Multiple Model Estimation With Uncertain Observation

As one of the most essential technologies in the field of estimation, state estimation plays an important part in the fields of national defense and national economy. Kalman filter is the optimal estimator in linear systems, and the unscented Kalman flter (UKF) becomes an effective approach of nonlinear filtering. The multiple model estimation approaches become effective for target tracking. Observations contain uncertainties in practice. The uncertainties refer to the wrong measurements generated by sensors in clutter, or measurements missing beacause of temporary sensor failures etc. which can not be modeled by measurements noise. This kind of uncertainties will deteriorate the estimation or even make the estimators divergent when using standard filters.Based on the above background, we first introduce a nonlinear estimation approach named randomized unscented Kalman filter (RUKF). RUKF eliminates the system errors caused by the UKF. Thus the accuracy of the estimation is improved. Then from the characteristics of the interacting multiple model estimation (IMM) algorithm, we apply it on estimating the states in systems with uncertain observations. The model set in the proposed IMM approach containes two sub model-sets. Final estimates are obtained basing on the fusion of the two model-sets’ estimates. This approach improves the stability of the estimator in systems with uncertain observations and performs better. Besides, the approach performs well when applied in target tracking. The simulation results show the proposed approach performs better than the traditional one-model estimation approaches in systems with uncertain observation.