In-vehicle intrusion detection model based on multi-channel graph convolutional network

With the rapid development of vehicle technology, innovative applications such as autonomous driving and the internet of vehicles are gradually becoming popular, and the in-vehicle network undoubtedly faces an increasing threat of malicious attacks.Therefore, building an efficient and accurate in-vehicle intrusion detection model is crucial.However, most existing methods rely on pre-set fixed features and lack the ability to dynamically adapt to network changes.This leads to their poor performance in dealing with complex and ever-changing attack patterns, making it difficult to cope with diverse attacks in actual in-vehicular networks.To address this challenge, we have designed an in-vehicle intrusion detection model based on multi-channel graph convolution.The core idea is to construct corresponding spatiotemporal topology graphs and feature coupling graphs from the fine-grained level of in-vehicle network data streams and use multi-channel graph convolution to learn static standard node embeddings and adaptive shared node embeddings from different graph structures, and fully integrate these embedding vectors for the final detection task.We conducted experimental validation of our designed model on a publicly available dataset Car Hacking Dataset, fully demonstrates the effectiveness and superiority of our model in-vehicle network intrusion detection.