Adhitya Shreyas SP
Sun is an eventful place. There are some active regions on the sun's surface from where a large amount of magnetic tension is released in the form of Solar Flares. Such active regions are called Sunspots. They are measured in terms of Sunspot Number and Sunspot Area. Solar Flares originating from these Sunspots can be catastrophic events. Along with them, they are often accompanied by CMEs, a significant release of plasma, which is a hot mixture of charged particles from the Sun's photosphere. The Solar Flares and CMEs when released with great energy can disrupt the communication system by causing malfunctions in the satellites and causing fire accidents by shorting the electric grid lines across the globe. Hence, it is wise to devise a method to predict the intensity of such events to be safe and well prepared beforehand. This was done using Machine Learning techniques and Curve-Fitting methods. The aim was to predict the Solar Flare Index, which roughly equates to its energy, based on the Sunspot Area, Sunspot Number, and the Magnetic Flux Density. Data of Sunspot Number, Sunspot Area, Solar Flare Index, and Magnetic Flux Density were collected from various credible websites like NCEI, Soho SDO, etc. and were averaged monthly. Thereafter, the Curve-Fitting method in MATLAB was used to study the relationship between the Solar Flare Index and the other parameters. Regression models were then created and optimized with an Ensemble method, which involves combining the base models to give better results for prediction. Upon performing the above, the prediction model was obtained with an accuracy of 0.79.
Sun is extremely hot, a gaseous ball made up of electrons and ions, also known
as Plasma. It also has a highly active magnetic field. During the 11-year half
cycle, it reaches a period where the magnetic field reaches such high tension
and energy, that it bursts and releases huge energy and hot plasma into the
atmosphere around it. These eruptions of energy radiation are termed as "Solar
Flares" and often burst out from "Active regions" where the temperature is low,
and magnetic field density(MFD) is relatively high hence they appear as dark
spots and are called "Sunspots." Sunspots are measured in terms of "Sunspot
Number(SSN)" and "Sunspot Area(SSA)."
Solar Flares can be calamitous events. Highly energetic solar flares can disrupt
the communication system of the Earth, short electric grids of the world, and
cause fires around the globe.
The deep study of such events in the past motivated our team, to explore various
computational methods to predict Solar Activity, to alarm the earth of such an
event.
Execution :
We started with data collection from various credible sources like NCEI,
SOHO, JSOC, etc, and alongside averaged them out to get monthly data. We
studied various parameters and chose MFD, SSA, SSN based on the strength of
relation and data availability.
Once parameters were sorted, we moved on to study various measurement
parameters for Solar Activity and collectively decided to take up the Solar Flare
Index (SFI), which is a rough measurement of energy of Solar Flares. Next, we studied the relation between SFI and various parameters through computational
(Curve-fitting) methods, using Matlab. This was followed by a lot of
well-thought trials and errors in testing various machine learning models with
our data to get the best accuracy. After working out a couple of such models, the
LS Boost Ensemble Regression Model gave us the best accuracy.
Results :
Thus, we obtained a system to predict Energy of Solar Flares using an easily
available dataset of SSN and SSA with an accuracy of 79%. This model, in the
future, can be used to predict energetic flares.