CN115393540A - Intelligent fusion method and system of three-dimensional marine environment field based on deep learning - Google Patents
Intelligent fusion method and system of three-dimensional marine environment field based on deep learning Download PDFInfo
- Publication number
- CN115393540A CN115393540A CN202211050583.XA CN202211050583A CN115393540A CN 115393540 A CN115393540 A CN 115393540A CN 202211050583 A CN202211050583 A CN 202211050583A CN 115393540 A CN115393540 A CN 115393540A
- Authority
- CN
- China
- Prior art keywords
- data
- dimensional
- marine
- ocean
- deep learning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/05—Geographic models
-
- 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)
- Software Systems (AREA)
- Geometry (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Remote Sensing (AREA)
- Computer Graphics (AREA)
- Testing Or Calibration Of Command Recording Devices (AREA)
Abstract
The invention discloses a three-dimensional marine environment field intelligent fusion method and a system based on deep learning, wherein the method comprises the following steps: s1, obtaining multi-source marine data, wherein the multi-source marine data comprise one-dimensional observation data, two-dimensional marine satellite observation data, a three-dimensional gridding marine mode and reanalysis data; s2, carrying out data processing on multi-source ocean data; s3, making a sample set; s4, building a deep learning network model, training the deep learning network model, and building an intelligent ocean data fusion model; and S5, inputting the multi-source ocean data to be fused into an ocean data intelligent fusion model, and outputting an intelligent fusion result. The invention comprehensively considers the spatial-temporal continuity and consistency of marine environment elements and the physical relevance among multiple elements, realizes the high-precision, high-resolution and spatial-temporal continuous intelligent fusion of multi-dimensional and multi-source heterogeneous marine data by constructing the marine data intelligent fusion model, and provides high-quality data resource support for scientific research and application guarantee in the marine field.
Description
Technical Field
The invention belongs to the field of marine science and machine learning, and particularly relates to an intelligent fusion method and system of a three-dimensional marine environment field based on deep learning.
Background
High-quality marine data are important prerequisites and requirements for marine scientific research, global change research and marine high-efficiency production operation. In order to further and deeply reveal and predict the climate change and the evolution of the marine environment, long-time sequence and high-resolution reliable marine environment data need to be obtained, and scientific research and numerical simulation are developed on the basis of the marine environment data, which become common knowledge of field research.
The current ocean observation means are increasingly abundant, the available ocean data volume is rapidly increased, and different ocean data (data) are good and bad respectively. The conventional ocean observation such as the conventional ocean station, buoy, submerged buoy, sailing and the like can detect the internal information of the ocean, has the advantages of high accuracy and the like, but is very sparse in the global scope and poor in space-time continuity due to the limited observation range; the ocean satellite remote sensing data has the advantages of wide detection range, good time-space continuity and the like, but the depth of an observation layer is limited, and the information of the ocean internal data cannot be detected; the existing mode simulation forecast or reanalysis data with the largest mass is three-dimensional gridding data, the space-time continuity is best, but the accuracy degree is far less than that of observation data. Therefore, the advantages of various ocean data need to be combined to obtain a three-dimensional ocean element fusion field with high precision, high resolution and space-time continuity.
With the increase of the types and the number of observation data, the fusion of multi-source ocean data provides huge challenges for computing power, numerical methods and assimilation technologies. The development of two traditional methods, namely marine statistical analysis and marine environment numerical prediction, meets certain bottlenecks: due to the facts that a traditional linear method model is too simple, understanding of an ocean process is not comprehensive and deep enough, uncertainty of a numerical mode and limitation of mode resolution are caused, and the requirements of increasingly refined and precise research and production cannot be met.
In recent years, an artificial intelligence technology taking deep learning as a core is introduced into the field of earth system science, good effects are obtained, a new thought and method are provided for marine science research, and an AI technology taking machine learning as a core becomes a new way for exerting the value of observation data and improving the guarantee level of marine environment.
Disclosure of Invention
Aiming at the problems in the prior art, the invention aims to provide an intelligent fusion method and system of a three-dimensional marine environment field based on deep learning, so as to solve the technical problems that the existing marine observation data has poor space-time continuity, limited observation layer depth, low precision and large calculated amount of mode and reanalysis data, and low generation efficiency, so that the three-dimensional marine element fusion field with high precision, high resolution and space-time continuity cannot be quickly obtained.
In order to achieve the purpose, the invention adopts the following technical scheme:
the invention provides an intelligent fusion method of a three-dimensional marine environment field based on deep learning, which comprises the following steps:
s1, obtaining multi-source marine data, wherein the multi-source marine data comprise one-dimensional observation data, two-dimensional marine satellite observation data, a three-dimensional gridding marine mode and reanalysis data;
s2, performing data processing on the obtained multi-source ocean data, including quality control processing, space-time interpolation processing and normalization processing on the multi-source ocean data;
s3, making a sample set, and dividing a training set, a testing set and a verification set;
s4, building a deep learning network model, training the deep learning network model by adopting a training set, and building an intelligent ocean data fusion model;
and S5, inputting the multi-source ocean data to be fused into an ocean data intelligent fusion model, and outputting an intelligent fusion result.
