High-Level Project Summary
Our project consists of an artificial intelligence system capable of predicting solar storms based on the data provided by the DSCOVR satellite, linked to an automatic alarm system to alert all energy, internet and GPS providers so as to predict solar storm consequences and elude damages.Its importance relies on the advantage of knowing when a solar storm will hit, and minimizing the consequences it would arrise on the era of information and communication. Not forseeing these effects would have an estimated cost of 41 billion dollars in damages (Just USA).
Link to Final Project
Link to Project "Demo"
Detailed Project Description
Our project
We deviced a System capable of predicting intense solar storms to avoid the consequences that such event would have in our modern connected society.
Consequences of a today Carrington event
In our modern world, which is so dependant on information, connections and systems, a Carrington event would impact in many aspects of our daily basis:
- Power grids would shut down massively
- Internet connection and optical fiber would be affected
- GPS systems would stop working
- Electrical components in general would be damaged
- Communications (HF, VHF & UHF frequencies) would be compromised
This damage would have an estimated cost of 41 billiion dollars per day (Just USA).
This is why it is so important for society to have a reliable system that could anticipate such disaster and minimizes damage.
The algorithm pipeline
The actual algorithm has "AD-HOC" functions ("AD-HOC" Wikipedia article in references), this nature is bad in many cases, so we replace it with an AI
Also we merge the satellite dataset with observatories datasets.
Our Data
The first thing we did was to take the datasets from the DSCOVR and Wind satellites.
The algorithm has been trained with the following input data:
- bz_gse: z component from the interplanetary magnetic field
- bt: interplanetary magnetic component (nT)
- density: the solar wind’s proton density
- speed: wind bulk speed
- Temperature: solar wind temperature
We dismiss bx, and by, because the most important information is in the z axis (at 90° with the sun).
Other variables are taken away to simplify the problem to allow us to solve the challenge in less than 48 hours.
These are the final variables we used from the datasets obtained from the satellites:
On the other hand we have the data gathered from different observatories around the world which give us information about the impact from the solar waves on the earth's electromagnetic field: Disturbance Storm-Time Index (DST):
The storm's intensity is classified by the DST as follows:
0 to -50 nT ---> Calm storm
-50 to -100 nT ---> Moderate storm
-100 to -250 nT ---> Intense storm
less than -250 nT ---> Super storm
Data Cleaning
We took care of all NaN data (which would interfere with the AI’s training) by filling those gaps with the interpolation of the adyacents values in the dataframe.
Finally, we realized there was a mismatch between the quantity of data that both, the satellites and the observatories, provided. As we can see on the upper images, the observatories messure the DST intex once per hour, while the satellites makes measurements every minute. To correct this, we shortened the longest dataset by calculating an average for every 60 rows.
In order to save time, we only chose 30000 (from the almost 140000) mesurements in the resulting dataset.
Our machine learning algorithm is given a collection of data from which it predicts a DST value (even before being measured by the observatories!). This way we could anticipate events of dangeous magnitude.
An example of some DST predictions made by the model is shown up next:
Alert system
The alerting system our project proposes consists of sending an email using a python script that alerts vital services providers and all providers whose activities may be interrupted by a solar storm, as soon as the AI detects a high intensity storm.
We give examples of regional providers like:
- EDEMSA (power supply providers)
- Cooperativa Eléctrica (power supply providers)
- ITC (Internet providers)
- Movistar (Internet providers)
- Google Maps (GPS provider)
Our system would advise these companies to undergo a course of action one hour before the actual solar storm hits the earth, thus creating the possibility of reaction and preparation, possibly saving lots of resources that would otherwise be destined to compensations and reparations.
Space Agency Data
- DSCOVR data - https://cdaweb.gsfc.nasa.gov/pub/data/dscovr/h0/mag/2022/
- WIND data - https://cdaweb.gsfc.nasa.gov/pub/data/wind/mfi/mfi_h2/2022/
Hackathon Journey
In the group, we had never made an AI, so we learn a lot about Machine learning, neural networks. Also the challenge had so many elements that it make us learn about the sun, the satellites, space fisics and many other things.
We choose this challenge because we wanted to learn IA and Data analisis and it was a very good ocasion to do it.
We want to thanks to Mars Society Argentina for organizing this event, and all the others teams who give a hand in projects parts.
References
- Carrington desaster consequences data - https://www.researchgate.net/publication/358282676_A_Future_Carrington_Event_Impact_on_International_Telecommunications
- Ad-Hoc Article - https://en.wikipedia.org/wiki/Ad_hoc
Tags
#IA #Data #DSCOVR #NASA #Sun #Astronaut