Deep Fuzzy Contrast-Set Deviation Point Representation and Trajectory Detection

Cooperative intelligent transportation systems (ITS) are on the rise in the field of transportation. The trajectory-based knowledge graph enables the ITS to have semantic and connectivity capabilities. This article presents the approach of embedding trajectory deviation points and deep clustering. We constructed the structural embedding by maintaining the relationship between the nodes based on the network structure and the neighbors of the nodes. This approach was then used to learn the latent representation based on the deviation points in the road network structure. We generated a set of sequences using a hierarchical multilayer network and a biased random walk. This research proposes a fuzzy contrast-based model that identifies deviation points using the deep network for weighted position nodes. This sequence is used to fine-tune the embedding of the nodes. We then averaged the embedding values of the nodes to obtain the travel embedding. Next, we extracted the embedding of the contrast set using a pairwise classification approach based on similarity metrics. Numerical studies show that the proposed learning trajectory embedding approach successfully captures the structural identity and outperforms competing strategies. The deep contrast set approach enables highly accurate detection of outliers in the trajectory and deviation locations.

Automated summary

This article presents the approach of embedding trajectory deviation points and deep clustering. The proposed learning trajectory embedding approach successfully captures the structural identity and outperforms competing strategies in detecting outliers in the trajectory and deviation locations.