CN117709529A - Ocean surface temperature prediction method based on graph neural network - Google Patents
Ocean surface temperature prediction method based on graph neural network Download PDFInfo
- Publication number
- CN117709529A CN117709529A CN202311713437.5A CN202311713437A CN117709529A CN 117709529 A CN117709529 A CN 117709529A CN 202311713437 A CN202311713437 A CN 202311713437A CN 117709529 A CN117709529 A CN 117709529A
- Authority
- CN
- China
- Prior art keywords
- surface temperature
- graph
- neural network
- ocean surface
- pearson correlation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 54
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 52
- 230000015654 memory Effects 0.000 claims abstract description 25
- 230000003068 static effect Effects 0.000 claims abstract description 17
- 230000008569 process Effects 0.000 claims description 13
- 238000007781 pre-processing Methods 0.000 claims description 10
- 230000002776 aggregation Effects 0.000 claims description 5
- 238000004220 aggregation Methods 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 3
- 230000007547 defect Effects 0.000 abstract description 5
- 238000013136 deep learning model Methods 0.000 abstract description 4
- 230000005540 biological transmission Effects 0.000 abstract description 3
- 238000012549 training Methods 0.000 description 8
- 238000012360 testing method Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 4
- 102100039956 Geminin Human genes 0.000 description 3
- 101000886596 Homo sapiens Geminin Proteins 0.000 description 3
- 230000004913 activation Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000001788 irregular Effects 0.000 description 3
- 230000004931 aggregating effect Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- KJONHKAYOJNZEC-UHFFFAOYSA-N nitrazepam Chemical compound C12=CC([N+](=O)[O-])=CC=C2NC(=O)CN=C1C1=CC=CC=C1 KJONHKAYOJNZEC-UHFFFAOYSA-N 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000002123 temporal effect Effects 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000013277 forecasting method Methods 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000006116 polymerization reaction Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000002759 z-score normalization Methods 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a marine surface temperature prediction method based on a graph neural network, and belongs to the field of deep learning model prediction. The invention converts the data into a graph structure based on the continuous edge strategy of the spatial distance and the pearson correlation coefficient, thereby realizing the complete expression of the sea area information; the method comprises the steps of integrating edge characteristics and forming multi-layer information transmission by designing a static graph encoder based on an iterative graph neural network, and deeply characterizing a heat transmission process; subsequently, a graph memory neural network comprising a static graph encoder, a timing encoder and a fully connected layer decoder is constructed to predict ocean surface temperatures. According to the invention, the ocean surface temperature data is coded in a graph mode, superior generalization capability compared with a comparison method is shown on a plurality of prediction step sizes, the defect that the space information is not fully considered in the existing model is overcome, and the accuracy of graph neural network on sea surface temperature prediction is improved.
Description
Technical Field
The invention belongs to the field of deep learning model prediction, and particularly relates to a marine surface temperature prediction method based on a graph neural network.
Background
Ocean surface temperature (Sea Surface Temperature, SST) is a fundamental physical quantity describing ocean temperature. The sea surface temperature change is closely related to typhoon formation, and the phenomena of early Nino, southern surges and the like. The method predicts the change condition of the sea surface temperature, can timely sense the abnormal sea surface temperature, and has important significance for monitoring the sea environment and preventing and reducing the disaster of the sea. The traditional numerical mode forecasting method needs to consume a large amount of computing resources, and is difficult to ensure timeliness. In recent years, the deep learning method exhibits excellent feature extraction capability and complex nonlinear relation fitting capability, and is widely used in prediction work of SST. However, in the existing ocean surface temperature prediction method based on deep learning, for the SST data with a space-time correlation structure, many researches do not fully consider spatial information, and only do time sequence prediction analysis based on a limited number of point values or area average values. In addition, some studies consider SST data as regular pictures by adopting the idea of convolution, but irregular sea areas with islands cannot be completely encoded, and missing values need to be filled in, so that the prediction accuracy of edge points is reduced.
As can be seen, most sea surface temperature prediction methods based on deep learning do not fully consider spatial information and lack data processing in the edge region. Therefore, the technology applied to deep learning for sea surface temperature prediction still has the defects of insufficient accuracy and the like.
Disclosure of Invention
The invention aims to overcome the defects of insufficient accuracy and the like in the existing sea surface temperature prediction, and provides a sea surface temperature prediction method based on a graph neural network.
