基于LSTM的工控入侵检测

Industrial Control Intrusion Detection based on LSTM

工业控制系统是社会运行和国家安全的重要基础。为提高工业控制系统的入侵检测能力,设计一个基于深度神经网络的工业入侵检测方案。采用与真实世界水处理厂非常相似的六级水处理模型SWaT,通过训练长短期记忆循环神经网络,对经过PCA降维的SWaT预处理数据进行检测和鉴别。结果显示,检测方案在测试集的表现上优于过去的OCSVM、IF、Matrix Profile这3种检测方案,并且网络训练速度更快,检测需要的上下文环境更少,具有一定的实用价值。

Industrial control system is an important foundation for social operation and national security.In order to improve the intrusion detection capability of industrial control system,an industrial intrusion detection scheme based on deep neural network is designed.By using the six level water treatment model SWaT,which is very similar to the real world water treatment plant,through training the LSTM(long and short term memory)loop neural network,the SWaT preprocessed data after PCA dimensionality reduction is detected and identified.The experiment results indicate that the performance of this detection scheme is better than that of the past three detection schemes OCSVM,IF and Matrix Profile,and moreover,the network training speed is faster,and the detection requires less context,and that this detection scheme thus has certain practical value.