High-Level Project Summary
The sun goes through many cycles, spewing tons of magnetic particles during its active phase. The Carrington event of 1859 is a daunting example of what these particles are capable of, shocking telegraph operators, and turning nights into days. Such an event in today's age of electricity would prove devastating with losses ranging up to trillions. The next Carrington size event is not a question of if, but a question of when. For that, we must be prepared. The solar translator aims to monitor the data obtained from the sun and extrapolate it to analyze what might lie in the future. This helps to anticipate a Carrington-like event in advance and take measures before it does its damage.
Link to Final Project
Link to Project "Demo"
Detailed Project Description
The Solar Translator
The solar translator provides tools to analyze the velocity of solar wind through a website. There are two Machine Learning models made available through the website. One does univariate forecasting on the speed of protons in the wind. Another model forecasts the proton velocity with the data of Magnetic Field density and previous proton velocity. The model consists of CNN, LSTM, and Dense layers with a window size of 24 hours.
These forecasts help to know about the developments in solar wind speed in advance which helps to predict potentially threatful solar flares. Predicting these flares in advance is the first step to minimizing the damage it causes to power grids and electrical equipment. If we were to face a Carrington-like event without any preparations, the losses would be astronomical, setting us back decades in terms of development.
So, our goal is to continuously monitor data from the sun to generate warnings about such an event.
The website provides continuous forecasting of solar wind speed based on recently available data. Furthermore, it also provides users the option to input data to generate predictions on it. This can be useful for researchers to analyze the sun under different scenarios just by entering the data.
Prediction of windspeed of next 24 hours obtained using the model using latest data from previous 24 hour
Prediction obtained from user's input
Tools Used: Machine Learning (CNN, LSTM), JavaScript
Coding Languages: Python, JavaScript
Software: Tensorflow, Keras, Numpy, Pandas, Sci-kit Learn
Space Agency Data
In 2012, a Carrington-sized solar flare missed earth approximately by the margin of 9 days. In the next Carrington-sized event the sun might mirror the activities from 2012. So, we decided to use the data from 2012. Since DSCOVR was launched in 2015 there is no data from 2012, we used the wind mission's data. We downloaded the WIND mission's magnetic field dataset and the WIND mission's ION parameters for the year 2012 using a python script. The amount of data was huge. So, we generated hourly average data from the original data so that only 24 data points were available for a day. For the velocity of the solar wind, we used the average velocity of the proton obtained from non-linear analysis and moment analysis. We trained the data for univariate time series forecasting of velocity. We used the data of Magnetic Field Density from the WIND mission's magnetic field dataset for magnetic field time series. We trained another model to forecast the velocity with magnetic field and previous values of velocity as features.
For forecasting, we used magnetic field data from both DISCOV and WIND missions.
Hackathon Journey
We learned a lot about dealing with CDF files, time series data, APIs, and many more. It was a fulfilling experience, We'd like to thank Mike from Smithsonian, Chaneil James, Maria Sha for helping us by answering our queries. The information we obtained from them was really valuable in completing this project.
References
- Wind mission's magnetic field dataset: https://cdaweb.gsfc.nasa.gov/pub/data/wind/mfi/mfi_h2/2022/
- DSCOVR magnetic field dataset: https://cdaweb.gsfc.nasa.gov/pub/data/dscovr/h0/mag/2022/
- Wind mission's ion parameters: https://cdaweb.gsfc.nasa.gov/pub/data/wind/swe/swe_h1/2022/
Tags
#machine-learning #carrrington-event