HELIO SURFERS

https://colab.research.google.com/drive/1akakPyBC22ER-z6mxvs9sJliBtRi6lKC?usp=sharing

https://docs.google.com/presentation/d/1BpslM1hM4jpbcV5Y_Gc_V3z9kfya75kZ/edit?usp=sharing&ouid=107026586004043348716&rtpof=true&sd=true

A machine learning algorithm for analyzing and predicting Carrington event using DSCOVR's Faraday Cup data and generating an early warning if event occurs.

Our primary Exploratory Data Analysis involves:

1. Data collection: We collected the required data from various sources including NASA( NOAA), NGDC(NOAA), SWPC(NOAA) with Interplanetary Magnetic Field (Bx GSM, By GSM, Bz GSM), Solar Wind Data (Temperature, Proton Density and velocity) and SYM-H index in .cdf and then formatted it to .csv using cdf viewer and Origin 2022b.
2. Data Cleaning: The data we collected was then filtered out to only contain the required parameters with a time interval of 5 minutes so as to facilitate higher resolution in accordance with SYM-H index for the first set of dataset and Bx GSM, By GSM, Bz GSM etc. for the second set of dataset with the use of numpy and pandas libraries.
3. Data Analysis: The relationship between various parameters and its influence on potential geomagnetic storm occurrence was analyzed. The relationship between various parameters inducing potential solar storm can be found that:

• SYM-H ≤-50nT , MODERATE
• SYM-H ≤-100nT , INTENSE
•  SYM-H ≤-250nT ,SUPERSTROMS

MACHINE LEARNING MODEL:

• ARIMA model (Autoregressive Integrated Moving Average) is used which combines three methods including autoregressive, moving average and integration.
• We also used logistic regression and TensorFlow(for data normalization using max scalar) for supervised prediction.
• Using sklearn library we had split the dataset to training, testing and validation and created a model which is giving an accuracy > 80%. for prediction with limited dataset. 
• For live prediction we planned to collect the real-time data using web scrapping tools like selenium, beautiful soup and pandas and feed those values to the pretrained model that we have generated using the above-mentioned datasets. So, if any solar storm occurs, we can predict that 45 min prior the coronal mass radiations traverse the Bx GSM of earth and can generate a warning or an alert to different authorities from the model.
•  By utilizing more dataset if we can forecast the changes for the above parameters like before 10 days the warning system can be improved which can reduce the impact of the event.

OUTPUT:

• For user given inputs for the above mentioned parameters,UI with React and Flask API will collect the user input and using a GET request it will be given to the model and prediction is made and it will display the graph for the parameters Bt, Bx GSM, By GSM, Bz GSM, velocity, temperature and density against time along with the predicted value whether event will occur or not. This is provide an alert mechanism for the user.

1. Interplanetary Magnetic Field (IMF) data ( NASA): https://cdaweb.gsfc.nasa.gov/pub/data/dscovr/h0/mag/2022/
2. Longitudinally Symmetric Components (SYM-H) data: (WDC, JAXA) https://wdc.kugi.kyoto-u.ac.jp/index.html
3. Real time Solar Wind data (NOAA, NASA): https://www.swpc.noaa.gov/products/real-time-solar-wind
4. Space Weather Prediction data (NOAA, NASA):https://www.swpc.noaa.gov/products-and-data
5. Deep Space Climate Observatory data (DSCOVR, NASA):https://www.ngdc.noaa.gov/dscovr/next/
6. SDO AIA(ESA) : https://www.lmsal.com/get_aia_data/?&wavelengths=&startDate=1998-01-01&startTime=06:48:09&stopDate=1998-12-31&stopTime=06:48:09

 Being a novice in hackathon, our entire team was just wrapped up with hope and confidence to solve the challenge we have selected and we are so glad that we have produced a synergistic outcome. We, being a coterie of astrophiles, were inclined towards the Carrington event and were more involved and curious in making the best possible solution to address the challenge.

 We started our project by collecting the required data from various sources including NASA( NOAA), NGDC(NOAA), SWPC(NOAA) and processed it through a series of Semi automated Exploratory Data Analysis (EDA) with Python visual studio and SKLEARN as our tool.

We then analyzed the relationship between various parameters of our data to process with the algorithm. We started a trial and error test of logistic regression algorithm and finally processed with ARIMA model with logistic regression .

We felt satisfactory to reach our current level of solution set with our intermediate machine learning skills. We are still determined in boosting our ML skills and working further on our algorithms to develop a more precise model with real time supervision!

  The hackathon journey didn't just serve us as an opportunity, but also as a crash course for us to improve our problem solving skills, divergent thinking ability and also gave us a chance to find our response time to riddle out the challenges.

RESOURCES:

• https://www.swpc.noaa.gov/products-and-data
• https://www.ngdc.noaa.gov/dscovr/next/
• https://wdc.kugi.kyoto-u.ac.jp/index.html
• https://www.lmsal.com/get_aia_data/?&wavelengths=&startDate=1998-01-01&startTime=06:48:09&stopDate=1998-12-31&stopTime=06:48:09
• https://www.swpc.noaa.gov/products/real-time-solar-wind
• https://wdc.kugi.kyoto-u.ac.jp///aeasy///wwwtmp/WWW_aeasy03949594.dat
• https://www.swsc-journal.org/articles/swsc/full_html/2019/01/swsc180011/swsc180011.html
• https://services.swpc.noaa.gov/products/solar-wind/mag-7-day.json
• https://www.space.com/the-carrington-event
• https://www.nasa.gov/mission_pages/sunearth/spaceweather/index.html
• https://www.helioviewer.org/
• https://ngdc.noaa.gov/dscovr/portal/index.html#/

TOOLS USED:

• CDF viewer
• Helioviewer
• Python
• Visual Studio
• pandas
• sklearn
• matplotlib
• Origin data viewer

#helio_surfers ,#carrington_event,#software,#machine_learning,#NARX