CN116306318A - Three-dimensional ocean thermal salt field forecasting method, system and equipment based on deep learning - Google Patents
Three-dimensional ocean thermal salt field forecasting method, system and equipment based on deep learning Download PDFInfo
- Publication number
- CN116306318A CN116306318A CN202310530474.6A CN202310530474A CN116306318A CN 116306318 A CN116306318 A CN 116306318A CN 202310530474 A CN202310530474 A CN 202310530474A CN 116306318 A CN116306318 A CN 116306318A
- Authority
- CN
- China
- Prior art keywords
- dimensional
- data
- model
- salt
- liquid crystal
- 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
- 150000003839 salts Chemical class 0.000 title claims abstract description 104
- 238000013135 deep learning Methods 0.000 title claims abstract description 37
- 238000013277 forecasting method Methods 0.000 title claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 45
- 238000012795 verification Methods 0.000 claims abstract description 41
- 230000008859 change Effects 0.000 claims abstract description 18
- 238000010586 diagram Methods 0.000 claims abstract description 14
- 238000013136 deep learning model Methods 0.000 claims abstract description 13
- 238000012805 post-processing Methods 0.000 claims abstract description 8
- 238000005259 measurement Methods 0.000 claims abstract description 6
- 239000004973 liquid crystal related substance Substances 0.000 claims description 42
- 230000003935 attention Effects 0.000 claims description 34
- 238000000034 method Methods 0.000 claims description 31
- 239000013598 vector Substances 0.000 claims description 20
- 238000007781 pre-processing Methods 0.000 claims description 14
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 13
- 230000006870 function Effects 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 238000012800 visualization Methods 0.000 claims description 5
- 238000004458 analytical method Methods 0.000 claims description 4
- 238000004590 computer program Methods 0.000 claims description 4
- 230000017105 transposition Effects 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 3
- 241000282414 Homo sapiens Species 0.000 claims description 2
- 230000003213 activating effect Effects 0.000 claims description 2
- 230000001186 cumulative effect Effects 0.000 claims description 2
- 238000005315 distribution function Methods 0.000 claims description 2
- 230000010332 selective attention Effects 0.000 claims description 2
- 238000010792 warming Methods 0.000 claims 1
- 230000009977 dual effect Effects 0.000 abstract description 5
- 230000007246 mechanism Effects 0.000 description 7
- 239000013535 sea water Substances 0.000 description 7
- 230000007613 environmental effect Effects 0.000 description 6
- 230000002123 temporal effect Effects 0.000 description 5
- 238000011160 research Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 238000003491 array Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000009499 grossing Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013079 data visualisation Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000008020 evaporation Effects 0.000 description 1
- 238000001704 evaporation Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 239000013505 freshwater Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 239000002352 surface water Substances 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/08—Thermal analysis or thermal optimisation
-
- 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)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computer Hardware Design (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- Geometry (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention belongs to the technical field of marine environment specific variable measurement, and discloses a three-dimensional marine temperature salt field forecasting method, a three-dimensional marine temperature salt field forecasting system and three-dimensional marine temperature salt field forecasting equipment based on deep learning, wherein a deep learning model training data set is constructed; constructing a future underwater three-dimensional Wen Yanchang forecasting model; performing multi-angle verification on the three-dimensional Wen Yanchang forecasting result by using measured values such as a data set or a buoy which are not trained by a model; and (3) carrying out post-processing on output data of the model, visualizing predicted ocean temperature and salinity in a two-dimensional thermodynamic diagram, an isotherm and a temperature-salt depth change curve, and calculating an error. According to the invention, a plurality of deep learning modules are integrated, the dual characteristics of time and space of three-dimensional warm salt calendar are fully considered, and the prediction accuracy is improved; the invention takes the sea surface environment data as input, has low data cost and is easy to obtain, and the practical application value is greatly improved.
Description
Technical Field
The invention belongs to the technical field of ocean engineering key supporting equipment and systems, and particularly relates to a three-dimensional ocean thermal salt field forecasting method, system and equipment based on deep learning.
Background
Seawater temperature and salinity are one of the most basic marine environmental parameters, and are important physical quantities for describing the properties of seawater. Some researches show that the rise of sea temperature can cause the rise of typhoon intensity, flood disasters, storm surge and other extreme weather occurrence frequency. Ocean salinity is a key factor affecting the marine power environment and the interaction of the sea. The salinity has very remarkable influence on the thermodynamic and dynamic processes in the ocean and is one of the driving factors of ocean hot salt circulation. Therefore, grasping the underwater temperature structure is an urgent need in the fields of marine disaster prevention and reduction, marine rights maintenance, marine ecological protection, marine area use management, marine law enforcement and supervision, marine disaster and emergency observation and the like, and is also a necessary way for researching physical marine phenomena such as jump layers, mesoscale vortex, internal waves and the like.
The current warm salt field forecast based on deep learning can be divided into sea surface warm salt data forecast and underwater three-dimensional warm salt field forecast. The method is mostly based on LSTM and CNN model architecture, and the underwater Wen Yanchang at the current moment or for a period of time in the future is forecasted through three-dimensional warm salt time calendar data. The sea surface salt temperature data cannot meet the needs of underwater research. However, the current prediction of Wen Yanchang under water is mostly performed only for the current time under water Wen Yanchang, and the environmental change of the future under water cannot be known in advance. The method for forecasting the temperature and salt in a future period of time is characterized in that the data required for forecasting is the time history of three-dimensional environment data of the underwater in the past, the forecast data is the month-to-month forecast of the temperature and salt in the future, the forecast is not fine in the time dimension, and the method has the defects of high cost and limitation of training data in practical application.
Through the above analysis, the problems and defects existing in the prior art are as follows:
(1) The prediction or simulation of sea surface temperature and salt data cannot know the condition of the underwater environment, and the requirement of underwater research is not met.
