CN111553393B - Solar radiation storm intensity judging method based on SVM multi-classification algorithm - Google Patents

Solar radiation storm intensity judging method based on SVM multi-classification algorithm Download PDF

Info

Publication number
CN111553393B
CN111553393B CN202010311617.0A CN202010311617A CN111553393B CN 111553393 B CN111553393 B CN 111553393B CN 202010311617 A CN202010311617 A CN 202010311617A CN 111553393 B CN111553393 B CN 111553393B
Authority
CN
China
Prior art keywords
classification
satellite
storm
noise ratio
sample
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
Application number
CN202010311617.0A
Other languages
Chinese (zh)
Other versions
CN111553393A (en
Inventor
祝雪芬
罗铱镅
林梦颖
杨帆
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 CN202010311617.0A priority Critical patent/CN111553393B/en
Publication of CN111553393A publication Critical patent/CN111553393A/en
Application granted granted Critical
Publication of CN111553393B publication Critical patent/CN111553393B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Position Fixing By Use Of Radio Waves (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a solar radiation storm intensity judging method based on an SVM multi-classification algorithm, wherein the basic principle of the SVM multi-classification algorithm is to combine every two types into a two-classifier aiming at a multi-classification problem, find out an optimal hyperplane for each classifier, divide a sample into two types, and finally obtain a classification result through a voting statistical method. In the process, firstly, data capable of reflecting the solar power storm intensity of the observation earth satellite are extracted, data preprocessing is carried out, feature vectors are obtained, and the type of the solar power storm intensity is marked. And then inputting the samples into an SVM classifier for learning to obtain an optimal classifier. When a new feature vector enters the classifier, the classification will be performed automatically. The judging method can automatically judge the intensity type of the solar power storm, has higher efficiency and accuracy, does not depend on a radio telescope, and has low cost.

