Detecting Temporal Attacks: An Intrusion Detection System for Train Communication Ethernet Based on Dynamic Temporal Convolutional Network

The train communication Ethernet (TCE) of modern intelligent trains is under an ever-increasing threat serious network attacks. Denial service (DoS) and man in the middle (MITM), two most destructive attacks against TCE, are difficult to detect by conventional methods. Aiming at their highly time-correlated properties, a novel dynamic temporal convolutional network-based intrusion detection system (DyTCN-IDS) proposed this paper these A semiphysical TCE testbed that capable simulating real situations TCE-based first built generate effective dataset for training testing. DyTCN-IDS consists phases, phase, systematic feature engineering designed optimize dataset. In second basic model good dealing with features utilizing several architectural optimizations. Then, order decrease computational consumption waste on packet sequences different lengths inner relationships, neural technology further adopted model. Diverse experiments were carried out evaluate from angles. experimental results indicate our easy train, converges fast, costs less resources, achieves satisfying performance macro false alarm rate 0.09%, F-score 99.39%, accuracy 99.40%. Compared some canonical DL-based IDS latest IDS, acquires best overall as well.