In order to achieve the aim of the invention, the technical scheme adopted by the invention is as follows:
s1, carrying out data preprocessing on the obtained ocean surface temperature raw data to obtain ocean surface temperature data to be detected;
s2, calculating node representation and continuous edge representation of ocean surface temperature data to be detected based on a space distance and a Pearson correlation coefficient space-time continuous edge method, and obtaining a graph sequence of the ocean surface temperature data;
s3, obtaining a trained graph memory neural network, wherein the graph memory neural network is formed by sequentially cascading a static graph encoder, a time sequence encoder and a full-connection layer decoder which are constructed based on an iterative graph neural network; and inputting a graph sequence of ocean surface temperature data into the graph memory neural network, and combining a multi-output direct prediction strategy to obtain ocean surface temperature prediction results at a plurality of future moments so as to complete ocean surface temperature prediction.
Preferably, in step S1, the data preprocessing process is as follows:
s11, for the missing values in the ocean surface temperature original data, assigning the missing values as NaN, and not executing the S2-S3 steps on grid points corresponding to the missing values to obtain primarily processed ocean surface temperature data;
s12, performing normalization processing on the primarily processed ocean surface temperature data to obtain ocean surface temperature data to be detected.
Preferably, the specific procedure of step S2 is as follows:
s21, respectively calculating Euclidean distances between two points of the same time dimension in ocean surface temperature data to be detected, sequentially judging the relative sizes of the Euclidean distances and a preset space distance threshold value, and connecting the points of the ocean surface temperature data to be detected according to the judging result: when the Euclidean distance is larger than a preset space distance threshold, two points corresponding to the Euclidean distance are not connected; when the Euclidean distance is smaller than or equal to a preset space distance threshold, connecting two points corresponding to the Euclidean distance;
obtaining a boundary connecting result based on a space distance threshold value after all points in the ocean surface temperature data to be detected are judged whether to be connected or not;
s22, respectively calculating pearson correlation coefficients between two points of the same space dimension and different time dimensions in ocean surface temperature data to be detected, sequentially judging the relative sizes of the pearson correlation coefficients and a preset pearson correlation coefficient threshold value, and connecting the points of the ocean surface temperature data to be detected according to a judging result: when the pearson correlation coefficient is larger than a preset pearson correlation coefficient threshold, connecting two points corresponding to the pearson correlation coefficient; when the pearson correlation coefficient is smaller than or equal to a preset pearson correlation coefficient threshold, two points corresponding to the pearson correlation coefficient are not connected;
when all points in the ocean surface temperature data to be detected are judged whether to be connected or not, obtaining a connecting edge result based on a Pearson correlation coefficient threshold;
s23, acquiring an intersection of a continuous edge result based on a spatial distance threshold and a continuous edge result based on a Pearson correlation coefficient threshold, obtaining a space-time continuous edge result based on a spatial distance and the Pearson correlation coefficient, and forming a graph sequence of ocean surface temperature data.
Preferably, the spatial distance threshold is set to 1.5.
Preferably, the pearson correlation coefficient threshold is set to 0.8.
Preferably, inputting a graph sequence of ocean surface temperature data into the static graph encoder to obtain a group of graph sequences extracted by spatial features; the static graph encoder comprises three layers of identical graph neural networks, in each layer of graph neural network, input node characteristics and edge characteristics are subjected to edge updating to obtain first edge output characteristics, the first edge output characteristics are subjected to edge aggregation to obtain second edge output characteristics, and the second edge output characteristics are subjected to node updating to obtain first node characteristics.
Preferably, both the edge update and the node update employ a multi-layer perceptron with a single hidden layer.
Preferably, the node state sequence is extracted from the graph sequence subjected to the spatial feature extraction, and the node state sequence is input into a time sequence encoder to obtain a feature sequence for completing time sequence encoding.
Preferably, the timing encoder employs a long and short term memory network.
Preferably, the fully-connected layer decoder uses a multi-layer perceptron with two hidden layers.
Compared with the prior art, the invention has the beneficial effects that:
according to the ocean surface temperature prediction method based on the graph neural network, according to the ocean surface temperature graph structure representation method of the spatial distance and the Pearson correlation coefficient, a static graph encoder based on the iterative graph neural network and a time sequence encoder based on the long-short-time memory network are designed to construct the graph memory neural network so as to predict the ocean surface temperature, so that the defect of a deep learning model under data driving can be overcome, the capacity of representing the spatial correlation of the data by the model is improved, and the model precision is improved.
