Parameter Identification And Its Application In Automatic Train Operation Control

In recent years, as the demand of the public transportation in the major city rapidly increases, the advanced train control technology receives a wide public attention from the industry. For the purpose of quality of the service (QOS) improvement, Automatic train operation (ATO) system, with a key role in the CBTC (Communication Based Train Control) system, is developed for providing passengers pleasant riding experiences. Au-tomatic train control has been regarded as a complex and multi-goal process where the variation of train parameters and external disturbances bring many difficulties to achieve good control performances.System identification and parameter estimation are the basis for the model-based control applications. In this work, we study the train control problems from a view of parameter estimation and derive several parameter estimation algorithms for a class of linear-in-parameters (LIP) systems that can represent several train control models. By using the filtering technique and the batch of data, the accurate parameter estimates of the systems with the colored noises are obtained. By using the multi-innovation identifica-tion theory, the balance of convergence rate and the computational efficiency is achieved. By using the decomposition technique, the computational costs of the algorithms are re-duced for the convenience of the real-time applications. Also, the parameter estimation algorithms are embedded into the control synthesis for solving the ATO tracking control problem and the train stop control problem. The analysis shows that parameter estimation algorithms in the control algorithms play an important role in improving the control ac-curacy such as the tracking accuracy and the stop accuracy. In addition, the identification and the control problems under two kinds of’input-output’abnormality are investigat-ed. We study parameter estimation algorithms with input-output non-uniformly sampling and the fault detection method with possible input or output sensor malfunction, which effectively improve the resilience of the ATO system.The main work and the contribution are as follows.1. For a class of general linear-in-parameters systems with the autoregressive mov-ing average colored noises, a multi-innovation stochastic gradient algorithm and a filter-based multi-innovation stochastic gradient algorithm are proposed. Based on the regressive identification model, a direct multi-innovation parameter identifica- tion based self-tune tracking control algorithm is derived. The convergence of the parameter identification algorithm is analyzed by using the martingale convergence theorem and the stability of the control algorithm is proved.2. For a class of general linear-in-parameters systems with autoregressive colored noises, a gradient-based iterative algorithm and a decomposed least squares-based iterative algorithm are proposed. By using the decomposition technique, the com-putational loads are reduced. Based on the iterative parameter identification algo-rithm, a two-stage adaptive terminal train stop control algorithm is derived. The proposed algorithm can ensure not only the train stop accuracy but also the riding comfort.3. Considering the input-output non-uniformly sampling and the missing data in the ATO system, we model the ATO system as a dual-rate system. Based on the linear-in-parameters model, we propose an auxiliary model based non-uniformly sam-pling parameter identification algorithm and a varying-interval based parameter identification algorithm. Both proposed algorithms do not need to decouple the parameters, which is convenient for the ATO control.4. For the ATO system with the input-output sensor malfunction or failure, a fault detection method using the unknown input observer is proposed. The convergence conditions for designing the observer are analyzed and the design procedure of the residual based fault detection method is obtained. The proposed method can detect three classes of sensor failures for the ATO system by setting appropriate thresholds of residuals. The detection information can be used in multiple train-related fields such as automatic train control and train maintenance.