CN111539453B - Global ionized layer electron total content prediction method based on deep cycle neural network - Google Patents
Global ionized layer electron total content prediction method based on deep cycle neural network Download PDFInfo
- Publication number
- CN111539453B CN111539453B CN202010238027.XA CN202010238027A CN111539453B CN 111539453 B CN111539453 B CN 111539453B CN 202010238027 A CN202010238027 A CN 202010238027A CN 111539453 B CN111539453 B CN 111539453B
- Authority
- CN
- China
- Prior art keywords
- thermodynamic diagram
- global
- thermodynamic
- prediction
- electron
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Analysing Materials By The Use Of Radiation (AREA)
Abstract
The invention discloses a global ionosphere total electron content prediction method based on a deep cycle neural network, which comprises a training phase and a prediction phase, wherein the training phase comprises the following steps: 1. acquiring global ionospheric electron total content thermodynamic diagrams, and forming an original image sequence after adjusting the horizontal position; 2. constructing a training sample set; 3. constructing a global ionosphere electron total content prediction model based on a deep circulation neural network, and training by utilizing a training sample set; the prediction phase comprises: 4. collecting K global total ionospheric electron content thermodynamic diagrams every day for t days continuously; adjusting the horizontal position of a pixel of the acquired image, establishing a prediction sample, and taking the prediction sample as the input of a global ionosphere electron total content prediction model to obtain a prediction thermodynamic diagram; 5. and carrying out longitude sequencing on the predicted thermodynamic diagrams to obtain the predicted global ionospheric electron total content thermodynamic diagrams. The method combines the changes of the ionized layer in space and time, fully and effectively utilizes the existing observation data, and improves the prediction precision.
Description
Technical Field
The invention belongs to the field of ionosphere detection, and particularly relates to a method for predicting the total electron content of a global ionosphere based on a deep cycle neural network.
Background
The ionosphere is an important component of the geospatial environment, wherein the electron content of the ionosphere is an important physical characteristic parameter of the ionosphere, and the ionosphere electron content not only needs to be researched on the space-time change rule of the electron content of the ionosphere, but also needs to be forecasted. At present, the ionospheric electron content forecast can be divided into long-period forecast and short-period forecast according to different forecast time lengths. The forecasting methods are used for establishing a mathematical model by using electronic content time sequence data observed in a certain geographic position for forecasting. The change of the electron content of the ionized layer is continuously changed in time dimension and space, the traditional method only considers the change of the electron content of the ionized layer in time and ignores the correlation in space, so that the time-space change of the electron content of the ionized layer needs to be analyzed and forecasted by fully utilizing the observed data of the electron content of the ionized layer, and the high-precision ionized layer electron content forecasting model has important significance for the related research of the ionized layer.
Disclosure of Invention
The purpose of the invention is as follows: the invention aims to provide a global ionosphere electron total content prediction method which can obviously improve the prediction precision.
The technical scheme is as follows: the invention adopts the following technical scheme:
the global ionospheric total electron content prediction method based on the deep cycle neural network comprises a training phase and a prediction phase, wherein the training phase comprises the following steps:
(1) collecting K global ionospheric electron total content thermodynamic diagrams at equal time intervals every day for N days continuously; for each collected image, adjusting the horizontal position according to the position of each pixel point, so that the positions of the pixel points with the same vertical coordinate are increased along the positive direction of the horizontal axis; constructing an original image sequence S ═ { Pic) according to an acquisition orderk,n}，k＝1,2,…,K，n＝1,2,…,N；
(2) Constructing a training sample set, wherein the mth sample s in the training sample setmThe method comprises the steps of inputting sample data and outputting the sample data, wherein the input sample data comprises t days of original image sequences starting from the mth day in an original image sequence S and t-1 days of primary image differential sequences formed by the t days of original image sequences; outputting sample data as an original image sequence of the m + t days in the S; t is an integer greater than 2;
(3) constructing a global ionosphere electron total content prediction model based on a deep cycle neural network, wherein the model comprises an original thermodynamic diagram branch, a differential thermodynamic diagram branch and a convolution subnet;
