【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。更多还原

【Abstract】 Hazard identification plays an important role in aviation safety management.The results of hazard identification must be highly accurate to ensure the safety of the flight.To this end,a hazard identification algorithm based on deep extreme learning machine(HIELM)was proposed,which consists of multiple deep stacked ELMs and a single ELM.There are a parallel structure and different number of hidden nodes among these deep ELMs,and the hazard information is gotten to produce deep features according to the hazard area.In addition,the way of generating input weights has been enhanced with recognition features.The single ELM receives the results as its input.With the help of the improved back propagation algorithm,the network can achieve much better accuracy.The thought that deep ELMs are trained respectively alleviates the memory pressure and the over fitting phenomenon when facing the high-dimensional datasets.更多还原