Model Identification And State Estimation Of High Speed Electrical Multiple Unit

With the continuously high speed operating, the dynamic environment of EMUoperation continues to deteriorate. The air resistance effect, nonlinear and couplingcharacteristics of carriages become more and more stronger. According to theproblems of complexity, uncertainty and nonlinear, the study on modeling of highspeed EMU and parameter optimization is carried out. The main contents of the paperare as follows:(1) A maximum likelihood identification method is proposed to parameterestimation of nonlinear dynamic EMU model subjected to non-gaussian noise. Firstly,a stochastic discrete nonlinear state-space model is established to describe thedynamic behavior of EMU as a single-point-mass object under analyzing the basicforce in a horizontal direction of EMU. And the expectation-maximization approach isemployed to compute the ML parameter estimates. In addition, particle filtering andparticle smoothing technique is given to estimate the state of EMU, which is used tocompute approximation of the conditional expectation. Furthermore, gradient-basedsearch method is presented to maximize the conditional expectation. And theidentification algorithm is given for parameter estimation of high-speed train.Theconvergence rate of the identification algorithm is also discussed in detail. Finally, thesimulation results show the efficiency of the proposed method.(2) A maximum likelihood (ML) solution to the problem of identifyingparameters for a multiple-point mass model of high-speed electrical multiple unit(EMU) is presented. A stochastic discrete nonlinear state-space model is proposedto describe the dynamic behavior of multiple-point mass model of high-speed EMU.And the expectation maximization (EM) algorithm is employed to solve the problemof ML parameter estimates. In addition, an improved particle filter approach is givento estimate the state of high-speed EMU, which is used to compute approximation ofconditional expectation. Then, the conditional expectation is optimized bygradient-based search method. Furthermore, the identification algorithm is given forparameter estimation of multiple-point mass model of high-speed EMU. The limit point of the identification algorithm is also analyzed. Finally, numerical simulationstudy of parameter estimation for multiple-point mass model of high-speed EMU isimplemented and the results show the effectiveness of the proposed ML identificationmethod.(3)The problem of estimate state that multiple-point mass model of high-speedelectrical multiple unit（EMU）under the abrupt change condition is disscussd. Thestate estimation problem for high-speed EMU is formulated as a least-squaresproblem with sum-of-norms regularization. The regularization parameter is used tocontrol the trade-off between the fit and the state changes over time, and the weight isemployed to improve the robustness of state estimation. Furthermore, the algorithm isgiven for state estimation of multiple-point mass model of high-speed EMU. Finally,the effectiveness of the proposed approach is confirmed through numericalsimulation.