CN114266954A - Vortex identification method and device based on graph neural network - Google Patents
Vortex identification method and device based on graph neural network Download PDFInfo
- Publication number
- CN114266954A CN114266954A CN202111675316.7A CN202111675316A CN114266954A CN 114266954 A CN114266954 A CN 114266954A CN 202111675316 A CN202111675316 A CN 202111675316A CN 114266954 A CN114266954 A CN 114266954A
- Authority
- CN
- China
- Prior art keywords
- buoy
- data
- neural network
- track
- graph
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 82
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 25
- 238000012545 processing Methods 0.000 claims abstract description 38
- 238000003062 neural network model Methods 0.000 claims abstract description 36
- 238000006243 chemical reaction Methods 0.000 claims description 12
- 230000008859 change Effects 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000003780 insertion Methods 0.000 claims description 8
- 230000037431 insertion Effects 0.000 claims description 8
- 238000007781 pre-processing Methods 0.000 claims description 8
- 238000010276 construction Methods 0.000 claims description 7
- 238000013507 mapping Methods 0.000 claims description 6
- 238000004140 cleaning Methods 0.000 claims description 4
- 230000002708 enhancing effect Effects 0.000 claims description 3
- 238000004806 packaging method and process Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 230000001131 transforming effect Effects 0.000 claims description 3
- 238000001514 detection method Methods 0.000 abstract description 10
- 239000011159 matrix material Substances 0.000 description 24
- 230000008569 process Effects 0.000 description 20
- 230000006870 function Effects 0.000 description 18
- 230000004913 activation Effects 0.000 description 9
- 238000003860 storage Methods 0.000 description 8
- 238000004422 calculation algorithm Methods 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 239000000463 material Substances 0.000 description 5
- 238000011160 research Methods 0.000 description 5
- 238000012549 training Methods 0.000 description 5
- 238000004590 computer program Methods 0.000 description 4
- 238000013178 mathematical model Methods 0.000 description 4
- 230000010339 dilation Effects 0.000 description 3
- 230000000977 initiatory effect Effects 0.000 description 3
- 108010003272 Hyaluronate lyase Proteins 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 2
- 230000001276 controlling effect Effects 0.000 description 2
- 230000000875 corresponding effect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000008034 disappearance Effects 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000001680 brushing effect Effects 0.000 description 1
- 230000001364 causal effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000013075 data extraction Methods 0.000 description 1
- 230000000916 dilatatory effect Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 238000012067 mathematical method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 230000010355 oscillation Effects 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000004804 winding Methods 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a vortex identification method and device based on a graph neural network, wherein the method comprises the following steps: constructing a graph neural network model according to the acquired sea surface drifting buoy track data; then acquiring a buoy track sequence data set to be detected; and finally, inputting the buoy trajectory sequence data set to be detected into the graph neural network model to obtain a buoy trajectory prediction trajectory. The invention improves the efficiency and the precision and is convenient for detection. The method can be widely applied to the technical field of image processing, and particularly comprises the technical field of ocean phenomenon recognition.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a vortex identification method and device based on a graph neural network.
Background
The secondary mesoscale process can change the potential vortex in the mixed layer, strengthen the communication between the density jump layer and the sea surface, change the upper ocean layering and the mixed layer structure within a few days, and help us to understand how the energy is from mesoscale vortex cascade to small-scale turbulence which can be dissipated as heat. As an intermediate process between meso-scale and micro-scale dynamics, a secondary meso-scale process plays an important role in forward energy cascades, is active at the periphery of meso-scale vortices, and makes a great contribution to the dissipation and improved mixing of the vortices.
In the last two decades, the emission and utilization of the satellite altimeter provides convenience for tracking and detecting the mesoscale vortex, and the method is applied by a plurality of scholars at China and foreign countries and is used for researching the spatial distribution and the propagation characteristics of the mesoscale vortex, the influence on the aspects of material and energy transportation and the like by utilizing satellite remote sensing data, mode data and a field observation data method.
Vortex detection methods are classified into two categories according to the different categories of data: eulerian (Eulerian) and Lagrangian (Lagrangian) methods.
Progress in the field of AI artificial intelligence and development of computer programming languages have brought new research ideas to the development of engineering in various fields in recent years. The machine learning research and construction is a special algorithm rather than a specific algorithm, and a computer can learn in data so as to predict the data at present. The graph neural network method has been primarily applied in the field of scientific engineering and achieves good effects. Therefore, the research on the vortex identification method based on the graph neural network technology is of great significance to the intellectualization of vortex identification and even other ocean phenomenon identification.
The most basic idea is to abstract the actual problem into a mathematical model, continuously improve the model by adjusting and optimizing various parameters in the model, and solve the mathematical model by using a mathematical method, thereby solving the problem. And finally, evaluating the selected mathematical model to judge whether the mathematical model is in accordance with the reality.
The important difficulty lies in the link of converting the practical problem into the mathematical problem. The data of various geometrical characteristics of the ocean secondary mesoscale vortex are further processed to establish a relevant mathematical relation, a selected model is used for fitting, and a fitting result is judged by a computer. Therefore, the problems of data screening loss, large data quantity, manpower and material resource consumption, lack of accuracy in experience judgment and the like can be solved.
The prior art scheme is as follows: the vortex detection method based on the geometric characteristics of the flow field comprises the following steps: the method is realized by judging the geometric characteristics of the numerical simulation field. Obtaining a speed matrix speed according to a warp speed matrix v and a weft speed matrix u of parameters of numerical simulation, searching a speed minimum value in the speed matrix, constraining flow field geometric conditions, judging whether the point is a vortex central point, if so, storing the point, otherwise, removing, and accordingly realizing vortex detection.
Vortex detection algorithm based on height outliers: since the ocean vortex is divided into cyclone (anti-cyclone) vortex, the height outliers of the two types of vortex are distributed differently, and the vortex center is a local minimum for the gas vortex. The center of the vortex of the gas vortex is maximum. And detecting a suspected vortex center according to the height abnormal value, performing further brushing selection by combining with the geometric characteristics of the flow field, reserving the points meeting the conditions, and otherwise, rejecting the points.
The traditional classical detection methods, such as OW algorithm, vector geometry algorithm (VG) and winding angle algorithm (WA), not only need to rely on the research experience of experts to set the threshold, but also need a large amount of calculation for the places with large detection areas, which not only consumes time and labor and has low detection efficiency, but also has unstable obtained detection accuracy.
Disclosure of Invention
In view of this, the embodiment of the present invention provides a method and an apparatus for vortex recognition based on a graph neural network, which are efficient and high in precision.
One aspect of the present invention provides a graph neural network-based vortex identification method, including:
constructing a graph neural network model according to the acquired sea surface drifting buoy track data;
acquiring a buoy track sequence data set to be detected;
and inputting the buoy trajectory sequence data set to be detected into the graph neural network model to obtain a buoy trajectory prediction trajectory.
Optionally, the constructing a graph neural network model according to the acquired sea surface drifting buoy trajectory data includes:
acquiring the track data of a sea surface drifting buoy;
preprocessing the sea surface drifting buoy track data to obtain a target data set;
and constructing a graph learning module, and establishing a graph neural network model according to the graph learning module.
Optionally, the preprocessing the sea surface drifting buoy trajectory data to obtain a target data set includes:
carrying out transformation and enhancement processing on the sea surface drifting buoy trajectory data to obtain a first data set;
and carrying out interpolation processing on the first data set to obtain a feature sequence data set, and taking the feature sequence data set as the target data set.
Optionally, the transforming and enhancing the sea surface drifting buoy trajectory data to obtain a first data set includes:
carrying out longitude and latitude coordinate conversion processing on the sea surface drifting buoy track data to obtain first data;
carrying out noise processing on the first data to obtain second data;
carrying out invalid data cleaning processing on the second data to obtain third data;
packaging the third data into the first data set;
the method comprises the following steps of carrying out longitude and latitude coordinate conversion processing on the sea surface drifting buoy track data to obtain first data, and specifically comprises the following steps: mapping the sea surface drifting buoy track data by adopting an ink card support projection method, and converting the longitude and latitude of the geographic coordinate into an ink card support plane coordinate;
the calculation formula of the longitude and latitude coordinate conversion processing is as follows:
x＝r0×β
y＝r0×q
wherein, the (beta) represents the longitude and latitude of the motion track point of the sea surface drifting buoy; r is0Representing a reference dimension circle radius; a represents the ellipse major radius of the earth; q represents the equivalent dimension; (x, y) rectangular coordinates representing coordinates of the mercator plane; e represents the elliptical first eccentricity.
Optionally, the method further comprises: determining the sea surface drifting buoy by a loop identification method, which specifically comprises the following steps:
judging whether the distance between the current position of the target sea surface drifting buoy and the historical position is smaller than a first threshold value or not, and if yes, determining that the target sea surface drifting buoy returns to the historical position;
wherein the first threshold is determined by a product of a background flow rate and a sample time interval.
Optionally, the interpolating the first data set to obtain a feature sequence data set includes:
constructing a time-space sequence of an original buoy track segment, and determining a time interval between adjacent buoy track points;
traversing a time-space sequence of a buoy track of a loop, and determining the position of an insertion point;
and generating a data point by an interpolation method, and inserting the data point to the position of the insertion point to obtain a feature sequence data set.
Optionally, the acquiring the data set of the trajectory sequence of the buoy to be detected includes:
subtracting the timestamp value of the previous moment from the timestamp value of the current moment to serve as the timestamp change attribute of the current buoy track point;
selecting 8 continuous buoy track points from the buoy track of each loop to form sample data, wherein the first 7 buoy track points are used as historical buoy track points, and the last 1 buoy track point is used as a future buoy track point;
completing construction of a buoy track sequence data set to be detected;
and the buoy track sequence data set to be detected can be input into a graph neural network model for direct calculation.
Optionally, the graph neural network model comprises a graph learning module, a graph convolution module, a time convolution module and an output module;
the graph learning module is used for extracting sparse adjacent matrixes of a plurality of variables from input data;
the graph convolution module is used for processing the spatial dependence among the variables, acquiring the interdependence among the variables and further constructing one-dimensional convolution;
a time convolution module for capturing a time law of a variable by the one-dimensional convolution;
the output module is used for outputting the prediction result of the model;
the graph convolution modules and the time convolution modules are alternately distributed, and a corresponding graph convolution module is connected behind each time convolution module;
the graph convolution module comprises two Mix-Hop propagation layers, wherein the Mix-Hop propagation layers are used for processing inflow information and outflow information of a single node, and the inflow information and the outflow information are added to obtain module output information.
In another aspect, an embodiment of the present invention further provides a vortex identification apparatus based on a graph neural network, including:
the first module is used for constructing a graph neural network model according to the acquired sea surface drifting buoy track data;
the second module is used for acquiring a buoy track sequence data set to be detected;
and the third module is used for inputting the data set of the buoy track sequence to be detected into the graph neural network model to obtain the buoy track prediction track.
Another aspect of the embodiments of the present invention further provides an electronic device, including a processor and a memory;
the memory is used for storing programs;
the processor executes the program to implement the method as described above.
The embodiment of the invention also discloses a computer program product or a computer program, which comprises computer instructions, and the computer instructions are stored in a computer readable storage medium. The computer instructions may be read by a processor of a computer device from a computer-readable storage medium, and the computer instructions executed by the processor cause the computer device to perform the foregoing method.
