作　　者： ; ;

机构地区： 深圳大学

出　　处： 《计算机应用》 2009年第6期1539-1543,共5页

摘　　要： 基于贝叶斯模型的结构学习是NP完全问题。给出了一种新颖的基于粒子群优化算法(PSO)的无序条件下的结构学习方法。迭代中,新算法用粒子表示网络结构,根据PSO群体规则进行更新,同时利用互信息进行修正,将所得到的最佳网络结构作为K2算法的先验知识。实验结果表明,新算法可以在无序的条件下,取得稳定的贝叶斯分数值。同时,在与K2算法和BN-PSO算法的比较中,新算法取得了更优异的网络结构和更快的收敛速度。 Structure learning of Bayesian network is a NP hard problem. A structure learning method without an ordering requirement based on novel Particle Swarm Optimization （PSO） was proposed. In this method, the structures were represented by particles, and updated according to swarm intelligence of PSO. Then the net structures were modified with mutual information. Finally the K2 algorithm was applied with the global best particle as the ordering and updating the best particle for next generation. Experimental results demonstrate that the performance of this new strategy is better than that of K2 and BNPSO, especially it has higher convergence rate and better Bayesian score without a previous ordering.

关 键 词： 基因调控网络 贝叶斯网络 粒子群优化算法 算法

领　　域： [自动化与计算机技术] [自动化与计算机技术]