摘要
为有效提高电网软件恶意攻击的检测效率,使用自动编码器进行异常恶意攻击检测的无监督深度学习方法,同时采用低采样和高采样的混合采样策略来平衡数据集,并对深度学习算法的检测性能及数据丢包率进行分析。实验结果表明,基于深度学习算法在检测电网运营平台恶意攻击时,准确率高达98.84%,真正例率和耗时均比较低(2.1%、11.28 ms),且深度学习算法的召回率高达99.2%;进一步表明基于深度学习所建立的自动编码器可以有效检测到电网运营平台恶意攻击,且检测综合性能优于支持向量机等其他机器学习算法。丢包率随着样本数的增加而降低,当样本数增加到40000个时,丢包率最小约为3%。
In order to effectively improve the detection efficiency of malicious attacks on power grid software,an unsupervised deep learning method for abnormal malicious attack detection using autoencoder was used,and a mixed sampling strategy of low sampling and high sampling was used to balance the data set,and the detection performance and data packet loss rate of the deep learning algorithm were analyzed.The experimental results showed that when detecting malicious attacks on power grid operation platforms based on deep learning algorithms,the accuracy rate was as high as 98.84%,the true case rate and time consumption were relatively low(2.1%,11.28 ms),and the recall rate of deep learning algorithms was as high as 99.2%.This further indicates that the automatic encoder established based on deep learning can effectively detect malicious attacks on power grid operation platforms,and the overall detection performance is superior to other machine learning algorithms such as support vector ma‐chines.The packet loss rate decreased as the number of samples increases.When the sample size increased to 40000,the minimum packet loss rate was about 3%.
作者
李强
张兴富
桂胜
胡博
LI Qiang;ZHANG Xingfu;GUI Sheng;HU Bo(State Grid Information Communications Industry Group Co.,Ltd.,Beijing 102200,China;Beijing CLP Puhua Information Technology Co.,Ltd.,Beijing 100089,China)
出处
《粘接》
CAS
2024年第7期140-143,共4页
Adhesion
关键词
深度学习
自动编码器
电网软件运营平台
恶意攻击
deep learning
automatic encoder
power grid software operation platform
malicious attacks
