Static Bayesian Network Parameter Learning Using Constraints

To solve the problem of the static Bayesian network parameter learning using small sample, a study under restrained condition is proposed in the light of backward recursive accumulation parameter algorithm with priori constraints. Based on the variable of prior parameters, the constraints of domain knowledge described by uniform distribution and optimization algorithm, a Dirichlet distribution of prior parameter that resembles the even distribution most is obtained. By substituting that prior parameter to a transition probability model, the parameter learning process is completed. The efficiency and accuracy of the algorithm can be authenticated by the evaluation model of UAV.