CN109100759A - A kind of ionosphere Amplitude scintillation detection method based on machine learning - Google Patents

A kind of ionosphere Amplitude scintillation detection method based on machine learning Download PDF

Info

Publication number
CN109100759A
CN109100759A CN201810861939.5A CN201810861939A CN109100759A CN 109100759 A CN109100759 A CN 109100759A CN 201810861939 A CN201810861939 A CN 201810861939A CN 109100759 A CN109100759 A CN 109100759A
Authority
CN
China
Prior art keywords
data
scintillation
sample
model
value
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201810861939.5A
Other languages
Chinese (zh)
Other versions
CN109100759B (en
Inventor
祝雪芬
林梦颖
陈熙源
汤新华
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Southeast University
Original Assignee
Southeast University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Southeast University filed Critical Southeast University
Priority to CN201810861939.5A priority Critical patent/CN109100759B/en
Publication of CN109100759A publication Critical patent/CN109100759A/en
Application granted granted Critical
Publication of CN109100759B publication Critical patent/CN109100759B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S19/00—Satellite radio beacon positioning systems; Determining position, velocity or attitude using signals transmitted by such systems
    • G01S19/01—Satellite radio beacon positioning systems transmitting time-stamped messages, e.g. GPS [Global Positioning System], GLONASS [Global Orbiting Navigation Satellite System] or GALILEO
    • G01S19/13—Receivers
    • G01S19/35—Constructional details or hardware or software details of the signal processing chain
    • G01S19/37—Hardware or software details of the signal processing chain
    • 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Remote Sensing (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Signal Processing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Position Fixing By Use Of Radio Waves (AREA)

Abstract

The invention discloses a kind of ionosphere Amplitude scintillation detection method of combination machine learning, this method detects ionosphere Amplitude scintillation signal using the SVM algorithm in machine learning, judges whether that scintillation event occurs according to this.In the process, the GPS signal received is handled and is calculated first Amplitude scintillation index S4 maximum value and average value, corresponding scintillation event is marked as learning sample, setting label indicates scintillation event for 1, does not occur for -1 expression scintillation event.Then sample is inputted in SVM classifier and is learnt, obtain optimum classifier.When new scintillation event feature vector enters SVM classifier, will classify automatically to it.The detection method can detect a large amount of scintillation events simultaneously, while improving detection efficiency, simplify detection process and obtain relatively high and stable accuracy.

