Study Of An Intrusion Detection Based On Quantium Neural Networks Technology

With the increasingly far-ranging application of network, the way of network intrusion becomes multitudinal, so it threatens the security of the network seriously. Only use static technology (Access Control, Firewall, Data encryption and so on) to build a security system is difficult to detect the rough intrusion. Intrusion detection as a dynamic security and defense technology not only can identify the internal network attack, but also have a higher efficiency to detect external attack.The neural networks technology obtains the widespread application in intrusion detection, the representative is the BP (Back Propagation, is called BP for short) neural networks, but the slow constringency and local minimum nature of itself has limited the detection performance enhancement. In order to solve the problem of low detection rate for novel attack and the difficulties in detecting unknown intrusions in traditional intrusion systems, the thesis will use the multi-layer excitation function's learning to detect invasion: First the thesis uses many implicit strata excitation function to adjust the neural networks implicit strata's weight, compared with the traditional weight adjustment, this method may cause the data-in to correspond in the different space; Next, the thesis makes the adjustment to the implicit strata quantum neuron's quantum gap,so it can embody the uncertainty of the data. These steps enable the implicit strata neuron to express more states or levels, thus enhances the implicit strata neuron's processing speed and the examination efficiency. Then, based the above, the thesis also proposes an effective method to solve the feature easy to sink into the local minimum of forward feed neural networks, and designs it.The thesis designs a Quantum Neural Networks intrusion detection model, and designs data collection module, data processing module and the detection engine module in detail. In the last, the KDD 1999 data set is applied to Quantum Neural Networks, before the simulation experiment is done, the thesis carrys on a treatment to the intrusion data: first the thesis maps the characters to the data, next the thesis briefs and normalizes the data. Then the thesis uses the disposed KDD 1999 data as the input to Quantum Neural Networks, the output is defined as the invasion type. Experimental results show that Quantum Neural Networks has a better detection rate of the intrusion to the conventional BP Neural Networks.