Description

Solar radiation storm intensity judging method based on SVM multi-classification algorithm
Technical Field
The invention relates to the technical field of wireless communication, in particular to a solar radiation storm intensity judging method based on an SVM multi-classification algorithm.
Background
With the wider and wider application of satellite technologies such as GNSS in modern society, the influence of solar storm intensity on GNSS signals has become a non-negligible important part. Because the radio burst phenomenon usually occurs very suddenly and is often accompanied by proton burst, X-ray burst or flare burst in the solar active area, the radiation intensity is large and the variation is severe, and when reaching the earth, a series of geophysical effects such as magnetic storm, laser, communication interference and the like can be caused. During the previous multi-sun radio explosion period, the solar radio storm may cause the conditions of reduced satellite carrier-to-noise ratio, increased positioning error, increased geometric precision factor, greatly reduced satellite unlocking and visible star count, etc., so that the noise signal generated by the solar radio storm is one of the influencing factors of navigation signals.
Because of different solar storm intensities, the interference on GNSS signals is different, so that the navigation system can work normally when the solar storm happens, the classification work of the solar storm is beneficial to the research on the solar storm, and has great significance for maintaining the normal operation of the satellite navigation system. Meanwhile, the intensity of the solar power storm also affects the weather, and the classification work of the solar power storm is favorable for early warning on disastrous weather, so that the solar power storm has good practical value.
The traditional classification method generally needs astronomists with expert knowledge to manually classify the intensity of the solar radiation storm, and is time-consuming and labor-consuming, and low in efficiency and accuracy.
Disclosure of Invention
In order to solve the above-mentioned problems. The invention provides a solar power storm intensity judging method based on an SVM multi-classification algorithm, which comprehensively examines influence generated by GNSS signals when solar power storm occurs and judges the intensity of the solar power storm by combining with the SVM algorithm in machine learning. The method has feasibility for classifying the solar radio storm intensity, and the judging result can simultaneously give out the solar radio storm intensity types of a plurality of stations. Compared with the traditional method, the method combines multiple factors, saves time and labor cost, and improves the identification accuracy and efficiency relatively. To achieve this object:
the invention provides a solar radiation storm intensity judging method based on an SVM multi-classification algorithm, which comprises the following specific steps:
(1) Carrying out data preprocessing, and calculating feature vectors at different moments of an observation ground, wherein the feature vectors comprise an observation ground carrier-to-noise ratio (carrier-to-noise ratio) decline value, an observation ground total positioning error, a geometric precision GDOP factor and a satellite unlocking number;
(2) Dividing the solar electric storm into 4 types of solar electric storm which do not occur, have weak intensity, strong intensity and strong intensity according to the change of the electric current flow, and respectively assigning labels corresponding to the 4 types of characteristic vectors as 1,2,3 and 4;
(3) Combining each two types to form a classifier to obtain 6 groups of training sets, constructing an unknown nonlinear SVM (support vector machine) classification model for each group of training sets, and performing cross validation on training samples to select optimal parameters to obtain 6 groups of trained classification models;
(4) And (3) inputting the feature vectors extracted from the satellite navigation signals to be identified into the classification models in the step (3), and outputting a result by each model, wherein if the judgment result is p, the p type counts one ticket, and finally the judgment result of the predicted sample is the type with the largest number of tickets.
As a further refinement of the present invention, step (1) specifically includes:
(1.1) inputting the carrier-to-noise ratio of each satellite in the observation ground, the positioning error of the observation ground, the geometric accuracy GDOP factor and the number of satellite unlocking;
(1.2) data preprocessing is performed on the observed ground load-to-noise ratio:
taking the average value of the current carrier-to-noise ratio decline values of each satellite as the carrier-to-noise ratio decline value of the station, and marking as x 1 Units: dBHz;
snr i mean value of the measured satellite carrier-to-noise ratio at the first 8 moments-value of carrier-to-noise ratio at the moment
Figure BDA0002458058950000021
Wherein, snr i N is the number of satellites for each satellite carrier-to-noise ratio reduction value;
(1.3) the observed ground positioning error consists of the positioning errors of X, Y, Z in the northeast and north coordinate system, and the standard deviation is taken as the positioning error of the station in the three directions and is recorded as x 2 Units: m;
Figure BDA0002458058950000022
wherein r is i (i=1, 2, 3) represents the positioning error in the direction of the observation ground X, Y, Z at that time, respectively;
(1.4) GDOP factor is denoted as x 3 The number of satellite lock loss is recorded as x 4 Units: and each.
As a further refinement of the present invention, step (2) specifically includes:
reducing the observed ground load to noise ratio by x 1 Total positioning error x of observation ground 2 GDOP factor x 3 Number of satellite lock loss x 4 As the feature vector x, the sample feature vector is:
Figure BDA0002458058950000023
classifying the solar radio storm intensities according to the following table;
TABLE 1 solar storm intensity class classification
Figure BDA0002458058950000024
Figure BDA0002458058950000031
Where i represents the ith sample, x (i) A feature vector representing the i-th sample, y (i) A label representing the ith sample, (x) (i) ,y (i) ) I.e. the sample point.
As a further refinement of the present invention, step (3) specifically includes:
(3.1) training sets of 6 groups are training samples corresponding to labels 1 and 2, 1 and 3, 1 and 4, 2 and 3, 2 and 4 respectively;
(3.2) assuming a classifier (k, l=1, 2,3,4 and k < l) constructed for training samples with labels k and l, the nonlinear SVM classification model is specifically:
f(x)＝w T Φ(x)+b
where Φ (x) is a high-dimensional linear mapping, mapping function mapping 4-dimensional eigenvector x to M dimensions, M > 4,
Figure BDA0002458058950000032
both are parameters to be solved;
(3.3) solving the following optimization problem:
Figure BDA0002458058950000033
Figure BDA0002458058950000034
in xi i For the relaxation variable of each sample i, ζ i The larger the representative sample point is, the farther the representative sample point is from the group, C is a super parameter, and the larger C represents the more importance is placed on the objective function loss brought by the outlier, y (i) Label for sample i, x (i) A feature vector representing sample i;
(3.4) solving for the lagrangian multiplier:
Figure BDA0002458058950000035
Figure BDA0002458058950000036
wherein alpha is i ，y (i) I=1, 2,3n are lagrange multipliers and data classification labels, respectively, and n is the number of samples;
further, obtain
Figure BDA0002458058950000037
Wherein x is (s) Is Lagrangian multiplier alpha i The eigenvectors of the samples corresponding to item not equal to 0, i.e. support vectors, y (s) Is a corresponding label;
(3.5) w is calculated 0 And b 0 Substituting the model expression of the nonlinear SVM classifier to obtain the following components:
Figure BDA0002458058950000041
in the method, in the process of the invention,
Figure BDA0002458058950000042
as a kernel function, taking a Gaussian kernel function as a kernel function
Figure BDA0002458058950000043
Sigma is a kernel parameter;
(3.6) inputting sample points of training samples labeled as k and l into MATLABIs selected for cross-validation and the fold k is set, i.e. the n samples of the matrix Z are taken randomly
Figure BDA0002458058950000044
k=2, 3, 4..machine learning was performed as a training sample, left->
Figure BDA0002458058950000045
The learned model is tested by the samples, a Gaussian nonlinear SVM classifier model is selected, and parameters C and +.>
Figure BDA0002458058950000046
Obtaining 6 groups of training sets, and respectively performing machine learning on each group of training sets;
(3.7) presetting a large enough times, changing the parameter C,
Figure BDA0002458058950000047
Returning to repeat the step (3.6), and recording the average accuracy rate until the preset times are reached; />
And (3.8) comparing the average accuracy corresponding to all the parameters, finding out the super parameter C and the nuclear parameter sigma corresponding to the maximum accuracy as optimal parameters, and training the obtained model under the parameter setting to serve as an optimal classification model.
As a further refinement of the present invention, step (4) specifically includes:
(4.1) extracting a feature vector from the carrier-to-noise ratio of the observation earth satellite to be determined, the positioning errors of three directions of the observation earth, the geometric accuracy GDOP factor and the number of satellite lock loss, wherein the feature vector is expressed as X= (X) (1) ,x (2) ,...,x (N) ) N is the total number of samples to be detected;
(4.2) inputting X into the 6 classification models, wherein X is calculated for each feature vector in each model t ) And outputting a judging result p (p=1, 2,3 and 4) corresponding to different solar storm intensity type labels, and after the 6 SVM models are classified, obtaining the type corresponding to the largest label occurrence number, namely the solar storm intensity type at the moment in the region.
Compared with the prior art, the invention has the remarkable advantages that:
the invention provides a solar radiation storm intensity judging method based on an SVM multi-classification algorithm. The method comprises the steps of firstly extracting and observing the ground satellite carrier-to-noise ratio, three-direction positioning errors, geometric accuracy (GDOP) factors and satellite unlocking number to perform data preprocessing, and obtaining a feature vector capable of effectively reflecting the intensity of solar radiation storm. Based on the SVM itself as a binary classifier, a one-to-one method is adopted for different solar radio storm intensity types, and four intensity types are respectively marked as 1,2,3 and 4, and form a sample with the feature vector. And combining samples with labels of 1,2,3 and 4 in pairs, and training an SVM classification model for each combination to obtain a corresponding optimal classification model, wherein the total number of the samples is 6. And for the data of the observation site to be judged, extracting the characteristic vector of the data, inputting the extracted characteristic vector into each group of optimal classification models for classification, outputting a label corresponding to the characteristic vector by each group of models, and taking a voting form, wherein the type with the largest appearance frequency of the label is the corresponding intensity type. Traditional solar radio storms require manual sorting by astronomists with specialized knowledge, which is time-consuming and laborious. The method has the advantages of low cost and relatively enhanced accuracy, can process a large amount of data at the same time, improves the detection efficiency, and has great significance for maintaining the normal operation of the satellite navigation system.
Drawings
FIG. 1 is a schematic flow diagram of one embodiment of the present invention.
Detailed Description
The invention is described in further detail below with reference to the attached drawings and detailed description:
the invention provides a solar power storm intensity judging method based on an SVM multi-classification algorithm. The method has feasibility for detecting the solar storm, and the detection result not only can display the occurrence time of the solar storm of a single station, but also can simultaneously give out whether a plurality of stations are influenced by the solar storm. Compared with the traditional method, the method has low cost, combines various factors, and relatively improves the identification accuracy and efficiency.
As a specific embodiment of the invention, the invention provides a solar radiation storm intensity judging method based on an SVM multi-classification algorithm, wherein a flow chart is shown in figure 1, and the specific steps are as follows;
comprising the following steps:
step one, data preprocessing is carried out, and feature vectors of different moments of an observation ground are calculated, wherein the feature vectors comprise an observation ground carrier-to-noise ratio reduction value, an observation ground total positioning error, a geometric precision GDOP factor and a satellite unlocking number.
The method specifically comprises the following steps:
(1.1) inputting the carrier-to-noise ratio of each satellite in the observation ground, the positioning error of the observation ground, the geometric accuracy GDOP factor and the number of satellite unlocking;
(1.2) data preprocessing is performed on the observed ground load-to-noise ratio:
