%0 Journal Article
%T 无人驾驶汽车RBF神经网络滑模 横向控制策略？？<br>Lateral control strategy of RBF neural sliding mode for autonomous vehicles
%A 贺伊琳
%A 马建
%A 赵丹
%A 刘晓东
%A 张一西
%A 张凯
%J 长安大学学报(自然科学版)
%D 2018
%X 为了加强无人驾驶汽车横向运动控制，提出了一种优化型径向基函数（RBF）神经网络的滑模控制策略。根据视觉导航无人驾驶汽车单点预瞄模型与车辆二自由度模型得出横向运动状态方程，在滑模控制的基础上，采用RBF神经网络在线拟合滑模变结构的切换控制量，并基于改进的粒子群算法优化网络结构，使其快速达到滑模面，减小抖振。基于MATLAB/CarSim联合仿真平台，对建立的无人驾驶汽车横向运动状态模型及提出的控制策略进行不同工况下的仿真验证；基于A&D 5435建立无人驾驶汽车快速原型开发平台，完成实车试验。结果表明：基于优化的RBF神经网络滑模横向控制策略能精确实现对车辆横向运动的控制，一定程度上减小了系统建模不确定性带来的影响，能有效抑制方向盘转角的抖振，将横向距离偏差与航向角偏差控制在一定范围内，可靠跟踪期望路径；车辆等速循迹行驶试验时，提出的控制策略的方向盘转角试验结果与仿真结果最大相对误差为5.8%，横向偏差的最大相对误差为6.2%，航向角偏差的最大相对误差为5.4%，试验结果与仿真结果一致性较好；Alt 3 from FHWA仿真行驶工况下，相比于传统滑模控制策略，提出控制策略下的最大横向距离偏差误差和航向角偏差误差分别降低了90.8%和67.6%，双移线工况下误差分别降低了63.4%和69.9%，蛇形工况下误差分别降低了54.4%和39.6%。<br>To enhance the lateral motion control of autonomous vehicles, a sliding mode control strategy based on improved RBF neural network was proposed. First, the state equation of lateral motion was obtained based on the single point preview model and the two？？freedom model of the vehicle. Subsequently, the RBF neural network was used to replace the switching control of the sliding mode structure. On the basis of sliding mode control, the RBF network was optimized using an improved particle swarm optimization method, which enabled a faster sliding mode plane reaching and a more efficient buffeting reduction. The vehicle lateral motion model and the proposed control strategy were verified under various conditions simulation experiments using the MATLAB/CarSim software. Further, a rapid prototype development system of autonomous vehicles was built with A&D 5435, on which vehicle tests were carried out. The results show that precise control for vehicle lateral motion can be realized through a sliding mode control strategy based on the improved RBF neural network. The uncertainty of system modeling is compensated to a certain extent and buffeting of the steering wheel angle can be reduced. Moreover, the lateral distance deviation and heading angle deviation are controlled within acceptable range. The vehicle？？s trajectory highly matches the desired path. When the vehicle is driven with a constant speed on the testing ground, the relative error of maximum difference of the steering wheel angle between the experimental and simulation results is 5.8%. The relative error of maximum difference of the lateral deviation between the experimental and simulation results is 6.2%, and the relative error of maximum difference of the heading angle deviation between the experimental and simulation results is 5.4%, based on the designed control strategy. The test results are consistent with the simulation results. 〖JP2〗Compared with the traditional sliding mode control strategy, under Alt 3 from FHWA
%K 汽车工程
%K 横向运动控制策略
%K 滑模控制
%K 无人驾驶汽车
%K RBF神经网络
%K 粒子群算法&lt
%K br&gt
%K automotive engineering
%K lateral motion control strategy
%K sliding mode control
%K autonomous vehicle
%K RBF neural network
%K particle swarm optimization
%U http://zzszrb.chd.edu.cn/oa/DArticle.aspx?type=view&id=1805030