基于一维卷积神经网络的WSN多攻击行为判别研究

Research on WSN multi-attack behavior discrimination based on onedimensional convolutional neural network

无线传感器网络一般部署在户外等无人值守的环境中,容易被攻击者在物理上接近,因此更容易遭受攻击。当入侵者意图攻击信标节点而提供错误的位置信息给网络内其他节点时,对基于位置服务的无线传感网络是毁灭性的破坏,因此能够准确判别各类攻击对保障无线传感器网络(WSN)安全具有重要的意义。针对节点定位的攻击问题提出一种深度学习的WSN多攻击行为判别方法,主要识别重放攻击、干扰攻击和女巫攻击三种类型。该方法基于信标节点的位置信息和网络的拓扑属性构建具有代表性的特征,然后利用一维卷积神经网络(CNN)从原始特征中获取更具有代表性的预处理特征,最后利用输出层激活算法通过随机梯度下降法更新深度学习模型的权重值,从而完成对攻击行为的分类。实验表明,该算法对信标节点4种状态的平均识别率达到了94.23%。

Wireless sensor networks are generally deployed in unattended environments such as outdoors,and are easily physically accessed by attackers,so they are more vulnerable to attacks.When an intruder intends to attack a beacon node and provides wrong location information to other nodes in the network,it is devastating to the wireless sensor network based on location services.Therefore,it is important to accurately identify various types of attacks to ensure the security of WSN significance.Aiming at the problem of node location attack,a deep learning WSN multi-attack behavior discrimination method is proposed,which mainly recognizes three types of replay attack,interference attack and witch attack.This method builds representative features based on the location information of beacon nodes and the topological properties of the network,then uses a one-dimensional convolutional neural network(CNN)to obtain more representative preprocessing features from the original features,and finally uses the output layer The activation algorithm updates the weight value of the deep learning model through the stochastic gradient descent method,thus completing the classification of the attack behavior.Experiments show that the average recognition rate of the algorithm for the four states of the beacon node reaches 94.23%.