Abstract: An observation scheme for quantifying tower-shadow effects on offshore wind measurement masts was proposed, and a one-year wind dataset was acquired from a mast deployment in Chinese sea area. The influence pattern of the mast tower shadow on critical parameters was analyzed, and a tower-shadow influence factor (T_SIF) was introduced to characterize the severity of shadowing within each wind-direction sector. A mast tower shadow effect correction algorithm was then developed combining a temporal convolutional network (TCN) for extracting wind-speed time-series patterns with a multi-head self-attention mechanism (MHSA) for capturing key shadow-region features. Validation at the installed mast shows that the T_SIF serves as an essential auxiliary feature that significantly improves correction accuracy. The mounting azimuth of anemometers strongly affects both wind-resource measurement precision and the efficacy of tower-shadow correction. The proposed algorithm is applicable to most masts used in engineering and requires only two anemometer data series at one measurement height to correct tower-shadow-induced errors at any single-anemometer height, thereby substantially enhancing the wind-resource assessment accuracy.