Forwards-backwards information repairing algorithm and appliance on discrete dynamic Bayesian networks

For the study about the missing data on the Bayesian networks, the repairing algorithm given up to the present time is aimed at the unknown structure and parameters, or at the known structure and the unknown parameters. However the research about the data repairing on Dynamic Bayesian Networks (DBNs) whose structure and parameters are known is still in the primary stage. We proposed the Forwards-Backwards Information Repairing (FBIR) algorithm for this kind of data repairing, which was to use the combined evidences before a certain time slice and the evidences after this time slice to estimate the missing data which were on this time slice. We applied this algorithm to Discrete Dynamic Bayesian Networks (DDBNs) to identify the airplane group. It is proved by the simulation results that this algorithm is very efficient, in addition the repaired networks can significantly improve its accuracy, reliability and robustness.