CN107563574A - A kind of earth's magnetic field Forecasting Methodology based on long memory models Recognition with Recurrent Neural Network in short-term - Google Patents
A kind of earth's magnetic field Forecasting Methodology based on long memory models Recognition with Recurrent Neural Network in short-term Download PDFInfo
- Publication number
- CN107563574A CN107563574A CN201710911282.4A CN201710911282A CN107563574A CN 107563574 A CN107563574 A CN 107563574A CN 201710911282 A CN201710911282 A CN 201710911282A CN 107563574 A CN107563574 A CN 107563574A
- Authority
- CN
- China
- Prior art keywords
- earth
- magnetic field
- data
- neural network
- recurrent neural
- 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
Abstract
The invention discloses a kind of earth's magnetic field Forecasting Methodology based on long memory models Recognition with Recurrent Neural Network in short-term, and compared to spherical harmonic analysis method, simple in construction, it is necessary to which the parameter solved is few, method for solving is simple;Recognition with Recurrent Neural Network is enhanced the processing of the sequential relationship of magnetic field data over the ground, can effectively be predicted the rule changed over time of earth magnetism field data using long memory models structure in short-term;Without do the processing of complexity to original earth magnetism field data, without in frequency spectrum and statistically magnetic field data is pre-processed over the ground, the processing procedure of data is facilitated;The time span of training sample and test sample, and the structure of design cycle neutral net according to this are selected for the periodic characteristics of earth magnetism field data so that neutral net can effectively analyze earth's magnetic field periodic feature, greatly improve the precision of prediction in earth's magnetic field.
Description
Technical field
The invention belongs to the space physics technical field of geoscience, and in particular to one kind is followed based on long memory models in short-term
The earth's magnetic field Forecasting Methodology of ring neutral net.
Background technology
Natural magnetic phenomena, as earth's magnetic field be present in the inside of the earth.The analysis and prediction of earth magnetism field data are in engineering
It is upper that there is very high application value, important work is especially played in fields such as abnormality detection, space physics, navigator fixs
With.Possessed magnetic field superposition on earth's magnetic field, makes earth's magnetic field exist to the magnetic targets such as vehicle, mineral metal products, submarine, submarine mine in itself
Occur in certain area abnormal.In the abnormality detection technology of magnetic field, magnetic field detection device can detect magnetic field abnormal signal, pass through
The abnormal signal is extracted and performed an analysis, you can obtains the relevant information (including the information such as position, shape) of abnormal object.
During magnetic anomaly detection, environmental magnetic field complicated and changeable can disturb the extraction of abnormal signal, influence to divide magnetic anomaly regular signal
Analysis, so as to have a strong impact on detection accuracy, in order to meet the performance requirement such as the high accuracy of magnetic anomaly detection and real-time, it is necessary to magnetic
Abnormality detection instrument environmental background magnetic field suffered when working carries out real-time high-accuracy compensation, and compensation can be by earth's magnetic field
High-precision forecast is completed.In earth-magnetism navigation technology, need early stage in the built-in on the spot magnetic field model in the moving region of carrier, draw
Go out earth magnetism reference figure, preserve in the memory unit.Carrier, as benchmark, passes through corresponding earth magnetism in motion process
Navigator fix information is obtained with technology.Therefore, region Geomagnetic Field Model establish it is particularly important, be determine earth-magnetism navigation precision
Key factor, the Geomagnetic Field Model of a real-time high-precision can be established to the high-precision forecast in earth's magnetic field, improve earth-magnetism navigation
Precision.In space physics and weather forecasting techniques, earth's magnetic field carries substantial amounts of information, is important analysis object, ground
The prediction in magnetic field can realize the forecast to geological disasters such as the meteorological forecast in space and earthquakes.
