Abstract:

An improved interactive multiple model filter based on Bayesian network (BN-IMM) is proposed. The aim is to resolve the problem when the noise of the autonomous underwater vehicle (AUV) integrated navigation system in the tough environment is uncertain or time varying. The proposed algorithm builds a Bayesian network according to the relationship of characteristic variables and the system model. The parameters of the Bayesian network are used to correct the model probabilities in the interactive multiple model (IMM)  algorithm which can reduce the dependence to the prior knowledge in the real mode recognition of the system. The proposed method can solve the problems of time lag in model transformation and probability jump in the IMM algorithm. The outputs of gyros and accelerometers are used as characteristic variables to establish the Bayesian network. Simulation results show that the BN-IMM algorithm can improve the model converting speed and the precision of estimation significantly when the AUV is in maneuvering state.