摘要
介绍了高速列车动模型试验装置独特的组成结构与原理;在分析弹射系统动模型列车运动状态及其影响因素基础上,构建3层前馈神经网络(ANN)模型,实现从输入试验给定速度、弹射质量和环境温度到控制弹射力输出的复杂非线性映射;采用把遗传算法(GA)与改进的误差反向传播(BP)算法有机结合的混合(GA-BP)优化方法求解ANN模型最佳参数,并提出了模型跟踪系统特性变化的自适应学习规则。检验样本数据验证了GA-BP网络模型良好的学习和泛化能力;重复试验应用结果显示,弹射试验速度控制精度优于2 m/s,表明了系统模型的有效性和稳定性能。
A unique moving model rig (MMR) developed for research on high-speed train aerodynamics is introduced. In modeling of MMR's rubber launching system, which is of high-order nonlinearity in terms of complicated mapping from its inputs to the output, a three-layer static feed-forward artificial neuron network (ANN) is adopted to predict optimal launching force according to the ambient temperatue, mass and expected running speed of the tested model train. ANN's optimized parameters are searched out by using a delicately designed hybrid learning (GA-BP) algorithm cooperating genetic algorithm with improved BP algorithm. Validation test of training data and experiment results show that excellent learning and generalization abilities of the ANN model are acquired, as a result of which MMR enables to perform accurate test speed control with an uncertainty of less than 2m/s.
出处
《测控技术》
CSCD
2005年第9期50-54,58,共6页
Measurement & Control Technology
基金
国家"九五"重点科技攻关计划顶目(96A070302)
铁道部科技研究开发计划项目(2003J027)
关键词
高速列车
动模型试验
系统建模
神经网络
遗传算法
BP算法
high-speed train
moving model test
system modeling
neuron network
genetic algorithm
BP algorithm