Preferably, the method also comprises S6, inputting the test set into the ocean data intelligent fusion model, outputting a model test result, and analyzing the performance of the model according to the speed and parameters of the model during testing; meanwhile, according to the model test result and the test sample label, error indexes are calculated and visualized, and the precision and the data fusion quality of the verification model are evaluated.
Preferably, the one-dimensional observation data comprises Argo, and sea Wen Pouxian data observed by buoys, submerged buoys and shipborne equipment, and the one-dimensional observation data is derived from an EN4-profiles data set of Hadelai center of the United kingdom weather bureau; the two-dimensional marine satellite observation data comprises sea surface temperature data, sea surface height anomaly data and sea surface wind field data; wherein the sea surface temperature data are derived from an ERSST data set, a GHRSST data set, a HadisT data set and the like; the sea surface height anomaly data is derived from an AVISO SLA data set; the sea surface wind field data is derived from a CCMP data set; the three-dimensional gridding ocean mode and reanalysis data comprise HYCOM data, SODA ocean reanalysis data and EN4-analysis data.
Preferably, in step S2, the data processing includes performing quality control processing, temporal-spatial interpolation processing, and normalization processing on the obtained multi-source ocean data in sequence.
Preferably, the quality control processing is performed on the multi-source ocean data, specifically: and carrying out query, elimination, correction and filling on abnormal values and missing values of one-dimensional observation data in the multi-source ocean data.
Preferably, the performing the spatial-temporal interpolation processing on the multi-source ocean data specifically includes the following steps:
a21, interpolating the one-dimensional observation data to a vertical direction standard layer { z1, z2, …, zk } by adopting a linear interpolation method;
a22, interpolating the one-dimensional observation data which are discretely distributed in the horizontal direction to grid points corresponding to a horizontal grid { x1, x2, …, xm } × { y1, y2, …, yn } by adopting a space-time weight interpolation method;
a23, interpolating the two-dimensional marine satellite observation data to a horizontal grid { x1, x2, …, xm } × { y1, y2, …, yn } by a bilinear interpolation method;
a24, interpolating the three-dimensional gridding ocean mode and the reanalysis data into a three-dimensional grid { x1, x2, …, xm } × { y1, y2, …, yn } × { z1, z2, …, zk } by a bilinear interpolation method;
and A25, interpolating the multisource ocean data subjected to the spatial interpolation to the same time resolution.
Preferably, the normalization processing is performed on the multi-source ocean data, and specifically includes: and carrying out normalization processing on the multi-source ocean data and the space-time information data after space-time interpolation, wherein the space-time information data comprises the year, month, longitude and latitude of the position.
Preferably, in step S3, the preparing a sample set specifically includes the following steps:
s31, taking the one-dimensional observation data interpolated to the vertical direction standard layer as a sample label, setting the number of layers of the vertical direction standard layer as k, and recording the dimension of a sample label array as (1,k);
s32, taking EN4-analysis data, HYCOM data and SODA reanalysis data of a horizontal grid position corresponding to the one-dimensional observation data, a difference value between the EN4-analysis data and the HYCOM data, a difference value between the HYCOM data and the SODA reanalysis data and a difference value between the EN4-analysis data and the SODA reanalysis data as sample characteristic data to form 6 characteristic channels;
s33, taking the space-time information data where the sample label is located and the two-dimensional marine satellite observation data of the grid corresponding to the sample label as sample characteristic data, forming 1 characteristic channel, making a sample set, and recording the dimension of the sample characteristic array as (7,k).
Preferably, a global attention mechanism and an adaptive parameterized ReLU activation function are added to the deep learning network model, wherein the global attention mechanism adopts a serial mode of firstly performing channel attention and then performing spatial attention.
Preferably, in step S4, the training set is used to train the built deep learning network model, and the specific steps are as follows: training the built deep learning network model by adopting a training set, comparing an output result with sample label data, and constructing an ocean data intelligent fusion model by minimizing the difference between the network model output and the sample label data and optimizing parameters.
The invention provides an intelligent system of a three-dimensional marine environment field based on deep learning, which comprises:
the data acquisition and processing module is used for acquiring multi-source marine data and processing the data;
the sample set making module is used for analyzing the multi-source ocean data and making the multi-source ocean data into a sample set;
the ocean data intelligent fusion model building module is used for building and training a deep learning network model;
the intelligent fusion result output module is used for inputting the multi-source ocean data to be fused into the ocean data intelligent fusion model and outputting an intelligent fusion result;
and the result verification module is used for evaluating the performance, the precision and the data fusion quality of the constructed ocean data intelligent fusion model.
The invention has the following beneficial effects:
the intelligent fusion method of the three-dimensional marine environment field based on deep learning can perform intelligent fusion on a plurality of marine element fields including temperature, salinity, flow fields, sea level height, sea waves, sea surface wind fields and the like, comprehensively considers the spatial-temporal continuity and consistency of marine environment elements and the physical relevance among multiple elements based on multi-source marine environment data such as field observation data, satellite remote sensing data, marine re-analysis data, mode data and the like, realizes the high-precision, high-resolution and spatial-temporal continuity intelligent fusion of multi-dimensional multi-source heterogeneous marine data, namely the three-dimensional marine environment (sea temperature) field, and provides high-quality data resource support for scientific research and application guarantee of the marine field.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flow chart of an intelligent fusion method of a three-dimensional marine environment field based on deep learning according to the present invention;
FIG. 2 is a schematic view of a sample label;
FIG. 3 is a diagram illustrating an arrangement of sample feature data;
FIG. 4 is a schematic diagram of a built ODF-Net network model structure;
FIG. 5 is a schematic diagram of a global attention mechanism;
FIG. 6 is a schematic diagram of a channel attention submodule;
FIG. 7 is a schematic structural diagram of a spatial attention submodule;
FIG. 8 is a diagram illustrating the basic principles of the APReLU activation function;
FIG. 9-a is a diagram a of the intelligent fusion result of the ocean temperature data of the global ocean standards layer in example 3;
FIG. 9-b is a graph b of the intelligent fusion result of the sea temperature data of the global ocean standard layer in example 3;
FIG. 9-c is a graph c of the intelligent fusion result of the ocean temperature data of the global ocean standards layer in example 3.
Detailed Description
In the following description, for purposes of explanation and not limitation, specific details are set forth, such as particular system structures, techniques, etc. in order to provide a thorough understanding of the embodiments of the invention. It will be apparent, however, to one skilled in the art that the present invention may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present invention with unnecessary detail.
Example 1
Referring to fig. 1, an embodiment of the present invention provides an intelligent fusion method for a three-dimensional marine environment field based on deep learning, including the following steps:
s1, acquiring multi-source marine data, wherein the multi-source marine data comprises one-dimensional observation data, two-dimensional marine satellite observation data, a three-dimensional gridding marine mode and reanalysis data;
specifically, the one-dimensional observation data includes Argo data, and data Wen Pouxian observed by buoys, submerged buoys, shipborne equipment and the like, and is derived from an EN4-profiles (sea Wen Kuoxian) data set of the hadley center of the united kingdom meteorological office;
the two-dimensional marine satellite observation data comprises sea surface temperature data, sea surface height anomaly data, sea surface wind field data and the like. Wherein, the sea surface temperature data is derived from an ERSST data set (expanded reconstructed sea surface temperature data set), a GHRSST data set (global high resolution sea surface temperature test plan), a HadisST data set (global sea surface temperature analysis data of Hadeli center), and the like; sea surface height anomaly data is derived from an AVISOSLA data set (French national space research center satellite oceanography archive data center); the sea surface wind farm data is derived from the CCMP dataset (NASA geosciences agency multi-platform cross-correction data).
The three-dimensional gridding ocean mode and re-analysis data comprise (American naval mixed coordinate ocean mode) HYCOM data, (American Maryland university simple ocean data assimilation) SODA re-analysis data, (British weather Hardelai center quality control temperature salt analysis)
EN4-analysis data, etc.
S2, carrying out data processing on the obtained multi-source ocean data;
specifically, the obtained multi-source ocean data is subjected to quality control processing, space-time interpolation processing and normalization processing in sequence;
performing quality control processing on the obtained multi-source marine data, wherein the quality control processing is mainly used for inquiring, eliminating, correcting and filling abnormal values and missing values of one-dimensional observation data in the multi-source marine data; specifically, all incorrect and uncertain sectioning lines are removed according to a flag quality identifier in an EN4-profiles data set, then the deepest depth of the EN4-profiles data set is selected to 1000m, and then abnormal sectioning lines with temperature values exceeding the range of minus 5-40 ℃ are screened out and removed;
(2) Performing space-time interpolation processing on multi-source ocean data;
firstly, interpolating multisource (heterogeneous) ocean data (data) to the same horizontal grid point, and then interpolating the multisource ocean data (data) subjected to spatial interpolation to the same time resolution; the method specifically comprises the following steps:
a21, interpolating the one-dimensional observation data to the vertical direction standard layer { z ] by adopting a linear interpolation method 1 ,z 2 ,…,z k K =23,z in the present embodiment 1 Has a vertical depth of 0m, z 23 Has a vertical depth of 1000m;
a22, interpolating the one-dimensional observation data which are discretely distributed in the horizontal direction to a horizontal grid { x ] by adopting a space-time weight interpolation method 1 ,x 2 ,…,x m }×{y 1 ,y 2 ,…,y n On the corresponding grid point; in the present embodiment, the resolution of the horizontal grid is 0.25 °, and the corresponding value of each horizontal grid to be interpolated is calculated according to the following formulas (i) and (ii):
wherein u is estimate For interpolation of ocean elements, w k As interpolation weights, x k ，y k ，t k Respectively the longitude and latitude values and the time, x, of the k-th point in the interpolation neighborhood o ，y o ，t o Respectively representing longitude and latitude values and time of a target interpolation point, wherein D is a space horizontal interpolation radius, T is a time dimension interpolation radius, and N is the number of all observation data contained in a space-time neighborhood of the space radius D and the time dimension radius T which take a target grid as a center; in the present embodiment, D =0.5 °, T =15D;
a23, interpolating the two-dimensional marine satellite observation data to a horizontal grid { x ] by adopting a bilinear interpolation method 1 ,x 2 ,…,x m }×{y 1 ,y 2 ,…,y n }；
A24, interpolating the three-dimensional gridding ocean mode and the reanalysis data (including HYCOM data, SODA data, EN4-analysis data and the like) into the three-dimensional grid { x ] by adopting a bilinear interpolation method 1 ,x 2 ,…,x m }×{y 1 ,y 2 ,…,y n }×{z 1 ,z 2 ,…,z k }；
And A25, interpolating the multi-source ocean data after the spatial interpolation to the same time resolution.
