Conceitos Básicos
Machine learning techniques can be effectively leveraged to improve the accuracy of orbit estimation and atmospheric density modeling, which are critical for enhancing space safety and mitigating the risks of collisions between space objects.
Resumo
This survey provides an overview of how machine learning has been applied to tackle key challenges in orbit estimation and atmospheric density modeling:
Orbit Determination:
- The Extended Kalman Filter (EKF) is the standard method, but has limitations due to linearization and Gaussian assumptions.
- Machine learning approaches like Gaussian Mixture Models, Gauss-von Mises distributions, and Physics-Informed Neural Networks have been used to better represent the non-Gaussian nature of the orbit distribution.
- These methods can propagate the orbit uncertainty more accurately compared to the EKF.
Orbit Prediction:
- Analytical and numerical methods like SGP4 have errors due to simplifying assumptions and lack of information about space object characteristics.
- Machine learning techniques have been used to correct the errors in physics-based propagators like SGP4, improving prediction accuracy.
- Data-driven models like Latent Force Models and Neural Networks have also been explored as alternatives to physics-based propagators.
Thermospheric Density Modeling:
- Empirical atmospheric density models like NRLMSISE-00 have limited predictive capability and uncertainty estimation.
- Machine learning has been applied to improve space weather forecasting, which drives these density models.
- Neural networks and other techniques have also been used to calibrate and combine multiple density models to reduce forecasting errors.
Overall, the survey highlights how machine learning can work in parallel with classical methods to enhance orbit estimation and space safety by better characterizing the uncertainty, improving prediction accuracy, and advancing atmospheric density modeling.
Estatísticas
The number of space objects larger than 1 cm is estimated to be around 1 million, with only 30,000 larger than 10 cm being tracked.
Orbit determination errors are in the order of kilometers for 7-day predictions, which is insufficient for space debris tracking.
Atmospheric density models have 10-15% one-sigma accuracy on average, and a 10% error in EUV light prediction can result in over 200 km uncertainty after 7 days.
Citações
"To avert a chain reaction of collisions, known as Kessler Syndrome, it is essential to accurately track and predict debris and satellites' orbits."
"Current approximate physics-based methods have errors in the order of kilometers for seven-day predictions, which is insufficient when considering space debris, typically with less than one meter."
"On average, these [atmospheric density] models have a one-sigma accuracy of 10-15%, depending on the model, solar activity, and location."