基于加性协同的离散贝叶斯网络参数学习

Parameter Learning of Discrete Bayesian Network Based on Additive Synergistic Constraints

针对小数据集条件下离散BN参数学习的问题,为了将加性协同约束融入到BN参数学习过程中,通过借鉴经典保序回归算法的思想,提出四种处理加性协同约束的方法,进而利用经典的草地湿润模型对改进算法进行仿真,并与最大似然估计算法进行对比,仿真结果表明,改进算法在精度上有一定优势,能够很好的对最大似然估计算法进行修正,得到相对准确的参数,然而时效性则劣于最大似然估计算法。进一步将改进算法应用到弹道导弹突防模型的参数学习中,通过推理分析验证算法的有效性。

In order to integrate additive synergistic constrains into the learning process of discrete Bayesian Net- work parameters, four methods to deal with the additive synergistic constraints are proposed based on the idea of clas- sical isotonic regression algorithm. The four methods are simulated by using the classic wet grass model, and com- pared with the maximum likelihood estimation algorithm. Simulation results show that the proposed methods have some advantages in accuracy, which can correct the results of the maximum likelihood estimation algorithm to obtain relatively accurate parameters, while timeliness is inferior to the maximum likelihood estimation algorithm. Further- more, the proposed methods are used to learn the parameters of the model for ballistic missile penetration, and the ef- fectiveness of the methods is verified by inference analysis.