Aiming at the problem of excessive state dimension of nonlinear filter in SINS filtering alignment process under large misalignment angles, the Kalman filtering/extended Kalman filtering (KF/EKF2) hybrid filtering method based on the model decomposition is proposed. The initial alignment model based on Euler platform error angle (EPEA) is decomposed into the linear part and nonlinear part, and is processed by the linear KF filter and nonlinear EKF2 filter respectively. The filtering process is designed to ensure the overall optimization of state variables. The experiment results show that the KF/EKF2 hybrid filtering algorithm is superior to the most commonly used unscented Kalman filtering (UKF) and EKF2 filtering in terms of computational complexity, real-time performance and accuracy.