Firstly, constructing a graph neural network model according to acquired sea surface drifting buoy track data; then acquiring a buoy track sequence data set to be detected; and finally, inputting the buoy trajectory sequence data set to be detected into the graph neural network model to obtain a buoy trajectory prediction trajectory. The invention improves the efficiency and the precision and is convenient for detection.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
FIG. 1 is a flow chart of a method of the present invention;
FIG. 2 is a diagram of the neural network architecture of the present invention;
FIG. 3 is a schematic diagram of a volume module architecture of the present invention;
FIG. 4 is a diagram of a Mix-Hop propagation layer structure of the present invention;
FIG. 5 is a block diagram of the time convolution module of the present invention;
FIG. 6 is a view showing the structure of the expansion initiation layer of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
Aiming at the problems in the prior art, the invention solves the problems of data screening loss, large data quantity, manpower and material resource consumption, lack of accuracy in judgment by depending on experience and the like commonly occurring in the traditional algorithm of vortex recognition by using the graph neural network technology.
Specifically, an embodiment of the present invention provides a vortex identification method based on a graph neural network, including:
constructing a graph neural network model according to the acquired sea surface drifting buoy track data;
acquiring a buoy track sequence data set to be detected;
and inputting the buoy trajectory sequence data set to be detected into the graph neural network model to obtain a buoy trajectory prediction trajectory.
Optionally, the constructing a graph neural network model according to the acquired sea surface drifting buoy trajectory data includes:
acquiring the track data of a sea surface drifting buoy;
preprocessing the sea surface drifting buoy track data to obtain a target data set;
and constructing a graph learning module, and establishing a graph neural network model according to the graph learning module.
Optionally, the preprocessing the sea surface drifting buoy trajectory data to obtain a target data set includes:
carrying out transformation and enhancement processing on the sea surface drifting buoy trajectory data to obtain a first data set;
and carrying out interpolation processing on the first data set to obtain a feature sequence data set, and taking the feature sequence data set as the target data set.
Optionally, the transforming and enhancing the sea surface drifting buoy trajectory data to obtain a first data set includes:
carrying out longitude and latitude coordinate conversion processing on the sea surface drifting buoy track data to obtain first data;
carrying out noise processing on the first data to obtain second data;
carrying out invalid data cleaning processing on the second data to obtain third data;
packaging the third data into the first data set;
the method comprises the following steps of carrying out longitude and latitude coordinate conversion processing on the sea surface drifting buoy track data to obtain first data, and specifically comprises the following steps: mapping the sea surface drifting buoy track data by adopting an ink card support projection method, and converting the longitude and latitude of the geographic coordinate into an ink card support plane coordinate;
the calculation formula of the longitude and latitude coordinate conversion processing is as follows:
x＝r0×β
y＝r0×q
wherein, (, beta) represents the sea chartThe longitude and latitude of the motion track point of the drift buoy; r is0Representing a reference dimension circle radius; a represents the ellipse major radius of the earth; q represents the equivalent dimension; (x, y) rectangular coordinates representing coordinates of the mercator plane; e represents the elliptical first eccentricity.
Optionally, the method further comprises: determining the sea surface drifting buoy by a loop identification method, which specifically comprises the following steps:
judging whether the distance between the current position of the target sea surface drifting buoy and the historical position is smaller than a first threshold value or not, and if yes, determining that the target sea surface drifting buoy returns to the historical position;
wherein the first threshold is determined by a product of a background flow rate and a sample time interval.
Optionally, the interpolating the first data set to obtain a feature sequence data set includes:
constructing a time-space sequence of an original buoy track segment, and determining a time interval between adjacent buoy track points;
traversing a time-space sequence of a buoy track of a loop, and determining the position of an insertion point;
and generating a data point by an interpolation method, and inserting the data point to the position of the insertion point to obtain a feature sequence data set.
Optionally, the acquiring the data set of the trajectory sequence of the buoy to be detected includes:
subtracting the timestamp value of the previous moment from the timestamp value of the current moment to serve as the timestamp change attribute of the current buoy track point;
selecting 8 continuous buoy track points from the buoy track of each loop to form sample data, wherein the first 7 buoy track points are used as historical buoy track points, and the last 1 buoy track point is used as a future buoy track point;
completing construction of a buoy track sequence data set to be detected;
and the buoy track sequence data set to be detected can be input into a graph neural network model for direct calculation.
Optionally, the graph neural network model comprises a graph learning module, a graph convolution module, a time convolution module and an output module;
the graph learning module is used for extracting sparse adjacent matrixes of a plurality of variables from input data;
the graph convolution module is used for processing the spatial dependence among the variables, acquiring the interdependence among the variables and further constructing one-dimensional convolution;
a time convolution module for capturing a time law of a variable by the one-dimensional convolution;
the output module is used for outputting the prediction result of the model;
the graph convolution modules and the time convolution modules are alternately distributed, and a corresponding graph convolution module is connected behind each time convolution module;
the graph convolution module comprises two Mix-Hop propagation layers, wherein the Mix-Hop propagation layers are used for processing inflow information and outflow information of a single node, and the inflow information and the outflow information are added to obtain module output information.
In another aspect, an embodiment of the present invention further provides a vortex identification apparatus based on a graph neural network, including:
the first module is used for constructing a graph neural network model according to the acquired sea surface drifting buoy track data;
the second module is used for acquiring a buoy track sequence data set to be detected;
and the third module is used for inputting the data set of the buoy track sequence to be detected into the graph neural network model to obtain the buoy track prediction track.
Another aspect of the embodiments of the present invention further provides an electronic device, including a processor and a memory;
the memory is used for storing programs;
the processor executes the program to implement the method as described above.
Yet another aspect of an embodiment of the present invention provides a computer-readable storage medium, including a processor;
the processor executes the program when running to implement the method as described previously.
