Velocity and position error compensation using strapdown inertial navigation system/celestial navigation system integration based on ensemble neural network

Strapdown inertial navigation system (SINS)/celestial navigation system (CNS) integrated navigation system can estimate attitude errors and gyro drift. However, prior to star sensor working, initial misalignments and accelerometer bias may cause large velocity and position errors which cannot be estimated by using CNS. Therefore, this paper aims to find an effective solution that can estimate and correct for the navigation errors caused by the initial misalignments as well as the inertial sensors errors at the start-up of CNS. This paper adopts an estimation method using time evaluation of the system's state transition matrix. Mathematical details for this efficient and novel idea are put forward in this research. The conventional Kalman filter assumes that the system model and the observation model are linear. The paper presents a method, which utilizes neural network ensembles to deal with the Kalman filter. Simulation results demonstrate validity of the proposed method and clearly show that integrated navigation solution can be used for extended periods without degradation.