In order to solve the problem that the error model of strapdown inertial navigation system (SINS) has high state dimension when the azimuth misalignment angle of SINS on static base is large, the method based on unscented particle filter (UPF) would bring the curse of dimensionality,a method based on the Kalman filter/unscented particle filter (KF)/UPF integrated filter is proposed. The initial alignment model is decomposed into the linear part and the nonlinear part,the optimal state estimation of the linear part is achieved by KF, and UPF is used to estimate the state value of the nonlinear part. Finally,the KF/UPF algorithm is compared with the UPF algorithm in different number of particles by use of the simulation experiment, simulation results prove that not only the KF/UPF integrated filter can maintain the indexes of the high alignment accuracy and the fast convergence speed,but also the state dimension of UPF filtering is reduced from ten to three, the computation time is reduced to 52.69% and 6.0% of the reference value, respectively, which improves the realtime performance of the initial alignment.