(2) At present, for Wen Yanchang prediction under water, most of the prediction is performed only on the underwater Wen Yanchang at the current moment, and environmental changes of the underwater in the future cannot be known in advance.
(3) Aiming at the method for forecasting the temperature and salt in a future period, the data required by forecasting is the three-dimensional environmental data time history under the water in the past, and the cost of training the data is high in practical application.
(4) The method for forecasting the warm salt in a future period has the advantages that forecast data are month-to-month warm salt data, time resolution is low, and practical application is limited.
Disclosure of Invention
Aiming at the problems existing in the prior art, the invention provides a three-dimensional ocean thermal salt field forecasting method, system and equipment based on deep learning.
The invention discloses a three-dimensional ocean thermal salt field forecasting method based on deep learning, which comprises the following steps:
firstly, constructing a deep learning model training data set, and preprocessing original ocean three-dimensional warm salt data and ocean surface data to enable the model training data set to meet the input requirements of a model;
step two, constructing an underwater three-dimensional Wen Yanchang forecasting model, wherein the model comprises 5 modules: a 3-dimensional convolution module, an attention module, a position coding module, a residual error module and a 3-dimensional transposition convolution module;
step three, performing multi-angle verification on the three-dimensional Wen Yanchang forecasting result by using measured values such as a data set or a buoy which are not trained by a model;
and fourthly, carrying out post-processing on output data of the model, visualizing predicted ocean temperature and salinity in a two-dimensional thermodynamic diagram, an isotherm and a temperature-salt depth change curve, and calculating an error.
Further, the first step specifically includes:
(1.1) acquiring marine environment data for many years through satellites, buoys, numerical modes and analysis data, wherein the marine environment data comprise sea surface temperature, sea surface salinity, sea surface north-south flow velocity, sea surface height and underwater 5500m internal temperature salt data;
(1.2) bilinear interpolation, setting a pixel point
Four points around it ∈ ->
，/>
，/>
，/>
The corresponding grey values are +.>
，/>
，/>
，/>
And->
Then->
；
(1.3) Min-Max normalization, specifically expressed as:
wherein, the liquid crystal display device comprises a liquid crystal display device,
for inputting data +.>
For outputting data;
(1.4) land and outlier handling, replacing all land and outliers with 0;
(1.5) uniformly converting the spatial resolution of the original data into 0.083 DEG, the temporal resolution into 3h and the depth of 0-5500m into 1 day and the depth of 0-1000m;
(1.6) dividing the sea area data to be forecasted into a plurality of sub-sea area data.
And (1.7) processing the processed sea surface data and the three-dimensional underwater salt temperature data into 5-dimensional matrixes (BxT x V x H x W, bxT x D x H x W) required by the model respectively, wherein B is the number of batches, V is the number of sea surface variables, T is the number of samples of a time sequence, D is a depth range, H is a latitude range and W is a longitude range.
Further, the underwater three-dimensional Wen Yanchang forecasting model is developed based on a Pytorch framework.
Further, the second step specifically includes:
(2.1) convolving the input multidimensional marine surface environment data by Conv3D, wherein the size of a convolution kernel in the 3D convolution is (2, 10, 10), and the step size is (1, 1, 1);
(2.2) flattening the input data into a 1-dimensional vector after 3D convolution, and adding a position code after flattening, wherein the specific expression of the position code is as follows:
and (2.3) extracting the mapping relation of the data of different positions on the sea surface to the influence under water by adopting an attention mechanism, wherein the attention mechanism module comprises four layers of networks, each layer of network comprises six attention heads, and the specific expression of the multi-head attention is as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,
is a trainable matrix>
Outputting a result for a single attention;
(2.4) adding a residual module before and after the attention module, specifically expressed as:
wherein, the liquid crystal display device comprises a liquid crystal display device,
is depth +.>
Is (are) unit features of->
Is depth +.>
Is (are) unit features of->
Weights transferred to the unit;
(2.5) recovering the size using a transposed convolution in which the size of the convolution kernel is (4, 10, 10) and the step size is (2, 1, 1);
and (2.6) selecting a GELU optimizer, and continuously optimizing the super parameters of the model by taking the average absolute error as a loss function to minimize the error.
Further, the multi-angle verification comprises early forecasting time length verification, vertical verification, horizontal verification and actual measurement verification.
Further, the three-dimensional Wen Yanchang forecasting result is subjected to multi-angle verification, and the forecasting precision is evaluated through two indexes, which are respectively:
(3.1) forecasting Root Mean Square Error (RMSE) of the thermal salt field and the open source dataset thermal salt field:
(3.2) average absolute percent error (MAPE) of the forecast temperature salt field and the open source dataset temperature salt field:
wherein, the liquid crystal display device comprises a liquid crystal display device,
and->
Is a predicted value and a true value.
Further, the output data post-processing includes:
(4.1) changing the corresponding position of the output of the model to nan by three-dimensional topographic data before training;
(4.2) interpolating ocean temperature/salt at each depth to a single point through bilinear interpolation to form temperature/salt data at different depths at the longitude and latitude point;
and (4.3) cutting the three-dimensional temperature/salt data output by the model along the warp and weft respectively to obtain a temperature/salt section on a vertical plane.
Another object of the present invention is to provide a three-dimensional marine thermal salt field prediction system based on deep learning, the three-dimensional marine thermal salt field prediction system based on deep learning comprising:
the input module is used for constructing a training data set of the deep learning model and preprocessing the training data set to enable the training data set to meet the input requirement of the model;
the training module is used for training a future underwater three-dimensional Wen Yanchang forecasting model;
the verification module is used for multi-angle verification of the three-dimensional Wen Yanchang forecasting result;
and the visualization module is used for visualizing the predicted ocean temperature and salinity according to a two-dimensional thermodynamic diagram, an isotherm and a temperature and salt depth change curve.