The embodiment of the invention also discloses a computer program product or a computer program, which comprises computer instructions, and the computer instructions are stored in a computer readable storage medium. The computer instructions may be read by a processor of a computer device from a computer-readable storage medium, and the computer instructions executed by the processor cause the computer device to perform the foregoing method.
The following describes in detail a specific implementation process of the vortex identification method according to the present invention with reference to the accompanying drawings:
as shown in fig. 1, an embodiment of the present invention provides a vortex identification method based on a graph neural network, including:
s1: constructing a graph neural network model;
s2: acquiring a vortex characteristic sequence data set to be detected;
s3: inputting the buoy trajectory sequence data set to be detected into a graph neural network model to obtain a buoy trajectory prediction trajectory.
Further, the specific method of S1 is as follows:
s1-1: acquiring the track data of a sea surface drifting buoy, wherein the data source can refer to a global ARGO real-time observation network and the like;
s1-2: processing the vortex characteristic data to obtain a transformed and enhanced data set;
s1-3: and adding a graph learning module to establish a graph neural network model.
Further, the graph neural network model constructed in step S1 includes a graph learning module, m graph convolution modules, m time convolution modules and an output module, the graph learning module calculates an adjacency matrix of input graph data according to the data, the adjacency matrix is used for all the graph convolution modules as the input of the module, the graph convolution modules and the time convolution modules are distributed alternately, and one graph convolution module is followed by another graph convolution module to capture the correlation of the input information in time and space respectively.
Further, the graph convolution module comprises two Mix-Hop propagation layers, and the Mix-Hop propagation layers are combined with the adjacency matrix to process the spatial correlation information of the nodes in the information flow.
The Mix-Hop propagation layer of the key structure is combined with the adjacency matrix to process the spatial correlation information of the nodes in the information flow. Whereas a Mix-Hop propagation layer typically comprises two steps: a propagation process and a selection process. The propagation process of the Mix-Hop propagation layer can be defined as:
wherein, beta is a hyper-parameter used for keeping original node information of what proportion. k denotes the depth of the propagation layer, HinHidden layer input (output of previous layer) representing current layer, HoitHidden layer output State of the Current layer, Hin＝H(0),
Wherein
Wherein A is adjacent matrix and I is unit matrix.
For the information selection process, it is defined as
K is the information propagation depth, HinRepresenting the input of the current layer, which is usually a hidden state of the output of the previous layer, HoutRepresenting the hidden state of the current layer output.
Furthermore, the time convolution module includes two expansion starting layers, in order to extract higher-level time features, in the time convolution module, a plurality of one-dimensional convolution filters are usually used for building the module, but the amount of transmitted information cannot be reasonably controlled by using the filters alone, so that the input of the whole time convolution module output divided into two parts of modules is filtered by the expansion starting layers respectively consisting of a group of one-dimensional convolution filters, and the difference lies in that the activation functions following the expansion starting layers are different. One part of the output is processed by tanh activation function after passing through the expansion start layer, which is used for filtering the input filter, and the other branch of the input is processed by Sigmoid activation function, which is used for controlling the information amount that the filter can pass to the next module. The output module is used for mapping the hidden features in the model into an output space with a proper size. As shown in FIG. 1, the invention is a graph neural network-based sub-mesoscale vortex identification method, which is totally divided into the following steps in the actual operation level:
referring to fig. 1, a graph neural network-based sub-mesoscale vortex prediction method includes:
s1: constructing a graph neural network model;
s2: acquiring a vortex characteristic sequence data set to be detected;
s3: inputting the data set of the buoy track sequence to be detected into a neural network model of the graph to obtain the predicted track of the buoy track, namely a complete vortex loop.
The S1 specifically includes:
s1-1: acquiring the track data of a sea surface drifting buoy;
s1-2: processing the vortex characteristic data to obtain a transformed and enhanced data set;
s1-3: and adding a graph learning module to establish a graph neural network model.
The S1-2 comprises:
s1-2-1: preprocessing the vortex data;
s1-2-2: and carrying out interpolation processing on the data to obtain a feature sequence data set.
The step S1-2-1 includes longitude and latitude coordinate conversion processing and noise and invalid data cleaning processing. The longitude and latitude coordinate conversion processing adopts an ink card support projection method for mapping, and the longitude and latitude of the geographic coordinate is converted into an ink card support plane coordinate, and the formula is as follows:
x＝r0×β
y＝r0×q
the longitude and latitude of the motion track point of the sea surface drift buoy are recorded as (alpha, beta), and the rectangular coordinates in the ink card tray coordinate system obtained by conversion are (x, y), r0The reference dimension is defined as the circle radius, q is defined as the equivalent dimension, a in the formula is the ellipse major radius of the earth, and e is the ellipse first eccentricity.
The ocean surface drift buoy data is one of the frequently used data in the sub-mesoscale vortex research, and vortex is mainly identified by detecting a 'loop' part in the lagrangian locus of the drift buoy. Grouping all the data of the ocean surface drifting buoys into one group according to the same loop, and arranging the data points in each group according to the ascending sequence of the time stamps to obtain a drifting track data point set of the ocean surface drifting buoys representing a certain vortex in a time period.
Loop identification: when the distance between the current position and the previous position of the buoy is less than a threshold value D0When it is time, the float is considered to return to the previous position. Threshold value D0Can be estimated from the product of the background flow rate and the sample time interval. Since the trace samples are evenly spaced in time, they are actually estimated by the average spatial separation of the traces in this region. Consider a series of points P (i) ([ 1, M ]) along the buoy trajectory Γ]And M is the total points. D (i, j) is the distance between P (i) and P (j). At point P (i), search for the first one whose distance D (i, k) is less than D0Point p (k). Search range time [ i + τ, min(i+N,M)]And τ is the number of data points for which the truncation time of the high frequency oscillation is removed. N is the number of data points used to search the longest time taken for the loop. That is, if P (i) is returned, the search is stopped if the number of search steps used is greater than N. Thus, P (k) satisfies the following conditions:
D(i，k)≤D0，i+τ<k<min(i+N,M)
wherein D (i, k) is the distance between P (i) and P (k). All points from P (i) to P (k) are recorded as a set of loop points.
