High-Level Project Summary
As the Carrington event was a catastrophe that could ruin the modern life we are in, making a model that predicts its occurrence is a necessity.So We build a deep learning RNN model using Bi-LSTM that tracks the change in the solar wind speed and magnetic field and exceeds all the previous attempts to predict the hazardous Solar flares with minimal loss , this model gives a one hour margined prediction as a warning sign for all human lives, for those who lay ground at our earth, and especially the ones in the sky from passengers to astronauts.
Link to Final Project
Link to Project "Demo"
Detailed Project Description
We wrote a Full descriptive research paper in this link: https://docs.google.com/document/d/1NEzU6QZCGNU4D1_4BzBnOzeUYfo3UzeGwGRSGtpF8w4/edit#
Abstract
Solar flares are large eruptions of electromagnetic radiation from the Sun lasting from minutes to hours.
The sudden outburst of electromagnetic energy travels at the speed of light, therefore any effect upon the sunlit side of Earth’s exposed outer atmosphere occurs at the same time the event is observed. Solar flares occur when a large magnetic energy builds up in the solar atmosphere and is released suddenly. These outbursts are intrinsically linked to the solar cycle — an approximately 11-year cycle of solar activity driven by the sun's magnetic field.
These solar flares are accompanied by Coronal Mass Ejections (CMEs), they are large expulsions of plasma and magnetic fields from the Sun's corona. They can eject billions of tons of coronal material and carry an embedded magnetic field (frozen in flux) that is stronger than the background solar wind interplanetary magnetic field (IMF) strength.
Solar flares can be the reason for the end of all human technology for more than 10 years, 10 years of total blackout, with the potential of causing radiation hazards to astronauts.
So it is crucial to build a model that predicts solar flares and CMEs, and in our thesis we discuss the journey of avoiding a global catastrophe by a time-series data consisting of active solar region magnetic field parameters acquired from the NOAA observatory satellite DSCOVR. Our approach to predict the Carrington event is building a LSTM model that predicts the value of the DST(disturbance storm index) feature.
Introduction
Part of the Intelligent design of the universe, is the fine tuned parameters of it, a fine tuned universe suggests that the occurrence of life in the universe is very sensitive to the values of certain fundamental physical constants and that the observed values are, for some reason, improbable. Important example of it is the distance between our star; the sun and earth, approximately 149.82 million km.
The Sun's temperature is exactly 5,778 K (15 million degrees Celsius), and this fine-tuned sun has many changes, disturbances, and phenomena that keep occurring in and around the Sun’s atmosphere, generally referred to as Solar activity. It has several variables that fluctuate even on a fraction of a millisecond. This solar activity happens when the energy stored in 'twisted' magnetic fields (usually above sunspots) is suddenly released. These outbursts are called “Solar flares”. Solar flares are one of the solar activities of the sun.
There are four primary forms of solar activity: solar flares, coronal mass ejections, high-speed solar winds, and solar energetic particles. A powerful Class-X solar flare accompanied by a Coronal Mass Ejection (CME) – a cascade of highly energetic particles accelerated from the sun’s corona by magnetic field collapse – could cause catastrophic damage to Earth’s ground electronic and orbital satellite infrastructures.
These giant eruptions have the capability to wreak havoc on GPS and other satellites, airplane communications, power grids, copper wiring in transformers, and even hand-held modern devices like smartphones, and a Carrington event can occur.
What is the Carrington event?
The Carrington Event was a large solar storm that took place at the beginning of September 1859. World was in a contrasting situation where people in the northern countries were amazed by the beauty of the aurora lights(Northern lights) and across the earth there was a terrifying storm.
Northern lights: Is one visible effect of solar radiation. Earth’s magnetic field, which protects us from most of the dangers of space radiation, directs the charged particles to the poles, where they enter our atmosphere and cause beautiful light displays. in the sky, In August 1859, astronomers around the world watched with fascination as the number of sunspots on the solar disk grew. Among them was Richard Carrington, an amateur skywatcher in a small town called Redhill, near London in England. On Sep. 1, as Carrington was sketching the sunspots, he was blinded by a sudden flash of light. Carrington described it as a "white light flare" according to NASA spaceflight. The whole event lasted about five minutes.
Preprocessing:
- Feature selection
- Feature engineering
- Scaling
- Flooring
Modelling:
- We used a Bi-LSTM model along with Uni-LSTM with GRU model.
RESULTS AND DISCUSSION:
The BI-LSTM has a longer run time compared to the LSTM, but it provides more reliable results and lower loss. GRU on the other hand provides similar results to the BI-LSTM in terms of loss but the downside is that it has a much larger runtime and it is a very complex model.
Due to the reasons mentioned above we decided to go with the BI-LSTM model to predict the DST and give the warning of solar flares we are looking for. This warning is set before an hour of the solar flare activity, giving scientists time to prepare for a new Carrington event.
Space Agency Data
Introduction to the Dataset
The data is made up of solar wind measurements acquired by two satellites: NASA's Advanced Composition Explorer (ACE) and NOAA's Deep Space Climate Observatory (DSCOVR).
Datasets
Geomagnetic storms form Due to the transfer of the energy from solar wind to the earth’s magnetic field. The magnetic navigation errors are increased by the resulting magnetic field changes. The geomagnetic storm's intensity is measured by the disturbance-storm-time index, or DST.
1- Solar Wind Data
2- Satellite Data
3- Sunspots Data
4- Labels Data
Hackathon Journey
- It was a tremendous experience full of challenges and knowledge, we got to know many great acquaintances from the colleges and teammates.
- We got to learn many concepts about space and about NASA, along with the challenges in AI knowledge.
- We were interested in Carrington event as we read about it before.
- Reading the challenge and understanding its requirements and using NASA resources to understand our data along with building an AI model to solve it.
- We would like to thank NASA space apps Cairo for their great efforts.
References
Main resources:
https://ngdc.noaa.gov/dscovr/portal/index.html#/
https://www.ngdc.noaa.gov/geomag/data/geomag/magnet/
Rest of references:
https://www.geeksforgeeks.org/standardscaler-minmaxscaler-and-robustscaler-techniques-ml/
https://analyticsindiamag.com/lstm-vs-gru-in-recurrent-neural-network-a-comparative-study/
https://www.space.com/22215-solar-wind.html
https://www.exploratorium.edu/spaceweather/flares.html
http://www.pas.rochester.edu/~blackman/ast104/wind.html
https://www.esa.int/Science_Exploration/Space_Science/What_are_solar_flares
https://theconversation.com/why-we-need-to-get-better-at-predicting-space-weather-157630
https://www.space.com/the-carrington-event
Inspiration:
https://www.kaggle.com/code/arashnic/eda-prep-and-keras-lstm/notebook
Tags
#ML #DL #LSTM #RNN #data_Anylsis #AI

