CN111680784A - Sea surface temperature deep learning prediction method based on time-space multidimensional influence - Google Patents
Sea surface temperature deep learning prediction method based on time-space multidimensional influence Download PDFInfo
- Publication number
- CN111680784A CN111680784A CN202010461315.1A CN202010461315A CN111680784A CN 111680784 A CN111680784 A CN 111680784A CN 202010461315 A CN202010461315 A CN 202010461315A CN 111680784 A CN111680784 A CN 111680784A
- Authority
- CN
- China
- Prior art keywords
- surface temperature
- sea surface
- time
- deep learning
- space
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 78
- 238000013135 deep learning Methods 0.000 title claims abstract description 41
- 238000012545 processing Methods 0.000 claims abstract description 23
- 238000005516 engineering process Methods 0.000 claims abstract description 12
- 238000012935 Averaging Methods 0.000 claims abstract description 11
- 238000011002 quantification Methods 0.000 claims abstract description 4
- 239000013598 vector Substances 0.000 claims description 33
- 230000008569 process Effects 0.000 claims description 22
- 241000288105 Grus Species 0.000 claims description 9
- 238000013507 mapping Methods 0.000 claims description 4
- PRQROPMIIGLWRP-UHFFFAOYSA-N N-formyl-methionyl-leucyl-phenylalanin Chemical compound CSCCC(NC=O)C(=O)NC(CC(C)C)C(=O)NC(C(O)=O)CC1=CC=CC=C1 PRQROPMIIGLWRP-UHFFFAOYSA-N 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 230000002123 temporal effect Effects 0.000 claims description 3
- 241001522296 Erithacus rubecula Species 0.000 claims 1
- 238000004364 calculation method Methods 0.000 claims 1
- 125000004122 cyclic group Chemical group 0.000 abstract description 3
- 238000013527 convolutional neural network Methods 0.000 description 9
- 230000003442 weekly effect Effects 0.000 description 9
- 238000002474 experimental method Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 3
- 230000007812 deficiency Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000001932 seasonal effect Effects 0.000 description 2
- 238000000547 structure data Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000004907 flux Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 description 1
- 239000010931 gold Substances 0.000 description 1
- 229910052737 gold Inorganic materials 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a sea surface temperature deep learning prediction method based on space-time multidimensional influence. The method comprises three steps: 1. quantification of spatial influence: and quantifying the spatial influence on the target observation point by using the sea surface temperature data of the neighbor observation point of the target observation point, and constructing a multi-dimensional space-time sea surface temperature data set on the target sea area. 2. And (3) data completion processing: and performing data completion processing on the data set by using a neighbor data averaging method. 3. Establishing a prediction model: a sea surface temperature deep learning prediction model based on space-time multidimensional influence, namely a Convolutional gated cyclic unit (Convolutional gated loop) Multilayer Perceptron (CGMP), is established by combining deep learning technologies such as GRU, CNN and MLP. The method integrates space-time multidimensional influence, combines a deep learning technology mode, establishes a high-precision sea surface temperature prediction model, and can be widely applied to sea surface temperature prediction of different sea areas and different scales.
Description
Technical Field
The invention relates to a sea surface temperature prediction method.
Background
Sea Surface Temperature (SST), also known as Surface Temperature, refers to the Temperature of water near the Surface of the ocean. The exact meaning of the sea surface will vary depending on the measurement method used, but is typically between 1 mm and 20 meters below the sea surface. The sea surface temperature plays a fundamental role in the energy exchange between the sea and the atmosphere, and is also a very important parameter in marine dynamics and climate change. Small changes in sea surface temperature have profound effects on global climate, marine ecosystems and even marine life, and the well-known phenomena of el nino and raney are caused by changes in sea surface temperature anomaly (SST). Climate models driven by observed or simulated sea surface temperatures or sea surface temperature anomalies are often used as standard tools for seasonal climate prediction. Sea surface temperature is also often used as a key factor in the study of marine ecosystems and marine life. Therefore, the method has important significance in accurately and effectively observing and predicting the sea surface temperature. It not only makes human know the global climate and marine ecosystem better, but also is an important component for the application of marine related fields, such as extreme weather forecast, seasonal climate forecast, marine organism research and sustainable development fishery. However, various physical and environmental factors (heat flux, radiation, and the weather near the sea surface) affect the sea surface temperature variations, making them highly random and uncertain. Therefore, it is still a challenge to provide a high-precision sea surface temperature prediction method.