It is a further object of the present invention to provide a computer device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of the deep learning based three-dimensional sea temperature salt field forecasting method.
In combination with the above technical solution and the technical problems to be solved, please analyze the following aspects to provide the following advantages and positive effects: (1) According to the invention, a plurality of deep learning modules are integrated, the dual characteristics of time and space of three-dimensional warm salt calendar are fully considered, and the prediction accuracy is improved. (2) The invention merges the attention mechanism, gives different attention to the ocean three-dimensional space and improves the generalization of the model. (3) The invention merges the residual network and prevents the situation of reducing the model forecasting precision caused by the phenomenon of overfitting when the model depth is deeper. (4) According to the invention, the large sea area is partitioned, so that the demand of model training on the video memory of the display card is greatly reduced, and only the forecast data of the target sub sea area can be output according to the demand after the training is finished, thereby improving the practicability of the model. (5) The method provided by the invention takes the sea surface environment data as input, has low data cost and is easy to obtain, and the practical application value is greatly improved.
According to the invention, by utilizing various deep learning modules, the time and space characteristics of the three-dimensional thermal salt are fully extracted, and the marine subsurface thermal salt of 1000m in the future 15 days can be accurately predicted. The method can well predict the ocean temperatures of different advance time lengths, different depths and different positions, and effectively improves the prediction accuracy. The invention can provide more accurate and reliable data support for the prediction of ocean weather events, thereby being beneficial to guaranteeing the life and property safety of ocean economy and people, knowing the trend and influence of climate change, and better understanding and exploring the problems in aspects of physics, chemistry, biology and the like of the ocean.
The invention provides a novel deep learning method for predicting marine subsurface temperature salt for 15 days in the future, and the method fully considers the dual characteristics of time and space. The input data are sea surface temperature, sea surface salinity, sea surface altitude, sea surface flow rate for the past 15 days. The model can accurately forecast the ocean temperature and salt at different advanced time lengths, different depths and different positions, and can provide powerful data support for ocean related events, climate change research and various underwater operation tasks.
At present, for Wen Yanchang prediction under water, most of the prediction is performed only on the underwater Wen Yanchang at the current moment, and environmental changes of the underwater in the future cannot be known in advance. The invention provides a novel deep learning method for predicting the marine subsurface temperature salt for 15 days in the future, which fully considers the dual characteristics of time and space and can accurately predict the marine subsurface temperature salt for 15 days in the future under water of 1000m.
Aiming at the method for forecasting the temperature and salt in a period of time in the future, the data required for forecasting are the time histories of the three-dimensional environment data under the water in the past, and the method has the defect of high cost of training data in practical application.
Aiming at the higher requirement of the integrated deep learning model on the display memory of the display card, the invention divides the large sea area into a plurality of sub sea areas, each sub sea area has independent weight, reduces the cost and difficulty of model training, simultaneously can forecast one or a plurality of sub sea area data according to the actual requirement, does not need to output the ocean temperature and salt data of the whole sea area, and reduces the time required by forecasting.
The three-dimensional salt temperature field forecasting method based on the integrated deep learning model can accurately forecast the sea subsurface salt temperature of 1000m in the future for 15 days. Compared with the traditional method, the method fully considers the dual characteristics of time and space, can well predict the ocean temperatures of different advance durations, different depths and different positions, and effectively improves the prediction accuracy.
Drawings
Fig. 1 is a flowchart of a three-dimensional ocean thermal salt field forecasting method based on deep learning according to an embodiment of the invention.
Fig. 2 is a block diagram of a three-dimensional ocean thermal salt field prediction system based on deep learning according to an embodiment of the present invention.
Fig. 3 is a schematic diagram of an overall workflow of a three-dimensional ocean thermal salt field forecasting method based on deep learning according to an embodiment of the invention.
Fig. 4 is a schematic diagram of an internal structure of a three-dimensional Wen Yanchang forecasting model of a sea according to an embodiment of the present invention.
The following steps are sequentially carried out from left to right in the figure: input data of the model, 3D convolution, position coding, attention, residual network, activation function GELU, regularized DropOut, 3D transpose convolution, output data of the model.
Detailed Description
The present invention will be described in further detail with reference to the following examples in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
As shown in fig. 1, the three-dimensional ocean thermal salt field prediction method based on deep learning provided by the embodiment of the invention comprises the following steps:
s101, constructing a deep learning model training data set, and preprocessing original ocean three-dimensional warm salt data and ocean surface data for many years to enable the model training data set to meet the input requirements of a model;
s102, constructing a future underwater three-dimensional Wen Yanchang forecasting model, wherein the model comprises 5 modules: a 3-dimensional convolution module, an attention module, a position coding module, a residual error module and a 3-dimensional transposition convolution module;
s103, performing multi-angle verification on the three-dimensional Wen Yanchang forecasting result by using measured values such as a data set or a buoy which are not trained by a model;
and S104, visualizing the predicted ocean temperature and salinity in a two-dimensional thermodynamic diagram, an isotherm and a temperature-salt depth change curve, and calculating an error.
As shown in fig. 2, the three-dimensional ocean thermal salt field prediction system based on deep learning provided by the embodiment of the invention comprises:
the input module is used for constructing a training data set of the deep learning model and preprocessing the training data set to enable the training data set to meet the input requirement of the model;
the training module is used for training a future underwater three-dimensional Wen Yanchang forecasting model;
the verification module is used for multi-angle verification of the three-dimensional Wen Yanchang forecasting result;
and the visualization module is used for visualizing the predicted ocean temperature and salinity according to a two-dimensional thermodynamic diagram, an isotherm and a temperature and salt depth change curve.
