Method of Learning Dynamic Bayesian Network Parameter Based on DEQPK Algorithm

For the problem of DBN parameter learning from small sample data, a differential evolution based on qualitative prior knowledge (DEQPK) is proposed. Firstly, the feasible region defined by the parameter qualitative constrains is sampled by the Monte Carlo to obtain the qualitative prior knowledge (QPK); secondly, the search space is reduced according to the QPK, and then the DE algorithm is used for parameter learning; finally, the QPK and the results of the DE algorithm are fused to acquire the real parameters. In the simulation experiment, three algorithms are used for parameter learning. The results show that the DEQPK is the most precise as well as the least time-consuming. At the same time, the parameters learned by the DEQPK algorithm is substituted into DBN, and the situations of battlefield targets are assessed.