Description

A kind of ionosphere Amplitude scintillation detection method based on machine learning
Technical field
The present invention relates to wireless communication field more particularly to a kind of ionosphere Amplitude scintillation detection sides based on machine learning Method.
Background technique
As the satellite technologies such as GNSS are in the more and more extensive application of modern society, ionospheric scintillation and its to radiofrequency signal Influence has become very important pith.Ionospheric scintillation is Refer to the rapid fluctuations of radiofrequency signal (such as GNSS) amplitude propagated in ionosphere and phase.Due to generating this phenomenon Have very much, including but not limited to solar activity, magnetic storm, internal field, conductivity, wave interaction etc..More specifically, high latitude The flashing for spending area is usually related with solar activity and magnetic storm period, therefore flashes prediction and model extremely difficult.Under the line and Low latitudes, the abnormal ionosphere bubble formed with post sunset of equator ionization are the potential causes of scintillation event.Flashing can shadow All spacing wireless electric signals for penetrating ionosphere are rung, and may cause precision and the decline of successional performance.Strong flashing can be tight Ghost image rings the signal acquisition and tracking of GNSS receiver, leads to losing lock and navigation failure.Therefore, seek accurately and efficiently to ionize Layer flicker detection method not only facilitates design performance more preferably shadow of the receiver to improve positioning accuracy, reduce ionospheric scintillation It rings, while also can provide help for ionosphere and space weather model foundation.
Traditional ionospheric scintillation detection is to observe detection data by researcher to carry out manual identified, this process consumption Duration and all data can not be traversed.Propose automatic flicker detection method thus to improve detection efficiency.It is most familiar of with Detection method based on experience is to judge Amplitude scintillation index S4, and scintillation event occurs if more than 0.2.Separately there is more complicated base In the detection algorithm of NP etection theory, such as wavelet decomposition method, i.e., by assuming that wavelet coefficient Gaussian distributed distinguishes flashing With non-flickering event；Separately there is a kind of integrated empirical mode decomposition method of complementation, this method passes through the precise measurement to carrier-to-noise ratio To detect flashing.Although above-mentioned algorithm can detect scintillation event, in the design process by unoptimizable experience The limitation of threshold value or ideal hypothesized model.
Summary of the invention
Goal of the invention: in view of the problems of the existing technology the present invention, provides a kind of ionosphere width based on machine learning Flicker detection method is spent, the SVM algorithm in the method combination machine learning is right by learning ionosphere Amplitude scintillation feature Scintillation event is detected automatically.Compared to other methods, detection accuracy had both been effectively increased, has simplified detecting step, again High-volume data can be handled simultaneously, effectively increase detection efficiency.
Technical solution: the ionosphere Amplitude scintillation detection method of the present invention based on machine learning includes:
(1) data collected for different location GPS receiver use length of window for a seconds and move b seconds every time Moving window obtains data, calculates Amplitude scintillation index S for the data that moving window each time obtains4；Wherein, a, b are positive Integer, and a > b；
(2) the collected data of different location GPS receiver are divided with every t minutes for a data block, it will be every Characteristic quantity of the maximum value and average value of Amplitude scintillation index in a data block as the data block, and should using label label Whether data block occurs scintillation event, wherein t is positive integer；
(3) using the characteristic quantity of partial data block and corresponding label as training sample, and it is divided into according to label and is flashed Event sample and two class of scintillation event sample does not occur, the characteristic quantity and corresponding label of remaining data block are as verification sample；
(4) Linear SVM sorter model is established；
(5) two class training sample input linear SVM classifier models are subjected to cross validation, obtain SVM classifier model In optimal hyper parameter, obtain optimal SVM classifier；