At present, earth's magnetic field Forecasting Methodology is according to known discrete measurement point geomagnetic data in global range, to earth's magnetic field
Mathematical modeling is carried out, so as to obtain earth's magnetic field global models, the numerical value in future time instance earth's magnetic field, realization pair are calculated according to the model
The prediction in earth's magnetic field.The conventional method of earth's magnetic field modeling is spherical harmonic analysis method, i.e., the limited individual humorous letter of ball increased successively with exponent number
Number sum approximate expression earth's magnetic field scalar potential, and global magnetic spy data is obtained best fit, obtain the model of Global Geomagnetic Field.
Existing method is limited by spherical harmonic series cutoff level, and the spatial resolution of spherical harmonic analysis method is not general high, far can not meet
The actual demands such as detection, navigation, although resolution ratio can be improved by improving cutoff level, when spherical harmonic series increase, wait to ask
The number of the spherical harmonic coefficient of solution sharply increases, and amount of calculation and amount of storage sharply increase.
The content of the invention
The present invention is directed to weak point of the spheric harmonic function when establishing Geomagnetic Field Model and being predicted to earth's magnetic field, proposes
A kind of earth's magnetic field Forecasting Methodology based on long memory models Recognition with Recurrent Neural Network in short-term, it is pre- high accuracy can be carried out to earth's magnetic field
Survey.
A kind of earth's magnetic field Forecasting Methodology, comprises the following steps:
Step 1: in the time span of setting, by the collection period of setting, the earth's magnetic field on each sampled point is obtained
Data, the output par, c as data set;The space simultaneously, obtained on each sampled point is sampled in magnetic field data over the ground
Physical index data, as the importation of data set, it is consequently formed data set;The time span is in units of day;
Step 2: access time span forms training set for the data in integer day in the data set;
Step 3: since the 1st sampled point of training set, the space physics index number on L continuous sampling point is chosen
According to the importation of first training sample of composition;M continuous samplings are chosen since next sampled point of the training sample
Earth magnetism field data on point, the output par, c as the training sample;
Then since the 2nd sampled point of training set, the space physics exponent data on L continuous sampling point is chosen,
The importation of second training sample is formed, and M continuous samplings are chosen since next sampled point of the training sample
Earth magnetism field data on point, the output par, c as the training sample;
By that analogy, importation and the output par, c of each training sample are obtained;Wherein, L be training sample when
Between length, be to be set according to the periodic feature in earth's magnetic field, M be prediction earth's magnetic field time span, predict as needed
Time span determines;
Step 4: Recognition with Recurrent Neural Network is built, wherein, input layer quantity is arranged to data set importation space thing
The species number of index is managed, hidden layer node number and the node number of output layer are set, wherein, the number of output layer is i.e. predictably
The time span M in magnetic field;Hidden layer node is arranged to long memory models structure in short-term;
It is input to Step 5: step 2 is obtained into training sample in the Recognition with Recurrent Neural Network that step 4 is built, it is carried out
Training;Judge whether to meet stop condition after the completion of training iteration each time, if it is satisfied, then the circulation nerve trained
Network, next step is performed, if be unsatisfactory for, continue repetitive exercise;
Step 6: it is pre- that the space physics index of input is input into progress earth's magnetic field in the Recognition with Recurrent Neural Network trained
Survey.
Preferably, space physics index input include sun air temperature, density, speed, geomagnetic activity index Dst with
And Kp.
Preferably, the data in the data set in the step 1 are normalized, data span is limited
System is between [0.2-0.8].
Preferably, in data set, the time span data in integer day are selected to form test set, according still further to the side of step 3
Method, obtain each test sample;The Recognition with Recurrent Neural Network that step 5 trains is tested using test sample, tested
As a result:If test result meets to require, earth magnetism field prediction is carried out using the Recognition with Recurrent Neural Network, if being unsatisfactory for requiring,
The hidden layer node number of Recognition with Recurrent Neural Network is then adjusted, step 5 is continued executing with and is trained, until meeting to require position.
Preferably, the Recognition with Recurrent Neural Network uses activation primitive of the Sigmoid functions as neutral net.
Preferably, the Recognition with Recurrent Neural Network is updated iteration using BP algorithm to the weights in network and biasing.
Preferably, the time span of data set is at least 30 days in step 1.
Preferably, the L is at least 7 days.