the original thermodynamic diagram branch and the differential thermodynamic diagram branch are both deep circulation neural networks with the same structure, and each deep circulation neural network comprises a first convolution long-time and short-time memory network, an expansion convolution long-time and short-time memory network and a second convolution long-time and short-time memory network which are sequentially connected;
the input of the original thermodynamic diagram branch is an original image sequence of t days, and the output is a one-day K thermodynamic diagram sequence predicted according to the input; the input of the differential thermodynamic diagram branch is a primary differential sequence of images for t-1 days, and the output is a predicted K differential thermodynamic diagram sequence for one day;
the convolution subnet recovers the differential thermodynamic diagrams output by the differential thermodynamic diagram branch into thermodynamic diagram data, the recovered thermodynamic diagram data and the thermodynamic diagram sequence output by the original thermodynamic diagram branch are stacked into a dual-channel image sequence, and the stacked dual-channel image sequence is convolved to obtain the predicted global ionospheric electron total content thermodynamic diagram;
training the global ionospheric total electron content prediction model by using a training sample set: taking t-day original image sequences in input sample data of a training sample as the input of an original thermodynamic diagram branch of the global ionospheric electronic total content prediction model, taking t-1-day primary image difference sequences as the input of a differential thermodynamic diagram branch, comparing an obtained prediction result with output sample data in the training sample to calculate errors, correcting parameters to be solved in the error back propagation model, and obtaining a final global ionospheric electronic total content prediction model through iterative training;
the prediction phase comprises:
(4) collecting K global ionospheric electron total content thermodynamic diagrams at equal time intervals every day for t days continuously; for each collected image, adjusting the horizontal position according to the position of each pixel point, so that the positions of the pixel points with the same vertical coordinate are positively increased along the horizontal axis to form a to-be-predicted image sequence T ═ Pk,ττ ═ 1,2, …, t; establishing prediction samples comprising a sequence of predicted images and a sequence of primary difference images { Δ P } of the sequence of predicted imagesk,τ+1}; will { Pk,τAnd { Δ P }k,τ+1Respectively as global ionized layer electron total content prediction modelInputting an original thermodynamic diagram branch and a differential thermodynamic diagram branch in the model, and outputting the model as a prediction thermodynamic diagram;
(5) and carrying out longitude sequencing on the predicted thermodynamic diagrams to obtain the predicted global ionospheric electron total content thermodynamic diagrams.
The step (1) further includes resizing and pixel value normalization of the images in the image sequence S, and the step (5) further includes resizing and inverse normalization of the predictive thermodynamic diagram.
Adjusting the horizontal position according to the position of the pixel point in the steps (1) and (4), comprising:
calculating the local time of each pixel point in the global ionosphere electron total content thermodynamic diagram in the range direction; keeping the vertical coordinates of the pixel points unchanged, sorting the pixel points from small to large according to places, and adjusting the horizontal positions of the pixels according to a sorting result to increase the positions of the pixel points with the same vertical coordinates along the positive direction of the horizontal axis.
And recovering the differential thermodynamic diagrams output by the global ionospheric electron total content prediction model convolution subnet according to the following formula into thermodynamic diagram data:
where K is 1,2, …, K,
a differential thermodynamic diagram sequence output for the differential thermodynamic diagram branch; f. oft,kInputting an original image sequence of the t day in the original image sequence for the original thermodynamic diagram branch;
is the recovered thermodynamic diagram data.
The longitude sorting of the predictive thermodynamic diagrams in the step (5) comprises the following steps:
calculating the longitude of each pixel point according to the local time and the world time of each pixel point; keeping the latitude direction coordinate unchanged, sorting the pixel points from small to large according to the longitude values, and adjusting the horizontal positions of the pixels according to the sorting result to increase the longitude values of the pixel points with the same vertical coordinate along the positive direction of the horizontal axis.
Has the advantages that: compared with the prior art, the method has the advantages that,
drawings
FIG. 1 is a diagram of a global ionospheric total electron content prediction model constructed in accordance with the present invention;
FIG. 2 is a comparison graph of predicted results of two methods in the examples.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention are described below with reference to the accompanying drawings.
The invention discloses a global ionosphere total electron content prediction method based on a deep cycle neural network, which comprises a training phase and a prediction phase, wherein the training phase comprises the following steps:
step 1, collecting K global ionospheric electron total content thermodynamic diagrams at equal time intervals every day for N consecutive days; for each collected image, adjusting the horizontal position according to the position of each pixel point, so that the positions of the pixel points with the same vertical coordinate are increased along the positive direction of the horizontal axis; constructing an original image sequence S ═ { Pic) according to an acquisition orderk,n}，k＝1,2,…,K，n＝1,2,…,N；
In this embodiment, ionosphere electron content grid data GIM provided by International GNSS Service (IGS) is adopted, the longitude range of the grid data coverage is-180 °, the latitude range is-87.5 °, the pixel scale of a single grid is latitude 2.5 ° × longitude 5.0 °, the time resolution is 2 hours, that is, data is collected every 2 hours, and 12 pieces of data are collected every day. According to the parameters of the GIM, the total pixel number of the global total ionospheric electron content thermodynamic diagram collected at each moment is 71 × 73, the horizontal direction is the longitudinal direction, and the vertical direction is the latitudinal direction. When each pixel can calculate the place according to the corresponding longitude and latitude, the place is only related to the longitude and is not related to the latitude.
Adjusting the horizontal position of the collected image according to the position of the pixel point, and specifically comprising the following steps:
calculating the local time of each pixel point in the global ionosphere electron total content thermodynamic diagram in the range direction; keeping the vertical coordinates of the pixel points unchanged, sorting the pixel points from small to large according to places, and adjusting the horizontal positions of the pixels according to a sorting result to increase the positions of the pixel points with the same vertical coordinates along the positive direction of the horizontal axis.
In the thermodynamic diagram with the horizontal position adjusted, the position corresponding to the pixel at the center of the image is 12 pm, and the position in the horizontal direction increases from 0 to 24 pm in the positive direction along the horizontal axis. For subsequent calculation, the thermal diagram after the horizontal position adjustment is adjusted to 72 × 72 in pixel resolution, and the pixel values are normalized.
In this example, 91 days from 2016 (04/01/04) to 2016 (04/04) were selected. Therefore, in this embodiment, K is 12 and N is 91.
Step 2, constructing a training sample set, wherein the mth sample s in the training sample setmIncluding inputting sample data and outputting sample data, sm＝(datain,dataout) Wherein sample data is inputinThe method comprises the following steps of (1) including t days of original image sequences from the m-th day in an original image sequence S, and t-1 days of primary image differential sequences formed by the t days of original image sequences:
datain＝(Pick,m,…,Pick,m+t-1,ΔPick,m+1,…,ΔPick,m+t-1)
t is an integer greater than 2, Δ Pick,m+1A global ionospheric electron total content thermodynamic diagram first difference image, delta Pic, for two adjacent days corresponding to the timek,m+1＝Pick,m+1-Pick,m；
Output sample dataoutFor the original image sequence on day m + t in S: dataout＝(Pick,m+t)；
Step 3, constructing a global ionosphere electron total content prediction model based on a deep cycle neural network, wherein the model comprises an original thermodynamic diagram branch, a differential thermodynamic diagram branch and a convolution subnet as shown in fig. 1;
the original thermodynamic diagram branch and the differential thermodynamic diagram branch are both depth cycle neural networks with the same structure, and each depth cycle neural network comprises a first Convolution Long-Term Memory network CLSTM1, an expansion Convolution Long-Term Memory network DCLSTM (dimension Convolution Long Short-Term Memory) and a second Convolution Long-Term Memory network CLSTM2 which are sequentially connected;
the dilated convolution long-time memory network replaces the convolution in CLSTM with dilated convolution, which has the same operation as general convolution except that delta input widths can be skipped at one time. δ is a coefficient of expansion, and when δ is 1, the expansion convolution becomes a general convolution operation.
Input I of original thermodynamic diagram branchonoOutputting a one-day K-tension chart sequence predicted according to input for an original image sequence of t days; input delta I of differential thermodynamic diagram branchonoOutputting a predicted K differential thermodynamic diagram sequence of one day for the primary differential sequence of the images of the t-1 day;
the convolution sub-network recovers the differential thermodynamic diagrams output by the differential thermodynamic diagram branches into thermodynamic diagram data, such as the thermodynamic diagram data shown in FIG. 1
A symbol represents a recovery operation of the differential thermodynamic diagram; stacking the recovered thermodynamic data with the thermodynamic sequence of the original thermodynamic tributary output as a two-channel image sequence, as in FIG. 1
The symbol represents a stacking operation. Convolving the stacked two-channel image sequence by convolution layer conv0 to obtain a predicted K global ionospheric electron total content thermodynamic diagram;
the output of the differential thermodynamic diagram branch is K differential thermodynamic diagram sequences
And recovering the output differential thermodynamic diagrams in the convolution sub-network into thermodynamic diagram data according to the following formula:
wherein K is 1,2, …, K, ft,kFor the original image sequence of the t day in the original image sequence input by the original thermodynamic branch, in the training phase, ft,kIs Pick,m+t-1；
Is the recovered thermodynamic diagram data.
