Heterogeneous Graph Neural Network-based Network Security Situation Assessment for In-Vehicle Network

In response to the issues in current neural network-based cybersecurity situation assessment methods for In-Vehicle Network (IVN), which ignore the associations between situation entities and lack representation learning of spatial structural information in situation features—thereby limiting the accuracy of situation assessment—this paper proposes a situation assessment scheme based on heterogeneous graph neural networks. This scheme divides the graph into multiple meta-path sequences through meta-paths and uses Bidirectional Long Short-Term Memory (Bi-LSTM) networks and attention mechanisms to obtain attribute features of each node. By combining these with the structural features obtained from Metapath2Vec, node feature representations are acquired. The overall representation of the vehicle nodes is then generated using an attention mechanism for situation assessment. Both theoretical analysis and experimental results indicate that this method can accurately evaluate the level of security situational level of IVN.