The trajectory of the vortices can be already represented by the buoy trajectory extraction at S1-2-2, but since some buoy points are discarded to some extent during the preprocessing stage and the buoy data extraction stage, the buoy points may cause the time intervals before and after the original buoy sequence to become uneven.
Therefore, it is necessary to generate and interpolate a data point by the interpolation point location identification and interpolation method as described in step S1-2-3.
Let the time-space sequence of the original buoy trace segment be T ═ P1P2P3···Pn}，PiAs the buoy track point, the adjacent buoy track points PiAnd Pi-1The time interval between i is noted as Δ tiAnd then: the insertion point position is identified as traversing a float track space-time sequence T of a loop to find out the position needing interpolation in the sequence, and the interpolation processing aims at smoothing the time interval difference, so the time interval delta T between the front float point and the rear float point in the float track sequence can be usediAs a recognition criterion for the location of the insertion point.
The buoy trajectory sequences obtained through the above steps cannot be directly used as input of the model, and the buoy trajectory sequences obtained through the above steps should be directly used as data sets for model training and prediction only through some construction steps, because the buoy trajectory sequences obtained through extraction have different lengths, and the rotation angles of loops are different, and therefore, the buoy trajectory sequences should be appropriately grouped and divided, so as to construct data sets which can be directly used for model training and prediction.
The construction of the data set comprises the following steps:
the method comprises the following steps: and converting the timestamp attribute in each buoy track sequence into a time stamp change attribute, namely subtracting the time stamp value at the previous moment from the time stamp value at the current moment to serve as the time stamp change attribute of the current buoy track point.
Step two: in each loop, every 8 continuous buoy track points form a sample data, the first 7 buoy track points are used as historical buoy track points, the last one is used as a future buoy track point, and n-7 sample data can be extracted from a buoy track sequence with n continuous buoy track points. And completing the construction of the data set through the steps. The constructed data may be used as an input data set for the model.
Referring to fig. 2, the graph neural network model constructed in the step S1 includes a graph learning module, m graph convolution modules, m time convolution modules and an output module.
The graph learning module is responsible for extracting sparse adjacency matrixes of a plurality of variables from input data, the graph convolution module (GC) is used for processing spatial dependence among the variables, obtaining interdependence among the variables and constructing one-dimensional convolution, and the time convolution module (TC) captures time regularity of the variables through the one-dimensional convolution to complete a prediction task. As shown in fig. 2, the whole model consists of a graph learning layer, m graph convolution modules, m time convolution modules and an output module. The graph learning layer computes from the data an adjacency matrix of the input graph data, which is then used on all graph convolution modules as input to this module. The graph convolution modules and the time convolution modules are distributed alternately, and one time convolution module must be followed by one graph convolution module to capture the correlation of input information in time and space respectively. Before each time convolution module, the input of the time convolution module is collected and is connected to the graph convolution module as a residual, and then the residual and the graph convolution module are used together as the input of the next time convolution module, and the residual connection has the function of avoiding the disappearance of the gradient to a certain extent. And the output module maps the hidden features in the model into an output space with a proper size.
The graph learning module computes from the data an adjacency matrix of the input graph data, which adjacency matrix will then be used on all graph convolution modules as input to this module. The method is used for adaptively constructing the graph adjacency matrix according to input information learning, so that the potential relation between each variable in the multivariate time series is obtained. In order to obtain such an adjacency matrix, a common method is to characterize the similarity between nodes by calculating the distances between variables, and the adjacency matrices obtained in this way are usually bidirectional or symmetrical, but in the prediction task of multivariate time series, the causal relationship that a change of one node causes a change of another node is usually adopted, so that the adjacency matrix to be learned should be unidirectional. In order to obtain a unidirectional adjacency matrix, the module adopts the following mathematical expression method. The graph learning module is obtained by the following method:
M1＝tanh(αE1θ1) (1)
M2＝tanh(αE2θ2) (2)
idx＝argtopk(A[i,:]),i＝1,2,...,N (4)
A[i,-idx]＝0 (5)
wherein E1 and E2 represent randomly initialized node embedding which can be modified during training; theta1And theta2Are parameters of the model; α is a saturation ratio hyperparameter for controlling the saturation state of the activation function. The Argtopk (. X) function returns the first k's large value in the vector.
Equation (3) is to calculate the asymmetry information of the adjacency matrix, where the effect of the adjacency matrix can be regularized using ReLU activation, such as AuvIs positive, then its diagonal element AvuWill be 0 (negative values are 0 under ReLU).
Equations (4) (5) act as sparse adjacency matrices, which can reduce the computational cost of the convolutional network of the subsequent graph. The nearest k nodes of the nodes are selected, so that the number of neighbor nodes can be reduced, and the computational complexity is reduced.
The graph convolution modules and the time convolution modules are distributed alternately, and one time convolution module is followed by one graph convolution module to capture the correlation of input information in time and space respectively.
As shown in fig. 3, the graph convolution module includes two Mix-Hop propagation layers, the two Mix-hops process the inflow information and the outflow information of a single node, respectively, and finally add the two pieces of information to the so-called final module output information. The Mix-Hop propagation layer will process the spatially correlated information of the nodes in the information stream in combination with the adjacency matrix.
The Mix-Hop propagation layer of the key structure is combined with the adjacency matrix to process the spatial correlation information of the nodes in the information flow. Whereas a Mix-Hop propagation layer typically comprises two steps: an information dissemination process and an information selection process. The propagation process of the Mix-Hop propagation layer can be defined as
Wherein, beta is a hyper-parameter used for keeping original node information of what proportion. k represents the depth of the propagation layer,inhidden layer input (output of previous layer) representing current layer, HoutRepresenting the hidden layer output state of the current layer.
