摘要
在传感器网络中,有监督的异常数据检测算法的检测准确率以及鲁棒性受限于有标注数据集的构建,无监督异常数据检测算法往往导致较高的误报率(FPR)。为解决上述问题,针对到达服务器端的传感器数据流提出了一种基于卷积神经网络(CNN)和长短时记忆网络(LSTM)的半监督在线异常检测算法。本算法利用K-means判别检测误差,并在检测中利用新数据重新训练机器学习模型,从而提高模型在长时间范围内的异常检测准确度。为了评估本算法的性能,使用因特尔伯克利实验室数据集IBRL(Intel Berkeley Research Lab)完成仿真实验,并与同类算法进行对比。实验结果表明,与同类算法相比,本算法对各个数据集都具有较高的召回率和F1-Score;应用K-means聚类的半监督模型,其异常检测结果更稳定。
In sensor networks,the accuracy and robustness of supervised anomaly detection algorithms are limited by the construction of labeled datasets.Unsupervised anomaly detection algorithms often result in high false positive rates(FPR).To solve this issue,a semi-supervised online anomaly detection algorithm of sensor data stream using CNN and LSTM is proposed.The proposed algorithm determines the detection error using K-means clustering and retrains the machine learning model with the newly inputted data to detect anomalies.We implemented the proposed algorithm on Intel Berkeley Research Lab datasets(IBRL),and compared it with the existing algorithms.The experimental results demonstrated that the proposed algorithm outperform other methods in term of recall and F1-Score on each dataset.Furthermore,it is found that the semi-supervised learning model using K-means clustering can significantly improve the model stability.
作者
唐海贤
李光辉
TANG Haixian;LI Guanghui(School of Artificial Intelligence and Computer Science,Jiangnan University,Wuxi Jiangsu 214122,China;Research Center of IoT Technology Application Engineering(MOE),Wuxi Jiangsu 214122,China)
出处
《传感技术学报》
CAS
CSCD
北大核心
2021年第3期330-339,共10页
Chinese Journal of Sensors and Actuators
基金
国家自然科学基金项目(62072216)
无锡市国际科技研发合作项目(CZE02H1706)
江苏省农业科技自主创新资金项目(CX(19)3087)。
