基于AMCNN-LSTM的电力无线接入专网异常流量检测

为了减轻电力无线专网系统因网络业务增多而带来的网络攻击以及异常流量入侵的安全事故隐患,提出了一种基于注意力机制的卷积-长短期记忆网络(convolution-long short-term memory network based on attention mecha-nism,AMCNN-LSTM)模型.该模型为避免序列特征稀疏分布的问题,采用卷积神经网络(convolutional neural net-work,CNN)提取时间序列数据特征并转化为维度固定的稠密向量;为防止记忆丢失和梯度分散问题,使用融合注意力机制的CNN单元来捕捉重要的时间序列细粒度特征;将CNN提取局部特征与长短期记忆网络(long short-term memory network,LSTM)提取序列特征的优势相结合,对电力接入专网流量数据进行异常检测.通过在电力网真实数据集上实验表明,基于注意力机制的算法能够在150轮次迭代下达到89.14％的召回率及89.67％的综合F-meas-ure得分.所提出的模型能够及时、准确地检测电力网络异常流量,有效提高检测效率及准确度.

Research on anomaly detection of power network based on AMCNN-LSTM

In order to reduce the hidden dangers of network attacks and abnormal traffic intrusion caused by the increase of network services in the electric power wireless private network system, a convolution-long short-term memory network (AMCNN-LSTM) model based on the attention mechanism is proposed. First, CNN is used to extract the features of time series data and converted into dense vectors with fixed dimensions; secondly, in order to prevent memory loss and gradient dispersion problems, the CNN unit with the attention mechanism is used to capture important time series fine-grained features; finally, anomaly detection is carried out on the traffic data of power access private network. Experiments on the real data set of the power grid show that the algorithm based on the attention mechanism can achieve a recall rate of 89.14% and a comprehensive F-measure score of 89.67% under 150 iterations. Experimental research shows that the model proposed in this paper can detect the abnormal flow of the power network in a timely and accurate manner, and effectively improve the detection efficiency and accuracy.