Drawings
FIG. 1 is a schematic diagram of the steps of the present invention;
FIG. 2 is a diagram of a memory neural network according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a static diagram encoder basic structure and encoding process according to an embodiment of the present invention;
fig. 4 shows a basic structure and an encoding process of a time-series encoder according to an embodiment of the present invention.
Detailed Description
In order that the above objects, features and advantages of the invention will be readily understood, a more particular description of the invention will be rendered by reference to the appended drawings. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. The present invention may be embodied in many other forms than described herein and similarly modified by those skilled in the art without departing from the spirit of the invention, whereby the invention is not limited to the specific embodiments disclosed below. The technical features of the embodiments of the invention can be combined correspondingly on the premise of no mutual conflict.
According to the invention, the conventional sea surface temperature prediction is researched, so that the conventional method focuses on the time dependence of sea surface temperature, and the utilization of space information is relatively poor. In order to fully consider the space information, a method for introducing a convolution idea is mostly adopted for research, but the sea surface temperature data are irregularly distributed due to the existence of non-ocean areas such as land, islands and the like, the convolution idea is difficult to carry out complete encoding on the irregular data, so that the sea-land boundary position accuracy is limited, meanwhile, the convolution method focuses on local feature extraction, and the space relevance characterization contained in the ocean surface heat transmission process is insufficient. Therefore, the sea surface temperature prediction method fully considers the spatial relevance of irregular sea surface temperature data, provides a sea surface temperature prediction method based on a graph neural network, designs a static graph encoder based on an iterative graph neural network and a time sequence encoder based on a long-short time memory network to construct a sea surface temperature prediction model so as to make up the defect of a deep learning model under data driving, and improves the capacity of the model for representing the spatial relevance of the data, thereby improving the model precision. Specific implementations of the method are described below.
As shown in fig. 1, in a preferred embodiment of the present invention, there is provided a marine surface temperature prediction method based on a graph neural network, which includes the following steps S1 to S3, the specific implementation of each step is as follows:
s1, performing data preprocessing on the obtained ocean surface temperature raw data to obtain ocean surface temperature data to be detected.
The ocean surface temperature raw data used in the invention is mixed data combining ship data, buoy observation data and satellite observation data. The ocean surface temperature raw data can be obtained through sea surface real-time measurement, and also can be obtained through inversion of remote sensing satellite products. In practical application, data with different sources and different time-space resolutions can be selected according to specific requirements. In an embodiment of the present invention, the ocean surface temperature raw data is the ocean surface temperature product OISST V2.1 (Optimum Interpolation Sea Surface Temperature, OISST) produced by the U.S. national ocean and atmosphere administration, which provides ocean surface temperature data with a spatial resolution of 0.25 ° worldwide from month 9 of 1981 to date. After the ocean surface temperature raw data are obtained in the mode, preprocessing of the ocean surface temperature raw data is carried out.
As an implementation manner of the embodiment of the present invention, in the above step S1, a specific method for performing data preprocessing on the raw data of the ocean surface temperature is as follows:
s11, for the missing values in the ocean surface temperature original data, assigning the missing values as NaN (infinite non-numerical value, not a Number), wherein grid points corresponding to the missing values do Not participate in the construction of a later graph model, namely, the steps S2-S3 are Not executed, and the primarily processed ocean surface temperature data are obtained.
In the process of acquiring the raw data of the ocean surface temperature, data of non-ocean areas such as land, islands and the like are doped, and the data of the non-ocean areas can cause the raw data of the ocean surface temperature to have a missing value.
S12, further, normalizing the primarily processed ocean surface temperature data to obtain ocean surface temperature data to be detected.
It should be noted that, the data normalization is to convert each variable with different value ranges to the same metric scale. As an implementation manner of the embodiment of the present invention, the data normalization manner adopted herein is Z-Score normalization, that is, scaling all samples of each feature of data to a space centered on 0 and having a standard deviation of 1.
It should be noted that the ocean surface temperature data to be detected includes corresponding time information and other attribute information.
S2, calculating node representation and continuous edge representation of the ocean surface temperature data to be detected based on a space distance threshold and a Pelson correlation coefficient space-time continuous edge method, and obtaining a graph sequence of the ocean surface temperature data.
The sequence of the marine surface temperature data includes a temporal feature, a spatial feature, and a self-attribute feature.
