高速列车纵向动力学模型时变参数在线辨识方法  

Online identification method of time-varying parameters for longitudinal dynamics model of high-speed train

作　　者：谢国[1] 张丹[1,2] 黑新宏[1] 钱富才[1] 曹源[3] 蔡伯根[3] 高橋聖 望月宽 XIE Guo ZHANG Dan HEI Xin-hong QIAN Fu-cai CAO Yuan CAI Bo-gen TAKAHASHI Sei MOCHIZUKI Hiroshi(Shaanxi Key Laboratory of Complex System Control and Intelligent Information Processing, Xi＇an University of Technology, Xi＇an 710048, Shaanxi, China Quanzhou Institute of Equipment Manufacturing, Chinese Academy of Sciences, Quanzhou 362200, Fujian, China School of Electronic and Information Engineering, Beijing Jiaotong University, Beijing 100044, China Department of Computer Engineering, Nihon University, Funabashi 274-8501, Chiba, Japan)

机构地区：[1]西安理工大学陕西省复杂系统控制与智能信息处理重点实验室,陕西西安710048 [2]中国科学院泉州装备制造研究所,福建泉州362200 [3]北京交通大学电子信息工程学院,北京100044 [4]日本大学,千叶船桥274-8501

出　　处：《交通运输工程学报》2017年第1期71-81,共11页Journal of Traffic and Transportation Engineering

基　　金：国家自然科学基金项目(U1534208,U1334211);陕西省青年科技新星计划项目(2016KJXX-79);陕西省科技统筹创新工程计划项目(2015KTZDGY01-04)

摘　　要：针对高速列车纵向动力学特性,分析了牵引力、制动力、阻力与速度和加速度的关系;考虑了天气和线路对高速列车运行状态造成的随机干扰,以及机械磨损和运行环境对列车模型结构参数造成的随机影响,建立了噪声干扰下的高速列车纵向动力学参数化状态空间模型,利用期望极大化准则,计算了列车模型参数的条件数学期望,并结合粒子滤波理论估计了参数粒子下的列车状态;基于贝叶斯后验概率理论,建立了高速列车非线性动力学模型的时变参数辨识方法,估计了列车的实时状态,并在噪声与参数分布均属于高斯分布、噪声属于高斯分布与参数属于指数分布、噪声属于伽玛分布与参数属于高斯分布的3种工况下,进行了蒙特卡洛仿真试验。仿真结果表明:在3种工况下,高速列车位移和速度的估计值与真实值的相对误差小于5%,列车模型参数估计值与真实值的相对误差小于10%,满足实际系统需求,因此,在高斯或伽玛噪声的干扰下,针对给定概率分布的时变参数,本方法均能实现系统状态的估计和模型参数的辨识。In view of the longitudinal dynamics characteristic of high-speed train,the relationships between traction,braking force,resistance and velocity and acceleration were analyzed.The random disturbance caused by weather and route in high-speed train running condition was studied,the random effect on the structural parameters of train model caused by mechanical wearand running environment was analyzed,and a nonlinear parametric state space model was established to describe the longitudinal dynamics characteristic of high-speed train under the disturbances of noises.The conditional mathematical expectations of the model parameters were calculated by the expectation maximization（EM）algorithm,and the state of the train with particles parameters was estimated by combining the particle filter theory.The online identification method of time-varying parameters for the nonlinear dynamics model of high-speed train was set up based on the Bayesian posterior probability theory,and the real-time state of train was estimated.The parameters of Gaussian and Exponential distribution under Gaussian or Gamma noise were identified by Monte Carlo simulation.The simulation result shows that the relative errors of estimated and real speeds and displacements of high-speed train are less than 5%,and the relative errors of estimated parameters and real parameters of train model are less than 10%,which meets the actual requirement of train system.So,when the probability distributions of time-varying parameters are given,the states and model parameters of train can be identified and estimated under Gaussian or Gamma noise by the method.

关 键 词：高速列车 非线性动力学模型 随机干扰 状态估计 参数辨识 

分 类 号：U270.11[机械工程—车辆工程]