摘要
首次把飞行故障检测视为一个非线性数据分类问题，从而可望借助人工神经网络来处理。为了克服ＭＬＦＮＮ在数据分类中存在的学习慢与分类精度低，发展了由ＭＬＦＮＮ和ＳＬＦＮＮ并联并可接收编码输入的ＤＰＦＮＮ模型，还将训练ＭＬＰ的有关算法推广到ＤＰＦＮＮ情形。用计算机仿真了若干飞行故障模式并用于测试ＤＰＦＮＮ。
Flight fault inspection which will be one of the key problems in pilot-aided equipment in the next generation of avionic system is first considered to be a nonlinear mapping,or classifi-cation problem,which provides a possibility to solve it with artificial neural networks(ANN).To overcome the problems of learning speed and classification accuracy in data classification with ANNs,the Double Parallel Feedforward Neural Network(DPFNN) model with encoded input is developed and training algorithms for MLP-like networks are extended to training the DPFNNs.TheDPFNN considered to be a parallel connection of an MLFNN with a SLFNN is expected to have better learning speed and classification accuracy,since the SLFNN can give a li near solution very fast and then the M LFNN uses its hidden nodes to adjust that solution to improve the DPFNN’ s performance. A few flight fauIt patterns as study cases are simulated on computer and used to test the DPFNN model.The experiment results have shown their good agreement with theoretical consideration。
出处
《航空学报》
EI
CAS
CSCD
北大核心
1994年第7期877-881,共5页
Acta Aeronautica et Astronautica Sinica
基金
航空科学基金
关键词
检验
神经网络
仿真
飞行故障
flight hazards-faults-inspection,neural nets, feedfomard control, simulations