Wherein the content of the first and second substances,
for the information selection process, it is defined as
K is the information propagation depth, HinRepresenting the input of the current layer, which is usually a hidden state of the output of the previous layer, HoutIndicating current layer inputThe hidden state is shown. The variables appearing in the above formula are defined as shown in formula (7):
wherein A is adjacent matrix and I is unit matrix.
For the Mix-Hop propagation layer as shown in fig. 4, information flow in the vertical direction is the propagation process of the network, and information transfer in the horizontal direction is the selection process. If equation (6) is used alone, some node information may be lost, especially in the extreme case where spatial dependencies are not present. The node information of the node fusion neighborhood only adds useless noise to the information of the current node, so that the formulas (4) to (7) are used as information selection steps, and the parameter matrix W (k) is introduced to select the characteristics, even if no or only no spatial dependency is needed among the nodes in the graph structure. And (5) adjusting W (k) when all k is greater than 0 to be 0, so that the self information of the original node can be reserved.
As shown in fig. 5, the temporal convolution module includes two dilation initiation layers. In the time convolution module, in order to extract higher-level time features, a plurality of one-dimensional convolution filters are generally used for building the module, but the amount of transmitted information cannot be reasonably controlled by simply using the filters, so that the output of the whole time convolution module is divided into two parts, the inputs of the module are respectively filtered through an expansion starting layer consisting of a group of one-dimensional convolution filters, and the difference lies in that the activation functions connected after the expansion starting layer are different. One part of the output is processed by tanh activation function after expanding the start layer, which acts as a filter to filter the input, while the other branch input is processed by Sigmoid activation function to control the amount of information the filter can pass to the next module.
In order to find various ranges of time patterns and process a long time sequence in the time convolution module, a filter with a plurality of convolutions is used in the expansion starting layer in the time convolution module, and the expansion convolution is adopted.
For a convolutional network, the size of the filter directly determines the performance of the network, a kernel that is too large to represent the short-term signal pattern finely, and a kernel that is too large to find the long-term signal pattern sufficiently, and in image processing, the kernel size is usually determined by an initial strategy, that is, the outputs of one-dimensional convolutional filters with 3 different sizes (1 × 1, 3 × 3, and 5 × 5) are connected together. It is inspired by this that since time signals tend to have several inherent periods, e.g. 7, 12, 24, 28 and 60, etc., which are not well contained by the stack of receive layers with filter sizes of 1 × 1, 1 × 3 and 1 × 5, an underlying structure of the expansion starting layer is composed of filter sizes of 1 × 2, 1 × 3, 1 × 6 and 1 × 7, as shown in fig. 6, so that the above-mentioned time period can be covered by a combination of these filter sizes. For example, to represent a time period of 12, the model may pass the input through a 1 × 7 filter and then through a 1 × 6 filter. On the other hand, the size of the acceptance domain of a convolution network increases linearly with the depth of the network and the size of the kernel, and for a one-dimensional convolution network with m convolution kernels having the size of c, the size of the acceptance domain is the formula:
R＝(c-1)+1 (4-11)
thus, if very long sequences are to be processed, a very deep network (m is large) or kernel filter (c is large) is required, the complexity of the model thus constructed is high, while the dilation convolution is used to reduce the complexity of the model, which runs a standard convolution filter on the down-sampled input with a certain frequency, e.g. when the dilation factor is 2, it applies the standard convolution on the input for every two samples. Letting the spreading factor of each layer grow exponentially at a rate of q (q >1), assuming an initial spreading factor of 1, the receive domain size of an m-layer expanded convolutional network with kernel size c is, according to the above equation (4-11):
R＝1+(c1)(qm1)/(q1)
the equations (4-11) indicate that the receive domain size of the network grows exponentially with the ratio q as the hidden layer increases. For a one-dimensional sequence input z ∈ RTAnd one is composed of1×2∈R2,f1×3∈R3,f1×6∈R6,f1×7∈R7Filter (2)Wave filter, form of the expansion initiation layer such as:
z＝concat(z*f1×2，z*f1×3，z*f1×6，z*f1×7) (4-12)
wherein d is a dilating factor. The output module is used for mapping the hidden features in the model into an output space with a proper size.
Before model training, hyper-parameters related to the model are set, for example, the model adopts an Adam optimizer with a gradual clipping threshold of 5, wherein the gradual clipping is a method for effectively solving the problem of model gradient explosion. Further, the model learning rate is set to 1 × 10-3L2 regularization penalty threshold set to 1 × 10-4And a Droupout layer with a discarding rate of 0.4 is connected after each time convolution module to prevent the over-fitting problem. And a Layer Norm normalization Layer is connected behind each graph convolution module, and the normalization Layer can effectively make the data distribution consistent and avoid the problem of gradient disappearance. The depth of the Mix-Hop propagation layer in the graph convolution module is set to 3, the information retention rate from the Mix-Hop propagation layer is set to 0.06, the saturation rate of the activation function in the graph learning layer is set to 3, and finally the dimension of the static feature is set to 40.
On the structure of the model, 5 graph convolution modules and 5 time convolution modules are connected in an interleaving sequence, and for 1 × 1 convolution of the inlet of the model, the graph convolution module and the time convolution module have one input channel and 16 output channels, and the graph convolution module and the time convolution module also have 16 output channels. In the output module, the first layer of the module is provided with 32 output channels. In some embodiments, the second level of output modules is set to 1 output channel, since the 6 th float track point is to be predicted from the 5 historical float track points. Setting the training Epoch to be 30, setting the number of the adjacent nodes to be 30, and finally setting the Batch Size to be 4, so as to construct the model structure of the prediction task to be executed in the embodiment.
