High-Level Project Summary
The main idea of the project was to improve the measurements from the DSCOVR satellite, which were experiencing noise, and get them as close as possible to the ones taken from the WIND. We noticed that trying to forecast such an event with the imprecise data from the first satellite would result in biased predictions about a potential solar storm.
Link to Final Project
Link to Project "Demo"
Detailed Project Description
The main idea of the project was to improve the measurements from the DSCOVR satellite, which were experiencing noise, and get them as close as possible to the ones taken from the WIND. We noticed that trying to forecast such an event with the imprecise data from the first satellite would result in biased predictions about a potential solar storm.
Our first approach was to do a mapping to correlate the measurements from DSCOVR and WIND with a DTW library, as it was suggested by the Notional DSCOVR Faraday Cup Instrument “Calibration” and Data Analysis Procedure. Then, the obtained graph would have shown this relation and from it we would have been able to determine the data needed to be compared to feed the neural network. Finally, the results given by the AI would have provided more precise data regarding the solar wind.
However, this method did not work as expected. The library added unnecessary complexity, which resulted in a more inefficient use of computational resources and time.
Based on the suggestions made by one of the instructors at our institution, we changed the main 'skeleton' of our algorithm to develop a simpler method of the application, with the objective of suppressing the mapping process of the previous algorithm.
In order to do this, we decided to train an AI model with the data provided by WIND satellite, to reproduce the conditions of solar wind. Afterwards, we used this model with the DSCOVR data (something similar to transfer learning). By employing this method, the AI should be able to figure out the difference in time shifts by itself.
The variables taken into account to calculate and compare the characteristics of the solar wind were 'Proton_Np_nonlinear' (density), 'Proton_V_nonlinear' (velocity), 'Proton_W_nonlinear' (temperature). Although the measurements with this metrics alone might not be as accurate, these were used with the purpose of simplifying the process
Space Agency Data
Data base for the WIND satellite
https://cdaweb.gsfc.nasa.gov/pub/data/wind/mfi/mfi_h2/2022/
https://cdaweb.gsfc.nasa.gov/pub/data/wind/swe/swe_h1/2022/
Data base for the DSCOVR satellite
Hackathon Journey
We are really grateful for having this opportunity. We started from scratch, not knowing much about the topic and we found our way to give our best effort and contribute to this amazing project. This was our first time, we didn't know each other. Despite the difficulty, the last two days were an unforgettable experience. Every one of us learned a lot of valuable things from working together and asking for more help. We are genuinely happy with our project and, hopefully, we’ll be able to participate in this event in the future again.
References
Libraries
When we tried to appply the DTW
https://dynamictimewarping.github.io/python/
A NN library from Tensorflow
AI framework used in the project
https://www.tensorflow.org/?hl=es-419
Machine Learning Library
https://scikit-learn.org/stable/
Data base for the WIND satellite
https://cdaweb.gsfc.nasa.gov/pub/data/wind/mfi/mfi_h2/2022/
https://cdaweb.gsfc.nasa.gov/pub/data/wind/swe/swe_h1/2022/
Data base for the DSCOVR satellite
Tags
#software #AI #solarWind #CarringtonEvent