(6) characteristic quantity verified in sample optimal SVM classifier is inputted to classify, by the classification results of output with it is right It answers label to compare, then thinks SVM classifier qualification when accuracy rate reaches preset value；
(7) characteristic quantity of unknown scintillation event data is input in qualified SVM classifier, SVM classifier it is defeated It is out classification results.
Further, step (1) specifically includes:
(1.1) digital medium-frequency signal that GPS receiver receives is multiplied with local quadrature carrier signals respectively generate I and I and Q baseband are multiplied to obtain I by Q two-way baseband signal with i.e. time-code respectivelypAnd Qp；
(1.2) according to IpAnd QpThe broadband power and narrow band power of signal are calculated using following formula:
In formula, Ip,i、Qp,iRespectively to Ip、QpThe ith sample value once obtained is sampled with every h milliseconds, Δ t is power Counting period time value, J are indicated IpAnd QpAll sampled values be divided into J segmentation, W by interval of Δ tBP,j、NBP,j Respectively indicate the broadband power and narrow band power of j-th of block signal；
(1.3) the normalized signal intensity of signal is calculated using following formula according to the broadband power of signal and narrow band power:
SI,raw,j=NBP,j-WBP,j
In formula, SI,norm,jIndicate the normalized signal intensity of j-th of block signal, SI,trend,jIt indicates according to SI,raw,j4 Trend signal strength is gone obtained by rank polynomial fitting；
(1.4) it uses length of window for a seconds and the moving window moved every time b seconds obtains data, for moving each time The data that window obtains calculate its Amplitude scintillation index S4, wherein the Amplitude scintillation index for the data that kth time moving window obtains Value S4Are as follows:
In formula,Indicate the normalized signal intensity total collection for a second data being located in k moving window, standardization Signal strength data total a*1000/ Δ t, mathematic expectaion is sought in E [] expression.
Further, step (2) specifically includes:
(2.1) the collected data of different location GPS receiver are carried out non-overlapping stroke with every t minutes for a data block Point；
(2.2) the multiple Amplitude scintillation index Ss being calculated according to data in each data block are obtained4, therefrom extract most Big value S4,maxWith average value S4,avgMark whether the data block flashes as the characteristic quantity of the data block, and using label Event, with being expressed mathematically as:
Characteristic quantity:
Label:
In formula, l indicates data block sequence number,Indicate two dimensional vector space.
Further, the Linear SVM sorter model established in step (4) are as follows:
Constraint condition:
In formula, w is parameter matrix to be asked, and b is parameter to be asked, For the one-dimensional space, ξlIt is first The slack variable of training sample, m are training samples number, and C is hyper parameter, are indicated to the training sample for being more than maximization boundary Tolerance, x(l)、y(l)Respectively indicate the characteristic quantity and label of first of training sample.
Further, step (5) specifically includes:
(5.1) Lagrange multiplier α is introducedl,βl, Linear SVM sorter model is indicated are as follows:
Respectively by L to wi,bi,ξiDerivation juxtaposition 0:
(5.2) the above results are substituted into the model that step (4) are established, is converted to dual form simultaneously further according to strong dual relationship Remove negative sign, obtain model:
Constraint condition:
Hidden conditional:
The model is solved in MATLAB using function quadprog, the optimal value w of w is obtained0；And according to w0It solves Obtain b optimal value b0=y(s)-w0 Tx(s), wherein ξs=0, x(s)It is α for supporting vectorlTraining sample feature corresponding to ≠ 0 Amount, y(s)For corresponding label；
(5.3) training sample characteristic quantity is denoted as X=(x(1),x(2),...,x(m)), i.e. the matrix of 2 × m；Label is denoted as Y =(y(1),y(2),...,y(m)), i.e. above-mentioned matrix and vector are combined into the matrix Z=(X of 3 × m by the row vector of 1 × m；Y), As sample input matrix；
(5.5) sample input matrix is integrally inputted into SVM classifier model, and cross validation broken number and hyper parameter C is set Value the sample number of input is divided into u parts at random in the training process, wherein every u-1 parts be used to model is learnt, Remaining 1 part is tested the model learnt, obtains test accuracy rate, obtains current hyper parameter C after successively carrying out u training altogether Corresponding average test accuracy rate；
(5.6) value of hyper parameter C is changed, and returns to step (5.5), so that the value for obtaining different hyper parameter C is corresponding Average test accuracy rate；
