Ruliang Xiao

The complexity of urban transportation networks continues to increase, and traffic congestion and accidents remain common. However, identifying critical nodes in intelligent transportation systems is challenging due to the instability of boundary sample determination in network clustering. To address this issue, this paper proposes a critical‐node identification method that integrates traffic semantic discriminability with a boundary‐aware dynamic caching mechanism. Specifically, a spatiotemporal traffic graph representation module performs encoder–decoder collaborative learning on a weighted spatiotemporal traffic graph that incorporates both road network topology and dynamic traffic flow, thereby generating node representations with strong traffic semantic discriminability. In addition, the proposed boundary‐aware dynamic caching mechanism (BA‐DC) conducts local dynamic re‐evaluation and closed‐loop optimization for low‐margin boundary samples and samples located in inter‐cluster transition regions, which enhances intra‐cluster cohesion and improves the stability of boundary sample classification. Meanwhile, the probabilistic structural divergence optimization module (PSD‐Opt) optimizes network structure by maximizing the divergence between category probability distributions, thereby increasing inter‐cluster separation. Furthermore, the Weighted Prototype Entropy Regularization strategy (WPE‐Reg) suppresses intra‐cluster dispersion and mitigates imbalanced cluster assignments through flow‐aware weighted class centers and entropy‐regularization constraints. Experiments on multiple public traffic datasets demonstrate that, compared with representative baseline methods, the proposed approach achieves substantial improvements in clustering performance, boundary sample stability, and critical node identification.

It remains a challenging to capture inter-variable correlations in time series forecasting (TSF), which leads researchers to focus more on channel-independent TSF methods. However, it is unwise to overlook variable interactions in real-world applications. To address this issue, we propose a novel TSF model, named Recurrent Low-rank Cross-attention Network (RLCN), which integrates channel independence and variable interaction. RLCN adopts a dual-branch feature extraction strategy: one branch learns an interpretable symmetric low-rank core matrix to model inter-variable dependencies, while the other employs an RNN-based module to capture global temporal patterns within variables. Subsequently, a cross-attention encoding module integrates these features. Experimental results demonstrate that RLCN achieves high forecasting accuracy with low memory consumption. The ablation study verifies the indispensability of components. In the context of real-world wind forecasting tasks, we analyze the correlations encoded in the low-rank core matrix by integrating meteorological domain knowledge. The matrix reflects inter-variable correlations, identifies key factors influencing wind speed forecasting, endows the model with good interpretability.