Abstract:This paper introduces surrogate models to estimate collision probability between two space objects. Previous approaches include Monte-Carlo simulation (MC) and multi-Gaussian distribution which integrats collision probability distribution function (PDF) into a defined encounter plane with some simplified assumptions, such as linearization of the velocity and the pdf of position. Compared with MC methods, surrogate models can reduce computing cost in two aspects:Firstly, orbit propagation with surrogate models is faster than in original mode; secondly, surrogate methods require less number of simulations to obtain the appropriate collision probability. Further, surrogate models require no basic linearization and Gaussian assumptions, and the PDF of collision probability lies in the propagation of input uncertainties, which might be shown as a non-Gaussian distribution Therefore, they can be widely used. The simulation shows that the surrogate models can precisely estimat collision probability while ensuring low computational cost.