Self-learning Neural Network-based Intrusion Detection System Design And Implementation

With the fast development of the Internet, network application becomesan important part of people's lives. The Internet security we face to is risingdue to its opening and frangibility as well as the growth of its applications.Whether the OS, software, or hardware has some security hidden troubles.Such hidden troubles give hackers chance to attack the system. The attacks toInternet have become more complex and flexible and its bugs are exploredmore in recent years, the fact of Internet security presents to people we mustface to.In order to solve these problems, a lot of secure policies, mechanismand tools (including IDS) are analyzed. IDS is one of well-know securefacilities, it could choice corresponding method to protect system securitybased on an analysis of attacking actions. But traditional IDS have somelimitations, such as it can't identify unknown attacking action. TraditionalIDS needs people to maintain its rule lib by updating its character lib. Thereis not a mature intelligence IDS product yet, which can identify unknownattacking actions automatically.In this thesis, author analyzes the actuality of network and informationsystem security, and indicates the limitation of current technology Based onstudying the existence and trends in IDS. Author gives out a self-study IDSmodel and solution based on neural network. In this model, a creative systemstructure about several iterative neural networks is presented, and thetransfer formulas about every input vector are also proved. In this research,author finished the detailed design and implement, and tested and trained thisnetwork by a lot of data including known and unknown attacking action. Thenew IDS makes a few improve on performance and breakthroughs somelimitation of traditional IDS.