Research On Intrusion Detection Method Of Urban Rail Transit Train Control System

With the modernization of China and the continuous development of urban rail transit,Communication Based Train Control(CBTC)system has been more widely applied to urban rail transit lines.The application of modern technology has not only contributed to the development of automation and information technology in CBTC system but also brought about safety problems.Due to the special characteristics of CBTC system,traditional intrusion detection methods in the field of information technology and industrial control cannot fully meet the needs of CBTC system for intrusion detection,with currently little domestic and international researches on intrusion detection methods for CBTC system.Therefore,it is necessary to carry out researches on intrusion detection methods for CBTC system.The main work in this dissertation is as follows:(1)CBTC system structure and its operation mechanism have been analyzed,information security hazards and typical attack methods of CBTC system have been studied,and CBTC system intrusion detection for its detection needs has been designed overally,including the design of intrusion detection process and specific detection methods.(2)Research on pre-processing methods of data set has been conducted.Meanwhile,with the characteristics of high dimensionality and high data repetition rate of dataset taken into consideration,natural number coding method,one-hot coding method,numerical normalization method and principal component analysis method have been combined to pre-process the data set analysis and improve the speed and accuracy of subsequent detection.(3)The CBTC packet characteristics have been analyzed through the CBTC system communication protocol,and appropriate security rules been set by means of the whitelist method to detect the communication data once and prohibit the passage of data outside the whitelist security rules.In order to improve the intrusion detection performance,BP neural network has been used for secondary detection,and the particle swarm algorithm with improved non-linear inertia weights have been used to optimize the BP neural network and improve the neural network convergence speed problem and the local minima problem.(4)The KDD CUP99 dataset and the CBTC system simulation dataset have been used as the experimental dataset respectively,and the comparative analysis shows that the intrusion detection method of whitelist combined with neural network has certain advantages over other algorithms.Therefore,the method has certain practical significance and application value.