Nasa Space Apps Challenge: Deep Neural Network approach for DSCOVR's Oracle project

This repository is intended for the Nasa Space Apps Challenge competition results. We cleaned and expanded the database, removing the non-numerical values (Not a Number) from the main dataset and relating this database to the information about the mission studied. As instructed, we ostensibly tested various adaptive neural networks (ANN).

In this challenge, the team's knowledge of data processing was exploited, with NASA providing various databases and sources of information on the measured data. By exploiting the databases and the information provided about the Faraday Cup Sensor, we were able to numerically infer the temperature, velocity, and density values of the plasma sent by the sun to the probe at the Lagrange point, thus turning the dataset with level 1 information into level 2. We intend to implement an AdaRNN to predict the Kp values since AdaRNN is an ANN-type network with proven effectiveness on data with evolutionary noise, as is the case with the data provided in the main database.

The presentation of what has been developed is contained within the file NASA_Space_Apps_Challenger_Presentation.pdf.

We intent to use for this problem:

• Cacth22 and XGBClassifer;
• AdaRNN;
• LSTM;

This work was produced by:

• Igor Braga Palhano;
• João Gabriel Soares Ferreira;
• Matheus da Almeida Esteban;
• Paulo Douglas Melo da Silva;
• Ruan Pereira Alves;