面向粒子群优化的贝叶斯网络结构学习算法

Structure learning algorithm of Bayesian networks on particle swarm optimization

提出了一种基于离散粒子群优化的贝叶斯网络结构学习算法——PSBN(Particle Swarm for Bayesian Network)。贝叶斯网络的结构被映射为一种符号编码,通过在迭代过程中对粒子的符号编码进行调整,从而进化得到具有更高适应度值的贝叶斯网络结构。根据贝叶斯网络的结构特点,粒子位置和速度的编码方案和基本操作被设计,使得算法对贝叶斯网络的结构学习有较好的收敛性。实验结果表明,与基于遗传算法的贝叶斯网络结构学习算法相比,PSBN算法具有较好的学习效果。

A discrete PSO（Particle Swarm Optimization） based Bayesian network structure learning algorithm—PSBN（Particle Swarm for Bayesian Network） is proposed.A fitness function is given to evaluate the possible BN structure.Based on the characteristics of BN structure,the definition and encoding of the position and velocity of particle in PSO are given,and the basic operations of PSO are designed,which provides guarantee of convergence.As BN structure is considered as a symbol encoding,the BN structure having higher fitness values can be gotten by changing the symbol encoding of particles.The experimental results show this algorithm has better performance than the BN structure learning algorithm based on genetic algorithm.