Preferably, the M is less than 3 days.
The present invention has the advantages that:
Recognition with Recurrent Neural Network Forecasting Methodology provided by the invention, compared to spherical harmonic analysis method, it is simple in construction, it is necessary to solve
Parameter is few, and method for solving is simple;Recognition with Recurrent Neural Network using long memory models structure in short-term, enhance magnetic field data over the ground when
The processing of order relation, it can effectively predict the rule changed over time of earth magnetism field data;Without to original earth magnetism number of fields
According to the processing for doing complexity, without in frequency spectrum and statistically magnetic field data is pre-processed over the ground, treating for data is facilitated
Journey;The time span of training sample and test sample, and design cycle according to this are selected for the periodic characteristics of earth magnetism field data
The structure of neutral net so that neutral net can effectively analyze earth's magnetic field periodic feature, greatly improve the pre- of earth's magnetic field
Survey precision;Importation of the selection space physics index relevant with terrestrial magnetic disturbance as neutral net so that neutral net can
To efficiently extract the variation characteristic in earth's magnetic field from terrestrial magnetic disturbance formation mechenism, so as to realize to the accurate prediction in earth's magnetic field.
Brief description of the drawings
Fig. 1 is neural metwork training and test flow chart in the present invention;
Fig. 2 is the structure chart of Recognition with Recurrent Neural Network;
Fig. 3 is the structure chart of LSTM (long memory models in short-term);
Fig. 4 is the neural network structure figure in the present invention.
Embodiment
The present invention will now be described in detail with reference to the accompanying drawings and examples.
A kind of earth's magnetic field Forecasting Methodology based on long memory models Recognition with Recurrent Neural Network in short-term comprises the following steps:
Step 1:Earth's magnetic field data set is built, the neutral net in this method uses the learning method of supervised learning, according to
This characteristic, earth's magnetic field data set are made up of output par, c and importation.Original earth magnetism field data uses some regularly
The data of magnetic field monitoring point, it is desirable to the magnetic field environment around this test point is more clean, i.e., other suffered magnetic interferences compared with
It is small, the sampling period is set, samples the output par, c that the earth magnetism field data obtained afterwards forms data set;Enter in magnetic field data over the ground
Row sampling simultaneously, obtains the space physics exponent data on each sampled point, for the formation mechenism of terrestrial magnetic disturbance, selects phase
Corresponding space physics index, including the solar wind index of correlation, geomagnetic activity index of correlation etc., these space physics indexes are made
For the importation of data set;Wherein, because the daily amplitude of variation in earth's magnetic field has certain periodicity, sampled data when
Between span in units of day, time span is at least 30 days.
Step 2:Data in data set obtained by step 1 are normalized, during to prevent data test
There is saturation situation in the output of neutral net, and the data area in data set is limited between [0.2-0.8];
Step 3:It is training set and test set by Segmentation of Data Set in units of day;
Step 4:Since the 1st sampled point of training set, the space physics index number on L continuous sampling point is chosen
According to the importation of one training sample of composition;M continuous sampled points are chosen since next sampled point of the training sample
On earth magnetism field data, the output par, c as the training sample;Again since the 2nd sampled point of training set, L is chosen even
Space physics exponent data on continuous sampled point, the importation of second training sample is formed, and under the training sample
One sampled point starts to choose the earth magnetism field data on M continuous sampled points, the output par, c as the training sample;With such
Push away, obtain importation and the output par, c of each training sample;Wherein, L is the time span of training sample, is base area
The periodic feature setting in magnetic field, at least 7 days, M was the time span in prediction earth's magnetic field, within may be selected 3 days.Using phase
Same method, importation and the output par, c of middle selection test sample are obtained from test set;The time span and instruction of test sample
The time span for practicing sample is consistent;
Step 5:Recognition with Recurrent Neural Network is built, input layer quantity is arranged to data set input quantity segment space physics
