CN105372723A - Solar flare forecasting method based on convolutional neural network model - Google Patents
Solar flare forecasting method based on convolutional neural network model Download PDFInfo
- Publication number
- CN105372723A CN105372723A CN201510727599.3A CN201510727599A CN105372723A CN 105372723 A CN105372723 A CN 105372723A CN 201510727599 A CN201510727599 A CN 201510727599A CN 105372723 A CN105372723 A CN 105372723A
- Authority
- CN
- China
- Prior art keywords
- solar
- flare
- sample
- forecast
- layer
- 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
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01W—METEOROLOGY
- G01W1/00—Meteorology
- G01W1/10—Devices for predicting weather conditions
Landscapes
- Environmental & Geological Engineering (AREA)
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Atmospheric Sciences (AREA)
- Biodiversity & Conservation Biology (AREA)
- Ecology (AREA)
- Environmental Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a solar flare forecasting method based on a convolutional neural network model, and the method comprises the steps: A, preparing observation raw data of an active region; B, building a depth forecasting model, extracting features from the observation data through employing a convolutional neural network, and forecasting whether the active region generates solar flare or not. The method can directly enable the observation raw data to serve as the input of the model, automatically extract a forecasting factor for the forecasting of the solar flare from the raw data through employing the strong learning capability of a depth neural network, builds a corresponding forecasting model, and achieves an ideal forecasting capability through the forecasting model.
Description
Technical field
The present invention relates to the investigative technique of solar activity, particularly relate to a kind of solar-flare forecast method based on convolutional neural networks model.
Background technology
The sun is the source of space weather, and violent solar activity may cause satellite failure, communication disruption, navigates malfunctioning, even power transmission network paralysis.The situation of following a period of time solar burst activity if can forecast with unerring accuracy, just can carry out protection and the process of disaster in time.
Solar flare is a kind of violent solar activity phenomenon, and the forecast of solar flare has longer investigation and application history.Basis is observed with sunspot, nineteen ninety McIntosh proposes the morphological classification (McIntosh somatotype) of sunspot in " McIntosh; P.S.1990; Sol.Phys.; 125; 251 ", based on the McIntosh somatotype of black mole, manually establishes the solar-flare forecast expert system (Theo) that comprises rule more than 500.Bornmann and Shaw in 1994 has added up the McIntsho somatotype of sunspot and the relation of solar flare in " Bornmann, P.L. & Shaw, D.1994; Sol.Phys.; 150,127 ", and establishes the regression model of McIntosh somatotype and solar flare.Li Rong in 2007 etc. use in " Li, R., Wang; H.-N., He, H.; Cui; Y.-M., & Du, Z.-L.2007; ChJAA; 7,441 " area of sunspot, magnetic classification, McIntosh somatotype, 10 centimetres of radio flows as mode input, utilize support vector machine method to set up solar-flare forecast model.Within 2009, Colak and Qahwaji utilizes image processing techniques and machine learning techniques to set up automatic sun activity prediction system (ASAP) in " Colak; T. & Qahwaji; R.2009; SpaceWeather; 7,06001 ", and this system detects sunspot and automatically to automatically identifying its McIntosh somatotype, on this basis, neural net method is utilized to set up solar-flare forecast model.Bloomfield in 2012 etc. use the McIntosh somatotype of black mole in " Bloomfield, D.S., Higgins, P.A.; McAteer, R.T.J., & Gallagher, P.T.2012; ApJL, 747, L41 ", utilize Poisson statistics technology to set up solar-flare forecast model.
Based on the solar flare event occurred, within 2005, Wheatland only uses the history observation data of solar flare itself in " Wheatland, M.S.2005; SpaceWeather; 3,07003 ", utilizes bayes method to establish solar-flare forecast model.
