Abstract
At the heart of the challenges for space situational awareness is to predict the orbit of a resident space object efficiently and accurately. Current orbit predictions that are solely grounded on physics-based models, however, fail to achieve required accuracy for collision avoidance and have already led to collisions. The failure of the physics-based prediction fundamentally arises from the lack of required information such as the space environment and resident space object’s characteristics that are challenging to acquire. With the hypothesis that a machine learning approach can learn the underlying pattern of the orbit prediction errors from historical data, in this paper, the use of support vector machines is explored for improving the orbit prediction accuracy. It is first demonstrated that the support vector machine model can reduce the orbit prediction errors with both good average and individual performances. Then, it is shown that the performance can be further improved with more training data, until adequate data information has been exploited. Last, it is illustrated that the support vector machine model shall be updated frequently, and orbit predictions should not be made too far in the future. It is concluded that the capabilities of the machine learning approach for improving orbit prediction accuracy are very promising.