As an implementation manner of the embodiment of the present invention, in the step S2, the method for obtaining the map sequence of the ocean surface temperature data is as follows:
s21, respectively calculating Euclidean distances between two points of the same time dimension in ocean surface temperature data to be detected, sequentially judging the relative sizes of the Euclidean distances and a preset space distance threshold value, and connecting the points of the ocean surface temperature data to be detected according to the judging result:
when the Euclidean distance is larger than a preset space distance threshold value, two points corresponding to the Euclidean distance are not connected; when the Euclidean distance is smaller than or equal to a preset space distance threshold value, connecting two points corresponding to the Euclidean distance;
and obtaining a boundary connecting result based on the space distance threshold after judging whether connection is needed or not after all points in the ocean surface temperature data to be detected are judged.
It should be noted that the above-mentioned S21 process may be specifically implemented according to the following functional form:
wherein e ij A boundary result between the point i and the point j based on a space distance threshold value; 1 represents two points for connection; 0 represents that two points are not connected; d, d ij Is the Euclidean distance between point i and point j; d, d min The preset distance threshold is preferably set to 1.5.
S22, respectively calculating pearson correlation coefficients between two points of the same space dimension and different time dimensions in ocean surface temperature data to be detected, sequentially judging the relative sizes of the pearson correlation coefficients and a preset pearson correlation coefficient threshold value, and connecting the points of the ocean surface temperature data to be detected according to a judging result:
when the pearson correlation coefficient is larger than a preset pearson correlation coefficient threshold, connecting two points corresponding to the pearson correlation coefficient;
when the pearson correlation coefficient is smaller than or equal to a preset pearson correlation coefficient threshold, two points corresponding to the pearson correlation coefficient are not connected;
and obtaining a connecting edge result based on the Pearson correlation coefficient threshold after judging whether connection is needed or not at all points in the ocean surface temperature data to be detected.
It should be noted that the above-mentioned S22 process may be specifically implemented according to the following functional form:
wherein f ij A continuous edge result between the point i and the point j based on the threshold value of the Pearson correlation coefficient; 1 represents two points for connection; 0 represents that two points are not connected; r is (r) ij Is the pearson correlation coefficient between point i and point j; r is (r) min Is a preset pearson correlation coefficient threshold, and is preferably set to 0.8.
In the present embodiment, the ocean surface temperature data to be detected from the same spatial dimension and different temporal dimensions are used as different time series, and the time series are consistent in length, so that the pearson correlation coefficient r ij The specific calculation mode of (2) is as follows:
wherein T is the time sequence length, I and J are the time sequence sets at the positions of point I and point J respectively, I t And J t The time series values at the t-th moment in the two time series sets are respectively,and->Each time series average of two time series sets.
S23, acquiring an intersection of a continuous edge result based on a spatial distance threshold and a continuous edge result based on a pearson correlation coefficient threshold, obtaining a space-time continuous edge result based on a spatial distance and a pearson correlation coefficient, and forming a graph sequence of marine surface temperature data of a space-time continuous edge method based on the spatial distance and the pearson correlation coefficient.
It should be noted that the above-mentioned S23 process may be specifically implemented according to the following functional form:
wherein g ij Is a spatiotemporal edge result between point i and point j based on the spatial distance and pearson correlation coefficient.
S3, obtaining a trained graph memory neural network, wherein the graph memory neural network is formed by sequentially cascading a static graph encoder constructed based on an iterative graph neural network, a time sequence encoder constructed based on a long-short time memory network and a full-connection layer decoder. Inputting a graph sequence of ocean surface temperature data into the graph memory neural network, and combining a multi-output direct prediction strategy to obtain an ocean surface temperature prediction result at a future moment so as to complete ocean surface temperature prediction; .
It should be noted that, in step S3 of the present invention, the map sequence of the ocean surface temperature data is first input into the static map encoder to obtain a set of map sequences extracted by spatial features, where the static map encoder includes three layers of identical map neural networks (GNNs), and each layer of map neural networks includes three stages of updating, aggregating and node updating. In each layer of graph neural network, the input node characteristics and the edge characteristics are subjected to edge updating to obtain first edge output characteristics, the first edge output characteristics are subjected to edge aggregation to obtain second edge output characteristics, and the second edge output characteristics are subjected to node updating to obtain first node characteristics.