Be observed basis with sunshine signal magnetic field, within 2006, Cui Yan U.S. waits in " Cui, Y.M.; Li; R., Wang, H.N.; & He; H.2007, Sol.Phys., 242; 1 " in from longitudinal magnetic field maximum horizontal gradient, centerline length, isolated singularity number 3 physical quantitys of MDI longitudinal magnetic field extracting data behaviour area, and added up the relation between these physical quantity and solar flares.Based on end user's artificial neural networks method establishment solar-flare forecast model in " Wang, H.N., Cui, Y.M., Li, R., Zhang, L.Y., & Han, H.2008, Adv.SpaceRes., 42,1464 " such as these physical parameters king Huaning.Within 2007, Georgoulis and Rust defines effective connection parameter Beff of active region magnetic fields in " Georgoulis, M.K. & Rust, D.M.2007; ApJ; 661,109 ", and this parameter reflects behaviour area photosphere magnetic flux distributions and photosphere magnetic field link properties.Within 2007, Schrijver is in " Schrijver, C.J.2007, ApJ; 655; 117 " literary composition, and author finds that large solar flare and the behaviour area strong gradient neutral line is relevant, and Sources of flare energy is in the free magnetic energy entrained by the fibre structure appeared in one's mind (fibrils).Appear entrained current characteristics in order to portray magnetic fiber in one's mind by photosphere, author define near high-intensity magnetic field in 15Mm, the large gradient neutral line without symbol magnetic fluxes R value, and added up the relation between this physical quantity and solar flare.Within 2007, Leka and Barnes utilizes behaviour area photosphere vector magnetic chart to calculate a large amount of magnetic fields parameter (comprise magnetic dip, magnetic field levels gradient, longitudinal current density, be wound around parameter, current helicity, magnetic shear angle, free magnetic energy density etc.) in " Leka; K.D. & Barnes; G.2007; ApJ; 656; 1173 ", find in these parameters, photosphere magnetic free energy has the strongest prediction ability, but any one single photosphere magnetic field parameter is all not enough to judge whether behaviour area produces large solar flare.Barnes and Leka in 2008 in " Barnes; G. & Leka; K.D.2008, ApJ, 688; L107 " based on identical data set, modeling method and evaluation index, test the prediction ability of the physical parameter proposed in document " Georgoulis, M.K. & Rust, D.M.2007; ApJ; 661,109 " and document " Schrijver, C.J.2007; ApJ; 655,117 ", point out that these physical parameters are when for solar-flare forecast, do not have significant difference, and the prediction ability of single parameter is limited.Within 2010, Mason and Hoeksema utilizes MDI longitudinal magnetic field to observe in " Mason; J.P. & Hoeksema; J.T.2010; ApJ; 723; 634 ", calculate behaviour area without symbol total magnetic flux, main centerline length, distance is effectively split in behaviour area, the neutral line length of gradient weighting, and find that the neutral line length of gradient weighting has merged the neutral line information of reflect structure and the magnetic field gradient information of reflection shearing, a situation arises thus can to reflect behaviour area solar flare better.McAteer in 2010 etc. in " McAteer; R.T.J.; Gallagher, P.T., & Conlon; P.A.2010; AdSpR, 45,1067 " from active region magnetic fields power spectrum and many fractal characteristics two angles, portray magnetic field, active photospheric region complicacy, and utilize the generation of these Parameters Forecasting solar flares.Ahmed in 2013 etc. use sun monitoring district tracker (SMART) to extract active region magnetic fields characteristic in " Ahmed; O.W.; Qahwaji; R., & Colak, T.etal.2013; SoPh; 283,157 ", and utilize neural net method to set up solar-flare forecast model.Within 2015, Bobra and Couvidat utilizes the Vector Magnetic Field of SDO to observe in " BobraM.G.andCouvidatS.2015ApJ798135 ", is extracted the physical parameter of 25 reflection behaviour area characteristics, and utilizes support vector machine to set up solar-flare forecast model.
In sum, existing solar-flare forecast model modeling flow process as shown in Figure 1.Existing forecasting model needs the parameter first extracting behaviour area to portray the characteristic of behaviour area, using the input of the behaviour area parameter of extraction as forecasting model, and then provides forecast result.
But the physical mechanism occurred solar flare is not at present also very clear and definite, and the parameter of Tectonic activity region has certain difficulty artificially, and the behaviour area parameter extracted does not reach desirable prediction ability.
Summary of the invention
In view of this, fundamental purpose of the present invention is to provide a kind of solar-flare forecast method based on convolutional neural networks model, directly using the input of the raw data of observation as this model, utilize the learning ability that deep neural network is powerful, from raw data, automatically extract the predictor being used for solar-flare forecast, and set up corresponding forecasting model, thus it is utilized to reach desirable prediction ability.
For achieving the above object, technical scheme of the present invention is achieved in that
Based on a solar-flare forecast method for convolutional neural networks model, this forecasting procedure comprises:
The preparation process of A, behaviour area original observed data;
B, set up degree of depth forecasting model, adopt convolutional neural networks to extract feature from observation data, and forecast whether this behaviour area produces solar flare.
Wherein, the taking a step forward to comprise and extract the parameter of behaviour area to portray the step of the characteristic of behaviour area of steps A.
Described steps A specifically comprises:
The raw data of A1, acquisition active region, i.e. SOHO/MDI full-time longitudinal magnetic chart;
The step of A2, acquisition solar flare sample;
A3, determine the step of solar flare intensity;
A4, described data are divided into training data and test data according to the time, and the observation data of behaviour area is converted into the image of value formed objects between 0 ~ 1;
Solar flare samples all in A5, reservation training data, and the non-solar flare sample that Stochastic choice is identical with solar flare sample size from non-solar flare sample, form the training dataset of two new classes balances.
Describedly determine solar flare intensity, be specially:
A31, solar flare intensity are determined by the weighted sum of the solar flare occurred in fixed time section, and its expression formula is:
I
tot=∑c+10∑m+100∑x
Wherein: c, m and x represent C level respectively, the coefficient of M level and X level solar flare.
Convolutional neural networks described in step B forms by 6 layers, is specially:
1st layer is input layer, and input layer is the photosphere magnetic field observation data of 100 × 100;
2nd layer is convolutional layer, and convolutional layer comprises 100 wave filters altogether, and filter size is 7, and step-length is 5; The output of convolutional layer is the characteristic pattern of 100 group 19 × 19;
3rd layer is pond layer, and pond layer wave filter is of a size of 3, and step-length is 2, and pond method is the maximal value of getting in wave filter; The output of pond layer is the figure of 100 group 9 × 9;
4th layer is the first full context layer, and interstitial content is 200;
5th layer is the second full context layer, and interstitial content is 20;
6th layer is output layer, and interstitial content is 2, respectively two kinds of output states of corresponding model, and described two kinds of output states are: future is by generation solar flare and do not produce solar flare;
In above-mentioned model training process, learning rate is set to 0.01, and momentum is set to 0.9, and largest loop number is set to 45000.
Describedly from observation data, extract feature, and forecast whether this behaviour area produces solar flare, is specially:
Whether described behaviour area produces the forecast of the solar flare being greater than certain threshold value, is a typical two-value forecasting problem, and for a two-value forecast system, its forecast result is following four kinds of possible results:
Itself be that the sample that positive class is almost always correctly predicted again as positive class is called as correct affirmative; Itself be that the sample that negative class is almost always correctly predicted again as negative class is called as correct negative; It itself is positive class is called as again mistake negative by the sample being predicted as negative class mistakenly; It itself is negative class is called as again mistake affirmative by the sample being predicted as positive class mistakenly.
In solar-flare forecast, using solar flare sample as positive class sample, non-solar flare sample is as negative class sample; Four classes of result export according to weather report, are defined as follows the performance that four indexs portray forecasting model:
Wherein: N
tPfor correct affirmative sample number, N
fNfor the negative sample number of mistake;
Wherein: N
tNfor correct negative sample number, N
fPfor the affirmative sample number of mistake;
TSS=TPrate-FPrate
Wherein: FPrate=1-TNrate.
Wherein:
N=N
TP+N
TP+N
TP+N
TP,
Described TPrate and TNrate is respectively used to evaluate the order of accuarcy of Forecast of Solar Flares and the order of accuarcy of non-Forecast of Solar Flares; The ratio of described index TSS to solar flare sample number and non-solar flare sample number is insensitive; Described HSS is for reflecting that the prediction ability of forecasting model compares the added value of random guess.
After described step B, comprise further:
The step of C, evaluation forecasting model.
Solar-flare forecast method based on convolutional neural networks provided by the present invention, has the following advantages:
The present invention is by utilizing newly-established solar-flare forecast model, no longer need the physical parameter of artificial extraction behaviour area, but directly from the feature representation of raw data learning behaviour area, greatly reduce the impact of human factor on forecast, reduce the application difficulty of forecasting model, improve the range of application of forecasting model.
Accompanying drawing explanation
Fig. 1 is existing solar-flare forecast model modeling flow process;
Fig. 2 is the model modeling schematic flow sheet of the solar-flare forecast method that the present invention is based on convolutional neural networks model;
The convolutional neural networks schematic diagram of Fig. 3 for using in the embodiment of the present invention;
Fig. 4 is solar-flare forecast model performance test result schematic diagram of the present invention.
Embodiment
Below in conjunction with accompanying drawing and embodiments of the invention, the solar-flare forecast method based on convolutional neural networks of the present invention is described in further detail.
Fig. 2 is the model modeling schematic flow sheet of the solar-flare forecast method that the present invention is based on convolutional neural networks model.
As shown in Figure 2, the present invention is based on the solar-flare forecast method of convolutional neural networks model, directly using the input of the raw data of observation as model, utilize the learning ability that deep neural network is powerful, from raw data, automatically extract the predictor being used for solar-flare forecast, and set up corresponding forecasting model.New solar-flare forecast model no longer needs the physical parameter of artificial extraction behaviour area, but directly from the feature representation of raw data learning behaviour area, like this, greatly can reduce the impact of human factor on forecast, and reduce the application difficulty of forecasting model, improve the range of application of forecasting model.
The present invention utilizes convolutional neural networks automatically to extract feature from observation data, and gives the result whether breaking out solar flare activity in prediction this behaviour area in 48 hours futures.The present invention utilizes the generation of active region Magnetic Field forecast solar flare.
Below in conjunction with Fig. 1 ~ Fig. 4, the solar-flare forecast method based on convolutional neural networks model of the present invention is described in detail.The method comprises the steps:
Step 10: extract the parameter of behaviour area to portray the step of the characteristic of behaviour area.This step is known technology, repeats no more here.
Step 11: the preparation process of behaviour area original observed data.
In an embodiment of the present invention, the raw data of active region is SOHO/MDI full-time longitudinal magnetic chart of 96 minutes from temporal resolution.Described SOHO/MDI full-time longitudinal magnetic chart can from ftp: //soi-ftp.stanford.edu/pub/magnetograms/ downloads and obtains.
The information of totally 1055 the NOAA behaviour areas within the scope of 30 °, day face is appeared at during present invention uses in May, 1996 in June, 2007.For the behaviour area within the scope of day 30 °, face, we ignore its projection effect.
Obtain the step of solar flare sample.Solar flare sample in the embodiment of the present invention is from U.S.'s day ground typical data center NationalGeophysicalDataCenter (NGDC) ftp: //ftp.ngdc.noaa.gov/STP/SOLAR_DATA/SOLAR_FLARES/FLARES_XRA Y/ obtains.
Determine solar flare intensity.Described solar flare intensity is determined by the weighted sum of the solar flare occurred in fixed time section.Its expression formula is:
I
tot=∑c+10∑m+100∑x
Wherein: c, m and x represent C level respectively, the coefficient of M level and X level solar flare.
Such as, in section, 3 solar flares (C4, M3 and X2) occur in preset time, the total intensity of solar flare is then 234 (i.e. 4+10 × 3+100 × 2).
In given forecasting period, if the total intensity of solar flare is greater than given threshold value, just think that this sample is solar flare sample, otherwise this sample is considered to be non-solar flare sample.In the present embodiment, forecasting period is 48 hours, and the threshold value of Itot is set to 10, is namely equivalent to generation M1.0 level solar flare.
In order to evaluate the performance of forecasting model objectively, be divided into training data and test data from all data 1996 to 2007 according to the time.Data between 1996 to 2006 are training data, and the data of 2007 are as test data.Training data is used for the foundation of forecasting model, and test data is used for the evaluation of forecasting model performance.
In order to adapt to the requirement of modeling method, the observation data of all behaviour areas is all converted into the image of value formed objects (100 × 100 pixel) between 0 ~ 1.
Because the quantity of solar flare sample non-in training sample is far more than the quantity of solar flare sample, if directly utilize raw data to train forecasting model, model can be partial to the many classifications of sample size usually.Therefore, the present invention retains solar flare samples all in training data, and the non-solar flare sample that Stochastic choice is identical with solar flare sample size from non-solar flare sample, together form the training dataset of two new class balances.Test data remains unchanged.
Step 12: the step setting up degree of depth forecasting model.
The present invention adopts convolutional neural networks to extract feature from observation data, and forecasts whether this behaviour area produces solar flare.The convolutional neural networks schematic diagram that Fig. 3 uses for the embodiment of the present invention.As shown in Figure 3, described convolutional neural networks forms by 6 layers, is specially:
1st layer is input layer, and input layer is the photosphere magnetic field observation data of 100 × 100.
2nd layer is convolutional layer, and convolutional layer comprises 100 wave filters altogether, and filter size is 7, and step-length is 5.Therefore, the output of convolutional layer is the characteristic pattern of 100 group 19 × 19.
3rd layer is pond layer, and pond layer wave filter is of a size of 3, and step-length is 2, and pond method is the maximal value of getting in wave filter.The output of pond layer is the figure of 100 group 9 × 9.
4th layer is the first full context layer, and interstitial content is 200.
5th layer is the second full context layer, and interstitial content is 20.
6th layer is output layer, and interstitial content is 2, respectively two kinds of output states of corresponding model.Namely following by generation solar flare with do not produce solar flare.
In above-mentioned model training process, learning rate is set to 0.01, and momentum is set to 0.9, and largest loop number is set to 45000.
The present invention provides the forecast whether behaviour area produces the solar flare being greater than certain threshold value, and this is a typical two-value forecasting problem.For a two-value forecast system, there are four kinds of possible results in its forecast result, as shown in table 1.
Itself be that the sample that positive class is almost always correctly predicted again as positive class is called as correct affirmative, itself be that the sample that negative class is almost always correctly predicted again as negative class is called as correct negative, itself being positive class is called as again mistake negative by the sample being predicted as negative class mistakenly, itself is negative class is called as again mistake affirmative by the sample being predicted as positive class mistakenly.
Table 1: four kinds of possible outcomes of two-value forecast result
The positive class of prediction | The negative class of prediction | |
Real positive class | Correct affirmative (TP) | The negative (FN) of mistake |
Real negative class | The affirmative (FP) of mistake | Correct negative (TN) |
In solar-flare forecast, we regard solar flare sample as positive class sample, and negative class sample regarded as by non-solar flare sample.Export based on four classes shown in table 1, be defined as follows the performance that four indexs portray forecasting model:
Wherein: N
tPfor correct affirmative sample number, N
fNfor the negative sample number of mistake.
Wherein: N
tNfor correct negative sample number, N
fPfor the affirmative sample number of mistake.
TSS=TPrate-FPrate
Wherein: FPrate=1-TNrate.
Wherein: N=N
tP+ N
tP+ N
tP+ N
tP,
In above-mentioned four evaluation indexes, TPrate and TNrate is respectively used to evaluate the order of accuarcy of Forecast of Solar Flares and the order of accuarcy of non-Forecast of Solar Flares.In order to the evaluation that whole forecasting model one can be given overall, we also need to use trueskillstatistic (TSS) and these two evaluation indexes of Heidkeskillscore (HSS).The ratio of TSS to solar flare sample number and non-solar flare sample number is insensitive, and the prediction ability that HSS reflects forecasting model compares the added value of random guess.
Step 13: the step evaluating forecasting model.
Embodiments of the invention using the behaviour area of 2007 and solar flare data as test data.1172 solar flare samples and 8828 non-solar flare samples are comprised in test data.Convolutional neural networks in the present invention trains 45000 times, and every the performance of 1000 tests model, test result as shown in Figure 4.
As can be seen from Figure 4, in front 5000 training, forecasting model is not from data learning to useful information;
From the 6000th time, convolutional neural networks is from observation data learning to useful feature, and model starts to have prediction ability.
6000th time and 8000 solar-flare forecast performances are presented in table 3 and table 5, forecasting model is from the training process of 6000 times to 8000 times, give solar flare sample more to pay close attention to, meanwhile, the forecast accuracy of non-solar flare sample reduces, this needs us according to the needs of different task, selects the forecasting model meeting mission requirements.Model tends towards stability after training 10000 times, as shown in Figure 4.
Table 2: during the 6000th step, the result that Forecast of Solar Flares model exports
The positive class of prediction | The negative class of prediction | |
Real positive class | Correct affirmative (707) | The negative (465) of mistake |
Real negative class | The affirmative (1129) of mistake | Correct negative (7699) |
Table 3: during the 6000th step, Forecast of Solar Flares model performance test result
Performance Evaluating Indexes | Test result |
TP rate | 60% |
TN rate | 87% |
ACC | 84% |
HSS | 0.38 |
TSS | 0.48 |
Table 4: during the 8000th step, the result that Forecast of Solar Flares model exports
The positive class of prediction | The negative class of prediction | |
Real positive class | Correct affirmative (753) | The negative (419) of mistake |
Real negative class | The affirmative (1884) of mistake | Correct negative (6944) |
Table 5: during the 8000th step, Forecast of Solar Flares model performance test result
Performance Evaluating Indexes | Test result |
TP rate | 64% |
TN rate | 79% |
ACC | 77% |
HSS | 0.28 |
TSS | 0.43 |
The above, be only preferred embodiment of the present invention, be not intended to limit protection scope of the present invention.
Claims (8)
1. based on a solar-flare forecast method for convolutional neural networks model, it is characterized in that, this forecasting procedure comprises:
The preparation process of A, behaviour area original observed data;
B, set up degree of depth forecasting model, adopt convolutional neural networks to extract feature from observation data, and forecast whether this behaviour area produces solar flare.
2. the solar-flare forecast method based on convolutional neural networks model according to claim 1, is characterized in that, the taking a step forward of steps A comprise extract behaviour area parameter to portray the step of the characteristic of behaviour area.
3. the solar-flare forecast method based on convolutional neural networks model according to claim 1, it is characterized in that, described steps A specifically comprises:
The raw data of A1, acquisition active region, i.e. SOHO/MDI full-time longitudinal magnetic chart;
The step of A2, acquisition solar flare sample;
A3, determine the step of solar flare intensity;
A4, described data are divided into training data and test data according to the time, and the observation data of behaviour area is converted into the image of value formed objects between 0 ~ 1;
Solar flare samples all in A5, reservation training data, and the non-solar flare sample that Stochastic choice is identical with solar flare sample size from non-solar flare sample, form the training dataset of two new classes balances.
4. the solar-flare forecast method based on convolutional neural networks model according to claim 3, is characterized in that, describedly determines solar flare intensity, is specially:
A31, solar flare intensity are determined by the weighted sum of the solar flare occurred in fixed time section, and its expression formula is:
I
tot=∑c+10∑m+100∑x
Wherein: c, m and x represent C level respectively, the coefficient of M level and X level solar flare.
5. the solar-flare forecast method based on convolutional neural networks model according to claim 1, it is characterized in that, convolutional neural networks described in step B forms by 6 layers, is specially:
1st layer is input layer, and input layer is the photosphere magnetic field observation data of 100 × 100;
2nd layer is convolutional layer, and convolutional layer comprises 100 wave filters altogether, and filter size is 7, and step-length is 5; The output of convolutional layer is the characteristic pattern of 100 group 19 × 19;
3rd layer is pond layer, and pond layer wave filter is of a size of 3, and step-length is 2, and pond method is the maximal value of getting in wave filter; The output of pond layer is the figure of 100 group 9 × 9;
4th layer is the first full context layer, and interstitial content is 200;
5th layer is the second full context layer, and interstitial content is 20;
6th layer is output layer, and interstitial content is 2, respectively two kinds of output states of corresponding model, and described two kinds of output states are: future is by generation solar flare and do not produce solar flare;
In above-mentioned model training process, learning rate is set to 0.01, and momentum is set to 0.9, and largest loop number is set to 45000.
6. the solar-flare forecast method based on convolutional neural networks model according to claim 5, is characterized in that, describedly from observation data, extracts feature, and forecasts whether this behaviour area produces solar flare, is specially:
Whether described behaviour area produces the forecast of the solar flare being greater than certain threshold value, is a typical two-value forecasting problem, and for a two-value forecast system, its forecast result is following four kinds of possible results:
Itself be that the sample that positive class is almost always correctly predicted again as positive class is called as correct affirmative; Itself be that the sample that negative class is almost always correctly predicted again as negative class is called as correct negative; It itself is positive class is called as again mistake negative by the sample being predicted as negative class mistakenly; It itself is negative class is called as again mistake affirmative by the sample being predicted as positive class mistakenly.
7. the solar-flare forecast method based on convolutional neural networks model according to claim 6, is characterized in that, in solar-flare forecast, using solar flare sample as positive class sample, non-solar flare sample is as negative class sample; Four classes of result export according to weather report, are defined as follows the performance that four indexs portray forecasting model:
Wherein: N
tPfor correct affirmative sample number, N
fNfor the negative sample number of mistake;
Wherein: N
tNfor correct negative sample number, N
fPfor the affirmative sample number of mistake;
TSS=TPrate-FPrate
Wherein: FPrate=1-TNrate.
Wherein: N=N
tP+ N
tP+ N
tP+ N
tP,
Described TPrate and TNrate is respectively used to evaluate the order of accuarcy of Forecast of Solar Flares and the order of accuarcy of non-Forecast of Solar Flares; The ratio of described index TSS to solar flare sample number and non-solar flare sample number is insensitive; Described HSS is for reflecting that the prediction ability of forecasting model compares the added value of random guess.
8. the solar-flare forecast method based on convolutional neural networks model according to claim 1, is characterized in that, after described step B, comprise further:
The step of C, evaluation forecasting model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510727599.3A CN105372723B (en) | 2015-10-30 | 2015-10-30 | Solar-flare forecast method based on convolutional neural networks model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510727599.3A CN105372723B (en) | 2015-10-30 | 2015-10-30 | Solar-flare forecast method based on convolutional neural networks model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105372723A true CN105372723A (en) | 2016-03-02 |
CN105372723B CN105372723B (en) | 2018-03-02 |
Family
ID=55375069
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510727599.3A Active CN105372723B (en) | 2015-10-30 | 2015-10-30 | Solar-flare forecast method based on convolutional neural networks model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105372723B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109447937A (en) * | 2017-08-29 | 2019-03-08 | 中国移动通信有限公司研究院 | A kind of determination method and related device of image processing model |
CN111079608A (en) * | 2019-12-09 | 2020-04-28 | 中国科学院新疆天文台 | Quick radio storm real-time searching method |
CN111965733A (en) * | 2020-08-25 | 2020-11-20 | 哈尔滨工业大学 | Method for evaluating correlation between forecasting factor and solar flare occurrence |
CN112327389A (en) * | 2020-10-30 | 2021-02-05 | 中国科学院国家空间科学中心 | Full-day face flare forecasting method and system based on solar 10.7cm radio flow |
CN112558187A (en) * | 2020-12-17 | 2021-03-26 | 昆明理工大学 | Method for forecasting solar flare outbreak based on 3D convolutional neural network |
CN113052202A (en) * | 2021-01-29 | 2021-06-29 | 昆明理工大学 | Method for classifying sun black subgroup in full-sun image |
KR20210104209A (en) * | 2020-02-14 | 2021-08-25 | 경희대학교 산학협력단 | Forecast method of Major Solar X-ray Flare Flux Profiles Using a Novel Deep Learning Method |
CN113537460A (en) * | 2021-06-29 | 2021-10-22 | 哈尔滨工业大学 | Method for constructing multithreading neural network model suitable for flare prediction problem |
CN113610762A (en) * | 2021-07-07 | 2021-11-05 | 中国科学院国家空间科学中心 | Early warning method and early warning system for solar flare |
CN114912674A (en) * | 2022-05-05 | 2022-08-16 | 山东大学 | Solar radio burst forecasting method and system based on full-sun photosphere magnetic map |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080071136A1 (en) * | 2003-09-18 | 2008-03-20 | Takenaka Corporation | Method and Apparatus for Environmental Setting and Data for Environmental Setting |
CN101266302A (en) * | 2007-03-15 | 2008-09-17 | 中国科学院国家天文台 | Computer sun activity prediction system |
CN101430388A (en) * | 2008-11-14 | 2009-05-13 | 中国科学院国家天文台 | Solar active longitude zone prediction method |
CN104850836A (en) * | 2015-05-15 | 2015-08-19 | 浙江大学 | Automatic insect image identification method based on depth convolutional neural network |
-
2015
- 2015-10-30 CN CN201510727599.3A patent/CN105372723B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080071136A1 (en) * | 2003-09-18 | 2008-03-20 | Takenaka Corporation | Method and Apparatus for Environmental Setting and Data for Environmental Setting |
CN101266302A (en) * | 2007-03-15 | 2008-09-17 | 中国科学院国家天文台 | Computer sun activity prediction system |
CN101430388A (en) * | 2008-11-14 | 2009-05-13 | 中国科学院国家天文台 | Solar active longitude zone prediction method |
CN104850836A (en) * | 2015-05-15 | 2015-08-19 | 浙江大学 | Automatic insect image identification method based on depth convolutional neural network |
Non-Patent Citations (1)
Title |
---|
裴世鑫等: "基于RBF人工神经网络的X级以上太阳耀斑预报研究", 《西北师范大学学报》 * |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109447937B (en) * | 2017-08-29 | 2021-07-06 | 中国移动通信有限公司研究院 | Image processing model determining method and related device thereof |
CN109447937A (en) * | 2017-08-29 | 2019-03-08 | 中国移动通信有限公司研究院 | A kind of determination method and related device of image processing model |
CN111079608A (en) * | 2019-12-09 | 2020-04-28 | 中国科学院新疆天文台 | Quick radio storm real-time searching method |
KR102378281B1 (en) * | 2020-02-14 | 2022-03-25 | 경희대학교 산학협력단 | Forecast method of Major Solar X-ray Flare Flux Profiles Using a Novel Deep Learning Method |
KR20210104209A (en) * | 2020-02-14 | 2021-08-25 | 경희대학교 산학협력단 | Forecast method of Major Solar X-ray Flare Flux Profiles Using a Novel Deep Learning Method |
CN111965733A (en) * | 2020-08-25 | 2020-11-20 | 哈尔滨工业大学 | Method for evaluating correlation between forecasting factor and solar flare occurrence |
CN112327389A (en) * | 2020-10-30 | 2021-02-05 | 中国科学院国家空间科学中心 | Full-day face flare forecasting method and system based on solar 10.7cm radio flow |
CN112327389B (en) * | 2020-10-30 | 2022-07-26 | 中国科学院国家空间科学中心 | Full-sun face flare forecasting method and system based on solar 10.7cm radio flux |
CN112558187A (en) * | 2020-12-17 | 2021-03-26 | 昆明理工大学 | Method for forecasting solar flare outbreak based on 3D convolutional neural network |
CN112558187B (en) * | 2020-12-17 | 2022-08-12 | 昆明理工大学 | Method for forecasting solar flare outbreak based on 3D convolutional neural network |
CN113052202A (en) * | 2021-01-29 | 2021-06-29 | 昆明理工大学 | Method for classifying sun black subgroup in full-sun image |
CN113537460A (en) * | 2021-06-29 | 2021-10-22 | 哈尔滨工业大学 | Method for constructing multithreading neural network model suitable for flare prediction problem |
CN113610762A (en) * | 2021-07-07 | 2021-11-05 | 中国科学院国家空间科学中心 | Early warning method and early warning system for solar flare |
CN113610762B (en) * | 2021-07-07 | 2024-02-23 | 中国科学院国家空间科学中心 | Early warning method and early warning system for solar flare |
CN114912674A (en) * | 2022-05-05 | 2022-08-16 | 山东大学 | Solar radio burst forecasting method and system based on full-sun photosphere magnetic map |
Also Published As
Publication number | Publication date |
---|---|
CN105372723B (en) | 2018-03-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105372723A (en) | Solar flare forecasting method based on convolutional neural network model | |
Arena et al. | Wave climate analysis for the design of wave energy harvesters in the Mediterranean Sea | |
Özger et al. | Long lead time drought forecasting using a wavelet and fuzzy logic combination model: a case study in Texas | |
CN103903452A (en) | Traffic flow short time predicting method | |
CN102945223B (en) | Method for constructing joint probability distribution function of output of a plurality of wind power plants | |
CN102736101B (en) | A kind of Distributed seismic early warning information processing method and system | |
CN105403680B (en) | A kind of short-term method for early warning of red tide based on buoy online monitoring data and purposes | |
CN103323755A (en) | Method and system for recognition of GIS ultrahigh frequency partial discharge signal | |
CN102323972B (en) | Method for managing phased array radar resource | |
CN103516912B (en) | A kind of mobile terminal location electricity saving method | |
CN105005833A (en) | Method for evaluating influence of hydropower cascade development on watershed ecology safety | |
Huang et al. | Real‐time reservoir flood operation during typhoon attacks | |
Luu et al. | Analysing flood fatalities in Vietnam using national disaster database and tree-based methods | |
Zhu et al. | A decision tree model for meteorological disasters grade evaluation of flood | |
CN102684713B (en) | Emitting machine of electromagnetic detection instrument and emitting method thereof | |
Hall et al. | Understanding flood regime changes in Europe: a state of the art assessment | |
Kang et al. | Disaster vulnerability assessment in coastal areas of Korea | |
Samuels et al. | The sound of silence: Exploring how decreases in tweets contribute to local crisis identification | |
Glamore | A decision support tool for assessing the impact of boat wake waves on inland waterways | |
CN103634310A (en) | Ocean network security risk assessment system and method | |
CN105678460A (en) | Relevancy analysis method for lightning disaster, equipment overload and section off-limit | |
Qiu et al. | Study on rockburst intensity prediction based on efficacy coefficient method | |
Luan | Research on geological disaster forecast method based on PCA structure BP model | |
Wan et al. | Optimization of the lightning warning model for distribution network lines based on multiple meteorological factor thresholds | |
Barruol et al. | Monitoring The River Activity of Two Rivers During Cyclones in La Réunion Island from a Seismic Network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |