This project is awarded the Excellent Youth Scholars Award at KSCY.
This project is submitted to the Journal of The Korean Earth Science Society and pending evaluation.
This project is awarded an Honorable mention at the State Science Fair.
Utilizing Deep Learning Models for Space Debris Reentry Prediction
Among 420 space debris falling toward the Earth annually, an average of 10 to 40 percent of objects fall to the surface of the Earth without reentry burn-up. Although space debris that has entered the Earth’s atmosphere so far has fallen into inhabited areas, chances are high that they will fall into human-occupied areas as the number of space debris continues to increase in the near future. Predicting the reentry orbit of objects entering the Earth is thus important as it is directly related to safety, but existing orbital forecasting techniques based on the Simplified General Perturbations 4 model have a problem of increasing errors over time. Therefore, this study came up with a method to predict the reentry orbit of space debris by introducing a deep learning model to reduce errors in prediction.
In this study, six orbital elements representing the position and the speed of a space object in the Two-Line Element data (TLE data) along with the time interval between the data were given input variables. The input variables were then trained on different deep learning models and calculated for prediction. The predicted TLE data and the true TLE data were then used to calculate the distance by obtaining the three-dimensional coordinates (X, Y, and Z) in an Earth-Centered Earth-Fixed (ECEF) coordinate system. Finally, the trajectories were visualized by a MATLAB add-on program.
The significance of this study lies in employing deep learning models to predict the trajectory of space debris. The distance between the true point and the predicted point approximately estimates as 99 kilometers, and the MATLAB visualization signifies no difference in trajectories. However, if one could change the variables and the number of data in the future, one could expect to see an improvement in the model performance.
View Project Paper
View Project Poster