基于卷积神经网络的网络攻击检测-毕业论文.docx

毕业设计论文 基于卷积神经网络的网络攻击检测 摘 要 传统的IDS在管理上存在着三大挑战：误报、人员配置和漏检问题。与此同时，机器学习技术的出现正改变着我们的世界观并冲击着我们日常生活的方方面面。它在信息安全的领域也正扮演着重要的角色，将它与网络攻击检测技术结合一直是网络安全研究的热点。设计实现一个基于卷积神经网络的网络攻击检测模拟软件正是本文的目的所在。 神经网络有多种，本文中将选择卷积神经网络作为网络攻击检测系统的神经网络。总体设计分为三个模块：即数据预处理模块、数据训练和测试模块、用户界面模块。对于数据预处理模块来说，选用KDD 99 数据集，将KDD 99 数据集中的字符型数据转换为非字符型数据。由于编程语言选用的Python，所以使用列表实现。将同一字段中字符型变量的所有类型写入一个列表，使用列表下标表示对应的类型，便将字符型数据转化为非字符型的数据。对于数据训练和测试模块，使用四层结构的神经网络，模型为单卷积层连接池化层后连接两个全连接层。使用带有标签的训练集数据训练出检测模型，通过模型对不带有标签的测试集数据进行检测，判断是否受到攻击。用户界面模块使用PyQt5设置文件路径的选择栏，添加测试、训练、退出功能按钮，并且显示训练的准确度。 对程序进行测试，选择5000条处理后的数据作为训练集，选择该训练集对网络进行训练，得到训练模型。分别选择1000条和20000条数据作为测试集1和测试集2，对两个测试集分别进行测试，准确率可达95%以上，达到设计目标与要求。 关键词：机器学习；网络攻击检测；KDD-99数据集；卷积神经网络 Network Attack Detection Based on Convolutional Neural Network Abstract Traditional IDS has three major challenges in management: false positives, staffing, and missed detection. However, the emergence of machine learning technology is changing our worldview and impacting every aspect of our daily lives. It is also playing an important role in the field of information security. Combining it with network attack detection technology has always been a hot topic in network security research. Designing and implementing a network attack detection simulation software based on convolutional neural network is the purpose of this paper. There are many kinds of neural networks. In this paper, convolutional neural networks will be chosen as the neural network of this network attack detection system. For the data preprocessing module, the KDD 99 data set is used to convert the character data in the KDD 99 data set to non-character data. Because of the choice of Python in the programming language, a list implementation is used. All types of character variables in the same field are writed to a list, and the list subscript is used to indicate the corresponding type, and convert the character data into non-character data. For the data training and test module, a four-layer neural network structure is used, and the