Research On Multiple Model Adaptive Estimation Filtering Mehtods And Application Study

With the increase of deep space exploration missions, technology of autonomous navigation becomes increasingly important. When it comes to the probes on transfer orbit, since they are relatively far away from the Sun and all planets, autonomous navigation method for Low-Earth orbit can’t satisfy their navigational requirements. So the celestial navigation becomes a very effective method. Celestial Navigation, a fully autonomous method, achieves high precision, doesn’t accumulate errors over time and bolds strong anti-jamming capability. Moreover,it can provide information on the position and attitude, and is well suited for the long-time navigation tasks with long-distance flight and complex external environment.On the transfer orbit of Mars probe, there are many low-Earth asteroids between the Earth’s and Mars’ s orbit, we can take advantage of asteroid observations to obtain information, and then confirm the navigation information of probe. Here we employ the four-body model as the navigation system state model, when using the image information or stars vector of asteroids as measurement, there will be attitude estimation error and sensor pointing error. However, taking the starlight angular distance as observation can avoid the effects on navigational accuracy in these two aspects, so observation model use the starlight angular distance as observations.Since errors will append inevitably in the state equation and observation equation of the navigation system, in order to get more accurate state estimation, we need to use filter estimation method to estimate the state variables of system. Due to the complex and changeable deep space environment, noise of navigation system state models change constantly. Compared with the single model, multiple model adaptive estimation method applies a set of parallel filter estimators, and calculates model probability constantly, which can meet need of process noise sequence and achieve adaptive effect.In this thesis, methods of combining multiple model adaptive estimation and extended Kalman filter/unscented Kalman filter are studied. Furthermore,the two methods are used for celestial navigation system on the basis of asteroids observations. By simulation, we compare celestial navigation system based on single model and celestial navigation system based on multiple model in detail, the results indicate that introducing multiple model and combining multiple model adaptive estimation theory with extended Kalman filter/unscented Kalman filter, when applying to celestial navigation system, can enhance the adaptability to the environment and greatly improve the accuracy of celestial navigation system.