The species number of index, hidden layer node number is set, it is desirable to more than 50, sets the number of output layer, the number of output layer is
Predict the time span M in earth's magnetic field;
Step 6:Setting hidden layer node is LSTM (long memory models in short-term) structure;
Step 7:The Recognition with Recurrent Neural Network of completion is built in training, according to the normalization pretreatment of data and data amplitude
Limitation, select activation primitive of the Sigmoid functions as neutral net, cost function used in selection training, set and instruct
Practice speed, the training sample in the data training set of earth's magnetic field is sent into Recognition with Recurrent Neural Network and is trained, in training process, lead to
The BP algorithm (back-propagation algorithm) of neutral net is crossed, the weights in network are updated with biasing;
Step 8:Iterations or training precision is set to judge whether to reach iteration stopping as iteration stopping condition
Condition, if not up to, return to step seven continues to train, if having reached, goes to step 9;
Step 9:Default iterations or default training precision are reached, according to related performance indicators, analysis
Its training result.Data test integrated test sample importation is input in the neutral net that training is completed, tested,
By the propagated forward of neutral net, exported accordingly, the output par, c exported again with test sample is contrasted, obtained
Go out performance indications, analyze its test result, and hidden layer number is adjusted according to test result, until obtaining the test knot of satisfaction
Fruit, complete Neural Network Optimization;
Step 10:The space physics index of input is input in neutral net, completes to follow based on long memory models in short-term
The earth magnetism field prediction of ring neutral net.
Embodiment:
This example provides a kind of earth's magnetic field Forecasting Methodology based on long memory models Recognition with Recurrent Neural Network in short-term, this method energy
It is enough that somewhere earth's magnetic field is predicted, it is real-time.
This example selects Ming Tombs, Beijing earth magnetism monitoring station 4 days in Augusts, 2017 data of 10 days of August in 2017, sampling week
Phase is 1 minute, and the space physics index input of selection includes sun air temperature, density and speed, and geomagnetic activity index
Dst, Kp, its sampling period are consistent with earth's magnetic field data sampling period, are 1 minute.Data set is time series, and length is
10080.Wherein, K indexes are the indexes that single geomagnetic observatory is used for describing the terrestrial magnetic disturbance intensity in daily each 3 hours, are referred to as
Three hours indexes or character figure.Kp indexes are to choose latitude in 44 ° to 60 ° of north latitude and 13 geomagnetic observatories of 44 ° to 60 ° of south latitude
The average value for the K values stood.Dst indexes are to choose equator distribution than more uniform and remote ionosphere E areas equatorial electrojet
4 geomagnetic observatories, horizontal component measurement data in each of which hour is averaging, take its hourly value with earth magnetism calm day it
Difference.
Geomagnetic data collection is represented by
[x1,x2,x3,x4,x5,y]
Wherein, xi=[xi,1,xi,2,...,xi,10080]T(5) i=1 2 ..., is expressed as input space physical index, y=
[y1,y2,...,y10080]T, it is expressed as exporting earth magnetism field data.
The data concentrated to data are normalized, to prevent the output of data test from saturation situation occur, by number
It is limited according to the data area of concentration between [0.2-0.8].New sample is obtained after normalization, the new samples after normalization
It is represented by:
[x1′,x2′,x3′,x4′,x5′,y′]
Wherein,
Cycle of training is set, will be arranged to cycle of training 6 days, data set is divided into training dataset and test data set, instruction
The time span for practicing data set is 6 days, and corresponding data set time sequence length is 8640, i.e. the of data set time sequence
1-8640 sample, is represented by
[xl1,xl2,xl3,xl4,xl5,yl]
Wherein, xli=[xi,1,xi,2,...,xi,8640]T(i=1,2 ..., 5), the input for being expressed as training dataset are empty
Between physical index amount, yl=[y1,y2,...,y8640]T, it is expressed as the output earth's magnetic field value of training dataset.
The time span of test data set be 1 day, corresponding data set time sequence length be 1440, i.e., data set when
Between sequence the 8641-10080 sample, be represented by:
[xt1,xt2,xt3,xt4,xt5,yt]
Wherein, xti=[xi,8641,xi,8642,...,xi,10080]T(i=1,2 ..., 5), are expressed as the defeated of test data set
Enter space physics exponential quantity, yt=[y8641,y8642,...,y10080]T, it is expressed as the output earth's magnetic field value of test data set.
