Abstract: Coning error compensation algorithm in the process of self-alignment is put forward in light of the strapdown intertial navigation system (SINS) self-alignment error on swaying base. It reduces coning error and increases the self-alignment accuracy. The quaternion fourth-order Runge-Kutta algorithm and two-sample and three-sample and other error compensation algorithm of equivalent rotation vector and their theoretical compensation effect are analyzed and compared in theory; through simulation and experiments, it is obtained that the self-alignment error decreases along with increasing of the number of sample. When the number of sample increases 1, the self-alignment error of north-angle is reduced nearly 1 time and the dispersibility of attitude angle is reduced; with the increasing of the swaying amplitude and frequency, the self-alignment accuracy of three-sample compensation algorithm is close to stability . Considering the sampling frequency and the number of sample and calculation and alignment accuracy, the three-sample coning error compensation algorithm should be selected and it can satisfy the self-alignment requirements of SINS on swaying base.