(5.7) the corresponding Average Accuracy of all hyper parameters is compared, finds out the corresponding hyper parameter C of maximum accuracy rate As best hyper parameter, training gained model is optimal classification model under the parameter setting.
The utility model has the advantages that compared with prior art, the present invention its remarkable advantage is: the invention proposes one kind to be based on engineering The ionosphere Amplitude scintillation detection method of habit, the method is first from the data for whether scintillation event occurring known to detecting It extracts feature simultaneously to mark, arranges as sample, input in established SVM classifier model and learnt and find out optimal classification device Model then tests the classifier learnt, can get compared with high-accuracy, which is better than other methods. Finally the classifier is applied in new scintillation event, classification and Detection can be carried out automatically, and refer to according to flashing compared to tradition Number S4 judges that the method whether flashing occurs, the method have higher accuracy, is able to detect that scintillation index lower than 0.2 Scintillation event, this is of great significance for ionospheric structure model and scintillation mechanisms research, and this method can be located simultaneously High-volume data are managed, detection efficiency is largely improved.
Detailed description of the invention
Fig. 1 is the flow chart of one embodiment of the present of invention.
Specific embodiment
As shown in Figure 1, the ionosphere Amplitude scintillation detection method of the present invention based on machine learning includes:
(1) data collected for different location GPS receiver use length of window for 10 seconds and move 1 second every time Moving window obtain data, for moving window each time obtain data calculate Amplitude scintillation index S4。
Step (1) specifically includes:
(1.1) digital medium-frequency signal that GPS receiver receives is multiplied with local quadrature carrier signals respectively generate I and I and Q baseband are multiplied to obtain I by Q two-way baseband signal with i.e. time-code respectivelypAnd Qp；
(1.2) according to IpAnd QpThe broadband power and narrow band power of signal are calculated using following formula:
In formula, Ip,i、Qp,iRespectively to Ip、QpThe ith sample value once obtained is sampled with every 1ms, Δ t is power meter Interval time value is calculated, 20ms, J is taken to indicate IpAnd QpAll sampled values be divided into J segmentation by interval of 20ms, WBP,j、NBP,jRespectively indicate the broadband power and narrow band power of j-th of 20ms signal；
(1.3) the normalized signal intensity of signal is calculated using following formula according to the broadband power of signal and narrow band power:
SI,raw,j=NBP,j-WBP,j
In formula, SI,norm,jIndicate the normalized signal intensity of j-th of block signal, SI,trend,jIt indicates according to SI,raw,j4 Trend signal strength is gone obtained by rank polynomial fitting；
(1.4) it uses length of window for 10 seconds and the moving window moved every time 1 second obtains data, for moving each time The data that window obtains calculate its Amplitude scintillation index S410 moving windows can be then moved in 10s, can be counted within every 1 second Calculation obtains 1 Amplitude scintillation index value S4。
The Amplitude scintillation index value S for the data that kth time moving window obtains4Are as follows:
In formula,Indicate the normalized signal intensity total collection for a second data being located in k moving window, standardization Signal strength data total 10*1000/20=500, mathematic expectaion is sought in E [] expression.
(2) the collected data of different location GPS receiver were divided with every 3 minutes for a data block, it will be every Characteristic quantity of the maximum value and average value of Amplitude scintillation index in a data block as the data block, and should using label label Whether data block occurs scintillation event.
Step (2) specifically includes:
(2.1) the collected data of different location GPS receiver were carried out non-overlapping stroke with every 3 minutes for a data block Point；
(2.2) the multiple Amplitude scintillation index Ss being calculated according to data in each data block are obtained4, therefrom extract most Big value S4,maxWith average value S4,avgMark whether the data block flashes as the characteristic quantity of the data block, and using label Event, with being expressed mathematically as:
Characteristic quantity:
Label:
In formula, l indicates data block sequence number,Indicate two dimensional vector space.
