Tropical cyclones (TC) are a critical hazard threatening the safety of wind power infrastructure along China’s southern coastal regions, as they generate extreme wind speeds that far exceed conventional design standards. There is an urgent need to develop improved methods for estimating design wind speeds in TC-prone areas. Using the CMA-STI best-track tropical cyclone dataset from 1949 to 2024 and focusing on Xuwen, Guangdong Province, this study employs the Jelesnianski wind field model to reconstruct hourly maximum wind speeds induced by TCs, yielding 183 effective extreme wind speed samples with values ≥12.0 m?s?1. Building upon this dataset, a new Poisson–Pearson Type Ⅲ compound extreme value distribution model is developed to jointly simulate the annual TC occurrence frequency and the maximum wind speed associated with individual TC events. The results show that Xuwen experiences an average of 2.4 TC impacts per year, predominantly occurring from July to October. The 50-year return period 10-minute design wind speed is estimated at 44.8 m?s?1—significantly higher than estimates from single extreme value distributions (e.g., Gumbel, Generalized Extreme Value) and the Poisson–Gumbel compound model, and slightly exceeding the recommended value of 43.6 m?s?1 in the current Wind Resistance Design Specification for Highway Bridges. By fully utilizing the TC event sample and achieving superior tail fitting, the proposed method effectively captures the statistical characteristics of TC-induced extreme winds, offering a practical and robust approach for the safe design of offshore wind farms, long-span bridges, and other critical infrastructure in high-risk TC zones.