宣传招生策划.pdf

第12卷第2期 2012年4月交通运输系统工程与信息 Journal ofTransportation Systems Engineering andInformation Technology April 2012 文章编号:1009-6744(2012)02-0034-07 基于人工神经网络的驾驶行为动态集成学习算法梁军+,沙志强,陈龙(江苏大学汽车工程研究院,江苏镇江212013) 摘要:针对传统驾驶决策模型难以体现驾驶员驾驶过程中对交通环境的感知、判断、决策、动作等环节存在不确定性和不一致性,提出了一种基于神经网络的驾驶行为动态集成学习算法——,然后动态选择泛化误差E最小的个体网络进行集成,采用拉格朗日函数法求解最优集成权系数tOi,并引入 agent联盟的思想,把联盟中的个体网络对应的神经元输出做加权平均后,,仿真实验中得到的驾驶员踩踏踏板的习惯行为仿真结果与实际采集的样本数据总体趋势基本吻合. 关键词: 综合交通运输;动态集成学习算法;人工神经网络;驾驶行为仿真中图分类号: 文献标识码:A Dynamic work—Based Integrated Learning Algorithm forDriver Behavior LIANG Jun,SHA Zhi—qiang,CHEN Long (AutomotiveEngineering Research Institute,Jian挚uUniversity。Zhenjiang 212013,Ji蛐gBu,China) Abstract: Vehicle driyer"’Sperception,judgment,decision and action towards trafficenvironment ale usually uncertain and itumnsistent during theirdriving difficultto眦the traditional driving decision model accurately predict thedriving behaviors under these paper proposes aDNNIA algorithm tOdescribe driving behavior bydynamically ,solne ANNs are firsttrained learndifferentkinds ofdriving behaviors basedonsample dataand鲫Iall amounts of theseANNs with minimal generalization elTor E are then selected andintegrated topredict thefinaldriving functionmethod used toresolvethecoefficient∞i foroptim