The static map encoder infrastructure and decoding process is shown in fig. 3, where a map sequence of ocean surface temperature data includes node features and edge features. Specifically, taking a layer of graph neural network as an example, node features numbered 1 in FIG. 3According to the step S2, the space-time edge connecting result based on the space distance and the Pearson correlation coefficient is obtained, namely the node characteristic of the number 1 and the node characteristics of the numbers 2, 3, 4 and 5 are respectively added with->All are connected to obtain the side characteristic +.> The rest of the reference nodes are selected as such. Then the node characteristics, the edge characteristics and the attribute characteristics are subjected to edge updating firstly, and the updated edge characteristics are output +.>And then the updated edge feature +.> Performing edge polymerization to obtain polymerized edge feature ∈>The polymerized edge feature->Updating the node to obtain updated node characteristics +.>
As an implementation manner of the embodiment of the present invention, in the above step S3, the method for constructing the static diagram encoder is as follows:
step 1, inputting a sea-surface temperature map sequence into a static map encoder, outputting an updating result by using a multi-layer perceptron comprising a single hidden layer in an edge updating stage, and capturing nonlinear characteristics by using a ReLU activation function. Here, the ReLU activation function f (x) is specifically implemented according to the following calculation formula:
f(x)＝max(0,x)
and 2, aggregating the node edge connection states of the edge updating stage. The embodiment of the invention considers that the heat change appears as a converging or dissipating process to a certain point on the ocean surface, so that the edge aggregation is realized in a summation mode.
And 3, taking the previous aggregation result and the current state of the node into consideration, updating the state of the node by using a multi-layer perceptron comprising a single hidden layer, capturing nonlinear characteristics through a ReLU activation function, acquiring hidden states of the edge and the node, and finally obtaining a group of graph sequences extracted through spatial characteristics.
As an implementation manner of the embodiment of the present invention, the time sequence encoder in the above step S3 adopts a long and short time memory network (LSTM), and it should be noted that, as shown in fig. 4, the graph sequence extracted by the spatial feature includes a set of node state sequences and a set of edge state sequences, when performing time sequence encoding, the node state sequences in the graph sequence extracted by the spatial feature need to be extracted, and the node state sequences are used as input of the time sequence encoder, and the forgetting gate, the input gate and the output gate of the long and short time memory network are used to extract the time sequence features, so as to obtain the feature sequence for completing the time sequence encoding.
The fully-connected layer decoder in the step S3 adopts a multi-layer perceptron including two hidden layers, the feature sequence after time sequence coding is input into the fully-connected layer decoder, and the features extracted by the space-time mode are converted into sea table temperature prediction results at future time, so that the conversion from time dimension, space dimension to time dimension is realized. In addition, in the sea surface temperature prediction scene, the prediction step length is fixed and sufficient historical data is provided, so that the method adopts a multi-output direct prediction strategy, namely, the prediction of multiple steps in the future is directly realized by constructing only one model.
The specific training, application process and achieved effect of the method of the invention are shown by a specific application example based on the marine surface temperature prediction method based on the graph neural network in the previous embodiment. Specific method steps are as described above, and are not repeated, and only specific effects thereof are shown below.
Examples
In this embodiment, a graph memory neural network (Graph Memory Neural Network, GMNN) is adopted, and the specific network structure is as described above and will not be described again. The overall process of predicting the ocean surface temperature data can be divided into three stages of data preprocessing, model training and data prediction.
1) Data preprocessing and constructing a data set: the present embodiment constructs training data and test data based on the OISST V2.1 (Optimum Interpolation Sea Surface Temperature, OISST) dataset. Firstly, according to the step S1, carrying out data preprocessing on original OSST data: all deletion values were NaN assigned and Z-score normalized. And then taking a certain node in the preprocessed ocean surface temperature data (SST) as an example, calculating Euclidean distance between the node and each peripheral node, connecting the node with the distance less than 1.5, further calculating Pelson correlation coefficients between the node and corresponding samples of each node connected with the node, deleting the connecting edges of the nodes with the correlation coefficients less than 0.8, obtaining a final connecting edge result, generating a graph sequence of the ocean surface temperature data, and completing construction of training samples in a training set and test samples in a test set.
2) According to the step S3, inputting the node, edge and attribute characteristics of the graph sequence in the training sample into an iterative graph neural network for training to obtain a graph sequence extracted by the spatial characteristics, and inputting the node state sequence of the graph sequence in the step S3 into a long-short-time memory network for training to obtain the characteristics of finishing the time sequence coding; and decoding by using a full-connection layer decoder to obtain the trained graph memory neural network.
