摘要
针对驾驶员建模中不确定因素的影响,采用操纵逆动力学方法,反求出驾驶员的操纵输入来避开驾驶员建模.神经网络作为一种较好的识别驾驶员输入的方法,其学习速度和收敛精度会影响识别精度.为了提高汽车操纵逆动力模型识别时神经网络的学习速度和收敛精度,基于Elman网络,采用一种新的动态神经网络结构——状态延迟输入动态递归神经网络(SDIDRNN).首先,建立三自由度人—车闭环模型并以实车试验数据验证了模型的正确性.然后,通过建立SDIDRNN网络模型,取闭环模型的仿真结果做为训练样本,对汽车操纵逆动力模型进行了识别,所得结果及误差分析说明了该神经网络在学习能力上的优越性及识别模型的有效性.
The method of vehicle handling inverse dynamics was used to identify driver modeling with unpredictable disturbances. Driver handling input was obtained to avoid modeling the model of driver by using inverse dynamics. Neural network was a kind of method to identify drivers＇ inputs, and the identification accuracy was affected by learning speed and convergence precision. Based on Elman network, a new dynamic neural network structure, namely state delay input dynamic recurrent neural network （ SDIDRNN）, was used to improve neural network＇ s learning speed and convergence precision of the vehicle handling inverse dynamic model recognition. Firstly, the three-degree-of-freedom driver-vehicle closed-loop model was established and the model＇ s accuracy was also verified by real vehicle test data. Then, the vehicle handling inverse dynamics model was recognized by building the SDIDRNN network model and taken the closed-loop model＇s simulation results as the training sample. The obtained results and the error analysis showed the neural network＇ s learning superiority and the model identification＇ s effectiveness.
出处
《应用基础与工程科学学报》
EI
CSCD
北大核心
2014年第3期606-615,共10页
Journal of Basic Science and Engineering
基金
国家自然科学基金项目(11072106)
关键词
汽车操纵动力学
逆问题
实车试验
神经网络
识别
vehicle handling dynamics
inverse problem
real vehicle test
neural network
identification