The sea surface temperature prediction method can be classified into two types. One is a numerical method and the other is a data-driven method. The traditional numerical method does not easily model the changes in the sea surface temperature because it requires much knowledge of the sea and atmospheric domains and predicts the sea surface temperature with relatively low resolution, usually on a marine or even global scale. Compared with a numerical method, a data-driven method, particularly a deep learning method, requires less knowledge of the sea and atmosphere fields and can predict the sea surface temperature with high resolution on a smaller scale, so it is easier to model to accurately predict the sea surface temperature of the target sea area. However, the sea surface temperature deep learning prediction method based on the time influence generally improves the prediction performance by enhancing the capability of processing time information, and such methods usually focus on capturing the time influence of the historical sea surface temperature, but neglect the overall spatial influence of the target sea area on the target observation point, so that such methods have obvious performance bottlenecks.
Disclosure of Invention
The sea surface temperature deep learning prediction model based on the space-time multidimensional influence is established by integrating the space-time multidimensional influence and combining deep learning technologies such as a Recurrent Neural Network (RNN), a Convolutional Neural Network (CNN) and a Multilayer Perceptron (MLP), so that the performance bottleneck problem is solved, and the accuracy of sea surface temperature prediction is further improved.
The invention is realized by the following technical scheme:
a sea surface temperature deep learning prediction method based on space-time multidimensional influence builds a high-precision sea surface temperature prediction model, namely a convolution gating circulation unit (CGMP) Multilayer Perceptron, by means of fusion of the space-time multidimensional influence and combination of a deep learning technology. The model can overcome the performance bottleneck of the sea surface temperature deep learning prediction model based on the time influence, and can further improve the accuracy of sea surface temperature prediction. The method comprises the following specific steps:
A. quantification of spatial influence: quantifying the space influence on the target observation point by using the sea surface temperature data of the neighbor observation point of the target observation point, and constructing a multi-dimensional space-time sea surface temperature data set on a target sea area;
B. and (3) data completion processing: performing data completion processing on the data set by using a neighbor data averaging method;
C. establishing a prediction model: and (3) establishing a sea surface temperature deep learning prediction model based on the time-space multi-dimensional influence by combining deep learning technologies such as GRU, CNN and MLP, namely the multilayer perceptron CGMP of the convolution gating circulation unit.
The method comprises the following steps of A, quantifying the space influence on the target observation point by using the neighbor observation point sea surface temperature data of the target observation point, and constructing a multi-dimensional space-time sea surface temperature data set on a target sea area, wherein the specific steps are as follows:
a1: the sea surface may be divided into individual grids according to longitude, latitude, and resolution information. One grid is a sea surface temperature observation point. The surface temperature prediction task may be expressed simply as using historical surface temperature data to predict future surface temperature data at the target surface temperature observation. In order to quantify the spatial influence of the target sea surface temperature observation point quantitatively, the spatial influence of the target observation point is assumed to be totally from eight adjacent sea surface temperature observation points, which are called neighbor points, so that a multi-dimensional space-time sea surface temperature data set can be constructed in a target sea area, namely a sea surface temperature (daily average value/weekly average value/monthly average value) complete data set and a sea surface temperature (daily average value/weekly average value/monthly average value) ocean data set, wherein the complete data set comprises data with missing spatial information, and the ocean data set excludes data with missing spatial information.
And B, performing data completion processing on the data set by using a neighbor data averaging method, and specifically comprising the following steps of:
and B1, performing data completion processing on the complete data set by using a neighbor data averaging method. For some target observation points close to the coast, some of their neighbors may be located in the land area, and therefore there is no sea surface temperature observation data at these neighbors, which are called land sites. Therefore, in order to unify and standardize the input of the prediction model, data completion processing is performed on the land points located in the land area. For a land site, the complemented sea surface temperature data is the average value of the real sea surface temperature observation data of the eight neighbor points, namely, a neighbor data averaging method. The data completion method has the advantages of strong practicability and suitability for all different deficiency conditions. If other land points exist in the eight neighbor points, the average value of the remaining sea surface points is calculated after the eight neighbor points are removed.
