中文摘要：

齿轮箱故障数据建模相当复杂,其计算量极大甚至是不可行的。提出一种基于蜂群算法的选择性神经网络集成算法来解决此问题。首先选取齿轮箱轴承故障数据训练各学习器,然后给每个学习器赋予权重系数,组成权值向量作为蜜源个体用于蜂群算法寻优,最后根据得到的最优权向量和阈值比较确定需要剔除的学习器。通过多种UCI数据集分析以及实际轴承故障数据集试验,结果表明新算法诊断效率明显高于基于遗传算法的选择性神经网络集成算法,同时这两种算法诊断精度相当,甚至新算法占优。

英文摘要：

Gearbox fault data modeling was very complex and its calculation was not feasible. Artificial bee colony algorithm based selective ensemble was proposed to solve the problem. First of all, gearbox bearing fault data was selected for training every learner. Secondly, all learners were given weight coefficients which compose the weight vector as nectar source individual for artificial bee colony algorithm optimization. Finally, comparison of the optimal weight vector and threshold was used to determine which learner should be eliminated. Through a variety of UCI data set analysis and actual bearing fault data set experiment, the results showed that the diagnosis efficiency of the new algorithm was obviously higher than that of genetic algorithm based selective ensemble and the diagnosis accuracy of both methods were fairly even new algorithm was dominant.