In summary, since the graph data has an extremely wide application scenario, the present invention utilizes the advantages of GNN, and has the following advantages:
1. GNNs have a strong ability to fit map data.
2. GNNs have powerful reasoning capabilities.
3. The combination of GNN with knowledge maps.
The vortex identification method based on the graph neural network technology has important significance for the prediction of vortex and the intellectualization of other ocean phenomenon identification.
The invention utilizes the huge calculation operation function of the computer, and can efficiently screen a large amount of ocean flow field data, thereby solving the problem that the traditional technology is time-consuming and labor-consuming. On the other hand, the machine also solves the problems that the traditional technology selects the threshold value by the experience judgment of expert scholars, and the like, which possibly causes judgment errors and large errors.
In alternative embodiments, the functions/acts noted in the block diagrams may occur out of the order noted in the operational illustrations. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality/acts involved. Furthermore, the embodiments presented and described in the flow charts of the present invention are provided by way of example in order to provide a more thorough understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and in which sub-operations described as part of larger operations are performed independently.
Furthermore, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise stated to the contrary, one or more of the described functions and/or features may be integrated in a single physical device and/or software module, or one or more functions and/or features may be implemented in a separate physical device or software module. It will also be appreciated that a detailed discussion of the actual implementation of each module is not necessary for an understanding of the present invention. Rather, the actual implementation of the various functional modules in the apparatus disclosed herein will be understood within the ordinary skill of an engineer, given the nature, function, and internal relationship of the modules. Accordingly, those skilled in the art can, using ordinary skill, practice the invention as set forth in the claims without undue experimentation. It is also to be understood that the specific concepts disclosed are merely illustrative of and not intended to limit the scope of the invention, which is defined by the appended claims and their full scope of equivalents.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
The logic and/or steps represented in the flowcharts or otherwise described herein, e.g., an ordered listing of executable instructions that can be considered to implement logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or more wires, a portable computer diskette (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium could even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, via for instance optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner if necessary, and then stored in a computer memory.
It should be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or combination of the following techniques, which are known in the art, may be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application specific integrated circuit having an appropriate combinational logic gate circuit, a Programmable Gate Array (PGA), a Field Programmable Gate Array (FPGA), or the like.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
While embodiments of the invention have been shown and described, it will be understood by those of ordinary skill in the art that: various changes, modifications, substitutions and alterations can be made to the embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.
While the preferred embodiments of the present invention have been illustrated and described, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (10)
1. The vortex identification method based on the graph neural network is characterized by comprising the following steps:
constructing a graph neural network model according to the acquired sea surface drifting buoy track data;
acquiring a buoy track sequence data set to be detected;
and inputting the buoy trajectory sequence data set to be detected into the graph neural network model to obtain a buoy trajectory prediction trajectory.
2. The vortex identification method based on the graph neural network according to claim 1, wherein the constructing of the graph neural network model according to the acquired sea surface drifting buoy trajectory data comprises:
acquiring the track data of a sea surface drifting buoy;
preprocessing the sea surface drifting buoy track data to obtain a target data set;
and constructing a graph learning module, and establishing a graph neural network model according to the graph learning module.
3. The vortex identification method based on the graph neural network according to claim 2, wherein the preprocessing the sea surface drifting buoy trajectory data to obtain a target data set comprises:
carrying out transformation and enhancement processing on the sea surface drifting buoy trajectory data to obtain a first data set;
and carrying out interpolation processing on the first data set to obtain a feature sequence data set, and taking the feature sequence data set as the target data set.
4. The graph neural network-based vortex identification method of claim 3, wherein the transforming and enhancing the sea surface drifting buoy trajectory data to obtain a first data set comprises:
carrying out longitude and latitude coordinate conversion processing on the sea surface drifting buoy track data to obtain first data;
carrying out noise processing on the first data to obtain second data;
carrying out invalid data cleaning processing on the second data to obtain third data;
packaging the third data into the first data set;
the method comprises the following steps of carrying out longitude and latitude coordinate conversion processing on the sea surface drifting buoy track data to obtain first data, and specifically comprises the following steps: mapping the sea surface drifting buoy track data by adopting an ink card support projection method, and converting the longitude and latitude of the geographic coordinate into an ink card support plane coordinate;
the calculation formula of the longitude and latitude coordinate conversion processing is as follows:
x＝r0×β
y＝r0×q
wherein, (a, beta) represents the longitude and latitude of the motion track point of the ocean surface drifting buoy; r is0Representing a reference dimension circle radius; a represents the ellipse major radius of the earth; q represents the equivalent dimension; (x, y) stands for mercatorRectangular coordinates of the plane coordinates; e represents the elliptical first eccentricity.
5. The graph neural network-based vortex identification method of claim 4, further comprising:
determining the sea surface drifting buoy by a loop identification method, which specifically comprises the following steps:
judging whether the distance between the current position of the target sea surface drifting buoy and the historical position is smaller than a first threshold value or not, and if yes, determining that the target sea surface drifting buoy returns to the historical position;
wherein the first threshold is determined by a product of a background flow rate and a sample time interval.
6. The method of claim 3, wherein the interpolating the first data set to obtain a feature sequence data set comprises:
constructing a time-space sequence of an original buoy track segment, and determining a time interval between adjacent buoy track points;
traversing a time-space sequence of a buoy track of a loop, and determining the position of an insertion point;
and generating a data point by an interpolation method, and inserting the data point to the position of the insertion point to obtain a feature sequence data set.
7. The graph neural network-based vortex identification method according to claim 1, wherein the acquiring the data set of the trajectory sequence of the buoy to be detected comprises:
subtracting the timestamp value of the previous moment from the timestamp value of the current moment to serve as the timestamp change attribute of the current buoy track point;
selecting 8 continuous buoy track points from the buoy track of each loop to form sample data, wherein the first 7 buoy track points are used as historical buoy track points, and the last 1 buoy track point is used as a future buoy track point;
completing construction of a buoy track sequence data set to be detected;
and the buoy track sequence data set to be detected can be input into a graph neural network model for direct calculation.