And C, establishing a sea surface temperature deep learning prediction model based on the space-time multidimensional influence by combining deep learning technologies such as GRU, CNN and MLP, namely a convolution gating cyclic unit multilayer perceptron CGMP, and specifically comprising the following steps:
c1, CGMP is an end-to-end sea surface temperature prediction model, the input of which is a historical SST sequence and the output of which is a predicted future SST sequence. Wherein the historical SST sequence comprises neighbor information and historical information of the observation point of the target sea surface temperature. The CGMP firstly processes the neighbor information of a target observation point on the spatial dimension through a convolution layer to obtain an intermediate vector, namely an empty SST sequence. The CGMP then processes the historical information of the spatio-temporal SST sequence in the time dimension through a GRU, resulting in a hidden state vector. Finally, the multi-layered perceptron maps the hidden state vector onto the prediction result, i.e., the future SST sequence. Convolutional layers are good at processing grid-like structure data and GRUs are good at processing sequence data, so the combinatorial collocation of convolutional layers and GRUs can sufficiently and effectively mine neighbor and historical information in the spatio-temporal dimension. Wherein:
c11, historical SST sequence X is a tensor with one dimension k × 3 × 3, and includes neighbor information (matrix with dimension 3 × 3) and historical information of k time units of the observation point of the target sea surface temperature. Since convolutional neural networks are good at handling grid-like structured data, CGMP first processes the neighbor information of the target observation point in the spatial dimension through one convolutional layer. The depth of the convolutional layer is 1, the size of the convolutional kernel K is 3 × 3, the step size is 1, and the padding is 0. After X is subjected to convolution operation, a space-time SST sequence S with the dimensionality of k multiplied by 1 is obtained, and the operation process is defined as follows:
S=(s1,s2,...,sk)
si=Xi★K+bs,i∈{1,2,...,k}
wherein s isiRepresenting the processed spatio-temporal history information, ★ representing a two-dimensional convolution operation, wk,1–wk,9Representing the weight of the convolution kernel K, bsIndicating the corresponding bias.
C12, the spatiotemporal SST sequence S contains historical information of k time units of the target sea surface temperature observation point. CGMP is subsequently communicated because GRU is good at processing sequence dataHistorical information of the target observation point is processed by one GRU in a time dimension. The GRU receives two inputs at each time step: one is spatio-temporal history information S from a spatio-temporal SST sequence SiThe other is the hidden state vector H from the previous time step(i-1)Or an initial hidden state vector H0. After the history information is processed, the GRU generates a new hidden state vector H at each time stepi. So GRUs generate k hidden state vectors in total, i.e., H1–Hk. The process is defined as follows:
ri=σ(Wrssi+WrHH(i-1)+br)
zi=σ(Wzssi+WzHH(i-1)+bz)
ni=tanh(Wnssi+bns+ri*(WnHH(i-1)+bnH))
Hi=(1-zi)*ni+zi*H(i-1)
wherein HiRepresenting the hidden state vector, s, of the GRU at time step iiRepresenting spatio-temporal history information at time step i, H(i-1)Representing the hidden state vector at time step i-1 or representing the initial hidden state vector at time step 0, ri、ziAnd niRespectively representing the reset, update and new gates of the GRU. σ denotes a sigmoid function, and σ denotes a Hadamard product. W represents the corresponding weight and b represents the corresponding bias.
Finally, CGMP uses a multi-layer perceptron to hide the state vector HkMapped onto the prediction, i.e. the future SST sequence Y. The mapping process is defined as follows:
Y=fMLP(WYHk+bY)
where WY represents the corresponding weight and bY represents the corresponding bias.
The invention has the beneficial effects that:
1. the traditional sea surface temperature deep learning and prediction method based on the time influence usually focuses on capturing the time influence of historical sea surface temperature, but neglects the overall spatial influence of a target sea area on a target observation point, so that the method has obvious performance bottleneck. The newly proposed sea surface temperature deep learning prediction method based on the space-time multidimensional influence solves the performance bottleneck problem by combining the space-time multidimensional influence and deep learning technologies such as RNN, CNN and MLP, and further improves the accuracy of sea surface temperature prediction.
2. According to the method, the spatial influence on the target observation point is quantified by using the sea surface temperature data of the neighbor observation point of the target observation point. The method carries out data completion processing on the spatial information by using a neighbor data averaging method.
3. A high-precision sea surface temperature prediction model is established by combining the space-time multidimensional influence and a deep learning technology. The model can overcome the performance bottleneck of the sea surface temperature deep learning prediction model based on the time influence, and can be widely applied to the sea surface temperature prediction of different sea areas and different scales.
Drawings
Fig. 1 is a general flowchart of the sea surface temperature deep learning prediction method based on the spatio-temporal multidimensional influence of the present invention.
Fig. 2 is a schematic diagram of the principle of quantifying the spatial influence on the target observation point by using the sea surface temperature data of the neighbor observation points of the target observation point, which is described in step a in fig. 1.
Fig. 3 is a schematic diagram of the data completion process performed on the data set by using the neighbor data averaging method in step B in fig. 1.