The time span of training sample is set, and because change of the earth's magnetic field within the time has periodically, noon disturbs
Dynamic larger, night disturbance is gentle, and the time span for setting training sample is 1440, i.e., the data of one day, training sample can represent
For:
Batch (j)=[xt1j,xt2j,xt3j,xt4j,xt5j,ytj]
Wherein, xtij=[xi,j,xi,j+1,...,xi,j+1440]T, ytj=[yj+1441,yj+1442,...,yj+step]T(i=1,
2 ..., 5, j=1,2 ..., 7200), step are the time step of earth magnetism field prediction.Test sample is represented by:
Batch (m)=[xt1m,xt2m,xt3m,xt4m,xt5m,ytm]
Wherein, xtim=[xi,m,xi,m+1,...,xi,m+1440]T, ytj=[ym+1441,ym+1442,...,ym+step]T(i=1,
2 ..., 5, m=8641,8642 ..., (10080-step)), step be earth magnetism field prediction time step.
Recognition with Recurrent Neural Network, such as Fig. 4 are built, input layer is arranged to the input space physical index in data set, including too
Positive air temperature, density and speed, geomagnetic activity index Dst, Kp, according to the species of data set input quantity segment space physical parameter
Number, the nodes for setting input layer are 5;It is 300 to set hidden layer number;The number for setting output layer is 10, i.e. step=
10, neutral net predictably magnetic field when a length of 10 minutes.
Setting hidden layer node is LSTM (long memory models in short-term) structure.
The Recognition with Recurrent Neural Network of completion is built in training, and the activation primitive for setting neutral net is Sigmoid functions, and it is expressed
Formula is:
Used cost function is secondary cost function when setting training, and the speed for setting training is 1 × 10-3, will enter
The training sample in the earth's magnetic field data training set of normalized of going, which is sent into Recognition with Recurrent Neural Network, to be trained, and is trained
Cheng Zhong, by the BP algorithm (back-propagation algorithm) of neutral net, using the method for stochastic gradient descent to the weights in network
It is updated with biasing.
Set iterations be 30000, judge whether to reach default 30000 iterationses, if not up to, continue into
Row training.
Default 30000 iterationses, the performance indications trained are reached, its training mean accuracy is
0.5nT/min, the test sample in the data test collection input quantity part in data set is input to the nerve net of training completion
In network, tested, by the propagated forward of neutral net, exported accordingly, carried out with output quantity part in test set
Contrast, it is 2.3nT/min that it, which tests mean accuracy, and it is pre- to complete the earth's magnetic field based on long memory models Recognition with Recurrent Neural Network in short-term
Survey.
In summary, presently preferred embodiments of the present invention is these are only, is not intended to limit the scope of the present invention.
Within the spirit and principles of the invention, any modification, equivalent substitution and improvements made etc., it should be included in the present invention's
Within protection domain.