taking the average value of the current carrier-to-noise ratio decline values of each satellite as the carrier-to-noise ratio decline value of the station, and marking as x 1 Units: dBHz;
snr i mean value of the measured satellite carrier-to-noise ratio at the first 8 moments-value of carrier-to-noise ratio at the moment
Figure BDA0002458058950000051
Wherein, snr i N is the number of satellites for each satellite carrier-to-noise ratio reduction value;
(1.3) the observed ground positioning error consists of the positioning errors of X, Y, Z in the northeast and north coordinate system, and the standard deviation is taken as the positioning error of the station in the three directions and is recorded as x 2 Units: m;
Figure BDA0002458058950000061
wherein r is i (i=1, 2, 3) represents the positioning error in the direction of the observation ground X, Y, Z at that time, respectively;
(1.4) GDOP factor is denoted as x 3 The number of satellite lock loss is recorded as x 4 Units: and each.
Examples: solar radiation storm occurs 12 and 13 in 2006, and KUMN is selected to be observed in 12 and 13 in 2006 2:00-4:00 The carrier-to-noise ratio of (UT) satellite, positioning error of three directions, geometric precision (GDOP) factor and number of out-of-lock satellites are taken as sampling intervals of 30s, and total time is 240. Taking three moments 2006/12/13 02:07:30, 2006/12/1302:46:00 and 2006/12/13 03:36:30 as examples, the other moments are calculated by the same method. Firstly, acquiring the obtained data from different KUMNs in the tables 2 and 3 according to the method of (1), and obtaining the carrier-to-noise ratio degradation value, the positioning error, the geometric accuracy (GDOP) factor and the satellite unlocking number of the KUMNs in the different times in the table 4.
TABLE 2 KUMN observed at different time instants and acquired carrier to noise ratio data
Figure BDA0002458058950000062
TABLE 3 acquisition of the obtained positioning error, GDOP factor, satellite Lock loss number data at different time instants of KUMN observed
Figure BDA0002458058950000071
TABLE 4 observed degradation of carrier to noise ratio for KUMN, positioning error, GDOP factor, number of satellite lock loss
Figure BDA0002458058950000072
Note that: when the station is out of lock, the station, the positioning error, GDOP factor, is defined as + -infinity.
And secondly, dividing the solar electric storm into 4 types of solar electric storm which do not occur, have weak intensity, strong intensity and strong intensity according to the change of the electric discharge quantity, and respectively assigning labels corresponding to the 4 types of feature vectors as 1,2,3 and 4.
The method specifically comprises the following steps:
reducing the observed ground load to noise ratio by x 1 Total positioning error x of observation ground 2 GDOP factor x 3 Number of satellite lock loss x 4 As the feature vector x, the sample feature vector is:
Figure BDA0002458058950000073
classifying the solar radio storm intensities according to the following table;
TABLE 1 solar storm intensity class classification
Figure BDA0002458058950000081
Where i represents the ith sample, x (i) A feature vector representing the i-th sample, y (i) A label representing the ith sample, (x) (i) ,y (i) ) I.e. the sample point.
Examples: in the above example, the sample points corresponding to the three moments are shown in the following table, and the sample points at other moments are obtained in the same way:
TABLE 5 sample points corresponding to KUMN observed at different times
Figure BDA0002458058950000082
And thirdly, combining each two types to construct a classifier to obtain 6 groups of training sets, constructing an unknown nonlinear SVM (support vector machine) classification model for each group of training sets, and carrying out cross validation on training samples to select optimal parameters to obtain 6 groups of trained classification models.
The method specifically comprises the following steps:
(3.1) training sets of 6 groups are training samples corresponding to labels 1 and 2, 1 and 3, 1 and 4, 2 and 3, 2 and 4 respectively;
(3.2) assuming a classifier (k, l=1, 2,3,4 and k < l) constructed for training samples with labels k and l, the nonlinear SVM classification model is specifically:
f(x)＝w T Φ(x)+b
where Φ (x) is a high-dimensional linear mapping, mapping function mapping 4-dimensional eigenvector x to M dimensions, M > 4,
Figure BDA0002458058950000083
both are parameters to be solved;
(3.3) solving the following optimization problem:
Figure BDA0002458058950000084
Figure BDA0002458058950000091
in xi i For the relaxation variable of each sample i, ζ i The larger the representative sample point is, the farther the representative sample point is from the group, C is a super parameter, and the larger C represents the more importance is placed on the objective function loss brought by the outlier, y (i) Label for sample i, x (i) A feature vector representing sample i;
(3.4) solving for the lagrangian multiplier:
Figure BDA0002458058950000092
Figure BDA0002458058950000093
wherein alpha is i ，y (i) I=1, 2,3n are lagrange multipliers and data classification labels, respectively, and n is the number of samples;
further, obtain
Figure BDA0002458058950000094
Wherein x is (s) Is Lagrangian multiplier alpha i The eigenvectors of the samples corresponding to item not equal to 0, i.e. support vectors, y (s) Is a corresponding label; />
(3.5) w is calculated 0 And b 0 Substituting the model expression of the nonlinear SVM classifier to obtain the following components:
Figure BDA0002458058950000095
in the method, in the process of the invention,
Figure BDA0002458058950000096
as a kernel function, taking a Gaussian kernel function as a kernel function
Figure BDA0002458058950000097
Sigma is a kernel parameter;
(3.6) inputting sample points of training samples corresponding to labels k and l into Classification Learner in MATLAB, selecting cross-validation and setting a fold number k, namely randomly taking n samples of a matrix Z
Figure BDA0002458058950000098
k=2, 3, 4..machine learning was performed as a training sample, left->
Figure BDA0002458058950000099
n samples are used for testing the learned model, a Gaussian nonlinear SVM classifier model is selected, and parameters C and +.>
Figure BDA00024580589500000910
Obtaining 6 groups of training sets, and respectively performing machine learning on each group of training sets;
(3.7) presetting a large enough times, changing the parameter C,
Figure BDA0002458058950000101
Returning to repeat the step (3.6), and recording the average accuracy rate until the preset times are reached;
and (3.8) comparing the average accuracy corresponding to all the parameters, finding out the super parameter C and the nuclear parameter sigma corresponding to the maximum accuracy as optimal parameters, and training the obtained model under the parameter setting to serve as an optimal classification model.
Examples: in the example above, the resulting 120 samples are input into matrix Z, cross-validation fold k=5 is set, parameter C is modified,
Figure BDA0002458058950000102
and obtaining the average accuracy of each group of models under different parameters.
The following table shows the optimal parameters of the group 6 classification models and the average accuracy corresponding to the optimal parameters, namely the classifier has the highest accuracy under the control of the parameters, and the obtained model is used as the optimal classification model.
Table 6 best parameters and average accuracy corresponding thereto
Figure BDA0002458058950000103
And step four, inputting the feature vector extracted from the satellite navigation signal to be identified into the classification model in the step 3, and outputting a result by each model, wherein if the judgment result is p, the p type counts one ticket, and finally the judgment result of the predicted sample is the type with the largest number of tickets.
The method specifically comprises the following steps:
(4.1) extracting a feature vector from the carrier-to-noise ratio of the observation earth satellite to be determined, the positioning errors of three directions of the observation earth, the geometric accuracy GDOP factor and the number of satellite lock loss, wherein the feature vector is expressed as X= (X) (1) ,x (2) ,...,x (N) ) N is the total number of samples to be detected;
(4.2) inputting X into the 6 classification models obtained. In each model, for each eigenvector x # t ) And outputting a judging result p (p=1, 2,3 and 4) corresponding to different solar storm intensity type labels, and after the 6 SVM models are classified, obtaining the type corresponding to the largest label occurrence number, namely the solar storm intensity type at the moment in the region.
The above description is only of the preferred embodiment of the present invention, and is not intended to limit the present invention in any other way, but is intended to cover any modifications or equivalent variations according to the technical spirit of the present invention, which fall within the scope of the present invention as defined by the appended claims.

Claims (4)

1. A solar radiation storm intensity judging method based on SVM multi-classification algorithm comprises the following specific steps:
(1) Carrying out data preprocessing, and calculating feature vectors at different moments of an observation ground, wherein the feature vectors comprise an observation ground carrier-to-noise ratio (carrier-to-noise ratio) decline value, an observation ground total positioning error, a geometric precision GDOP factor and a satellite unlocking number;
(2) Dividing the solar electric storm into 4 types of solar electric storm which do not occur, have weak intensity, strong intensity and strong intensity according to the change of the electric current flow, and respectively assigning labels corresponding to the 4 types of characteristic vectors as 1,2,3 and 4;
(3) Combining each two types to form a classifier to obtain 6 groups of training sets, constructing an unknown nonlinear SVM (support vector machine) classification model for each group of training sets, and performing cross validation on training samples to select optimal parameters to obtain 6 groups of trained classification models;
the step (3) specifically comprises:
(3.1) training sets of 6 groups are training samples corresponding to labels 1 and 2, 1 and 3, 1 and 4, 2 and 3, 2 and 4, and 3 and 4 respectively;
(3.2) assuming a classifier (k, l=1, 2,3,4 and k < l) constructed for training samples with labels k and l, the nonlinear SVM classification model is specifically:
f(x)＝w T Φ(x)+b
where Φ (x) is a high-dimensional linear mapping, mapping function mapping 4-dimensional eigenvector x to M dimensions, M > 4,
Figure QLYQS_1
both are parameters to be solved;
(3.3) solving the following optimization problem:
Figure QLYQS_2
Figure QLYQS_3
in xi i For the relaxation variable of each sample i, ζ i The larger the representative sample point is, the farther the representative sample point is from the group, C is a super parameter, and the larger C represents the more importance is placed on the objective function loss brought by the outlier, y (i) Label for sample i, x (i) A feature vector representing sample i;
(3.4) solving for the lagrangian multiplier:
Figure QLYQS_4
Figure QLYQS_5
wherein alpha is i ，y (i) I=1, 2,3 … n are lagrange multiplier and data classification label, respectively, n is the number of samples;
further, obtain
Figure QLYQS_6
b 0 ＝y (s) -w 0 T Φ(x (s) ) Wherein x is (s) Is Lagrangian multiplier alpha i The eigenvectors of the samples corresponding to item not equal to 0, i.e. support vectors, y (s) Is a corresponding label;
(3.5) w is calculated 0 And b 0 Substituting the model expression of the nonlinear SVM classifier to obtain the following components:
Figure QLYQS_7
wherein k (x (i) ,x)＝Φ T (x (i) ) Phi (x) is taken as a kernel function, and a Gaussian kernel function is taken as a kernel function
Figure QLYQS_8
Sigma is a kernel parameter;
(3.6) inputting the sample points of training samples labeled as k and l into Classification Learner in MATLAB, selecting cross-validation and setting the fold number k, i.e. randomly taking n samples of matrix Z
Figure QLYQS_9
k=2, 3, 4..machine learning was performed as a training sample, left->
Figure QLYQS_10
The learned model is tested by the samples, a Gaussian nonlinear SVM classifier model is selected, and parameters C and +.>
Figure QLYQS_11
Obtaining 6 groups of training sets, and respectively performing machine learning on each group of training sets;
(3.7) presetting a large enough times, changing the parameter C,
Figure QLYQS_12
Returning to repeat the step (3.6), and recording the average accuracy rate until the preset times are reached;
(3.8) comparing the average accuracy corresponding to all the parameters, finding out the super parameter C and the nuclear parameter sigma corresponding to the maximum accuracy as optimal parameters, and training the obtained model under the parameter setting to serve as an optimal classification model;
(4) And (3) inputting the feature vectors extracted from the satellite navigation signals to be identified into the classification models in the step (3), and outputting a result by each model, wherein if the judgment result is p, the p type counts one ticket, and finally the judgment result of the predicted sample is the type with the largest number of tickets.
2. The solar radiation storm intensity judging method based on the SVM multi-classification algorithm according to claim 1, wherein the method comprises the following steps: the step (1) specifically comprises:
(1.1) inputting the carrier-to-noise ratio of each satellite in the observation ground, the positioning error of the observation ground, the geometric accuracy GDOP factor and the number of satellite unlocking;
(1.2) data preprocessing is performed on the observed ground load-to-noise ratio:
taking the average value of the current carrier-to-noise ratio decline values of each satellite as the station carrier-to-noise ratio decline value, and marking as x 1 Units: dBHz;
snr i mean value of the measured satellite carrier-to-noise ratio at the first 8 moments-value of carrier-to-noise ratio at the moment
Figure QLYQS_13
Wherein, snr i N is the number of satellites for each satellite carrier-to-noise ratio reduction value;
(1.3) the observed ground positioning error consists of the positioning errors of X, Y, Z in the northeast and north coordinate system, and the standard deviation is taken as the positioning error of the station in the three directions and is recorded as x 2 Units: m;
Figure QLYQS_14
wherein r is i (i=1, 2, 3) represents the positioning error in the direction of the observation ground X, Y, Z at that time, respectively;
(1.4) GDOP factor is denoted as x 3 The number of satellite lock loss is recorded as x 4 Units: and each.
3. The solar radiation storm intensity judging method based on the SVM multi-classification algorithm according to claim 1, wherein the method comprises the following steps: the step (2) specifically comprises:
reducing the observed ground load to noise ratio by x 1 Total positioning error x of observation ground 2 GDOP factor x 3 Number of satellite lock loss x 4 As the feature vector x, the sample feature vector is:
Figure QLYQS_15
classifying the solar radio storm intensities according to the following table;
TABLE 1 solar storm intensity class classification
Figure QLYQS_16
Where i represents the ith sample, x (i) A feature vector representing the i-th sample, y (i) A label representing the ith sample, (x) (i) ,y (i) ) I.e. the sample point.
4. The solar radiation storm intensity judging method based on the SVM multi-classification algorithm according to claim 1, wherein the method comprises the following steps: the step (4) specifically comprises:
(4.1) extracting a feature vector from the carrier-to-noise ratio of the observation earth satellite to be determined, the positioning errors of three directions of the observation earth, the geometric accuracy GDOP factor and the number of satellite lock loss, wherein the feature vector is expressed as X= (X) (1) ,x (2) ,...,x (N) ) N is the total number of samples to be detected;
(4.2) inputting X into 6 classification models, in each model, for each feature vector X (t) And outputting a judging result p, p= (1, 2,3, 4), and corresponding to different solar storm intensity type labels, wherein after the 6 SVM models are all classified, the type corresponding to the largest label occurrence number is the solar storm intensity type corresponding to the region corresponding to the moment.
CN202010311617.0A 2020-04-20 2020-04-20 Solar radiation storm intensity judging method based on SVM multi-classification algorithm Active CN111553393B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010311617.0A CN111553393B (en) 2020-04-20 2020-04-20 Solar radiation storm intensity judging method based on SVM multi-classification algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010311617.0A CN111553393B (en) 2020-04-20 2020-04-20 Solar radiation storm intensity judging method based on SVM multi-classification algorithm

Publications (2)

Publication Number Publication Date
CN111553393A CN111553393A (en) 2020-08-18
CN111553393B true CN111553393B (en) 2023-05-19

Family

ID=72002426

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010311617.0A Active CN111553393B (en) 2020-04-20 2020-04-20 Solar radiation storm intensity judging method based on SVM multi-classification algorithm

Country Status (1)

Country Link
CN (1) CN111553393B (en)

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108802502B (en) * 2018-07-24 2020-11-10 山东大学 Solar radio flow monitoring system and method based on radio technology
CN110389361B (en) * 2019-06-24 2021-04-06 东南大学 Solar radio storm detection method based on GPS/BDS carrier-to-noise ratio reduction

Also Published As

Publication number Publication date
CN111553393A (en) 2020-08-18

Similar Documents

Publication Publication Date Title
CN113156391B (en) Radar signal multi-dimensional feature intelligent sorting method
CN103889051B (en) Indoor WLAN fingerprint positioning methods based on AP ID filterings and Kalman filtering
CN108055094B (en) Unmanned aerial vehicle manipulator frequency spectrum feature identification and positioning method
CN108830312B (en) Integrated learning method based on sample adaptive expansion
Wu et al. Superpixel-based unsupervised change detection using multi-dimensional change vector analysis and SVM-based classification
CN109508733A (en) A kind of method for detecting abnormality based on distribution probability measuring similarity
CN105225227A (en) The method and system that remote sensing image change detects
CN110598730A (en) Flight path association algorithm based on decision tree
CN110414566B (en) Land coverage type change detection method based on time sequence PolSAR image
CN107390259A (en) A kind of nuclide identification method based on SVD and SVM
CN106933977A (en) It is a kind of that the method that flight parameter outlier is rejected in classification is excavated based on big data
CN105472733A (en) Indoor positioning method based on AP selection
CN104408472A (en) Wishart and SVM (support vector machine)-based polarimetric SAR (synthetic aperture radar) image classification method
CN112149502A (en) Unfavorable geology positioning forecasting method based on convolutional neural network
CN103593852A (en) Hyperspectral image abnormality detection method based on homogeneous patches
CN109450573A (en) A kind of frequency spectrum sensing method based on deep neural network
CN117725448A (en) Cluster analysis method for meteorological navigation signal characteristics
CN104198998B (en) Clustering treatment based CFAR (Constant False Alarm Rate) detection method under non-uniform background
CN111553393B (en) Solar radiation storm intensity judging method based on SVM multi-classification algorithm
CN111428794B (en) Solar radio storm detection method based on support vector machine model
CN105354845A (en) Method for semi-supervised detection on changes in remote sensing images
CN107895361A (en) A kind of hyperspectral abnormity detection method based on local density&#39;s purifying background
CN109100759B (en) Ionosphere amplitude flicker detection method based on machine learning
Hu et al. Statistical analysis of massive AIS trajectories using Gaussian mixture models
CN114897010A (en) Heavy ionospheric scintillation detection method based on resampling improvement for gradient lifting tree

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