Fig. 4 is a schematic model principle diagram of the convolution gating cyclic unit multi-layered perceptron CGMP described in step C of fig. 1.
Fig. 5 lists the performance comparison of CGMP, NGMP, CLMP and NLMP on the bohai sea surface temperature daily mean complete data set.
Fig. 6 lists the performance comparison of CGMP, NGMP, CLMP and NLMP on the bohai sea surface temperature cycle mean complete data set.
FIG. 7 lists the comparison of CGMP, FC-LSTM and GED performance on the Bohai sea surface temperature complete data set.
FIG. 8 lists the performance comparison of CGMP, FC-LSTM and GED on the Bohai sea surface temperature ocean data set.
FIG. 9 lists the performance of CGMP, FC-LSTM and GED on a south sea surface temperature integrity data set.
FIG. 10 lists the performance of CGMP, FC-LSTM and GED on the south sea surface temperature ocean data set versus the time of day.
Detailed Description
Preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
The conception of the invention is as follows: firstly, the space influence on a target sea surface temperature observation point is quantified, and a multi-dimensional space-time sea surface temperature data set is constructed on a target sea area. And then carrying out data preprocessing and data completion processing on the data set. And finally, establishing a sea surface temperature deep learning prediction model based on the space-time multi-dimensional influence by combining deep learning technologies such as RNN (LSTM or GRU), CNN and MLP.
In this embodiment, the experiments are all completed on a machine learning platform of Shanghai university, which has two Intel Xeon Gold 6130 CPUs, 192GB RAMs, four Nvidia Tesla V100 GPUs, and eighteen Nvidia Tesla P100 GPUs. In the aspect of establishing the model, the CGMP and the comparison model are realized based on an open source deep learning tool PyTorch. In terms of experimental setup, Adam was chosen as the optimization algorithm for the experiment, the batch size was set to 256, and the number of iterations was set to 1000. In terms of performance evaluation, the experiment evaluated the sea surface temperature prediction performance of the model by calculating Mean Square Error (MSE) and Mean Absolute Error (MAE). The smaller the mean square error or mean absolute error obtained by the model is, the better the prediction performance of the model is represented. The detailed definitions of MSE and MAE are as follows:
wherein Y represents a predicted future SST sequence, Y 'represents a corresponding observed SST sequence, Y represents a predicted future SST, Y' represents a corresponding true value, and l represents the size of the prediction scale.
The sea surface temperature deep learning prediction method based on the space-time multidimensional influence disclosed by the invention is shown in figures 1-4 and comprises the following steps:
A. quantification of spatial influence: and quantifying the spatial influence on the target observation point by using the sea surface temperature data of the neighbor observation point of the target observation point, and constructing a multi-dimensional space-time sea surface temperature data set on the target sea area. The method comprises the following specific steps:
a1, the sea surface can be divided into individual grids according to longitude, latitude and resolution information. One grid is a sea surface temperature observation point. The surface temperature prediction task may be expressed simply as using historical surface temperature data to predict future surface temperature data at the target surface temperature observation. In order to quantify the spatial influence of the target sea surface temperature observation point quantitatively, the spatial influence of the target observation point is assumed to be totally from eight adjacent sea surface temperature observation points, called neighbor points, so that a multi-dimensional space-time sea surface temperature data set can be constructed on the Bohai sea and the south sea area, namely a Bohai sea surface temperature (daily average/weekly average/monthly average) complete data set, a Bohai sea surface temperature (daily average/weekly average/monthly average) marine data set, a south sea surface temperature (daily average/weekly average/monthly average) complete data set and a south sea surface temperature (daily average/weekly average/monthly average) marine data set. The complete data set contains data with missing spatial information, and the ocean data set excludes data with missing spatial information.
B. And (3) data completion processing: and performing data completion processing on the data set by using a neighbor data averaging method. The method comprises the following specific steps:
and B1, performing data completion processing on the complete data set by using a neighbor data averaging method. For some target observation points close to the coast, some of their neighbors may be located in the land area, and therefore there is no sea surface temperature observation data at these neighbors, which are called land sites. Therefore, in order to unify and standardize the input of the prediction model, this chapter performs data completion processing on the land points located in the land area. For a land site, the complemented sea surface temperature data is the average value of the real sea surface temperature observation data of the eight neighbor points, namely, a neighbor data averaging method. The data completion method has the advantages of strong practicability and suitability for all different deficiency conditions.