8. The graph neural network-based vortex identification method of claim 1, wherein the graph neural network model comprises a graph learning module, a graph convolution module, a time convolution module, and an output module;
the graph learning module is used for extracting sparse adjacent matrixes of a plurality of variables from input data;
the graph convolution module is used for processing the spatial dependence among the variables, acquiring the interdependence among the variables and further constructing one-dimensional convolution;
a time convolution module for capturing a time law of a variable by the one-dimensional convolution;
the output module is used for outputting the prediction result of the model;
the graph convolution modules and the time convolution modules are alternately distributed, and a corresponding graph convolution module is connected behind each time convolution module;
the graph convolution module comprises two Mix-Hop propagation layers, wherein the Mix-Hop propagation layers are used for processing inflow information and outflow information of a single node, and the inflow information and the outflow information are added to obtain module output information.
9. Vortex recognition device based on graph neural network, characterized by, includes:
the first module is used for constructing a graph neural network model according to the acquired sea surface drifting buoy track data;
the second module is used for acquiring a buoy track sequence data set to be detected;
and the third module is used for inputting the data set of the buoy track sequence to be detected into the graph neural network model to obtain the buoy track prediction track.
10. An electronic device comprising a processor and a memory;
the memory is used for storing programs;
the processor executing the program realizes the method of any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111675316.7A CN114266954A (en) | 2021-12-31 | 2021-12-31 | Vortex identification method and device based on graph neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111675316.7A CN114266954A (en) | 2021-12-31 | 2021-12-31 | Vortex identification method and device based on graph neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114266954A true CN114266954A (en) | 2022-04-01 |
Family
ID=80832398
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111675316.7A Pending CN114266954A (en) | 2021-12-31 | 2021-12-31 | Vortex identification method and device based on graph neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114266954A (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180308371A1 (en) * | 2017-04-19 | 2018-10-25 | Beihang University | Joint search method for uav multiobjective path planning in urban low altitude environment |
| CN110836675A (en) * | 2019-10-25 | 2020-02-25 | 北京航空航天大学 | Decision tree-based automatic driving search decision method |
| CN113505878A (en) * | 2021-07-09 | 2021-10-15 | 兰州大学 | Flight path prediction method based on graph neural network |
-
2021
- 2021-12-31 CN CN202111675316.7A patent/CN114266954A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180308371A1 (en) * | 2017-04-19 | 2018-10-25 | Beihang University | Joint search method for uav multiobjective path planning in urban low altitude environment |
| CN110836675A (en) * | 2019-10-25 | 2020-02-25 | 北京航空航天大学 | Decision tree-based automatic driving search decision method |
| CN113505878A (en) * | 2021-07-09 | 2021-10-15 | 兰州大学 | Flight path prediction method based on graph neural network |
Non-Patent Citations (3)
| Title |
|---|
| 董昌明 等: "海洋涡旋自动探测几何方法、涡旋数据库及其应用", 海洋科学进展, pages 439 - 453 * |
| 董昌明: "《海洋涡旋探测与分析》", 31 December 2015, 科学出版社, pages: 26 - 35 * |
| 黄艳松;宋金宝;范聪慧;: "海气通量涡相关法计算中的时间尺度分析", 海洋科学, no. 11, pages 116 - 121 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Asadollah et al. | River water quality index prediction and uncertainty analysis: A comparative study of machine learning models | |
| Choi et al. | Real-time significant wave height estimation from raw ocean images based on 2D and 3D deep neural networks | |
| Liu et al. | Fg-net: A fast and accurate framework for large-scale lidar point cloud understanding | |
| Garland et al. | Exploring the topology of dynamical reconstructions | |
| CN106250931A (en) | A kind of high-definition picture scene classification method based on random convolutional neural networks | |
| Partal | Modelling evapotranspiration using discrete wavelet transform and neural networks | |
| Do et al. | Parallel computing flow accumulation in large digital elevation models | |
| Meyer | The watershed concept and its use in segmentation: a brief history | |
| CN111753682B (en) | Hoisting area dynamic monitoring method based on target detection algorithm | |
| Guo et al. | Network pruning for remote sensing images classification based on interpretable CNNs | |
| CN101299233A (en) | Device and method for realizing moving object identification and track based on FPGA | |
| Thongniran et al. | Spatio-temporal deep learning for ocean current prediction based on HF radar data | |
| Li et al. | A spatiotemporal directed graph convolution network for ultra-short-term wind power prediction | |
| CN113505878B (en) | Flight path prediction method based on graph neural network | |
| Mao et al. | Deep learning-enhanced extraction of drainage networks from digital elevation models | |
| Du et al. | Srh-net: Stacked recurrent hourglass network for stereo matching | |
| Karbasi et al. | Developing a novel hybrid Auto Encoder Decoder Bidirectional Gated Recurrent Unit model enhanced with empirical wavelet transform and Boruta-Catboost to forecast significant wave height | |
| CN114266954A (en) | Vortex identification method and device based on graph neural network | |
| Feng et al. | Graph convolution based spatial-temporal attention lstm model for flood forecasting | |
| Pang et al. | A novel multivariable hybrid model to improve short and long-term significant wave height prediction | |
| CN111830595A (en) | Meteorological element prediction method and equipment | |
| Giang et al. | Adaptive Spatial Complex Fuzzy Inference Systems With Complex Fuzzy Measures | |
| Zeng et al. | Adversarial network for edge detection | |
| CN109558819A (en) | A kind of depth network light weight method for Remote Sensing Target detection | |
| Brun et al. | Hierarchical graph encodings |
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 |