(3) The method comprises the following steps of carrying out normalization processing on multi-source ocean data, specifically: normalizing the multi-source ocean data and the spatio-temporal information data after the spatio-temporal interpolation to unify dimensions, wherein the spatio-temporal information data comprise the year, month, longitude and latitude of the place, and due to the particularity of longitude characteristics (180 degrees E and 180 degrees W are overlapped), the longitude is normalized according to the following formula (III):
the remaining spatiotemporal information data (including year, month and latitude) are normalized for maximum and minimum values according to the following equation (iv):
wherein x is norm Representing the normalized value of a spatio-temporal information data (feature), X representing the original value of the spatio-temporal information data, X max And X min Respectively representing the maximum value and the minimum value of the spatio-temporal information data in all samples;
s3, making a sample set, and dividing the sample set into a training set, a testing set and a verification set;
by analyzing the relation among the spatio-temporal information data, the sea surface information data and the multi-source observation data, extracting key elements to manufacture a sample set, and specifically comprising the following steps:
s31, taking the one-dimensional observation data (from an EN4-profiles data set) interpolated to the vertical standard layer as a sample label;
since the number of the vertical standard layers in this embodiment is 23, the dimension of the label in one sample is denoted as 1 × 23, and as shown in fig. 2, li represents an observed value in the EN4-profiles data of the i-th layer;
s32, taking EN4-analysis data, HYCOM data, SODA reanalysis data, a difference value between the EN4-analysis data and the HYCOM data, a difference value between the HYCOM data and the SODA reanalysis data, a difference value between the EN4-analysis data and the SODA reanalysis data, spatio-temporal information data where a sample label is located and two-dimensional marine satellite observation data of a grid corresponding to the sample label as sample characteristic data, wherein the EN4-analysis data, the HYCOM data, the SODA reanalysis data, a difference value between the EN4-analysis data and the HYCOM data, a difference value between the HYCOM data and the SODA reanalysis data, and a difference value between the EN4-analysis data and the SODA reanalysis data of a horizontal grid position corresponding to the one-dimensional observation data form 6 characteristic channels, the arrangement mode of the sample feature data is shown in fig. 2, wherein Ei, hi, and Si respectively represent analysis data of an ith layer in a same longitude and latitude grid as the tag, (E-H) i represents a difference value between EN4-analysis and HYCOM data of the ith layer, (E-S) i represents a difference value between EN4-analysis data and SODA reanalysis data of the ith layer, (H-S) i represents a difference value between HYCOM data and SODA reanalysis data of the ith layer, hydro includes spatio-temporal information where the sample tag is located and two-dimensional marine satellite observation data of the corresponding grid, lat is a latitude of the corresponding grid of the sample tag, lon1 and lon2 represent longitudes of the corresponding grid of the sample tag, year is a year of the sample tag, month is the month of the sample label, sla is the sea level height anomaly of the position corresponding to the sample label, ccmp _ u and ccmp _ v are the latitudinal and longitudinal components of the sea level wind field of the position corresponding to the sample label, ersst, ghrsst and hadisst are three sets of different sea level temperature data of the position corresponding to the sample label, and finally, 0 is used for compensating the Hydro to the length of 23, so the dimension of the sample feature array can be marked as 7 × 23.
And S4, building a deep learning network model, training the deep learning network model by adopting a training set, comparing an output result with tag data, and constructing an ocean data intelligent fusion model by minimizing the difference between the network model output and the tag data and optimizing parameters.
