Research And Application Of BP Neural Network Optimization Based On Improved Ant Colony Algorithm

With the development of the Internet, network intrusion becomes increasingly complicated and frequent. The conventional security strategy can no longer meet the requirements of customers. The firewall which is generally considered effective has a lot of disadvantages, so it is not enough to cope with the growing network crisis. Therefore, real-time intrusion detection systems have been proposed as an effective complement to firewall. The only way to get a successful protection from complex blended threats is the effective combination of a variety of network infrastructure equipments and state-of-the-art network security defense technologies.For the emerging network intrusion, this thesis proposed an intrusion detection system based on combination of misuse detection and anomaly detection with high accuracy detection. Anomaly detection uses neural network module which has highly self-learning and adaptive capacity to identify intrusion behaviors accurately according to the existed intrusion detection instances in training samples and to identify and summarize the new features of the intrusion behaviors. Faced with the problems of slow convergence speed and easy to fall into local minimum point of BP neural network, this thesis proposed an improved ant colony algorithm for effective improvements. The weights and thresholds of BP neural network can be compressed to a certain extent by the global search ability of the improved ant colony algorithm. Neural network module uses a different protocol packet respectively handling mechanism. This thesis focuses on the major transport layer protocols (TCP and UDP) to design the modules. A TCP neural network and an UDP neural network were designed according to the different network intrusion feature information of the two protocols, thus improving detection efficiency targeted.From the current requirement of customers for network security and intrusion detection development direction, this thesis certained research project direction and thesis organizational structure. This thesis firstly have a brief introduction on the new designed system model, then have a detailed analysis on the functions of the six modules, including packet capture module, protocol analysis module, pre-processing module, rule matching module, improved BP neural network module and the response module, finally have a detailed analysis on the module structure and shortcomings of the BP neural network of the anomaly detection. For the defects of BP neural network of slow convergence speed and easy to fall into local minimum point, an ant colony algorithm with good global search capability is used to handle. This thesis further optimizes the global update formula which based on the combining global and local pheromone update in the improvement strategy, and applies this global and local update pheromone scheme to BP neural network. Simulation results show that the scheme improved detection efficiency. TCP and UDP neural network were simulated respectively, and the results show that the targeted network detection has higher accuracy.