Abstract: Based on the hourly rainfall data from 158 rainstorm events with storm-center daily precipitation exceeding 200 mm in Anhui Province from 1953 to 2024, this study constructed a multidimensional characteristic index system integrating peak time coefficient, average rainfall intensity, rain peak coefficient and skewness coefficient, and applied a coupled algorithm combining Ward hierarchical clustering and fuzzy C-means (FCM) clustering to systematically classify and analyze the spatiotemporal evolution of rainstorm patterns during the flood season in Anhui Province. The results show that rainstorms in Anhui Province can be divided into four typical rainfall patterns: persistent early-peak, classical mid-peak, extreme late-peak and stepwise late-peak patterns. Among them, the persistent early-peak pattern has the characteristics of “early peaking and long-tail attenuation”, and the disaster risk is high; the rainfall centroid of classical mid-peak rainfall is in the middle, and the shape is rounded, making it the most common pattern. The stepwise late-peak pattern is characterized by a wide peak with a stepwise rise; extreme late-peak precipitation is highly concentrated at the end of the period, and its abruptness is pronounced, making it the main reason for the formation of steep flood peaks. In terms of spatial and temporal distribution, the occurrence frequencies of various rainfall patterns showed significant seasonal concentration during the year, with events highly concentrated in the main flood season from June to August, especially in July, when late-peak rainfall dominated. From the perspective of spatial pattern, the frequency of rainstorms shows the spatial pattern of “more in the south and less in the north, more in mountainous areas and less in the plains”. The Dabie Mountain region is the high-frequency center of rainstorms, accounting for 27.8% of the total. Differences in topographic uplift and water vapor conditions led to regional variations in rainfall-pattern composition: early-peak rainfall accounted for a relatively high proportion in the Huaibei area, while late-peak rainfall was the main type in the Jianghuai, Jiangnan and mountainous areas. The interdecadal evolution analysis reveals that the frequency of regional rainfall events has increased significantly since 2010, with a significant abrupt change around 2020. After this change, the dominance of extreme late-peak rainfall was further strengthened, accounting for up to 57.7% in the recent period, indicating that rainstorm patterns in Anhui Province are evolving toward greater extremes and later peaks. In terms of research methods, this study applies a coupled approach combining Ward hierarchical clustering and FCM clustering. To address the limitation that the traditional FCM algorithm is sensitive to initial cluster centers and can easily converge to a local optimum, this study first uses Ward hierarchical clustering to determine the optimal number of clusters and the corresponding initial cluster centers, and uses them as initial inputs to the FCM algorithm for soft partitioning. This strategy effectively avoids iterative instability caused by the random initialization of the FCM algorithm, significantly improves the objectivity and stability of rainstorm-pattern classification results, and lays a solid methodological foundation for accurately revealing the spatiotemporal evolution of rainfall patterns. The research conclusions reveal a clear shift toward extreme late-peak patterns with high hazard potential in Anhui Province, providing an important scientific basis for the revision of regional water conservancy project design standards, the precise prevention and control of flood disasters and water resources management. In particular, attention should be paid to the risk of delayed flood peaks caused by a lagging rainfall centroid in the future.