基于混合简化粒子群算法的贝叶斯网络结构学习研究OACSTPCD

In order to improve the problems that the current Bayesian network structure learning algorithm tends to fall into local optimization,premature convergence and low optimization efficiency,a hybrid simplified particle swarm algorithm is proposed to optimize Bayesian network structure learning(BNs-HsPSO).The algorithm uses the maximum support tree to constrain the search space,and proposes an initial orientation strategy combining V-structure and conditional relative average entropy(CRAE),and then uses the mountain-climbing strategy to establish the initial particle swarm,uses the improved particle swarm optimization algorithm and genetic algorithm to iteratively optimize the initial population,proposes a conditional crossing and mutation strategy in the iterative process to avoid random divergence update of particles,and updates the unoptimized particles in combination with the sub-particle slowing strategy to avoid the algorithm falling into local optimum.The algorithm is compared with other algorithms in simulation experiments under four standard networks.The experimental results show that the proposed algorithm has an average higher BIC score of 5.775％,5.8％,0.475％,and 2.75％compared to MMHC,GS,BNC-PSO,and PC-PSO algorithms in ASIA,CAR,CHILD,and ARM networks,respectively;the Hamming distance HD is lower and the accuracy ACC is higher.