(3) using the characteristic quantity of 80% data block and corresponding label as training sample, and it is divided into according to label and is flashed Event sample and two class of scintillation event sample does not occur, the characteristic quantity and corresponding label of remaining data block are as verification sample.
(4) Linear SVM sorter model model y=w is establishedTX+b finds out parameter w0And b0So that y=w0 Tx+b0As one Sample is divided into two classes by a plane, and guarantees that the sample nearest apart from the hyperplane two sides has farthest spacing, has these The sample point of feature is " supporting vector (SV) ".Above-mentioned maximization border issue is expressed as following mathematics shape through a series of conversions Formula:
Constraint condition:
In formula, w is parameter matrix to be asked, and b is parameter to be asked, For the one-dimensional space, ξlIt is first The slack variable of training sample, m are training samples number, and C is hyper parameter, are indicated to the training sample for being more than maximization boundary Tolerance, x(l)、y(l)Respectively indicate the characteristic quantity and label of first of training sample.
(5) two class training sample input linear SVM classifier models are subjected to cross validation, obtain SVM classifier model In optimal hyper parameter, obtain optimal SVM classifier.
Step (5) specifically includes:
(5.1) Lagrange multiplier α is introducedl,βl, Linear SVM sorter model is indicated are as follows:
Respectively by L to wi,bi,ξiDerivation juxtaposition 0:
(5.2) the above results are substituted into the model that step (4) are established, is converted to dual form simultaneously further according to strong dual relationship Remove negative sign, obtain model:
Constraint condition:
Hidden conditional:
I.e. be converted into quadratic programming problem seeks α to the above probleml, it can be solved in MATLAB using function quadprog, It substitutes into hidden conditional and acquires the w for meeting and maximizing border issue0Vector.In addition, Lagrangian feature determines αl≠ 0 I-th corresponding of training sample is supporting vector SV (with x(s)Indicate), it is corresponding to find out b0=y(s)-w0 Tx(s), wherein ξs=0.
(5.3) by maximization problems by MATLAB Classification Learner model carry out sample learning come Find optimal classification device.Training sample characteristic quantity is denoted as X=(x(1),x(2),...,x(m)), i.e. the matrix of 2 × m；Label is denoted as Y=(y(1),y(2),...,y(m)), i.e. above-mentioned matrix and vector are combined into the matrix Z=(X of 3 × m by the row vector of 1 × m；Y), As sample input matrix；
(5.5) sample input matrix is integrally inputted matrix to the Classification Learner in MATLAB, and Linear SVM classifier model is selected, the value of cross validation broken number and hyper parameter C is set, in the training process, at random by input Sample number is divided into u parts, wherein every u-1 parts is used to learn model, remaining 1 part is tested the model learnt, Test accuracy rate is obtained, obtains the corresponding average test accuracy rate of current hyper parameter C after successively carrying out u training altogether；
(5.6) value of hyper parameter C is changed, and returns to step (5.5), so that the value for obtaining different hyper parameter C is corresponding Average test accuracy rate；Change hyper parameter C value can approximate 3 times of growth rate parameter C values, such as 0.001,0.003 are set, 0.01,0.03,0.1,0.3,1,3,10,30；
(5.7) the corresponding Average Accuracy of all hyper parameters is compared, finds out the corresponding hyper parameter C of maximum accuracy rate As best hyper parameter, training gained model is optimal classification model under the parameter setting.
(6) characteristic quantity verified in sample optimal SVM classifier is inputted to classify, by the classification results of output with it is right It answers label to compare, then thinks SVM classifier qualification when accuracy rate reaches preset value.
(7) characteristic quantity of unknown scintillation event data is input in qualified SVM classifier, SVM classifier it is defeated It is out classification results.Operation result is the label of new scintillation event, is judged as there is scintillation event for 1, is -1 It is judged as that scintillation event does not occur.
Data sample of the present invention chooses difference, and accuracy rate can generate small deviation, count after tested, and accuracy rate can reach 96%, compared to conventional method, improve accuracy in detection.
Above disclosed is only a preferred embodiment of the present invention, and the right model of the present invention cannot be limited with this It encloses, therefore equivalent changes made in accordance with the claims of the present invention, is still within the scope of the present invention.

