LSTM for SCADA Intrusion Detection

Abstract

We present recurrent neural networks (RNN) for supervisory control and data acquisition (SCADA) Intrusion Detection System (IDS). Using long short term memory (LSTM) with many-to-many (MTM) and a novel many-to-one (MTO) architectures, both IDSs display excellent performance in detecting temporal uncorrelated attacks while MTO showing superior performance on temporal correlated attacks.