C. Establishing a prediction model: and (3) establishing a sea surface temperature deep learning prediction model based on the time-space multi-dimensional influence by combining deep learning technologies such as GRU, CNN and MLP, namely the multilayer perceptron CGMP of the convolution gating circulation unit. The method comprises the following specific steps:
c1, CGMP is an end-to-end sea surface temperature prediction model, the input of which is a historical SST sequence and the output of which is a predicted future SST sequence. Wherein the historical SST sequence comprises neighbor information and historical information of the observation point of the target sea surface temperature. The CGMP firstly processes the neighbor information of a target observation point on the spatial dimension through a convolution layer to obtain an intermediate vector, namely an empty SST sequence. The CGMP then processes the historical information of the spatio-temporal SST sequence in the time dimension through a GRU, resulting in a hidden state vector. Finally, the multi-layered perceptron maps the hidden state vector onto the prediction result, i.e., the future SST sequence. Convolutional layers are good at processing grid-like structure data and GRUs are good at processing sequence data, so the combinatorial collocation of convolutional layers and GRUs can sufficiently and effectively mine neighbor and historical information in the spatio-temporal dimension. Wherein:
c11, historical SST sequence X is a tensor with one dimension k × 3 × 3, and includes neighbor information (matrix with dimension 3 × 3) and historical information of k time units of the observation point of the target sea surface temperature. Since convolutional neural networks are good at handling grid-like structured data, CGMP first processes the neighbor information of the target observation point in the spatial dimension through one convolutional layer. The depth of the convolutional layer is 1, the size of the convolutional kernel K is 3 × 3, the step size is 1, and the padding is 0. After X is subjected to convolution operation, a space-time SST sequence S with the dimensionality of k multiplied by 1 is obtained, and the operation process is defined as follows:
S=(s1,s2,...,sk)
si=Xi★K+bs,i∈{1,2,...,k}
wherein s isiRepresenting the processed spatio-temporal history information, ★ representing a two-dimensional convolution operation, wk,1–wk,9Representing the weight of the convolution kernel K, bsIndicating the corresponding bias.
C12, the spatiotemporal SST sequence S contains historical information of k time units of the target sea surface temperature observation point. Since GRUs are good at processing sequence data, CGMP then processes the historical information of the target observation point in the time dimension through one GRU. The GRU receives two inputs at each time step: one is spatio-temporal history information S from a spatio-temporal SST sequence SiThe other is the hidden state vector H from the previous time step(i-1)Or an initial hidden state vector H0. After the history information is processed, the GRU generates a new hidden state vector H at each time stepi. So GRUs generate k hidden state vectors in total, i.e., H1–Hk. The process is defined as follows:
ri=σ(Wrssi+WrHH(i-1)+br)
zi=σ(Wzssi+WzHH(i-1)+bz)
ni=tanh(Wnssi+bns+ri*(WnHH(i-1)+bnH))
Hi=(1-zi)*ni+zi*H(i-1)
wherein HiRepresenting the hidden state vector, s, of the GRU at time step iiRepresenting spatio-temporal history information at time step i, H(i-1)Hidden state at time step i-1The state vector either represents the initial hidden state vector at time step 0, ri、ziAnd niRespectively representing the reset, update and new gates of the GRU. σ denotes a sigmoid function, and σ denotes a Hadamard product. W represents the corresponding weight and b represents the corresponding bias.
Finally, CGMP uses a multi-layer perceptron to hide the state vector HkMapped onto the prediction, i.e. the future SST sequence Y.
The mapping process is defined as follows:
Y=fMLP(WYHk+bY)
wherein, WYRepresenting the corresponding weight, bYIndicating the corresponding bias.
Referring to fig. 5 and fig. 6, the performance comparison of CGMP, NGMP, CLMP and NLMP on the bohai sea surface temperature daily mean and weekly mean complete data sets is shown. In the experiment, the convolution layer and the full link layer are compared in spatial information extraction performance, and the GRU and the LSTM are compared in temporal information extraction performance. Therefore, in addition to CGMP, there are three other corresponding comparison methods for experiments, namely Neighbor Full-Connected gated round-robin unit Multilayer Perceptron (NGMP), Convolutional long-short term memory Multilayer Perceptron (CLMP), and Neighbor Full-Connected long-short term memory Multilayer Perceptron (NLMP). Meanwhile, the influence of the depth of the multilayer perceptron on the prediction performance is also studied through experiments. Comparing the experimental results of the CGMP and the NGMP, it can be seen that under the same depth of the multilayer perceptron, both the MSE and the MAE of the CGMP are less than those of the NGMP, that is to say, the prediction performance of the CGMP on the Bohai sea surface temperature daily average complete data set is superior to that of the NGMP. This also indirectly demonstrates that convolutional layers have better ability to capture spatial influences in the spatial dimension than fully connected layers. Comparing the CGMP and the CLMP, the experimental result shows that under the same depth of the multilayer perceptron, both the MSE and the MAE of the CGMP are smaller than those of the CLMP, namely the prediction performance of the CGMP on a complete data set of the Bohai sea surface temperature daily average is superior to that of the CLMP. This also indirectly demonstrates that the GRU's ability to process historical information in the time dimension is superior to LSTM. By comparing the four methods together, the CGMP using the convolution layer and the GRU at the same time shows the optimal prediction performance on the depth of the multilayer sensor of 3. In summary, convolutional layers may capture spatial impact in the spatial dimension more efficiently than fully-connected layers, and GRUs may process historical information in the temporal dimension more efficiently than LSTM.
