遗传神经网络的民用航空器SDR预测.pdf

南京航空航天大学
士学位论文
基于遗传神经网络的民用航空器SDR预测
姓名:解海涛
申请学位级别:硕士
专业:交通信息工程及控制
指导教师:孙建红
20090201
南京航空航天人学士学位论
摘要
随着困内航空业持续高速发展,近几年运输类飞机数量每年以100多架的速度增长。安全
是民用航空运输水恒的主越,民用航空器SDR(使用困难报告)能够反映出民用航空的管理
水平和安全状况,因此对SDR进行预测有着重要意义。
论文的研究作主要是基于人.神经网络理论的分析,针对民用航器SDR预测问题,
在分析影响SDR发生的诸多因素基础上,指出了采用传统预测方法对SDR进行预测存在的弊
,并提出了应用BP神经网络来预测SDR的新方法:通过建立预测模型,对波音737机2007
年1-12月SDR数量进行了预测,预测结某的精度很高。表明用BP神经网络对民用航空器SDR
预测完全可行,并且效果很好。
针对BP神经网络在权值调整过程中存在的收敛速度慢、易陷入局部极小点的固有缺陷,
木文采用遗传算法米优化BP神经网络,并对遗传BP神经网络进行了实证研究,明了遗传
BP神经?络在预测精度、运行速度方面均有很火提高。该方法在对波737飞机SDR进行预
测測时,也有很好的适用性,并进一步提高了精度。本文在遗传BP神经网络模型的基础上:预测
了该机刑木来儿年的SDR数量,具有?定的参考意义。最后得出结论,遗传BP神经网络应
用」民用航空器SDR预测方面有着很好的前景和应用价值
关键词:SDR預测,神经网络,BP算法,遗传算法,遗传神经网络
基于遗传神经?络的民用航空器SDR预测
Abstract
With the sustained and rapid development of domestic aviation industry, the number of car o
planc has bccn incrcascd by more than100 pcr ycar in reccnt ycars. Sccurity is an ctcrnal thcmc of
Civil Aviation
The research of my paper mainly deals with civil aircal SDR prediction, which is based on the
analysis to BP neural nctwork thcory. It analyzes many factors Influncing SDR, points out thc
malady existing in using traditional forecasting methods to do the prediction on the SDR, and puts
forward a new way by usin BP neural network to predicl SDR; through the establishment of the
mocl, it ucccssfully forecasted thc SDR quantity from January 2007 to Dcccmbcr on the Bocing
10
Aim al the inherent defects of Slow convergence and easy to [all into local minimum point
during BP nural nctwork adjusts weights, the papcr trics the approach of gcnctic algorithm to
optimize thc BP nural nctwork, docs the cmpirical rcscarch on gcnctic BP nural nctwork, and
proves that there is a great improvement of using on prediction accuracy and running velocity Using
this mcthod to Prcdict SDR quantity in 2008 and 2009 on the Bocing 737 modc, it has good
applicability and ulteriorly improves the accuracy, Finally comc to the conclusion, genctic BP ncural
network in prediction applies to civil aircraft SDR has a very good prospect and application value
Keywords: SDR predict, Neural Network, BP algorithm, Genetic algorithm, GA-BP algorithm