摘要
随着物联网设备规模激增,网络攻击手段日益复杂,传统恶意流量检测方法因特征单一、动态适应性差而难以有效识别新型攻击。为此,本文提出一种融合多传感器数据的网络恶意流量监测框架,通过程序化生成涵盖网络流量与设备状态的多维数据,结合具有噪声的基于密度的空间聚类应用(Density-Based Spatial Clustering of Applications with Noise,DBSCAN)与时间序列分析实现异常检测。实验结果表明,流量大小在特征权重中占比达0.79,主导异常判别,但单一特征依赖可能导致协议滥用型攻击漏检;可视化分析揭示正常流量呈宽峰分布,异常流量在中高区间形成密集双峰,且时序尖峰事件呈现瞬时突增与阶梯累积两种模式。模型通过滑动窗口与动态基线增强对脉冲式攻击的捕捉,但在持续性高强度攻击检测中受限于数据生成预设的截断阈值。研究成果为物联网环境下的动态安全监测提供了可行的技术路径。
With the rapid increase in the scale of Internet of Things(IoT)devices and the growing complexity of network attack methods,traditional malicious traffic detection methods are struggling to effectively identify new attacks due to their single feature set and poor dynamic adaptability.Therefore,this paper proposes a network malicious traffic monitoring framework that integrates multi-sensor data.It programmatically generates multi-dimensional data covering network traffic and device status,combining density-based spatial clustering of applications with noise(DBSCAN)with time series analysis to achieve anomaly detection.Experimental results show that traffic size accounts for 0.79 of the feature weight,dominating anomaly detection;however,reliance on a single feature may lead to missed detections of protocol abuse attacks.Visual analysis reveals that normal traffic exhibits a broad peak distribution,while abnormal traffic forms a dense double peak in the mid-to-high range,with time-series spikes showing both instantaneous surges and stepped accumulation patterns.The model enhances the capture of pulse attacks through sliding windows and dynamic baselines,but its detection of persistent high-intensity attacks is limited by the preset truncation threshold in data generation.This research provides a feasible technical path for dynamic security monitoring in the IoT environment.
作者
潘科
王强
PAN Ke;WANG Qiang(State Grid Jinhua Power Supply Company,Jinhua 321001,China)
出处
《国外电子测量技术》
2025年第11期71-77,共7页
Foreign Electronic Measurement Technology
关键词
恶意流量检测
多传感器
时间序列
特征权重
malicious traffic detection
multi-sensor
time series
feature weight
