基于CapsNet和SRU的工业互联网入侵检测方法

Intrusion Detection Model Based on CapsNet and SRU for Industrial Internet

随着工业互联网的普及,工业系统中大量的基础设施和设备接入互联网,使得工业系统更容易受到外部攻击,工业互联网入侵检测成为保障工业网络安全和稳定运营的重要手段。针对现有的深度学习方法在工业互联网入侵检测中存在数据特征提取不全和对罕见攻击检测准确率低的问题,提出一种基于胶囊网络(CapsNet)和简单循环单元(SRU)融合的工业互联网入侵检测模型。采用SMOTE-ENN算法完成数据的平衡处理,结合简单循环单元神经网络和引入残差块的胶囊网络分别提取流量数据的时间和空间特征,通过自注意力机制进行特征加权,进而提高模型的检测性能。在气体管道数据集上,对比经过SMOTE-ENN算法处理前后的预测结果,该模型对MSCI和MFCI类别的识别精度分别提高4.69百分点和4.41百分点,表明数据平衡算法提高了分类器对少数类样本的预测能力,对比其他模型,该模型的准确率达到99.36%,误报率为0.73%。

With the popularization of the industrial Internet,a large number of infrastructure and equipment in industrial systems are connected to the Internet,making industrial systems more vulnerable to external attacks.Industrial Internet intrusion detection has become an important means to ensure the security and stable operation of industrial networks.Existing deep learning methods have problems with incomplete data feature extraction and low accuracy in detecting rare attacks in industrial Internet intrusion detection.Therefore,an industrial Internet intrusion detection model based on the fusion of Capsule Network(CapsNet)and Simple Recurrent Unit(SRU)is proposed.The SMOTE-ENN algorithm is used to complete the balanced processing of data,combining the simple recurrent unit neural network and the capsule that introduces the residual block.The network extracts the temporal and spatial features of the traffic data respectively,and weights the features through the self-attention mechanism,thereby improving the detection performance of the model.On the gas pipeline data set,comparing the prediction results before and after processing by the SMOTE-ENN algorithm,the proposed model’s recognition accuracy for the MSCI and MFCI categories is increased by 4.69 percentage points and 4.41 percentage points respectively,indicating that the data balancing algorithm improves the classifier’s prediction ability for a few categories of samples.Compared with other models,the accuracy of the proposed model reaches 99.36%and the false positive rate is 0.73%.