摘要
扩展卡尔曼滤波(Extended Kalman filter,EKF)的准确性依赖于观测的质量、观测对象的非线性程度及动态模型的准确性.该方法通常假设其动态模型是不变的,而且默认为非线性程度较弱,这些在实际的车辆运动中都是不可靠的处理方式.本文提出了一种利用最小二乘支持向量机(Least squares support vector machine,LSSVM)的技术增强扩展卡尔曼滤波的新算法.LSSVM改进后的EKF算法(LSSVM-EKF)一定程度上弥补了EKF处理强非线性问题的不足;而且可以自适应地估计历史数据的动态建模偏差,并使用估计偏差来补偿动态模型.开发了一种引入Allan方差的K折交叉验证方法来确定LSSVM的训练参数;将动态模型偏差通过有限数据集与LSSVM一起训练;并引入无损变换将LSSVM与EKF进行了集成.为了验证算法,最后设计了车载试验,并采用列车数据验证了文中所提的方法,结果表明LSSVMEKF可以较好地适应实际车辆运动环境,可以提供一种可用的车辆定位方法.
The accuracy of extended Kalman filter(EKF)depends on the quality of the observation,the degree of nonlinearity of the observed object,and the accuracy of the dynamic model.The default is that the degree of nonlinearity is weak,and it is usually assumed that the dynamic model is constant,which is an unreliable treatment in actual vehicle motion.This paper proposes a new algorithm for enhancing the extended Kalman filtering using the technique of least squares support vector machine(LSSVM).The improved EKF algorithm(LSSVM-EKF)of LSSVM compensates for the insufficiency of the strong nonlinear problem of the EKF processing to some extent;and it can adaptively estimate the dynamic modeling deviation of historical data and use the estimated bias to compensate the dynamic model.A K-fold cross-validation method that introduces the Allan variance is developed to determine the training parameters of the LSSVM.The dynamic model deviation is trained with the LSSVM through the finite data set.The lossless transform is introduced to integrate the LSSVM with the EKF.In order to verify the algorithm,the paper finally designs the vehicle test,and uses the train data to verify the proposed method.The results show that LSSVM-EKF can adapt to the actual vehicle motion environment and provide a usable vehicle positioning method.
作者
陈光武
刘昊
李少远
杨菊花
魏宗寿
CHEN Guang-Wu;LIU Hao;LI Shao-Yuan;YANG Ju-Hua;WEI Zong-Shou(Gansu Provincial Key Laboratory of Traffic Information Engineering and Control,Lanzhou 730070;School of Electronic Information and Electrical Engineering,Shanghai Jiaotong University,Shanghai 200240;School of Traffic and Transportation,Lanzhou Jiaotong University,Lanzhou 730070)
出处
《自动化学报》
EI
CSCD
北大核心
2019年第12期2281-2293,共13页
Acta Automatica Sinica
基金
国家自然科学基金(61863024,71761023)
甘肃省高等学校科研项目(2018C-11,2018A-22)
甘肃省自然基金(17JR5RA089,18JR3RA130)资助~~