Training the global ionospheric total electron content prediction model by using a training sample set:
setting input channel, output channel, convolution kernel, stride and boundary filling parameters of each layer, where the parameters of each layer in this embodiment are shown in table 1:
TABLE 1 prediction model parameters of the total electron content of the global ionosphere
| Layer name | Input channel | Output channel | Convolution kernel | Coefficient of expansion | Stride length | Boundary filling |
| CLSTM1 | 1 | 8 | 3X3 | 1 | -- | 1 |
| DCLSTM | 8 | 8 | 3X3 | 2 | -- | 2 |
| CLSTM2 | 8 | 1 | 3X3 | 1 | -- | 1 |
| Conv0 | 2 | 1 | 3X3 | -- | 1 | 1 |
In Table 1, "- -" indicates that this parameter is not necessarily set.
Secondly, setting a training round, a model loss function, a model optimization mode, a learning rate and a single training sample number; in the embodiment, the training round is 15, and an L1 loss function is used as a model loss function; selecting Adam as a model optimization mode, and setting the learning rate to be 0.001; the number of samples for a single training is 16.
Input sample number of training samplesT days from the original image sequence Pick,m,…,Pick,m+t-1The primary difference sequence delta Pic of t-1 day images is used as the input of an original thermodynamic diagram branch of a global ionospheric electron total content prediction modelk,m+1,…,ΔPick,m+t-1As the input of the differential thermodynamic diagram branch, the obtained prediction result and the output sample data in the training sample are obtainedoutComparing the calculation errors, performing iterative training on parameters to be solved in the error back propagation correction model to obtain a final global total ionospheric electron content prediction model;
the prediction phase comprises:
step 4, collecting K global ionospheric electron total content thermodynamic diagrams at equal time intervals every day for t days continuously; for each collected image, according to the method in step 1, the horizontal position is adjusted according to the position of each pixel point, so that the positions of the pixel points with the same vertical coordinate are increased along the positive direction of the horizontal axis, the pixel resolution is adjusted, the pixels are normalized, and a to-be-predicted image sequence T is formed, wherein the to-be-predicted image sequence T is { P ═ Pk,ττ ═ 1,2, …, t; establishing prediction samples comprising a sequence of predicted images and a sequence of primary difference images { Δ P } of the sequence of predicted imagesk,τ+1}; will { Pk,τAnd { Δ P }k,τ+1Respectively serving as the input of an original thermodynamic diagram branch and a differential thermodynamic diagram branch in a global ionospheric electronic total content prediction model, wherein the output of the model is a prediction thermodynamic diagram;
and 5, carrying out longitude sequencing on the predicted thermodynamic diagrams to obtain the predicted global ionospheric total electron content thermodynamic diagrams.
In the predictive thermodynamic diagram, the horizontal direction is the direction of increase when the horizontal direction is a place, and is not consistent with the longitude information of the space, so the longitudes need to be sorted to make the horizontal direction consistent with the longitude direction, and the specific steps are as follows:
calculating the longitude of each pixel point according to the local time and the world time of each pixel point; keeping the latitude direction coordinate unchanged, sorting the pixel points from small to large according to the longitude values, and adjusting the horizontal positions of the pixels according to the sorting result to increase the longitude values of the pixel points with the same vertical coordinate along the positive direction of the horizontal axis.
And (3) recovering the size of the prediction thermodynamic diagrams after the longitude sorting, adjusting the pixel resolution of the prediction thermodynamic diagrams to 71 x 73, and performing inverse normalization on pixel values to obtain a global ionospheric electron total content thermodynamic diagram sequence predicted for the t +1 th day.
According to the global ionosphere electron total content prediction model based on the deep cyclic neural network, the depth of the cyclic neural network is increased through the first convolution long-time memory network, the expansion convolution long-time memory network and the second convolution long-time memory network, the learning capacity of the network is enhanced, and then the time-space change rule in an ionosphere map sequence is analyzed and refined, wherein the expansion convolution long-time memory network improves the receptive field through expansion convolution; and finally, further processing the image characteristics output by the two branches through a convolution subnet to obtain a final global ionosphere electron total content prediction image.
In the embodiment, a global ionospheric total electron content thermodynamic diagram collected from 2016, 04, 05 and 15 to 2016, 12 and 15 is used as a verification sample to verify the feasibility and the accuracy of the prediction method disclosed by the invention. The following three methods are respectively adopted to predict the period from 2016, 04/05/2016 to 2016, 12/2016 and 15/2016:
the method A comprises the following steps: the global total ionospheric electron content prediction is carried out by using a training sample without adopting primary image differential data, and a global total ionospheric electron content prediction model constructed in the method only has an original thermodynamic diagram branch;
the method B comprises the following steps: predicting the total electron content of the global ionized layer only by adopting a convolution long-time memory network;
the method C comprises the following steps: the method of the invention is adopted to predict the total electron content of the global ionized layer.
The root mean square error (RMS) ratio of the total electron content prediction results of the method a, the method B and the method C is shown in fig. 2, wherein (a) in fig. 2 shows the RMS of each graph in the prediction result graph sequence of the three methods; the percentage improvement in accuracy of method C over methods a and B is given in fig. 2 (B). The average RMS of the method A is 3.055TECU, the average RMS of the method B is 3.664TECU, the average RMS of the method C is 2.544TECU, the accuracy of the method C is improved by 16.74% in comparison with the method A on average, the accuracy is improved by 41.85% to the maximum extent, the accuracy of the method C is improved by 30.57% in comparison with the method B on average, and the accuracy is improved by 36.45% to the maximum extent.
Claims (5)
1. The global ionospheric total electron content prediction method based on the deep cycle neural network comprises a training phase and a prediction phase, and is characterized in that the training phase comprises the following steps:
acquiring K global ionospheric electron total content thermodynamic diagrams at equal time intervals every day for N consecutive days; for each collected image, adjusting the horizontal position according to the position of each pixel point, so that the positions of the pixel points with the same vertical coordinate are increased along the positive direction of the horizontal axis; constructing an original image sequence S ═ { Pic) according to an acquisition orderk,n}，k＝1,2,…,K，n＝1,2,…,N；
Step (2) constructing a training sample set, wherein the mth sample s in the training sample setmThe method comprises the steps of inputting sample data and outputting the sample data, wherein the input sample data comprises t days of original image sequences starting from the mth day in an original image sequence S and t-1 days of primary image differential sequences formed by the t days of original image sequences; outputting sample data as an original image sequence of the m + t days in the S; t is an integer greater than 2;
constructing a global ionosphere electron total content prediction model based on a deep cycle neural network, wherein the model comprises an original thermodynamic diagram branch, a differential thermodynamic diagram branch and a convolution subnet;
the original thermodynamic diagram branch and the differential thermodynamic diagram branch are both deep circulation neural networks with the same structure, and each deep circulation neural network comprises a first convolution long-time and short-time memory network, an expansion convolution long-time and short-time memory network and a second convolution long-time and short-time memory network which are sequentially connected;
the input of the original thermodynamic diagram branch is an original image sequence of t days, and the output is a one-day K thermodynamic diagram sequence predicted according to the input; the input of the differential thermodynamic diagram branch is a primary differential sequence of images for t-1 days, and the output is a predicted K differential thermodynamic diagram sequence for one day;
the convolution subnet recovers the differential thermodynamic diagrams output by the differential thermodynamic diagram branch into thermodynamic diagram data, the recovered thermodynamic diagram data and the thermodynamic diagram sequence output by the original thermodynamic diagram branch are stacked into a dual-channel image sequence, and the stacked dual-channel image sequence is convolved to obtain the predicted global ionospheric electron total content thermodynamic diagram;
training the global ionospheric total electron content prediction model by using a training sample set: taking t-day original image sequences in input sample data of a training sample as the input of an original thermodynamic diagram branch of the global ionospheric electronic total content prediction model, taking t-1-day primary image difference sequences as the input of a differential thermodynamic diagram branch, comparing an obtained prediction result with output sample data in the training sample to calculate errors, correcting parameters to be solved in the error back propagation model, and obtaining a final global ionospheric electronic total content prediction model through iterative training;
the prediction phase comprises:
collecting K global ionospheric electron total content thermodynamic diagrams at equal time intervals every day for t days continuously; for each collected image, adjusting the horizontal position according to the position of each pixel point, so that the positions of the pixel points with the same vertical coordinate are positively increased along the horizontal axis to form a to-be-predicted image sequence T ═ Pk,ττ ═ 1,2, …, t; establishing prediction samples comprising a sequence of predicted images and a sequence of primary difference images { Δ P } of the sequence of predicted imagesk,τ+1}; will { Pk,τAnd { Δ P }k,τ+1Respectively serving as the input of an original thermodynamic diagram branch and a differential thermodynamic diagram branch in a global ionospheric electronic total content prediction model, wherein the output of the model is a prediction thermodynamic diagram;
and (5) carrying out longitude sequencing on the predicted thermodynamic diagrams to obtain the predicted global ionospheric electron total content thermodynamic diagrams.
2. The global total ionospheric electron content prediction method according to claim 1, wherein the step (1) further comprises resizing and pixel value normalizing the images in the image sequence S, and the step (5) further comprises resizing and denormalizing the predictive thermodynamic diagram.
3. The method for predicting the total electron content in the global ionosphere according to claim 1, wherein the adjusting of the horizontal position according to the location of the pixel points in the steps (1) and (4) comprises:
calculating the local time of each pixel point in the global ionosphere electron total content thermodynamic diagram in the range direction; keeping the vertical coordinates of the pixel points unchanged, sorting the pixel points from small to large according to places, and adjusting the horizontal positions of the pixels according to a sorting result to increase the positions of the pixel points with the same vertical coordinates along the positive direction of the horizontal axis.
4. The global total ionospheric electron content prediction method of claim 1, wherein the differential thermodynamic diagrams output by the global total ionospheric electron content prediction model convolutional subnetworks are restored to thermodynamic diagram data according to the following formula:
where K is 1,2, …, K,
a differential thermodynamic diagram sequence output for the differential thermodynamic diagram branch; f. oft,kInputting an original image sequence of the t day in the original image sequence for the original thermodynamic diagram branch;
is the recovered thermodynamic diagram data.
5. The global total ionospheric electron content prediction method according to claim 1, wherein said step (5) of longitudinally ordering the predicted thermodynamic diagrams comprises:
calculating the longitude of each pixel point according to the local time and the world time of each pixel point; keeping the latitude direction coordinate unchanged, sorting the pixel points from small to large according to the longitude values, and adjusting the horizontal positions of the pixels according to the sorting result to increase the longitude values of the pixel points with the same vertical coordinate along the positive direction of the horizontal axis.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010238027.XA CN111539453B (en) | 2020-03-30 | 2020-03-30 | Global ionized layer electron total content prediction method based on deep cycle neural network |
| PCT/CN2020/114998 WO2021196528A1 (en) | 2020-03-30 | 2020-09-14 | Global ionospheric total electron content prediction method based on deep recurrent neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010238027.XA CN111539453B (en) | 2020-03-30 | 2020-03-30 | Global ionized layer electron total content prediction method based on deep cycle neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111539453A CN111539453A (en) | 2020-08-14 |
| CN111539453B true CN111539453B (en) | 2022-04-26 |
Family
ID=71976907
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010238027.XA Active CN111539453B (en) | 2020-03-30 | 2020-03-30 | Global ionized layer electron total content prediction method based on deep cycle neural network |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN111539453B (en) |
| WO (1) | WO2021196528A1 (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111539453B (en) * | 2020-03-30 | 2022-04-26 | 东南大学 | Global ionized layer electron total content prediction method based on deep cycle neural network |
| CN112085068B (en) * | 2020-08-18 | 2022-11-01 | 东南大学 | Global ionospheric electron total content anomaly detection method based on image difference |
| CN112700007B (en) * | 2020-12-30 | 2023-06-20 | 千寻位置网络有限公司 | Training method, forecasting method and device of ionosphere electronic content forecasting model |
| CN114652326B (en) * | 2022-01-30 | 2024-06-14 | 天津大学 | Real-time brain fatigue monitoring device based on deep learning and data processing method |
| CN116738159B (en) * | 2023-08-16 | 2023-11-14 | 北京航空航天大学 | Global ionosphere space weather response extraction method based on complex network |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102508211A (en) * | 2011-11-04 | 2012-06-20 | 西安电子科技大学 | Method for estimating total electron content in ionized layer based on double-frequency correction method |
| CN103197340A (en) * | 2013-04-01 | 2013-07-10 | 东南大学 | Gridding real-time monitoring method for total electron content of ionized layer |
| CN104992054A (en) * | 2015-06-19 | 2015-10-21 | 东南大学 | Method for forecasting ionospheric vertical total electron content based on time-series two-dimensionalization |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8930299B2 (en) * | 2010-12-15 | 2015-01-06 | Vaisala, Inc. | Systems and methods for wind forecasting and grid management |
| CN110751057A (en) * | 2019-09-27 | 2020-02-04 | 五邑大学 | Finger vein verification method and device based on long-time and short-time memory cyclic neural network |
| CN111539453B (en) * | 2020-03-30 | 2022-04-26 | 东南大学 | Global ionized layer electron total content prediction method based on deep cycle neural network |
-
2020
- 2020-03-30 CN CN202010238027.XA patent/CN111539453B/en active Active
- 2020-09-14 WO PCT/CN2020/114998 patent/WO2021196528A1/en active Application Filing
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102508211A (en) * | 2011-11-04 | 2012-06-20 | 西安电子科技大学 | Method for estimating total electron content in ionized layer based on double-frequency correction method |
| CN103197340A (en) * | 2013-04-01 | 2013-07-10 | 东南大学 | Gridding real-time monitoring method for total electron content of ionized layer |
| CN104992054A (en) * | 2015-06-19 | 2015-10-21 | 东南大学 | Method for forecasting ionospheric vertical total electron content based on time-series two-dimensionalization |
Non-Patent Citations (2)
| Title |
|---|
| 基于Holt-Winters的电离层总电子含量预报;谢劭峰等;《大地测量与地球动力学》;20170115;全文 * |
| 深度学习LSTM模型的电离层总电子含量预报;吉长东;《导航定位学报》;20190829;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111539453A (en) | 2020-08-14 |
| WO2021196528A1 (en) | 2021-10-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111539453B (en) | Global ionized layer electron total content prediction method based on deep cycle neural network | |
| US11333796B2 (en) | Spatial autocorrelation machine learning-based downscaling method and system of satellite precipitation data | |
| CN107688906B (en) | Multi-method fused transmission line meteorological element downscaling analysis system and method | |
| CN114943963A (en) | Remote sensing image cloud and cloud shadow segmentation method based on double-branch fusion network | |
| CN110874630B (en) | Deep learning-based numerical model product downscaling refinement method | |
| CN111105068B (en) | Numerical mode correction method based on sequential regression learning | |
| CN114547017A (en) | Meteorological big data fusion method based on deep learning | |
| CN113379107A (en) | Regional ionized layer TEC forecasting method based on LSTM and GCN | |
| Sampath et al. | Estimation of rooftop solar energy generation using satellite image segmentation | |
| CN110909447A (en) | High-precision short-term prediction method for ionization layer region | |
| CN113344180A (en) | Neural network training and image processing method, device, equipment and storage medium | |
| CN111539455B (en) | Global ionosphere electron total content prediction method based on image primary difference | |
| CN111539433B (en) | Semantic segmentation based global ionosphere total electron content prediction method | |
| CN112990354B (en) | Method and device for constructing deep convolution regression network for wind speed prediction | |
| CN110595477A (en) | Method for positioning according to sun shadow in video based on genetic algorithm | |
| CN112285808B (en) | Method for reducing scale of APHRODITE precipitation data | |
| CN113989612A (en) | Remote sensing image target detection method based on attention and generation countermeasure network | |
| CN116541681A (en) | Composite disaster space variability identification method based on collaborative kriging interpolation | |
| CN112990609B (en) | Air quality prediction method based on space-time bandwidth self-adaptive geographical weighted regression | |
| CN112085068B (en) | Global ionospheric electron total content anomaly detection method based on image difference | |
| CN109886497B (en) | Ground air temperature interpolation method based on latitude improved inverse distance weighting method | |
| CN113159426A (en) | Weather type similarity judgment method and device, electronic equipment and readable storage medium | |
| CN111814855A (en) | Global ionospheric total electron content prediction method based on residual seq2seq neural network | |
| CN113034363A (en) | Nitrogen oxide rapid reduction method based on GEE depth space-time experience kriging regional scale | |
| CN114048783B (en) | Cellular signal map construction method based on mobile group perception |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |