Bayesian Network Enhanced Multiple Model Algorithm and Its Application in Integrated Navigation System

The uncertainty of the model parameters of integrated navigation system and the instability of the system model are characteristic of unstructured environment. For these systems, large estimation errors are likely to occur if a fixed single model is used for navigation solutions. To solve this problem, a Bayesian network enhanced interacting multiple model (BN-IMM) filtering algorithm is proposed. In the proposed algorithm, certain motion characteristic variables are introduced on the basis of multiple model estimation, and Bayesian networks are established according to the causal relationship between variables and the system model. Bayesian network parameters are used to modify the model switching probability in multiple model estimation, which can reduce the dependence of real model recognition on prior knowledge in multiple model algorithm. The proposed algorithm can solve the problems such as model conversion lag and model probability mutation in the interacting multiple model (IMM) algorithm, and enhance the adaptive ability of the multiple model algorithm. The proposed BN-IMM was utilized as a local sub-filter within a federated filter, establishing an information fusion algorithm architecture for the strapdown inertial navigation system (SINS)/global positioning system (GPS)/odometer integrated navigation system. In the test, the output of gyro and accelerometer was taken as characteristic variables to build a Bayesian network. The established Bayesian network was used to dynamically predict the uncertainties in the integrated navigation system. The actual road tests demonstrate that the proposed federated BN-IMM algorithm can significantly enhance the stability and accuracy of state estimation in the integrated navigation system.