Referring to fig. 7 and 8, the CGMP, FC-LSTM and GED performance comparison on the bohai sea surface temperature integrity and marine data set is shown. As is evident from the table, the prediction performance of CGMP compared to FC-LSTM and GED remains optimal in all daily, weekly and monthly mean sea surface temperature predictions
Referring to FIGS. 9 and 10, a comparison of CGMP, FC-LSTM and GED performance on south sea ocean surface temperature integer and ocean datasets is shown. As is evident from the table, the prediction performance of CGMP compared to FC-LSTM and GED remains optimal in all daily, weekly and monthly mean sea surface temperature predictions.
In summary, the sea surface temperature deep learning prediction method based on the time-space multidimensional influence establishes a high-precision sea surface temperature prediction model by combining the time-space multidimensional influence and the deep learning technology. The model can overcome the performance bottleneck of the sea surface temperature deep learning prediction model based on the time influence, and can be widely applied to the sea surface temperature prediction of different sea areas and different scales.
The accompanying drawings and the detailed description are included to provide a further understanding of the invention. The method of the present invention is not limited to the examples described in the specific embodiments, and other embodiments derived from the method and idea of the present invention by those skilled in the art also belong to the technical innovation scope of the present invention. The description is not to be construed as limiting the invention.
Claims (7)
1. A sea surface temperature deep learning prediction method based on space-time multidimensional influence is characterized by comprising the following steps:
A. quantification of spatial influence: quantifying the space influence on the target observation point by using the sea surface temperature data of the neighbor observation point of the target observation point, and constructing a multi-dimensional space-time sea surface temperature data set on a target sea area;
B. and (3) data completion processing: performing data completion processing on the data set by using a neighbor data averaging method;
C. establishing a prediction model: a sea surface temperature deep learning prediction model based on space-time multidimensional influence is established by combining deep learning technologies such as convolution operation, a gated circulation unit, a multilayer perceptron and the like, namely a convolution gated circulation unit multilayer perceptron is called CGMP for short.
2. The sea surface temperature deep learning prediction method based on the spatiotemporal multidimensional influence is characterized by comprising the following steps of: in the step A, the sea surface is divided into grids according to longitude, latitude and resolution information, one grid is a sea surface temperature observation point, eight adjacent sea surface temperature observation points of the target observation point are called neighbor points, and a multi-dimensional space-time sea surface temperature data set of the nine points can be constructed in the target sea area.
3. The sea surface temperature deep learning prediction method based on the spatiotemporal multidimensional influence is characterized by comprising the following steps of: in the step B, the data of the missing point in the nine points is completed, the average value of the real sea surface temperature observation data of 8 neighbor points around the missing point is used as the missing point, and if the neighbor point of the missing point also has data missing, the point is excluded during calculation.
4. The sea surface temperature deep learning prediction method based on the space-time multidimensional influence is characterized by comprising the following steps of: in step C, the CGMP is an end-to-end sea surface temperature prediction model, the input of the model is a historical SST sequence, and the output of the model is a predicted future SST sequence, wherein the historical SST sequence includes historical sea surface temperature data of a target sea surface temperature observation point and neighbor points, the CGMP processes the neighbor information of the target observation point in a spatial dimension through a convolution layer to obtain an intermediate vector, i.e., a space SST sequence, the CGMP processes the historical information of the space SST sequence in a temporal dimension through a gated cycle unit to obtain a hidden state vector, and finally, the hidden state vector is mapped to a prediction result, i.e., the future SST sequence, by using a multi-layer sensor.
