Research On Intrusion Detection Technology Based On Genetic Algorithm And Neural Network In The TDCS Network

In the environment of rapid development of railway’s information construction, Train operation Dispatching Command System(TDCS) quickly covers the entire way in order to improve the railway’s service capability. At present, this system can realize automation of train running, automatic adjustment of running plan, data sharing and so on. It is the effective protection of the railway transport command of information technology and automation.The level of modernization of railway transportation dispatch and command has greatly improved by TDCS used. It means that only promoting the whole system’s confidentiality, integrity and availability can keep traffic safe.With the widespread implementation of TDCS, there are some safety facilities for TDCS. But they can not protect the system from security threats and computer viruses. Hence, the existing network security system only plays a role of protection, it can not completely solve the whole network system security problem. The intrusion detection technology applied in the TDCS, can be achieved to control the whole system anytime. On the other hand, monitoring and limiting the data through the network can protect the network system from every aspect and provide a solid foundation more for improving TDCS security.Currently, there are a lot of intrusion detection searchs to be used, but technically, there are still some problems to be solved. For example, how to make the high missing report rate and false-alarm rate of existing intrusion detection system low. Facing the problems of intrusion detection and the characteristics of TDCS, this thesis established a new intrusion detection model, which was used to protect TDCS network securely. This model combined anomaly detection with misuse detection. Taking the attributes characteristics, which were collected from data pre-processing, served as the inputs of neural network. Additionally, in order to overcome the shortages of falling into local minimum quickly and premature convergence, the neural network used genetic algorithm to optimize the weights. Because each behavior of KDD99data set is described with some features, in order to reduce the input dimension and to improve detection rate, this thesis used principal component analysis method to abandon the attributes characteristics with no contribution or low contribution in.training and testing data set.At last, with the help of Malab simulation platform, having genetic algorithm altered the weights and putting the trained neural network applied to the improved intrusion detection model for TDCS testing. The results showed that using principal component analysis method can greatly reduce the input dimension and improve the data processing speed effectively. After the neural network optimized by genetic algorithm, intrusion detection rate had been greatly improved, the missing report rate and false-alarm rate were significantly reduced.