Intrusion detection method for communication behaviors in smart substations based on soft-membership graph partitioning and sequence analysis
Abstract
Smart substation communication networks are characterized by sparse topologies, strong service correlations, and highly regular communication behaviors, making it difficult for conventional intrusion detection methods to adequately characterize communication topologies and cross-protocol behavioral correlations. To address this issue, an intrusion detection method for communication behaviors is proposed by integrating soft-membership graph partitioning with sequence analysis. A directed information-flow graph is constructed from communication packets, and communication regions are identified through weighted random-walk spectral clustering and soft-membership optimization. Node2Vec-based node representation and frequent sequence mining are then employed to establish a normal communication sequence library, while sequence matching is used to identify anomalous behaviors. Experimental results demonstrate that the proposed method preserves communication topology integrity and effectively detects random replay, reverse replay, and random message injection attacks, achieving an overall detection accuracy of 92.87%.