Claims (9)
1. a kind of earth's magnetic field Forecasting Methodology, it is characterised in that comprise the following steps:
Step 1: in the time span of setting, by the collection period of setting, the earth magnetism number of fields on each sampled point is obtained
According to the output par, c as data set;The space thing simultaneously, obtained on each sampled point is sampled in magnetic field data over the ground
Exponent data is managed, as the importation of data set, is consequently formed data set;The time span is in units of day;
Step 2: access time span forms training set for the data in integer day in the data set;
Step 3: since the 1st sampled point of training set, the space physics exponent data on L continuous sampling point, group are chosen
Into the importation of first training sample;Chosen since next sampled point of the training sample on M continuous sampled points
Earth magnetism field data, the output par, c as the training sample;
Then since the 2nd sampled point of training set, the space physics exponent data on L continuous sampling point, composition are chosen
The importation of second training sample, and chosen since next sampled point of the training sample on M continuous sampled points
Earth magnetism field data, the output par, c as the training sample;
By that analogy, importation and the output par, c of each training sample are obtained;Wherein, L is the time length of training sample
Degree, is set according to the periodic feature in earth's magnetic field, and M is the time span in prediction earth's magnetic field, the time predicted as needed
Length determines;
Step 4: Recognition with Recurrent Neural Network is built, wherein, input layer quantity is arranged to data set importation space physics and referred to
Several species numbers, hidden layer node number and the node number of output layer are set, wherein, the number of output layer predicts earth's magnetic field
Time span M;Hidden layer node is arranged to long memory models structure in short-term;
It is input to Step 5: step 2 is obtained into training sample in the Recognition with Recurrent Neural Network that step 4 is built, it is trained;
Judge whether to meet stop condition after the completion of training iteration each time, if it is satisfied, then the Recognition with Recurrent Neural Network trained,
Next step is performed, if be unsatisfactory for, continues repetitive exercise;
Step 6: the space physics index of input is input to progress earth magnetism field prediction in the Recognition with Recurrent Neural Network trained.
2. a kind of earth's magnetic field Forecasting Methodology as claimed in claim 1, it is characterised in that the space physics index input includes
Sun air temperature, density, speed, geomagnetic activity index Dst and Kp.
3. a kind of earth's magnetic field Forecasting Methodology as claimed in claim 1, it is characterised in that in the data set in the step 1
Data be normalized, data span is limited between [0.2-0.8].
4. a kind of earth's magnetic field Forecasting Methodology as claimed in claim 1, it is characterised in that in data set, select integer day
Time span data form test set, according still further to the method for step 3, obtain each test sample;Using test sample to step
Five Recognition with Recurrent Neural Network trained are tested, and obtain test result:If test result meets to require, the circulation is utilized
Neutral net carries out earth magnetism field prediction, if being unsatisfactory for requiring, adjusts the hidden layer node number of Recognition with Recurrent Neural Network, continues
Perform step 5 to be trained, until meeting to require position.
5. a kind of earth's magnetic field Forecasting Methodology as claimed in claim 1, it is characterised in that the Recognition with Recurrent Neural Network uses
Activation primitive of the Sigmoid functions as neutral net.
6. a kind of earth's magnetic field Forecasting Methodology as claimed in claim 1, it is characterised in that the Recognition with Recurrent Neural Network is calculated using BP
Method is updated iteration to the weights in network and biasing.
A kind of 7. earth's magnetic field Forecasting Methodology as claimed in claim 1, it is characterised in that the time span of data set in step 1
At least 30 days.
8. a kind of earth's magnetic field Forecasting Methodology as claimed in claim 1, it is characterised in that the L is at least 7 days.
