Research On Parameter Identification For Train Control Model And Its Online Learning Algorithms

In order to satisfy the high efficiency and high density request of urban railway traffic system, automatic train operation (ATO) can replace the experienced driver to complete the driving task. And the ATO system has high requirements for train control system. we have to know that, more accurate the parameters are, more precise the controller is. Considering the parameters will have small changes because of various internal and external factors, the online learning method was presented in this paper to update the parameters. So the task of this paper is to identify the parameters of train control system and update them based on the experiment data of Dalian railway line three, that analysis lays the foundation for designing the controller.The paper's research emphasized on several aspects as follows:1. The experiment data of Dalian railway line three are analyzed. Using data of train coasting state, the genetic algorithm was carried on the identification to train basic resistance parameters and got good performance.2. Analyzed the relationship between the calculation interval and error to determine the reasonable calculation interval. Compared the nonlinear least squares algorithm and genetic algorithm based on cross-validation. At last chose nonlinear least squares algorithm to identify the parameters of train brake model. The traction characteristics have a relationship with train velocity which was taken into account. And the accuracy of the model was verified.3. Widrow-Hoff was used to update the train basic resistance parameters and Gradient method was used to update the brake and traction parameters and reduce the system error.4. The parameter identification system program development and interface were completed, so users can use it easily.