基于支持向量机的道路坡度实时预测方法试验

摘要

Prediction of road slope is a key technology to vehicles' electronic real-time control system,such as ABS,AMT and hybrid torque distribution,and so on.In this paper,a real-time prediction method of road slope was put forward based on support vector machine (SVM),in which the input parameters of SVM module included engine speed,engine output torque,vehicle speed and longitudinal acceleration,and could be extracted from controller CAN network in real-time.The vehicle roadway test system and the CarSim simulation platform were built up respectively,and the samples required for SVM model learning,generalization performance test were achieved by the systematic tests.The squared correlation coefficient of SVM model from CarSim tests was 0.99,while it was 0.9 from roadway tests.The main reason for the difference could be that the GPS method in road slope test may add in a body pitch angle which could not be eliminated systematically.Furthermore,the SVM model of roadway test was imported into the real-time virtual controller PXIe using LabVIEW programming method.For the equivalent prediction time of one point to the single-chip computer selected by automotive electronic controller was only 1.33 ms,which met the requirements of real-time control.The road slope prediction method proposed in this paper is effective and practicable.%道路坡度预测是汽车ABS、AMT、混合动力汽车扭矩分配等实时控制的关键技术.提出一种基于支持向量机(SVM)的道路坡度实时预测方法,输入参数为发动机转速、输出扭矩、纵向车速和纵向加速度,均从控制器CAN网络中实时提取.分别构建实车道路试验系统和CarSim仿真平台,通过系统试验分别得到的样本对SVM模型进行学习和泛化能力测试.结果表明:CarSim试验数据建立的SVM模型预测平方相关系数达到0.99,实车试验数据建立的SVM模型预测平方相关系数在0.9左右,二者差异的主要原因是实车试验GPS方法获取道路坡度信息时叠加了不易消除的车体俯仰角的影响.基于LabVIEW编程将实车试验SVM模型导入虚拟仪器PXIe实时控制器中,其预测一个点的耗时等效到汽车电控ECU单片机为1.33 ms,完全满足实时控制要求.证明所提出道路坡度预测方法是有效、可行的.