Research on Anomaly Detection in Vehicular CAN Based on Bi-LSTM

Abstract

Controller Area Network (CAN) is one of the most widely used in-vehicle networks in modern vehicles. Due to the lack of security mechanisms such as encryption and authentication, CAN is vulnerable to external hackers in the intelligent network environment. In the paper, a lightweight CAN bus anomaly detection model based on the Bi-LSTM model is proposed. The Bi-LSTM model learns ID sequence correlation features to detect anomalies. At the same time, the Attention mechanism is introduced to improve the model’s efficiency. The paper focuses on replay attacks, denial of service attacks and fuzzing attacks. The experimental results show that the anomaly detection model based on Bi-LSTM can detect three attack types quickly and accurately.