5. The sea surface temperature deep learning prediction method based on the space-time multidimensional influence is characterized in that: the historical SST sequence X is a tensor with one dimension of K multiplied by 3, and comprises neighbor information of K time units of a target sea surface temperature observation point and historical data information, the CGMP firstly processes the neighbor information of the target observation point on the spatial dimension through a convolution layer, the depth of the convolution layer is 1, the size of a convolution kernel K is 3 multiplied by 3, the step length is 1, and the filling is 0. After X is subjected to convolution operation, a space-time SST sequence S with the dimensionality of k multiplied by 1 is obtained, and the operation process is defined as follows:
S=(S1,S2,...,Sk)
si=Xi★K+bs,∈{1,2,...,k}
wherein s isiRepresenting the processed spatio-temporal history information, ★ representing a two-dimensional convolution operation, wk,1–wk,9Representing the weight of the convolution kernel K, bsIndicating the corresponding bias.
6. The sea surface temperature deep learning prediction method based on the spatiotemporal multidimensional influence is characterized by comprising the following steps of: a gated round robin unit, GRU, receives two inputs at each time step: one is spatio-temporal history information S from a spatio-temporal SST sequence SiThe other is the hidden state vector H from the previous time step(i-1)Or an initial hidden state vector H0History information is passed throughAfter processing, the GRU generates a new hidden state vector H at each time stepiThe GRUs generate a total of k hidden state vectors, i.e., H1–HkThe process is defined as follows:
ri=σ(Wrssi+WrHH(i-1)+br)
zi=σ(Wzssi+WzHH(i-1)+bz)
ni=tanh(Wnssi+bns+ri*(WnHH(i-1)+bnH))
Hi=(1-zi)*ni+zi*H(i-1)
wherein HiRepresenting the hidden state vector, s, of the GRU at time step iiRepresenting spatio-temporal history information at time step i, H(i-1)Representing the hidden state vector at time step i-1 or representing the initial hidden state vector at time step 0, ri、ziAnd niRespectively, a reset gate, an update gate and a new gate of the GRU, a represents a sigmoid function, a represents a Hadamard product, W represents a corresponding weight, and b represents a corresponding offset.
7. The sea surface temperature deep learning prediction method based on the space-time multidimensional influence is characterized by comprising the following steps: CGMP (China general packet radio service) enables hidden state vector H to pass through a multi-layer perceptronkMapping onto the prediction, i.e. future SST sequence Y, the mapping process is defined as follows:
Y=fMLP(WYHk+bY)
wherein, WYRepresenting the corresponding weight, bYIndicating the corresponding bias.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010461315.1A CN111680784B (en) | 2020-05-27 | 2020-05-27 | Sea surface temperature deep learning prediction method based on space-time multidimensional influence |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010461315.1A CN111680784B (en) | 2020-05-27 | 2020-05-27 | Sea surface temperature deep learning prediction method based on space-time multidimensional influence |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111680784A true CN111680784A (en) | 2020-09-18 |
CN111680784B CN111680784B (en) | 2023-10-24 |
Family
ID=72434400
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010461315.1A Active CN111680784B (en) | 2020-05-27 | 2020-05-27 | Sea surface temperature deep learning prediction method based on space-time multidimensional influence |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111680784B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112465203A (en) * | 2020-11-19 | 2021-03-09 | 中国石油大学(华东) | Sea level height intelligent prediction and forecast system based on gate control circulation unit neural network, computer equipment and storage medium |
CN117633712A (en) * | 2024-01-24 | 2024-03-01 | 国家卫星海洋应用中心 | Sea level height data fusion method, device and equipment based on multi-source data |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180336460A1 (en) * | 2017-05-22 | 2018-11-22 | Sap Se | Predicting wildfires on the basis of biophysical indicators and spatiotemporal properties using a convolutional neural network |
US20180336452A1 (en) * | 2017-05-22 | 2018-11-22 | Sap Se | Predicting wildfires on the basis of biophysical indicators and spatiotemporal properties using a long short term memory network |
CN109685767A (en) * | 2018-11-26 | 2019-04-26 | 西北工业大学 | A kind of bimodal brain tumor MRI dividing method based on Cluster-Fusion algorithm |
CN109754605A (en) * | 2019-02-27 | 2019-05-14 | 中南大学 | A kind of traffic forecast method based on attention temporal diagram convolutional network |
CN110176309A (en) * | 2019-05-28 | 2019-08-27 | 上海大学 | It is a kind of for predicting the medical data processing method of cardiovascular disease |
CN110321603A (en) * | 2019-06-18 | 2019-10-11 | 大连理工大学 | A kind of depth calculation model for Fault Diagnosis of Aircraft Engine Gas Path |
-
2020
- 2020-05-27 CN CN202010461315.1A patent/CN111680784B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180336460A1 (en) * | 2017-05-22 | 2018-11-22 | Sap Se | Predicting wildfires on the basis of biophysical indicators and spatiotemporal properties using a convolutional neural network |
US20180336452A1 (en) * | 2017-05-22 | 2018-11-22 | Sap Se | Predicting wildfires on the basis of biophysical indicators and spatiotemporal properties using a long short term memory network |
CN109685767A (en) * | 2018-11-26 | 2019-04-26 | 西北工业大学 | A kind of bimodal brain tumor MRI dividing method based on Cluster-Fusion algorithm |
CN109754605A (en) * | 2019-02-27 | 2019-05-14 | 中南大学 | A kind of traffic forecast method based on attention temporal diagram convolutional network |
CN110176309A (en) * | 2019-05-28 | 2019-08-27 | 上海大学 | It is a kind of for predicting the medical data processing method of cardiovascular disease |
CN110321603A (en) * | 2019-06-18 | 2019-10-11 | 大连理工大学 | A kind of depth calculation model for Fault Diagnosis of Aircraft Engine Gas Path |
Non-Patent Citations (3)
Title |
---|
CHANGJIANG XIAO: "A spatiotemporal deep learning model for sea surface temperature field prediction using time-series satellite data" * |
YUTING YANG等: "A CFCC-LSTM Model for Sea Surface Temperature Prediction" * |
姜英超: "基于深度学习的海表温度遥感反演模型" * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112465203A (en) * | 2020-11-19 | 2021-03-09 | 中国石油大学(华东) | Sea level height intelligent prediction and forecast system based on gate control circulation unit neural network, computer equipment and storage medium |
CN117633712A (en) * | 2024-01-24 | 2024-03-01 | 国家卫星海洋应用中心 | Sea level height data fusion method, device and equipment based on multi-source data |
CN117633712B (en) * | 2024-01-24 | 2024-04-19 | 国家卫星海洋应用中心 | Sea level height data fusion method, device and equipment based on multi-source data |
Also Published As
Publication number | Publication date |
---|---|
CN111680784B (en) | 2023-10-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110705457B (en) | Remote sensing image building change detection method | |
Li et al. | Missing value imputation for traffic-related time series data based on a multi-view learning method | |
Braakmann-Folgmann et al. | Sea level anomaly prediction using recurrent neural networks | |
CN111680784B (en) | Sea surface temperature deep learning prediction method based on space-time multidimensional influence | |
CN110796634A (en) | Dam state detection method and device, computer equipment and readable storage medium | |
Jin et al. | Neu-nbv: Next best view planning using uncertainty estimation in image-based neural rendering | |
CN114399073A (en) | Ocean surface temperature field prediction method based on deep learning | |
Kim et al. | Scalable learning with a structural recurrent neural network for short-term traffic prediction | |
Chen et al. | WSN sampling optimization for signal reconstruction using spatiotemporal autoencoder | |
Battrawy et al. | Rms-flownet: Efficient and robust multi-scale scene flow estimation for large-scale point clouds | |
Abubakr et al. | Application of deep learning in damage classification of reinforced concrete bridges | |
Xie et al. | An evolving sea surface temperature predicting method based on multidimensional spatiotemporal influences | |
Saffari et al. | Robust wind speed forecasting: A deep spatio-temporal approach | |
Baydaroğlu et al. | Temporal and spatial satellite data augmentation for deep learning-based rainfall nowcasting | |
Wang et al. | Optimized lightweight CA-transformer: Using transformer for fine-grained visual categorization | |
CN114399628A (en) | Insulator high-efficiency detection system under complex space environment | |
CN114676887A (en) | River water quality prediction method based on graph convolution STG-LSTM | |
Duan et al. | Deep spatio-temporal attention model for grain storage temperature forecasting | |
Champenois et al. | Real-time reconstruction of 3D ocean temperature fields from reanalysis data and satellite and buoy surface measurements | |
CN117909719B (en) | Terminal internet of things equipment monitoring and early warning method and system based on autonomous learning algorithm | |
Sergeev et al. | On the hybrid models of soil contaminants concentrations predicting in subarctic region | |
Wang et al. | Remote Sensing Image Prediction of Water Environment Based on 3D CNN and ConvLSTM | |
Ma et al. | application of a combinatorial DIC algorithm in sub-pixel displacement measurement | |
CN116484921B (en) | Crack size accurate quantification method for multi-physical-quantity feature fusion convolutional neural network | |
CN117575111B (en) | Agricultural remote sensing image space-time sequence prediction method based on transfer learning |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |