基于深度极限学习机的危险源识别算法HIELM

发布时间：2018-03-01 05:19

  本文关键词： 危险源识别 深度学习 极限学习机(ELM) 分类　出处：《计算机科学》2017年05期 　论文类型：期刊论文

【摘要】：危险源识别是民用航空管理的重要环节之一,危险源识别结果必须高度准确才能确保飞行的安全。为此,提出了一种基于深度极限学习机的危险源识别算法HIELM(Hazard Identification Algorithm Based on Extreme Learning Machine),设计了一种由多个深层栈式极限学习机(S-ELM)和一个单隐藏层极限学习机(ELM)构成的深层网络结构。算法中,多个深层S-ELM使用平行结构,各自可以拥有不同的隐藏结点个数,按照危险源领域分类接受危险源状态信息完成预学习,并结合识别特征改进网络输入权重的产生方式。在单隐藏层ELM中,深层ELM的预学习结果作为其输入,改进了反向传播算法,提高了网络识别的精确度。同时,分别训练各深层S-ELM,缓解了高维数据训练的内存压力和节点过多产生的过拟合现象。
[Abstract]:Hazard source identification is one of the important links in civil aviation management. The result of hazard source identification must be highly accurate in order to ensure the safety of flight. In this paper, an algorithm for identifying hazard sources based on depth limit learning machine (HIELM(Hazard Identification Algorithm Based on Extreme Learning machine) is proposed. A deep network structure is designed, which is composed of multiple deep stack extreme learning machines (S-ELM) and a single hidden layer extreme learning machine (ELM). Multiple deep S-ELM use parallel structure, each can have different number of hidden nodes, according to the classification of dangerous source domain to receive risk source state information to complete the pre-learning, In single hidden layer ELM, the pre-learning result of deep ELM is used as its input, and the back propagation algorithm is improved, and the accuracy of network recognition is improved. The S-ELMs are trained separately to alleviate the memory pressure and the over-fitting caused by the excessive number of nodes in the high-dimensional data training.
【作者单位】： 南京航空航天大学计算机科学与技术学院;
【基金】：江苏省产学研联合创新资金项目(SBY201320423)资助
【分类号】：TP18;V328
，

本文编号：1550583