The method comprises the steps of independently building a deep learning network model, and naming the built deep learning network model as ODF-Net, wherein the structure of the ODF-Net network model is shown in FIG. 2 and comprises 4 basic blocks (block 1, block2, block3 and block 4) with residual error structures; wherein Input represents an Input layer; GAM represents a global attention mechanism; conv1d represents a 1-dimensional convolution layer, the convolution kernel size being 3 × 1; BN (Batch Normalization) denotes Batch Normalization layer; aprlu denotes the adaptive parameterized ReLU activation function; the Dropout layer is to temporarily drop neurons from the network with a probability of p =0.5 to prevent the network from overfitting; max-Pool denotes the maximum pooling layer; the Linear layer is used for processing data into one dimension; FC denotes a full connection layer. Blocks indicated by dashed rectangles represent basic blocks with residual structure, block1 having one BN layer, one aprellu and one Dropout layer less than the other blocks. Experiments prove that the 4 blocks (basic blocks) built by the invention can bring the best fusion effect.
A Global Attention Mechanism (GAM) is introduced to amplify the interaction across dimensions, so as to capture important features in all three dimensions. The structure of the GAM Attention mechanism is shown in fig. 5, in which Channel Attention represents a Channel Attention submodule, spatial Attention represents a Spatial Attention submodule, and GAM adopts a serial mode of firstly performing Channel Attention and then performing Spatial Attention. Given an input mapping F1, F ∈ R C×H×W Intermediate state F 2 And an output F 3 Is defined as:
wherein M is C And M S A channel attention diagram and a space attention diagram are respectively provided;
indicating multiplication by element.
The structure of the channel attention submodule is shown in fig. 6, the channel attention submodule uses three-dimensional arrangement to retain information in three dimensions; it then amplifies the cross-dimensional channel spatial dependence with a two-layer MLP (multi-layer perceptron). Here, MLP is a codec structure with a compression ratio of r. Specifically, input data with the dimension of C × W × H is subjected to dimension conversion, the converted dimension is W × H × C, then the input data passes through a simple MLP structure, then the dimension W × H × C is converted back to the original dimension C × W × H, and finally the weight value of each channel is obtained through a Sigmoid function.
The spatial attention submodule has a structure as shown in fig. 7, and in the spatial attention submodule, spatial information fusion is performed using two convolutional layers in order to focus on spatial information. Specifically, input data with the dimension of C × H × W first passes through a convolution layer with a convolution kernel of 7 × 7, the data dimension becomes C/r × H × W after passing through the convolution layer, the input data passes through a convolution layer with a convolution kernel of 7 × 7 again, the data dimension returns to the original C × H × W after passing through the convolution layer, and finally the weight of each feature value is obtained through a Sigmoid function.
An apreflu (adaptive Parametric Rectifier Unit, abbreviated as apreflu), that is, an adaptive Parametric ReLU activation function is an optimization improvement made on the basis of a prellu (Parametric ReLU) activation function. The PReLU sets the coefficient of the feature less than zero as a parameter which can be obtained by training, and a gradient descent method is adopted to train the PReLU together with other parameters in the training process of the neural network. The aprerlu obtains the weight corresponding to each sample through a small fully-connected network, and then takes the group of weights as the coefficient of the original eigenvalue when the eigenvalue is less than zero, namely the weight of the negative part in the prilu function.
The basic principle of aprlu is shown in fig. 8, reLU denoting the ReLU activation function; min (x, 0) represents selecting the smaller of the two values x and 0; max (x, 0) represents selecting the larger of the two values x and 0; GAP represents global average pooling; concat represents channel splicing; FC denotes a full connection layer; BN represents batch standardization operations; sigmoid represents a Sigmoid function used to find a set of weight values;
and S5, inputting the multi-source ocean data to be fused into an ocean data intelligent fusion model, and outputting an intelligent fusion result.
S6, inputting the test set into the ocean data intelligent fusion model obtained in the step S4, and outputting a model test result; and analyzing the performance of the network model according to the speed, parameter quantity and the like of the model during testing, calculating and visualizing error indexes according to the test result of the model and the test sample label, and evaluating the precision and data fusion quality of the verification model.
Example 2
The embodiment provides an intelligent system of a three-dimensional marine environment field based on deep learning, which comprises:
the data acquisition and processing module is used for acquiring multi-source marine data and processing the data;
the sample set making module is used for analyzing the multi-source ocean data and making the multi-source ocean data into a sample set;
the ocean data intelligent fusion model building module is used for building and training a deep learning network model;
the intelligent fusion result output module is used for inputting the multi-source ocean data to be fused into the ocean data intelligent fusion model and outputting an intelligent fusion result;
and the result verification module is used for evaluating the performance and the precision of the constructed intelligent ocean data fusion model and the data fusion quality.
