Relation Networks of Dynamic Bayesian Structure Learning in Non-stationary Random System

The continuous variable non-stationary systems learning model is problem for dynamic Bayesian networks (DBN) with variable structure for the problem of graph-model-based environment perception in autonomous control. Firstly, constant DBN structure learning model in smooth random system is discussed, and the Bayesian information criterion (BIC) score to continuous hide variable DBN and structure learning frame are researched. Secondly, on the basis of constant DBN, the fuzzy self-adapt measure algorithm is presented to learn the variable DBN structure in unsmooth random system. It is capabe of inferring walk modulus k and window modulus b in term of unsmooth grade modulus rb and adjusting modulus m, data time Δt through fuzzy logical and the variable structure DBN can be gained through the designed frame. In this paper, the general variable structure DBN learning model frame and the whole algorithm are presented. Simulation results are provided to demonstrate the effectiveness of the proposed algorithm.