3) And verifying the prediction performance of the graph memory neural network by using a test sample in the test set, and inputting a graph sequence of ocean surface temperature data in the test sample into the trained graph memory neural network to obtain the prediction results of the ocean surface temperature in the future 1 day, 3 days and 7 days.
In this example, to test the prediction performance of the GMNN model in the present invention, the common time-series prediction models FC-LSTM, FC-GRU and the space-time prediction model GCN-LSTM were selected as a comparison, and performance evaluation was performed using the mean absolute error (mean absolute error, MAE) and root mean square error (root mean squared error, RMSE), and the specific calculation results are shown in table 1.
Compared with the prior art, the chart memory neural network (GMNN) provided by the invention predicts the sea surface temperature, and can improve the prediction accuracy of the sea surface temperature.
Table 1 results of different model comparison experiments
The above embodiment is only a preferred embodiment of the present invention, but it is not intended to limit the present invention. Various changes and modifications may be made by one of ordinary skill in the pertinent art without departing from the spirit and scope of the present invention. Therefore, all the technical schemes obtained by adopting the equivalent substitution or equivalent transformation are within the protection scope of the invention.
Claims (10)
1. The marine surface temperature prediction method based on the graph neural network is characterized by comprising the following specific steps of:
s1, carrying out data preprocessing on the obtained ocean surface temperature raw data to obtain ocean surface temperature data to be detected;
s2, calculating node representation and continuous edge representation of ocean surface temperature data to be detected based on a space distance and a Pearson correlation coefficient space-time continuous edge method, and obtaining a graph sequence of the ocean surface temperature data;
s3, obtaining a trained graph memory neural network, wherein the graph memory neural network is formed by sequentially cascading a static graph encoder, a time sequence encoder and a full-connection layer decoder which are constructed based on an iterative graph neural network; and inputting a graph sequence of ocean surface temperature data into the graph memory neural network, and combining a multi-output direct prediction strategy to obtain ocean surface temperature prediction results at a plurality of future moments so as to complete ocean surface temperature prediction.
2. The marine surface temperature prediction method based on a graph neural network according to claim 1, wherein in step S1, the data preprocessing process is as follows:
s11, for the missing values in the ocean surface temperature original data, assigning the missing values as NaN, and not executing the S2-S3 steps on grid points corresponding to the missing values to obtain primarily processed ocean surface temperature data;
s12, performing normalization processing on the primarily processed ocean surface temperature data to obtain ocean surface temperature data to be detected.
3. The marine surface temperature prediction method based on the graph neural network according to claim 1, wherein the specific process of step S2 is as follows:
s21, respectively calculating Euclidean distances between two points of the same time dimension in ocean surface temperature data to be detected, sequentially judging the relative sizes of the Euclidean distances and a preset space distance threshold value, and connecting the points of the ocean surface temperature data to be detected according to the judging result: when the Euclidean distance is larger than a preset space distance threshold, two points corresponding to the Euclidean distance are not connected; when the Euclidean distance is smaller than or equal to a preset space distance threshold, connecting two points corresponding to the Euclidean distance;
obtaining a boundary connecting result based on a space distance threshold value after all points in the ocean surface temperature data to be detected are judged whether to be connected or not;
s22, respectively calculating pearson correlation coefficients between two points of the same space dimension and different time dimensions in ocean surface temperature data to be detected, sequentially judging the relative sizes of the pearson correlation coefficients and a preset pearson correlation coefficient threshold value, and connecting the points of the ocean surface temperature data to be detected according to a judging result: when the pearson correlation coefficient is larger than a preset pearson correlation coefficient threshold, connecting two points corresponding to the pearson correlation coefficient; when the pearson correlation coefficient is smaller than or equal to a preset pearson correlation coefficient threshold, two points corresponding to the pearson correlation coefficient are not connected;
when all points in the ocean surface temperature data to be detected are judged whether to be connected or not, obtaining a connecting edge result based on a Pearson correlation coefficient threshold;
s23, acquiring an intersection of a continuous edge result based on a spatial distance threshold and a continuous edge result based on a Pearson correlation coefficient threshold, obtaining a space-time continuous edge result based on a spatial distance and the Pearson correlation coefficient, and forming a graph sequence of ocean surface temperature data.
4. A marine surface temperature prediction method based on a graph neural network as claimed in claim 3, wherein the spatial distance threshold is set to 1.5.
