the joint learning of distributed multi-scale features. By fusing the multi-scale feature information between different nodes and establishing a hierarchical feature interaction criterion, it is possible to adaptively fuse local refined features with global macro features. Joint learning of feature information at different scales can achieve accurate capture and collaborative optimization of key scale features in the evolution of the sea temperature field, significantly improving the spatiotemporal modeling capabilities of cross-regional sea temperature forecasts.

[0017] The specific implementation methods of the present application are further described in detail below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings are part of this application and are used to provide a further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application, but do not constitute an improper limitation on this application. Obviously, the drawings described below are only some embodiments. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0019] In the drawings of the specification: Figure 1 is a schematic diagram of the architecture of the local system and the distributed system in this specific implementation mode; Figure 2 is a schematic diagram of the architecture of a local feature learning module in a local system in this specific implementation mode; Figure 3Schematic diagram of the architecture of the distributed graph convolution module in the local system and the cross-scale joint learning module in the distributed system in this specific implementation mode; Figure 4 This is a schematic diagram of the architecture of the temporal feature learning module in this specific implementation mode; Figure 5 It is a logical schematic diagram of the graph learning module in this specific implementation method. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. The following embodiments are used to illustrate the present application but are not used to limit the scope of the present application.

[0021] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present application, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0022] See also Figure 1 The present application provides a sea surface temperature prediction method based on distributed cross-scale joint learning, and constructs a sea surface temperature prediction model based on distributed cross-scale joint learning. The prediction model adopts a local system and a distributed system architecture. The local system includes a number of local nodes. The communication relationship between each local node is defined by a connection matrix. The matrix elements contain 0 or 1 to indicate whether the local nodes can communicate. Each local node is distributed with a sea surface temperature image sequence from different sources but with the same data structure as the input of the prediction model. The prediction model includes: A local feature learning module is deployed in each local node of the local system, and is used to perform local feature learning on the sea surface temperature image sequence stored on the local node and output the sea surface temperature prediction result of the local node; The cross-scale joint learning module is deployed in a distributed system. It learns the sea surface temperature image features of different scales on different local nodes through the graph learning module, dynamically optimizes the communication weights between each local node, realizes the gradient aggregation of different local nodes, obtains the global gradient update, and distributes the updated gradient value to each local node to guide the cross-scale joint learning.

[0023] It should be noted that the execution subject of the prediction method in this embodiment is a sea surface temperature prediction device based on distributed cross-scale joint learning, which can be an electronic device, a component in an electronic device, an integrated circuit, or a chip. The electronic device can be a mobile electronic device or a non-mobile electronic device. Exemplarily, the mobile electronic device can be a mobile phone, a tablet computer, a laptop computer, a PDA, a vehicle-mounted electronic device, a wearable device, etc., and the non-mobile electronic device can be a server and a personal computer, etc., which are not specifically limited in this application. The following takes the execution subject as an example of a server to describe the sea surface temperature prediction method based on distributed cross-scale joint learning in this embodiment.

[0024] In an achievable implementation, the local feature learning module performs local feature learning on the sea surface temperature image sequence stored on the local node, including: constructing the sampling points of the sea surface temperature image as graph nodes, each sampling point includes latitude and longitude coordinates and temperature values, and generating an adjacency matrix based on the Euclidean distance between the sampling points. The adjacency matrix is ​​expressed as: ; in, represents the Euclidean distance between sampling points i and j, It is expressed as a threshold distance. When the distance between two sampling points is less than or equal to the threshold distance, the edge is connected. When the distance between two sampling points is greater than the threshold distance, the edge is not connected.

[0025] Specifically, for any local node i, the SST image sequence it has is expressed as , for any of the images , with a resolution of k k pixels, each pixel contains latitude and longitude information and sea temperature information. In the horizontal direction, sampling is done every s kilometers, and in the vertical direction, sampling is done every m kilometers. Sampling points, consider the sampling points as graphs Nodes, Represents the Euclidean distance between sampling points i and j, setting the threshold distance , when the Euclidean distance is greater than or equal to the threshold distance, the two points are considered connected, otherwise they are not connected, so the adjacency matrix is ​​defined.

[0026] See also Figure 2 and Figure 3 In an achievable implementation, the local feature learning module performs local feature learning on the sea surface temperature image sequence stored on the local node, specifically including: Feature extraction is performed through a distributed graph convolution module, including the sea surface temperature image feature at time t on one of the local nodes and the adjacency matrix A as the input of the distributed graph convolution module, and the output of the local node i is obtained through GCN. , define the network weight as W, and GCN is expressed as: ; Among them, the initial input , Represents the feature representation of the sea surface temperature image at time t on the local node i, and defines , is the identity matrix, D is the degree matrix, , is an activation function, where the model gradient of a local node i is defined as ; The temporal feature learning module is used to perform operations, and the output of the distributed graph convolution module is learned through the LSTM network; The fully connected layer is used to predict the output of the time series feature learning module and output the prediction result of the local node.

[0027] It should be noted that is the feature representation sequence of the time series sea temperature image on the local node i, Represents the feature representation of the SST image at time t on the local node i (i is an integer greater than or equal to 1), represents the number of the graph node on the SST image, and A is the adjacency matrix of the graph.

[0028] See also Figure 3 and Figure 5 In an achievable implementation, the cross-scale joint learning module learns the sea surface temperature image features of different scales on different local nodes through the graph learning module, including: the output of the distributed graph convolution module learns the weight of the distributed gradient aggregation through the graph learning module, and the cross-scale joint learning is performed according to the value returned by the graph learning module to optimize the network weight. The graph learning module is expressed as: ; in, represents the graph convolutional network output from local node i, Represents the F-norm of the matrix.

[0029] In another achievable implementation, the cross-scale joint learning module dynamically optimizes the communication weight between each local node, including: optimizing the communication weight matrix through the graph learning module , the communication weight matrix Aggregation parameters used to guide each local node in gradient aggregation. Aggregation parameters include sea surface temperature image features at different scales. Distributed gradient aggregation is expressed as: ; in, represents the set of nodes that can communicate with the local node i, represents the gradient of node j that communicates with local node i, represents the communication weight of local nodes i and j, Represents the gradient of the local node i after distributed aggregation.

[0030] It should be noted that this embodiment optimizes the communication weight matrix, utilizes the feature information of different scales of each node, performs joint learning on information in different scale spaces, uses the optimized communication weight matrix to guide cross-scale joint learning, and calculates the pre-gradient parameters in the distributed gradient update process by calculating the similarity of feature spaces of different scales, rather than simply taking the average value, thereby obtaining better model performance.

[0031] Specifically, in the local system, during the model initialization phase, the initial input feature of the local node i at time step t is expressed as ,in is the original feature vector of the node, and l represents the number of layers of the neural network. Input the graph convolution layer GCN with the adjacency matrix A to get the output of the local node i The output features of local node i The distributed system is used by the graph learning module (GL) to receive output features from different local nodes. , and calculate the optimized communication weight matrix based on this value , according to the values ​​in the communication weight matrix Guide distributed gradient aggregation; then perform gradient aggregation of local nodes according to the distributed graph convolution (D-GCNAvg) strategy, and send the optimized gradient value back to the local system. The local system optimizes the network weights according to the updated gradient value.

[0032] Then pass through the output features of the double-layer GCN It will enter the time series feature learning module, which has multiple LSTM structures in series, and the output features at different time steps It will enter different LSTMs in sequence, and the final output features will be obtained after passing through the layers of the LSTM network. , this value can get the final prediction result after passing through the fully connected layer .

[0033] See also Figure 2 , Figure 3 and Figure 4 In practical applications, the output features obtained by the distributed graph convolution module are It will be used as the input of the temporal feature learning module. The temporal feature learning module uses the LSTM network to learn the temporal information of the input features. The LSTM network performs operations including forget gate calculation, input gate calculation, candidate cell state generation, cell state update, output gate calculation and hidden state update.

[0034] Specifically, when the output of the distributed graph convolution module enters the LSTM network, the input data first passes through the forget gate, which determines which information should be forgotten or retained. At each time step t, the calculation formula of the forget gate is as follows: ; in, is the sigmoid function, where is the weight matrix of the forget gate, is the bias term, represents the input of the current time step, is the hidden state of the previous time step, the output of the forget gate is a value between 0 and 1, indicating how much past information should be forgotten. When it is close to 1, the past information will be completely retained. When it approaches 0, past information will be completely forgotten.

[0035] When new input enters the LSTM network, the input gate determines which information should be retained and updates the cell state. At each time step t, the calculation formula of the input gate is as follows: ; in, is the weight matrix of the input gate, is the bias term, represents the input of the current time step, is the hidden state of the previous time step. Next, LSTM calculates the candidate cell state , which indicates how much influence the new input at the current time step can have on the cell state. The calculation formula for the candidate cell state is as follows: in, is the weight matrix of the input gate, is the bias term, represents the input of the current time step, is the hidden state at the previous time step, It is a hyperbolic tangent function, which is used to map the input value to between -1 and 1. The function of the input gate is to control the weight of the new input at the current time step. Through the input gate, LSTM can better handle long sequence data, avoid the problem of gradient disappearance and gradient explosion, and thus improve the effect and stability of the model.

