Accident Prediction Of Interlocking System Based On Stochastic Hybrid Automata

As one of the core subsystems of the rail transit system,the safety of the interlocking system should be the goal of the development.The accident prediction of interlocking system is an important way to ensure the safety of interlocking system,and it is a hot issue in the research of rail traffic safety.In this thesis,we propose an approach to establish system model of interlocking system by using stochastic hybrid automata and predict accidents in interlocking system.The main work of this thesis includes:1)An accident prediction framework for interlocking system based on stochastic hybrid automata is constructed.The framework contains three main stages,which are system model generation,faults discovery and accident prediction.The system model,the monitor model and the prediction model are generated in the three stages.2)The discrete modeling and accident prediction method of interlocking system are presented.The patterns of the system model are first extracted,consisting of the discrete environment entity patterns and the controller pattern.Based on these patterns,the system model construction process is provided.Then,a specific discrete system model is constructed by inputting parameters.Secondly,the automatic generation algorithm of the monitor in the discrete model is designed and implemented.Finally,according to the domain knowledge,the train accident is defined,and a prediction model based on time difference is established.3)The hybrid modeling and accident prediction method of interlocking system are presented.Firstly,continuous variables are added to the discrete model for extracting the train entity pattern.On this basis,the system model construction process is defined.Then,a specific hybrid system model is conctructed by inputting parameters.Secondly,the automatic generation algorithm of the monitor in the hybrid model is designed and implemented.Finally,an accident prediction model based on distance is established.Finally,the characteristics and application situations of discrete modeling and hybrid modeling are summarized and compared.The feasibility and effectiveness of the proposed method are illustrated through case studies.