Abstract: Association analysis of urban anomalous events is crucial for enhancing city management efficacy and advancing smart city development. Addressing the heterogeneity and complex associations inherent in multidimensional event features, this paper proposes an analytical framework integrating clustering and causal reasoning. On one hand, a density based spatial-temporal clustering algorithm identified explicit spatial-temporal associations among events. On the other hand, large language models facilitated implicit association inference between events. To meet the demand for clustering large-scale urban anomalous events, this paper proposed an improved spatial-temporal clustering algorithm building upon DBSCAN and ST_DBSCAN. It utilized a BallTree index structure and significantly enhanced clustering efficiency through spatial-temporal neighborhood density analysis, enabling rapid processing of million-scale anomalous events. Furthermore, an optimization method for the core clustering parameters of the spatial radius threshold and temporal radius threshold was proposed. Experimental results demonstrate the superior clustering performance of the proposed algorithm when handling massive urban anomalous events.