[0036] LSTM cell state Will be updated and passed to the next time step. At each time step t, the update formula of the cell state is as follows: ; in, is the forget gate, which indicates the weight of forgetting the cell state. is the cell state at the previous time step, is the input gate, which represents the weight for updating the cell state. is the candidate cell state at the current time step, indicating how much influence the new input at the current time step can have on the cell state. During the training process, the LSTM network can adaptively update the cell state through the learned weights, retaining and transmitting important information.

[0037] When the information of the current time step needs to be passed to the next layer or output layer, the output gate is needed to control which information should be output. The calculation formula of the output gate is as follows: ; in, is the weight matrix of the output gate, is the bias term, represents the input of the current time step, is the hidden state of the previous time step. After that, LSTM will pass the cell state through a Function is used to process and obtain the hidden state of the current time step , the specific calculation formula is as follows: ; Through the output gate, LSTM can adaptively control the output of information according to the cell state and hidden state of the current time step. In this way, in the LSTM network, important information can be automatically screened out and passed to the next layer or output layer.

[0038] In one feasible implementation, the output of the temporal feature learning module is The prediction is made through a fully connected layer, which consists of a multi-layer perceptron with two hidden layers. The hidden state of the LSTM network output is mapped to the sea surface temperature prediction at the future moment. .

[0039] In one feasible implementation, the loss function is expressed as: ; in, To predict the loss, is the loss of the cross-scale distributed joint learning module, and the parameters is a trade-off parameter; Predicting Losses It is expressed as: ; in, represents the prediction result on the local node i, represents the truth value; Loss of the Cross-Scale Distributed Joint Learning Module It is expressed as: ; in, represents the graph convolutional network output on the local node i, represents the results of the graph learning module, is a trade-off parameter.

[0040] Based on the same inventive concept, the present application also provides a sea surface temperature prediction system based on distributed cross-scale joint learning, the prediction system includes a sea surface temperature prediction model based on distributed cross-scale joint learning, the prediction model adopts a local system and a distributed system architecture, the local system includes a number of local nodes, the communication relationship between each local node is defined by a connection matrix, the matrix elements contain 0 or 1 to indicate whether the local nodes can communicate, each local node is distributed with a sea surface temperature image sequence from different sources but with the same data structure as the input of the prediction model, the prediction model includes: A local feature learning module is deployed in each local node of the local system, and is used to perform local feature learning on the sea surface temperature image sequence stored on the local node and output the sea surface temperature prediction result of the local node; The cross-scale joint learning module is deployed in a distributed system. It learns the sea surface temperature image features of different scales on different local nodes through the graph learning module, dynamically optimizes the communication weights between each local node, realizes the gradient aggregation of different local nodes, obtains the global gradient update, and distributes the updated gradient value to each local node to guide the cross-scale joint learning.

[0041] The above are only preferred embodiments of the present application, and are not intended to limit the present application in any form. Although the present application has been disclosed as above with preferred embodiments, it is not intended to limit the present application. Any technician familiar with the present application can make some changes or modifications to equivalent embodiments of equivalent changes using the above-mentioned technical contents without departing from the scope of the technical solution of the present application. The implementation schemes in the above embodiments can also be further combined or replaced. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present application without departing from the content of the technical solution of the present application are still within the scope of the solution of the present application.

Claims

1. A sea surface temperature prediction method based on distributed cross-scale joint learning, characterized in that: A sea surface temperature prediction model based on distributed cross-scale joint learning is constructed. The prediction model adopts a local system and a distributed system architecture. The local system includes a number of local nodes. The communication relationship between each local node is defined by a connection matrix. The matrix elements contain 0 or 1 to indicate whether the local nodes can communicate. Each local node is distributed with a sea surface temperature image sequence from different sources but with the same data structure as the input of the prediction model. The prediction model includes: A local feature learning module, deployed on each local node in the local system, for performing local feature learning on a sea surface temperature image sequence stored on the local node and outputting a sea surface temperature prediction result of the local node; The cross-scale joint learning module is deployed in the distributed system, and learns the sea surface temperature image features of different scales on different local nodes through the graph learning module, dynamically optimizes the communication weight between each local node, realizes the gradient aggregation of different local nodes, obtains the global gradient update, and distributes the updated gradient value to each local node to guide the cross-scale joint learning.

