针对直接使用粒子群算法进行结构学习效率较低的缺陷,基于无约束优化,提出一种贝叶斯网络结构学习的混合粒子群算法。该算法首先构造并求解一无约束优化问题,其最优解对应的无向图中的边可为结构学习提供一搜索范围,缩小粒子群算法的搜索空间,然后在缩小的空间中完成对贝叶斯网络的结构学习,从而提高了粒子群算法的学习效率。仿真试验结果表明,该混合粒子群算法可以快速、准确地学习到最优贝叶斯网络结构。

In order to overcome the defects existing in the lower efficiency of structure learning caused by directly applying particle swarm algorithm to it, i.e. the search space is too large; a hybrid particle swarm algorithm for Bayesian network structure learning is presented based on unconstrained optimization problem. Firstly, for the algorithm, an unconstrained optimization problem is established and solved; the edges in the undirected graph corresponding to the optimal solution could provide a search range for structure learning and reduce the search space of particle swarm algorithm; then, Bayesian network structure learning is completed in the reduced space. Therefore, the leaning efficiency of particle swarm algorithm is raised. The simulation results indicate that the proposed method is able to quickly and accurately learn the optimum Bayesian network structure.