innovator

Add idea


Calendar

«    May 2019    »
MonTueWedThuFriSatSun
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
 

 

Advert

 

Payment

 

Advert

 

Authorization

Стартап

DateDate: 22-05-2019, 05:57

Forecasts come true with an accuracy of 75% and 92%
The machine learning system has studied thousands of recent solar flares and determined that their occurrence could have been predicted several days before emissions from a sharp increase in the strength of electric currents in the future zone of their origin. This was told by Mark Chung from Stanford University, PNAS reports.
American scientists have suggested that all flares are generated by similar processes in the depths of the sun.
For artificial intelligence training, specialists manually analyzed more than 1000 photos of active regions on the surface of the star that caused flashes in 2010–2016 and were studied using the GOES meteorological probe tools.
The machine learning system rather quickly learned to recognize the “quiet” regions of the Sun and the future centers of flare activity with an accuracy of 75% and 92%.
At the same time, differences between them occurred and persisted for 3 days before and after the birth of the release, which indicates an extremely “long-playing” character of such activity zones.
In their predictions, AI relied on how the electrical activity of the solar photosphere changed in those zones where flashes appeared.
The subsequent analysis of data from solar satellites will help scientists understand what exactly forces solar matter to “eject” from the surface of the star and how such cataclysms can be predicted with an accuracy of 100%.