In the embodiment of the invention, the three-dimensional ocean thermal salt field prediction system based on deep learning provided by the invention can adopt the following specific scheme:
1. an input module: the module is mainly responsible for constructing a deep learning model training data set and preprocessing data so as to enable the data to meet the input requirements of the model. The method comprises the following specific steps:
and (3) data acquisition: collecting ocean temperature and salt field data from various sensors and observation equipment;
data preprocessing: preprocessing the acquired data, including data cleaning, denoising, interpolation, smoothing and the like;
dividing data: the preprocessed data set is divided into a training set, a validation set and a test set. The training set is used for training the model, the verification set is used for model parameter adjustment and model performance verification, and the test set is used for final model evaluation.
2. Training module: the module is mainly responsible for training a future underwater three-dimensional Wen Yanchang forecasting model. The method comprises the following specific steps:
model selection: deep learning models suitable for forecasting the ocean temperature and salt field, such as convolutional neural networks, cyclic neural networks, converters and the like, are selected;
model construction: constructing a deep learning model, including defining a model architecture, selecting a loss function, an optimizer and the like;
model training: training the model by using the training set, and optimizing model parameters;
model preservation: the trained model is saved for subsequent prediction and verification.
3. And (3) a verification module: the module is mainly used for multi-angle verification of three-dimensional Wen Yanchang forecasting results so as to evaluate the performance and reliability of the model. The method comprises the following specific steps:
data preprocessing: preprocessing the verification data to enable the verification data to meet the input requirement of the model;
prediction data: predicting the verification data by using the trained model;
evaluation of results: the prediction results are evaluated, including calculating prediction errors, generating error distribution maps, calculating correlation coefficients, and the like.
4. And a visualization module: the module is mainly used for visualizing the predicted ocean temperature and salinity in the forms of a two-dimensional thermodynamic diagram, an isotherm, a temperature and salt depth change curve and the like. The method comprises the following specific steps:
data preprocessing: post-processing the prediction result, including reduction, smoothing and the like;
data visualization: the processed data is visualized in the form of a two-dimensional thermodynamic diagram, an isotherm, a temperature and salt depth change curve and the like by using a visualization tool, so that a user can more intuitively know the change condition of the ocean temperature and salt field.
In the embodiment of the invention, the three-dimensional ocean thermal salt field forecasting method based on deep learning provided by the embodiment of the invention specifically comprises the following steps:
(1) Constructing a deep learning model training data set, and preprocessing original ocean three-dimensional temperature salt data and ocean surface data for many years to enable the model training data set to meet the input requirements of the model;
(2) Constructing a future underwater three-dimensional Wen Yanchang forecasting model, wherein the model comprises 5 modules: the input of the model is 5-dimensional tensor (BxT x V x H x W), the output is 5-dimensional tensor (BxT x D x H x W), B is batch, T is time, V is variable, D is depth, H is longitude, and W is latitude;
(3) Performing multi-angle verification on the three-dimensional Wen Yanchang forecasting result by using measured values such as a data set or a buoy which are not trained by a model;
(4) And (3) carrying out post-processing on output data of the model, visualizing predicted ocean temperature and salinity in a two-dimensional thermodynamic diagram, an isotherm and a temperature-salt depth change curve, and calculating an error.
The first step provided by the embodiment of the invention specifically comprises the following steps:
(1.1) acquiring marine environment data for many years through satellites, buoys, numerical modes and analysis data, wherein the marine environment data comprises sea surface temperature, sea surface salinity, sea surface north-south flow velocity, sea surface height and underwater 5500m internal temperature salt data.
(1.2) bilinear interpolation, setting a pixel point
Four points around it ∈ ->
，/>
，/>
，/>
The corresponding grey values are +.>
，/>
，/>
，/>
And->
Then->
。
(1.3) Min-Max normalization, wherein all data ranges after the Min-Max normalization are between [0, 1], and the Min-Max normalization is specifically expressed as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,
for inputting data +.>
For outputting data.
(1.4) land and outlier handling, land will be replaced with nan values, outliers will be replaced with-999, for example, and all land and outliers will be replaced with 0 in order not to affect model training.
And (1.5) taking the space-time change relation of the obtained warm salt and the actual application requirement into consideration, uniformly converting the spatial resolution of the original data into the temporal resolution of 1 day and the depth of 0-1000m, wherein the temporal resolution is 0.083 degrees, the temporal resolution is 3h and the depth of 0-5500 m.
(1.6) dividing the sea area data to be forecasted into a plurality of sub-sea area data.
And (1.7) processing the processed sea surface data and the three-dimensional underwater salt temperature data into 5-dimensional matrixes (BxT x V x H x W, bxT x D x H x W) required by the model respectively, wherein B is the number of batches, V is the number of sea surface variables, T is the number of samples of a time sequence, D is a depth range, H is a latitude range and W is a longitude range.
The future underwater three-dimensional Wen Yanchang forecasting model provided by the embodiment of the invention is developed based on a Pytorch framework.
The second step provided by the embodiment of the invention specifically comprises the following steps:
(2.1) the input multidimensional marine surface environmental data is convolved by Conv3D, the 3D convolution being changed on the basis of the 2D convolution. For 2D convolution, the same filter output is a two-dimensional feature map, and the information of multiple channels is fully compressed, so that the information on the time sequence cannot be captured well. The output of the 3D convolution is still a 3D feature map, enabling better capture of data spatial and temporal features. The three-dimensional convolution is specifically expressed as:
the size of the convolution kernel in the 3D convolution is (2, 10, 10), and the step size is (1, 1, 1).
(2.2) flattening the input data into a 1-dimensional vector after 3D convolution, and adding position codes after flattening in order to prevent the position information from being lost after data flattening, wherein the specific expression of the position codes is as follows:
(2.3) because the influence of the data of different positions on the sea surface on the water is different, the invention adopts an attention mechanism to extract the mapping relation. The attention mechanism simulates the selective attention of human beings, more key information is gradually selected in the training process, different weights are given according to the importance degree of different information, and therefore data characteristics are better learned. The specific expression of the attention mechanism is as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,
for inputting vectors, ++>
、/>
、/>
Is a trainable weight matrix, +.>
Is->
Square root of the dimension of (c).
The attention mechanism module provided by the embodiment of the invention comprises four layers of networks, each layer of network comprises six attention heads, and the specific expression of the multi-head attention is as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,
is a trainable matrix>
The results are output for a single attention.
(2.4) in order to prevent the model from being over-fitted when using multi-head attention, a residual module is added before and after the attention module. The residual network is used for solving the phenomenon of gradient dispersion or gradient explosion when the hidden layers of the neural network are too many. For a residual function
In other words, in the case of the gradient of the function pair x by the back-propagation chain law, the gradient is due to +.>
The derivative is 1, so the formula ensures that gradient disappearance does not occur in the propagation process. The specific expression of the residual network is:
wherein, the liquid crystal display device comprises a liquid crystal display device,
is depth +.>
Is (are) unit features of->
Is depth +.>
Is (are) unit features of->
For the weight transferred to the unit.
(2.5) the matrix edges are not filled when the input data is subjected to 3D convolution, so that the size will change according to the size of the convolution kernel during the convolution, for which we use transposed convolution to recover the size, the size of the convolution kernel in 3D transposed convolution is (4, 10, 10), and the step size is (2, 1, 1). Compared with other methods for recovering the size, the transposed convolution can be regarded as an inverse process of the convolution, has a learnable parameter, and can acquire an optimal up-sampling mode through training. The transposed convolution is specifically expressed as:
wherein, the liquid crystal display device comprises a liquid crystal display device,
for outputting the vector +.>
For inputting vectors, ++>
Is a convolution kernel matrix.
(2.6) selecting GELThe U optimizer takes the average absolute error as a loss function to continuously optimize the super parameters of the model so as to minimize the error,
specifically expressed as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,
for activating the input vector of the function, +.>
Is a gaussian cumulative distribution function.
The multi-angle verification provided by the embodiment of the invention comprises early forecasting time length verification, vertical verification, horizontal verification and actual measurement verification.
The embodiment of the invention provides a multi-angle verification method for a three-dimensional Wen Yanchang forecasting result, which evaluates forecasting accuracy through two indexes, wherein the first index is Root Mean Square Error (RMSE) of a forecasting temperature salt field and an open source data set temperature salt field:
the second index is the Mean Absolute Percentage Error (MAPE) of the forecast temperature salt field and the open source dataset temperature salt field:
wherein, the liquid crystal display device comprises a liquid crystal display device,
and->
Is a predicted value and a true value.
The post-processing of the output data provided by the embodiment of the invention comprises the following steps:
(3.1) changing the corresponding position of the output of the model to nan by three-dimensional topographic data before training;
and (3.2) interpolating the ocean temperature/salt at each depth to a single point through bilinear interpolation to form temperature/salt data at different depths at the longitude and latitude point. The bilinear interpolation is specifically expressed as:
pixel point setting
Four points around it ∈ ->
，/>
，/>
，/>
The corresponding gray values are respectively
，/>
，/>
，/>
And->
Then->
；
And (3.3) cutting the three-dimensional temperature/salt data output by the model along the warp and weft respectively to obtain a temperature/salt section on a vertical plane.
As shown in fig. 3, the overall workflow of the embodiment of the present invention is: preprocessing original sea surface environment data to meet the input requirement of a model, sequentially passing through a 3D convolution module, a position coding, an attention module, a residual module and a transposition convolution module to extract time and space characteristics, finally obtaining underwater three-dimensional warm salt data, comparing the warm salt data output by the model with actual warm salt data, thereby obtaining a forecast error at the time, continuously optimizing super parameters of the model according to the change of the forecast error, and finally obtaining an optimal warm salt forecast model.
In order to prove the inventive and technical value of the technical solution of the present invention, this section is an application example on specific products or related technologies of the claim technical solution.
The three-dimensional ocean thermal salt field forecasting method based on the deep learning, which is provided by the application embodiment of the invention, is applied to computer equipment, wherein the computer equipment comprises a memory and a processor, the memory stores a computer program, and the computer program, when being executed by the processor, enables the processor to execute the steps of the three-dimensional ocean thermal salt field forecasting method based on the deep learning.
The three-dimensional ocean thermal salt field forecasting method based on the deep learning, which is provided by the application embodiment, is applied to an information data processing terminal, and the information data processing terminal is used for realizing the three-dimensional ocean thermal salt field forecasting system based on the deep learning.
TABLE 1 error in prediction of underwater temperature
Table 1 is the prediction errors of the underwater temperature under different advance prediction durations and depths, and for convenience of presentation, the table only contains the prediction errors under the advance durations of 5, 10, 15 days and 10 depths. According to experimental results, the average MAPE for forecasting the underwater temperature is lower than 2%, and the RMSE is lower than 0.4, so that excellent forecasting precision is shown. From the aspect of the early forecasting time, the error keeps stable along with the increase of the early forecasting time, and the precision is not obviously reduced. From the prediction effect of different depths, as the thermocline of a selected sea area is positioned at 50-200 m, the underwater temperature at the thermocline can change sharply, the prediction difficulty is increased, the prediction precision of the thermocline is slightly lower than that of other depths, and the error is still lower than 3%. The invention fully illustrates the good underwater temperature forecasting performance.
TABLE 2 error in salinity forecast under water
Table 2 is the prediction error of the salinity under water at different lengths and depths of advance prediction, and for convenience of presentation, the table only contains the prediction errors at the lengths of advance of 5, 10, 15 days and 10 depths. According to experimental results, the average MAPE for forecasting the underwater salinity is lower than 1%, and the RMSE is lower than 0.3, so that excellent forecasting precision is shown. From the aspect of the early forecast time, the error slightly rises along with the increase of the early forecast time, and the accuracy is not obviously reduced. From the forecast effect of different depths, the salinity of the shallow sea water is mainly influenced by the evaporation of the surface water body and the input of fresh water, and the salinity of the deep sea water is mainly influenced by the pressure of the sea water, so that the salinity forecast precision of the surface sea water is slightly lower than that of the deep sea water, but is at a high precision level. The invention fully demonstrates the good underwater salinity forecasting performance.
It should be noted that the embodiments of the present invention can be realized in hardware, software, or a combination of software and hardware. The hardware portion may be implemented using dedicated logic; the software portions may be stored in a memory and executed by a suitable instruction execution system, such as a microprocessor or special purpose design hardware. Those of ordinary skill in the art will appreciate that the apparatus and methods described above may be implemented using computer executable instructions and/or embodied in processor control code, such as provided on a carrier medium such as a magnetic disk, CD or DVD-ROM, a programmable memory such as read only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The device of the present invention and its modules may be implemented by hardware circuitry, such as very large scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., as well as software executed by various types of processors, or by a combination of the above hardware circuitry and software, such as firmware.
The foregoing is merely illustrative of specific embodiments of the present invention, and the scope of the invention is not limited thereto, but any modifications, equivalents, improvements and alternatives falling within the spirit and principles of the present invention will be apparent to those skilled in the art within the scope of the present invention.
Claims (9)
1. A three-dimensional ocean thermal salt field forecasting method based on deep learning is characterized by comprising the following steps:
firstly, constructing a deep learning model training data set, and preprocessing original ocean three-dimensional warm salt data and ocean surface data;
step two, constructing an underwater three-dimensional Wen Yanchang forecasting model, wherein the model comprises 5 modules: a 3-dimensional convolution module, an attention module, a position coding module, a residual error module and a 3-dimensional transposition convolution module;
thirdly, performing multi-angle verification including early prediction duration verification, vertical verification, horizontal verification and actual measurement verification on a three-dimensional Wen Yanchang prediction result by using a data set or buoy actual measurement value which is not trained by the model;
and fourthly, carrying out post-processing on output data of the model, visualizing predicted ocean temperature and salinity in a two-dimensional thermodynamic diagram, an isotherm and a temperature-salt depth change curve, and calculating an error.
2. The three-dimensional ocean thermal salt field prediction method based on deep learning according to claim 1, wherein the first step comprises:
(1) Acquiring marine environment data of many years through satellites, buoys, numerical modes and analysis data, wherein the marine environment data comprises sea surface temperature, sea surface salinity, sea surface north-south flow velocity, sea surface height and sea surface internal temperature salt data of 2000m under water;
(2) Bilinear interpolation with pixel points
Four points around it ∈ ->
，/>
，/>
，/>
The corresponding grey values are +.>
；
(3) Min-Max normalization is specifically expressed as:
wherein, the liquid crystal display device comprises a liquid crystal display device,
for inputting data +.>
For outputting data +.>
Minimum value, +.>
Is at a maximum value;
(4) Land and outlier processing, replacing all land and outliers with 0;
(5) Uniformly converting the spatial resolution of the original data into 0.083 degrees, the time resolution of 3h and the depth of 0-5500m into 1 day of time resolution and the depth of 0-1000m;
(6) Dividing sea area data to be forecasted into 16 sub sea area data;
(7) And respectively processing the processed sea surface data and the underwater three-dimensional salt warming data into 5-dimensional matrixes (BxT x V x H x W, bxT x D x H x W) required by the model, wherein B is the number of batches, V is the sea surface variable number, T is the sample number of a time sequence, D is a depth range, H is a latitude range and W is a longitude range.
3. The three-dimensional ocean thermal salt field forecasting method based on deep learning of claim 1, wherein the three-dimensional Wen Yanchang forecasting model is developed based on a Pytorch framework.
4. The three-dimensional ocean thermal salt field forecasting method based on deep learning according to claim 1, wherein the second step specifically comprises:
(1) The input multidimensional sea surface environment data is convolved by a 3D convolution module, and the output of the 3D convolution is still a 3D characteristic diagram; the size of the convolution kernel in the 3D convolution is (2, 10, 10), the step size is (1, 1, 1), and the specific expression of the three-dimensional convolution is:
wherein, the liquid crystal display device comprises a liquid crystal display device,
，/>
，/>
for the length of the data in the horizontal direction and in the depth direction, < >>
，/>
，/>
Respectively->
，/>
，/>
Index of->
For inputting samples, < >>
For convolution kernel +.>
Is->
The convolution kernels are +.>
The weight value of the time-out is calculated,
is->
The individual samples are->
Data value at->
Is->
Bias term of the convolution kernel +.>
Is->
The sample is at->
Convolving the convolved result under a convolution signal;
(2) The input data is flattened into a 1-dimensional vector after 3D convolution, and position codes are added after flattening, wherein the specific expression of the position codes is as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,
for the position index of the input vector, < >>
For the dimension index in the position-coding vector, +.>
For feature vector dimension, < >>
Is a sine function +.>
As cosine function +.>
Position coding for even positions, +.>
Position encoding for odd positions;
(3) After the 1-dimensional vector is added into the position code, the vector is transmitted into an attention module; the attention module simulates the selective attention of human beings, more key information is gradually selected in the training process, different weights are given according to the importance degree of different information, and the specific expression of the attention is as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,
is->
Input vector->
Is->
Individual query vectors->
Is->
Personal key vector->
Is->
Personal value vector->
、/>
、/>
Is a trainable weight matrix, +.>
Is->
Square root of dimension, & gt>
For the input query vector, +.>
For the key vector entered, +.>
For the input value vector, +.>
Indicating transpose,/->
In order to be able to take care of the size of the attention,
the expression of (2) is:
wherein, the liquid crystal display device comprises a liquid crystal display device,
indicate->
Similarity of the individual key vector and the query vector, < ->
Representing a natural exponential function;
for capturing data characteristics, multi-head attention is adopted to extract the mapping relation of the data of different positions on the sea surface to the underwater influence; the multi-head attention comprises a 4-layer network, each layer of network comprises 6 attention heads, and the multi-head attention is specifically expressed as:
wherein, the liquid crystal display device comprises a liquid crystal display device,
is a trainable matrix>
Outputting a result for a single attention, < >>
Indicating that the output results of a plurality of attentions are spliced, < >>
The multi-head attention size;
(4) The residual error modules are added before and after the attention module, and are specifically expressed as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,
is depth +.>
Is (are) unit features of->
Is depth +.>
Is (are) unit features of->
Is->
Input vector->
For the weight transferred to the unit, +.>
Is an arbitrary function;
(5) Recovering the size using transpose convolution; the size of the convolution kernel in the 3D transpose convolution is (4, 10, 10), the step size is (2, 1, 1), and the transpose convolution is specifically expressed as:
wherein, the liquid crystal display device comprises a liquid crystal display device,
for outputting the vector +.>
For inputting vectors, ++>
For convolution kernel matrix, ++>
Representing a transpose;
(6) Selecting a GELU optimizer, and continuously optimizing super parameters of the model by taking the average absolute error as a loss function to minimize the error; the specific expression of the gel is:
wherein, the liquid crystal display device comprises a liquid crystal display device,
for activating the input vector of the function, +.>
Is a gaussian cumulative distribution function.
5. The deep learning-based three-dimensional ocean salt field prediction method according to claim 1, wherein the multi-angle verification comprises early prediction duration verification, vertical verification, horizontal verification and actual measurement verification.
6. The three-dimensional ocean thermal salt field forecasting method based on deep learning according to claim 1, wherein the three-dimensional Wen Yanchang forecasting result is subjected to multi-angle verification, and forecasting accuracy is evaluated by two indexes, namely:
(1) Root mean square error of forecast temperature salt field and open source data set temperature salt field:
(2) Average absolute percentage error of the forecast temperature salt field and the open source data set temperature salt field:
wherein, the liquid crystal display device comprises a liquid crystal display device,
and->
Is->
Predicted and actual values, +.>
For the total number of data>
Is the root mean square error (rms) error,
is the mean absolute percentage error.
7. The deep learning-based three-dimensional ocean thermal salt field prediction method of claim 1, wherein post-processing the output data of the model comprises:
(1) Changing the corresponding position of the output of the model into nan through three-dimensional topographic data before training;
(2) Interpolating ocean temperature/salt at each depth to a single point through bilinear interpolation to form temperature/salt data at different depths at the single point;
(3) And cutting the three-dimensional temperature/salt data output by the model along the warp and weft respectively to obtain a temperature/salt section on a vertical plane.
8. A three-dimensional ocean thermal salt field forecasting system based on deep learning, which implements the three-dimensional ocean thermal salt field forecasting method based on deep learning as set forth in any one of claims 1 to 7, and is characterized by comprising:
the input module is used for constructing a training data set of the deep learning model and preprocessing the training data set to enable the training data set to meet the input requirement of the model;
the training module is used for training a future underwater three-dimensional Wen Yanchang forecasting model;
the verification module is used for multi-angle verification of the three-dimensional Wen Yanchang forecasting result;
and the visualization module is used for visualizing the predicted ocean temperature and salinity according to a two-dimensional thermodynamic diagram, an isotherm and a temperature and salt depth change curve.
