Discrete dynamic BN parameter learning under small sample and incomplete information

Aiming at the discrete dynamic Bayesian network parameter learning under the situation of small sample and incomplete information, a constraint recursion learning algorithm is presented. The forward algorithm is used to establish a parameter recursion estimation model of discrete dynamic Bayesian network with hidden variables. A prior parameter constraint model with uniform distribution is established with the present network parameters as variables. Then the approximate Beta distribution could be acquired through the optimization algorithm. Finally, the distribution of prior parameter knowledge could be used in the above model of recursive estimation to finish the parameter learning process. The method is applied to the unmanned aerial vehicle dynamic model of threat assessment. The results show the effectiveness and accuracy of the proposed algorithm.