5. A graph neural network based ocean surface temperature prediction method according to claim 3, wherein the pearson correlation coefficient threshold is set to 0.8.
6. The marine surface temperature prediction method based on the graph neural network according to claim 1, wherein a graph sequence of marine surface temperature data is input into the static graph encoder to obtain a set of graph sequences extracted by spatial features; the static graph encoder comprises three layers of identical graph neural networks, in each layer of graph neural network, input node characteristics and edge characteristics are subjected to edge updating to obtain first edge output characteristics, the first edge output characteristics are subjected to edge aggregation to obtain second edge output characteristics, and the second edge output characteristics are subjected to node updating to obtain first node characteristics.
7. The method of claim 6, wherein the edge update and the node update each employ a multi-layer perceptron with a single hidden layer.
8. The marine surface temperature prediction method based on the graph neural network according to claim 6, wherein the node state sequence is extracted from the graph sequence subjected to the spatial feature extraction, and the node state sequence is input into a time sequence encoder to obtain the feature sequence subjected to the time sequence encoding.
9. The marine surface temperature prediction method based on a graph neural network of claim 8, wherein the timing encoder employs a long-short-term memory network.
10. The marine surface temperature prediction method based on a graph neural network according to claim 1, wherein the fully connected layer decoder uses a multi-layer perceptron with two hidden layers.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311713437.5A CN117709529A (en) | 2023-12-13 | 2023-12-13 | Ocean surface temperature prediction method based on graph neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311713437.5A CN117709529A (en) | 2023-12-13 | 2023-12-13 | Ocean surface temperature prediction method based on graph neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117709529A true CN117709529A (en) | 2024-03-15 |
Family
ID=90154735
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311713437.5A Pending CN117709529A (en) | 2023-12-13 | 2023-12-13 | Ocean surface temperature prediction method based on graph neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117709529A (en) |
-
2023
- 2023-12-13 CN CN202311713437.5A patent/CN117709529A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111091233B (en) | Short-term wind power prediction modeling method for wind power plant | |
| CN114662788B (en) | Seawater quality three-dimensional time-space sequence multi-parameter accurate prediction method and system | |
| CN112183742B (en) | Neural network hybrid quantization method based on progressive quantization and Hessian information | |
| CN115393540A (en) | Intelligent fusion method and system of three-dimensional marine environment field based on deep learning | |
| CN114549925A (en) | Sea wave effective wave height time sequence prediction method based on deep learning | |
| CN114724012B (en) | Tropical unstable wave early warning method and device based on space-time cross-scale attention fusion | |
| CN110533591B (en) | Super-resolution image reconstruction method based on codec structure | |
| CN111209968B (en) | Multi-meteorological-factor mode prediction temperature correction method and system based on deep learning | |
| CN114445634A (en) | Sea wave height prediction method and system based on deep learning model | |
| CN114817773A (en) | Time sequence prediction system and method based on multi-stage decomposition and fusion | |
| CN115307780A (en) | Sea surface temperature prediction method, system and application based on time-space information interaction fusion | |
| CN115810149A (en) | High-resolution remote sensing image building extraction method based on superpixel and image convolution | |
| CN114764602A (en) | Short-term rainfall prediction method based on space-time attention and data fusion | |
| CN111505706B (en) | Microseism P wave first arrival pickup method and device based on deep T-Net network | |
| Song et al. | Sea level simulation with signal decomposition and machine learning | |
| Kisi et al. | Wind speed prediction by using different wavelet conjunction models | |
| Luo et al. | Reconstruction of chlorophyll-a satellite data in Bohai and Yellow sea based on DINCAE method | |
| CN117668743A (en) | Time sequence data prediction method of association time-space relation | |
| CN117709529A (en) | Ocean surface temperature prediction method based on graph neural network | |
| CN110020379B (en) | Link prediction method based on deep dynamic network embedded representation model | |
| CN115032682A (en) | Multi-station seismic source parameter estimation method based on graph theory | |
| CN113341419A (en) | Weather extrapolation method and system based on VAN-ConvLSTM | |
| Ouyang et al. | Significant Wave Height Forecasting Based on EMD-TimesNet Networks | |
| CN115267883B (en) | Earthquake response prediction model training and predicting method, system, equipment and medium | |
| CN115755227B (en) | Three-dimensional radar extrapolation method based on deep neural network model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |