自動網路入侵偵測及防禦之設計－運用倒傳遞類神經網路

隨著網路日益發達，人們對網路各種服務也隨之增加，相對著網路安全也是人們重視議題之一，且網路安全的防範與維護也是網路管理者的重責大任。因此，本論文研究目的是想透過倒傳遞類神經網路學習模式，能針對linux系列伺服器web記錄檔，來分類正常和危險行為，並進一步做好網路安全回應與防範。本論文實驗過程有包含類神經網路學習過程、防火牆阻隔危險行為等，首先將伺服器web記錄檔內容轉換成數值，經過倒傳遞類神經網路訓練與學習，並驗證其學習模式的準確度，找出最適合本論文的類神經網路最佳學習模式，再將有危險行為的IP篩選出，加入伺服器防火牆阻擋名單，能達成維護伺服器的網路安全。本篇論文是從分析記錄檔來提昇管理網路伺服器安全性，藉此能減輕管理者管控伺服器系統安全的工作，希望降低網路中發生資安問題可能性。

關鍵字

伺服器安全； web記錄檔；類神經網路學習

並列摘要

As the Internet increasingly is developed, people have needed increasingly the services of network, and network security is the one of the important issues for people. The prevention of network security is also the administrator’s responsibility. Therefore, the purpose of this thesis through the neural network learning model on linux web server log can be classify normal and dangerous behavior, and further improve the prevention of network security. This thesis’s experiment has included the learning process of neural network, and firewall blocking dangerous behavior . First , web server log will be converted to numeric values. Second, back-propagation neural network will be trained and learned, and be verified the accuracy of the learning mode to identify the most suitable neural network learning mode in this paper.Then the filtering firewall add the ip of the ricky behavior to block , and the server can be reached to maintain the network security. This paper is the analysis of web server log to achieve the management of the server’s network security. The way makes the system manager control the server security easily, and reduce the possibility of the problem of network security occurred.