基于分类优化贝叶斯结构算法的篦冷机参数状态分析及其算法收敛性分析

针对种群算法建立贝叶斯结构存在参数多、易陷入局部最优的问题,提出一种改进贝叶斯结构学习算法。该算法将候选结构分为优劣解集,利用师生交流机制优化优解集保留精英个体,利用变异机制优化劣解集来增加结构多样性,从而加快算法收敛速度,并在准确率和运行时间上达到平衡。最后不仅利用马尔科夫链证明该算法是全局收敛的,而且通过仿真实验验证了所提出算法的性能。将该算法应用到水泥篦冷机的实际数据中,构建水泥篦冷机工艺参数的贝叶斯网络结构,并完成篦冷机参数状态分析。

Parameter state analysis of grate cooler based on Bayesian structure algorithm based on classification optimization& and convergence analysis

Aiming at the problem that population algorithms of Bayesian structure learning have many parameters and easily fall into local optimum, an improved Bayesian structure learning algorithms is proposed. The algorithm combines the advantage of teaching learning based optimization without parameters and the random search of mutation mechanism. Teaching learning based optimization and mutation mechanism contrapuntally optimize candidate structures to learn the best Bayesian network structure. Teaching learning based optimization optimizes excellent set of structures to preserve entity. Mutation mechanism optimizes poor set of structures to increase structural diversity. By these operations, this algorithm not only accelerates the convergence speed , but also the balance between solutions quality and computational effort. Finally, the convergence of the algorithm is analyzed by Markov chain. The simulation results have shown that these properties can be achieved.