Claims (5)

1. a kind of ionosphere Amplitude scintillation detection method based on machine learning characterized by comprising
(1) data collected for different location GPS receiver use length of window for a seconds and move movement in b seconds every time Window obtains data, calculates Amplitude scintillation index S for the data that moving window each time obtains4；Wherein, a, b are positive integer, And a > b；
(2) the collected data of different location GPS receiver are divided with every t minutes for a data block, by every number Characteristic quantity according to the maximum value and average value of the Amplitude scintillation index in block as the data block, and the data are marked using label Whether block occurs scintillation event, wherein t is positive integer；
(3) using the characteristic quantity of partial data block and corresponding label as training sample, and generation scintillation event is divided into according to label Sample and two class of scintillation event sample does not occur, the characteristic quantity and corresponding label of remaining data block are as verification sample；
(4) Linear SVM sorter model is established；
(5) two class training sample input linear SVM classifier models are subjected to cross validation, obtained in SVM classifier model Optimal hyper parameter obtains optimal SVM classifier；
(6) characteristic quantity verified in sample is inputted optimal SVM classifier to classify, by the classification results of output and corresponding mark Label compare, and then think SVM classifier qualification when accuracy rate reaches preset value；
(7) characteristic quantity of unknown scintillation event data is input in qualified SVM classifier, the output of SVM classifier is For classification results.
2. the ionosphere Amplitude scintillation detection method according to claim 1 based on machine learning, it is characterised in that: step (1) it specifically includes:
(1.1) digital medium-frequency signal that GPS receiver receives is multiplied with local quadrature carrier signals respectively and generates I and Q two I and Q baseband are multiplied to obtain I by roadbed band signal with i.e. time-code respectivelypAnd Qp；
(1.2) according to IpAnd QpThe broadband power and narrow band power of signal are calculated using following formula:
In formula, Ip,i、Qp,iRespectively to Ip、QpThe ith sample value once obtained is sampled with every h milliseconds, Δ t is power calculation Interval time value, J are indicated IpAnd QpAll sampled values be divided into J segmentation, W by interval of Δ tBP,j、NBP,jRespectively Indicate the broadband power and narrow band power of j-th of block signal；
(1.3) the normalized signal intensity of signal is calculated using following formula according to the broadband power of signal and narrow band power:
SI,raw,j=NBP,j-WBP,j
In formula, SI,norm,jIndicate the normalized signal intensity of j-th of block signal, SI,trend,jIt indicates according to SI,raw,j4 ranks it is quasi- It closes and goes trend signal strength obtained by multinomial；
(1.4) it uses length of window for a seconds and the moving window moved every time b seconds obtains data, for moving window each time The data of acquisition calculate its Amplitude scintillation index S4, wherein the Amplitude scintillation index value S for the data that kth time moving window obtains4 Are as follows:
In formula,Indicate the normalized signal intensity total collection for a second data being located in k moving window, normalized signal Intensity data total a*1000/ Δ t, mathematic expectaion is sought in E [] expression.
3. the ionosphere Amplitude scintillation detection method according to claim 1 based on machine learning, it is characterised in that: step (2) it specifically includes:
(2.1) the collected data of different location GPS receiver are subjected to non-overlapping division with every t minutes for a data block；
(2.2) the multiple Amplitude scintillation index Ss being calculated according to data in each data block are obtained4, therefrom extract maximum value S4,maxWith average value S4,avgMark whether the data block occurs scintillation event as the characteristic quantity of the data block, and using label, With being expressed mathematically as:
Characteristic quantity:
Label:
In formula, l indicates data block sequence number,Indicate two dimensional vector space.
4. the ionosphere Amplitude scintillation detection method according to claim 1 based on machine learning, it is characterised in that: step (4) the Linear SVM sorter model established in are as follows:
Constraint condition:
In formula, w is parameter matrix to be asked, and b is parameter to be asked, For the one-dimensional space, ξlFor first of training The slack variable of sample, m are training samples number, and C is hyper parameter, indicate the tolerance to the training sample for being more than maximization boundary Degree, x(l)、y(l)Respectively indicate the characteristic quantity and label of first of training sample.
5. the ionosphere Amplitude scintillation detection method according to claim 4 based on machine learning, it is characterised in that: step (5) it specifically includes:
(5.1) Lagrange multiplier α is introducedl,βl, Linear SVM sorter model is indicated are as follows:
Respectively by L to wi,bi,ξiDerivation juxtaposition 0:
(5.2) the above results are substituted into the model that step (4) are established, is converted to dual form further according to strong dual relationship and removes Negative sign obtains model:
Constraint condition:
Hidden conditional:
The model is solved in MATLAB using function quadprog, the optimal value w of w is obtained0；And according to w0Solution obtains b Optimal value b0=y(s)-w0 Tx(s), wherein x(s)Corresponding ξs=0, x(s)It is α for supporting vectorlTraining sample corresponding to ≠ 0 Characteristic quantity, y(s)For corresponding label；
(5.3) training sample characteristic quantity is denoted as X=(x(1),x(2),...,x(m)), i.e. the matrix of 2 × m；Label is denoted as Y=(y(1),y(2),...,y(m)), i.e. above-mentioned matrix and vector are combined into the matrix Z=(X of 3 × m by the row vector of 1 × m；Y), as Sample input matrix；
(5.5) sample input matrix is integrally inputted into SVM classifier model, and the value of cross validation broken number and hyper parameter C is set, In the training process, the sample number of input is divided into u parts at random, wherein every u-1 parts is used to learn model, is left 1 Part tests the model learnt, obtains test accuracy rate, and it is corresponding successively to obtain current hyper parameter C after u training of progress altogether Average test accuracy rate；
(5.6) value of hyper parameter C is changed, and returns to step (5.5), so that the value for obtaining different hyper parameter C is corresponding flat Equal test accuracy rate；
(5.7) the corresponding Average Accuracy of all hyper parameters is compared, finding out the corresponding hyper parameter C of maximum accuracy rate is Best hyper parameter, training gained model is optimal classification model under the parameter setting.
CN201810861939.5A 2018-08-01 2018-08-01 Ionosphere amplitude flicker detection method based on machine learning Active CN109100759B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810861939.5A CN109100759B (en) 2018-08-01 2018-08-01 Ionosphere amplitude flicker detection method based on machine learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810861939.5A CN109100759B (en) 2018-08-01 2018-08-01 Ionosphere amplitude flicker detection method based on machine learning

