欢迎来到天天文库
浏览记录
ID:37377929
大小:413.45 KB
页数:6页
时间:2019-05-22
《基于改进RBF网的汽车侧偏角估计方法试验研究》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、学兔兔www.xuetutu.com第46卷第22期机械工程学报VO1.46NO.222010年1l川JOURNALOFMECHANICALENGINEERINGNOV.2O10DoI:10.3901/JME.2010.22.105基于改进RBF网的汽车侧偏角估计方法试验研究术张小龙,2李亮李红志贺林宋健(1.清华大学汽车安全与节能国家重点实验室北京100084;2.安徽农业大学工学院合肥230036)摘要:基于汽车稳定性控制系统配置传感器信号,利用改进径向基神经网络技术对车身和车轮侧偏角进行估计。对径向基网络本最小乘算
2、法提⋯3条改进措施以获得合适的网络结构、提高网络的泛化能力和计算实时性。构建车身和前轮侧偏角、电子稳定程序(Electronicstabilityprogram,ESP)传感器信号测试道路试验系统,进行典型高附路面试验,并提取数据样本用于网络的学习和测试。通过网络结构和性能参数交叉验证,确定网络结构为4.12—2,扩展常数为9,}_j标学>j误差及其梯度分别为0.025和O.O5。由验证样本测试网络对车身和前轮侧偏角的估计精度分别为0.5。和O.8。。基丁PC平台对网络预测实时性进行测试。结果表明所构建的网络在精度和实时
3、性方面能够较好地满足ESP控制器对侧偏角的监控要求。关键词:汽车测试侧偏角估计改进径向基网络中图分类号:U270.7ExperimentalResearchonVehicleSideslipAngleEstimationBasedonImprovedRBFNeuralNetworksZHANGXiaolong-LILiangLIHongzhiHELinSONGJian(1.StateKeyLaboratoryofAutomotiveSafetyandEnergy,TsinghuaUniversity,Beijing100
4、084;2.SchoolofEngineering,AnhuiAgriculturalUniversity,Hefei230036)Abstract:Thetheoryandalgorithmoftheimprovedradialbasisfunctionneuralnetwork(IRBFNN)areappliedinestimationofvehiclebodyandwheelsideslipangles,basedontheconfigurationsensorsignalofelectronicstability
5、program(ESP)system.InordertosimplifytheRBFNNconstructionandimproveitsgeneralizationandreal-timecalculationperformance,threemethodsareputforwardtomodifytheradialbasisfunctionnetworkorthogonalleastsquares(OLS)learningalgorithm.Aroadtestsystemisdesignedtomainlyacqui
6、rethesignalsofthebodyandwheelsideslipangles,andofthekinemicsparametersofESPconfigurationsensors.Severaltypicalvehiclemanipulabilitytestsareconductedonhighadhesionroad,andthetestdataareusedtotrainandconfiguretheNNconstruction.ThefinalNNconstructionandlearningparam
7、etersaredeterminedbycross—verificationmethod,suchasthenetworkconstructionbeing4—12—2,theexpansionconstant9,thegoallearningelToranditsgradient0.025and0.05respectively.Theestimationaccuraciesofbodyandwheelsideslipanglesare0.5。and0.8。provedbytheverificationtestdata.
8、Finally,theNNreal—timepredictionperformanceistestedonthePCmachine.ThestudyshowsthattheconstructedRBFNNwithitsgoodaccuracyandreal—timecalculationperformancecanm
此文档下载收益归作者所有