9. a kind of earth's magnetic field Forecasting Methodology as claimed in claim 1, it is characterised in that the M is less than 3 days.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710911282.4A CN107563574A (en) | 2017-09-29 | 2017-09-29 | A kind of earth's magnetic field Forecasting Methodology based on long memory models Recognition with Recurrent Neural Network in short-term |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710911282.4A CN107563574A (en) | 2017-09-29 | 2017-09-29 | A kind of earth's magnetic field Forecasting Methodology based on long memory models Recognition with Recurrent Neural Network in short-term |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107563574A true CN107563574A (en) | 2018-01-09 |
Family
ID=60984805
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710911282.4A Pending CN107563574A (en) | 2017-09-29 | 2017-09-29 | A kind of earth's magnetic field Forecasting Methodology based on long memory models Recognition with Recurrent Neural Network in short-term |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107563574A (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108957418A (en) * | 2018-05-30 | 2018-12-07 | 西安电子科技大学 | A kind of radar target identification method based on Recognition with Recurrent Neural Network model |
CN109194423A (en) * | 2018-08-13 | 2019-01-11 | 中国人民解放军陆军工程大学 | The single-frequency point spectrum prediction method of shot and long term memory models based on optimization |
CN109711022A (en) * | 2018-12-17 | 2019-05-03 | 哈尔滨工程大学 | A kind of submarine anti-sinking system based on deep learning |
CN109738939A (en) * | 2019-03-21 | 2019-05-10 | 蔡寅 | A kind of Precursory Observational Data method for detecting abnormality |
CN109781094A (en) * | 2018-12-24 | 2019-05-21 | 上海交通大学 | Earth magnetism positioning system based on Recognition with Recurrent Neural Network |
CN110255303A (en) * | 2019-06-13 | 2019-09-20 | 天津大学 | A kind of elevator waiting floor optimization method |
CN111121759A (en) * | 2019-12-30 | 2020-05-08 | 杭州电子科技大学 | Geomagnetic indoor positioning method based on multilayer long-short term memory network |
CN111224726A (en) * | 2019-11-21 | 2020-06-02 | 南京邮电大学 | Video live broadcast system based on long-term and short-term memory network and implementation method thereof |
CN113139327A (en) * | 2020-01-19 | 2021-07-20 | 中国科学院国家空间科学中心 | Ionized layer TEC single-point prediction method and system based on GRU network model |
CN113344032A (en) * | 2021-05-13 | 2021-09-03 | 中国科学院国家空间科学中心 | Geomagnetic Kp index forecasting method and system based on similarity algorithm |
US11651839B2 (en) | 2020-03-02 | 2023-05-16 | Uchicago Argonne, Llc | Systems and methods for generating phase diagrams for metastable material states |
CN116136975A (en) * | 2023-02-24 | 2023-05-19 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | Neutral atmosphere temperature density prediction method and equipment based on LSTM neural network |
US11663494B2 (en) | 2019-12-05 | 2023-05-30 | Uchicago Argonne, Llc | Systems and methods for hierarchical multi-objective optimization |
US11710038B2 (en) * | 2020-04-13 | 2023-07-25 | Uchicago Argonne, Llc | Systems and methods for active learning from sparse training data |
CN116882594A (en) * | 2023-09-06 | 2023-10-13 | 数字太空(北京)智能技术研究院有限公司 | Geomagnetic ap index medium-term forecasting method based on limited recursion and deep learning |
-
2017
- 2017-09-29 CN CN201710911282.4A patent/CN107563574A/en active Pending
Non-Patent Citations (8)
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108957418A (en) * | 2018-05-30 | 2018-12-07 | 西安电子科技大学 | A kind of radar target identification method based on Recognition with Recurrent Neural Network model |
CN109194423A (en) * | 2018-08-13 | 2019-01-11 | 中国人民解放军陆军工程大学 | The single-frequency point spectrum prediction method of shot and long term memory models based on optimization |
CN109711022A (en) * | 2018-12-17 | 2019-05-03 | 哈尔滨工程大学 | A kind of submarine anti-sinking system based on deep learning |
CN109781094A (en) * | 2018-12-24 | 2019-05-21 | 上海交通大学 | Earth magnetism positioning system based on Recognition with Recurrent Neural Network |
CN109738939A (en) * | 2019-03-21 | 2019-05-10 | 蔡寅 | A kind of Precursory Observational Data method for detecting abnormality |
CN110255303A (en) * | 2019-06-13 | 2019-09-20 | 天津大学 | A kind of elevator waiting floor optimization method |
CN111224726A (en) * | 2019-11-21 | 2020-06-02 | 南京邮电大学 | Video live broadcast system based on long-term and short-term memory network and implementation method thereof |
US11663494B2 (en) | 2019-12-05 | 2023-05-30 | Uchicago Argonne, Llc | Systems and methods for hierarchical multi-objective optimization |
CN111121759A (en) * | 2019-12-30 | 2020-05-08 | 杭州电子科技大学 | Geomagnetic indoor positioning method based on multilayer long-short term memory network |
