Intrusion Detection System for Vehicle Network based on Embedded Convolutional LSTM

The CAN bus is a mainstream explicit protocol for data transmission in intelligent vehicles. Therefore, the secure transmission of CAN bus is an essential means to ensure the safety of intelligent vehicles. Traditional models ignore the CAN message-to-message's time sequence and the continuity of the internal characteristics of CAN data frames. To this end, this paper proposes an intrusion detection system for vehicle network based on an embedded convolutional LSTM (EC-LSTM-VNIDS). Firstly, a new embedded convolutional gate is inserted between the cellular state and the implicit state to extract the correlation of the internal features of the CAN data frame via a one-dimensional convolutional operation. Secondly, the model extracts the time relationship of CAN message-to-message by LSTM. Finally, the model analyzes the continuity of the internal features of the CAN data frames and the time relationship of the message-to-message to classify them by the fully connected layer. Experiments show that the EC-LSTM-VNIDS proposed in this paper can achieve more than 99% in all four evaluation metrics on four real-world in-vehicle CAN bus message datasets.