Publications (2)

Publication Number Publication Date
CN109100759A true CN109100759A (en) 2018-12-28
CN109100759B CN109100759B (en) 2020-05-05

Family

ID=64848151

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810861939.5A Active CN109100759B (en) 2018-08-01 2018-08-01 Ionosphere amplitude flicker detection method based on machine learning

Country Status (1)

Country Link
CN (1) CN109100759B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113031036A (en) * 2021-03-01 2021-06-25 中国矿业大学 Ionosphere phase flicker factor construction method based on GNSS 30s sampling frequency data
CN114417742A (en) * 2022-04-01 2022-04-29 中国工程物理研究院流体物理研究所 Laser atmospheric flicker index prediction method and system

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106249256A (en) * 2016-07-08 2016-12-21 辽宁工程技术大学 Real-time GLONASS phase deviation estimation method based on particle swarm optimization algorithm
CN107991690A (en) * 2017-06-28 2018-05-04 江苏省测绘工程院 A kind of computational methods of the scintillation index based on Beidou navigation

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106249256A (en) * 2016-07-08 2016-12-21 辽宁工程技术大学 Real-time GLONASS phase deviation estimation method based on particle swarm optimization algorithm
CN107991690A (en) * 2017-06-28 2018-05-04 江苏省测绘工程院 A kind of computational methods of the scintillation index based on Beidou navigation

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
ZHU XUEFEN 等: "Analysis of GPS ionospheric scintillation signal amplitude fading characteristics at low latitude", 《JOURNAL OF SOUTHEAST UNIVERSITY ( ENGLISH EDITION)》 *
ZHU XUE-FEN: "Influence of strong ionospheric scintillation on GPS software receiver", 《中国惯性技术学报》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113031036A (en) * 2021-03-01 2021-06-25 中国矿业大学 Ionosphere phase flicker factor construction method based on GNSS 30s sampling frequency data
CN113031036B (en) * 2021-03-01 2021-09-24 中国矿业大学 Ionosphere phase flicker factor construction method based on GNSS 30s sampling frequency data
CN114417742A (en) * 2022-04-01 2022-04-29 中国工程物理研究院流体物理研究所 Laser atmospheric flicker index prediction method and system
CN114417742B (en) * 2022-04-01 2022-06-10 中国工程物理研究院流体物理研究所 Laser atmospheric flicker index prediction method and system

Also Published As

Publication number Publication date
CN109100759B (en) 2020-05-05

Similar Documents

Publication Publication Date Title
Ruwali et al. Implementation of hybrid deep learning model (LSTM-CNN) for ionospheric TEC forecasting using GPS data
Jiao et al. Automatic equatorial GPS amplitude scintillation detection using a machine learning algorithm
CN109508730B (en) Ionosphere phase scintillation detection method based on nonlinear SVM algorithm
CN111832220A (en) Lithium ion battery health state estimation method based on codec model
CN109274624B (en) Carrier frequency offset estimation method based on convolutional neural network
CN110532932A (en) A kind of multi -components radar emitter signal intra-pulse modulation mode recognition methods
CN109658948A (en) One kind is towards the movable acoustic monitoring method of migratory bird moving
CN103093244B (en) A kind of Radar Signal Recognition method based on Its Sparse Decomposition
CN110132373A (en) A kind of large-scale inland lake and wetland water level inversion method based on Altimetry Data
CN110555841A (en) SAR image change detection method based on self-attention image fusion and DEC
CN109100759A (en) A kind of ionosphere Amplitude scintillation detection method based on machine learning
US20240078413A1 (en) Massive data-driven method for automatically locating mine microseismic source
CN109450573A (en) A kind of frequency spectrum sensing method based on deep neural network
CN115951430A (en) Geomagnetic index forecasting method based on LSTM
Zhong et al. Multi-factor intensity estimation for tropical cyclones in the western North Pacific based on the deviation angle variance technique
US20220051075A1 (en) Methods and apparatuses for tracking weak signal traces
Wang et al. Radar HRRP target recognition in frequency domain based on autoregressive model
Darmawan et al. GRU and XGBoost Performance with Hyperparameter Tuning Using GridSearchCV and Bayesian Optimization on an IoT-Based Weather Prediction System.
Fu et al. Navigation jamming signal recognition based on long short-term memory neural networks
CN117272783A (en) Power battery life prediction method based on cloud edge collaborative multi-model fusion
CN115952410A (en) Landslide hazard detection system based on deep learning
CN105738950A (en) Directive bunching detection method aiming at specific regions
CN116127354A (en) High-voltage cable partial discharge fault positioning method and system based on deep learning
Luwanga et al. Automatic spread-F detection using deep learning
Qiu et al. A short-term precipitation prediction model based on spatiotemporal convolution network and ensemble empirical mode decomposition

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