Awards & Nominations
Artificial ASME has received the following awards and nominations. Way to go!
Global Finalists Honorable Mentions
Using RNN as classifier and forecasting models to predict solar storms
High-Level Project Summary
We in “Save Earth From Another Carrington Event” have a challenge to predict the power of next geomagnetic event before it happened, simply, after searching and to see what parameters can help us to predict this event, we found that magnetic field vector in 3d, and solar wind speed can be our target, we built forecasting RNN model to get today’s data, to try getting next data points in the future, then feed this data to a supervised RNN model with kp-index as label to predict if there is a storm or not, and the strength of it, and then make an API, to deploy this system of models to production, to let anyone to check this phenomena every date.
Link to Final Project
Link to Project "Demo"
Detailed Project Description
Our System of models relay on Wind’s mission datasets, mfi, to get the data of magnetic field, as geomagnetic storm affect magnetosphere, and change the magnetic field of the earth, we selected BGSM, and BGSE each field contains 3 components, with about 900000 data point in every day, first we decided to clean it from outliers and errors in sensors, then down sample our data, to just 70 point a day to make it easier to be classified, and then collected about 2 years of data with the same preprocesses, then concatenate it, to make a one csv files, our classifier model uses LSTM, with dense and dropout layer, and Kp-index as labels, this model comes with f1 score of 97%, means that it can get 97% of all data points, the second model is using lstm also like the first one, but at this time, we process data this time differently, we first make every row of this data from last 70 point, to predict the next point, and then enter a closed loop to repeat this process and predict the next points, we then feed this data to the first model, how we can use this model, we decided to deploy this model in website, this website based on Flask framework with python, this method will help anyone to know if any storm close, in the future we will use SWE dataset to know if storm comes days before, and instead of using just 2 years, we will use all the data from 1994 to 2022, we used in this project Python as the main language to develop models and backend of the website, with tensorflow, pandas, numpy, spacepy and Flask, and we used java script, html and css to build the frontend of the website.
Space Agency Data
We used Wind’s mission mfi dataset, as main dataset in this prototype, but we processed also swe dataset from wind’s mission also.
Hackathon Journey
This Space Apps Hackathon will be a great memory to us to remember in the future, it was fun to challenge your self and work as a team to discover a new fields in space, this is our first time to deal with this amount of row data from satellite, it is difficult but also fun, in the beginning of this challenge we decided to choose a challenge with AI application, as that’s our passion, so this is the main reason of choosing this challenge, we also concorded other challenges, but this challenge was different as it solves a main issue we face.
References
C. Möstl,R. L. Bailey,H. T. Rüdisser,U. V. Amerstorfer,T. Amerstorfer,A. J. Weiss,J. Hinterreiter,A. Windisch (2021, Nov 02). Machine learning for predicting the BZ magnetic ... - wiley online library. (n.d.). Retrieved from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2021SW002859
NASA. (n.d.). Index of /pub/Data/Wind/MFI/mfi_h2/2022. NASA. Retrieved from https://cdaweb.gsfc.nasa.gov/pub/data/wind/mfi/mfi_h2/2022/
Chakraborty, S., & Morley, S. K. (2020, July 30). Probabilistic prediction of geomagnetic storms and the KP index. Journal of Space Weather and Space Climate. Retrieved from https://www.swsc-journal.org/articles/swsc/full_html/2020/01/swsc190086/swsc190086.html
Space weather archive | spaceweatherlive.com. Space weather live. (n.d.). Retrieved from https://www.spaceweatherlive.com/en/archive.html
Tags
#Ai,#RNN, #forecasting,