9. An apparatus for deep learning based three-dimensional marine thermal salt field prediction, the apparatus comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of the deep learning based three-dimensional marine thermal salt field prediction method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310530474.6A CN116306318B (en) | 2023-05-12 | 2023-05-12 | Three-dimensional ocean thermal salt field forecasting method, system and equipment based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310530474.6A CN116306318B (en) | 2023-05-12 | 2023-05-12 | Three-dimensional ocean thermal salt field forecasting method, system and equipment based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116306318A true CN116306318A (en) | 2023-06-23 |
| CN116306318B CN116306318B (en) | 2023-08-01 |
Family
ID=86801636
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310530474.6A Active CN116306318B (en) | 2023-05-12 | 2023-05-12 | Three-dimensional ocean thermal salt field forecasting method, system and equipment based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116306318B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116588282A (en) * | 2023-07-17 | 2023-08-15 | 青岛哈尔滨工程大学创新发展中心 | AUV intelligent operation and maintenance system and method |
| CN117647808A (en) * | 2024-01-25 | 2024-03-05 | 青岛哈尔滨工程大学创新发展中心 | Non-coherent radar phase analysis wave calendar inversion method based on deep learning |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190155973A1 (en) * | 2017-11-02 | 2019-05-23 | Airworks Solutions, Inc. | Methods and apparatus for automatically defining computer-aided design files using machine learning, image analytics, and/or computer vision |
| CN111951553A (en) * | 2020-08-17 | 2020-11-17 | 上海电科智能系统股份有限公司 | Prediction method based on traffic big data platform and mesoscopic simulation model |
| CN112446170A (en) * | 2020-11-06 | 2021-03-05 | 山东大学 | Method for establishing regional ocean bottom temperature thermal infrared remote sensing monitoring and forecasting model |
| CN112525166A (en) * | 2020-11-24 | 2021-03-19 | 浙江大学 | Tidal flat health multi-parameter profile real-time synchronous monitoring device and method |
| CN113072490A (en) * | 2021-03-26 | 2021-07-06 | 中国海洋大学 | High-efficiency synthesis method of tipifarnib quinolinone intermediate |
| CN113821986A (en) * | 2021-11-22 | 2021-12-21 | 中国海洋大学 | Vortex sea surface signal and underwater key parameter inversion method based on random forest |
| CN114417693A (en) * | 2021-11-26 | 2022-04-29 | 中国石油大学(华东) | Ocean three-dimensional temperature field inversion method based on deep learning |
-
2023
- 2023-05-12 CN CN202310530474.6A patent/CN116306318B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190155973A1 (en) * | 2017-11-02 | 2019-05-23 | Airworks Solutions, Inc. | Methods and apparatus for automatically defining computer-aided design files using machine learning, image analytics, and/or computer vision |
| CN111951553A (en) * | 2020-08-17 | 2020-11-17 | 上海电科智能系统股份有限公司 | Prediction method based on traffic big data platform and mesoscopic simulation model |
| CN112446170A (en) * | 2020-11-06 | 2021-03-05 | 山东大学 | Method for establishing regional ocean bottom temperature thermal infrared remote sensing monitoring and forecasting model |
| CN112525166A (en) * | 2020-11-24 | 2021-03-19 | 浙江大学 | Tidal flat health multi-parameter profile real-time synchronous monitoring device and method |
| CN113072490A (en) * | 2021-03-26 | 2021-07-06 | 中国海洋大学 | High-efficiency synthesis method of tipifarnib quinolinone intermediate |
| CN113821986A (en) * | 2021-11-22 | 2021-12-21 | 中国海洋大学 | Vortex sea surface signal and underwater key parameter inversion method based on random forest |
| CN114417693A (en) * | 2021-11-26 | 2022-04-29 | 中国石油大学(华东) | Ocean three-dimensional temperature field inversion method based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| S.TORRES LOPEZ等: "Residual circulation and thermohaline distribution of the ria de vigo: a 3-D hydrodynamical model", 《SCIENTIA MARINA》 * |
| 黄润琪等: "湛江湾三维温盐特征季节变化观测分析", 《万方》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116588282A (en) * | 2023-07-17 | 2023-08-15 | 青岛哈尔滨工程大学创新发展中心 | AUV intelligent operation and maintenance system and method |
| CN116588282B (en) * | 2023-07-17 | 2023-10-13 | 青岛哈尔滨工程大学创新发展中心 | AUV intelligent operation and maintenance system and method |
| CN117647808A (en) * | 2024-01-25 | 2024-03-05 | 青岛哈尔滨工程大学创新发展中心 | Non-coherent radar phase analysis wave calendar inversion method based on deep learning |
| CN117647808B (en) * | 2024-01-25 | 2024-04-30 | 青岛哈尔滨工程大学创新发展中心 | Non-coherent radar phase analysis wave calendar inversion method based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116306318B (en) | 2023-08-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116306318B (en) | Three-dimensional ocean thermal salt field forecasting method, system and equipment based on deep learning | |
| Lou et al. | Application of machine learning in ocean data | |
| US20200356839A1 (en) | Systems and methods for selecting global climate simulation models for training neural network climate forecasting models | |
| Lemos et al. | A spatio-temporal model for mean, anomaly, and trend fields of North Atlantic sea surface temperature | |
| Jing et al. | Numerical study of a CNN-based model for regional wave prediction | |
| Zhang et al. | Deep learning for 3-D inversion of gravity data | |
| Chattopadhyay et al. | Towards physically consistent data-driven weather forecasting: Integrating data assimilation with equivariance-preserving deep spatial transformers | |
| Li et al. | Efficient estimation of nonstationary spatial covariance functions with application to high-resolution climate model emulation | |
| CN115248992A (en) | Ocean three-dimensional temperature and salt space-time prediction method based on compressed excitation PredRNN | |
| Grönquist et al. | Predicting weather uncertainty with deep convnets | |
| Chen et al. | WSN sampling optimization for signal reconstruction using spatiotemporal autoencoder | |
| Xie et al. | Developing a deep learning-based storm surge forecasting model | |
| Jiang et al. | Ocean observation data prediction for Argo data quality control using deep bidirectional LSTM network | |
| CN117350171B (en) | Mesoscale vortex three-dimensional subsurface structure inversion method and system based on double-flow model | |
| CN115238937A (en) | Ocean temperature salt prediction method and system | |
| Petrou et al. | Prediction of sea ice motion with recurrent neural networks | |
| CN117253344A (en) | Seawater acidification early warning and forecasting method, system and electronic equipment | |
| Menaka et al. | Prediction of dominant ocean parameters for sustainable marine environment | |
| Zhang et al. | A Gaussian process regression-based sea surface temperature interpolation algorithm | |
| CN116739161A (en) | Sea surface temperature prediction method integrating empirical mode decomposition and Informir model | |
| Jiang et al. | Prediction of sea temperature using temporal convolutional network and LSTM-GRU network | |
| Ou et al. | Estimation of sound speed profiles using a random forest model with satellite surface observations | |
| CN112541292B (en) | Submarine cable burial depth estimation algorithm based on distributed optical fiber temperature measurement principle | |
| Yu et al. | Inversion of the three-dimensional temperature structure of mesoscale eddies in the Northwest Pacific based on deep learning | |
| Chen et al. | TemproNet: A transformer-based deep learning model for seawater temperature prediction |
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 |