2. The method according to claim 1, characterized in that The local feature learning module performs local feature learning on the sea surface temperature image sequence stored on the local node, including: constructing the sampling points of the sea surface temperature image as graph nodes, each sampling point includes latitude and longitude coordinates and temperature values, and generating an adjacency matrix based on the Euclidean distance between the sampling points. The adjacency matrix is ​​expressed as: ； in, represents the Euclidean distance between sampling points i and j, It is expressed as a threshold distance. When the distance between two sampling points is less than or equal to the threshold distance, the edge is connected. When the distance between two sampling points is greater than the threshold distance, the edge is not connected.

3. The method according to claim 2, characterized in that The local feature learning module performs local feature learning on the sea surface temperature image sequence stored on the local node, specifically including: Feature extraction is performed through a distributed graph convolution module, including the sea surface temperature image feature at time t on one of the local nodes and the adjacency matrix A as the input of the distributed graph convolution module, and the output of the local node i is obtained through GCN , define the network weight as W, and the GCN is expressed as: ； Among them, the initial input , Represents the feature representation of the sea surface temperature image at time t on the local node i, and defines , is the identity matrix, D is the degree matrix, the degree matrix is ​​a diagonal matrix, and the diagonal elements , is the sum of the elements in the i-th row or i-th column of the adjacency matrix A, is an activation function, where the model gradient of a local node i is defined as ; The operation is performed using a temporal feature learning module, and the output of the distributed graph convolution module is learned through an LSTM network; The fully connected layer is used to predict the output of the time series feature learning module and output the prediction result of the local node.

4. The method according to claim 3, characterized in that The cross-scale joint learning module learns the sea surface temperature image features of different scales on different local nodes through the graph learning module, including: the output of the distributed graph convolution module learns the communication weight during distributed gradient aggregation through the graph learning module, and cross-scale joint learning is performed according to the value returned by the graph learning module to optimize the network weight. The graph learning module is expressed as: ； in, represents the graph convolutional network output from local node i, represents the F-norm of the matrix, represents the activation function, represents the communication weight matrix.

5. The method according to claim 4, characterized in that The cross-scale joint learning module dynamically optimizes the communication weight between each local node, including: optimizing the communication weight matrix through the graph learning module , the communication weight matrix Aggregation parameters used to guide each local node in performing gradient aggregation, the aggregation parameters include sea surface temperature image features at different scales, and distributed gradient aggregation is expressed as: ； in, represents the set of nodes that can communicate with the local node i, represents the gradient of node j that communicates with local node i, represents the communication weight of local nodes i and j, Represents the gradient of the local node i after distributed aggregation.

6. The method according to claim 3, characterized in that: Output features obtained by the distributed graph convolution module It will be used as the input of the temporal feature learning module, and the temporal feature learning module adopts the LSTM network to learn the temporal information of the input features. The LSTM network performs operations including forget gate calculation, input gate calculation, candidate cell state generation, cell state update, output gate calculation and hidden state update.

7. The method according to claim 6, characterized in that The output of the time series feature learning module is predicted by a fully connected layer, which is composed of a multi-layer perceptron with two hidden layers, and the hidden state of the LSTM network output is mapped to the sea surface temperature prediction at the future time. .

8. The method according to claim 1, characterized in that The loss function is expressed as: ； in, To predict the loss, is the loss of the cross-scale distributed joint learning module, and the parameters is a trade-off parameter; Predicting Losses It is expressed as: ； in, represents the prediction result on the local node i, represents the truth value; Loss of the Cross-Scale Distributed Joint Learning Module It is expressed as: ； in, represents the graph convolutional network output on the local node i, represents the results of the graph learning module, is a trade-off parameter.

9. A sea surface temperature prediction system based on distributed cross-scale joint learning, characterized in that: The prediction system includes a sea surface temperature prediction model based on distributed cross-scale joint learning. The prediction model adopts a local system and a distributed system architecture. The local system includes a plurality of local nodes. The communication relationship between each local node is defined by a connection matrix. The matrix elements contain 0 or 1 to indicate whether the local nodes can communicate. Each local node is distributed with a sea surface temperature image sequence from different sources but with the same data structure as the input of the prediction model. The prediction model includes: A local feature learning module, deployed on each local node in the local system, for performing local feature learning on a sea surface temperature image sequence stored on the local node and outputting a sea surface temperature prediction result of the local node; The cross-scale joint learning module is deployed in the distributed system, and learns the sea surface temperature image features of different scales on different local nodes through the graph learning module, dynamically optimizes the communication weight between each local node, realizes the gradient aggregation of different local nodes, obtains the global gradient update, and distributes the updated gradient value to each local node to guide the cross-scale joint learning.

Images