CN111121759B (en) * | 2019-12-30 | 2021-07-27 | 杭州电子科技大学 | Geomagnetic indoor positioning method based on multilayer long-short term memory network |
CN113139327B (en) * | 2020-01-19 | 2024-03-12 | 中国科学院国家空间科学中心 | Ionized layer TEC single-point prediction method and system based on GRU network model |
CN113139327A (en) * | 2020-01-19 | 2021-07-20 | 中国科学院国家空间科学中心 | Ionized layer TEC single-point prediction method and system based on GRU network model |
US11651839B2 (en) | 2020-03-02 | 2023-05-16 | Uchicago Argonne, Llc | Systems and methods for generating phase diagrams for metastable material states |
US11710038B2 (en) * | 2020-04-13 | 2023-07-25 | Uchicago Argonne, Llc | Systems and methods for active learning from sparse training data |
CN113344032B (en) * | 2021-05-13 | 2021-12-28 | 中国科学院国家空间科学中心 | Geomagnetic Kp index forecasting method and system based on similarity algorithm |
CN113344032A (en) * | 2021-05-13 | 2021-09-03 | 中国科学院国家空间科学中心 | Geomagnetic Kp index forecasting method and system based on similarity algorithm |
CN116136975A (en) * | 2023-02-24 | 2023-05-19 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | Neutral atmosphere temperature density prediction method and equipment based on LSTM neural network |
CN116136975B (en) * | 2023-02-24 | 2023-08-15 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | Neutral atmosphere temperature density prediction method and equipment based on LSTM neural network |
CN116882594A (en) * | 2023-09-06 | 2023-10-13 | 数字太空(北京)智能技术研究院有限公司 | Geomagnetic ap index medium-term forecasting method based on limited recursion and deep learning |
CN116882594B (en) * | 2023-09-06 | 2023-12-01 | 数字太空(北京)智能技术研究院有限公司 | Geomagnetic ap index medium-term forecasting method based on limited recursion and deep learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107563574A (en) | A kind of earth's magnetic field Forecasting Methodology based on long memory models Recognition with Recurrent Neural Network in short-term | |
CN107390298B (en) | A kind of analogy method and device of Complex Mountain underlying surface strong wind | |
CN105911037B (en) | Manganese and association heavy metal distribution Forecasting Methodology in the soil water termination contaminated stream of Manganese Ore District | |
CN105243435B (en) | A kind of soil moisture content prediction technique based on deep learning cellular Automation Model | |
Chauhan et al. | Landslide susceptibility zonation through ratings derived from artificial neural network | |
CN109085641B (en) | Monitoring data processing method for earthquake prediction, earthquake prediction method and system | |
CN107703564A (en) | A kind of precipitation predicting method, system and electronic equipment | |
CN108460739A (en) | A kind of thin cloud in remote sensing image minimizing technology based on generation confrontation network | |
CN113222316B (en) | Variation scenario simulation method based on FLUS model and biodiversity model | |
CN110442666A (en) | A kind of mineral resource prediction method and system based on neural network model | |
CN103455702A (en) | Method for determining regional ionospheric layer delay | |
CN104361255B (en) | It is a kind of to improve cellular automata urban sprawl analogy method | |
CN107783185A (en) | A kind of processing method and processing device of tomographic statics | |
CN111859054B (en) | Meteorological satellite data processing method and device | |
CN114399073A (en) | Ocean surface temperature field prediction method based on deep learning | |
CN107944550A (en) | A kind of space-time big data Forecasting Methodology based on detection type depth network | |
CN110927793B (en) | Reservoir prediction method and system based on sequential random fuzzy simulation | |
Adnan Ikram et al. | Novel evolutionary-optimized neural network for predicting landslide susceptibility | |
CN107016205A (en) | A kind of multi-model construction method of groundwater Numerical Simulation | |
CN113902580A (en) | Historical farmland distribution reconstruction method based on random forest model | |
CN114861840A (en) | Multi-source precipitation data fusion method | |
CN104424373B (en) | A kind of fine expression of space variable correlation | |
Priatna et al. | Precipitation prediction using recurrent neural networks and long short-term memory | |
Huang et al. | Uncertainties of landslide susceptibility prediction: influences of random errors in landslide conditioning factors and errors reduction by low pass filter method | |
CN112668606B (en) | Step type landslide displacement prediction method based on gradient elevator and quadratic programming |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180109 |