High-Level Project Summary
The project includes software that takes in the live data stream from NOAA's DSCOVR and uses a deep neural network model which is trained using multiple factors like Solar Cycle trends and using plasma datasets and solar wind datasets to early predict such events. The project adds one more layer of innovation which is the addition of hardware (a Microcontroller) with a pub sub-model that connects with the internet and listens to a topic. Whenever such events occur it sends a message on a topic and all the devices automatically follow it. Our accuracy was about 97% (Validation ). We feel such events today will take us decades back. The hardware tinge is to make faster shutdown and evacuation
Link to Final Project
Link to Project "Demo"
Detailed Project Description
The software developed by us is supposed to be a background application as making a mobile application wouldn't interest nor benefit anyone as just reading the data as the data might be redundant and events like Carrington are quite rare. That's one of the reasons why we wanted to keep it that way. The application was running and fetching live data from DSCOVR and Deep Neural Network which was based on LSTM (Long, Short Term Memory) to predict regressive values of solar wind and magnetic strength in the direction of the earth. All the model was trained by keeping the values of the solar cycles into account. The validation accuracy of the model was found to be 97% and training accuracy 98% and the mean squared error was 1.1132e-4.
So, a program runs and automatically detects solar flare events using the DNN model and then alerts the authority and ISP providers as we provide an SDK for other companies and organizations that may want to install and use it for broadcasting in times of crisis. The broadcast-based evacuation from hospitals or universities where there might be the chance of short circuits may create havoc.
Then comes the hardware in the picture. The solar flare events will be known to us only before 1 hour to 45 minutes so the only way to prevent huge disasters and millions being at the risk of losing a life is to at least break the connection of the huge transformers and the power grid to prevent them from being damaged by the electromagnetic disturbances. The hardware uses the MQTT protocol which provides features such as one publish and many subscriptions and in a single go everybody receives the data. The entire system is supposed to be autonomous as soon as the application detects a threatening event. It will figure out places of impact and try to send those devices (surge protectors) a signal to break the connection. This will be extremely fast and administrations can spend that time handling and managing people. The IoT-based approach can be effective in dealing with autonomous connections and triggers.
We hope to further develop it to be much more reliable and secure. The other parameters could be explored more and we can be better at predicting and analyzing space weather.
We used Jupyter notebook and python for data visualizations and understanding of the spread of the data and then Keras and TensorFlow to build an LSTM model. We then used embedded C and Arduino for developing software for the IoT. For the MQTT protocol, we used Mosquitto broker which we hosted on a Virtual Machine. For hardware, we used an ESP8266 board and we intended to use relays for switching but unfortunately, the relay we had was a defective one so we need to come up with a conceptually similar system so we used an NPN transistor BC547B and a couple of resistors and a breadboard.
The software used for programming the logic and SDK involved Microsoft VS code and PyCharm.
Space Agency Data
The dataset used for the project are of the NOAA and NASA's DSCOVR and ACE programs
https://www.ngdc.noaa.gov/stp/solar/solarflares.html
https://services.swpc.noaa.gov/products/solar-wind/mag-1-day.json
https://services.swpc.noaa.gov/products/solar-wind/plasma-1-day.json
Hackathon Journey
It was our first offline hackathon as we are currently engineering undergraduates and we really enjoyed it. The experience was amazing we explored a lot of things about the space, our skills, and working together as a team. The ideation phase of the hackathon urged us to push our boundaries and we read a lot about heliophysics and solar cycles. But to be honest it was tiring and demanding as well but we did our best and stayed on our toes through the entire competition will very little sleep. The feeling to forget and just work on one problem proved to be a lot more effective than working on various problems while swimming through life.
It was really a dream for us to someday solve the challenges that space offers and during this hackathon we got a chance to fulfill that childhood desire. We are quite content with our efforts and understanding of the entire topic but we will surely like to explore it further on our own. We even took away all the notes that we wrote during those hours with us and told each other that we will be saving them for life. It was really amazing. There were quite setbacks when we were looking for the data. We tried scrapping it out. We used image data but we soon knew it was not going to be providing a solution. But anyways we come over it just to be better. We would like to thank our university professors Mr. Anshul Sharma and Mr. Rashmit Singh who encouraged us to participate in this competition. The major motivation for this topic was that we were not aware enough of it but when we studied and googled it, it just left us in shock and awe. An event like that could cause a lot of damage. It's one of those doomsday events. The severity inspired a lot, it looked like a problem wanting to be solved not us looking for a problem to solve.
References
Programming:
https://github.com/knolleary/pubsubclient
https://pypi.org/project/schedule/
https://scikit-learn.org/stable/
https://towardsdatascience.com/the-complete-guide-to-time-series-analysis-and-forecasting-70d476bfe775s
Datasets:
https://www.ngdc.noaa.gov/stp/solar/solarflares.html
https://services.swpc.noaa.gov/products/solar-wind/mag-1-day.json
https://services.swpc.noaa.gov/products/solar-wind/plasma-1-day.json
Tags
#AI #earth #sun #saveearth #space #hardware #software #iot #machinelearning