Factoring Dynamic Bayesian Networks based on structural observability

Dynamic Bayesian networks (DBNs) provide a systematic framework for robust online monitoring of dynamic systems. This paper presents an approach for increasing the efficiency of online estimation by partitioning a system DBN into a set of smaller factors, such that estimation algorithms can be applied to each factor independently. Our factoring scheme is based on the analysis of structural observability of the dynamic system. We establish the theoretical background for structural observability and derive an algorithm for generating the factors using structural observability analysis. We present experimental results to demonstrate the effectiveness of our factoring approach for accurate estimation of system behavior.