Example 3
In this embodiment, the intelligent fusion method of the three-dimensional marine environment field based on deep learning in embodiment 1 is adopted to perform intelligent fusion on the global marine standard layer sea temperature data to obtain a three-dimensional fusion sea temperature field, and specific results are shown in table 1, table 2, fig. 9-a, fig. 9-b, and fig. 9-c.
The results in table 1 show that compared with the model without APReLU and GAM, the ODF-Net model constructed by the method provided by the invention has a significantly smaller prediction error, which indicates that APReLU and GAM have positive effects on the prediction of the ocean data intelligent fusion model;
as can be seen from the results of Table 2, FIG. 9-a, FIG. 9-b and FIG. 9-c, the fused data obtained by the method of the present invention is closer to the observed data of the label than the En4_ Anal, HYCOM and SODA data.
TABLE 1
| APReLU | GAM | MAE | RMSE |
| × | × | 0.384 | 0.687 |
| √ | × | 0.369 | 0.662 |
| √ | √ | 0.349 | 0.628 |
Remarking: v represents addition; x represents no addition.
TABLE 2
| Name of data | MAE | RMSE |
| En4 Anal | 0.468 | 0.799 |
| HYCOM | 0.576 | 0.966 |
| SODA | 0.476 | 0.870 |
| Sea temperature fusion data of global ocean standard layer | 0.349 | 0.628 |
The present invention is not limited to the above-described embodiments, and those skilled in the art will be able to make various modifications without creative efforts from the above-described conception, and fall within the scope of the present invention.
Claims (10)
1. The intelligent fusion method of the three-dimensional marine environment field based on deep learning is characterized by comprising the following steps:
s1, acquiring multi-source marine data, wherein the multi-source marine data comprises one-dimensional observation data, two-dimensional marine satellite observation data, a three-dimensional gridding marine mode and reanalysis data;
s2, carrying out data processing on the obtained multi-source ocean data, including quality control processing, space-time interpolation processing and normalization processing on the multi-source ocean data;
s3, making a sample set, and dividing a training set, a testing set and a verification set;
s4, building a deep learning network model, training the deep learning network model by adopting a training set, and building an ocean data intelligent fusion model;
and S5, inputting the multi-source ocean data to be fused into an ocean data intelligent fusion model, and outputting an intelligent fusion result.
2. The intelligent fusion method of the three-dimensional marine environment field based on the deep learning of claim 1, further comprising S6, inputting the test set into the marine data intelligent fusion model, outputting the test result of the model, and analyzing the performance of the model according to the speed and parameters of the model during the test; and meanwhile, according to the model test result and the test sample label, calculating and visualizing an error index, and evaluating the precision and the data fusion quality of the verification model.
3. The intelligent fusion method for the deep learning-based three-dimensional marine environment field according to claim 1, wherein the one-dimensional observation data comprises Argo, and sea Wen Pouxian data observed by buoys, submerged buoys and shipborne equipment; the two-dimensional marine satellite observation data comprises sea surface temperature data, sea surface height anomaly data and sea surface wind field data; the three-dimensional gridding ocean mode and reanalysis data comprise HYCOM data, SODA ocean reanalysis data and EN4-analysis data.
4. The intelligent fusion method for the three-dimensional marine environment field based on the deep learning of claim 3, wherein in the step S2, the data processing comprises performing quality control processing, temporal-spatial interpolation processing and normalization processing on the obtained multi-source marine data in sequence.
5. The intelligent fusion method of the three-dimensional marine environment field based on deep learning of claim 4, wherein the quality control processing is performed on the multi-source marine data, specifically: and carrying out query, elimination, correction and filling on abnormal values and missing values of one-dimensional observation data in the multi-source ocean data.
6. The intelligent fusion method of the three-dimensional marine environment field based on the deep learning of claim 5, wherein the spatial-temporal interpolation processing is performed on the multi-source marine data, and specifically comprises the following steps:
a21, interpolating the one-dimensional observation data to a vertical standard layer { z1, z2, …, zk } by adopting a linear interpolation method;
a22, interpolating the one-dimensional observation data which are discretely distributed in the horizontal direction to corresponding grid points of a horizontal grid { x1, x2, …, xm } × { y1, y2, …, yn } by adopting a space-time weight interpolation method;
a23, interpolating the two-dimensional marine satellite observation data to a horizontal grid { x1, x2, …, xm } × { y1, y2, …, yn } by a bilinear interpolation method;
a24, interpolating the three-dimensional gridding ocean mode and the reanalysis data into a three-dimensional grid { x1, x2, …, xm } × { y1, y2, …, yn } × { z1, z2, …, zk } by a bilinear interpolation method;
and A25, interpolating the multisource ocean data subjected to the spatial interpolation to the same time resolution.
7. The intelligent fusion method of the three-dimensional marine environment field based on the deep learning of claim 6, wherein the multi-source marine data is normalized by the following specific steps: and carrying out normalization processing on the multi-source ocean data and the spatio-temporal information data after the spatio-temporal interpolation, wherein the spatio-temporal information data comprise the year, month, longitude and latitude of the place.
8. The intelligent fusion method for the deep learning-based three-dimensional marine environment field according to claim 7, wherein in the step S3, the preparing of the sample set specifically includes the following steps:
s31, taking the one-dimensional observation data interpolated to the vertical direction standard layer as a sample label, setting the number of layers of the vertical direction standard layer as k, and recording the array dimension of the sample label as (1,k);
s32, taking EN4-analysis data, HYCOM data, SODA reanalysis data, a difference value between the EN4-analysis data and the HYCOM data, a difference value between the HYCOM data and the SODA reanalysis data, and a difference value between the EN4-analysis data and the SODA reanalysis data of a horizontal grid position corresponding to the one-dimensional observation data as sample characteristic data to form 6 characteristic channels;
s33, taking the space-time information data where the sample label is located and the two-dimensional marine satellite observation data of the grid corresponding to the sample label as sample characteristic data, forming 1 characteristic channel, making a sample set, and recording the dimension of the sample characteristic array as (7,k).
9. The intelligent fusion method for the three-dimensional marine environment field based on the deep learning according to claim 1, wherein in step S4, a global attention mechanism and an adaptive parameterized ReLU activation function are added to the deep learning network model, wherein the global attention mechanism adopts a serial mode of firstly performing channel attention and then performing spatial attention;
the method is characterized in that the built deep learning network model is trained by adopting a training set, and the method comprises the following specific steps: training the built deep learning network model by adopting a training set, comparing an output result with sample label data, and constructing an intelligent ocean data fusion model by minimizing the difference between the output of the network model and the sample label data and optimizing parameters.
10. Intelligent fusion system of three-dimensional marine environment field based on degree of deep learning, its characterized in that includes:
the data acquisition and processing module is used for acquiring multi-source marine data and processing the data;
the sample set making module is used for analyzing the multi-source ocean data and making the multi-source ocean data into a sample set;
the ocean data intelligent fusion model building module is used for building and training a deep learning network model;
the intelligent fusion result output module is used for inputting the multi-source ocean data to be fused into the ocean data intelligent fusion model and outputting an intelligent fusion result;
and the result verification module is used for evaluating and verifying the performance and the precision of the constructed intelligent ocean data fusion model and the data fusion quality.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211050583.XA CN115393540A (en) | 2022-08-30 | 2022-08-30 | Intelligent fusion method and system of three-dimensional marine environment field based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211050583.XA CN115393540A (en) | 2022-08-30 | 2022-08-30 | Intelligent fusion method and system of three-dimensional marine environment field based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115393540A true CN115393540A (en) | 2022-11-25 |
Family
ID=84123674
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211050583.XA Pending CN115393540A (en) | 2022-08-30 | 2022-08-30 | Intelligent fusion method and system of three-dimensional marine environment field based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115393540A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116597085A (en) * | 2023-05-18 | 2023-08-15 | 中山大学 | Three-dimensional flow field reconstruction method, system, electronic equipment and storage medium |
| CN116758224A (en) * | 2023-03-29 | 2023-09-15 | 国家海洋环境预报中心 | Fusion assimilation method and device for multi-source ocean observation data |
| CN116797756A (en) * | 2023-08-16 | 2023-09-22 | 海博泰科技(青岛)有限公司 | Holographic ocean background field implementation method |
| CN116912459A (en) * | 2023-09-13 | 2023-10-20 | 中国海洋大学 | Variable-grid multi-scale mixed data assimilation method |
| CN117009913A (en) * | 2023-05-05 | 2023-11-07 | 中国人民解放军61741部队 | Sea surface height abnormal data fusion method based on satellite altimeter and tide station |
| CN117150437A (en) * | 2023-10-31 | 2023-12-01 | 航天宏图信息技术股份有限公司 | Multi-source satellite sea surface wind field data fusion method, device, equipment and medium |
| CN117633441A (en) * | 2024-01-24 | 2024-03-01 | 航天宏图信息技术股份有限公司 | Sea surface temperature data filling and correcting method and device and electronic equipment |
| 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 |
| CN117909691A (en) * | 2024-03-15 | 2024-04-19 | 广东海洋大学 | Ocean engineering design data acquisition system and method |
| CN117909691B (en) * | 2024-03-15 | 2024-06-07 | 广东海洋大学 | Ocean engineering design data acquisition system and method |
-
2022
- 2022-08-30 CN CN202211050583.XA patent/CN115393540A/en active Pending
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116758224A (en) * | 2023-03-29 | 2023-09-15 | 国家海洋环境预报中心 | Fusion assimilation method and device for multi-source ocean observation data |
| CN116758224B (en) * | 2023-03-29 | 2024-01-19 | 国家海洋环境预报中心 | Fusion assimilation method and device for multi-source ocean observation data |
| CN117009913B (en) * | 2023-05-05 | 2024-01-30 | 中国人民解放军61741部队 | Sea surface height abnormal data fusion method based on satellite altimeter and tide station |
| CN117009913A (en) * | 2023-05-05 | 2023-11-07 | 中国人民解放军61741部队 | Sea surface height abnormal data fusion method based on satellite altimeter and tide station |
| CN116597085A (en) * | 2023-05-18 | 2023-08-15 | 中山大学 | Three-dimensional flow field reconstruction method, system, electronic equipment and storage medium |
| CN116597085B (en) * | 2023-05-18 | 2023-12-12 | 中山大学 | Three-dimensional flow field reconstruction method, system, electronic equipment and storage medium |
| CN116797756A (en) * | 2023-08-16 | 2023-09-22 | 海博泰科技(青岛)有限公司 | Holographic ocean background field implementation method |
| CN116797756B (en) * | 2023-08-16 | 2023-11-07 | 海博泰科技(青岛)有限公司 | Holographic ocean background field implementation method |
| CN116912459B (en) * | 2023-09-13 | 2023-12-19 | 中国海洋大学 | Variable-grid multi-scale mixed data assimilation method |
| CN116912459A (en) * | 2023-09-13 | 2023-10-20 | 中国海洋大学 | Variable-grid multi-scale mixed data assimilation method |
| CN117150437A (en) * | 2023-10-31 | 2023-12-01 | 航天宏图信息技术股份有限公司 | Multi-source satellite sea surface wind field data fusion method, device, equipment and medium |
| CN117150437B (en) * | 2023-10-31 | 2024-01-30 | 航天宏图信息技术股份有限公司 | Multi-source satellite sea surface wind field data fusion method, device, equipment and medium |
| CN117633441A (en) * | 2024-01-24 | 2024-03-01 | 航天宏图信息技术股份有限公司 | Sea surface temperature data filling and correcting method and device and electronic equipment |
| 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 |
| CN117633441B (en) * | 2024-01-24 | 2024-05-10 | 航天宏图信息技术股份有限公司 | Sea surface temperature data filling and correcting method and device and electronic equipment |
| CN117909691A (en) * | 2024-03-15 | 2024-04-19 | 广东海洋大学 | Ocean engineering design data acquisition system and method |
| CN117909691B (en) * | 2024-03-15 | 2024-06-07 | 广东海洋大学 | Ocean engineering design data acquisition system and method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115393540A (en) | Intelligent fusion method and system of three-dimensional marine environment field based on deep learning | |
| CN110245709B (en) | 3D point cloud data semantic segmentation method based on deep learning and self-attention | |
| Yang et al. | Solar irradiance forecasting using spatio-temporal empirical kriging and vector autoregressive models with parameter shrinkage | |
| CN112818969B (en) | Knowledge distillation-based face pose estimation method and system | |
| CN110555841B (en) | SAR image change detection method based on self-attention image fusion and DEC | |
| CN114724012B (en) | Tropical unstable wave early warning method and device based on space-time cross-scale attention fusion | |
| CN114549925A (en) | Sea wave effective wave height time sequence prediction method based on deep learning | |
| CN111461423B (en) | High-precision gridding air quality inference method, system, terminal equipment and storage medium | |
| CN111428862B (en) | Polar unbalanced space-time combined convection primary short-term prediction method | |
| CN113935249B (en) | Upper-layer ocean thermal structure inversion method based on compression and excitation network | |
| CN112946784B (en) | SuperDARN radar convection diagram short-term forecasting method based on deep learning | |
| CN112308973A (en) | Three-dimensional ocean element field intelligent construction method and system based on machine learning | |
| CN114399073A (en) | Ocean surface temperature field prediction method based on deep learning | |
| Zuo et al. | Ocean temperature prediction based on stereo spatial and temporal 4-D convolution model | |
| CN112183625A (en) | PM based on deep learning2.5High-precision time-space prediction method | |
| CN116720156A (en) | Weather element forecasting method based on graph neural network multi-mode weather data fusion | |
| CN116306203A (en) | Intelligent simulation generation method for marine target track | |
| Meng et al. | Tropical cyclone intensity probabilistic forecasting system based on deep learning | |
| CN114330641A (en) | Method for establishing short-term wind speed correction model based on deep learning of complex terrain | |
| CN117233869A (en) | Site short-term wind speed prediction method based on GRU-BiTCN | |
| CN110826810B (en) | Regional rainfall prediction method combining spatial reasoning and machine learning | |
| Zhang et al. | A Gaussian process regression-based sea surface temperature interpolation algorithm | |
| CN111932017A (en) | Short-term forecasting method suitable for single-station ionized layer TEC | |
| Bosma et al. | Estimating solar and wind power production using computer vision deep learning techniques on weather maps | |
| Xie et al. | Stcanet: Spatiotemporal coupled attention network for ocean surface Current 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 |