High-Level Project Summary
We developed an ML-based and ANN model predict or estimate the characteristic values of solar wind. This model emulate the behavior of DISCOVR station to help it to still producing data for prevent a high solar torment event. This is very important because if the Carrington event could be occur again the damage over our planet could be very serious and danger.
Link to Final Project
Link to Project "Demo"
Detailed Project Description
In this work, we present a ML-based and ANN able to predict protons' velocity, density and Temperature, which is needed to estimate the main parameters of solar winds. It's input data are composed by the sensors' measurements of the DSCOVR spacecraft. In other words, we are comparing these two models, regression model and a Neural Network to achieve a parameters classification.
The most challenging part of this project was Data analysis. We have used two main tools: Python programming language, and Matlab (only to understand raw data). Our entire framework was written in Python (notebook and scripts) with the following hardware specs: Intel® Core™ i5-7 CPU @ 2.60GHz × 4 and 12GB of RAM.
After some tests, we have noticed that this method could play a crucial role in preventing Carrington-like events in future .
Space Agency Data
We use the NASA's datasets in order to understand the Ion Wind complex behavior. We tried to comprise the relation between the magnetic field measurement from Spacecraft and the magnetic field measurement station base. We were able to indentify the meaning of all input variables and the expected results. We noticed that one dataset had peaks with very big values and we tried to filter them. At the same time, we decided to use only few variables. That because we noticed some values (like sensor states) will not be significant for the training.
Hackathon Journey
The experience was enriching and challenging, because we had the opportunity to work with new people in a totally new case of studio that allows us to study and be creative to solve the problem. We learned many skills specially in manage of datasets and exploring machine learning models. We tried to solve the project using the most rellevant cuantity of data in order to obtain a simpler but functionall model. To solve setbacks we had to dabble in much online investigation and also the discord comunity was very usefull.
References
- https://cdaweb.gsfc.nasa.gov/pub/data/wind/mfi/mfi_h2/2022/
- https://cdaweb.gsfc.nasa.gov/pub/data/dscovr/h0/mag/2022/
- https://cdaweb.gsfc.nasa.gov/pub/data/wind/swe/swe_h1/2022/
- https://cdaweb.gsfc.nasa.gov/
- https://www.tensorflow.org/?gclid=CjwKCAjw7eSZBhB8EiwA60kCWxu5C_MiLfJKobtMP78Y46zUIU1Ea2atUAGIzPv8ssFQBbrrulDOTBoCPdkQAvD_BwE
- https://scikit-learn.org/stable/
- https://www.python.org/
- https://www.mathworks.com/products/matlab.html
Tags
Carrington event, TensorFlow, DSCOVR, Python, IA